diff options
Diffstat (limited to 'libopus/celt')
93 files changed, 26058 insertions, 0 deletions
diff --git a/libopus/celt/_kiss_fft_guts.h b/libopus/celt/_kiss_fft_guts.h new file mode 100644 index 0000000..17392b3 --- /dev/null +++ b/libopus/celt/_kiss_fft_guts.h @@ -0,0 +1,182 @@ +/*Copyright (c) 2003-2004, Mark Borgerding + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_GUTS_H +#define KISS_FFT_GUTS_H + +#define MIN(a,b) ((a)<(b) ? (a):(b)) +#define MAX(a,b) ((a)>(b) ? (a):(b)) + +/* kiss_fft.h + defines kiss_fft_scalar as either short or a float type + and defines + typedef struct { kiss_fft_scalar r; kiss_fft_scalar i; }kiss_fft_cpx; */ +#include "kiss_fft.h" + +/* + Explanation of macros dealing with complex math: + + C_MUL(m,a,b) : m = a*b + C_FIXDIV( c , div ) : if a fixed point impl., c /= div. noop otherwise + C_SUB( res, a,b) : res = a - b + C_SUBFROM( res , a) : res -= a + C_ADDTO( res , a) : res += a + * */ +#ifdef FIXED_POINT +#include "arch.h" + + +#define SAMP_MAX 2147483647 +#define TWID_MAX 32767 +#define TRIG_UPSCALE 1 + +#define SAMP_MIN -SAMP_MAX + + +# define S_MUL(a,b) MULT16_32_Q15(b, a) + +# define C_MUL(m,a,b) \ + do{ (m).r = SUB32_ovflw(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \ + (m).i = ADD32_ovflw(S_MUL((a).r,(b).i) , S_MUL((a).i,(b).r)); }while(0) + +# define C_MULC(m,a,b) \ + do{ (m).r = ADD32_ovflw(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \ + (m).i = SUB32_ovflw(S_MUL((a).i,(b).r) , S_MUL((a).r,(b).i)); }while(0) + +# define C_MULBYSCALAR( c, s ) \ + do{ (c).r = S_MUL( (c).r , s ) ;\ + (c).i = S_MUL( (c).i , s ) ; }while(0) + +# define DIVSCALAR(x,k) \ + (x) = S_MUL( x, (TWID_MAX-((k)>>1))/(k)+1 ) + +# define C_FIXDIV(c,div) \ + do { DIVSCALAR( (c).r , div); \ + DIVSCALAR( (c).i , div); }while (0) + +#define C_ADD( res, a,b)\ + do {(res).r=ADD32_ovflw((a).r,(b).r); (res).i=ADD32_ovflw((a).i,(b).i); \ + }while(0) +#define C_SUB( res, a,b)\ + do {(res).r=SUB32_ovflw((a).r,(b).r); (res).i=SUB32_ovflw((a).i,(b).i); \ + }while(0) +#define C_ADDTO( res , a)\ + do {(res).r = ADD32_ovflw((res).r, (a).r); (res).i = ADD32_ovflw((res).i,(a).i);\ + }while(0) + +#define C_SUBFROM( res , a)\ + do {(res).r = ADD32_ovflw((res).r,(a).r); (res).i = SUB32_ovflw((res).i,(a).i); \ + }while(0) + +#if defined(OPUS_ARM_INLINE_ASM) +#include "arm/kiss_fft_armv4.h" +#endif + +#if defined(OPUS_ARM_INLINE_EDSP) +#include "arm/kiss_fft_armv5e.h" +#endif +#if defined(MIPSr1_ASM) +#include "mips/kiss_fft_mipsr1.h" +#endif + +#else /* not FIXED_POINT*/ + +# define S_MUL(a,b) ( (a)*(b) ) +#define C_MUL(m,a,b) \ + do{ (m).r = (a).r*(b).r - (a).i*(b).i;\ + (m).i = (a).r*(b).i + (a).i*(b).r; }while(0) +#define C_MULC(m,a,b) \ + do{ (m).r = (a).r*(b).r + (a).i*(b).i;\ + (m).i = (a).i*(b).r - (a).r*(b).i; }while(0) + +#define C_MUL4(m,a,b) C_MUL(m,a,b) + +# define C_FIXDIV(c,div) /* NOOP */ +# define C_MULBYSCALAR( c, s ) \ + do{ (c).r *= (s);\ + (c).i *= (s); }while(0) +#endif + +#ifndef CHECK_OVERFLOW_OP +# define CHECK_OVERFLOW_OP(a,op,b) /* noop */ +#endif + +#ifndef C_ADD +#define C_ADD( res, a,b)\ + do { \ + CHECK_OVERFLOW_OP((a).r,+,(b).r)\ + CHECK_OVERFLOW_OP((a).i,+,(b).i)\ + (res).r=(a).r+(b).r; (res).i=(a).i+(b).i; \ + }while(0) +#define C_SUB( res, a,b)\ + do { \ + CHECK_OVERFLOW_OP((a).r,-,(b).r)\ + CHECK_OVERFLOW_OP((a).i,-,(b).i)\ + (res).r=(a).r-(b).r; (res).i=(a).i-(b).i; \ + }while(0) +#define C_ADDTO( res , a)\ + do { \ + CHECK_OVERFLOW_OP((res).r,+,(a).r)\ + CHECK_OVERFLOW_OP((res).i,+,(a).i)\ + (res).r += (a).r; (res).i += (a).i;\ + }while(0) + +#define C_SUBFROM( res , a)\ + do {\ + CHECK_OVERFLOW_OP((res).r,-,(a).r)\ + CHECK_OVERFLOW_OP((res).i,-,(a).i)\ + (res).r -= (a).r; (res).i -= (a).i; \ + }while(0) +#endif /* C_ADD defined */ + +#ifdef FIXED_POINT +/*# define KISS_FFT_COS(phase) TRIG_UPSCALE*floor(MIN(32767,MAX(-32767,.5+32768 * cos (phase)))) +# define KISS_FFT_SIN(phase) TRIG_UPSCALE*floor(MIN(32767,MAX(-32767,.5+32768 * sin (phase))))*/ +# define KISS_FFT_COS(phase) floor(.5+TWID_MAX*cos (phase)) +# define KISS_FFT_SIN(phase) floor(.5+TWID_MAX*sin (phase)) +# define HALF_OF(x) ((x)>>1) +#elif defined(USE_SIMD) +# define KISS_FFT_COS(phase) _mm_set1_ps( cos(phase) ) +# define KISS_FFT_SIN(phase) _mm_set1_ps( sin(phase) ) +# define HALF_OF(x) ((x)*_mm_set1_ps(.5f)) +#else +# define KISS_FFT_COS(phase) (kiss_fft_scalar) cos(phase) +# define KISS_FFT_SIN(phase) (kiss_fft_scalar) sin(phase) +# define HALF_OF(x) ((x)*.5f) +#endif + +#define kf_cexp(x,phase) \ + do{ \ + (x)->r = KISS_FFT_COS(phase);\ + (x)->i = KISS_FFT_SIN(phase);\ + }while(0) + +#define kf_cexp2(x,phase) \ + do{ \ + (x)->r = TRIG_UPSCALE*celt_cos_norm((phase));\ + (x)->i = TRIG_UPSCALE*celt_cos_norm((phase)-32768);\ +}while(0) + +#endif /* KISS_FFT_GUTS_H */ diff --git a/libopus/celt/arch.h b/libopus/celt/arch.h new file mode 100644 index 0000000..08b07db --- /dev/null +++ b/libopus/celt/arch.h @@ -0,0 +1,288 @@ +/* Copyright (c) 2003-2008 Jean-Marc Valin + Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file arch.h + @brief Various architecture definitions for CELT +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef ARCH_H +#define ARCH_H + +#include "opus_types.h" +#include "opus_defines.h" + +# if !defined(__GNUC_PREREQ) +# if defined(__GNUC__)&&defined(__GNUC_MINOR__) +# define __GNUC_PREREQ(_maj,_min) \ + ((__GNUC__<<16)+__GNUC_MINOR__>=((_maj)<<16)+(_min)) +# else +# define __GNUC_PREREQ(_maj,_min) 0 +# endif +# endif + +#if OPUS_GNUC_PREREQ(3, 0) +#define opus_likely(x) (__builtin_expect(!!(x), 1)) +#define opus_unlikely(x) (__builtin_expect(!!(x), 0)) +#else +#define opus_likely(x) (!!(x)) +#define opus_unlikely(x) (!!(x)) +#endif + +#define CELT_SIG_SCALE 32768.f + +#define CELT_FATAL(str) celt_fatal(str, __FILE__, __LINE__); + +#if defined(ENABLE_ASSERTIONS) || defined(ENABLE_HARDENING) +#ifdef __GNUC__ +__attribute__((noreturn)) +#endif +void celt_fatal(const char *str, const char *file, int line); + +#if defined(CELT_C) && !defined(OVERRIDE_celt_fatal) +#include <stdio.h> +#include <stdlib.h> +#ifdef __GNUC__ +__attribute__((noreturn)) +#endif +void celt_fatal(const char *str, const char *file, int line) +{ + fprintf (stderr, "Fatal (internal) error in %s, line %d: %s\n", file, line, str); + abort(); +} +#endif + +#define celt_assert(cond) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond);}} +#define celt_assert2(cond, message) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond "\n" message);}} +#define MUST_SUCCEED(call) celt_assert((call) == OPUS_OK) +#else +#define celt_assert(cond) +#define celt_assert2(cond, message) +#define MUST_SUCCEED(call) do {if((call) != OPUS_OK) {RESTORE_STACK; return OPUS_INTERNAL_ERROR;} } while (0) +#endif + +#if defined(ENABLE_ASSERTIONS) +#define celt_sig_assert(cond) {if (!(cond)) {CELT_FATAL("signal assertion failed: " #cond);}} +#else +#define celt_sig_assert(cond) +#endif + +#define IMUL32(a,b) ((a)*(b)) + +#define MIN16(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum 16-bit value. */ +#define MAX16(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 16-bit value. */ +#define MIN32(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum 32-bit value. */ +#define MAX32(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 32-bit value. */ +#define IMIN(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum int value. */ +#define IMAX(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum int value. */ +#define UADD32(a,b) ((a)+(b)) +#define USUB32(a,b) ((a)-(b)) + +/* Set this if opus_int64 is a native type of the CPU. */ +/* Assume that all LP64 architectures have fast 64-bit types; also x86_64 + (which can be ILP32 for x32) and Win64 (which is LLP64). */ +#if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64) +#define OPUS_FAST_INT64 1 +#else +#define OPUS_FAST_INT64 0 +#endif + +#define PRINT_MIPS(file) + +#ifdef FIXED_POINT + +typedef opus_int16 opus_val16; +typedef opus_int32 opus_val32; +typedef opus_int64 opus_val64; + +typedef opus_val32 celt_sig; +typedef opus_val16 celt_norm; +typedef opus_val32 celt_ener; + +#define celt_isnan(x) 0 + +#define Q15ONE 32767 + +#define SIG_SHIFT 12 +/* Safe saturation value for 32-bit signals. Should be less than + 2^31*(1-0.85) to avoid blowing up on DC at deemphasis.*/ +#define SIG_SAT (300000000) + +#define NORM_SCALING 16384 + +#define DB_SHIFT 10 + +#define EPSILON 1 +#define VERY_SMALL 0 +#define VERY_LARGE16 ((opus_val16)32767) +#define Q15_ONE ((opus_val16)32767) + +#define SCALEIN(a) (a) +#define SCALEOUT(a) (a) + +#define ABS16(x) ((x) < 0 ? (-(x)) : (x)) +#define ABS32(x) ((x) < 0 ? (-(x)) : (x)) + +static OPUS_INLINE opus_int16 SAT16(opus_int32 x) { + return x > 32767 ? 32767 : x < -32768 ? -32768 : (opus_int16)x; +} + +#ifdef FIXED_DEBUG +#include "fixed_debug.h" +#else + +#include "fixed_generic.h" + +#ifdef OPUS_ARM_PRESUME_AARCH64_NEON_INTR +#include "arm/fixed_arm64.h" +#elif defined (OPUS_ARM_INLINE_EDSP) +#include "arm/fixed_armv5e.h" +#elif defined (OPUS_ARM_INLINE_ASM) +#include "arm/fixed_armv4.h" +#elif defined (BFIN_ASM) +#include "fixed_bfin.h" +#elif defined (TI_C5X_ASM) +#include "fixed_c5x.h" +#elif defined (TI_C6X_ASM) +#include "fixed_c6x.h" +#endif + +#endif + +#else /* FIXED_POINT */ + +typedef float opus_val16; +typedef float opus_val32; +typedef float opus_val64; + +typedef float celt_sig; +typedef float celt_norm; +typedef float celt_ener; + +#ifdef FLOAT_APPROX +/* This code should reliably detect NaN/inf even when -ffast-math is used. + Assumes IEEE 754 format. */ +static OPUS_INLINE int celt_isnan(float x) +{ + union {float f; opus_uint32 i;} in; + in.f = x; + return ((in.i>>23)&0xFF)==0xFF && (in.i&0x007FFFFF)!=0; +} +#else +#ifdef __FAST_MATH__ +#error Cannot build libopus with -ffast-math unless FLOAT_APPROX is defined. This could result in crashes on extreme (e.g. NaN) input +#endif +#define celt_isnan(x) ((x)!=(x)) +#endif + +#define Q15ONE 1.0f + +#define NORM_SCALING 1.f + +#define EPSILON 1e-15f +#define VERY_SMALL 1e-30f +#define VERY_LARGE16 1e15f +#define Q15_ONE ((opus_val16)1.f) + +/* This appears to be the same speed as C99's fabsf() but it's more portable. */ +#define ABS16(x) ((float)fabs(x)) +#define ABS32(x) ((float)fabs(x)) + +#define QCONST16(x,bits) (x) +#define QCONST32(x,bits) (x) + +#define NEG16(x) (-(x)) +#define NEG32(x) (-(x)) +#define NEG32_ovflw(x) (-(x)) +#define EXTRACT16(x) (x) +#define EXTEND32(x) (x) +#define SHR16(a,shift) (a) +#define SHL16(a,shift) (a) +#define SHR32(a,shift) (a) +#define SHL32(a,shift) (a) +#define PSHR32(a,shift) (a) +#define VSHR32(a,shift) (a) + +#define PSHR(a,shift) (a) +#define SHR(a,shift) (a) +#define SHL(a,shift) (a) +#define SATURATE(x,a) (x) +#define SATURATE16(x) (x) + +#define ROUND16(a,shift) (a) +#define SROUND16(a,shift) (a) +#define HALF16(x) (.5f*(x)) +#define HALF32(x) (.5f*(x)) + +#define ADD16(a,b) ((a)+(b)) +#define SUB16(a,b) ((a)-(b)) +#define ADD32(a,b) ((a)+(b)) +#define SUB32(a,b) ((a)-(b)) +#define ADD32_ovflw(a,b) ((a)+(b)) +#define SUB32_ovflw(a,b) ((a)-(b)) +#define MULT16_16_16(a,b) ((a)*(b)) +#define MULT16_16(a,b) ((opus_val32)(a)*(opus_val32)(b)) +#define MAC16_16(c,a,b) ((c)+(opus_val32)(a)*(opus_val32)(b)) + +#define MULT16_32_Q15(a,b) ((a)*(b)) +#define MULT16_32_Q16(a,b) ((a)*(b)) + +#define MULT32_32_Q31(a,b) ((a)*(b)) + +#define MAC16_32_Q15(c,a,b) ((c)+(a)*(b)) +#define MAC16_32_Q16(c,a,b) ((c)+(a)*(b)) + +#define MULT16_16_Q11_32(a,b) ((a)*(b)) +#define MULT16_16_Q11(a,b) ((a)*(b)) +#define MULT16_16_Q13(a,b) ((a)*(b)) +#define MULT16_16_Q14(a,b) ((a)*(b)) +#define MULT16_16_Q15(a,b) ((a)*(b)) +#define MULT16_16_P15(a,b) ((a)*(b)) +#define MULT16_16_P13(a,b) ((a)*(b)) +#define MULT16_16_P14(a,b) ((a)*(b)) +#define MULT16_32_P16(a,b) ((a)*(b)) + +#define DIV32_16(a,b) (((opus_val32)(a))/(opus_val16)(b)) +#define DIV32(a,b) (((opus_val32)(a))/(opus_val32)(b)) + +#define SCALEIN(a) ((a)*CELT_SIG_SCALE) +#define SCALEOUT(a) ((a)*(1/CELT_SIG_SCALE)) + +#define SIG2WORD16(x) (x) + +#endif /* !FIXED_POINT */ + +#ifndef GLOBAL_STACK_SIZE +#ifdef FIXED_POINT +#define GLOBAL_STACK_SIZE 120000 +#else +#define GLOBAL_STACK_SIZE 120000 +#endif +#endif + +#endif /* ARCH_H */ diff --git a/libopus/celt/arm/arm2gnu.pl b/libopus/celt/arm/arm2gnu.pl new file mode 100755 index 0000000..a2895f7 --- /dev/null +++ b/libopus/celt/arm/arm2gnu.pl @@ -0,0 +1,353 @@ +#!/usr/bin/perl +# Copyright (C) 2002-2013 Xiph.org Foundation +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions +# are met: +# +# - Redistributions of source code must retain the above copyright +# notice, this list of conditions and the following disclaimer. +# +# - Redistributions in binary form must reproduce the above copyright +# notice, this list of conditions and the following disclaimer in the +# documentation and/or other materials provided with the distribution. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +# ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER +# OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +# LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +# NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +# SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +my $bigend; # little/big endian +my $nxstack; +my $apple = 0; +my $symprefix = ""; + +$nxstack = 0; + +eval 'exec /usr/local/bin/perl -S $0 ${1+"$@"}' + if $running_under_some_shell; + +while ($ARGV[0] =~ /^-/) { + $_ = shift; + last if /^--$/; + if (/^-n$/) { + $nflag++; + next; + } + if (/^--apple$/) { + $apple = 1; + $symprefix = "_"; + next; + } + die "I don't recognize this switch: $_\\n"; +} +$printit++ unless $nflag; + +$\ = "\n"; # automatically add newline on print +$n=0; + +$thumb = 0; # ARM mode by default, not Thumb. +@proc_stack = (); + +printf (" .syntax unified\n"); + +LINE: +while (<>) { + + # For ADRLs we need to add a new line after the substituted one. + $addPadding = 0; + + # First, we do not dare to touch *anything* inside double quotes, do we? + # Second, if you want a dollar character in the string, + # insert two of them -- that's how ARM C and assembler treat strings. + s/^([A-Za-z_]\w*)[ \t]+DCB[ \t]*\"/$1: .ascii \"/ && do { s/\$\$/\$/g; next }; + s/\bDCB\b[ \t]*\"/.ascii \"/ && do { s/\$\$/\$/g; next }; + s/^(\S+)\s+RN\s+(\S+)/$1 .req r$2/ && do { s/\$\$/\$/g; next }; + # If there's nothing on a line but a comment, don't try to apply any further + # substitutions (this is a cheap hack to avoid mucking up the license header) + s/^([ \t]*);/$1@/ && do { s/\$\$/\$/g; next }; + # If substituted -- leave immediately ! + + s/@/,:/; + s/;/@/; + while ( /@.*'/ ) { + s/(@.*)'/$1/g; + } + s/\{FALSE\}/0/g; + s/\{TRUE\}/1/g; + s/\{(\w\w\w\w+)\}/$1/g; + s/\bINCLUDE[ \t]*([^ \t\n]+)/.include \"$1\"/; + s/\bGET[ \t]*([^ \t\n]+)/.include \"${ my $x=$1; $x =~ s|\.s|-gnu.S|; \$x }\"/; + s/\bIMPORT\b/.extern/; + s/\bEXPORT\b\s*/.global $symprefix/; + s/^(\s+)\[/$1IF/; + s/^(\s+)\|/$1ELSE/; + s/^(\s+)\]/$1ENDIF/; + s/IF *:DEF:/ .ifdef/; + s/IF *:LNOT: *:DEF:/ .ifndef/; + s/ELSE/ .else/; + s/ENDIF/ .endif/; + + if( /\bIF\b/ ) { + s/\bIF\b/ .if/; + s/=/==/; + } + if ( $n == 2) { + s/\$/\\/g; + } + if ($n == 1) { + s/\$//g; + s/label//g; + $n = 2; + } + if ( /MACRO/ ) { + s/MACRO *\n/.macro/; + $n=1; + } + if ( /\bMEND\b/ ) { + s/\bMEND\b/.endm/; + $n=0; + } + + # ".rdata" doesn't work in 'as' version 2.13.2, as it is ".rodata" there. + # + if ( /\bAREA\b/ ) { + my $align; + $align = "2"; + if ( /ALIGN=(\d+)/ ) { + $align = $1; + } + if ( /CODE/ ) { + $nxstack = 1; + } + s/^(.+)CODE(.+)READONLY(.*)/ .text/; + s/^(.+)DATA(.+)READONLY(.*)/ .section .rdata/; + s/^(.+)\|\|\.data\|\|(.+)/ .data/; + s/^(.+)\|\|\.bss\|\|(.+)/ .bss/; + s/$/; .p2align $align/; + # Enable NEON instructions but don't produce a binary that requires + # ARMv7. RVCT does not have equivalent directives, so we just do this + # for all CODE areas. + if ( /.text/ ) { + # Separating .arch, .fpu, etc., by semicolons does not work (gas + # thinks the semicolon is part of the arch name, even when there's + # whitespace separating them). Sadly this means our line numbers + # won't match the original source file (we could use the .line + # directive, which is documented to be obsolete, but then gdb will + # show the wrong line in the translated source file). + s/$/; .arch armv7-a\n .fpu neon\n .object_arch armv4t/ unless ($apple); + } + } + + s/\|\|\.constdata\$(\d+)\|\|/.L_CONST$1/; # ||.constdata$3|| + s/\|\|\.bss\$(\d+)\|\|/.L_BSS$1/; # ||.bss$2|| + s/\|\|\.data\$(\d+)\|\|/.L_DATA$1/; # ||.data$2|| + s/\|\|([a-zA-Z0-9_]+)\@([a-zA-Z0-9_]+)\|\|/@ $&/; + s/^(\s+)\%(\s)/ .space $1/; + + s/\|(.+)\.(\d+)\|/\.$1_$2/; # |L80.123| -> .L80_123 + s/\bCODE32\b/.code 32/ && do {$thumb = 0}; + s/\bCODE16\b/.code 16/ && do {$thumb = 1}; + if (/\bPROC\b/) + { + my $prefix; + my $proc; + /^([A-Za-z_\.]\w+)\b/; + $proc = $1; + $prefix = ""; + if ($proc) + { + $prefix = $prefix.sprintf("\t.type\t%s, %%function", $proc) unless ($apple); + # Make sure we $prefix isn't empty here (for the $apple case). + # We handle mangling the label here, make sure it doesn't match + # the label handling below (if $prefix would be empty). + $prefix = $prefix."; "; + push(@proc_stack, $proc); + s/^[A-Za-z_\.]\w+/$symprefix$&:/; + } + $prefix = $prefix."\t.thumb_func; " if ($thumb); + s/\bPROC\b/@ $&/; + $_ = $prefix.$_; + } + s/^(\s*)(S|Q|SH|U|UQ|UH)ASX\b/$1$2ADDSUBX/; + s/^(\s*)(S|Q|SH|U|UQ|UH)SAX\b/$1$2SUBADDX/; + if (/\bENDP\b/) + { + my $proc; + s/\bENDP\b/@ $&/; + $proc = pop(@proc_stack); + $_ = "\t.size $proc, .-$proc".$_ if ($proc && !$apple); + } + s/\bSUBT\b/@ $&/; + s/\bDATA\b/@ $&/; # DATA directive is deprecated -- Asm guide, p.7-25 + s/\bKEEP\b/@ $&/; + s/\bEXPORTAS\b/@ $&/; + s/\|\|(.)+\bEQU\b/@ $&/; + s/\|\|([\w\$]+)\|\|/$1/; + s/\bENTRY\b/@ $&/; + s/\bASSERT\b/@ $&/; + s/\bGBLL\b/@ $&/; + s/\bGBLA\b/@ $&/; + s/^\W+OPT\b/@ $&/; + s/:OR:/|/g; + s/:SHL:/<</g; + s/:SHR:/>>/g; + s/:AND:/&/g; + s/:LAND:/&&/g; + s/CPSR/cpsr/; + s/SPSR/spsr/; + s/ALIGN$/.balign 4/; + s/ALIGN\s+([0-9x]+)$/.balign $1/; + s/psr_cxsf/psr_all/; + s/LTORG/.ltorg/; + s/^([A-Za-z_]\w*)[ \t]+EQU/ .set $1,/; + s/^([A-Za-z_]\w*)[ \t]+SETL/ .set $1,/; + s/^([A-Za-z_]\w*)[ \t]+SETA/ .set $1,/; + s/^([A-Za-z_]\w*)[ \t]+\*/ .set $1,/; + + # {PC} + 0xdeadfeed --> . + 0xdeadfeed + s/\{PC\} \+/ \. +/; + + # Single hex constant on the line ! + # + # >>> NOTE <<< + # Double-precision floats in gcc are always mixed-endian, which means + # bytes in two words are little-endian, but words are big-endian. + # So, 0x0000deadfeed0000 would be stored as 0x0000dead at low address + # and 0xfeed0000 at high address. + # + s/\bDCFD\b[ \t]+0x([a-fA-F0-9]{8})([a-fA-F0-9]{8})/.long 0x$1, 0x$2/; + # Only decimal constants on the line, no hex ! + s/\bDCFD\b[ \t]+([0-9\.\-]+)/.double $1/; + + # Single hex constant on the line ! +# s/\bDCFS\b[ \t]+0x([a-f0-9]{8})([a-f0-9]{8})/.long 0x$1, 0x$2/; + # Only decimal constants on the line, no hex ! +# s/\bDCFS\b[ \t]+([0-9\.\-]+)/.double $1/; + s/\bDCFS[ \t]+0x/.word 0x/; + s/\bDCFS\b/.float/; + + s/^([A-Za-z_]\w*)[ \t]+DCD/$1 .word/; + s/\bDCD\b/.word/; + s/^([A-Za-z_]\w*)[ \t]+DCW/$1 .short/; + s/\bDCW\b/.short/; + s/^([A-Za-z_]\w*)[ \t]+DCB/$1 .byte/; + s/\bDCB\b/.byte/; + s/^([A-Za-z_]\w*)[ \t]+\%/.comm $1,/; + s/^[A-Za-z_\.]\w+/$&:/; + s/^(\d+)/$1:/; + s/\%(\d+)/$1b_or_f/; + s/\%[Bb](\d+)/$1b/; + s/\%[Ff](\d+)/$1f/; + s/\%[Ff][Tt](\d+)/$1f/; + s/&([\dA-Fa-f]+)/0x$1/; + if ( /\b2_[01]+\b/ ) { + s/\b2_([01]+)\b/conv$1&&&&/g; + while ( /[01][01][01][01]&&&&/ ) { + s/0000&&&&/&&&&0/g; + s/0001&&&&/&&&&1/g; + s/0010&&&&/&&&&2/g; + s/0011&&&&/&&&&3/g; + s/0100&&&&/&&&&4/g; + s/0101&&&&/&&&&5/g; + s/0110&&&&/&&&&6/g; + s/0111&&&&/&&&&7/g; + s/1000&&&&/&&&&8/g; + s/1001&&&&/&&&&9/g; + s/1010&&&&/&&&&A/g; + s/1011&&&&/&&&&B/g; + s/1100&&&&/&&&&C/g; + s/1101&&&&/&&&&D/g; + s/1110&&&&/&&&&E/g; + s/1111&&&&/&&&&F/g; + } + s/000&&&&/&&&&0/g; + s/001&&&&/&&&&1/g; + s/010&&&&/&&&&2/g; + s/011&&&&/&&&&3/g; + s/100&&&&/&&&&4/g; + s/101&&&&/&&&&5/g; + s/110&&&&/&&&&6/g; + s/111&&&&/&&&&7/g; + s/00&&&&/&&&&0/g; + s/01&&&&/&&&&1/g; + s/10&&&&/&&&&2/g; + s/11&&&&/&&&&3/g; + s/0&&&&/&&&&0/g; + s/1&&&&/&&&&1/g; + s/conv&&&&/0x/g; + } + + if ( /commandline/) + { + if( /-bigend/) + { + $bigend=1; + } + } + + if ( /\bDCDU\b/ ) + { + my $cmd=$_; + my $value; + my $prefix; + my $w1; + my $w2; + my $w3; + my $w4; + + s/\s+DCDU\b/@ $&/; + + $cmd =~ /\bDCDU\b\s+0x(\d+)/; + $value = $1; + $value =~ /(\w\w)(\w\w)(\w\w)(\w\w)/; + $w1 = $1; + $w2 = $2; + $w3 = $3; + $w4 = $4; + + if( $bigend ne "") + { + # big endian + $prefix = "\t.byte\t0x".$w1.";". + "\t.byte\t0x".$w2.";". + "\t.byte\t0x".$w3.";". + "\t.byte\t0x".$w4."; "; + } + else + { + # little endian + $prefix = "\t.byte\t0x".$w4.";". + "\t.byte\t0x".$w3.";". + "\t.byte\t0x".$w2.";". + "\t.byte\t0x".$w1."; "; + } + $_=$prefix.$_; + } + + if ( /\badrl\b/i ) + { + s/\badrl\s+(\w+)\s*,\s*(\w+)/ldr $1,=$2/i; + $addPadding = 1; + } + s/\bEND\b/@ END/; +} continue { + printf ("%s", $_) if $printit; + if ($addPadding != 0) + { + printf (" mov r0,r0\n"); + $addPadding = 0; + } +} +#If we had a code section, mark that this object doesn't need an executable +# stack. +if ($nxstack && !$apple) { + printf (" .section\t.note.GNU-stack,\"\",\%\%progbits\n"); +} diff --git a/libopus/celt/arm/arm_celt_map.c b/libopus/celt/arm/arm_celt_map.c new file mode 100644 index 0000000..ca988b6 --- /dev/null +++ b/libopus/celt/arm/arm_celt_map.c @@ -0,0 +1,160 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "pitch.h" +#include "kiss_fft.h" +#include "mdct.h" + +#if defined(OPUS_HAVE_RTCD) + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR) +opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, const opus_val16 *y, int N) = { + celt_inner_prod_c, /* ARMv4 */ + celt_inner_prod_c, /* EDSP */ + celt_inner_prod_c, /* Media */ + celt_inner_prod_neon /* NEON */ +}; + +void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) = { + dual_inner_prod_c, /* ARMv4 */ + dual_inner_prod_c, /* EDSP */ + dual_inner_prod_c, /* Media */ + dual_inner_prod_neon /* NEON */ +}; +# endif + +# if defined(FIXED_POINT) +# if ((defined(OPUS_ARM_MAY_HAVE_NEON) && !defined(OPUS_ARM_PRESUME_NEON)) || \ + (defined(OPUS_ARM_MAY_HAVE_MEDIA) && !defined(OPUS_ARM_PRESUME_MEDIA)) || \ + (defined(OPUS_ARM_MAY_HAVE_EDSP) && !defined(OPUS_ARM_PRESUME_EDSP))) +opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, + const opus_val16 *, opus_val32 *, int, int, int) = { + celt_pitch_xcorr_c, /* ARMv4 */ + MAY_HAVE_EDSP(celt_pitch_xcorr), /* EDSP */ + MAY_HAVE_MEDIA(celt_pitch_xcorr), /* Media */ + MAY_HAVE_NEON(celt_pitch_xcorr) /* NEON */ +}; + +# endif +# else /* !FIXED_POINT */ +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR) +void (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, + const opus_val16 *, opus_val32 *, int, int, int) = { + celt_pitch_xcorr_c, /* ARMv4 */ + celt_pitch_xcorr_c, /* EDSP */ + celt_pitch_xcorr_c, /* Media */ + celt_pitch_xcorr_float_neon /* Neon */ +}; +# endif +# endif /* FIXED_POINT */ + +#if defined(FIXED_POINT) && defined(OPUS_HAVE_RTCD) && \ + defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR) + +void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len +) = { + xcorr_kernel_c, /* ARMv4 */ + xcorr_kernel_c, /* EDSP */ + xcorr_kernel_c, /* Media */ + xcorr_kernel_neon_fixed, /* Neon */ +}; + +#endif + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +# if defined(HAVE_ARM_NE10) +# if defined(CUSTOM_MODES) +int (*const OPUS_FFT_ALLOC_ARCH_IMPL[OPUS_ARCHMASK+1])(kiss_fft_state *st) = { + opus_fft_alloc_arch_c, /* ARMv4 */ + opus_fft_alloc_arch_c, /* EDSP */ + opus_fft_alloc_arch_c, /* Media */ + opus_fft_alloc_arm_neon /* Neon with NE10 library support */ +}; + +void (*const OPUS_FFT_FREE_ARCH_IMPL[OPUS_ARCHMASK+1])(kiss_fft_state *st) = { + opus_fft_free_arch_c, /* ARMv4 */ + opus_fft_free_arch_c, /* EDSP */ + opus_fft_free_arch_c, /* Media */ + opus_fft_free_arm_neon /* Neon with NE10 */ +}; +# endif /* CUSTOM_MODES */ + +void (*const OPUS_FFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout) = { + opus_fft_c, /* ARMv4 */ + opus_fft_c, /* EDSP */ + opus_fft_c, /* Media */ + opus_fft_neon /* Neon with NE10 */ +}; + +void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout) = { + opus_ifft_c, /* ARMv4 */ + opus_ifft_c, /* EDSP */ + opus_ifft_c, /* Media */ + opus_ifft_neon /* Neon with NE10 */ +}; + +void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l, + kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, + int overlap, int shift, + int stride, int arch) = { + clt_mdct_forward_c, /* ARMv4 */ + clt_mdct_forward_c, /* EDSP */ + clt_mdct_forward_c, /* Media */ + clt_mdct_forward_neon /* Neon with NE10 */ +}; + +void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l, + kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, + int overlap, int shift, + int stride, int arch) = { + clt_mdct_backward_c, /* ARMv4 */ + clt_mdct_backward_c, /* EDSP */ + clt_mdct_backward_c, /* Media */ + clt_mdct_backward_neon /* Neon with NE10 */ +}; + +# endif /* HAVE_ARM_NE10 */ +# endif /* OPUS_ARM_MAY_HAVE_NEON_INTR */ + +#endif /* OPUS_HAVE_RTCD */ diff --git a/libopus/celt/arm/armcpu.c b/libopus/celt/arm/armcpu.c new file mode 100644 index 0000000..694a63b --- /dev/null +++ b/libopus/celt/arm/armcpu.c @@ -0,0 +1,185 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* Original code from libtheora modified to suit to Opus */ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifdef OPUS_HAVE_RTCD + +#include "armcpu.h" +#include "cpu_support.h" +#include "os_support.h" +#include "opus_types.h" +#include "arch.h" + +#define OPUS_CPU_ARM_V4_FLAG (1<<OPUS_ARCH_ARM_V4) +#define OPUS_CPU_ARM_EDSP_FLAG (1<<OPUS_ARCH_ARM_EDSP) +#define OPUS_CPU_ARM_MEDIA_FLAG (1<<OPUS_ARCH_ARM_MEDIA) +#define OPUS_CPU_ARM_NEON_FLAG (1<<OPUS_ARCH_ARM_NEON) + +#if defined(_MSC_VER) +/*For GetExceptionCode() and EXCEPTION_ILLEGAL_INSTRUCTION.*/ +# define WIN32_LEAN_AND_MEAN +# define WIN32_EXTRA_LEAN +# include <windows.h> + +static OPUS_INLINE opus_uint32 opus_cpu_capabilities(void){ + opus_uint32 flags; + flags=0; + /* MSVC has no OPUS_INLINE __asm support for ARM, but it does let you __emit + * instructions via their assembled hex code. + * All of these instructions should be essentially nops. */ +# if defined(OPUS_ARM_MAY_HAVE_EDSP) || defined(OPUS_ARM_MAY_HAVE_MEDIA) \ + || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + __try{ + /*PLD [r13]*/ + __emit(0xF5DDF000); + flags|=OPUS_CPU_ARM_EDSP_FLAG; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) \ + || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + __try{ + /*SHADD8 r3,r3,r3*/ + __emit(0xE6333F93); + flags|=OPUS_CPU_ARM_MEDIA_FLAG; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# if defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + __try{ + /*VORR q0,q0,q0*/ + __emit(0xF2200150); + flags|=OPUS_CPU_ARM_NEON_FLAG; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# endif +# endif +# endif + return flags; +} + +#elif defined(__linux__) +/* Linux based */ +opus_uint32 opus_cpu_capabilities(void) +{ + opus_uint32 flags = 0; + FILE *cpuinfo; + + /* Reading /proc/self/auxv would be easier, but that doesn't work reliably on + * Android */ + cpuinfo = fopen("/proc/cpuinfo", "r"); + + if(cpuinfo != NULL) + { + /* 512 should be enough for anybody (it's even enough for all the flags that + * x86 has accumulated... so far). */ + char buf[512]; + + while(fgets(buf, 512, cpuinfo) != NULL) + { +# if defined(OPUS_ARM_MAY_HAVE_EDSP) || defined(OPUS_ARM_MAY_HAVE_MEDIA) \ + || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + /* Search for edsp and neon flag */ + if(memcmp(buf, "Features", 8) == 0) + { + char *p; + p = strstr(buf, " edsp"); + if(p != NULL && (p[5] == ' ' || p[5] == '\n')) + flags |= OPUS_CPU_ARM_EDSP_FLAG; + +# if defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + p = strstr(buf, " neon"); + if(p != NULL && (p[5] == ' ' || p[5] == '\n')) + flags |= OPUS_CPU_ARM_NEON_FLAG; +# endif + } +# endif + +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) \ + || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + /* Search for media capabilities (>= ARMv6) */ + if(memcmp(buf, "CPU architecture:", 17) == 0) + { + int version; + version = atoi(buf+17); + + if(version >= 6) + flags |= OPUS_CPU_ARM_MEDIA_FLAG; + } +# endif + } + + fclose(cpuinfo); + } + return flags; +} +#else +/* The feature registers which can tell us what the processor supports are + * accessible in priveleged modes only, so we can't have a general user-space + * detection method like on x86.*/ +# error "Configured to use ARM asm but no CPU detection method available for " \ + "your platform. Reconfigure with --disable-rtcd (or send patches)." +#endif + +int opus_select_arch(void) +{ + opus_uint32 flags = opus_cpu_capabilities(); + int arch = 0; + + if(!(flags & OPUS_CPU_ARM_EDSP_FLAG)) { + /* Asserts ensure arch values are sequential */ + celt_assert(arch == OPUS_ARCH_ARM_V4); + return arch; + } + arch++; + + if(!(flags & OPUS_CPU_ARM_MEDIA_FLAG)) { + celt_assert(arch == OPUS_ARCH_ARM_EDSP); + return arch; + } + arch++; + + if(!(flags & OPUS_CPU_ARM_NEON_FLAG)) { + celt_assert(arch == OPUS_ARCH_ARM_MEDIA); + return arch; + } + arch++; + + celt_assert(arch == OPUS_ARCH_ARM_NEON); + return arch; +} + +#endif diff --git a/libopus/celt/arm/armcpu.h b/libopus/celt/arm/armcpu.h new file mode 100644 index 0000000..820262f --- /dev/null +++ b/libopus/celt/arm/armcpu.h @@ -0,0 +1,77 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(ARMCPU_H) +# define ARMCPU_H + +# if defined(OPUS_ARM_MAY_HAVE_EDSP) +# define MAY_HAVE_EDSP(name) name ## _edsp +# else +# define MAY_HAVE_EDSP(name) name ## _c +# endif + +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) +# define MAY_HAVE_MEDIA(name) name ## _media +# else +# define MAY_HAVE_MEDIA(name) MAY_HAVE_EDSP(name) +# endif + +# if defined(OPUS_ARM_MAY_HAVE_NEON) +# define MAY_HAVE_NEON(name) name ## _neon +# else +# define MAY_HAVE_NEON(name) MAY_HAVE_MEDIA(name) +# endif + +# if defined(OPUS_ARM_PRESUME_EDSP) +# define PRESUME_EDSP(name) name ## _edsp +# else +# define PRESUME_EDSP(name) name ## _c +# endif + +# if defined(OPUS_ARM_PRESUME_MEDIA) +# define PRESUME_MEDIA(name) name ## _media +# else +# define PRESUME_MEDIA(name) PRESUME_EDSP(name) +# endif + +# if defined(OPUS_ARM_PRESUME_NEON) +# define PRESUME_NEON(name) name ## _neon +# else +# define PRESUME_NEON(name) PRESUME_MEDIA(name) +# endif + +# if defined(OPUS_HAVE_RTCD) +int opus_select_arch(void); + +#define OPUS_ARCH_ARM_V4 (0) +#define OPUS_ARCH_ARM_EDSP (1) +#define OPUS_ARCH_ARM_MEDIA (2) +#define OPUS_ARCH_ARM_NEON (3) + +# endif + +#endif diff --git a/libopus/celt/arm/armopts.s.in b/libopus/celt/arm/armopts.s.in new file mode 100644 index 0000000..3d8aaf2 --- /dev/null +++ b/libopus/celt/arm/armopts.s.in @@ -0,0 +1,37 @@ +/* Copyright (C) 2013 Mozilla Corporation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +; Set the following to 1 if we have EDSP instructions +; (LDRD/STRD, etc., ARMv5E and later). +OPUS_ARM_MAY_HAVE_EDSP * @OPUS_ARM_MAY_HAVE_EDSP@ + +; Set the following to 1 if we have ARMv6 media instructions. +OPUS_ARM_MAY_HAVE_MEDIA * @OPUS_ARM_MAY_HAVE_MEDIA@ + +; Set the following to 1 if we have NEON (some ARMv7) +OPUS_ARM_MAY_HAVE_NEON * @OPUS_ARM_MAY_HAVE_NEON@ + +END diff --git a/libopus/celt/arm/celt_fft_ne10.c b/libopus/celt/arm/celt_fft_ne10.c new file mode 100644 index 0000000..ea5fd78 --- /dev/null +++ b/libopus/celt/arm/celt_fft_ne10.c @@ -0,0 +1,173 @@ +/* Copyright (c) 2015 Xiph.Org Foundation + Written by Viswanath Puttagunta */ +/** + @file celt_fft_ne10.c + @brief ARM Neon optimizations for fft using NE10 library + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef SKIP_CONFIG_H +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#endif + +#include <NE10_dsp.h> +#include "os_support.h" +#include "kiss_fft.h" +#include "stack_alloc.h" + +#if !defined(FIXED_POINT) +# define NE10_FFT_ALLOC_C2C_TYPE_NEON ne10_fft_alloc_c2c_float32_neon +# define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_float32_t +# define NE10_FFT_STATE_TYPE_T ne10_fft_state_float32_t +# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_float32 +# define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_float32_t +# define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_float32_neon +#else +# define NE10_FFT_ALLOC_C2C_TYPE_NEON(nfft) ne10_fft_alloc_c2c_int32_neon(nfft) +# define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_int32_t +# define NE10_FFT_STATE_TYPE_T ne10_fft_state_int32_t +# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32 +# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32 +# define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_int32_t +# define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_int32_neon +#endif + +#if defined(CUSTOM_MODES) + +/* nfft lengths in NE10 that support scaled fft */ +# define NE10_FFTSCALED_SUPPORT_MAX 4 +static const int ne10_fft_scaled_support[NE10_FFTSCALED_SUPPORT_MAX] = { + 480, 240, 120, 60 +}; + +int opus_fft_alloc_arm_neon(kiss_fft_state *st) +{ + int i; + size_t memneeded = sizeof(struct arch_fft_state); + + st->arch_fft = (arch_fft_state *)opus_alloc(memneeded); + if (!st->arch_fft) + return -1; + + for (i = 0; i < NE10_FFTSCALED_SUPPORT_MAX; i++) { + if(st->nfft == ne10_fft_scaled_support[i]) + break; + } + if (i == NE10_FFTSCALED_SUPPORT_MAX) { + /* This nfft length (scaled fft) is not supported in NE10 */ + st->arch_fft->is_supported = 0; + st->arch_fft->priv = NULL; + } + else { + st->arch_fft->is_supported = 1; + st->arch_fft->priv = (void *)NE10_FFT_ALLOC_C2C_TYPE_NEON(st->nfft); + if (st->arch_fft->priv == NULL) { + return -1; + } + } + return 0; +} + +void opus_fft_free_arm_neon(kiss_fft_state *st) +{ + NE10_FFT_CFG_TYPE_T cfg; + + if (!st->arch_fft) + return; + + cfg = (NE10_FFT_CFG_TYPE_T)st->arch_fft->priv; + if (cfg) + NE10_FFT_DESTROY_C2C_TYPE(cfg); + opus_free(st->arch_fft); +} +#endif + +void opus_fft_neon(const kiss_fft_state *st, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout) +{ + NE10_FFT_STATE_TYPE_T state; + NE10_FFT_CFG_TYPE_T cfg = &state; + VARDECL(NE10_FFT_CPX_TYPE_T, buffer); + SAVE_STACK; + ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T); + + if (!st->arch_fft->is_supported) { + /* This nfft length (scaled fft) not supported in NE10 */ + opus_fft_c(st, fin, fout); + } + else { + memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T)); + state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0]; +#if !defined(FIXED_POINT) + state.is_forward_scaled = 1; + + NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, + (NE10_FFT_CPX_TYPE_T *)fin, + cfg, 0); +#else + NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, + (NE10_FFT_CPX_TYPE_T *)fin, + cfg, 0, 1); +#endif + } + RESTORE_STACK; +} + +void opus_ifft_neon(const kiss_fft_state *st, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout) +{ + NE10_FFT_STATE_TYPE_T state; + NE10_FFT_CFG_TYPE_T cfg = &state; + VARDECL(NE10_FFT_CPX_TYPE_T, buffer); + SAVE_STACK; + ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T); + + if (!st->arch_fft->is_supported) { + /* This nfft length (scaled fft) not supported in NE10 */ + opus_ifft_c(st, fin, fout); + } + else { + memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T)); + state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0]; +#if !defined(FIXED_POINT) + state.is_backward_scaled = 0; + + NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, + (NE10_FFT_CPX_TYPE_T *)fin, + cfg, 1); +#else + NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, + (NE10_FFT_CPX_TYPE_T *)fin, + cfg, 1, 0); +#endif + } + RESTORE_STACK; +} diff --git a/libopus/celt/arm/celt_mdct_ne10.c b/libopus/celt/arm/celt_mdct_ne10.c new file mode 100644 index 0000000..3531d02 --- /dev/null +++ b/libopus/celt/arm/celt_mdct_ne10.c @@ -0,0 +1,258 @@ +/* Copyright (c) 2015 Xiph.Org Foundation + Written by Viswanath Puttagunta */ +/** + @file celt_mdct_ne10.c + @brief ARM Neon optimizations for mdct using NE10 library + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef SKIP_CONFIG_H +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#endif + +#include "kiss_fft.h" +#include "_kiss_fft_guts.h" +#include "mdct.h" +#include "stack_alloc.h" + +void clt_mdct_forward_neon(const mdct_lookup *l, + kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, + int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + VARDECL(kiss_fft_scalar, f); + VARDECL(kiss_fft_cpx, f2); + const kiss_fft_state *st = l->kfft[shift]; + const kiss_twiddle_scalar *trig; + + SAVE_STACK; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + ALLOC(f, N2, kiss_fft_scalar); + ALLOC(f2, N4, kiss_fft_cpx); + + /* Consider the input to be composed of four blocks: [a, b, c, d] */ + /* Window, shuffle, fold */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1); + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1); + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1); + const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1; + for(i=0;i<((overlap+3)>>2);i++) + { + /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/ + *yp++ = MULT16_32_Q15(*wp2, xp1[N2]) + MULT16_32_Q15(*wp1,*xp2); + *yp++ = MULT16_32_Q15(*wp1, *xp1) - MULT16_32_Q15(*wp2, xp2[-N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + wp1 = window; + wp2 = window+overlap-1; + for(;i<N4-((overlap+3)>>2);i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = *xp2; + *yp++ = *xp1; + xp1+=2; + xp2-=2; + } + for(;i<N4;i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = -MULT16_32_Q15(*wp1, xp1[-N2]) + MULT16_32_Q15(*wp2, *xp2); + *yp++ = MULT16_32_Q15(*wp2, *xp1) + MULT16_32_Q15(*wp1, xp2[N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + } + /* Pre-rotation */ + { + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const kiss_twiddle_scalar *t = &trig[0]; + for(i=0;i<N4;i++) + { + kiss_fft_cpx yc; + kiss_twiddle_scalar t0, t1; + kiss_fft_scalar re, im, yr, yi; + t0 = t[i]; + t1 = t[N4+i]; + re = *yp++; + im = *yp++; + yr = S_MUL(re,t0) - S_MUL(im,t1); + yi = S_MUL(im,t0) + S_MUL(re,t1); + yc.r = yr; + yc.i = yi; + f2[i] = yc; + } + } + + opus_fft(st, f2, (kiss_fft_cpx *)f, arch); + + /* Post-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_cpx * OPUS_RESTRICT fp = (kiss_fft_cpx *)f; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1); + const kiss_twiddle_scalar *t = &trig[0]; + /* Temp pointers to make it really clear to the compiler what we're doing */ + for(i=0;i<N4;i++) + { + kiss_fft_scalar yr, yi; + yr = S_MUL(fp->i,t[N4+i]) - S_MUL(fp->r,t[i]); + yi = S_MUL(fp->r,t[N4+i]) + S_MUL(fp->i,t[i]); + *yp1 = yr; + *yp2 = yi; + fp++; + yp1 += 2*stride; + yp2 -= 2*stride; + } + } + RESTORE_STACK; +} + +void clt_mdct_backward_neon(const mdct_lookup *l, + kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 * OPUS_RESTRICT window, + int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + VARDECL(kiss_fft_scalar, f); + const kiss_twiddle_scalar *trig; + const kiss_fft_state *st = l->kfft[shift]; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + ALLOC(f, N2, kiss_fft_scalar); + + /* Pre-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in; + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1); + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0]; + for(i=0;i<N4;i++) + { + kiss_fft_scalar yr, yi; + yr = S_MUL(*xp2, t[i]) + S_MUL(*xp1, t[N4+i]); + yi = S_MUL(*xp1, t[i]) - S_MUL(*xp2, t[N4+i]); + yp[2*i] = yr; + yp[2*i+1] = yi; + xp1+=2*stride; + xp2-=2*stride; + } + } + + opus_ifft(st, (kiss_fft_cpx *)f, (kiss_fft_cpx*)(out+(overlap>>1)), arch); + + /* Post-rotate and de-shuffle from both ends of the buffer at once to make + it in-place. */ + { + kiss_fft_scalar * yp0 = out+(overlap>>1); + kiss_fft_scalar * yp1 = out+(overlap>>1)+N2-2; + const kiss_twiddle_scalar *t = &trig[0]; + /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the + middle pair will be computed twice. */ + for(i=0;i<(N4+1)>>1;i++) + { + kiss_fft_scalar re, im, yr, yi; + kiss_twiddle_scalar t0, t1; + re = yp0[0]; + im = yp0[1]; + t0 = t[i]; + t1 = t[N4+i]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = S_MUL(re,t0) + S_MUL(im,t1); + yi = S_MUL(re,t1) - S_MUL(im,t0); + re = yp1[0]; + im = yp1[1]; + yp0[0] = yr; + yp1[1] = yi; + + t0 = t[(N4-i-1)]; + t1 = t[(N2-i-1)]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = S_MUL(re,t0) + S_MUL(im,t1); + yi = S_MUL(re,t1) - S_MUL(im,t0); + yp1[0] = yr; + yp0[1] = yi; + yp0 += 2; + yp1 -= 2; + } + } + + /* Mirror on both sides for TDAC */ + { + kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + const opus_val16 * OPUS_RESTRICT wp1 = window; + const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1; + + for(i = 0; i < overlap/2; i++) + { + kiss_fft_scalar x1, x2; + x1 = *xp1; + x2 = *yp1; + *yp1++ = MULT16_32_Q15(*wp2, x2) - MULT16_32_Q15(*wp1, x1); + *xp1-- = MULT16_32_Q15(*wp1, x2) + MULT16_32_Q15(*wp2, x1); + wp1++; + wp2--; + } + } + RESTORE_STACK; +} diff --git a/libopus/celt/arm/celt_neon_intr.c b/libopus/celt/arm/celt_neon_intr.c new file mode 100644 index 0000000..effda76 --- /dev/null +++ b/libopus/celt/arm/celt_neon_intr.c @@ -0,0 +1,211 @@ +/* Copyright (c) 2014-2015 Xiph.Org Foundation + Written by Viswanath Puttagunta */ +/** + @file celt_neon_intr.c + @brief ARM Neon Intrinsic optimizations for celt + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <arm_neon.h> +#include "../pitch.h" + +#if defined(FIXED_POINT) +void xcorr_kernel_neon_fixed(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len) +{ + int j; + int32x4_t a = vld1q_s32(sum); + /* Load y[0...3] */ + /* This requires len>0 to always be valid (which we assert in the C code). */ + int16x4_t y0 = vld1_s16(y); + y += 4; + + for (j = 0; j + 8 <= len; j += 8) + { + /* Load x[0...7] */ + int16x8_t xx = vld1q_s16(x); + int16x4_t x0 = vget_low_s16(xx); + int16x4_t x4 = vget_high_s16(xx); + /* Load y[4...11] */ + int16x8_t yy = vld1q_s16(y); + int16x4_t y4 = vget_low_s16(yy); + int16x4_t y8 = vget_high_s16(yy); + int32x4_t a0 = vmlal_lane_s16(a, y0, x0, 0); + int32x4_t a1 = vmlal_lane_s16(a0, y4, x4, 0); + + int16x4_t y1 = vext_s16(y0, y4, 1); + int16x4_t y5 = vext_s16(y4, y8, 1); + int32x4_t a2 = vmlal_lane_s16(a1, y1, x0, 1); + int32x4_t a3 = vmlal_lane_s16(a2, y5, x4, 1); + + int16x4_t y2 = vext_s16(y0, y4, 2); + int16x4_t y6 = vext_s16(y4, y8, 2); + int32x4_t a4 = vmlal_lane_s16(a3, y2, x0, 2); + int32x4_t a5 = vmlal_lane_s16(a4, y6, x4, 2); + + int16x4_t y3 = vext_s16(y0, y4, 3); + int16x4_t y7 = vext_s16(y4, y8, 3); + int32x4_t a6 = vmlal_lane_s16(a5, y3, x0, 3); + int32x4_t a7 = vmlal_lane_s16(a6, y7, x4, 3); + + y0 = y8; + a = a7; + x += 8; + y += 8; + } + + for (; j < len; j++) + { + int16x4_t x0 = vld1_dup_s16(x); /* load next x */ + int32x4_t a0 = vmlal_s16(a, y0, x0); + + int16x4_t y4 = vld1_dup_s16(y); /* load next y */ + y0 = vext_s16(y0, y4, 1); + a = a0; + x++; + y++; + } + + vst1q_s32(sum, a); +} + +#else +/* + * Function: xcorr_kernel_neon_float + * --------------------------------- + * Computes 4 correlation values and stores them in sum[4] + */ +static void xcorr_kernel_neon_float(const float32_t *x, const float32_t *y, + float32_t sum[4], int len) { + float32x4_t YY[3]; + float32x4_t YEXT[3]; + float32x4_t XX[2]; + float32x2_t XX_2; + float32x4_t SUMM; + const float32_t *xi = x; + const float32_t *yi = y; + + celt_assert(len>0); + + YY[0] = vld1q_f32(yi); + SUMM = vdupq_n_f32(0); + + /* Consume 8 elements in x vector and 12 elements in y + * vector. However, the 12'th element never really gets + * touched in this loop. So, if len == 8, then we only + * must access y[0] to y[10]. y[11] must not be accessed + * hence make sure len > 8 and not len >= 8 + */ + while (len > 8) { + yi += 4; + YY[1] = vld1q_f32(yi); + yi += 4; + YY[2] = vld1q_f32(yi); + + XX[0] = vld1q_f32(xi); + xi += 4; + XX[1] = vld1q_f32(xi); + xi += 4; + + SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0); + YEXT[0] = vextq_f32(YY[0], YY[1], 1); + SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1); + YEXT[1] = vextq_f32(YY[0], YY[1], 2); + SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0); + YEXT[2] = vextq_f32(YY[0], YY[1], 3); + SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1); + + SUMM = vmlaq_lane_f32(SUMM, YY[1], vget_low_f32(XX[1]), 0); + YEXT[0] = vextq_f32(YY[1], YY[2], 1); + SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[1]), 1); + YEXT[1] = vextq_f32(YY[1], YY[2], 2); + SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[1]), 0); + YEXT[2] = vextq_f32(YY[1], YY[2], 3); + SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[1]), 1); + + YY[0] = YY[2]; + len -= 8; + } + + /* Consume 4 elements in x vector and 8 elements in y + * vector. However, the 8'th element in y never really gets + * touched in this loop. So, if len == 4, then we only + * must access y[0] to y[6]. y[7] must not be accessed + * hence make sure len>4 and not len>=4 + */ + if (len > 4) { + yi += 4; + YY[1] = vld1q_f32(yi); + + XX[0] = vld1q_f32(xi); + xi += 4; + + SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0); + YEXT[0] = vextq_f32(YY[0], YY[1], 1); + SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1); + YEXT[1] = vextq_f32(YY[0], YY[1], 2); + SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0); + YEXT[2] = vextq_f32(YY[0], YY[1], 3); + SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1); + + YY[0] = YY[1]; + len -= 4; + } + + while (--len > 0) { + XX_2 = vld1_dup_f32(xi++); + SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0); + YY[0]= vld1q_f32(++yi); + } + + XX_2 = vld1_dup_f32(xi); + SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0); + + vst1q_f32(sum, SUMM); +} + +void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch) { + int i; + (void)arch; + celt_assert(max_pitch > 0); + celt_sig_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0); + + for (i = 0; i < (max_pitch-3); i += 4) { + xcorr_kernel_neon_float((const float32_t *)_x, (const float32_t *)_y+i, + (float32_t *)xcorr+i, len); + } + + /* In case max_pitch isn't a multiple of 4, do non-unrolled version. */ + for (; i < max_pitch; i++) { + xcorr[i] = celt_inner_prod_neon(_x, _y+i, len); + } +} +#endif diff --git a/libopus/celt/arm/celt_pitch_xcorr_arm-gnu.S b/libopus/celt/arm/celt_pitch_xcorr_arm-gnu.S new file mode 100644 index 0000000..10668e5 --- /dev/null +++ b/libopus/celt/arm/celt_pitch_xcorr_arm-gnu.S @@ -0,0 +1,555 @@ + .syntax unified +@ Copyright (c) 2007-2008 CSIRO +@ Copyright (c) 2007-2009 Xiph.Org Foundation +@ Copyright (c) 2013 Parrot +@ Written by Aurélien Zanelli +@ +@ Redistribution and use in source and binary forms, with or without +@ modification, are permitted provided that the following conditions +@ are met: +@ +@ - Redistributions of source code must retain the above copyright +@ notice, this list of conditions and the following disclaimer. +@ +@ - Redistributions in binary form must reproduce the above copyright +@ notice, this list of conditions and the following disclaimer in the +@ documentation and/or other materials provided with the distribution. +@ +@ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +@ ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +@ LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +@ A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER +@ OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +@ EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +@ PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +@ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +@ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +@ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + + .text; .p2align 2; .arch armv7-a + .fpu neon + .object_arch armv4t + + .include "celt/arm/armopts-gnu.S" + + .if OPUS_ARM_MAY_HAVE_EDSP + .global celt_pitch_xcorr_edsp + .endif + + .if OPUS_ARM_MAY_HAVE_NEON + .global celt_pitch_xcorr_neon + .endif + + .if OPUS_ARM_MAY_HAVE_NEON + +@ Compute sum[k]=sum(x[j]*y[j+k],j=0...len-1), k=0...3 + .type xcorr_kernel_neon, %function; xcorr_kernel_neon: @ PROC +xcorr_kernel_neon_start: + @ input: + @ r3 = int len + @ r4 = opus_val16 *x + @ r5 = opus_val16 *y + @ q0 = opus_val32 sum[4] + @ output: + @ q0 = opus_val32 sum[4] + @ preserved: r0-r3, r6-r11, d2, q4-q7, q9-q15 + @ internal usage: + @ r12 = int j + @ d3 = y_3|y_2|y_1|y_0 + @ q2 = y_B|y_A|y_9|y_8|y_7|y_6|y_5|y_4 + @ q3 = x_7|x_6|x_5|x_4|x_3|x_2|x_1|x_0 + @ q8 = scratch + @ + @ Load y[0...3] + @ This requires len>0 to always be valid (which we assert in the C code). + VLD1.16 {d5}, [r5]! + SUBS r12, r3, #8 + BLE xcorr_kernel_neon_process4 +@ Process 8 samples at a time. +@ This loop loads one y value more than we actually need. Therefore we have to +@ stop as soon as there are 8 or fewer samples left (instead of 7), to avoid +@ reading past the end of the array. +xcorr_kernel_neon_process8: + @ This loop has 19 total instructions (10 cycles to issue, minimum), with + @ - 2 cycles of ARM insrtuctions, + @ - 10 cycles of load/store/byte permute instructions, and + @ - 9 cycles of data processing instructions. + @ On a Cortex A8, we dual-issue the maximum amount (9 cycles) between the + @ latter two categories, meaning the whole loop should run in 10 cycles per + @ iteration, barring cache misses. + @ + @ Load x[0...7] + VLD1.16 {d6, d7}, [r4]! + @ Unlike VMOV, VAND is a data processsing instruction (and doesn't get + @ assembled to VMOV, like VORR would), so it dual-issues with the prior VLD1. + VAND d3, d5, d5 + SUBS r12, r12, #8 + @ Load y[4...11] + VLD1.16 {d4, d5}, [r5]! + VMLAL.S16 q0, d3, d6[0] + VEXT.16 d16, d3, d4, #1 + VMLAL.S16 q0, d4, d7[0] + VEXT.16 d17, d4, d5, #1 + VMLAL.S16 q0, d16, d6[1] + VEXT.16 d16, d3, d4, #2 + VMLAL.S16 q0, d17, d7[1] + VEXT.16 d17, d4, d5, #2 + VMLAL.S16 q0, d16, d6[2] + VEXT.16 d16, d3, d4, #3 + VMLAL.S16 q0, d17, d7[2] + VEXT.16 d17, d4, d5, #3 + VMLAL.S16 q0, d16, d6[3] + VMLAL.S16 q0, d17, d7[3] + BGT xcorr_kernel_neon_process8 +@ Process 4 samples here if we have > 4 left (still reading one extra y value). +xcorr_kernel_neon_process4: + ADDS r12, r12, #4 + BLE xcorr_kernel_neon_process2 + @ Load x[0...3] + VLD1.16 d6, [r4]! + @ Use VAND since it's a data processing instruction again. + VAND d4, d5, d5 + SUB r12, r12, #4 + @ Load y[4...7] + VLD1.16 d5, [r5]! + VMLAL.S16 q0, d4, d6[0] + VEXT.16 d16, d4, d5, #1 + VMLAL.S16 q0, d16, d6[1] + VEXT.16 d16, d4, d5, #2 + VMLAL.S16 q0, d16, d6[2] + VEXT.16 d16, d4, d5, #3 + VMLAL.S16 q0, d16, d6[3] +@ Process 2 samples here if we have > 2 left (still reading one extra y value). +xcorr_kernel_neon_process2: + ADDS r12, r12, #2 + BLE xcorr_kernel_neon_process1 + @ Load x[0...1] + VLD2.16 {d6[],d7[]}, [r4]! + @ Use VAND since it's a data processing instruction again. + VAND d4, d5, d5 + SUB r12, r12, #2 + @ Load y[4...5] + VLD1.32 {d5[]}, [r5]! + VMLAL.S16 q0, d4, d6 + VEXT.16 d16, d4, d5, #1 + @ Replace bottom copy of {y5,y4} in d5 with {y3,y2} from d4, using VSRI + @ instead of VEXT, since it's a data-processing instruction. + VSRI.64 d5, d4, #32 + VMLAL.S16 q0, d16, d7 +@ Process 1 sample using the extra y value we loaded above. +xcorr_kernel_neon_process1: + @ Load next *x + VLD1.16 {d6[]}, [r4]! + ADDS r12, r12, #1 + @ y[0...3] are left in d5 from prior iteration(s) (if any) + VMLAL.S16 q0, d5, d6 + MOVLE pc, lr +@ Now process 1 last sample, not reading ahead. + @ Load last *y + VLD1.16 {d4[]}, [r5]! + VSRI.64 d4, d5, #16 + @ Load last *x + VLD1.16 {d6[]}, [r4]! + VMLAL.S16 q0, d4, d6 + MOV pc, lr + .size xcorr_kernel_neon, .-xcorr_kernel_neon @ ENDP + +@ opus_val32 celt_pitch_xcorr_neon(opus_val16 *_x, opus_val16 *_y, +@ opus_val32 *xcorr, int len, int max_pitch, int arch) + .type celt_pitch_xcorr_neon, %function; celt_pitch_xcorr_neon: @ PROC + @ input: + @ r0 = opus_val16 *_x + @ r1 = opus_val16 *_y + @ r2 = opus_val32 *xcorr + @ r3 = int len + @ output: + @ r0 = int maxcorr + @ internal usage: + @ r4 = opus_val16 *x (for xcorr_kernel_neon()) + @ r5 = opus_val16 *y (for xcorr_kernel_neon()) + @ r6 = int max_pitch + @ r12 = int j + @ q15 = int maxcorr[4] (q15 is not used by xcorr_kernel_neon()) + @ ignored: + @ int arch + STMFD sp!, {r4-r6, lr} + LDR r6, [sp, #16] + VMOV.S32 q15, #1 + @ if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done + SUBS r6, r6, #4 + BLT celt_pitch_xcorr_neon_process4_done +celt_pitch_xcorr_neon_process4: + @ xcorr_kernel_neon parameters: + @ r3 = len, r4 = _x, r5 = _y, q0 = {0, 0, 0, 0} + MOV r4, r0 + MOV r5, r1 + VEOR q0, q0, q0 + @ xcorr_kernel_neon only modifies r4, r5, r12, and q0...q3. + @ So we don't save/restore any other registers. + BL xcorr_kernel_neon_start + SUBS r6, r6, #4 + VST1.32 {q0}, [r2]! + @ _y += 4 + ADD r1, r1, #8 + VMAX.S32 q15, q15, q0 + @ if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done + BGE celt_pitch_xcorr_neon_process4 +@ We have less than 4 sums left to compute. +celt_pitch_xcorr_neon_process4_done: + ADDS r6, r6, #4 + @ Reduce maxcorr to a single value + VMAX.S32 d30, d30, d31 + VPMAX.S32 d30, d30, d30 + @ if (max_pitch <= 0) goto celt_pitch_xcorr_neon_done + BLE celt_pitch_xcorr_neon_done +@ Now compute each remaining sum one at a time. +celt_pitch_xcorr_neon_process_remaining: + MOV r4, r0 + MOV r5, r1 + VMOV.I32 q0, #0 + SUBS r12, r3, #8 + BLT celt_pitch_xcorr_neon_process_remaining4 +@ Sum terms 8 at a time. +celt_pitch_xcorr_neon_process_remaining_loop8: + @ Load x[0...7] + VLD1.16 {q1}, [r4]! + @ Load y[0...7] + VLD1.16 {q2}, [r5]! + SUBS r12, r12, #8 + VMLAL.S16 q0, d4, d2 + VMLAL.S16 q0, d5, d3 + BGE celt_pitch_xcorr_neon_process_remaining_loop8 +@ Sum terms 4 at a time. +celt_pitch_xcorr_neon_process_remaining4: + ADDS r12, r12, #4 + BLT celt_pitch_xcorr_neon_process_remaining4_done + @ Load x[0...3] + VLD1.16 {d2}, [r4]! + @ Load y[0...3] + VLD1.16 {d3}, [r5]! + SUB r12, r12, #4 + VMLAL.S16 q0, d3, d2 +celt_pitch_xcorr_neon_process_remaining4_done: + @ Reduce the sum to a single value. + VADD.S32 d0, d0, d1 + VPADDL.S32 d0, d0 + ADDS r12, r12, #4 + BLE celt_pitch_xcorr_neon_process_remaining_loop_done +@ Sum terms 1 at a time. +celt_pitch_xcorr_neon_process_remaining_loop1: + VLD1.16 {d2[]}, [r4]! + VLD1.16 {d3[]}, [r5]! + SUBS r12, r12, #1 + VMLAL.S16 q0, d2, d3 + BGT celt_pitch_xcorr_neon_process_remaining_loop1 +celt_pitch_xcorr_neon_process_remaining_loop_done: + VST1.32 {d0[0]}, [r2]! + VMAX.S32 d30, d30, d0 + SUBS r6, r6, #1 + @ _y++ + ADD r1, r1, #2 + @ if (--max_pitch > 0) goto celt_pitch_xcorr_neon_process_remaining + BGT celt_pitch_xcorr_neon_process_remaining +celt_pitch_xcorr_neon_done: + VMOV.32 r0, d30[0] + LDMFD sp!, {r4-r6, pc} + .size celt_pitch_xcorr_neon, .-celt_pitch_xcorr_neon @ ENDP + + .endif + + .if OPUS_ARM_MAY_HAVE_EDSP + +@ This will get used on ARMv7 devices without NEON, so it has been optimized +@ to take advantage of dual-issuing where possible. + .type xcorr_kernel_edsp, %function; xcorr_kernel_edsp: @ PROC +xcorr_kernel_edsp_start: + @ input: + @ r3 = int len + @ r4 = opus_val16 *_x (must be 32-bit aligned) + @ r5 = opus_val16 *_y (must be 32-bit aligned) + @ r6...r9 = opus_val32 sum[4] + @ output: + @ r6...r9 = opus_val32 sum[4] + @ preserved: r0-r5 + @ internal usage + @ r2 = int j + @ r12,r14 = opus_val16 x[4] + @ r10,r11 = opus_val16 y[4] + STMFD sp!, {r2,r4,r5,lr} + LDR r10, [r5], #4 @ Load y[0...1] + SUBS r2, r3, #4 @ j = len-4 + LDR r11, [r5], #4 @ Load y[2...3] + BLE xcorr_kernel_edsp_process4_done + LDR r12, [r4], #4 @ Load x[0...1] + @ Stall +xcorr_kernel_edsp_process4: + @ The multiplies must issue from pipeline 0, and can't dual-issue with each + @ other. Every other instruction here dual-issues with a multiply, and is + @ thus "free". There should be no stalls in the body of the loop. + SMLABB r6, r12, r10, r6 @ sum[0] = MAC16_16(sum[0],x_0,y_0) + LDR r14, [r4], #4 @ Load x[2...3] + SMLABT r7, r12, r10, r7 @ sum[1] = MAC16_16(sum[1],x_0,y_1) + SUBS r2, r2, #4 @ j-=4 + SMLABB r8, r12, r11, r8 @ sum[2] = MAC16_16(sum[2],x_0,y_2) + SMLABT r9, r12, r11, r9 @ sum[3] = MAC16_16(sum[3],x_0,y_3) + SMLATT r6, r12, r10, r6 @ sum[0] = MAC16_16(sum[0],x_1,y_1) + LDR r10, [r5], #4 @ Load y[4...5] + SMLATB r7, r12, r11, r7 @ sum[1] = MAC16_16(sum[1],x_1,y_2) + SMLATT r8, r12, r11, r8 @ sum[2] = MAC16_16(sum[2],x_1,y_3) + SMLATB r9, r12, r10, r9 @ sum[3] = MAC16_16(sum[3],x_1,y_4) + LDRGT r12, [r4], #4 @ Load x[0...1] + SMLABB r6, r14, r11, r6 @ sum[0] = MAC16_16(sum[0],x_2,y_2) + SMLABT r7, r14, r11, r7 @ sum[1] = MAC16_16(sum[1],x_2,y_3) + SMLABB r8, r14, r10, r8 @ sum[2] = MAC16_16(sum[2],x_2,y_4) + SMLABT r9, r14, r10, r9 @ sum[3] = MAC16_16(sum[3],x_2,y_5) + SMLATT r6, r14, r11, r6 @ sum[0] = MAC16_16(sum[0],x_3,y_3) + LDR r11, [r5], #4 @ Load y[6...7] + SMLATB r7, r14, r10, r7 @ sum[1] = MAC16_16(sum[1],x_3,y_4) + SMLATT r8, r14, r10, r8 @ sum[2] = MAC16_16(sum[2],x_3,y_5) + SMLATB r9, r14, r11, r9 @ sum[3] = MAC16_16(sum[3],x_3,y_6) + BGT xcorr_kernel_edsp_process4 +xcorr_kernel_edsp_process4_done: + ADDS r2, r2, #4 + BLE xcorr_kernel_edsp_done + LDRH r12, [r4], #2 @ r12 = *x++ + SUBS r2, r2, #1 @ j-- + @ Stall + SMLABB r6, r12, r10, r6 @ sum[0] = MAC16_16(sum[0],x,y_0) + LDRHGT r14, [r4], #2 @ r14 = *x++ + SMLABT r7, r12, r10, r7 @ sum[1] = MAC16_16(sum[1],x,y_1) + SMLABB r8, r12, r11, r8 @ sum[2] = MAC16_16(sum[2],x,y_2) + SMLABT r9, r12, r11, r9 @ sum[3] = MAC16_16(sum[3],x,y_3) + BLE xcorr_kernel_edsp_done + SMLABT r6, r14, r10, r6 @ sum[0] = MAC16_16(sum[0],x,y_1) + SUBS r2, r2, #1 @ j-- + SMLABB r7, r14, r11, r7 @ sum[1] = MAC16_16(sum[1],x,y_2) + LDRH r10, [r5], #2 @ r10 = y_4 = *y++ + SMLABT r8, r14, r11, r8 @ sum[2] = MAC16_16(sum[2],x,y_3) + LDRHGT r12, [r4], #2 @ r12 = *x++ + SMLABB r9, r14, r10, r9 @ sum[3] = MAC16_16(sum[3],x,y_4) + BLE xcorr_kernel_edsp_done + SMLABB r6, r12, r11, r6 @ sum[0] = MAC16_16(sum[0],tmp,y_2) + CMP r2, #1 @ j-- + SMLABT r7, r12, r11, r7 @ sum[1] = MAC16_16(sum[1],tmp,y_3) + LDRH r2, [r5], #2 @ r2 = y_5 = *y++ + SMLABB r8, r12, r10, r8 @ sum[2] = MAC16_16(sum[2],tmp,y_4) + LDRHGT r14, [r4] @ r14 = *x + SMLABB r9, r12, r2, r9 @ sum[3] = MAC16_16(sum[3],tmp,y_5) + BLE xcorr_kernel_edsp_done + SMLABT r6, r14, r11, r6 @ sum[0] = MAC16_16(sum[0],tmp,y_3) + LDRH r11, [r5] @ r11 = y_6 = *y + SMLABB r7, r14, r10, r7 @ sum[1] = MAC16_16(sum[1],tmp,y_4) + SMLABB r8, r14, r2, r8 @ sum[2] = MAC16_16(sum[2],tmp,y_5) + SMLABB r9, r14, r11, r9 @ sum[3] = MAC16_16(sum[3],tmp,y_6) +xcorr_kernel_edsp_done: + LDMFD sp!, {r2,r4,r5,pc} + .size xcorr_kernel_edsp, .-xcorr_kernel_edsp @ ENDP + + .type celt_pitch_xcorr_edsp, %function; celt_pitch_xcorr_edsp: @ PROC + @ input: + @ r0 = opus_val16 *_x (must be 32-bit aligned) + @ r1 = opus_val16 *_y (only needs to be 16-bit aligned) + @ r2 = opus_val32 *xcorr + @ r3 = int len + @ output: + @ r0 = maxcorr + @ internal usage + @ r4 = opus_val16 *x + @ r5 = opus_val16 *y + @ r6 = opus_val32 sum0 + @ r7 = opus_val32 sum1 + @ r8 = opus_val32 sum2 + @ r9 = opus_val32 sum3 + @ r1 = int max_pitch + @ r12 = int j + @ ignored: + @ int arch + STMFD sp!, {r4-r11, lr} + MOV r5, r1 + LDR r1, [sp, #36] + MOV r4, r0 + TST r5, #3 + @ maxcorr = 1 + MOV r0, #1 + BEQ celt_pitch_xcorr_edsp_process1u_done +@ Compute one sum at the start to make y 32-bit aligned. + SUBS r12, r3, #4 + @ r14 = sum = 0 + MOV r14, #0 + LDRH r8, [r5], #2 + BLE celt_pitch_xcorr_edsp_process1u_loop4_done + LDR r6, [r4], #4 + MOV r8, r8, LSL #16 +celt_pitch_xcorr_edsp_process1u_loop4: + LDR r9, [r5], #4 + SMLABT r14, r6, r8, r14 @ sum = MAC16_16(sum, x_0, y_0) + LDR r7, [r4], #4 + SMLATB r14, r6, r9, r14 @ sum = MAC16_16(sum, x_1, y_1) + LDR r8, [r5], #4 + SMLABT r14, r7, r9, r14 @ sum = MAC16_16(sum, x_2, y_2) + SUBS r12, r12, #4 @ j-=4 + SMLATB r14, r7, r8, r14 @ sum = MAC16_16(sum, x_3, y_3) + LDRGT r6, [r4], #4 + BGT celt_pitch_xcorr_edsp_process1u_loop4 + MOV r8, r8, LSR #16 +celt_pitch_xcorr_edsp_process1u_loop4_done: + ADDS r12, r12, #4 +celt_pitch_xcorr_edsp_process1u_loop1: + LDRHGE r6, [r4], #2 + @ Stall + SMLABBGE r14, r6, r8, r14 @ sum = MAC16_16(sum, *x, *y) + SUBSGE r12, r12, #1 + LDRHGT r8, [r5], #2 + BGT celt_pitch_xcorr_edsp_process1u_loop1 + @ Restore _x + SUB r4, r4, r3, LSL #1 + @ Restore and advance _y + SUB r5, r5, r3, LSL #1 + @ maxcorr = max(maxcorr, sum) + CMP r0, r14 + ADD r5, r5, #2 + MOVLT r0, r14 + SUBS r1, r1, #1 + @ xcorr[i] = sum + STR r14, [r2], #4 + BLE celt_pitch_xcorr_edsp_done +celt_pitch_xcorr_edsp_process1u_done: + @ if (max_pitch < 4) goto celt_pitch_xcorr_edsp_process2 + SUBS r1, r1, #4 + BLT celt_pitch_xcorr_edsp_process2 +celt_pitch_xcorr_edsp_process4: + @ xcorr_kernel_edsp parameters: + @ r3 = len, r4 = _x, r5 = _y, r6...r9 = sum[4] = {0, 0, 0, 0} + MOV r6, #0 + MOV r7, #0 + MOV r8, #0 + MOV r9, #0 + BL xcorr_kernel_edsp_start @ xcorr_kernel_edsp(_x, _y+i, xcorr+i, len) + @ maxcorr = max(maxcorr, sum0, sum1, sum2, sum3) + CMP r0, r6 + @ _y+=4 + ADD r5, r5, #8 + MOVLT r0, r6 + CMP r0, r7 + MOVLT r0, r7 + CMP r0, r8 + MOVLT r0, r8 + CMP r0, r9 + MOVLT r0, r9 + STMIA r2!, {r6-r9} + SUBS r1, r1, #4 + BGE celt_pitch_xcorr_edsp_process4 +celt_pitch_xcorr_edsp_process2: + ADDS r1, r1, #2 + BLT celt_pitch_xcorr_edsp_process1a + SUBS r12, r3, #4 + @ {r10, r11} = {sum0, sum1} = {0, 0} + MOV r10, #0 + MOV r11, #0 + LDR r8, [r5], #4 + BLE celt_pitch_xcorr_edsp_process2_loop_done + LDR r6, [r4], #4 + LDR r9, [r5], #4 +celt_pitch_xcorr_edsp_process2_loop4: + SMLABB r10, r6, r8, r10 @ sum0 = MAC16_16(sum0, x_0, y_0) + LDR r7, [r4], #4 + SMLABT r11, r6, r8, r11 @ sum1 = MAC16_16(sum1, x_0, y_1) + SUBS r12, r12, #4 @ j-=4 + SMLATT r10, r6, r8, r10 @ sum0 = MAC16_16(sum0, x_1, y_1) + LDR r8, [r5], #4 + SMLATB r11, r6, r9, r11 @ sum1 = MAC16_16(sum1, x_1, y_2) + LDRGT r6, [r4], #4 + SMLABB r10, r7, r9, r10 @ sum0 = MAC16_16(sum0, x_2, y_2) + SMLABT r11, r7, r9, r11 @ sum1 = MAC16_16(sum1, x_2, y_3) + SMLATT r10, r7, r9, r10 @ sum0 = MAC16_16(sum0, x_3, y_3) + LDRGT r9, [r5], #4 + SMLATB r11, r7, r8, r11 @ sum1 = MAC16_16(sum1, x_3, y_4) + BGT celt_pitch_xcorr_edsp_process2_loop4 +celt_pitch_xcorr_edsp_process2_loop_done: + ADDS r12, r12, #2 + BLE celt_pitch_xcorr_edsp_process2_1 + LDR r6, [r4], #4 + @ Stall + SMLABB r10, r6, r8, r10 @ sum0 = MAC16_16(sum0, x_0, y_0) + LDR r9, [r5], #4 + SMLABT r11, r6, r8, r11 @ sum1 = MAC16_16(sum1, x_0, y_1) + SUB r12, r12, #2 + SMLATT r10, r6, r8, r10 @ sum0 = MAC16_16(sum0, x_1, y_1) + MOV r8, r9 + SMLATB r11, r6, r9, r11 @ sum1 = MAC16_16(sum1, x_1, y_2) +celt_pitch_xcorr_edsp_process2_1: + LDRH r6, [r4], #2 + ADDS r12, r12, #1 + @ Stall + SMLABB r10, r6, r8, r10 @ sum0 = MAC16_16(sum0, x_0, y_0) + LDRHGT r7, [r4], #2 + SMLABT r11, r6, r8, r11 @ sum1 = MAC16_16(sum1, x_0, y_1) + BLE celt_pitch_xcorr_edsp_process2_done + LDRH r9, [r5], #2 + SMLABT r10, r7, r8, r10 @ sum0 = MAC16_16(sum0, x_0, y_1) + SMLABB r11, r7, r9, r11 @ sum1 = MAC16_16(sum1, x_0, y_2) +celt_pitch_xcorr_edsp_process2_done: + @ Restore _x + SUB r4, r4, r3, LSL #1 + @ Restore and advance _y + SUB r5, r5, r3, LSL #1 + @ maxcorr = max(maxcorr, sum0) + CMP r0, r10 + ADD r5, r5, #2 + MOVLT r0, r10 + SUB r1, r1, #2 + @ maxcorr = max(maxcorr, sum1) + CMP r0, r11 + @ xcorr[i] = sum + STR r10, [r2], #4 + MOVLT r0, r11 + STR r11, [r2], #4 +celt_pitch_xcorr_edsp_process1a: + ADDS r1, r1, #1 + BLT celt_pitch_xcorr_edsp_done + SUBS r12, r3, #4 + @ r14 = sum = 0 + MOV r14, #0 + BLT celt_pitch_xcorr_edsp_process1a_loop_done + LDR r6, [r4], #4 + LDR r8, [r5], #4 + LDR r7, [r4], #4 + LDR r9, [r5], #4 +celt_pitch_xcorr_edsp_process1a_loop4: + SMLABB r14, r6, r8, r14 @ sum = MAC16_16(sum, x_0, y_0) + SUBS r12, r12, #4 @ j-=4 + SMLATT r14, r6, r8, r14 @ sum = MAC16_16(sum, x_1, y_1) + LDRGE r6, [r4], #4 + SMLABB r14, r7, r9, r14 @ sum = MAC16_16(sum, x_2, y_2) + LDRGE r8, [r5], #4 + SMLATT r14, r7, r9, r14 @ sum = MAC16_16(sum, x_3, y_3) + LDRGE r7, [r4], #4 + LDRGE r9, [r5], #4 + BGE celt_pitch_xcorr_edsp_process1a_loop4 +celt_pitch_xcorr_edsp_process1a_loop_done: + ADDS r12, r12, #2 + LDRGE r6, [r4], #4 + LDRGE r8, [r5], #4 + @ Stall + SMLABBGE r14, r6, r8, r14 @ sum = MAC16_16(sum, x_0, y_0) + SUBGE r12, r12, #2 + SMLATTGE r14, r6, r8, r14 @ sum = MAC16_16(sum, x_1, y_1) + ADDS r12, r12, #1 + LDRHGE r6, [r4], #2 + LDRHGE r8, [r5], #2 + @ Stall + SMLABBGE r14, r6, r8, r14 @ sum = MAC16_16(sum, *x, *y) + @ maxcorr = max(maxcorr, sum) + CMP r0, r14 + @ xcorr[i] = sum + STR r14, [r2], #4 + MOVLT r0, r14 +celt_pitch_xcorr_edsp_done: + LDMFD sp!, {r4-r11, pc} + .size celt_pitch_xcorr_edsp, .-celt_pitch_xcorr_edsp @ ENDP + + .endif + +@ END: + .section .note.GNU-stack,"",%progbits diff --git a/libopus/celt/arm/celt_pitch_xcorr_arm.s b/libopus/celt/arm/celt_pitch_xcorr_arm.s new file mode 100644 index 0000000..6e873af --- /dev/null +++ b/libopus/celt/arm/celt_pitch_xcorr_arm.s @@ -0,0 +1,551 @@ +; Copyright (c) 2007-2008 CSIRO +; Copyright (c) 2007-2009 Xiph.Org Foundation +; Copyright (c) 2013 Parrot +; Written by Aurélien Zanelli +; +; Redistribution and use in source and binary forms, with or without +; modification, are permitted provided that the following conditions +; are met: +; +; - Redistributions of source code must retain the above copyright +; notice, this list of conditions and the following disclaimer. +; +; - Redistributions in binary form must reproduce the above copyright +; notice, this list of conditions and the following disclaimer in the +; documentation and/or other materials provided with the distribution. +; +; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +; ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +; LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +; A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER +; OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +; EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +; PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +; PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +; LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +; NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +; SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + + AREA |.text|, CODE, READONLY + + GET celt/arm/armopts.s + +IF OPUS_ARM_MAY_HAVE_EDSP + EXPORT celt_pitch_xcorr_edsp +ENDIF + +IF OPUS_ARM_MAY_HAVE_NEON + EXPORT celt_pitch_xcorr_neon +ENDIF + +IF OPUS_ARM_MAY_HAVE_NEON + +; Compute sum[k]=sum(x[j]*y[j+k],j=0...len-1), k=0...3 +xcorr_kernel_neon PROC +xcorr_kernel_neon_start + ; input: + ; r3 = int len + ; r4 = opus_val16 *x + ; r5 = opus_val16 *y + ; q0 = opus_val32 sum[4] + ; output: + ; q0 = opus_val32 sum[4] + ; preserved: r0-r3, r6-r11, d2, q4-q7, q9-q15 + ; internal usage: + ; r12 = int j + ; d3 = y_3|y_2|y_1|y_0 + ; q2 = y_B|y_A|y_9|y_8|y_7|y_6|y_5|y_4 + ; q3 = x_7|x_6|x_5|x_4|x_3|x_2|x_1|x_0 + ; q8 = scratch + ; + ; Load y[0...3] + ; This requires len>0 to always be valid (which we assert in the C code). + VLD1.16 {d5}, [r5]! + SUBS r12, r3, #8 + BLE xcorr_kernel_neon_process4 +; Process 8 samples at a time. +; This loop loads one y value more than we actually need. Therefore we have to +; stop as soon as there are 8 or fewer samples left (instead of 7), to avoid +; reading past the end of the array. +xcorr_kernel_neon_process8 + ; This loop has 19 total instructions (10 cycles to issue, minimum), with + ; - 2 cycles of ARM insrtuctions, + ; - 10 cycles of load/store/byte permute instructions, and + ; - 9 cycles of data processing instructions. + ; On a Cortex A8, we dual-issue the maximum amount (9 cycles) between the + ; latter two categories, meaning the whole loop should run in 10 cycles per + ; iteration, barring cache misses. + ; + ; Load x[0...7] + VLD1.16 {d6, d7}, [r4]! + ; Unlike VMOV, VAND is a data processsing instruction (and doesn't get + ; assembled to VMOV, like VORR would), so it dual-issues with the prior VLD1. + VAND d3, d5, d5 + SUBS r12, r12, #8 + ; Load y[4...11] + VLD1.16 {d4, d5}, [r5]! + VMLAL.S16 q0, d3, d6[0] + VEXT.16 d16, d3, d4, #1 + VMLAL.S16 q0, d4, d7[0] + VEXT.16 d17, d4, d5, #1 + VMLAL.S16 q0, d16, d6[1] + VEXT.16 d16, d3, d4, #2 + VMLAL.S16 q0, d17, d7[1] + VEXT.16 d17, d4, d5, #2 + VMLAL.S16 q0, d16, d6[2] + VEXT.16 d16, d3, d4, #3 + VMLAL.S16 q0, d17, d7[2] + VEXT.16 d17, d4, d5, #3 + VMLAL.S16 q0, d16, d6[3] + VMLAL.S16 q0, d17, d7[3] + BGT xcorr_kernel_neon_process8 +; Process 4 samples here if we have > 4 left (still reading one extra y value). +xcorr_kernel_neon_process4 + ADDS r12, r12, #4 + BLE xcorr_kernel_neon_process2 + ; Load x[0...3] + VLD1.16 d6, [r4]! + ; Use VAND since it's a data processing instruction again. + VAND d4, d5, d5 + SUB r12, r12, #4 + ; Load y[4...7] + VLD1.16 d5, [r5]! + VMLAL.S16 q0, d4, d6[0] + VEXT.16 d16, d4, d5, #1 + VMLAL.S16 q0, d16, d6[1] + VEXT.16 d16, d4, d5, #2 + VMLAL.S16 q0, d16, d6[2] + VEXT.16 d16, d4, d5, #3 + VMLAL.S16 q0, d16, d6[3] +; Process 2 samples here if we have > 2 left (still reading one extra y value). +xcorr_kernel_neon_process2 + ADDS r12, r12, #2 + BLE xcorr_kernel_neon_process1 + ; Load x[0...1] + VLD2.16 {d6[],d7[]}, [r4]! + ; Use VAND since it's a data processing instruction again. + VAND d4, d5, d5 + SUB r12, r12, #2 + ; Load y[4...5] + VLD1.32 {d5[]}, [r5]! + VMLAL.S16 q0, d4, d6 + VEXT.16 d16, d4, d5, #1 + ; Replace bottom copy of {y5,y4} in d5 with {y3,y2} from d4, using VSRI + ; instead of VEXT, since it's a data-processing instruction. + VSRI.64 d5, d4, #32 + VMLAL.S16 q0, d16, d7 +; Process 1 sample using the extra y value we loaded above. +xcorr_kernel_neon_process1 + ; Load next *x + VLD1.16 {d6[]}, [r4]! + ADDS r12, r12, #1 + ; y[0...3] are left in d5 from prior iteration(s) (if any) + VMLAL.S16 q0, d5, d6 + MOVLE pc, lr +; Now process 1 last sample, not reading ahead. + ; Load last *y + VLD1.16 {d4[]}, [r5]! + VSRI.64 d4, d5, #16 + ; Load last *x + VLD1.16 {d6[]}, [r4]! + VMLAL.S16 q0, d4, d6 + MOV pc, lr + ENDP + +; opus_val32 celt_pitch_xcorr_neon(opus_val16 *_x, opus_val16 *_y, +; opus_val32 *xcorr, int len, int max_pitch, int arch) +celt_pitch_xcorr_neon PROC + ; input: + ; r0 = opus_val16 *_x + ; r1 = opus_val16 *_y + ; r2 = opus_val32 *xcorr + ; r3 = int len + ; output: + ; r0 = int maxcorr + ; internal usage: + ; r4 = opus_val16 *x (for xcorr_kernel_neon()) + ; r5 = opus_val16 *y (for xcorr_kernel_neon()) + ; r6 = int max_pitch + ; r12 = int j + ; q15 = int maxcorr[4] (q15 is not used by xcorr_kernel_neon()) + ; ignored: + ; int arch + STMFD sp!, {r4-r6, lr} + LDR r6, [sp, #16] + VMOV.S32 q15, #1 + ; if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done + SUBS r6, r6, #4 + BLT celt_pitch_xcorr_neon_process4_done +celt_pitch_xcorr_neon_process4 + ; xcorr_kernel_neon parameters: + ; r3 = len, r4 = _x, r5 = _y, q0 = {0, 0, 0, 0} + MOV r4, r0 + MOV r5, r1 + VEOR q0, q0, q0 + ; xcorr_kernel_neon only modifies r4, r5, r12, and q0...q3. + ; So we don't save/restore any other registers. + BL xcorr_kernel_neon_start + SUBS r6, r6, #4 + VST1.32 {q0}, [r2]! + ; _y += 4 + ADD r1, r1, #8 + VMAX.S32 q15, q15, q0 + ; if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done + BGE celt_pitch_xcorr_neon_process4 +; We have less than 4 sums left to compute. +celt_pitch_xcorr_neon_process4_done + ADDS r6, r6, #4 + ; Reduce maxcorr to a single value + VMAX.S32 d30, d30, d31 + VPMAX.S32 d30, d30, d30 + ; if (max_pitch <= 0) goto celt_pitch_xcorr_neon_done + BLE celt_pitch_xcorr_neon_done +; Now compute each remaining sum one at a time. +celt_pitch_xcorr_neon_process_remaining + MOV r4, r0 + MOV r5, r1 + VMOV.I32 q0, #0 + SUBS r12, r3, #8 + BLT celt_pitch_xcorr_neon_process_remaining4 +; Sum terms 8 at a time. +celt_pitch_xcorr_neon_process_remaining_loop8 + ; Load x[0...7] + VLD1.16 {q1}, [r4]! + ; Load y[0...7] + VLD1.16 {q2}, [r5]! + SUBS r12, r12, #8 + VMLAL.S16 q0, d4, d2 + VMLAL.S16 q0, d5, d3 + BGE celt_pitch_xcorr_neon_process_remaining_loop8 +; Sum terms 4 at a time. +celt_pitch_xcorr_neon_process_remaining4 + ADDS r12, r12, #4 + BLT celt_pitch_xcorr_neon_process_remaining4_done + ; Load x[0...3] + VLD1.16 {d2}, [r4]! + ; Load y[0...3] + VLD1.16 {d3}, [r5]! + SUB r12, r12, #4 + VMLAL.S16 q0, d3, d2 +celt_pitch_xcorr_neon_process_remaining4_done + ; Reduce the sum to a single value. + VADD.S32 d0, d0, d1 + VPADDL.S32 d0, d0 + ADDS r12, r12, #4 + BLE celt_pitch_xcorr_neon_process_remaining_loop_done +; Sum terms 1 at a time. +celt_pitch_xcorr_neon_process_remaining_loop1 + VLD1.16 {d2[]}, [r4]! + VLD1.16 {d3[]}, [r5]! + SUBS r12, r12, #1 + VMLAL.S16 q0, d2, d3 + BGT celt_pitch_xcorr_neon_process_remaining_loop1 +celt_pitch_xcorr_neon_process_remaining_loop_done + VST1.32 {d0[0]}, [r2]! + VMAX.S32 d30, d30, d0 + SUBS r6, r6, #1 + ; _y++ + ADD r1, r1, #2 + ; if (--max_pitch > 0) goto celt_pitch_xcorr_neon_process_remaining + BGT celt_pitch_xcorr_neon_process_remaining +celt_pitch_xcorr_neon_done + VMOV.32 r0, d30[0] + LDMFD sp!, {r4-r6, pc} + ENDP + +ENDIF + +IF OPUS_ARM_MAY_HAVE_EDSP + +; This will get used on ARMv7 devices without NEON, so it has been optimized +; to take advantage of dual-issuing where possible. +xcorr_kernel_edsp PROC +xcorr_kernel_edsp_start + ; input: + ; r3 = int len + ; r4 = opus_val16 *_x (must be 32-bit aligned) + ; r5 = opus_val16 *_y (must be 32-bit aligned) + ; r6...r9 = opus_val32 sum[4] + ; output: + ; r6...r9 = opus_val32 sum[4] + ; preserved: r0-r5 + ; internal usage + ; r2 = int j + ; r12,r14 = opus_val16 x[4] + ; r10,r11 = opus_val16 y[4] + STMFD sp!, {r2,r4,r5,lr} + LDR r10, [r5], #4 ; Load y[0...1] + SUBS r2, r3, #4 ; j = len-4 + LDR r11, [r5], #4 ; Load y[2...3] + BLE xcorr_kernel_edsp_process4_done + LDR r12, [r4], #4 ; Load x[0...1] + ; Stall +xcorr_kernel_edsp_process4 + ; The multiplies must issue from pipeline 0, and can't dual-issue with each + ; other. Every other instruction here dual-issues with a multiply, and is + ; thus "free". There should be no stalls in the body of the loop. + SMLABB r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x_0,y_0) + LDR r14, [r4], #4 ; Load x[2...3] + SMLABT r7, r12, r10, r7 ; sum[1] = MAC16_16(sum[1],x_0,y_1) + SUBS r2, r2, #4 ; j-=4 + SMLABB r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x_0,y_2) + SMLABT r9, r12, r11, r9 ; sum[3] = MAC16_16(sum[3],x_0,y_3) + SMLATT r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x_1,y_1) + LDR r10, [r5], #4 ; Load y[4...5] + SMLATB r7, r12, r11, r7 ; sum[1] = MAC16_16(sum[1],x_1,y_2) + SMLATT r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x_1,y_3) + SMLATB r9, r12, r10, r9 ; sum[3] = MAC16_16(sum[3],x_1,y_4) + LDRGT r12, [r4], #4 ; Load x[0...1] + SMLABB r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],x_2,y_2) + SMLABT r7, r14, r11, r7 ; sum[1] = MAC16_16(sum[1],x_2,y_3) + SMLABB r8, r14, r10, r8 ; sum[2] = MAC16_16(sum[2],x_2,y_4) + SMLABT r9, r14, r10, r9 ; sum[3] = MAC16_16(sum[3],x_2,y_5) + SMLATT r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],x_3,y_3) + LDR r11, [r5], #4 ; Load y[6...7] + SMLATB r7, r14, r10, r7 ; sum[1] = MAC16_16(sum[1],x_3,y_4) + SMLATT r8, r14, r10, r8 ; sum[2] = MAC16_16(sum[2],x_3,y_5) + SMLATB r9, r14, r11, r9 ; sum[3] = MAC16_16(sum[3],x_3,y_6) + BGT xcorr_kernel_edsp_process4 +xcorr_kernel_edsp_process4_done + ADDS r2, r2, #4 + BLE xcorr_kernel_edsp_done + LDRH r12, [r4], #2 ; r12 = *x++ + SUBS r2, r2, #1 ; j-- + ; Stall + SMLABB r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x,y_0) + LDRHGT r14, [r4], #2 ; r14 = *x++ + SMLABT r7, r12, r10, r7 ; sum[1] = MAC16_16(sum[1],x,y_1) + SMLABB r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x,y_2) + SMLABT r9, r12, r11, r9 ; sum[3] = MAC16_16(sum[3],x,y_3) + BLE xcorr_kernel_edsp_done + SMLABT r6, r14, r10, r6 ; sum[0] = MAC16_16(sum[0],x,y_1) + SUBS r2, r2, #1 ; j-- + SMLABB r7, r14, r11, r7 ; sum[1] = MAC16_16(sum[1],x,y_2) + LDRH r10, [r5], #2 ; r10 = y_4 = *y++ + SMLABT r8, r14, r11, r8 ; sum[2] = MAC16_16(sum[2],x,y_3) + LDRHGT r12, [r4], #2 ; r12 = *x++ + SMLABB r9, r14, r10, r9 ; sum[3] = MAC16_16(sum[3],x,y_4) + BLE xcorr_kernel_edsp_done + SMLABB r6, r12, r11, r6 ; sum[0] = MAC16_16(sum[0],tmp,y_2) + CMP r2, #1 ; j-- + SMLABT r7, r12, r11, r7 ; sum[1] = MAC16_16(sum[1],tmp,y_3) + LDRH r2, [r5], #2 ; r2 = y_5 = *y++ + SMLABB r8, r12, r10, r8 ; sum[2] = MAC16_16(sum[2],tmp,y_4) + LDRHGT r14, [r4] ; r14 = *x + SMLABB r9, r12, r2, r9 ; sum[3] = MAC16_16(sum[3],tmp,y_5) + BLE xcorr_kernel_edsp_done + SMLABT r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],tmp,y_3) + LDRH r11, [r5] ; r11 = y_6 = *y + SMLABB r7, r14, r10, r7 ; sum[1] = MAC16_16(sum[1],tmp,y_4) + SMLABB r8, r14, r2, r8 ; sum[2] = MAC16_16(sum[2],tmp,y_5) + SMLABB r9, r14, r11, r9 ; sum[3] = MAC16_16(sum[3],tmp,y_6) +xcorr_kernel_edsp_done + LDMFD sp!, {r2,r4,r5,pc} + ENDP + +celt_pitch_xcorr_edsp PROC + ; input: + ; r0 = opus_val16 *_x (must be 32-bit aligned) + ; r1 = opus_val16 *_y (only needs to be 16-bit aligned) + ; r2 = opus_val32 *xcorr + ; r3 = int len + ; output: + ; r0 = maxcorr + ; internal usage + ; r4 = opus_val16 *x + ; r5 = opus_val16 *y + ; r6 = opus_val32 sum0 + ; r7 = opus_val32 sum1 + ; r8 = opus_val32 sum2 + ; r9 = opus_val32 sum3 + ; r1 = int max_pitch + ; r12 = int j + ; ignored: + ; int arch + STMFD sp!, {r4-r11, lr} + MOV r5, r1 + LDR r1, [sp, #36] + MOV r4, r0 + TST r5, #3 + ; maxcorr = 1 + MOV r0, #1 + BEQ celt_pitch_xcorr_edsp_process1u_done +; Compute one sum at the start to make y 32-bit aligned. + SUBS r12, r3, #4 + ; r14 = sum = 0 + MOV r14, #0 + LDRH r8, [r5], #2 + BLE celt_pitch_xcorr_edsp_process1u_loop4_done + LDR r6, [r4], #4 + MOV r8, r8, LSL #16 +celt_pitch_xcorr_edsp_process1u_loop4 + LDR r9, [r5], #4 + SMLABT r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + LDR r7, [r4], #4 + SMLATB r14, r6, r9, r14 ; sum = MAC16_16(sum, x_1, y_1) + LDR r8, [r5], #4 + SMLABT r14, r7, r9, r14 ; sum = MAC16_16(sum, x_2, y_2) + SUBS r12, r12, #4 ; j-=4 + SMLATB r14, r7, r8, r14 ; sum = MAC16_16(sum, x_3, y_3) + LDRGT r6, [r4], #4 + BGT celt_pitch_xcorr_edsp_process1u_loop4 + MOV r8, r8, LSR #16 +celt_pitch_xcorr_edsp_process1u_loop4_done + ADDS r12, r12, #4 +celt_pitch_xcorr_edsp_process1u_loop1 + LDRHGE r6, [r4], #2 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, *x, *y) + SUBSGE r12, r12, #1 + LDRHGT r8, [r5], #2 + BGT celt_pitch_xcorr_edsp_process1u_loop1 + ; Restore _x + SUB r4, r4, r3, LSL #1 + ; Restore and advance _y + SUB r5, r5, r3, LSL #1 + ; maxcorr = max(maxcorr, sum) + CMP r0, r14 + ADD r5, r5, #2 + MOVLT r0, r14 + SUBS r1, r1, #1 + ; xcorr[i] = sum + STR r14, [r2], #4 + BLE celt_pitch_xcorr_edsp_done +celt_pitch_xcorr_edsp_process1u_done + ; if (max_pitch < 4) goto celt_pitch_xcorr_edsp_process2 + SUBS r1, r1, #4 + BLT celt_pitch_xcorr_edsp_process2 +celt_pitch_xcorr_edsp_process4 + ; xcorr_kernel_edsp parameters: + ; r3 = len, r4 = _x, r5 = _y, r6...r9 = sum[4] = {0, 0, 0, 0} + MOV r6, #0 + MOV r7, #0 + MOV r8, #0 + MOV r9, #0 + BL xcorr_kernel_edsp_start ; xcorr_kernel_edsp(_x, _y+i, xcorr+i, len) + ; maxcorr = max(maxcorr, sum0, sum1, sum2, sum3) + CMP r0, r6 + ; _y+=4 + ADD r5, r5, #8 + MOVLT r0, r6 + CMP r0, r7 + MOVLT r0, r7 + CMP r0, r8 + MOVLT r0, r8 + CMP r0, r9 + MOVLT r0, r9 + STMIA r2!, {r6-r9} + SUBS r1, r1, #4 + BGE celt_pitch_xcorr_edsp_process4 +celt_pitch_xcorr_edsp_process2 + ADDS r1, r1, #2 + BLT celt_pitch_xcorr_edsp_process1a + SUBS r12, r3, #4 + ; {r10, r11} = {sum0, sum1} = {0, 0} + MOV r10, #0 + MOV r11, #0 + LDR r8, [r5], #4 + BLE celt_pitch_xcorr_edsp_process2_loop_done + LDR r6, [r4], #4 + LDR r9, [r5], #4 +celt_pitch_xcorr_edsp_process2_loop4 + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDR r7, [r4], #4 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + SUBS r12, r12, #4 ; j-=4 + SMLATT r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_1, y_1) + LDR r8, [r5], #4 + SMLATB r11, r6, r9, r11 ; sum1 = MAC16_16(sum1, x_1, y_2) + LDRGT r6, [r4], #4 + SMLABB r10, r7, r9, r10 ; sum0 = MAC16_16(sum0, x_2, y_2) + SMLABT r11, r7, r9, r11 ; sum1 = MAC16_16(sum1, x_2, y_3) + SMLATT r10, r7, r9, r10 ; sum0 = MAC16_16(sum0, x_3, y_3) + LDRGT r9, [r5], #4 + SMLATB r11, r7, r8, r11 ; sum1 = MAC16_16(sum1, x_3, y_4) + BGT celt_pitch_xcorr_edsp_process2_loop4 +celt_pitch_xcorr_edsp_process2_loop_done + ADDS r12, r12, #2 + BLE celt_pitch_xcorr_edsp_process2_1 + LDR r6, [r4], #4 + ; Stall + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDR r9, [r5], #4 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + SUB r12, r12, #2 + SMLATT r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_1, y_1) + MOV r8, r9 + SMLATB r11, r6, r9, r11 ; sum1 = MAC16_16(sum1, x_1, y_2) +celt_pitch_xcorr_edsp_process2_1 + LDRH r6, [r4], #2 + ADDS r12, r12, #1 + ; Stall + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDRHGT r7, [r4], #2 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + BLE celt_pitch_xcorr_edsp_process2_done + LDRH r9, [r5], #2 + SMLABT r10, r7, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_1) + SMLABB r11, r7, r9, r11 ; sum1 = MAC16_16(sum1, x_0, y_2) +celt_pitch_xcorr_edsp_process2_done + ; Restore _x + SUB r4, r4, r3, LSL #1 + ; Restore and advance _y + SUB r5, r5, r3, LSL #1 + ; maxcorr = max(maxcorr, sum0) + CMP r0, r10 + ADD r5, r5, #2 + MOVLT r0, r10 + SUB r1, r1, #2 + ; maxcorr = max(maxcorr, sum1) + CMP r0, r11 + ; xcorr[i] = sum + STR r10, [r2], #4 + MOVLT r0, r11 + STR r11, [r2], #4 +celt_pitch_xcorr_edsp_process1a + ADDS r1, r1, #1 + BLT celt_pitch_xcorr_edsp_done + SUBS r12, r3, #4 + ; r14 = sum = 0 + MOV r14, #0 + BLT celt_pitch_xcorr_edsp_process1a_loop_done + LDR r6, [r4], #4 + LDR r8, [r5], #4 + LDR r7, [r4], #4 + LDR r9, [r5], #4 +celt_pitch_xcorr_edsp_process1a_loop4 + SMLABB r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + SUBS r12, r12, #4 ; j-=4 + SMLATT r14, r6, r8, r14 ; sum = MAC16_16(sum, x_1, y_1) + LDRGE r6, [r4], #4 + SMLABB r14, r7, r9, r14 ; sum = MAC16_16(sum, x_2, y_2) + LDRGE r8, [r5], #4 + SMLATT r14, r7, r9, r14 ; sum = MAC16_16(sum, x_3, y_3) + LDRGE r7, [r4], #4 + LDRGE r9, [r5], #4 + BGE celt_pitch_xcorr_edsp_process1a_loop4 +celt_pitch_xcorr_edsp_process1a_loop_done + ADDS r12, r12, #2 + LDRGE r6, [r4], #4 + LDRGE r8, [r5], #4 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + SUBGE r12, r12, #2 + SMLATTGE r14, r6, r8, r14 ; sum = MAC16_16(sum, x_1, y_1) + ADDS r12, r12, #1 + LDRHGE r6, [r4], #2 + LDRHGE r8, [r5], #2 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, *x, *y) + ; maxcorr = max(maxcorr, sum) + CMP r0, r14 + ; xcorr[i] = sum + STR r14, [r2], #4 + MOVLT r0, r14 +celt_pitch_xcorr_edsp_done + LDMFD sp!, {r4-r11, pc} + ENDP + +ENDIF + +END diff --git a/libopus/celt/arm/fft_arm.h b/libopus/celt/arm/fft_arm.h new file mode 100644 index 0000000..0b78175 --- /dev/null +++ b/libopus/celt/arm/fft_arm.h @@ -0,0 +1,71 @@ +/* Copyright (c) 2015 Xiph.Org Foundation + Written by Viswanath Puttagunta */ +/** + @file fft_arm.h + @brief ARM Neon Intrinsic optimizations for fft using NE10 library + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + + +#if !defined(FFT_ARM_H) +#define FFT_ARM_H + +#include "kiss_fft.h" + +#if defined(HAVE_ARM_NE10) + +int opus_fft_alloc_arm_neon(kiss_fft_state *st); +void opus_fft_free_arm_neon(kiss_fft_state *st); + +void opus_fft_neon(const kiss_fft_state *st, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout); + +void opus_ifft_neon(const kiss_fft_state *st, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout); + +#if !defined(OPUS_HAVE_RTCD) +#define OVERRIDE_OPUS_FFT (1) + +#define opus_fft_alloc_arch(_st, arch) \ + ((void)(arch), opus_fft_alloc_arm_neon(_st)) + +#define opus_fft_free_arch(_st, arch) \ + ((void)(arch), opus_fft_free_arm_neon(_st)) + +#define opus_fft(_st, _fin, _fout, arch) \ + ((void)(arch), opus_fft_neon(_st, _fin, _fout)) + +#define opus_ifft(_st, _fin, _fout, arch) \ + ((void)(arch), opus_ifft_neon(_st, _fin, _fout)) + +#endif /* OPUS_HAVE_RTCD */ + +#endif /* HAVE_ARM_NE10 */ + +#endif diff --git a/libopus/celt/arm/fixed_arm64.h b/libopus/celt/arm/fixed_arm64.h new file mode 100644 index 0000000..c6fbd3d --- /dev/null +++ b/libopus/celt/arm/fixed_arm64.h @@ -0,0 +1,35 @@ +/* Copyright (C) 2015 Vidyo */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_ARM64_H +#define FIXED_ARM64_H + +#include <arm_neon.h> + +#undef SIG2WORD16 +#define SIG2WORD16(x) (vqmovns_s32(PSHR32((x), SIG_SHIFT))) + +#endif diff --git a/libopus/celt/arm/fixed_armv4.h b/libopus/celt/arm/fixed_armv4.h new file mode 100644 index 0000000..d84888a --- /dev/null +++ b/libopus/celt/arm/fixed_armv4.h @@ -0,0 +1,80 @@ +/* Copyright (C) 2013 Xiph.Org Foundation and contributors */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_ARMv4_H +#define FIXED_ARMv4_H + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q16 +static OPUS_INLINE opus_val32 MULT16_32_Q16_armv4(opus_val16 a, opus_val32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#MULT16_32_Q16\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(b),"r"(SHL32(a,16)) + ); + return rd_hi; +} +#define MULT16_32_Q16(a, b) (MULT16_32_Q16_armv4(a, b)) + + +/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q15 +static OPUS_INLINE opus_val32 MULT16_32_Q15_armv4(opus_val16 a, opus_val32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#MULT16_32_Q15\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(b), "r"(SHL32(a,16)) + ); + /*We intentionally don't OR in the high bit of rd_lo for speed.*/ + return SHL32(rd_hi,1); +} +#define MULT16_32_Q15(a, b) (MULT16_32_Q15_armv4(a, b)) + + +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ +#undef MAC16_32_Q15 +#define MAC16_32_Q15(c, a, b) ADD32(c, MULT16_32_Q15(a, b)) + +/** 16x32 multiply, followed by a 16-bit shift right and 32-bit add. + Result fits in 32 bits. */ +#undef MAC16_32_Q16 +#define MAC16_32_Q16(c, a, b) ADD32(c, MULT16_32_Q16(a, b)) + +/** 32x32 multiplication, followed by a 31-bit shift right. Results fits in 32 bits */ +#undef MULT32_32_Q31 +#define MULT32_32_Q31(a,b) (opus_val32)((((opus_int64)(a)) * ((opus_int64)(b)))>>31) + +#endif diff --git a/libopus/celt/arm/fixed_armv5e.h b/libopus/celt/arm/fixed_armv5e.h new file mode 100644 index 0000000..6bf73cb --- /dev/null +++ b/libopus/celt/arm/fixed_armv5e.h @@ -0,0 +1,151 @@ +/* Copyright (C) 2007-2009 Xiph.Org Foundation + Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2008 CSIRO + Copyright (C) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_ARMv5E_H +#define FIXED_ARMv5E_H + +#include "fixed_armv4.h" + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q16 +static OPUS_INLINE opus_val32 MULT16_32_Q16_armv5e(opus_val16 a, opus_val32 b) +{ + int res; + __asm__( + "#MULT16_32_Q16\n\t" + "smulwb %0, %1, %2\n\t" + : "=r"(res) + : "r"(b),"r"(a) + ); + return res; +} +#define MULT16_32_Q16(a, b) (MULT16_32_Q16_armv5e(a, b)) + + +/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q15 +static OPUS_INLINE opus_val32 MULT16_32_Q15_armv5e(opus_val16 a, opus_val32 b) +{ + int res; + __asm__( + "#MULT16_32_Q15\n\t" + "smulwb %0, %1, %2\n\t" + : "=r"(res) + : "r"(b), "r"(a) + ); + return SHL32(res,1); +} +#define MULT16_32_Q15(a, b) (MULT16_32_Q15_armv5e(a, b)) + + +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ +#undef MAC16_32_Q15 +static OPUS_INLINE opus_val32 MAC16_32_Q15_armv5e(opus_val32 c, opus_val16 a, + opus_val32 b) +{ + int res; + __asm__( + "#MAC16_32_Q15\n\t" + "smlawb %0, %1, %2, %3;\n" + : "=r"(res) + : "r"(SHL32(b,1)), "r"(a), "r"(c) + ); + return res; +} +#define MAC16_32_Q15(c, a, b) (MAC16_32_Q15_armv5e(c, a, b)) + +/** 16x32 multiply, followed by a 16-bit shift right and 32-bit add. + Result fits in 32 bits. */ +#undef MAC16_32_Q16 +static OPUS_INLINE opus_val32 MAC16_32_Q16_armv5e(opus_val32 c, opus_val16 a, + opus_val32 b) +{ + int res; + __asm__( + "#MAC16_32_Q16\n\t" + "smlawb %0, %1, %2, %3;\n" + : "=r"(res) + : "r"(b), "r"(a), "r"(c) + ); + return res; +} +#define MAC16_32_Q16(c, a, b) (MAC16_32_Q16_armv5e(c, a, b)) + +/** 16x16 multiply-add where the result fits in 32 bits */ +#undef MAC16_16 +static OPUS_INLINE opus_val32 MAC16_16_armv5e(opus_val32 c, opus_val16 a, + opus_val16 b) +{ + int res; + __asm__( + "#MAC16_16\n\t" + "smlabb %0, %1, %2, %3;\n" + : "=r"(res) + : "r"(a), "r"(b), "r"(c) + ); + return res; +} +#define MAC16_16(c, a, b) (MAC16_16_armv5e(c, a, b)) + +/** 16x16 multiplication where the result fits in 32 bits */ +#undef MULT16_16 +static OPUS_INLINE opus_val32 MULT16_16_armv5e(opus_val16 a, opus_val16 b) +{ + int res; + __asm__( + "#MULT16_16\n\t" + "smulbb %0, %1, %2;\n" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define MULT16_16(a, b) (MULT16_16_armv5e(a, b)) + +#ifdef OPUS_ARM_INLINE_MEDIA + +#undef SIG2WORD16 +static OPUS_INLINE opus_val16 SIG2WORD16_armv6(opus_val32 x) +{ + celt_sig res; + __asm__( + "#SIG2WORD16\n\t" + "ssat %0, #16, %1, ASR #12\n\t" + : "=r"(res) + : "r"(x+2048) + ); + return EXTRACT16(res); +} +#define SIG2WORD16(x) (SIG2WORD16_armv6(x)) + +#endif /* OPUS_ARM_INLINE_MEDIA */ + +#endif diff --git a/libopus/celt/arm/kiss_fft_armv4.h b/libopus/celt/arm/kiss_fft_armv4.h new file mode 100644 index 0000000..e4faad6 --- /dev/null +++ b/libopus/celt/arm/kiss_fft_armv4.h @@ -0,0 +1,121 @@ +/*Copyright (c) 2013, Xiph.Org Foundation and contributors. + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_ARMv4_H +#define KISS_FFT_ARMv4_H + +#if !defined(KISS_FFT_GUTS_H) +#error "This file should only be included from _kiss_fft_guts.h" +#endif + +#ifdef FIXED_POINT + +#undef C_MUL +#define C_MUL(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MUL\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mi], r1, %[br]\n\t" \ + "smlal %[tt], %[mi], r0, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mr], r0, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #15\n\t" \ + "smlal %[br], %[mr], r1, %[bi]\n\t" \ + "orr %[mi], %[tt], %[mi], lsl #17\n\t" \ + "mov %[br], %[br], lsr #15\n\t" \ + "orr %[mr], %[br], %[mr], lsl #17\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#undef C_MUL4 +#define C_MUL4(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MUL4\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mi], r1, %[br]\n\t" \ + "smlal %[tt], %[mi], r0, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mr], r0, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #17\n\t" \ + "smlal %[br], %[mr], r1, %[bi]\n\t" \ + "orr %[mi], %[tt], %[mi], lsl #15\n\t" \ + "mov %[br], %[br], lsr #17\n\t" \ + "orr %[mr], %[br], %[mr], lsl #15\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#undef C_MULC +#define C_MULC(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MULC\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mr], r0, %[br]\n\t" \ + "smlal %[tt], %[mr], r1, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mi], r1, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #15\n\t" \ + "smlal %[br], %[mi], r0, %[bi]\n\t" \ + "orr %[mr], %[tt], %[mr], lsl #17\n\t" \ + "mov %[br], %[br], lsr #15\n\t" \ + "orr %[mi], %[br], %[mi], lsl #17\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#endif /* FIXED_POINT */ + +#endif /* KISS_FFT_ARMv4_H */ diff --git a/libopus/celt/arm/kiss_fft_armv5e.h b/libopus/celt/arm/kiss_fft_armv5e.h new file mode 100644 index 0000000..9eca183 --- /dev/null +++ b/libopus/celt/arm/kiss_fft_armv5e.h @@ -0,0 +1,118 @@ +/*Copyright (c) 2013, Xiph.Org Foundation and contributors. + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_ARMv5E_H +#define KISS_FFT_ARMv5E_H + +#if !defined(KISS_FFT_GUTS_H) +#error "This file should only be included from _kiss_fft_guts.h" +#endif + +#ifdef FIXED_POINT + +#if defined(__thumb__)||defined(__thumb2__) +#define LDRD_CONS "Q" +#else +#define LDRD_CONS "Uq" +#endif + +#undef C_MUL +#define C_MUL(m,a,b) \ + do{ \ + int mr1__; \ + int mr2__; \ + int mi__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MUL\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mi], %H[aval], %[bval]\n\t" \ + "smulwb %[mr1], %[aval], %[bval]\n\t" \ + "smulwt %[mr2], %H[aval], %[bval]\n\t" \ + "smlawt %[mi], %[aval], %[bval], %[mi]\n\t" \ + : [mr1]"=r"(mr1__), [mr2]"=r"(mr2__), [mi]"=r"(mi__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHL32(SUB32(mr1__, mr2__), 1); \ + (m).i = SHL32(mi__, 1); \ + } \ + while(0) + +#undef C_MUL4 +#define C_MUL4(m,a,b) \ + do{ \ + int mr1__; \ + int mr2__; \ + int mi__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MUL4\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mi], %H[aval], %[bval]\n\t" \ + "smulwb %[mr1], %[aval], %[bval]\n\t" \ + "smulwt %[mr2], %H[aval], %[bval]\n\t" \ + "smlawt %[mi], %[aval], %[bval], %[mi]\n\t" \ + : [mr1]"=r"(mr1__), [mr2]"=r"(mr2__), [mi]"=r"(mi__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHR32(SUB32(mr1__, mr2__), 1); \ + (m).i = SHR32(mi__, 1); \ + } \ + while(0) + +#undef C_MULC +#define C_MULC(m,a,b) \ + do{ \ + int mr__; \ + int mi1__; \ + int mi2__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MULC\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mr], %[aval], %[bval]\n\t" \ + "smulwb %[mi1], %H[aval], %[bval]\n\t" \ + "smulwt %[mi2], %[aval], %[bval]\n\t" \ + "smlawt %[mr], %H[aval], %[bval], %[mr]\n\t" \ + : [mr]"=r"(mr__), [mi1]"=r"(mi1__), [mi2]"=r"(mi2__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHL32(mr__, 1); \ + (m).i = SHL32(SUB32(mi1__, mi2__), 1); \ + } \ + while(0) + +#endif /* FIXED_POINT */ + +#endif /* KISS_FFT_GUTS_H */ diff --git a/libopus/celt/arm/mdct_arm.h b/libopus/celt/arm/mdct_arm.h new file mode 100644 index 0000000..14200ba --- /dev/null +++ b/libopus/celt/arm/mdct_arm.h @@ -0,0 +1,59 @@ +/* Copyright (c) 2015 Xiph.Org Foundation + Written by Viswanath Puttagunta */ +/** + @file arm_mdct.h + @brief ARM Neon Intrinsic optimizations for mdct using NE10 library + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(MDCT_ARM_H) +#define MDCT_ARM_H + +#include "mdct.h" + +#if defined(HAVE_ARM_NE10) +/** Compute a forward MDCT and scale by 4/N, trashes the input array */ +void clt_mdct_forward_neon(const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, + int shift, int stride, int arch); + +void clt_mdct_backward_neon(const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, + int shift, int stride, int arch); + +#if !defined(OPUS_HAVE_RTCD) +#define OVERRIDE_OPUS_MDCT (1) +#define clt_mdct_forward(_l, _in, _out, _window, _int, _shift, _stride, _arch) \ + clt_mdct_forward_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch) +#define clt_mdct_backward(_l, _in, _out, _window, _int, _shift, _stride, _arch) \ + clt_mdct_backward_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch) +#endif /* OPUS_HAVE_RTCD */ +#endif /* HAVE_ARM_NE10 */ + +#endif diff --git a/libopus/celt/arm/pitch_arm.h b/libopus/celt/arm/pitch_arm.h new file mode 100644 index 0000000..bed8b04 --- /dev/null +++ b/libopus/celt/arm/pitch_arm.h @@ -0,0 +1,160 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(PITCH_ARM_H) +# define PITCH_ARM_H + +# include "armcpu.h" + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +opus_val32 celt_inner_prod_neon(const opus_val16 *x, const opus_val16 *y, int N); +void dual_inner_prod_neon(const opus_val16 *x, const opus_val16 *y01, + const opus_val16 *y02, int N, opus_val32 *xy1, opus_val32 *xy2); + +# if !defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_PRESUME_NEON) +# define OVERRIDE_CELT_INNER_PROD (1) +# define OVERRIDE_DUAL_INNER_PROD (1) +# define celt_inner_prod(x, y, N, arch) ((void)(arch), PRESUME_NEON(celt_inner_prod)(x, y, N)) +# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) ((void)(arch), PRESUME_NEON(dual_inner_prod)(x, y01, y02, N, xy1, xy2)) +# endif +# endif + +# if !defined(OVERRIDE_CELT_INNER_PROD) +# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, const opus_val16 *y, int N); +# define OVERRIDE_CELT_INNER_PROD (1) +# define celt_inner_prod(x, y, N, arch) ((*CELT_INNER_PROD_IMPL[(arch)&OPUS_ARCHMASK])(x, y, N)) +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) +# define OVERRIDE_CELT_INNER_PROD (1) +# define celt_inner_prod(x, y, N, arch) ((void)(arch), celt_inner_prod_neon(x, y, N)) +# endif +# endif + +# if !defined(OVERRIDE_DUAL_INNER_PROD) +# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, + const opus_val16 *y01, const opus_val16 *y02, int N, opus_val32 *xy1, opus_val32 *xy2); +# define OVERRIDE_DUAL_INNER_PROD (1) +# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) ((*DUAL_INNER_PROD_IMPL[(arch)&OPUS_ARCHMASK])(x, y01, y02, N, xy1, xy2)) +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) +# define OVERRIDE_DUAL_INNER_PROD (1) +# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) ((void)(arch), dual_inner_prod_neon(x, y01, y02, N, xy1, xy2)) +# endif +# endif + +# if defined(FIXED_POINT) + +# if defined(OPUS_ARM_MAY_HAVE_NEON) +opus_val32 celt_pitch_xcorr_neon(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch); +# endif + +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) +# define celt_pitch_xcorr_media MAY_HAVE_EDSP(celt_pitch_xcorr) +# endif + +# if defined(OPUS_ARM_MAY_HAVE_EDSP) +opus_val32 celt_pitch_xcorr_edsp(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch); +# endif + +# if defined(OPUS_HAVE_RTCD) && \ + ((defined(OPUS_ARM_MAY_HAVE_NEON) && !defined(OPUS_ARM_PRESUME_NEON)) || \ + (defined(OPUS_ARM_MAY_HAVE_MEDIA) && !defined(OPUS_ARM_PRESUME_MEDIA)) || \ + (defined(OPUS_ARM_MAY_HAVE_EDSP) && !defined(OPUS_ARM_PRESUME_EDSP))) +extern opus_val32 +(*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, + const opus_val16 *, opus_val32 *, int, int, int); +# define OVERRIDE_PITCH_XCORR (1) +# define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \ + ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \ + xcorr, len, max_pitch, arch)) + +# elif defined(OPUS_ARM_PRESUME_EDSP) || \ + defined(OPUS_ARM_PRESUME_MEDIA) || \ + defined(OPUS_ARM_PRESUME_NEON) +# define OVERRIDE_PITCH_XCORR (1) +# define celt_pitch_xcorr (PRESUME_NEON(celt_pitch_xcorr)) + +# endif + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +void xcorr_kernel_neon_fixed( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len); +# endif + +# if defined(OPUS_HAVE_RTCD) && \ + (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) + +extern void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len); + +# define OVERRIDE_XCORR_KERNEL (1) +# define xcorr_kernel(x, y, sum, len, arch) \ + ((*XCORR_KERNEL_IMPL[(arch) & OPUS_ARCHMASK])(x, y, sum, len)) + +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) +# define OVERRIDE_XCORR_KERNEL (1) +# define xcorr_kernel(x, y, sum, len, arch) \ + ((void)arch, xcorr_kernel_neon_fixed(x, y, sum, len)) + +# endif + +#else /* Start !FIXED_POINT */ +/* Float case */ +#if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch); +#endif + +# if defined(OPUS_HAVE_RTCD) && \ + (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern void +(*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, + const opus_val16 *, opus_val32 *, int, int, int); + +# define OVERRIDE_PITCH_XCORR (1) +# define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \ + ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \ + xcorr, len, max_pitch, arch)) + +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) + +# define OVERRIDE_PITCH_XCORR (1) +# define celt_pitch_xcorr celt_pitch_xcorr_float_neon + +# endif + +#endif /* end !FIXED_POINT */ + +#endif diff --git a/libopus/celt/arm/pitch_neon_intr.c b/libopus/celt/arm/pitch_neon_intr.c new file mode 100644 index 0000000..1ac38c4 --- /dev/null +++ b/libopus/celt/arm/pitch_neon_intr.c @@ -0,0 +1,290 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <arm_neon.h> +#include "pitch.h" + +#ifdef FIXED_POINT + +opus_val32 celt_inner_prod_neon(const opus_val16 *x, const opus_val16 *y, int N) +{ + int i; + opus_val32 xy; + int16x8_t x_s16x8, y_s16x8; + int32x4_t xy_s32x4 = vdupq_n_s32(0); + int64x2_t xy_s64x2; + int64x1_t xy_s64x1; + + for (i = 0; i < N - 7; i += 8) { + x_s16x8 = vld1q_s16(&x[i]); + y_s16x8 = vld1q_s16(&y[i]); + xy_s32x4 = vmlal_s16(xy_s32x4, vget_low_s16 (x_s16x8), vget_low_s16 (y_s16x8)); + xy_s32x4 = vmlal_s16(xy_s32x4, vget_high_s16(x_s16x8), vget_high_s16(y_s16x8)); + } + + if (N - i >= 4) { + const int16x4_t x_s16x4 = vld1_s16(&x[i]); + const int16x4_t y_s16x4 = vld1_s16(&y[i]); + xy_s32x4 = vmlal_s16(xy_s32x4, x_s16x4, y_s16x4); + i += 4; + } + + xy_s64x2 = vpaddlq_s32(xy_s32x4); + xy_s64x1 = vadd_s64(vget_low_s64(xy_s64x2), vget_high_s64(xy_s64x2)); + xy = vget_lane_s32(vreinterpret_s32_s64(xy_s64x1), 0); + + for (; i < N; i++) { + xy = MAC16_16(xy, x[i], y[i]); + } + +#ifdef OPUS_CHECK_ASM + celt_assert(celt_inner_prod_c(x, y, N) == xy); +#endif + + return xy; +} + +void dual_inner_prod_neon(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + opus_val32 xy01, xy02; + int16x8_t x_s16x8, y01_s16x8, y02_s16x8; + int32x4_t xy01_s32x4 = vdupq_n_s32(0); + int32x4_t xy02_s32x4 = vdupq_n_s32(0); + int64x2_t xy01_s64x2, xy02_s64x2; + int64x1_t xy01_s64x1, xy02_s64x1; + + for (i = 0; i < N - 7; i += 8) { + x_s16x8 = vld1q_s16(&x[i]); + y01_s16x8 = vld1q_s16(&y01[i]); + y02_s16x8 = vld1q_s16(&y02[i]); + xy01_s32x4 = vmlal_s16(xy01_s32x4, vget_low_s16 (x_s16x8), vget_low_s16 (y01_s16x8)); + xy02_s32x4 = vmlal_s16(xy02_s32x4, vget_low_s16 (x_s16x8), vget_low_s16 (y02_s16x8)); + xy01_s32x4 = vmlal_s16(xy01_s32x4, vget_high_s16(x_s16x8), vget_high_s16(y01_s16x8)); + xy02_s32x4 = vmlal_s16(xy02_s32x4, vget_high_s16(x_s16x8), vget_high_s16(y02_s16x8)); + } + + if (N - i >= 4) { + const int16x4_t x_s16x4 = vld1_s16(&x[i]); + const int16x4_t y01_s16x4 = vld1_s16(&y01[i]); + const int16x4_t y02_s16x4 = vld1_s16(&y02[i]); + xy01_s32x4 = vmlal_s16(xy01_s32x4, x_s16x4, y01_s16x4); + xy02_s32x4 = vmlal_s16(xy02_s32x4, x_s16x4, y02_s16x4); + i += 4; + } + + xy01_s64x2 = vpaddlq_s32(xy01_s32x4); + xy02_s64x2 = vpaddlq_s32(xy02_s32x4); + xy01_s64x1 = vadd_s64(vget_low_s64(xy01_s64x2), vget_high_s64(xy01_s64x2)); + xy02_s64x1 = vadd_s64(vget_low_s64(xy02_s64x2), vget_high_s64(xy02_s64x2)); + xy01 = vget_lane_s32(vreinterpret_s32_s64(xy01_s64x1), 0); + xy02 = vget_lane_s32(vreinterpret_s32_s64(xy02_s64x1), 0); + + for (; i < N; i++) { + xy01 = MAC16_16(xy01, x[i], y01[i]); + xy02 = MAC16_16(xy02, x[i], y02[i]); + } + *xy1 = xy01; + *xy2 = xy02; + +#ifdef OPUS_CHECK_ASM + { + opus_val32 xy1_c, xy2_c; + dual_inner_prod_c(x, y01, y02, N, &xy1_c, &xy2_c); + celt_assert(xy1_c == *xy1); + celt_assert(xy2_c == *xy2); + } +#endif +} + +#else /* !FIXED_POINT */ + +/* ========================================================================== */ + +#ifdef OPUS_CHECK_ASM + +/* This part of code simulates floating-point NEON operations. */ + +/* celt_inner_prod_neon_float_c_simulation() simulates the floating-point */ +/* operations of celt_inner_prod_neon(), and both functions should have bit */ +/* exact output. */ +static opus_val32 celt_inner_prod_neon_float_c_simulation(const opus_val16 *x, const opus_val16 *y, int N) +{ + int i; + opus_val32 xy, xy0 = 0, xy1 = 0, xy2 = 0, xy3 = 0; + for (i = 0; i < N - 3; i += 4) { + xy0 = MAC16_16(xy0, x[i + 0], y[i + 0]); + xy1 = MAC16_16(xy1, x[i + 1], y[i + 1]); + xy2 = MAC16_16(xy2, x[i + 2], y[i + 2]); + xy3 = MAC16_16(xy3, x[i + 3], y[i + 3]); + } + xy0 += xy2; + xy1 += xy3; + xy = xy0 + xy1; + for (; i < N; i++) { + xy = MAC16_16(xy, x[i], y[i]); + } + return xy; +} + +/* dual_inner_prod_neon_float_c_simulation() simulates the floating-point */ +/* operations of dual_inner_prod_neon(), and both functions should have bit */ +/* exact output. */ +static void dual_inner_prod_neon_float_c_simulation(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + opus_val32 xy01, xy02, xy01_0 = 0, xy01_1 = 0, xy01_2 = 0, xy01_3 = 0, xy02_0 = 0, xy02_1 = 0, xy02_2 = 0, xy02_3 = 0; + for (i = 0; i < N - 3; i += 4) { + xy01_0 = MAC16_16(xy01_0, x[i + 0], y01[i + 0]); + xy01_1 = MAC16_16(xy01_1, x[i + 1], y01[i + 1]); + xy01_2 = MAC16_16(xy01_2, x[i + 2], y01[i + 2]); + xy01_3 = MAC16_16(xy01_3, x[i + 3], y01[i + 3]); + xy02_0 = MAC16_16(xy02_0, x[i + 0], y02[i + 0]); + xy02_1 = MAC16_16(xy02_1, x[i + 1], y02[i + 1]); + xy02_2 = MAC16_16(xy02_2, x[i + 2], y02[i + 2]); + xy02_3 = MAC16_16(xy02_3, x[i + 3], y02[i + 3]); + } + xy01_0 += xy01_2; + xy02_0 += xy02_2; + xy01_1 += xy01_3; + xy02_1 += xy02_3; + xy01 = xy01_0 + xy01_1; + xy02 = xy02_0 + xy02_1; + for (; i < N; i++) { + xy01 = MAC16_16(xy01, x[i], y01[i]); + xy02 = MAC16_16(xy02, x[i], y02[i]); + } + *xy1 = xy01; + *xy2 = xy02; +} + +#endif /* OPUS_CHECK_ASM */ + +/* ========================================================================== */ + +opus_val32 celt_inner_prod_neon(const opus_val16 *x, const opus_val16 *y, int N) +{ + int i; + opus_val32 xy; + float32x4_t xy_f32x4 = vdupq_n_f32(0); + float32x2_t xy_f32x2; + + for (i = 0; i < N - 7; i += 8) { + float32x4_t x_f32x4, y_f32x4; + x_f32x4 = vld1q_f32(&x[i]); + y_f32x4 = vld1q_f32(&y[i]); + xy_f32x4 = vmlaq_f32(xy_f32x4, x_f32x4, y_f32x4); + x_f32x4 = vld1q_f32(&x[i + 4]); + y_f32x4 = vld1q_f32(&y[i + 4]); + xy_f32x4 = vmlaq_f32(xy_f32x4, x_f32x4, y_f32x4); + } + + if (N - i >= 4) { + const float32x4_t x_f32x4 = vld1q_f32(&x[i]); + const float32x4_t y_f32x4 = vld1q_f32(&y[i]); + xy_f32x4 = vmlaq_f32(xy_f32x4, x_f32x4, y_f32x4); + i += 4; + } + + xy_f32x2 = vadd_f32(vget_low_f32(xy_f32x4), vget_high_f32(xy_f32x4)); + xy_f32x2 = vpadd_f32(xy_f32x2, xy_f32x2); + xy = vget_lane_f32(xy_f32x2, 0); + + for (; i < N; i++) { + xy = MAC16_16(xy, x[i], y[i]); + } + +#ifdef OPUS_CHECK_ASM + celt_assert(ABS32(celt_inner_prod_neon_float_c_simulation(x, y, N) - xy) <= VERY_SMALL); +#endif + + return xy; +} + +void dual_inner_prod_neon(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + opus_val32 xy01, xy02; + float32x4_t xy01_f32x4 = vdupq_n_f32(0); + float32x4_t xy02_f32x4 = vdupq_n_f32(0); + float32x2_t xy01_f32x2, xy02_f32x2; + + for (i = 0; i < N - 7; i += 8) { + float32x4_t x_f32x4, y01_f32x4, y02_f32x4; + x_f32x4 = vld1q_f32(&x[i]); + y01_f32x4 = vld1q_f32(&y01[i]); + y02_f32x4 = vld1q_f32(&y02[i]); + xy01_f32x4 = vmlaq_f32(xy01_f32x4, x_f32x4, y01_f32x4); + xy02_f32x4 = vmlaq_f32(xy02_f32x4, x_f32x4, y02_f32x4); + x_f32x4 = vld1q_f32(&x[i + 4]); + y01_f32x4 = vld1q_f32(&y01[i + 4]); + y02_f32x4 = vld1q_f32(&y02[i + 4]); + xy01_f32x4 = vmlaq_f32(xy01_f32x4, x_f32x4, y01_f32x4); + xy02_f32x4 = vmlaq_f32(xy02_f32x4, x_f32x4, y02_f32x4); + } + + if (N - i >= 4) { + const float32x4_t x_f32x4 = vld1q_f32(&x[i]); + const float32x4_t y01_f32x4 = vld1q_f32(&y01[i]); + const float32x4_t y02_f32x4 = vld1q_f32(&y02[i]); + xy01_f32x4 = vmlaq_f32(xy01_f32x4, x_f32x4, y01_f32x4); + xy02_f32x4 = vmlaq_f32(xy02_f32x4, x_f32x4, y02_f32x4); + i += 4; + } + + xy01_f32x2 = vadd_f32(vget_low_f32(xy01_f32x4), vget_high_f32(xy01_f32x4)); + xy02_f32x2 = vadd_f32(vget_low_f32(xy02_f32x4), vget_high_f32(xy02_f32x4)); + xy01_f32x2 = vpadd_f32(xy01_f32x2, xy01_f32x2); + xy02_f32x2 = vpadd_f32(xy02_f32x2, xy02_f32x2); + xy01 = vget_lane_f32(xy01_f32x2, 0); + xy02 = vget_lane_f32(xy02_f32x2, 0); + + for (; i < N; i++) { + xy01 = MAC16_16(xy01, x[i], y01[i]); + xy02 = MAC16_16(xy02, x[i], y02[i]); + } + *xy1 = xy01; + *xy2 = xy02; + +#ifdef OPUS_CHECK_ASM + { + opus_val32 xy1_c, xy2_c; + dual_inner_prod_neon_float_c_simulation(x, y01, y02, N, &xy1_c, &xy2_c); + celt_assert(ABS32(xy1_c - *xy1) <= VERY_SMALL); + celt_assert(ABS32(xy2_c - *xy2) <= VERY_SMALL); + } +#endif +} + +#endif /* FIXED_POINT */ diff --git a/libopus/celt/bands.c b/libopus/celt/bands.c new file mode 100644 index 0000000..a9aa78d --- /dev/null +++ b/libopus/celt/bands.c @@ -0,0 +1,1673 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008-2009 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <math.h> +#include "bands.h" +#include "modes.h" +#include "vq.h" +#include "cwrs.h" +#include "stack_alloc.h" +#include "os_support.h" +#include "mathops.h" +#include "rate.h" +#include "quant_bands.h" +#include "pitch.h" + +int hysteresis_decision(opus_val16 val, const opus_val16 *thresholds, const opus_val16 *hysteresis, int N, int prev) +{ + int i; + for (i=0;i<N;i++) + { + if (val < thresholds[i]) + break; + } + if (i>prev && val < thresholds[prev]+hysteresis[prev]) + i=prev; + if (i<prev && val > thresholds[prev-1]-hysteresis[prev-1]) + i=prev; + return i; +} + +opus_uint32 celt_lcg_rand(opus_uint32 seed) +{ + return 1664525 * seed + 1013904223; +} + +/* This is a cos() approximation designed to be bit-exact on any platform. Bit exactness + with this approximation is important because it has an impact on the bit allocation */ +opus_int16 bitexact_cos(opus_int16 x) +{ + opus_int32 tmp; + opus_int16 x2; + tmp = (4096+((opus_int32)(x)*(x)))>>13; + celt_sig_assert(tmp<=32767); + x2 = tmp; + x2 = (32767-x2) + FRAC_MUL16(x2, (-7651 + FRAC_MUL16(x2, (8277 + FRAC_MUL16(-626, x2))))); + celt_sig_assert(x2<=32766); + return 1+x2; +} + +int bitexact_log2tan(int isin,int icos) +{ + int lc; + int ls; + lc=EC_ILOG(icos); + ls=EC_ILOG(isin); + icos<<=15-lc; + isin<<=15-ls; + return (ls-lc)*(1<<11) + +FRAC_MUL16(isin, FRAC_MUL16(isin, -2597) + 7932) + -FRAC_MUL16(icos, FRAC_MUL16(icos, -2597) + 7932); +} + +#ifdef FIXED_POINT +/* Compute the amplitude (sqrt energy) in each of the bands */ +void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + (void)arch; + N = m->shortMdctSize<<LM; + c=0; do { + for (i=0;i<end;i++) + { + int j; + opus_val32 maxval=0; + opus_val32 sum = 0; + + maxval = celt_maxabs32(&X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM); + if (maxval > 0) + { + int shift = celt_ilog2(maxval) - 14 + (((m->logN[i]>>BITRES)+LM+1)>>1); + j=eBands[i]<<LM; + if (shift>0) + { + do { + sum = MAC16_16(sum, EXTRACT16(SHR32(X[j+c*N],shift)), + EXTRACT16(SHR32(X[j+c*N],shift))); + } while (++j<eBands[i+1]<<LM); + } else { + do { + sum = MAC16_16(sum, EXTRACT16(SHL32(X[j+c*N],-shift)), + EXTRACT16(SHL32(X[j+c*N],-shift))); + } while (++j<eBands[i+1]<<LM); + } + /* We're adding one here to ensure the normalized band isn't larger than unity norm */ + bandE[i+c*m->nbEBands] = EPSILON+VSHR32(EXTEND32(celt_sqrt(sum)),-shift); + } else { + bandE[i+c*m->nbEBands] = EPSILON; + } + /*printf ("%f ", bandE[i+c*m->nbEBands]);*/ + } + } while (++c<C); + /*printf ("\n");*/ +} + +/* Normalise each band such that the energy is one. */ +void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + N = M*m->shortMdctSize; + c=0; do { + i=0; do { + opus_val16 g; + int j,shift; + opus_val16 E; + shift = celt_zlog2(bandE[i+c*m->nbEBands])-13; + E = VSHR32(bandE[i+c*m->nbEBands], shift); + g = EXTRACT16(celt_rcp(SHL32(E,3))); + j=M*eBands[i]; do { + X[j+c*N] = MULT16_16_Q15(VSHR32(freq[j+c*N],shift-1),g); + } while (++j<M*eBands[i+1]); + } while (++i<end); + } while (++c<C); +} + +#else /* FIXED_POINT */ +/* Compute the amplitude (sqrt energy) in each of the bands */ +void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + N = m->shortMdctSize<<LM; + c=0; do { + for (i=0;i<end;i++) + { + opus_val32 sum; + sum = 1e-27f + celt_inner_prod(&X[c*N+(eBands[i]<<LM)], &X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM, arch); + bandE[i+c*m->nbEBands] = celt_sqrt(sum); + /*printf ("%f ", bandE[i+c*m->nbEBands]);*/ + } + } while (++c<C); + /*printf ("\n");*/ +} + +/* Normalise each band such that the energy is one. */ +void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + N = M*m->shortMdctSize; + c=0; do { + for (i=0;i<end;i++) + { + int j; + opus_val16 g = 1.f/(1e-27f+bandE[i+c*m->nbEBands]); + for (j=M*eBands[i];j<M*eBands[i+1];j++) + X[j+c*N] = freq[j+c*N]*g; + } + } while (++c<C); +} + +#endif /* FIXED_POINT */ + +/* De-normalise the energy to produce the synthesis from the unit-energy bands */ +void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, + celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandLogE, int start, + int end, int M, int downsample, int silence) +{ + int i, N; + int bound; + celt_sig * OPUS_RESTRICT f; + const celt_norm * OPUS_RESTRICT x; + const opus_int16 *eBands = m->eBands; + N = M*m->shortMdctSize; + bound = M*eBands[end]; + if (downsample!=1) + bound = IMIN(bound, N/downsample); + if (silence) + { + bound = 0; + start = end = 0; + } + f = freq; + x = X+M*eBands[start]; + for (i=0;i<M*eBands[start];i++) + *f++ = 0; + for (i=start;i<end;i++) + { + int j, band_end; + opus_val16 g; + opus_val16 lg; +#ifdef FIXED_POINT + int shift; +#endif + j=M*eBands[i]; + band_end = M*eBands[i+1]; + lg = SATURATE16(ADD32(bandLogE[i], SHL32((opus_val32)eMeans[i],6))); +#ifndef FIXED_POINT + g = celt_exp2(MIN32(32.f, lg)); +#else + /* Handle the integer part of the log energy */ + shift = 16-(lg>>DB_SHIFT); + if (shift>31) + { + shift=0; + g=0; + } else { + /* Handle the fractional part. */ + g = celt_exp2_frac(lg&((1<<DB_SHIFT)-1)); + } + /* Handle extreme gains with negative shift. */ + if (shift<0) + { + /* For shift <= -2 and g > 16384 we'd be likely to overflow, so we're + capping the gain here, which is equivalent to a cap of 18 on lg. + This shouldn't trigger unless the bitstream is already corrupted. */ + if (shift <= -2) + { + g = 16384; + shift = -2; + } + do { + *f++ = SHL32(MULT16_16(*x++, g), -shift); + } while (++j<band_end); + } else +#endif + /* Be careful of the fixed-point "else" just above when changing this code */ + do { + *f++ = SHR32(MULT16_16(*x++, g), shift); + } while (++j<band_end); + } + celt_assert(start <= end); + OPUS_CLEAR(&freq[bound], N-bound); +} + +/* This prevents energy collapse for transients with multiple short MDCTs */ +void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_masks, int LM, int C, int size, + int start, int end, const opus_val16 *logE, const opus_val16 *prev1logE, + const opus_val16 *prev2logE, const int *pulses, opus_uint32 seed, int arch) +{ + int c, i, j, k; + for (i=start;i<end;i++) + { + int N0; + opus_val16 thresh, sqrt_1; + int depth; +#ifdef FIXED_POINT + int shift; + opus_val32 thresh32; +#endif + + N0 = m->eBands[i+1]-m->eBands[i]; + /* depth in 1/8 bits */ + celt_sig_assert(pulses[i]>=0); + depth = celt_udiv(1+pulses[i], (m->eBands[i+1]-m->eBands[i]))>>LM; + +#ifdef FIXED_POINT + thresh32 = SHR32(celt_exp2(-SHL16(depth, 10-BITRES)),1); + thresh = MULT16_32_Q15(QCONST16(0.5f, 15), MIN32(32767,thresh32)); + { + opus_val32 t; + t = N0<<LM; + shift = celt_ilog2(t)>>1; + t = SHL32(t, (7-shift)<<1); + sqrt_1 = celt_rsqrt_norm(t); + } +#else + thresh = .5f*celt_exp2(-.125f*depth); + sqrt_1 = celt_rsqrt(N0<<LM); +#endif + + c=0; do + { + celt_norm *X; + opus_val16 prev1; + opus_val16 prev2; + opus_val32 Ediff; + opus_val16 r; + int renormalize=0; + prev1 = prev1logE[c*m->nbEBands+i]; + prev2 = prev2logE[c*m->nbEBands+i]; + if (C==1) + { + prev1 = MAX16(prev1,prev1logE[m->nbEBands+i]); + prev2 = MAX16(prev2,prev2logE[m->nbEBands+i]); + } + Ediff = EXTEND32(logE[c*m->nbEBands+i])-EXTEND32(MIN16(prev1,prev2)); + Ediff = MAX32(0, Ediff); + +#ifdef FIXED_POINT + if (Ediff < 16384) + { + opus_val32 r32 = SHR32(celt_exp2(-EXTRACT16(Ediff)),1); + r = 2*MIN16(16383,r32); + } else { + r = 0; + } + if (LM==3) + r = MULT16_16_Q14(23170, MIN32(23169, r)); + r = SHR16(MIN16(thresh, r),1); + r = SHR32(MULT16_16_Q15(sqrt_1, r),shift); +#else + /* r needs to be multiplied by 2 or 2*sqrt(2) depending on LM because + short blocks don't have the same energy as long */ + r = 2.f*celt_exp2(-Ediff); + if (LM==3) + r *= 1.41421356f; + r = MIN16(thresh, r); + r = r*sqrt_1; +#endif + X = X_+c*size+(m->eBands[i]<<LM); + for (k=0;k<1<<LM;k++) + { + /* Detect collapse */ + if (!(collapse_masks[i*C+c]&1<<k)) + { + /* Fill with noise */ + for (j=0;j<N0;j++) + { + seed = celt_lcg_rand(seed); + X[(j<<LM)+k] = (seed&0x8000 ? r : -r); + } + renormalize = 1; + } + } + /* We just added some energy, so we need to renormalise */ + if (renormalize) + renormalise_vector(X, N0<<LM, Q15ONE, arch); + } while (++c<C); + } +} + +/* Compute the weights to use for optimizing normalized distortion across + channels. We use the amplitude to weight square distortion, which means + that we use the square root of the value we would have been using if we + wanted to minimize the MSE in the non-normalized domain. This roughly + corresponds to some quick-and-dirty perceptual experiments I ran to + measure inter-aural masking (there doesn't seem to be any published data + on the topic). */ +static void compute_channel_weights(celt_ener Ex, celt_ener Ey, opus_val16 w[2]) +{ + celt_ener minE; +#ifdef FIXED_POINT + int shift; +#endif + minE = MIN32(Ex, Ey); + /* Adjustment to make the weights a bit more conservative. */ + Ex = ADD32(Ex, minE/3); + Ey = ADD32(Ey, minE/3); +#ifdef FIXED_POINT + shift = celt_ilog2(EPSILON+MAX32(Ex, Ey))-14; +#endif + w[0] = VSHR32(Ex, shift); + w[1] = VSHR32(Ey, shift); +} + +static void intensity_stereo(const CELTMode *m, celt_norm * OPUS_RESTRICT X, const celt_norm * OPUS_RESTRICT Y, const celt_ener *bandE, int bandID, int N) +{ + int i = bandID; + int j; + opus_val16 a1, a2; + opus_val16 left, right; + opus_val16 norm; +#ifdef FIXED_POINT + int shift = celt_zlog2(MAX32(bandE[i], bandE[i+m->nbEBands]))-13; +#endif + left = VSHR32(bandE[i],shift); + right = VSHR32(bandE[i+m->nbEBands],shift); + norm = EPSILON + celt_sqrt(EPSILON+MULT16_16(left,left)+MULT16_16(right,right)); + a1 = DIV32_16(SHL32(EXTEND32(left),14),norm); + a2 = DIV32_16(SHL32(EXTEND32(right),14),norm); + for (j=0;j<N;j++) + { + celt_norm r, l; + l = X[j]; + r = Y[j]; + X[j] = EXTRACT16(SHR32(MAC16_16(MULT16_16(a1, l), a2, r), 14)); + /* Side is not encoded, no need to calculate */ + } +} + +static void stereo_split(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT Y, int N) +{ + int j; + for (j=0;j<N;j++) + { + opus_val32 r, l; + l = MULT16_16(QCONST16(.70710678f, 15), X[j]); + r = MULT16_16(QCONST16(.70710678f, 15), Y[j]); + X[j] = EXTRACT16(SHR32(ADD32(l, r), 15)); + Y[j] = EXTRACT16(SHR32(SUB32(r, l), 15)); + } +} + +static void stereo_merge(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT Y, opus_val16 mid, int N, int arch) +{ + int j; + opus_val32 xp=0, side=0; + opus_val32 El, Er; + opus_val16 mid2; +#ifdef FIXED_POINT + int kl, kr; +#endif + opus_val32 t, lgain, rgain; + + /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */ + dual_inner_prod(Y, X, Y, N, &xp, &side, arch); + /* Compensating for the mid normalization */ + xp = MULT16_32_Q15(mid, xp); + /* mid and side are in Q15, not Q14 like X and Y */ + mid2 = SHR16(mid, 1); + El = MULT16_16(mid2, mid2) + side - 2*xp; + Er = MULT16_16(mid2, mid2) + side + 2*xp; + if (Er < QCONST32(6e-4f, 28) || El < QCONST32(6e-4f, 28)) + { + OPUS_COPY(Y, X, N); + return; + } + +#ifdef FIXED_POINT + kl = celt_ilog2(El)>>1; + kr = celt_ilog2(Er)>>1; +#endif + t = VSHR32(El, (kl-7)<<1); + lgain = celt_rsqrt_norm(t); + t = VSHR32(Er, (kr-7)<<1); + rgain = celt_rsqrt_norm(t); + +#ifdef FIXED_POINT + if (kl < 7) + kl = 7; + if (kr < 7) + kr = 7; +#endif + + for (j=0;j<N;j++) + { + celt_norm r, l; + /* Apply mid scaling (side is already scaled) */ + l = MULT16_16_P15(mid, X[j]); + r = Y[j]; + X[j] = EXTRACT16(PSHR32(MULT16_16(lgain, SUB16(l,r)), kl+1)); + Y[j] = EXTRACT16(PSHR32(MULT16_16(rgain, ADD16(l,r)), kr+1)); + } +} + +/* Decide whether we should spread the pulses in the current frame */ +int spreading_decision(const CELTMode *m, const celt_norm *X, int *average, + int last_decision, int *hf_average, int *tapset_decision, int update_hf, + int end, int C, int M, const int *spread_weight) +{ + int i, c, N0; + int sum = 0, nbBands=0; + const opus_int16 * OPUS_RESTRICT eBands = m->eBands; + int decision; + int hf_sum=0; + + celt_assert(end>0); + + N0 = M*m->shortMdctSize; + + if (M*(eBands[end]-eBands[end-1]) <= 8) + return SPREAD_NONE; + c=0; do { + for (i=0;i<end;i++) + { + int j, N, tmp=0; + int tcount[3] = {0,0,0}; + const celt_norm * OPUS_RESTRICT x = X+M*eBands[i]+c*N0; + N = M*(eBands[i+1]-eBands[i]); + if (N<=8) + continue; + /* Compute rough CDF of |x[j]| */ + for (j=0;j<N;j++) + { + opus_val32 x2N; /* Q13 */ + + x2N = MULT16_16(MULT16_16_Q15(x[j], x[j]), N); + if (x2N < QCONST16(0.25f,13)) + tcount[0]++; + if (x2N < QCONST16(0.0625f,13)) + tcount[1]++; + if (x2N < QCONST16(0.015625f,13)) + tcount[2]++; + } + + /* Only include four last bands (8 kHz and up) */ + if (i>m->nbEBands-4) + hf_sum += celt_udiv(32*(tcount[1]+tcount[0]), N); + tmp = (2*tcount[2] >= N) + (2*tcount[1] >= N) + (2*tcount[0] >= N); + sum += tmp*spread_weight[i]; + nbBands+=spread_weight[i]; + } + } while (++c<C); + + if (update_hf) + { + if (hf_sum) + hf_sum = celt_udiv(hf_sum, C*(4-m->nbEBands+end)); + *hf_average = (*hf_average+hf_sum)>>1; + hf_sum = *hf_average; + if (*tapset_decision==2) + hf_sum += 4; + else if (*tapset_decision==0) + hf_sum -= 4; + if (hf_sum > 22) + *tapset_decision=2; + else if (hf_sum > 18) + *tapset_decision=1; + else + *tapset_decision=0; + } + /*printf("%d %d %d\n", hf_sum, *hf_average, *tapset_decision);*/ + celt_assert(nbBands>0); /* end has to be non-zero */ + celt_assert(sum>=0); + sum = celt_udiv((opus_int32)sum<<8, nbBands); + /* Recursive averaging */ + sum = (sum+*average)>>1; + *average = sum; + /* Hysteresis */ + sum = (3*sum + (((3-last_decision)<<7) + 64) + 2)>>2; + if (sum < 80) + { + decision = SPREAD_AGGRESSIVE; + } else if (sum < 256) + { + decision = SPREAD_NORMAL; + } else if (sum < 384) + { + decision = SPREAD_LIGHT; + } else { + decision = SPREAD_NONE; + } +#ifdef FUZZING + decision = rand()&0x3; + *tapset_decision=rand()%3; +#endif + return decision; +} + +/* Indexing table for converting from natural Hadamard to ordery Hadamard + This is essentially a bit-reversed Gray, on top of which we've added + an inversion of the order because we want the DC at the end rather than + the beginning. The lines are for N=2, 4, 8, 16 */ +static const int ordery_table[] = { + 1, 0, + 3, 0, 2, 1, + 7, 0, 4, 3, 6, 1, 5, 2, + 15, 0, 8, 7, 12, 3, 11, 4, 14, 1, 9, 6, 13, 2, 10, 5, +}; + +static void deinterleave_hadamard(celt_norm *X, int N0, int stride, int hadamard) +{ + int i,j; + VARDECL(celt_norm, tmp); + int N; + SAVE_STACK; + N = N0*stride; + ALLOC(tmp, N, celt_norm); + celt_assert(stride>0); + if (hadamard) + { + const int *ordery = ordery_table+stride-2; + for (i=0;i<stride;i++) + { + for (j=0;j<N0;j++) + tmp[ordery[i]*N0+j] = X[j*stride+i]; + } + } else { + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + tmp[i*N0+j] = X[j*stride+i]; + } + OPUS_COPY(X, tmp, N); + RESTORE_STACK; +} + +static void interleave_hadamard(celt_norm *X, int N0, int stride, int hadamard) +{ + int i,j; + VARDECL(celt_norm, tmp); + int N; + SAVE_STACK; + N = N0*stride; + ALLOC(tmp, N, celt_norm); + if (hadamard) + { + const int *ordery = ordery_table+stride-2; + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + tmp[j*stride+i] = X[ordery[i]*N0+j]; + } else { + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + tmp[j*stride+i] = X[i*N0+j]; + } + OPUS_COPY(X, tmp, N); + RESTORE_STACK; +} + +void haar1(celt_norm *X, int N0, int stride) +{ + int i, j; + N0 >>= 1; + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + { + opus_val32 tmp1, tmp2; + tmp1 = MULT16_16(QCONST16(.70710678f,15), X[stride*2*j+i]); + tmp2 = MULT16_16(QCONST16(.70710678f,15), X[stride*(2*j+1)+i]); + X[stride*2*j+i] = EXTRACT16(PSHR32(ADD32(tmp1, tmp2), 15)); + X[stride*(2*j+1)+i] = EXTRACT16(PSHR32(SUB32(tmp1, tmp2), 15)); + } +} + +static int compute_qn(int N, int b, int offset, int pulse_cap, int stereo) +{ + static const opus_int16 exp2_table8[8] = + {16384, 17866, 19483, 21247, 23170, 25267, 27554, 30048}; + int qn, qb; + int N2 = 2*N-1; + if (stereo && N==2) + N2--; + /* The upper limit ensures that in a stereo split with itheta==16384, we'll + always have enough bits left over to code at least one pulse in the + side; otherwise it would collapse, since it doesn't get folded. */ + qb = celt_sudiv(b+N2*offset, N2); + qb = IMIN(b-pulse_cap-(4<<BITRES), qb); + + qb = IMIN(8<<BITRES, qb); + + if (qb<(1<<BITRES>>1)) { + qn = 1; + } else { + qn = exp2_table8[qb&0x7]>>(14-(qb>>BITRES)); + qn = (qn+1)>>1<<1; + } + celt_assert(qn <= 256); + return qn; +} + +struct band_ctx { + int encode; + int resynth; + const CELTMode *m; + int i; + int intensity; + int spread; + int tf_change; + ec_ctx *ec; + opus_int32 remaining_bits; + const celt_ener *bandE; + opus_uint32 seed; + int arch; + int theta_round; + int disable_inv; + int avoid_split_noise; +}; + +struct split_ctx { + int inv; + int imid; + int iside; + int delta; + int itheta; + int qalloc; +}; + +static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx, + celt_norm *X, celt_norm *Y, int N, int *b, int B, int B0, + int LM, + int stereo, int *fill) +{ + int qn; + int itheta=0; + int delta; + int imid, iside; + int qalloc; + int pulse_cap; + int offset; + opus_int32 tell; + int inv=0; + int encode; + const CELTMode *m; + int i; + int intensity; + ec_ctx *ec; + const celt_ener *bandE; + + encode = ctx->encode; + m = ctx->m; + i = ctx->i; + intensity = ctx->intensity; + ec = ctx->ec; + bandE = ctx->bandE; + + /* Decide on the resolution to give to the split parameter theta */ + pulse_cap = m->logN[i]+LM*(1<<BITRES); + offset = (pulse_cap>>1) - (stereo&&N==2 ? QTHETA_OFFSET_TWOPHASE : QTHETA_OFFSET); + qn = compute_qn(N, *b, offset, pulse_cap, stereo); + if (stereo && i>=intensity) + qn = 1; + if (encode) + { + /* theta is the atan() of the ratio between the (normalized) + side and mid. With just that parameter, we can re-scale both + mid and side because we know that 1) they have unit norm and + 2) they are orthogonal. */ + itheta = stereo_itheta(X, Y, stereo, N, ctx->arch); + } + tell = ec_tell_frac(ec); + if (qn!=1) + { + if (encode) + { + if (!stereo || ctx->theta_round == 0) + { + itheta = (itheta*(opus_int32)qn+8192)>>14; + if (!stereo && ctx->avoid_split_noise && itheta > 0 && itheta < qn) + { + /* Check if the selected value of theta will cause the bit allocation + to inject noise on one side. If so, make sure the energy of that side + is zero. */ + int unquantized = celt_udiv((opus_int32)itheta*16384, qn); + imid = bitexact_cos((opus_int16)unquantized); + iside = bitexact_cos((opus_int16)(16384-unquantized)); + delta = FRAC_MUL16((N-1)<<7,bitexact_log2tan(iside,imid)); + if (delta > *b) + itheta = qn; + else if (delta < -*b) + itheta = 0; + } + } else { + int down; + /* Bias quantization towards itheta=0 and itheta=16384. */ + int bias = itheta > 8192 ? 32767/qn : -32767/qn; + down = IMIN(qn-1, IMAX(0, (itheta*(opus_int32)qn + bias)>>14)); + if (ctx->theta_round < 0) + itheta = down; + else + itheta = down+1; + } + } + /* Entropy coding of the angle. We use a uniform pdf for the + time split, a step for stereo, and a triangular one for the rest. */ + if (stereo && N>2) + { + int p0 = 3; + int x = itheta; + int x0 = qn/2; + int ft = p0*(x0+1) + x0; + /* Use a probability of p0 up to itheta=8192 and then use 1 after */ + if (encode) + { + ec_encode(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); + } else { + int fs; + fs=ec_decode(ec,ft); + if (fs<(x0+1)*p0) + x=fs/p0; + else + x=x0+1+(fs-(x0+1)*p0); + ec_dec_update(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); + itheta = x; + } + } else if (B0>1 || stereo) { + /* Uniform pdf */ + if (encode) + ec_enc_uint(ec, itheta, qn+1); + else + itheta = ec_dec_uint(ec, qn+1); + } else { + int fs=1, ft; + ft = ((qn>>1)+1)*((qn>>1)+1); + if (encode) + { + int fl; + + fs = itheta <= (qn>>1) ? itheta + 1 : qn + 1 - itheta; + fl = itheta <= (qn>>1) ? itheta*(itheta + 1)>>1 : + ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); + + ec_encode(ec, fl, fl+fs, ft); + } else { + /* Triangular pdf */ + int fl=0; + int fm; + fm = ec_decode(ec, ft); + + if (fm < ((qn>>1)*((qn>>1) + 1)>>1)) + { + itheta = (isqrt32(8*(opus_uint32)fm + 1) - 1)>>1; + fs = itheta + 1; + fl = itheta*(itheta + 1)>>1; + } + else + { + itheta = (2*(qn + 1) + - isqrt32(8*(opus_uint32)(ft - fm - 1) + 1))>>1; + fs = qn + 1 - itheta; + fl = ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); + } + + ec_dec_update(ec, fl, fl+fs, ft); + } + } + celt_assert(itheta>=0); + itheta = celt_udiv((opus_int32)itheta*16384, qn); + if (encode && stereo) + { + if (itheta==0) + intensity_stereo(m, X, Y, bandE, i, N); + else + stereo_split(X, Y, N); + } + /* NOTE: Renormalising X and Y *may* help fixed-point a bit at very high rate. + Let's do that at higher complexity */ + } else if (stereo) { + if (encode) + { + inv = itheta > 8192 && !ctx->disable_inv; + if (inv) + { + int j; + for (j=0;j<N;j++) + Y[j] = -Y[j]; + } + intensity_stereo(m, X, Y, bandE, i, N); + } + if (*b>2<<BITRES && ctx->remaining_bits > 2<<BITRES) + { + if (encode) + ec_enc_bit_logp(ec, inv, 2); + else + inv = ec_dec_bit_logp(ec, 2); + } else + inv = 0; + /* inv flag override to avoid problems with downmixing. */ + if (ctx->disable_inv) + inv = 0; + itheta = 0; + } + qalloc = ec_tell_frac(ec) - tell; + *b -= qalloc; + + if (itheta == 0) + { + imid = 32767; + iside = 0; + *fill &= (1<<B)-1; + delta = -16384; + } else if (itheta == 16384) + { + imid = 0; + iside = 32767; + *fill &= ((1<<B)-1)<<B; + delta = 16384; + } else { + imid = bitexact_cos((opus_int16)itheta); + iside = bitexact_cos((opus_int16)(16384-itheta)); + /* This is the mid vs side allocation that minimizes squared error + in that band. */ + delta = FRAC_MUL16((N-1)<<7,bitexact_log2tan(iside,imid)); + } + + sctx->inv = inv; + sctx->imid = imid; + sctx->iside = iside; + sctx->delta = delta; + sctx->itheta = itheta; + sctx->qalloc = qalloc; +} +static unsigned quant_band_n1(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, int b, + celt_norm *lowband_out) +{ + (void)b; + int c; + int stereo; + celt_norm *x = X; + int encode; + ec_ctx *ec; + + encode = ctx->encode; + ec = ctx->ec; + + stereo = Y != NULL; + c=0; do { + int sign=0; + if (ctx->remaining_bits>=1<<BITRES) + { + if (encode) + { + sign = x[0]<0; + ec_enc_bits(ec, sign, 1); + } else { + sign = ec_dec_bits(ec, 1); + } + ctx->remaining_bits -= 1<<BITRES; + b-=1<<BITRES; + } + if (ctx->resynth) + x[0] = sign ? -NORM_SCALING : NORM_SCALING; + x = Y; + } while (++c<1+stereo); + if (lowband_out) + lowband_out[0] = SHR16(X[0],4); + return 1; +} + +/* This function is responsible for encoding and decoding a mono partition. + It can split the band in two and transmit the energy difference with + the two half-bands. It can be called recursively so bands can end up being + split in 8 parts. */ +static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X, + int N, int b, int B, celt_norm *lowband, + int LM, + opus_val16 gain, int fill) +{ + const unsigned char *cache; + int q; + int curr_bits; + int imid=0, iside=0; + int B0=B; + opus_val16 mid=0, side=0; + unsigned cm=0; + celt_norm *Y=NULL; + int encode; + const CELTMode *m; + int i; + int spread; + ec_ctx *ec; + + encode = ctx->encode; + m = ctx->m; + i = ctx->i; + spread = ctx->spread; + ec = ctx->ec; + + /* If we need 1.5 more bit than we can produce, split the band in two. */ + cache = m->cache.bits + m->cache.index[(LM+1)*m->nbEBands+i]; + if (LM != -1 && b > cache[cache[0]]+12 && N>2) + { + int mbits, sbits, delta; + int itheta; + int qalloc; + struct split_ctx sctx; + celt_norm *next_lowband2=NULL; + opus_int32 rebalance; + + N >>= 1; + Y = X+N; + LM -= 1; + if (B==1) + fill = (fill&1)|(fill<<1); + B = (B+1)>>1; + + compute_theta(ctx, &sctx, X, Y, N, &b, B, B0, LM, 0, &fill); + imid = sctx.imid; + iside = sctx.iside; + delta = sctx.delta; + itheta = sctx.itheta; + qalloc = sctx.qalloc; +#ifdef FIXED_POINT + mid = imid; + side = iside; +#else + mid = (1.f/32768)*imid; + side = (1.f/32768)*iside; +#endif + + /* Give more bits to low-energy MDCTs than they would otherwise deserve */ + if (B0>1 && (itheta&0x3fff)) + { + if (itheta > 8192) + /* Rough approximation for pre-echo masking */ + delta -= delta>>(4-LM); + else + /* Corresponds to a forward-masking slope of 1.5 dB per 10 ms */ + delta = IMIN(0, delta + (N<<BITRES>>(5-LM))); + } + mbits = IMAX(0, IMIN(b, (b-delta)/2)); + sbits = b-mbits; + ctx->remaining_bits -= qalloc; + + if (lowband) + next_lowband2 = lowband+N; /* >32-bit split case */ + + rebalance = ctx->remaining_bits; + if (mbits >= sbits) + { + cm = quant_partition(ctx, X, N, mbits, B, lowband, LM, + MULT16_16_P15(gain,mid), fill); + rebalance = mbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=0) + sbits += rebalance - (3<<BITRES); + cm |= quant_partition(ctx, Y, N, sbits, B, next_lowband2, LM, + MULT16_16_P15(gain,side), fill>>B)<<(B0>>1); + } else { + cm = quant_partition(ctx, Y, N, sbits, B, next_lowband2, LM, + MULT16_16_P15(gain,side), fill>>B)<<(B0>>1); + rebalance = sbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=16384) + mbits += rebalance - (3<<BITRES); + cm |= quant_partition(ctx, X, N, mbits, B, lowband, LM, + MULT16_16_P15(gain,mid), fill); + } + } else { + /* This is the basic no-split case */ + q = bits2pulses(m, i, LM, b); + curr_bits = pulses2bits(m, i, LM, q); + ctx->remaining_bits -= curr_bits; + + /* Ensures we can never bust the budget */ + while (ctx->remaining_bits < 0 && q > 0) + { + ctx->remaining_bits += curr_bits; + q--; + curr_bits = pulses2bits(m, i, LM, q); + ctx->remaining_bits -= curr_bits; + } + + if (q!=0) + { + int K = get_pulses(q); + + /* Finally do the actual quantization */ + if (encode) + { + cm = alg_quant(X, N, K, spread, B, ec, gain, ctx->resynth, ctx->arch); + } else { + cm = alg_unquant(X, N, K, spread, B, ec, gain); + } + } else { + /* If there's no pulse, fill the band anyway */ + int j; + if (ctx->resynth) + { + unsigned cm_mask; + /* B can be as large as 16, so this shift might overflow an int on a + 16-bit platform; use a long to get defined behavior.*/ + cm_mask = (unsigned)(1UL<<B)-1; + fill &= cm_mask; + if (!fill) + { + OPUS_CLEAR(X, N); + } else { + if (lowband == NULL) + { + /* Noise */ + for (j=0;j<N;j++) + { + ctx->seed = celt_lcg_rand(ctx->seed); + X[j] = (celt_norm)((opus_int32)ctx->seed>>20); + } + cm = cm_mask; + } else { + /* Folded spectrum */ + for (j=0;j<N;j++) + { + opus_val16 tmp; + ctx->seed = celt_lcg_rand(ctx->seed); + /* About 48 dB below the "normal" folding level */ + tmp = QCONST16(1.0f/256, 10); + tmp = (ctx->seed)&0x8000 ? tmp : -tmp; + X[j] = lowband[j]+tmp; + } + cm = fill; + } + renormalise_vector(X, N, gain, ctx->arch); + } + } + } + } + + return cm; +} + + +/* This function is responsible for encoding and decoding a band for the mono case. */ +static unsigned quant_band(struct band_ctx *ctx, celt_norm *X, + int N, int b, int B, celt_norm *lowband, + int LM, celt_norm *lowband_out, + opus_val16 gain, celt_norm *lowband_scratch, int fill) +{ + int N0=N; + int N_B=N; + int N_B0; + int B0=B; + int time_divide=0; + int recombine=0; + int longBlocks; + unsigned cm=0; + int k; + int encode; + int tf_change; + + encode = ctx->encode; + tf_change = ctx->tf_change; + + longBlocks = B0==1; + + N_B = celt_udiv(N_B, B); + + /* Special case for one sample */ + if (N==1) + { + return quant_band_n1(ctx, X, NULL, b, lowband_out); + } + + if (tf_change>0) + recombine = tf_change; + /* Band recombining to increase frequency resolution */ + + if (lowband_scratch && lowband && (recombine || ((N_B&1) == 0 && tf_change<0) || B0>1)) + { + OPUS_COPY(lowband_scratch, lowband, N); + lowband = lowband_scratch; + } + + for (k=0;k<recombine;k++) + { + static const unsigned char bit_interleave_table[16]={ + 0,1,1,1,2,3,3,3,2,3,3,3,2,3,3,3 + }; + if (encode) + haar1(X, N>>k, 1<<k); + if (lowband) + haar1(lowband, N>>k, 1<<k); + fill = bit_interleave_table[fill&0xF]|bit_interleave_table[fill>>4]<<2; + } + B>>=recombine; + N_B<<=recombine; + + /* Increasing the time resolution */ + while ((N_B&1) == 0 && tf_change<0) + { + if (encode) + haar1(X, N_B, B); + if (lowband) + haar1(lowband, N_B, B); + fill |= fill<<B; + B <<= 1; + N_B >>= 1; + time_divide++; + tf_change++; + } + B0=B; + N_B0 = N_B; + + /* Reorganize the samples in time order instead of frequency order */ + if (B0>1) + { + if (encode) + deinterleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); + if (lowband) + deinterleave_hadamard(lowband, N_B>>recombine, B0<<recombine, longBlocks); + } + + cm = quant_partition(ctx, X, N, b, B, lowband, LM, gain, fill); + + /* This code is used by the decoder and by the resynthesis-enabled encoder */ + if (ctx->resynth) + { + /* Undo the sample reorganization going from time order to frequency order */ + if (B0>1) + interleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); + + /* Undo time-freq changes that we did earlier */ + N_B = N_B0; + B = B0; + for (k=0;k<time_divide;k++) + { + B >>= 1; + N_B <<= 1; + cm |= cm>>B; + haar1(X, N_B, B); + } + + for (k=0;k<recombine;k++) + { + static const unsigned char bit_deinterleave_table[16]={ + 0x00,0x03,0x0C,0x0F,0x30,0x33,0x3C,0x3F, + 0xC0,0xC3,0xCC,0xCF,0xF0,0xF3,0xFC,0xFF + }; + cm = bit_deinterleave_table[cm]; + haar1(X, N0>>k, 1<<k); + } + B<<=recombine; + + /* Scale output for later folding */ + if (lowband_out) + { + int j; + opus_val16 n; + n = celt_sqrt(SHL32(EXTEND32(N0),22)); + for (j=0;j<N0;j++) + lowband_out[j] = MULT16_16_Q15(n,X[j]); + } + cm &= (1<<B)-1; + } + return cm; +} + + +/* This function is responsible for encoding and decoding a band for the stereo case. */ +static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, + int N, int b, int B, celt_norm *lowband, + int LM, celt_norm *lowband_out, + celt_norm *lowband_scratch, int fill) +{ + int imid=0, iside=0; + int inv = 0; + opus_val16 mid=0, side=0; + unsigned cm=0; + int mbits, sbits, delta; + int itheta; + int qalloc; + struct split_ctx sctx; + int orig_fill; + int encode; + ec_ctx *ec; + + encode = ctx->encode; + ec = ctx->ec; + + /* Special case for one sample */ + if (N==1) + { + return quant_band_n1(ctx, X, Y, b, lowband_out); + } + + orig_fill = fill; + + compute_theta(ctx, &sctx, X, Y, N, &b, B, B, LM, 1, &fill); + inv = sctx.inv; + imid = sctx.imid; + iside = sctx.iside; + delta = sctx.delta; + itheta = sctx.itheta; + qalloc = sctx.qalloc; +#ifdef FIXED_POINT + mid = imid; + side = iside; +#else + mid = (1.f/32768)*imid; + side = (1.f/32768)*iside; +#endif + + /* This is a special case for N=2 that only works for stereo and takes + advantage of the fact that mid and side are orthogonal to encode + the side with just one bit. */ + if (N==2) + { + int c; + int sign=0; + celt_norm *x2, *y2; + mbits = b; + sbits = 0; + /* Only need one bit for the side. */ + if (itheta != 0 && itheta != 16384) + sbits = 1<<BITRES; + mbits -= sbits; + c = itheta > 8192; + ctx->remaining_bits -= qalloc+sbits; + + x2 = c ? Y : X; + y2 = c ? X : Y; + if (sbits) + { + if (encode) + { + /* Here we only need to encode a sign for the side. */ + sign = x2[0]*y2[1] - x2[1]*y2[0] < 0; + ec_enc_bits(ec, sign, 1); + } else { + sign = ec_dec_bits(ec, 1); + } + } + sign = 1-2*sign; + /* We use orig_fill here because we want to fold the side, but if + itheta==16384, we'll have cleared the low bits of fill. */ + cm = quant_band(ctx, x2, N, mbits, B, lowband, LM, lowband_out, Q15ONE, + lowband_scratch, orig_fill); + /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse), + and there's no need to worry about mixing with the other channel. */ + y2[0] = -sign*x2[1]; + y2[1] = sign*x2[0]; + if (ctx->resynth) + { + celt_norm tmp; + X[0] = MULT16_16_Q15(mid, X[0]); + X[1] = MULT16_16_Q15(mid, X[1]); + Y[0] = MULT16_16_Q15(side, Y[0]); + Y[1] = MULT16_16_Q15(side, Y[1]); + tmp = X[0]; + X[0] = SUB16(tmp,Y[0]); + Y[0] = ADD16(tmp,Y[0]); + tmp = X[1]; + X[1] = SUB16(tmp,Y[1]); + Y[1] = ADD16(tmp,Y[1]); + } + } else { + /* "Normal" split code */ + opus_int32 rebalance; + + mbits = IMAX(0, IMIN(b, (b-delta)/2)); + sbits = b-mbits; + ctx->remaining_bits -= qalloc; + + rebalance = ctx->remaining_bits; + if (mbits >= sbits) + { + /* In stereo mode, we do not apply a scaling to the mid because we need the normalized + mid for folding later. */ + cm = quant_band(ctx, X, N, mbits, B, lowband, LM, lowband_out, Q15ONE, + lowband_scratch, fill); + rebalance = mbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=0) + sbits += rebalance - (3<<BITRES); + + /* For a stereo split, the high bits of fill are always zero, so no + folding will be done to the side. */ + cm |= quant_band(ctx, Y, N, sbits, B, NULL, LM, NULL, side, NULL, fill>>B); + } else { + /* For a stereo split, the high bits of fill are always zero, so no + folding will be done to the side. */ + cm = quant_band(ctx, Y, N, sbits, B, NULL, LM, NULL, side, NULL, fill>>B); + rebalance = sbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=16384) + mbits += rebalance - (3<<BITRES); + /* In stereo mode, we do not apply a scaling to the mid because we need the normalized + mid for folding later. */ + cm |= quant_band(ctx, X, N, mbits, B, lowband, LM, lowband_out, Q15ONE, + lowband_scratch, fill); + } + } + + + /* This code is used by the decoder and by the resynthesis-enabled encoder */ + if (ctx->resynth) + { + if (N!=2) + stereo_merge(X, Y, mid, N, ctx->arch); + if (inv) + { + int j; + for (j=0;j<N;j++) + Y[j] = -Y[j]; + } + } + return cm; +} + +static void special_hybrid_folding(const CELTMode *m, celt_norm *norm, celt_norm *norm2, int start, int M, int dual_stereo) +{ + int n1, n2; + const opus_int16 * OPUS_RESTRICT eBands = m->eBands; + n1 = M*(eBands[start+1]-eBands[start]); + n2 = M*(eBands[start+2]-eBands[start+1]); + /* Duplicate enough of the first band folding data to be able to fold the second band. + Copies no data for CELT-only mode. */ + OPUS_COPY(&norm[n1], &norm[2*n1 - n2], n2-n1); + if (dual_stereo) + OPUS_COPY(&norm2[n1], &norm2[2*n1 - n2], n2-n1); +} + +void quant_all_bands(int encode, const CELTMode *m, int start, int end, + celt_norm *X_, celt_norm *Y_, unsigned char *collapse_masks, + const celt_ener *bandE, int *pulses, int shortBlocks, int spread, + int dual_stereo, int intensity, int *tf_res, opus_int32 total_bits, + opus_int32 balance, ec_ctx *ec, int LM, int codedBands, + opus_uint32 *seed, int complexity, int arch, int disable_inv) +{ + int i; + opus_int32 remaining_bits; + const opus_int16 * OPUS_RESTRICT eBands = m->eBands; + celt_norm * OPUS_RESTRICT norm, * OPUS_RESTRICT norm2; + VARDECL(celt_norm, _norm); + VARDECL(celt_norm, _lowband_scratch); + VARDECL(celt_norm, X_save); + VARDECL(celt_norm, Y_save); + VARDECL(celt_norm, X_save2); + VARDECL(celt_norm, Y_save2); + VARDECL(celt_norm, norm_save2); + int resynth_alloc; + celt_norm *lowband_scratch; + int B; + int M; + int lowband_offset; + int update_lowband = 1; + int C = Y_ != NULL ? 2 : 1; + int norm_offset; + int theta_rdo = encode && Y_!=NULL && !dual_stereo && complexity>=8; +#ifdef RESYNTH + int resynth = 1; +#else + int resynth = !encode || theta_rdo; +#endif + struct band_ctx ctx; + SAVE_STACK; + + M = 1<<LM; + B = shortBlocks ? M : 1; + norm_offset = M*eBands[start]; + /* No need to allocate norm for the last band because we don't need an + output in that band. */ + ALLOC(_norm, C*(M*eBands[m->nbEBands-1]-norm_offset), celt_norm); + norm = _norm; + norm2 = norm + M*eBands[m->nbEBands-1]-norm_offset; + + /* For decoding, we can use the last band as scratch space because we don't need that + scratch space for the last band and we don't care about the data there until we're + decoding the last band. */ + if (encode && resynth) + resynth_alloc = M*(eBands[m->nbEBands]-eBands[m->nbEBands-1]); + else + resynth_alloc = ALLOC_NONE; + ALLOC(_lowband_scratch, resynth_alloc, celt_norm); + if (encode && resynth) + lowband_scratch = _lowband_scratch; + else + lowband_scratch = X_+M*eBands[m->nbEBands-1]; + ALLOC(X_save, resynth_alloc, celt_norm); + ALLOC(Y_save, resynth_alloc, celt_norm); + ALLOC(X_save2, resynth_alloc, celt_norm); + ALLOC(Y_save2, resynth_alloc, celt_norm); + ALLOC(norm_save2, resynth_alloc, celt_norm); + + lowband_offset = 0; + ctx.bandE = bandE; + ctx.ec = ec; + ctx.encode = encode; + ctx.intensity = intensity; + ctx.m = m; + ctx.seed = *seed; + ctx.spread = spread; + ctx.arch = arch; + ctx.disable_inv = disable_inv; + ctx.resynth = resynth; + ctx.theta_round = 0; + /* Avoid injecting noise in the first band on transients. */ + ctx.avoid_split_noise = B > 1; + for (i=start;i<end;i++) + { + opus_int32 tell; + int b; + int N; + opus_int32 curr_balance; + int effective_lowband=-1; + celt_norm * OPUS_RESTRICT X, * OPUS_RESTRICT Y; + int tf_change=0; + unsigned x_cm; + unsigned y_cm; + int last; + + ctx.i = i; + last = (i==end-1); + + X = X_+M*eBands[i]; + if (Y_!=NULL) + Y = Y_+M*eBands[i]; + else + Y = NULL; + N = M*eBands[i+1]-M*eBands[i]; + celt_assert(N > 0); + tell = ec_tell_frac(ec); + + /* Compute how many bits we want to allocate to this band */ + if (i != start) + balance -= tell; + remaining_bits = total_bits-tell-1; + ctx.remaining_bits = remaining_bits; + if (i <= codedBands-1) + { + curr_balance = celt_sudiv(balance, IMIN(3, codedBands-i)); + b = IMAX(0, IMIN(16383, IMIN(remaining_bits+1,pulses[i]+curr_balance))); + } else { + b = 0; + } + +#ifndef DISABLE_UPDATE_DRAFT + if (resynth && (M*eBands[i]-N >= M*eBands[start] || i==start+1) && (update_lowband || lowband_offset==0)) + lowband_offset = i; + if (i == start+1) + special_hybrid_folding(m, norm, norm2, start, M, dual_stereo); +#else + if (resynth && M*eBands[i]-N >= M*eBands[start] && (update_lowband || lowband_offset==0)) + lowband_offset = i; +#endif + + tf_change = tf_res[i]; + ctx.tf_change = tf_change; + if (i>=m->effEBands) + { + X=norm; + if (Y_!=NULL) + Y = norm; + lowband_scratch = NULL; + } + if (last && !theta_rdo) + lowband_scratch = NULL; + + /* Get a conservative estimate of the collapse_mask's for the bands we're + going to be folding from. */ + if (lowband_offset != 0 && (spread!=SPREAD_AGGRESSIVE || B>1 || tf_change<0)) + { + int fold_start; + int fold_end; + int fold_i; + /* This ensures we never repeat spectral content within one band */ + effective_lowband = IMAX(0, M*eBands[lowband_offset]-norm_offset-N); + fold_start = lowband_offset; + while(M*eBands[--fold_start] > effective_lowband+norm_offset); + fold_end = lowband_offset-1; +#ifndef DISABLE_UPDATE_DRAFT + while(++fold_end < i && M*eBands[fold_end] < effective_lowband+norm_offset+N); +#else + while(M*eBands[++fold_end] < effective_lowband+norm_offset+N); +#endif + x_cm = y_cm = 0; + fold_i = fold_start; do { + x_cm |= collapse_masks[fold_i*C+0]; + y_cm |= collapse_masks[fold_i*C+C-1]; + } while (++fold_i<fold_end); + } + /* Otherwise, we'll be using the LCG to fold, so all blocks will (almost + always) be non-zero. */ + else + x_cm = y_cm = (1<<B)-1; + + if (dual_stereo && i==intensity) + { + int j; + + /* Switch off dual stereo to do intensity. */ + dual_stereo = 0; + if (resynth) + for (j=0;j<M*eBands[i]-norm_offset;j++) + norm[j] = HALF32(norm[j]+norm2[j]); + } + if (dual_stereo) + { + x_cm = quant_band(&ctx, X, N, b/2, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, x_cm); + y_cm = quant_band(&ctx, Y, N, b/2, B, + effective_lowband != -1 ? norm2+effective_lowband : NULL, LM, + last?NULL:norm2+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, y_cm); + } else { + if (Y!=NULL) + { + if (theta_rdo && i < intensity) + { + ec_ctx ec_save, ec_save2; + struct band_ctx ctx_save, ctx_save2; + opus_val32 dist0, dist1; + unsigned cm, cm2; + int nstart_bytes, nend_bytes, save_bytes; + unsigned char *bytes_buf; + unsigned char bytes_save[1275]; + opus_val16 w[2]; + compute_channel_weights(bandE[i], bandE[i+m->nbEBands], w); + /* Make a copy. */ + cm = x_cm|y_cm; + ec_save = *ec; + ctx_save = ctx; + OPUS_COPY(X_save, X, N); + OPUS_COPY(Y_save, Y, N); + /* Encode and round down. */ + ctx.theta_round = -1; + x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, lowband_scratch, cm); + dist0 = MULT16_32_Q15(w[0], celt_inner_prod(X_save, X, N, arch)) + MULT16_32_Q15(w[1], celt_inner_prod(Y_save, Y, N, arch)); + + /* Save first result. */ + cm2 = x_cm; + ec_save2 = *ec; + ctx_save2 = ctx; + OPUS_COPY(X_save2, X, N); + OPUS_COPY(Y_save2, Y, N); + if (!last) + OPUS_COPY(norm_save2, norm+M*eBands[i]-norm_offset, N); + nstart_bytes = ec_save.offs; + nend_bytes = ec_save.storage; + bytes_buf = ec_save.buf+nstart_bytes; + save_bytes = nend_bytes-nstart_bytes; + OPUS_COPY(bytes_save, bytes_buf, save_bytes); + + /* Restore */ + *ec = ec_save; + ctx = ctx_save; + OPUS_COPY(X, X_save, N); + OPUS_COPY(Y, Y_save, N); +#ifndef DISABLE_UPDATE_DRAFT + if (i == start+1) + special_hybrid_folding(m, norm, norm2, start, M, dual_stereo); +#endif + /* Encode and round up. */ + ctx.theta_round = 1; + x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, lowband_scratch, cm); + dist1 = MULT16_32_Q15(w[0], celt_inner_prod(X_save, X, N, arch)) + MULT16_32_Q15(w[1], celt_inner_prod(Y_save, Y, N, arch)); + if (dist0 >= dist1) { + x_cm = cm2; + *ec = ec_save2; + ctx = ctx_save2; + OPUS_COPY(X, X_save2, N); + OPUS_COPY(Y, Y_save2, N); + if (!last) + OPUS_COPY(norm+M*eBands[i]-norm_offset, norm_save2, N); + OPUS_COPY(bytes_buf, bytes_save, save_bytes); + } + } else { + ctx.theta_round = 0; + x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, lowband_scratch, x_cm|y_cm); + } + } else { + x_cm = quant_band(&ctx, X, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, x_cm|y_cm); + } + y_cm = x_cm; + } + collapse_masks[i*C+0] = (unsigned char)x_cm; + collapse_masks[i*C+C-1] = (unsigned char)y_cm; + balance += pulses[i] + tell; + + /* Update the folding position only as long as we have 1 bit/sample depth. */ + update_lowband = b>(N<<BITRES); + /* We only need to avoid noise on a split for the first band. After that, we + have folding. */ + ctx.avoid_split_noise = 0; + } + *seed = ctx.seed; + + RESTORE_STACK; +} + diff --git a/libopus/celt/bands.h b/libopus/celt/bands.h new file mode 100644 index 0000000..422b32c --- /dev/null +++ b/libopus/celt/bands.h @@ -0,0 +1,123 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008-2009 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef BANDS_H +#define BANDS_H + +#include "arch.h" +#include "modes.h" +#include "entenc.h" +#include "entdec.h" +#include "rate.h" + +opus_int16 bitexact_cos(opus_int16 x); +int bitexact_log2tan(int isin,int icos); + +/** Compute the amplitude (sqrt energy) in each of the bands + * @param m Mode data + * @param X Spectrum + * @param bandE Square root of the energy for each band (returned) + */ +void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch); + +/*void compute_noise_energies(const CELTMode *m, const celt_sig *X, const opus_val16 *tonality, celt_ener *bandE);*/ + +/** Normalise each band of X such that the energy in each band is + equal to 1 + * @param m Mode data + * @param X Spectrum (returned normalised) + * @param bandE Square root of the energy for each band + */ +void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M); + +/** Denormalise each band of X to restore full amplitude + * @param m Mode data + * @param X Spectrum (returned de-normalised) + * @param bandE Square root of the energy for each band + */ +void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, + celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandE, int start, + int end, int M, int downsample, int silence); + +#define SPREAD_NONE (0) +#define SPREAD_LIGHT (1) +#define SPREAD_NORMAL (2) +#define SPREAD_AGGRESSIVE (3) + +int spreading_decision(const CELTMode *m, const celt_norm *X, int *average, + int last_decision, int *hf_average, int *tapset_decision, int update_hf, + int end, int C, int M, const int *spread_weight); + +#ifdef MEASURE_NORM_MSE +void measure_norm_mse(const CELTMode *m, float *X, float *X0, float *bandE, float *bandE0, int M, int N, int C); +#endif + +void haar1(celt_norm *X, int N0, int stride); + +/** Quantisation/encoding of the residual spectrum + * @param encode flag that indicates whether we're encoding (1) or decoding (0) + * @param m Mode data + * @param start First band to process + * @param end Last band to process + 1 + * @param X Residual (normalised) + * @param Y Residual (normalised) for second channel (or NULL for mono) + * @param collapse_masks Anti-collapse tracking mask + * @param bandE Square root of the energy for each band + * @param pulses Bit allocation (per band) for PVQ + * @param shortBlocks Zero for long blocks, non-zero for short blocks + * @param spread Amount of spreading to use + * @param dual_stereo Zero for MS stereo, non-zero for dual stereo + * @param intensity First band to use intensity stereo + * @param tf_res Time-frequency resolution change + * @param total_bits Total number of bits that can be used for the frame (including the ones already spent) + * @param balance Number of unallocated bits + * @param en Entropy coder state + * @param LM log2() of the number of 2.5 subframes in the frame + * @param codedBands Last band to receive bits + 1 + * @param seed Random generator seed + * @param arch Run-time architecture (see opus_select_arch()) + */ +void quant_all_bands(int encode, const CELTMode *m, int start, int end, + celt_norm * X, celt_norm * Y, unsigned char *collapse_masks, + const celt_ener *bandE, int *pulses, int shortBlocks, int spread, + int dual_stereo, int intensity, int *tf_res, opus_int32 total_bits, + opus_int32 balance, ec_ctx *ec, int M, int codedBands, opus_uint32 *seed, + int complexity, int arch, int disable_inv); + +void anti_collapse(const CELTMode *m, celt_norm *X_, + unsigned char *collapse_masks, int LM, int C, int size, int start, + int end, const opus_val16 *logE, const opus_val16 *prev1logE, + const opus_val16 *prev2logE, const int *pulses, opus_uint32 seed, + int arch); + +opus_uint32 celt_lcg_rand(opus_uint32 seed); + +int hysteresis_decision(opus_val16 val, const opus_val16 *thresholds, const opus_val16 *hysteresis, int N, int prev); + +#endif /* BANDS_H */ diff --git a/libopus/celt/celt.c b/libopus/celt/celt.c new file mode 100644 index 0000000..9ce2346 --- /dev/null +++ b/libopus/celt/celt.c @@ -0,0 +1,316 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_C + +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + +#ifndef PACKAGE_VERSION +#define PACKAGE_VERSION "unknown" +#endif + +#if defined(MIPSr1_ASM) +#include "mips/celt_mipsr1.h" +#endif + + +int resampling_factor(opus_int32 rate) +{ + int ret; + switch (rate) + { + case 48000: + ret = 1; + break; + case 24000: + ret = 2; + break; + case 16000: + ret = 3; + break; + case 12000: + ret = 4; + break; + case 8000: + ret = 6; + break; + default: +#ifndef CUSTOM_MODES + celt_assert(0); +#endif + ret = 0; + break; + } + return ret; +} + +#if !defined(OVERRIDE_COMB_FILTER_CONST) || defined(NON_STATIC_COMB_FILTER_CONST_C) +/* This version should be faster on ARM */ +#ifdef OPUS_ARM_ASM +#ifndef NON_STATIC_COMB_FILTER_CONST_C +static +#endif +void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12) +{ + opus_val32 x0, x1, x2, x3, x4; + int i; + x4 = SHL32(x[-T-2], 1); + x3 = SHL32(x[-T-1], 1); + x2 = SHL32(x[-T], 1); + x1 = SHL32(x[-T+1], 1); + for (i=0;i<N-4;i+=5) + { + opus_val32 t; + x0=SHL32(x[i-T+2],1); + t = MAC16_32_Q16(x[i], g10, x2); + t = MAC16_32_Q16(t, g11, ADD32(x1,x3)); + t = MAC16_32_Q16(t, g12, ADD32(x0,x4)); + t = SATURATE(t, SIG_SAT); + y[i] = t; + x4=SHL32(x[i-T+3],1); + t = MAC16_32_Q16(x[i+1], g10, x1); + t = MAC16_32_Q16(t, g11, ADD32(x0,x2)); + t = MAC16_32_Q16(t, g12, ADD32(x4,x3)); + t = SATURATE(t, SIG_SAT); + y[i+1] = t; + x3=SHL32(x[i-T+4],1); + t = MAC16_32_Q16(x[i+2], g10, x0); + t = MAC16_32_Q16(t, g11, ADD32(x4,x1)); + t = MAC16_32_Q16(t, g12, ADD32(x3,x2)); + t = SATURATE(t, SIG_SAT); + y[i+2] = t; + x2=SHL32(x[i-T+5],1); + t = MAC16_32_Q16(x[i+3], g10, x4); + t = MAC16_32_Q16(t, g11, ADD32(x3,x0)); + t = MAC16_32_Q16(t, g12, ADD32(x2,x1)); + t = SATURATE(t, SIG_SAT); + y[i+3] = t; + x1=SHL32(x[i-T+6],1); + t = MAC16_32_Q16(x[i+4], g10, x3); + t = MAC16_32_Q16(t, g11, ADD32(x2,x4)); + t = MAC16_32_Q16(t, g12, ADD32(x1,x0)); + t = SATURATE(t, SIG_SAT); + y[i+4] = t; + } +#ifdef CUSTOM_MODES + for (;i<N;i++) + { + opus_val32 t; + x0=SHL32(x[i-T+2],1); + t = MAC16_32_Q16(x[i], g10, x2); + t = MAC16_32_Q16(t, g11, ADD32(x1,x3)); + t = MAC16_32_Q16(t, g12, ADD32(x0,x4)); + t = SATURATE(t, SIG_SAT); + y[i] = t; + x4=x3; + x3=x2; + x2=x1; + x1=x0; + } +#endif +} +#else +#ifndef NON_STATIC_COMB_FILTER_CONST_C +static +#endif +void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12) +{ + opus_val32 x0, x1, x2, x3, x4; + int i; + x4 = x[-T-2]; + x3 = x[-T-1]; + x2 = x[-T]; + x1 = x[-T+1]; + for (i=0;i<N;i++) + { + x0=x[i-T+2]; + y[i] = x[i] + + MULT16_32_Q15(g10,x2) + + MULT16_32_Q15(g11,ADD32(x1,x3)) + + MULT16_32_Q15(g12,ADD32(x0,x4)); + y[i] = SATURATE(y[i], SIG_SAT); + x4=x3; + x3=x2; + x2=x1; + x1=x0; + } + +} +#endif +#endif + +#ifndef OVERRIDE_comb_filter +void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, + opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, + const opus_val16 *window, int overlap, int arch) +{ + int i; + /* printf ("%d %d %f %f\n", T0, T1, g0, g1); */ + opus_val16 g00, g01, g02, g10, g11, g12; + opus_val32 x0, x1, x2, x3, x4; + static const opus_val16 gains[3][3] = { + {QCONST16(0.3066406250f, 15), QCONST16(0.2170410156f, 15), QCONST16(0.1296386719f, 15)}, + {QCONST16(0.4638671875f, 15), QCONST16(0.2680664062f, 15), QCONST16(0.f, 15)}, + {QCONST16(0.7998046875f, 15), QCONST16(0.1000976562f, 15), QCONST16(0.f, 15)}}; + + if (g0==0 && g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y, x, N); + return; + } + /* When the gain is zero, T0 and/or T1 is set to zero. We need + to have then be at least 2 to avoid processing garbage data. */ + T0 = IMAX(T0, COMBFILTER_MINPERIOD); + T1 = IMAX(T1, COMBFILTER_MINPERIOD); + g00 = MULT16_16_P15(g0, gains[tapset0][0]); + g01 = MULT16_16_P15(g0, gains[tapset0][1]); + g02 = MULT16_16_P15(g0, gains[tapset0][2]); + g10 = MULT16_16_P15(g1, gains[tapset1][0]); + g11 = MULT16_16_P15(g1, gains[tapset1][1]); + g12 = MULT16_16_P15(g1, gains[tapset1][2]); + x1 = x[-T1+1]; + x2 = x[-T1 ]; + x3 = x[-T1-1]; + x4 = x[-T1-2]; + /* If the filter didn't change, we don't need the overlap */ + if (g0==g1 && T0==T1 && tapset0==tapset1) + overlap=0; + for (i=0;i<overlap;i++) + { + opus_val16 f; + x0=x[i-T1+2]; + f = MULT16_16_Q15(window[i],window[i]); + y[i] = x[i] + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g00),x[i-T0]) + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g01),ADD32(x[i-T0+1],x[i-T0-1])) + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g02),ADD32(x[i-T0+2],x[i-T0-2])) + + MULT16_32_Q15(MULT16_16_Q15(f,g10),x2) + + MULT16_32_Q15(MULT16_16_Q15(f,g11),ADD32(x1,x3)) + + MULT16_32_Q15(MULT16_16_Q15(f,g12),ADD32(x0,x4)); + y[i] = SATURATE(y[i], SIG_SAT); + x4=x3; + x3=x2; + x2=x1; + x1=x0; + + } + if (g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y+overlap, x+overlap, N-overlap); + return; + } + + /* Compute the part with the constant filter. */ + comb_filter_const(y+i, x+i, T1, N-i, g10, g11, g12, arch); +} +#endif /* OVERRIDE_comb_filter */ + +/* TF change table. Positive values mean better frequency resolution (longer + effective window), whereas negative values mean better time resolution + (shorter effective window). The second index is computed as: + 4*isTransient + 2*tf_select + per_band_flag */ +const signed char tf_select_table[4][8] = { + /*isTransient=0 isTransient=1 */ + {0, -1, 0, -1, 0,-1, 0,-1}, /* 2.5 ms */ + {0, -1, 0, -2, 1, 0, 1,-1}, /* 5 ms */ + {0, -2, 0, -3, 2, 0, 1,-1}, /* 10 ms */ + {0, -2, 0, -3, 3, 0, 1,-1}, /* 20 ms */ +}; + + +void init_caps(const CELTMode *m,int *cap,int LM,int C) +{ + int i; + for (i=0;i<m->nbEBands;i++) + { + int N; + N=(m->eBands[i+1]-m->eBands[i])<<LM; + cap[i] = (m->cache.caps[m->nbEBands*(2*LM+C-1)+i]+64)*C*N>>2; + } +} + + + +const char *opus_strerror(int error) +{ + static const char * const error_strings[8] = { + "success", + "invalid argument", + "buffer too small", + "internal error", + "corrupted stream", + "request not implemented", + "invalid state", + "memory allocation failed" + }; + if (error > 0 || error < -7) + return "unknown error"; + else + return error_strings[-error]; +} + +const char *opus_get_version_string(void) +{ + return "libopus " PACKAGE_VERSION + /* Applications may rely on the presence of this substring in the version + string to determine if they have a fixed-point or floating-point build + at runtime. */ +#ifdef FIXED_POINT + "-fixed" +#endif +#ifdef FUZZING + "-fuzzing" +#endif + ; +} diff --git a/libopus/celt/celt.h b/libopus/celt/celt.h new file mode 100644 index 0000000..24b6b2b --- /dev/null +++ b/libopus/celt/celt.h @@ -0,0 +1,251 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/** + @file celt.h + @brief Contains all the functions for encoding and decoding audio + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CELT_H +#define CELT_H + +#include "opus_types.h" +#include "opus_defines.h" +#include "opus_custom.h" +#include "entenc.h" +#include "entdec.h" +#include "arch.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define CELTEncoder OpusCustomEncoder +#define CELTDecoder OpusCustomDecoder +#define CELTMode OpusCustomMode + +#define LEAK_BANDS 19 + +typedef struct { + int valid; + float tonality; + float tonality_slope; + float noisiness; + float activity; + float music_prob; + float music_prob_min; + float music_prob_max; + int bandwidth; + float activity_probability; + float max_pitch_ratio; + /* Store as Q6 char to save space. */ + unsigned char leak_boost[LEAK_BANDS]; +} AnalysisInfo; + +typedef struct { + int signalType; + int offset; +} SILKInfo; + +#define __celt_check_mode_ptr_ptr(ptr) ((ptr) + ((ptr) - (const CELTMode**)(ptr))) + +#define __celt_check_analysis_ptr(ptr) ((ptr) + ((ptr) - (const AnalysisInfo*)(ptr))) + +#define __celt_check_silkinfo_ptr(ptr) ((ptr) + ((ptr) - (const SILKInfo*)(ptr))) + +/* Encoder/decoder Requests */ + + +#define CELT_SET_PREDICTION_REQUEST 10002 +/** Controls the use of interframe prediction. + 0=Independent frames + 1=Short term interframe prediction allowed + 2=Long term prediction allowed + */ +#define CELT_SET_PREDICTION(x) CELT_SET_PREDICTION_REQUEST, __opus_check_int(x) + +#define CELT_SET_INPUT_CLIPPING_REQUEST 10004 +#define CELT_SET_INPUT_CLIPPING(x) CELT_SET_INPUT_CLIPPING_REQUEST, __opus_check_int(x) + +#define CELT_GET_AND_CLEAR_ERROR_REQUEST 10007 +#define CELT_GET_AND_CLEAR_ERROR(x) CELT_GET_AND_CLEAR_ERROR_REQUEST, __opus_check_int_ptr(x) + +#define CELT_SET_CHANNELS_REQUEST 10008 +#define CELT_SET_CHANNELS(x) CELT_SET_CHANNELS_REQUEST, __opus_check_int(x) + + +/* Internal */ +#define CELT_SET_START_BAND_REQUEST 10010 +#define CELT_SET_START_BAND(x) CELT_SET_START_BAND_REQUEST, __opus_check_int(x) + +#define CELT_SET_END_BAND_REQUEST 10012 +#define CELT_SET_END_BAND(x) CELT_SET_END_BAND_REQUEST, __opus_check_int(x) + +#define CELT_GET_MODE_REQUEST 10015 +/** Get the CELTMode used by an encoder or decoder */ +#define CELT_GET_MODE(x) CELT_GET_MODE_REQUEST, __celt_check_mode_ptr_ptr(x) + +#define CELT_SET_SIGNALLING_REQUEST 10016 +#define CELT_SET_SIGNALLING(x) CELT_SET_SIGNALLING_REQUEST, __opus_check_int(x) + +#define CELT_SET_TONALITY_REQUEST 10018 +#define CELT_SET_TONALITY(x) CELT_SET_TONALITY_REQUEST, __opus_check_int(x) +#define CELT_SET_TONALITY_SLOPE_REQUEST 10020 +#define CELT_SET_TONALITY_SLOPE(x) CELT_SET_TONALITY_SLOPE_REQUEST, __opus_check_int(x) + +#define CELT_SET_ANALYSIS_REQUEST 10022 +#define CELT_SET_ANALYSIS(x) CELT_SET_ANALYSIS_REQUEST, __celt_check_analysis_ptr(x) + +#define OPUS_SET_LFE_REQUEST 10024 +#define OPUS_SET_LFE(x) OPUS_SET_LFE_REQUEST, __opus_check_int(x) + +#define OPUS_SET_ENERGY_MASK_REQUEST 10026 +#define OPUS_SET_ENERGY_MASK(x) OPUS_SET_ENERGY_MASK_REQUEST, __opus_check_val16_ptr(x) + +#define CELT_SET_SILK_INFO_REQUEST 10028 +#define CELT_SET_SILK_INFO(x) CELT_SET_SILK_INFO_REQUEST, __celt_check_silkinfo_ptr(x) + +/* Encoder stuff */ + +int celt_encoder_get_size(int channels); + +int celt_encode_with_ec(OpusCustomEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes, ec_enc *enc); + +int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels, + int arch); + + + +/* Decoder stuff */ + +int celt_decoder_get_size(int channels); + + +int celt_decoder_init(CELTDecoder *st, opus_int32 sampling_rate, int channels); + +int celt_decode_with_ec(OpusCustomDecoder * OPUS_RESTRICT st, const unsigned char *data, + int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec, int accum); + +#define celt_encoder_ctl opus_custom_encoder_ctl +#define celt_decoder_ctl opus_custom_decoder_ctl + + +#ifdef CUSTOM_MODES +#define OPUS_CUSTOM_NOSTATIC +#else +#define OPUS_CUSTOM_NOSTATIC static OPUS_INLINE +#endif + +static const unsigned char trim_icdf[11] = {126, 124, 119, 109, 87, 41, 19, 9, 4, 2, 0}; +/* Probs: NONE: 21.875%, LIGHT: 6.25%, NORMAL: 65.625%, AGGRESSIVE: 6.25% */ +static const unsigned char spread_icdf[4] = {25, 23, 2, 0}; + +static const unsigned char tapset_icdf[3]={2,1,0}; + +#ifdef CUSTOM_MODES +static const unsigned char toOpusTable[20] = { + 0xE0, 0xE8, 0xF0, 0xF8, + 0xC0, 0xC8, 0xD0, 0xD8, + 0xA0, 0xA8, 0xB0, 0xB8, + 0x00, 0x00, 0x00, 0x00, + 0x80, 0x88, 0x90, 0x98, +}; + +static const unsigned char fromOpusTable[16] = { + 0x80, 0x88, 0x90, 0x98, + 0x40, 0x48, 0x50, 0x58, + 0x20, 0x28, 0x30, 0x38, + 0x00, 0x08, 0x10, 0x18 +}; + +static OPUS_INLINE int toOpus(unsigned char c) +{ + int ret=0; + if (c<0xA0) + ret = toOpusTable[c>>3]; + if (ret == 0) + return -1; + else + return ret|(c&0x7); +} + +static OPUS_INLINE int fromOpus(unsigned char c) +{ + if (c<0x80) + return -1; + else + return fromOpusTable[(c>>3)-16] | (c&0x7); +} +#endif /* CUSTOM_MODES */ + +#define COMBFILTER_MAXPERIOD 1024 +#define COMBFILTER_MINPERIOD 15 + +extern const signed char tf_select_table[4][8]; + +#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS) +void validate_celt_decoder(CELTDecoder *st); +#define VALIDATE_CELT_DECODER(st) validate_celt_decoder(st) +#else +#define VALIDATE_CELT_DECODER(st) +#endif + +int resampling_factor(opus_int32 rate); + +void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RESTRICT inp, + int N, int CC, int upsample, const opus_val16 *coef, celt_sig *mem, int clip); + +void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, + opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, + const opus_val16 *window, int overlap, int arch); + +#ifdef NON_STATIC_COMB_FILTER_CONST_C +void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12); +#endif + +#ifndef OVERRIDE_COMB_FILTER_CONST +# define comb_filter_const(y, x, T, N, g10, g11, g12, arch) \ + ((void)(arch),comb_filter_const_c(y, x, T, N, g10, g11, g12)) +#endif + +void init_caps(const CELTMode *m,int *cap,int LM,int C); + +#ifdef RESYNTH +void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem); +void celt_synthesis(const CELTMode *mode, celt_norm *X, celt_sig * out_syn[], + opus_val16 *oldBandE, int start, int effEnd, int C, int CC, int isTransient, + int LM, int downsample, int silence); +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* CELT_H */ diff --git a/libopus/celt/celt_decoder.c b/libopus/celt/celt_decoder.c new file mode 100644 index 0000000..46a1f50 --- /dev/null +++ b/libopus/celt/celt_decoder.c @@ -0,0 +1,1372 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_DECODER_C + +#include "cpu_support.h" +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + +/* The maximum pitch lag to allow in the pitch-based PLC. It's possible to save + CPU time in the PLC pitch search by making this smaller than MAX_PERIOD. The + current value corresponds to a pitch of 66.67 Hz. */ +#define PLC_PITCH_LAG_MAX (720) +/* The minimum pitch lag to allow in the pitch-based PLC. This corresponds to a + pitch of 480 Hz. */ +#define PLC_PITCH_LAG_MIN (100) + +#if defined(SMALL_FOOTPRINT) && defined(FIXED_POINT) +#define NORM_ALIASING_HACK +#endif +/**********************************************************************/ +/* */ +/* DECODER */ +/* */ +/**********************************************************************/ +#define DECODE_BUFFER_SIZE 2048 + +/** Decoder state + @brief Decoder state + */ +struct OpusCustomDecoder { + const OpusCustomMode *mode; + int overlap; + int channels; + int stream_channels; + + int downsample; + int start, end; + int signalling; + int disable_inv; + int arch; + + /* Everything beyond this point gets cleared on a reset */ +#define DECODER_RESET_START rng + + opus_uint32 rng; + int error; + int last_pitch_index; + int loss_count; + int skip_plc; + int postfilter_period; + int postfilter_period_old; + opus_val16 postfilter_gain; + opus_val16 postfilter_gain_old; + int postfilter_tapset; + int postfilter_tapset_old; + + celt_sig preemph_memD[2]; + + celt_sig _decode_mem[1]; /* Size = channels*(DECODE_BUFFER_SIZE+mode->overlap) */ + /* opus_val16 lpc[], Size = channels*LPC_ORDER */ + /* opus_val16 oldEBands[], Size = 2*mode->nbEBands */ + /* opus_val16 oldLogE[], Size = 2*mode->nbEBands */ + /* opus_val16 oldLogE2[], Size = 2*mode->nbEBands */ + /* opus_val16 backgroundLogE[], Size = 2*mode->nbEBands */ +}; + +#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS) +/* Make basic checks on the CELT state to ensure we don't end + up writing all over memory. */ +void validate_celt_decoder(CELTDecoder *st) +{ +#ifndef CUSTOM_MODES + celt_assert(st->mode == opus_custom_mode_create(48000, 960, NULL)); + celt_assert(st->overlap == 120); +#endif + celt_assert(st->channels == 1 || st->channels == 2); + celt_assert(st->stream_channels == 1 || st->stream_channels == 2); + celt_assert(st->downsample > 0); + celt_assert(st->start == 0 || st->start == 17); + celt_assert(st->start < st->end); + celt_assert(st->end <= 21); +#ifdef OPUS_ARCHMASK + celt_assert(st->arch >= 0); + celt_assert(st->arch <= OPUS_ARCHMASK); +#endif + celt_assert(st->last_pitch_index <= PLC_PITCH_LAG_MAX); + celt_assert(st->last_pitch_index >= PLC_PITCH_LAG_MIN || st->last_pitch_index == 0); + celt_assert(st->postfilter_period < MAX_PERIOD); + celt_assert(st->postfilter_period >= COMBFILTER_MINPERIOD || st->postfilter_period == 0); + celt_assert(st->postfilter_period_old < MAX_PERIOD); + celt_assert(st->postfilter_period_old >= COMBFILTER_MINPERIOD || st->postfilter_period_old == 0); + celt_assert(st->postfilter_tapset <= 2); + celt_assert(st->postfilter_tapset >= 0); + celt_assert(st->postfilter_tapset_old <= 2); + celt_assert(st->postfilter_tapset_old >= 0); +} +#endif + +int celt_decoder_get_size(int channels) +{ + const CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); + return opus_custom_decoder_get_size(mode, channels); +} + +int opus_custom_decoder_get_size(const CELTMode *mode, int channels) +{ + int size = sizeof(struct CELTDecoder) + + (channels*(DECODE_BUFFER_SIZE+mode->overlap)-1)*sizeof(celt_sig) + + channels*LPC_ORDER*sizeof(opus_val16) + + 4*2*mode->nbEBands*sizeof(opus_val16); + return size; +} + +#ifdef CUSTOM_MODES +CELTDecoder *opus_custom_decoder_create(const CELTMode *mode, int channels, int *error) +{ + int ret; + CELTDecoder *st = (CELTDecoder *)opus_alloc(opus_custom_decoder_get_size(mode, channels)); + ret = opus_custom_decoder_init(st, mode, channels); + if (ret != OPUS_OK) + { + opus_custom_decoder_destroy(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} +#endif /* CUSTOM_MODES */ + +int celt_decoder_init(CELTDecoder *st, opus_int32 sampling_rate, int channels) +{ + int ret; + ret = opus_custom_decoder_init(st, opus_custom_mode_create(48000, 960, NULL), channels); + if (ret != OPUS_OK) + return ret; + st->downsample = resampling_factor(sampling_rate); + if (st->downsample==0) + return OPUS_BAD_ARG; + else + return OPUS_OK; +} + +int opus_custom_decoder_init(CELTDecoder *st, const CELTMode *mode, int channels) +{ + if (channels < 0 || channels > 2) + return OPUS_BAD_ARG; + + if (st==NULL) + return OPUS_ALLOC_FAIL; + + OPUS_CLEAR((char*)st, opus_custom_decoder_get_size(mode, channels)); + + st->mode = mode; + st->overlap = mode->overlap; + st->stream_channels = st->channels = channels; + + st->downsample = 1; + st->start = 0; + st->end = st->mode->effEBands; + st->signalling = 1; +#ifndef DISABLE_UPDATE_DRAFT + st->disable_inv = channels == 1; +#else + st->disable_inv = 0; +#endif + st->arch = opus_select_arch(); + + opus_custom_decoder_ctl(st, OPUS_RESET_STATE); + + return OPUS_OK; +} + +#ifdef CUSTOM_MODES +void opus_custom_decoder_destroy(CELTDecoder *st) +{ + opus_free(st); +} +#endif /* CUSTOM_MODES */ + +#ifndef CUSTOM_MODES +/* Special case for stereo with no downsampling and no accumulation. This is + quite common and we can make it faster by processing both channels in the + same loop, reducing overhead due to the dependency loop in the IIR filter. */ +static void deemphasis_stereo_simple(celt_sig *in[], opus_val16 *pcm, int N, const opus_val16 coef0, + celt_sig *mem) +{ + celt_sig * OPUS_RESTRICT x0; + celt_sig * OPUS_RESTRICT x1; + celt_sig m0, m1; + int j; + x0=in[0]; + x1=in[1]; + m0 = mem[0]; + m1 = mem[1]; + for (j=0;j<N;j++) + { + celt_sig tmp0, tmp1; + /* Add VERY_SMALL to x[] first to reduce dependency chain. */ + tmp0 = x0[j] + VERY_SMALL + m0; + tmp1 = x1[j] + VERY_SMALL + m1; + m0 = MULT16_32_Q15(coef0, tmp0); + m1 = MULT16_32_Q15(coef0, tmp1); + pcm[2*j ] = SCALEOUT(SIG2WORD16(tmp0)); + pcm[2*j+1] = SCALEOUT(SIG2WORD16(tmp1)); + } + mem[0] = m0; + mem[1] = m1; +} +#endif + +#ifndef RESYNTH +static +#endif +void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, + celt_sig *mem, int accum) +{ + int c; + int Nd; + int apply_downsampling=0; + opus_val16 coef0; + VARDECL(celt_sig, scratch); + SAVE_STACK; +#ifndef CUSTOM_MODES + /* Short version for common case. */ + if (downsample == 1 && C == 2 && !accum) + { + deemphasis_stereo_simple(in, pcm, N, coef[0], mem); + return; + } +#endif +#ifndef FIXED_POINT + (void)accum; + celt_assert(accum==0); +#endif + ALLOC(scratch, N, celt_sig); + coef0 = coef[0]; + Nd = N/downsample; + c=0; do { + int j; + celt_sig * OPUS_RESTRICT x; + opus_val16 * OPUS_RESTRICT y; + celt_sig m = mem[c]; + x =in[c]; + y = pcm+c; +#ifdef CUSTOM_MODES + if (coef[1] != 0) + { + opus_val16 coef1 = coef[1]; + opus_val16 coef3 = coef[3]; + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + m + VERY_SMALL; + m = MULT16_32_Q15(coef0, tmp) + - MULT16_32_Q15(coef1, x[j]); + tmp = SHL32(MULT16_32_Q15(coef3, tmp), 2); + scratch[j] = tmp; + } + apply_downsampling=1; + } else +#endif + if (downsample>1) + { + /* Shortcut for the standard (non-custom modes) case */ + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + VERY_SMALL + m; + m = MULT16_32_Q15(coef0, tmp); + scratch[j] = tmp; + } + apply_downsampling=1; + } else { + /* Shortcut for the standard (non-custom modes) case */ +#ifdef FIXED_POINT + if (accum) + { + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + m + VERY_SMALL; + m = MULT16_32_Q15(coef0, tmp); + y[j*C] = SAT16(ADD32(y[j*C], SCALEOUT(SIG2WORD16(tmp)))); + } + } else +#endif + { + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + VERY_SMALL + m; + m = MULT16_32_Q15(coef0, tmp); + y[j*C] = SCALEOUT(SIG2WORD16(tmp)); + } + } + } + mem[c] = m; + + if (apply_downsampling) + { + /* Perform down-sampling */ +#ifdef FIXED_POINT + if (accum) + { + for (j=0;j<Nd;j++) + y[j*C] = SAT16(ADD32(y[j*C], SCALEOUT(SIG2WORD16(scratch[j*downsample])))); + } else +#endif + { + for (j=0;j<Nd;j++) + y[j*C] = SCALEOUT(SIG2WORD16(scratch[j*downsample])); + } + } + } while (++c<C); + RESTORE_STACK; +} + +#ifndef RESYNTH +static +#endif +void celt_synthesis(const CELTMode *mode, celt_norm *X, celt_sig * out_syn[], + opus_val16 *oldBandE, int start, int effEnd, int C, int CC, + int isTransient, int LM, int downsample, + int silence, int arch) +{ + int c, i; + int M; + int b; + int B; + int N, NB; + int shift; + int nbEBands; + int overlap; + VARDECL(celt_sig, freq); + SAVE_STACK; + + overlap = mode->overlap; + nbEBands = mode->nbEBands; + N = mode->shortMdctSize<<LM; + ALLOC(freq, N, celt_sig); /**< Interleaved signal MDCTs */ + M = 1<<LM; + + if (isTransient) + { + B = M; + NB = mode->shortMdctSize; + shift = mode->maxLM; + } else { + B = 1; + NB = mode->shortMdctSize<<LM; + shift = mode->maxLM-LM; + } + + if (CC==2&&C==1) + { + /* Copying a mono streams to two channels */ + celt_sig *freq2; + denormalise_bands(mode, X, freq, oldBandE, start, effEnd, M, + downsample, silence); + /* Store a temporary copy in the output buffer because the IMDCT destroys its input. */ + freq2 = out_syn[1]+overlap/2; + OPUS_COPY(freq2, freq, N); + for (b=0;b<B;b++) + clt_mdct_backward(&mode->mdct, &freq2[b], out_syn[0]+NB*b, mode->window, overlap, shift, B, arch); + for (b=0;b<B;b++) + clt_mdct_backward(&mode->mdct, &freq[b], out_syn[1]+NB*b, mode->window, overlap, shift, B, arch); + } else if (CC==1&&C==2) + { + /* Downmixing a stereo stream to mono */ + celt_sig *freq2; + freq2 = out_syn[0]+overlap/2; + denormalise_bands(mode, X, freq, oldBandE, start, effEnd, M, + downsample, silence); + /* Use the output buffer as temp array before downmixing. */ + denormalise_bands(mode, X+N, freq2, oldBandE+nbEBands, start, effEnd, M, + downsample, silence); + for (i=0;i<N;i++) + freq[i] = ADD32(HALF32(freq[i]), HALF32(freq2[i])); + for (b=0;b<B;b++) + clt_mdct_backward(&mode->mdct, &freq[b], out_syn[0]+NB*b, mode->window, overlap, shift, B, arch); + } else { + /* Normal case (mono or stereo) */ + c=0; do { + denormalise_bands(mode, X+c*N, freq, oldBandE+c*nbEBands, start, effEnd, M, + downsample, silence); + for (b=0;b<B;b++) + clt_mdct_backward(&mode->mdct, &freq[b], out_syn[c]+NB*b, mode->window, overlap, shift, B, arch); + } while (++c<CC); + } + /* Saturate IMDCT output so that we can't overflow in the pitch postfilter + or in the */ + c=0; do { + for (i=0;i<N;i++) + out_syn[c][i] = SATURATE(out_syn[c][i], SIG_SAT); + } while (++c<CC); + RESTORE_STACK; +} + +static void tf_decode(int start, int end, int isTransient, int *tf_res, int LM, ec_dec *dec) +{ + int i, curr, tf_select; + int tf_select_rsv; + int tf_changed; + int logp; + opus_uint32 budget; + opus_uint32 tell; + + budget = dec->storage*8; + tell = ec_tell(dec); + logp = isTransient ? 2 : 4; + tf_select_rsv = LM>0 && tell+logp+1<=budget; + budget -= tf_select_rsv; + tf_changed = curr = 0; + for (i=start;i<end;i++) + { + if (tell+logp<=budget) + { + curr ^= ec_dec_bit_logp(dec, logp); + tell = ec_tell(dec); + tf_changed |= curr; + } + tf_res[i] = curr; + logp = isTransient ? 4 : 5; + } + tf_select = 0; + if (tf_select_rsv && + tf_select_table[LM][4*isTransient+0+tf_changed] != + tf_select_table[LM][4*isTransient+2+tf_changed]) + { + tf_select = ec_dec_bit_logp(dec, 1); + } + for (i=start;i<end;i++) + { + tf_res[i] = tf_select_table[LM][4*isTransient+2*tf_select+tf_res[i]]; + } +} + +static int celt_plc_pitch_search(celt_sig *decode_mem[2], int C, int arch) +{ + int pitch_index; + VARDECL( opus_val16, lp_pitch_buf ); + SAVE_STACK; + ALLOC( lp_pitch_buf, DECODE_BUFFER_SIZE>>1, opus_val16 ); + pitch_downsample(decode_mem, lp_pitch_buf, + DECODE_BUFFER_SIZE, C, arch); + pitch_search(lp_pitch_buf+(PLC_PITCH_LAG_MAX>>1), lp_pitch_buf, + DECODE_BUFFER_SIZE-PLC_PITCH_LAG_MAX, + PLC_PITCH_LAG_MAX-PLC_PITCH_LAG_MIN, &pitch_index, arch); + pitch_index = PLC_PITCH_LAG_MAX-pitch_index; + RESTORE_STACK; + return pitch_index; +} + +static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) +{ + int c; + int i; + const int C = st->channels; + celt_sig *decode_mem[2]; + celt_sig *out_syn[2]; + opus_val16 *lpc; + opus_val16 *oldBandE, *oldLogE, *oldLogE2, *backgroundLogE; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + int start; + int loss_count; + int noise_based; + const opus_int16 *eBands; + SAVE_STACK; + + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + + c=0; do { + decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+overlap); + out_syn[c] = decode_mem[c]+DECODE_BUFFER_SIZE-N; + } while (++c<C); + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+overlap)*C); + oldBandE = lpc+C*LPC_ORDER; + oldLogE = oldBandE + 2*nbEBands; + oldLogE2 = oldLogE + 2*nbEBands; + backgroundLogE = oldLogE2 + 2*nbEBands; + + loss_count = st->loss_count; + start = st->start; + noise_based = loss_count >= 5 || start != 0 || st->skip_plc; + if (noise_based) + { + /* Noise-based PLC/CNG */ +#ifdef NORM_ALIASING_HACK + celt_norm *X; +#else + VARDECL(celt_norm, X); +#endif + opus_uint32 seed; + int end; + int effEnd; + opus_val16 decay; + end = st->end; + effEnd = IMAX(start, IMIN(end, mode->effEBands)); + +#ifdef NORM_ALIASING_HACK + /* This is an ugly hack that breaks aliasing rules and would be easily broken, + but it saves almost 4kB of stack. */ + X = (celt_norm*)(out_syn[C-1]+overlap/2); +#else + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ +#endif + + /* Energy decay */ + decay = loss_count==0 ? QCONST16(1.5f, DB_SHIFT) : QCONST16(.5f, DB_SHIFT); + c=0; do + { + for (i=start;i<end;i++) + oldBandE[c*nbEBands+i] = MAX16(backgroundLogE[c*nbEBands+i], oldBandE[c*nbEBands+i] - decay); + } while (++c<C); + seed = st->rng; + for (c=0;c<C;c++) + { + for (i=start;i<effEnd;i++) + { + int j; + int boffs; + int blen; + boffs = N*c+(eBands[i]<<LM); + blen = (eBands[i+1]-eBands[i])<<LM; + for (j=0;j<blen;j++) + { + seed = celt_lcg_rand(seed); + X[boffs+j] = (celt_norm)((opus_int32)seed>>20); + } + renormalise_vector(X+boffs, blen, Q15ONE, st->arch); + } + } + st->rng = seed; + + c=0; do { + OPUS_MOVE(decode_mem[c], decode_mem[c]+N, + DECODE_BUFFER_SIZE-N+(overlap>>1)); + } while (++c<C); + + celt_synthesis(mode, X, out_syn, oldBandE, start, effEnd, C, C, 0, LM, st->downsample, 0, st->arch); + } else { + int exc_length; + /* Pitch-based PLC */ + const opus_val16 *window; + opus_val16 *exc; + opus_val16 fade = Q15ONE; + int pitch_index; + VARDECL(opus_val32, etmp); + VARDECL(opus_val16, _exc); + VARDECL(opus_val16, fir_tmp); + + if (loss_count == 0) + { + st->last_pitch_index = pitch_index = celt_plc_pitch_search(decode_mem, C, st->arch); + } else { + pitch_index = st->last_pitch_index; + fade = QCONST16(.8f,15); + } + + /* We want the excitation for 2 pitch periods in order to look for a + decaying signal, but we can't get more than MAX_PERIOD. */ + exc_length = IMIN(2*pitch_index, MAX_PERIOD); + + ALLOC(etmp, overlap, opus_val32); + ALLOC(_exc, MAX_PERIOD+LPC_ORDER, opus_val16); + ALLOC(fir_tmp, exc_length, opus_val16); + exc = _exc+LPC_ORDER; + window = mode->window; + c=0; do { + opus_val16 decay; + opus_val16 attenuation; + opus_val32 S1=0; + celt_sig *buf; + int extrapolation_offset; + int extrapolation_len; + int j; + + buf = decode_mem[c]; + for (i=0;i<MAX_PERIOD+LPC_ORDER;i++) + exc[i-LPC_ORDER] = ROUND16(buf[DECODE_BUFFER_SIZE-MAX_PERIOD-LPC_ORDER+i], SIG_SHIFT); + + if (loss_count == 0) + { + opus_val32 ac[LPC_ORDER+1]; + /* Compute LPC coefficients for the last MAX_PERIOD samples before + the first loss so we can work in the excitation-filter domain. */ + _celt_autocorr(exc, ac, window, overlap, + LPC_ORDER, MAX_PERIOD, st->arch); + /* Add a noise floor of -40 dB. */ +#ifdef FIXED_POINT + ac[0] += SHR32(ac[0],13); +#else + ac[0] *= 1.0001f; +#endif + /* Use lag windowing to stabilize the Levinson-Durbin recursion. */ + for (i=1;i<=LPC_ORDER;i++) + { + /*ac[i] *= exp(-.5*(2*M_PI*.002*i)*(2*M_PI*.002*i));*/ +#ifdef FIXED_POINT + ac[i] -= MULT16_32_Q15(2*i*i, ac[i]); +#else + ac[i] -= ac[i]*(0.008f*0.008f)*i*i; +#endif + } + _celt_lpc(lpc+c*LPC_ORDER, ac, LPC_ORDER); +#ifdef FIXED_POINT + /* For fixed-point, apply bandwidth expansion until we can guarantee that + no overflow can happen in the IIR filter. This means: + 32768*sum(abs(filter)) < 2^31 */ + while (1) { + opus_val16 tmp=Q15ONE; + opus_val32 sum=QCONST16(1., SIG_SHIFT); + for (i=0;i<LPC_ORDER;i++) + sum += ABS16(lpc[c*LPC_ORDER+i]); + if (sum < 65535) break; + for (i=0;i<LPC_ORDER;i++) + { + tmp = MULT16_16_Q15(QCONST16(.99f,15), tmp); + lpc[c*LPC_ORDER+i] = MULT16_16_Q15(lpc[c*LPC_ORDER+i], tmp); + } + } +#endif + } + /* Initialize the LPC history with the samples just before the start + of the region for which we're computing the excitation. */ + { + /* Compute the excitation for exc_length samples before the loss. We need the copy + because celt_fir() cannot filter in-place. */ + celt_fir(exc+MAX_PERIOD-exc_length, lpc+c*LPC_ORDER, + fir_tmp, exc_length, LPC_ORDER, st->arch); + OPUS_COPY(exc+MAX_PERIOD-exc_length, fir_tmp, exc_length); + } + + /* Check if the waveform is decaying, and if so how fast. + We do this to avoid adding energy when concealing in a segment + with decaying energy. */ + { + opus_val32 E1=1, E2=1; + int decay_length; +#ifdef FIXED_POINT + int shift = IMAX(0,2*celt_zlog2(celt_maxabs16(&exc[MAX_PERIOD-exc_length], exc_length))-20); +#endif + decay_length = exc_length>>1; + for (i=0;i<decay_length;i++) + { + opus_val16 e; + e = exc[MAX_PERIOD-decay_length+i]; + E1 += SHR32(MULT16_16(e, e), shift); + e = exc[MAX_PERIOD-2*decay_length+i]; + E2 += SHR32(MULT16_16(e, e), shift); + } + E1 = MIN32(E1, E2); + decay = celt_sqrt(frac_div32(SHR32(E1, 1), E2)); + } + + /* Move the decoder memory one frame to the left to give us room to + add the data for the new frame. We ignore the overlap that extends + past the end of the buffer, because we aren't going to use it. */ + OPUS_MOVE(buf, buf+N, DECODE_BUFFER_SIZE-N); + + /* Extrapolate from the end of the excitation with a period of + "pitch_index", scaling down each period by an additional factor of + "decay". */ + extrapolation_offset = MAX_PERIOD-pitch_index; + /* We need to extrapolate enough samples to cover a complete MDCT + window (including overlap/2 samples on both sides). */ + extrapolation_len = N+overlap; + /* We also apply fading if this is not the first loss. */ + attenuation = MULT16_16_Q15(fade, decay); + for (i=j=0;i<extrapolation_len;i++,j++) + { + opus_val16 tmp; + if (j >= pitch_index) { + j -= pitch_index; + attenuation = MULT16_16_Q15(attenuation, decay); + } + buf[DECODE_BUFFER_SIZE-N+i] = + SHL32(EXTEND32(MULT16_16_Q15(attenuation, + exc[extrapolation_offset+j])), SIG_SHIFT); + /* Compute the energy of the previously decoded signal whose + excitation we're copying. */ + tmp = ROUND16( + buf[DECODE_BUFFER_SIZE-MAX_PERIOD-N+extrapolation_offset+j], + SIG_SHIFT); + S1 += SHR32(MULT16_16(tmp, tmp), 10); + } + { + opus_val16 lpc_mem[LPC_ORDER]; + /* Copy the last decoded samples (prior to the overlap region) to + synthesis filter memory so we can have a continuous signal. */ + for (i=0;i<LPC_ORDER;i++) + lpc_mem[i] = ROUND16(buf[DECODE_BUFFER_SIZE-N-1-i], SIG_SHIFT); + /* Apply the synthesis filter to convert the excitation back into + the signal domain. */ + celt_iir(buf+DECODE_BUFFER_SIZE-N, lpc+c*LPC_ORDER, + buf+DECODE_BUFFER_SIZE-N, extrapolation_len, LPC_ORDER, + lpc_mem, st->arch); +#ifdef FIXED_POINT + for (i=0; i < extrapolation_len; i++) + buf[DECODE_BUFFER_SIZE-N+i] = SATURATE(buf[DECODE_BUFFER_SIZE-N+i], SIG_SAT); +#endif + } + + /* Check if the synthesis energy is higher than expected, which can + happen with the signal changes during our window. If so, + attenuate. */ + { + opus_val32 S2=0; + for (i=0;i<extrapolation_len;i++) + { + opus_val16 tmp = ROUND16(buf[DECODE_BUFFER_SIZE-N+i], SIG_SHIFT); + S2 += SHR32(MULT16_16(tmp, tmp), 10); + } + /* This checks for an "explosion" in the synthesis. */ +#ifdef FIXED_POINT + if (!(S1 > SHR32(S2,2))) +#else + /* The float test is written this way to catch NaNs in the output + of the IIR filter at the same time. */ + if (!(S1 > 0.2f*S2)) +#endif + { + for (i=0;i<extrapolation_len;i++) + buf[DECODE_BUFFER_SIZE-N+i] = 0; + } else if (S1 < S2) + { + opus_val16 ratio = celt_sqrt(frac_div32(SHR32(S1,1)+1,S2+1)); + for (i=0;i<overlap;i++) + { + opus_val16 tmp_g = Q15ONE + - MULT16_16_Q15(window[i], Q15ONE-ratio); + buf[DECODE_BUFFER_SIZE-N+i] = + MULT16_32_Q15(tmp_g, buf[DECODE_BUFFER_SIZE-N+i]); + } + for (i=overlap;i<extrapolation_len;i++) + { + buf[DECODE_BUFFER_SIZE-N+i] = + MULT16_32_Q15(ratio, buf[DECODE_BUFFER_SIZE-N+i]); + } + } + } + + /* Apply the pre-filter to the MDCT overlap for the next frame because + the post-filter will be re-applied in the decoder after the MDCT + overlap. */ + comb_filter(etmp, buf+DECODE_BUFFER_SIZE, + st->postfilter_period, st->postfilter_period, overlap, + -st->postfilter_gain, -st->postfilter_gain, + st->postfilter_tapset, st->postfilter_tapset, NULL, 0, st->arch); + + /* Simulate TDAC on the concealed audio so that it blends with the + MDCT of the next frame. */ + for (i=0;i<overlap/2;i++) + { + buf[DECODE_BUFFER_SIZE+i] = + MULT16_32_Q15(window[i], etmp[overlap-1-i]) + + MULT16_32_Q15(window[overlap-i-1], etmp[i]); + } + } while (++c<C); + } + + st->loss_count = loss_count+1; + + RESTORE_STACK; +} + +int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, + int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec, int accum) +{ + int c, i, N; + int spread_decision; + opus_int32 bits; + ec_dec _dec; +#ifdef NORM_ALIASING_HACK + celt_norm *X; +#else + VARDECL(celt_norm, X); +#endif + VARDECL(int, fine_quant); + VARDECL(int, pulses); + VARDECL(int, cap); + VARDECL(int, offsets); + VARDECL(int, fine_priority); + VARDECL(int, tf_res); + VARDECL(unsigned char, collapse_masks); + celt_sig *decode_mem[2]; + celt_sig *out_syn[2]; + opus_val16 *lpc; + opus_val16 *oldBandE, *oldLogE, *oldLogE2, *backgroundLogE; + + int shortBlocks; + int isTransient; + int intra_ener; + const int CC = st->channels; + int LM, M; + int start; + int end; + int effEnd; + int codedBands; + int alloc_trim; + int postfilter_pitch; + opus_val16 postfilter_gain; + int intensity=0; + int dual_stereo=0; + opus_int32 total_bits; + opus_int32 balance; + opus_int32 tell; + int dynalloc_logp; + int postfilter_tapset; + int anti_collapse_rsv; + int anti_collapse_on=0; + int silence; + int C = st->stream_channels; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + const opus_int16 *eBands; + ALLOC_STACK; + + VALIDATE_CELT_DECODER(st); + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + start = st->start; + end = st->end; + frame_size *= st->downsample; + + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+overlap)*CC); + oldBandE = lpc+CC*LPC_ORDER; + oldLogE = oldBandE + 2*nbEBands; + oldLogE2 = oldLogE + 2*nbEBands; + backgroundLogE = oldLogE2 + 2*nbEBands; + +#ifdef CUSTOM_MODES + if (st->signalling && data!=NULL) + { + int data0=data[0]; + /* Convert "standard mode" to Opus header */ + if (mode->Fs==48000 && mode->shortMdctSize==120) + { + data0 = fromOpus(data0); + if (data0<0) + return OPUS_INVALID_PACKET; + } + st->end = end = IMAX(1, mode->effEBands-2*(data0>>5)); + LM = (data0>>3)&0x3; + C = 1 + ((data0>>2)&0x1); + data++; + len--; + if (LM>mode->maxLM) + return OPUS_INVALID_PACKET; + if (frame_size < mode->shortMdctSize<<LM) + return OPUS_BUFFER_TOO_SMALL; + else + frame_size = mode->shortMdctSize<<LM; + } else { +#else + { +#endif + for (LM=0;LM<=mode->maxLM;LM++) + if (mode->shortMdctSize<<LM==frame_size) + break; + if (LM>mode->maxLM) + return OPUS_BAD_ARG; + } + M=1<<LM; + + if (len<0 || len>1275 || pcm==NULL) + return OPUS_BAD_ARG; + + N = M*mode->shortMdctSize; + c=0; do { + decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+overlap); + out_syn[c] = decode_mem[c]+DECODE_BUFFER_SIZE-N; + } while (++c<CC); + + effEnd = end; + if (effEnd > mode->effEBands) + effEnd = mode->effEBands; + + if (data == NULL || len<=1) + { + celt_decode_lost(st, N, LM); + deemphasis(out_syn, pcm, N, CC, st->downsample, mode->preemph, st->preemph_memD, accum); + RESTORE_STACK; + return frame_size/st->downsample; + } + + /* Check if there are at least two packets received consecutively before + * turning on the pitch-based PLC */ + st->skip_plc = st->loss_count != 0; + + if (dec == NULL) + { + ec_dec_init(&_dec,(unsigned char*)data,len); + dec = &_dec; + } + + if (C==1) + { + for (i=0;i<nbEBands;i++) + oldBandE[i]=MAX16(oldBandE[i],oldBandE[nbEBands+i]); + } + + total_bits = len*8; + tell = ec_tell(dec); + + if (tell >= total_bits) + silence = 1; + else if (tell==1) + silence = ec_dec_bit_logp(dec, 15); + else + silence = 0; + if (silence) + { + /* Pretend we've read all the remaining bits */ + tell = len*8; + dec->nbits_total+=tell-ec_tell(dec); + } + + postfilter_gain = 0; + postfilter_pitch = 0; + postfilter_tapset = 0; + if (start==0 && tell+16 <= total_bits) + { + if(ec_dec_bit_logp(dec, 1)) + { + int qg, octave; + octave = ec_dec_uint(dec, 6); + postfilter_pitch = (16<<octave)+ec_dec_bits(dec, 4+octave)-1; + qg = ec_dec_bits(dec, 3); + if (ec_tell(dec)+2<=total_bits) + postfilter_tapset = ec_dec_icdf(dec, tapset_icdf, 2); + postfilter_gain = QCONST16(.09375f,15)*(qg+1); + } + tell = ec_tell(dec); + } + + if (LM > 0 && tell+3 <= total_bits) + { + isTransient = ec_dec_bit_logp(dec, 3); + tell = ec_tell(dec); + } + else + isTransient = 0; + + if (isTransient) + shortBlocks = M; + else + shortBlocks = 0; + + /* Decode the global flags (first symbols in the stream) */ + intra_ener = tell+3<=total_bits ? ec_dec_bit_logp(dec, 3) : 0; + /* Get band energies */ + unquant_coarse_energy(mode, start, end, oldBandE, + intra_ener, dec, C, LM); + + ALLOC(tf_res, nbEBands, int); + tf_decode(start, end, isTransient, tf_res, LM, dec); + + tell = ec_tell(dec); + spread_decision = SPREAD_NORMAL; + if (tell+4 <= total_bits) + spread_decision = ec_dec_icdf(dec, spread_icdf, 5); + + ALLOC(cap, nbEBands, int); + + init_caps(mode,cap,LM,C); + + ALLOC(offsets, nbEBands, int); + + dynalloc_logp = 6; + total_bits<<=BITRES; + tell = ec_tell_frac(dec); + for (i=start;i<end;i++) + { + int width, quanta; + int dynalloc_loop_logp; + int boost; + width = C*(eBands[i+1]-eBands[i])<<LM; + /* quanta is 6 bits, but no more than 1 bit/sample + and no less than 1/8 bit/sample */ + quanta = IMIN(width<<BITRES, IMAX(6<<BITRES, width)); + dynalloc_loop_logp = dynalloc_logp; + boost = 0; + while (tell+(dynalloc_loop_logp<<BITRES) < total_bits && boost < cap[i]) + { + int flag; + flag = ec_dec_bit_logp(dec, dynalloc_loop_logp); + tell = ec_tell_frac(dec); + if (!flag) + break; + boost += quanta; + total_bits -= quanta; + dynalloc_loop_logp = 1; + } + offsets[i] = boost; + /* Making dynalloc more likely */ + if (boost>0) + dynalloc_logp = IMAX(2, dynalloc_logp-1); + } + + ALLOC(fine_quant, nbEBands, int); + alloc_trim = tell+(6<<BITRES) <= total_bits ? + ec_dec_icdf(dec, trim_icdf, 7) : 5; + + bits = (((opus_int32)len*8)<<BITRES) - ec_tell_frac(dec) - 1; + anti_collapse_rsv = isTransient&&LM>=2&&bits>=((LM+2)<<BITRES) ? (1<<BITRES) : 0; + bits -= anti_collapse_rsv; + + ALLOC(pulses, nbEBands, int); + ALLOC(fine_priority, nbEBands, int); + + codedBands = clt_compute_allocation(mode, start, end, offsets, cap, + alloc_trim, &intensity, &dual_stereo, bits, &balance, pulses, + fine_quant, fine_priority, C, LM, dec, 0, 0, 0); + + unquant_fine_energy(mode, start, end, oldBandE, fine_quant, dec, C); + + c=0; do { + OPUS_MOVE(decode_mem[c], decode_mem[c]+N, DECODE_BUFFER_SIZE-N+overlap/2); + } while (++c<CC); + + /* Decode fixed codebook */ + ALLOC(collapse_masks, C*nbEBands, unsigned char); + +#ifdef NORM_ALIASING_HACK + /* This is an ugly hack that breaks aliasing rules and would be easily broken, + but it saves almost 4kB of stack. */ + X = (celt_norm*)(out_syn[CC-1]+overlap/2); +#else + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ +#endif + + quant_all_bands(0, mode, start, end, X, C==2 ? X+N : NULL, collapse_masks, + NULL, pulses, shortBlocks, spread_decision, dual_stereo, intensity, tf_res, + len*(8<<BITRES)-anti_collapse_rsv, balance, dec, LM, codedBands, &st->rng, 0, + st->arch, st->disable_inv); + + if (anti_collapse_rsv > 0) + { + anti_collapse_on = ec_dec_bits(dec, 1); + } + + unquant_energy_finalise(mode, start, end, oldBandE, + fine_quant, fine_priority, len*8-ec_tell(dec), dec, C); + + if (anti_collapse_on) + anti_collapse(mode, X, collapse_masks, LM, C, N, + start, end, oldBandE, oldLogE, oldLogE2, pulses, st->rng, st->arch); + + if (silence) + { + for (i=0;i<C*nbEBands;i++) + oldBandE[i] = -QCONST16(28.f,DB_SHIFT); + } + + celt_synthesis(mode, X, out_syn, oldBandE, start, effEnd, + C, CC, isTransient, LM, st->downsample, silence, st->arch); + + c=0; do { + st->postfilter_period=IMAX(st->postfilter_period, COMBFILTER_MINPERIOD); + st->postfilter_period_old=IMAX(st->postfilter_period_old, COMBFILTER_MINPERIOD); + comb_filter(out_syn[c], out_syn[c], st->postfilter_period_old, st->postfilter_period, mode->shortMdctSize, + st->postfilter_gain_old, st->postfilter_gain, st->postfilter_tapset_old, st->postfilter_tapset, + mode->window, overlap, st->arch); + if (LM!=0) + comb_filter(out_syn[c]+mode->shortMdctSize, out_syn[c]+mode->shortMdctSize, st->postfilter_period, postfilter_pitch, N-mode->shortMdctSize, + st->postfilter_gain, postfilter_gain, st->postfilter_tapset, postfilter_tapset, + mode->window, overlap, st->arch); + + } while (++c<CC); + st->postfilter_period_old = st->postfilter_period; + st->postfilter_gain_old = st->postfilter_gain; + st->postfilter_tapset_old = st->postfilter_tapset; + st->postfilter_period = postfilter_pitch; + st->postfilter_gain = postfilter_gain; + st->postfilter_tapset = postfilter_tapset; + if (LM!=0) + { + st->postfilter_period_old = st->postfilter_period; + st->postfilter_gain_old = st->postfilter_gain; + st->postfilter_tapset_old = st->postfilter_tapset; + } + + if (C==1) + OPUS_COPY(&oldBandE[nbEBands], oldBandE, nbEBands); + + /* In case start or end were to change */ + if (!isTransient) + { + opus_val16 max_background_increase; + OPUS_COPY(oldLogE2, oldLogE, 2*nbEBands); + OPUS_COPY(oldLogE, oldBandE, 2*nbEBands); + /* In normal circumstances, we only allow the noise floor to increase by + up to 2.4 dB/second, but when we're in DTX, we allow up to 6 dB + increase for each update.*/ + if (st->loss_count < 10) + max_background_increase = M*QCONST16(0.001f,DB_SHIFT); + else + max_background_increase = QCONST16(1.f,DB_SHIFT); + for (i=0;i<2*nbEBands;i++) + backgroundLogE[i] = MIN16(backgroundLogE[i] + max_background_increase, oldBandE[i]); + } else { + for (i=0;i<2*nbEBands;i++) + oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]); + } + c=0; do + { + for (i=0;i<start;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + for (i=end;i<nbEBands;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + } while (++c<2); + st->rng = dec->rng; + + deemphasis(out_syn, pcm, N, CC, st->downsample, mode->preemph, st->preemph_memD, accum); + st->loss_count = 0; + RESTORE_STACK; + if (ec_tell(dec) > 8*len) + return OPUS_INTERNAL_ERROR; + if(ec_get_error(dec)) + st->error = 1; + return frame_size/st->downsample; +} + + +#ifdef CUSTOM_MODES + +#ifdef FIXED_POINT +int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size) +{ + return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL, 0); +} + +#ifndef DISABLE_FLOAT_API +int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, float * OPUS_RESTRICT pcm, int frame_size) +{ + int j, ret, C, N; + VARDECL(opus_int16, out); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + + ALLOC(out, C*N, opus_int16); + ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL, 0); + if (ret>0) + for (j=0;j<C*ret;j++) + pcm[j]=out[j]*(1.f/32768.f); + + RESTORE_STACK; + return ret; +} +#endif /* DISABLE_FLOAT_API */ + +#else + +int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, float * OPUS_RESTRICT pcm, int frame_size) +{ + return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL, 0); +} + +int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size) +{ + int j, ret, C, N; + VARDECL(celt_sig, out); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + ALLOC(out, C*N, celt_sig); + + ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL, 0); + + if (ret>0) + for (j=0;j<C*ret;j++) + pcm[j] = FLOAT2INT16 (out[j]); + + RESTORE_STACK; + return ret; +} + +#endif +#endif /* CUSTOM_MODES */ + +int opus_custom_decoder_ctl(CELTDecoder * OPUS_RESTRICT st, int request, ...) +{ + va_list ap; + + va_start(ap, request); + switch (request) + { + case CELT_SET_START_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<0 || value>=st->mode->nbEBands) + goto bad_arg; + st->start = value; + } + break; + case CELT_SET_END_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>st->mode->nbEBands) + goto bad_arg; + st->end = value; + } + break; + case CELT_SET_CHANNELS_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>2) + goto bad_arg; + st->stream_channels = value; + } + break; + case CELT_GET_AND_CLEAR_ERROR_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value=st->error; + st->error = 0; + } + break; + case OPUS_GET_LOOKAHEAD_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value = st->overlap/st->downsample; + } + break; + case OPUS_RESET_STATE: + { + int i; + opus_val16 *lpc, *oldBandE, *oldLogE, *oldLogE2; + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+st->overlap)*st->channels); + oldBandE = lpc+st->channels*LPC_ORDER; + oldLogE = oldBandE + 2*st->mode->nbEBands; + oldLogE2 = oldLogE + 2*st->mode->nbEBands; + OPUS_CLEAR((char*)&st->DECODER_RESET_START, + opus_custom_decoder_get_size(st->mode, st->channels)- + ((char*)&st->DECODER_RESET_START - (char*)st)); + for (i=0;i<2*st->mode->nbEBands;i++) + oldLogE[i]=oldLogE2[i]=-QCONST16(28.f,DB_SHIFT); + st->skip_plc = 1; + } + break; + case OPUS_GET_PITCH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value = st->postfilter_period; + } + break; + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (value==0) + goto bad_arg; + *value=st->mode; + } + break; + case CELT_SET_SIGNALLING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->signalling = value; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 * value = va_arg(ap, opus_uint32 *); + if (value==0) + goto bad_arg; + *value=st->rng; + } + break; + case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->disable_inv = value; + } + break; + case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->disable_inv; + } + break; + default: + goto bad_request; + } + va_end(ap); + return OPUS_OK; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +bad_request: + va_end(ap); + return OPUS_UNIMPLEMENTED; +} diff --git a/libopus/celt/celt_encoder.c b/libopus/celt/celt_encoder.c new file mode 100644 index 0000000..34e395c --- /dev/null +++ b/libopus/celt/celt_encoder.c @@ -0,0 +1,2607 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_ENCODER_C + +#include "cpu_support.h" +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + + +/** Encoder state + @brief Encoder state + */ +struct OpusCustomEncoder { + const OpusCustomMode *mode; /**< Mode used by the encoder */ + int channels; + int stream_channels; + + int force_intra; + int clip; + int disable_pf; + int complexity; + int upsample; + int start, end; + + opus_int32 bitrate; + int vbr; + int signalling; + int constrained_vbr; /* If zero, VBR can do whatever it likes with the rate */ + int loss_rate; + int lsb_depth; + int lfe; + int disable_inv; + int arch; + + /* Everything beyond this point gets cleared on a reset */ +#define ENCODER_RESET_START rng + + opus_uint32 rng; + int spread_decision; + opus_val32 delayedIntra; + int tonal_average; + int lastCodedBands; + int hf_average; + int tapset_decision; + + int prefilter_period; + opus_val16 prefilter_gain; + int prefilter_tapset; +#ifdef RESYNTH + int prefilter_period_old; + opus_val16 prefilter_gain_old; + int prefilter_tapset_old; +#endif + int consec_transient; + AnalysisInfo analysis; + SILKInfo silk_info; + + opus_val32 preemph_memE[2]; + opus_val32 preemph_memD[2]; + + /* VBR-related parameters */ + opus_int32 vbr_reservoir; + opus_int32 vbr_drift; + opus_int32 vbr_offset; + opus_int32 vbr_count; + opus_val32 overlap_max; + opus_val16 stereo_saving; + int intensity; + opus_val16 *energy_mask; + opus_val16 spec_avg; + +#ifdef RESYNTH + /* +MAX_PERIOD/2 to make space for overlap */ + celt_sig syn_mem[2][2*MAX_PERIOD+MAX_PERIOD/2]; +#endif + + celt_sig in_mem[1]; /* Size = channels*mode->overlap */ + /* celt_sig prefilter_mem[], Size = channels*COMBFILTER_MAXPERIOD */ + /* opus_val16 oldBandE[], Size = channels*mode->nbEBands */ + /* opus_val16 oldLogE[], Size = channels*mode->nbEBands */ + /* opus_val16 oldLogE2[], Size = channels*mode->nbEBands */ + /* opus_val16 energyError[], Size = channels*mode->nbEBands */ +}; + +int celt_encoder_get_size(int channels) +{ + CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); + return opus_custom_encoder_get_size(mode, channels); +} + +int opus_custom_encoder_get_size(const CELTMode *mode, int channels) +{ + int size = sizeof(struct CELTEncoder) + + (channels*mode->overlap-1)*sizeof(celt_sig) /* celt_sig in_mem[channels*mode->overlap]; */ + + channels*COMBFILTER_MAXPERIOD*sizeof(celt_sig) /* celt_sig prefilter_mem[channels*COMBFILTER_MAXPERIOD]; */ + + 4*channels*mode->nbEBands*sizeof(opus_val16); /* opus_val16 oldBandE[channels*mode->nbEBands]; */ + /* opus_val16 oldLogE[channels*mode->nbEBands]; */ + /* opus_val16 oldLogE2[channels*mode->nbEBands]; */ + /* opus_val16 energyError[channels*mode->nbEBands]; */ + return size; +} + +#ifdef CUSTOM_MODES +CELTEncoder *opus_custom_encoder_create(const CELTMode *mode, int channels, int *error) +{ + int ret; + CELTEncoder *st = (CELTEncoder *)opus_alloc(opus_custom_encoder_get_size(mode, channels)); + /* init will handle the NULL case */ + ret = opus_custom_encoder_init(st, mode, channels); + if (ret != OPUS_OK) + { + opus_custom_encoder_destroy(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} +#endif /* CUSTOM_MODES */ + +static int opus_custom_encoder_init_arch(CELTEncoder *st, const CELTMode *mode, + int channels, int arch) +{ + if (channels < 0 || channels > 2) + return OPUS_BAD_ARG; + + if (st==NULL || mode==NULL) + return OPUS_ALLOC_FAIL; + + OPUS_CLEAR((char*)st, opus_custom_encoder_get_size(mode, channels)); + + st->mode = mode; + st->stream_channels = st->channels = channels; + + st->upsample = 1; + st->start = 0; + st->end = st->mode->effEBands; + st->signalling = 1; + st->arch = arch; + + st->constrained_vbr = 1; + st->clip = 1; + + st->bitrate = OPUS_BITRATE_MAX; + st->vbr = 0; + st->force_intra = 0; + st->complexity = 5; + st->lsb_depth=24; + + opus_custom_encoder_ctl(st, OPUS_RESET_STATE); + + return OPUS_OK; +} + +#ifdef CUSTOM_MODES +int opus_custom_encoder_init(CELTEncoder *st, const CELTMode *mode, int channels) +{ + return opus_custom_encoder_init_arch(st, mode, channels, opus_select_arch()); +} +#endif + +int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels, + int arch) +{ + int ret; + ret = opus_custom_encoder_init_arch(st, + opus_custom_mode_create(48000, 960, NULL), channels, arch); + if (ret != OPUS_OK) + return ret; + st->upsample = resampling_factor(sampling_rate); + return OPUS_OK; +} + +#ifdef CUSTOM_MODES +void opus_custom_encoder_destroy(CELTEncoder *st) +{ + opus_free(st); +} +#endif /* CUSTOM_MODES */ + + +static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int C, + opus_val16 *tf_estimate, int *tf_chan, int allow_weak_transients, + int *weak_transient) +{ + int i; + VARDECL(opus_val16, tmp); + opus_val32 mem0,mem1; + int is_transient = 0; + opus_int32 mask_metric = 0; + int c; + opus_val16 tf_max; + int len2; + /* Forward masking: 6.7 dB/ms. */ +#ifdef FIXED_POINT + int forward_shift = 4; +#else + opus_val16 forward_decay = QCONST16(.0625f,15); +#endif + /* Table of 6*64/x, trained on real data to minimize the average error */ + static const unsigned char inv_table[128] = { + 255,255,156,110, 86, 70, 59, 51, 45, 40, 37, 33, 31, 28, 26, 25, + 23, 22, 21, 20, 19, 18, 17, 16, 16, 15, 15, 14, 13, 13, 12, 12, + 12, 12, 11, 11, 11, 10, 10, 10, 9, 9, 9, 9, 9, 9, 8, 8, + 8, 8, 8, 7, 7, 7, 7, 7, 7, 6, 6, 6, 6, 6, 6, 6, + 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, + 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, + 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 2, + }; + SAVE_STACK; + ALLOC(tmp, len, opus_val16); + + *weak_transient = 0; + /* For lower bitrates, let's be more conservative and have a forward masking + decay of 3.3 dB/ms. This avoids having to code transients at very low + bitrate (mostly for hybrid), which can result in unstable energy and/or + partial collapse. */ + if (allow_weak_transients) + { +#ifdef FIXED_POINT + forward_shift = 5; +#else + forward_decay = QCONST16(.03125f,15); +#endif + } + len2=len/2; + for (c=0;c<C;c++) + { + opus_val32 mean; + opus_int32 unmask=0; + opus_val32 norm; + opus_val16 maxE; + mem0=0; + mem1=0; + /* High-pass filter: (1 - 2*z^-1 + z^-2) / (1 - z^-1 + .5*z^-2) */ + for (i=0;i<len;i++) + { + opus_val32 x,y; + x = SHR32(in[i+c*len],SIG_SHIFT); + y = ADD32(mem0, x); +#ifdef FIXED_POINT + mem0 = mem1 + y - SHL32(x,1); + mem1 = x - SHR32(y,1); +#else + mem0 = mem1 + y - 2*x; + mem1 = x - .5f*y; +#endif + tmp[i] = SROUND16(y, 2); + /*printf("%f ", tmp[i]);*/ + } + /*printf("\n");*/ + /* First few samples are bad because we don't propagate the memory */ + OPUS_CLEAR(tmp, 12); + +#ifdef FIXED_POINT + /* Normalize tmp to max range */ + { + int shift=0; + shift = 14-celt_ilog2(MAX16(1, celt_maxabs16(tmp, len))); + if (shift!=0) + { + for (i=0;i<len;i++) + tmp[i] = SHL16(tmp[i], shift); + } + } +#endif + + mean=0; + mem0=0; + /* Grouping by two to reduce complexity */ + /* Forward pass to compute the post-echo threshold*/ + for (i=0;i<len2;i++) + { + opus_val16 x2 = PSHR32(MULT16_16(tmp[2*i],tmp[2*i]) + MULT16_16(tmp[2*i+1],tmp[2*i+1]),16); + mean += x2; +#ifdef FIXED_POINT + /* FIXME: Use PSHR16() instead */ + tmp[i] = mem0 + PSHR32(x2-mem0,forward_shift); +#else + tmp[i] = mem0 + MULT16_16_P15(forward_decay,x2-mem0); +#endif + mem0 = tmp[i]; + } + + mem0=0; + maxE=0; + /* Backward pass to compute the pre-echo threshold */ + for (i=len2-1;i>=0;i--) + { + /* Backward masking: 13.9 dB/ms. */ +#ifdef FIXED_POINT + /* FIXME: Use PSHR16() instead */ + tmp[i] = mem0 + PSHR32(tmp[i]-mem0,3); +#else + tmp[i] = mem0 + MULT16_16_P15(QCONST16(0.125f,15),tmp[i]-mem0); +#endif + mem0 = tmp[i]; + maxE = MAX16(maxE, mem0); + } + /*for (i=0;i<len2;i++)printf("%f ", tmp[i]/mean);printf("\n");*/ + + /* Compute the ratio of the "frame energy" over the harmonic mean of the energy. + This essentially corresponds to a bitrate-normalized temporal noise-to-mask + ratio */ + + /* As a compromise with the old transient detector, frame energy is the + geometric mean of the energy and half the max */ +#ifdef FIXED_POINT + /* Costs two sqrt() to avoid overflows */ + mean = MULT16_16(celt_sqrt(mean), celt_sqrt(MULT16_16(maxE,len2>>1))); +#else + mean = celt_sqrt(mean * maxE*.5*len2); +#endif + /* Inverse of the mean energy in Q15+6 */ + norm = SHL32(EXTEND32(len2),6+14)/ADD32(EPSILON,SHR32(mean,1)); + /* Compute harmonic mean discarding the unreliable boundaries + The data is smooth, so we only take 1/4th of the samples */ + unmask=0; + /* We should never see NaNs here. If we find any, then something really bad happened and we better abort + before it does any damage later on. If these asserts are disabled (no hardening), then the table + lookup a few lines below (id = ...) is likely to crash dur to an out-of-bounds read. DO NOT FIX + that crash on NaN since it could result in a worse issue later on. */ + celt_assert(!celt_isnan(tmp[0])); + celt_assert(!celt_isnan(norm)); + for (i=12;i<len2-5;i+=4) + { + int id; +#ifdef FIXED_POINT + id = MAX32(0,MIN32(127,MULT16_32_Q15(tmp[i]+EPSILON,norm))); /* Do not round to nearest */ +#else + id = (int)MAX32(0,MIN32(127,floor(64*norm*(tmp[i]+EPSILON)))); /* Do not round to nearest */ +#endif + unmask += inv_table[id]; + } + /*printf("%d\n", unmask);*/ + /* Normalize, compensate for the 1/4th of the sample and the factor of 6 in the inverse table */ + unmask = 64*unmask*4/(6*(len2-17)); + if (unmask>mask_metric) + { + *tf_chan = c; + mask_metric = unmask; + } + } + is_transient = mask_metric>200; + /* For low bitrates, define "weak transients" that need to be + handled differently to avoid partial collapse. */ + if (allow_weak_transients && is_transient && mask_metric<600) { + is_transient = 0; + *weak_transient = 1; + } + /* Arbitrary metric for VBR boost */ + tf_max = MAX16(0,celt_sqrt(27*mask_metric)-42); + /* *tf_estimate = 1 + MIN16(1, sqrt(MAX16(0, tf_max-30))/20); */ + *tf_estimate = celt_sqrt(MAX32(0, SHL32(MULT16_16(QCONST16(0.0069,14),MIN16(163,tf_max)),14)-QCONST32(0.139,28))); + /*printf("%d %f\n", tf_max, mask_metric);*/ + RESTORE_STACK; +#ifdef FUZZING + is_transient = rand()&0x1; +#endif + /*printf("%d %f %d\n", is_transient, (float)*tf_estimate, tf_max);*/ + return is_transient; +} + +/* Looks for sudden increases of energy to decide whether we need to patch + the transient decision */ +static int patch_transient_decision(opus_val16 *newE, opus_val16 *oldE, int nbEBands, + int start, int end, int C) +{ + int i, c; + opus_val32 mean_diff=0; + opus_val16 spread_old[26]; + /* Apply an aggressive (-6 dB/Bark) spreading function to the old frame to + avoid false detection caused by irrelevant bands */ + if (C==1) + { + spread_old[start] = oldE[start]; + for (i=start+1;i<end;i++) + spread_old[i] = MAX16(spread_old[i-1]-QCONST16(1.0f, DB_SHIFT), oldE[i]); + } else { + spread_old[start] = MAX16(oldE[start],oldE[start+nbEBands]); + for (i=start+1;i<end;i++) + spread_old[i] = MAX16(spread_old[i-1]-QCONST16(1.0f, DB_SHIFT), + MAX16(oldE[i],oldE[i+nbEBands])); + } + for (i=end-2;i>=start;i--) + spread_old[i] = MAX16(spread_old[i], spread_old[i+1]-QCONST16(1.0f, DB_SHIFT)); + /* Compute mean increase */ + c=0; do { + for (i=IMAX(2,start);i<end-1;i++) + { + opus_val16 x1, x2; + x1 = MAX16(0, newE[i + c*nbEBands]); + x2 = MAX16(0, spread_old[i]); + mean_diff = ADD32(mean_diff, EXTEND32(MAX16(0, SUB16(x1, x2)))); + } + } while (++c<C); + mean_diff = DIV32(mean_diff, C*(end-1-IMAX(2,start))); + /*printf("%f %f %d\n", mean_diff, max_diff, count);*/ + return mean_diff > QCONST16(1.f, DB_SHIFT); +} + +/** Apply window and compute the MDCT for all sub-frames and + all channels in a frame */ +static void compute_mdcts(const CELTMode *mode, int shortBlocks, celt_sig * OPUS_RESTRICT in, + celt_sig * OPUS_RESTRICT out, int C, int CC, int LM, int upsample, + int arch) +{ + const int overlap = mode->overlap; + int N; + int B; + int shift; + int i, b, c; + if (shortBlocks) + { + B = shortBlocks; + N = mode->shortMdctSize; + shift = mode->maxLM; + } else { + B = 1; + N = mode->shortMdctSize<<LM; + shift = mode->maxLM-LM; + } + c=0; do { + for (b=0;b<B;b++) + { + /* Interleaving the sub-frames while doing the MDCTs */ + clt_mdct_forward(&mode->mdct, in+c*(B*N+overlap)+b*N, + &out[b+c*N*B], mode->window, overlap, shift, B, + arch); + } + } while (++c<CC); + if (CC==2&&C==1) + { + for (i=0;i<B*N;i++) + out[i] = ADD32(HALF32(out[i]), HALF32(out[B*N+i])); + } + if (upsample != 1) + { + c=0; do + { + int bound = B*N/upsample; + for (i=0;i<bound;i++) + out[c*B*N+i] *= upsample; + OPUS_CLEAR(&out[c*B*N+bound], B*N-bound); + } while (++c<C); + } +} + + +void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RESTRICT inp, + int N, int CC, int upsample, const opus_val16 *coef, celt_sig *mem, int clip) +{ + int i; + opus_val16 coef0; + celt_sig m; + int Nu; + + coef0 = coef[0]; + m = *mem; + + /* Fast path for the normal 48kHz case and no clipping */ + if (coef[1] == 0 && upsample == 1 && !clip) + { + for (i=0;i<N;i++) + { + opus_val16 x; + x = SCALEIN(pcmp[CC*i]); + /* Apply pre-emphasis */ + inp[i] = SHL32(x, SIG_SHIFT) - m; + m = SHR32(MULT16_16(coef0, x), 15-SIG_SHIFT); + } + *mem = m; + return; + } + + Nu = N/upsample; + if (upsample!=1) + { + OPUS_CLEAR(inp, N); + } + for (i=0;i<Nu;i++) + inp[i*upsample] = SCALEIN(pcmp[CC*i]); + +#ifndef FIXED_POINT + if (clip) + { + /* Clip input to avoid encoding non-portable files */ + for (i=0;i<Nu;i++) + inp[i*upsample] = MAX32(-65536.f, MIN32(65536.f,inp[i*upsample])); + } +#else + (void)clip; /* Avoids a warning about clip being unused. */ +#endif +#ifdef CUSTOM_MODES + if (coef[1] != 0) + { + opus_val16 coef1 = coef[1]; + opus_val16 coef2 = coef[2]; + for (i=0;i<N;i++) + { + celt_sig x, tmp; + x = inp[i]; + /* Apply pre-emphasis */ + tmp = MULT16_16(coef2, x); + inp[i] = tmp + m; + m = MULT16_32_Q15(coef1, inp[i]) - MULT16_32_Q15(coef0, tmp); + } + } else +#endif + { + for (i=0;i<N;i++) + { + opus_val16 x; + x = inp[i]; + /* Apply pre-emphasis */ + inp[i] = SHL32(x, SIG_SHIFT) - m; + m = SHR32(MULT16_16(coef0, x), 15-SIG_SHIFT); + } + } + *mem = m; +} + + + +static opus_val32 l1_metric(const celt_norm *tmp, int N, int LM, opus_val16 bias) +{ + int i; + opus_val32 L1; + L1 = 0; + for (i=0;i<N;i++) + L1 += EXTEND32(ABS16(tmp[i])); + /* When in doubt, prefer good freq resolution */ + L1 = MAC16_32_Q15(L1, LM*bias, L1); + return L1; + +} + +static int tf_analysis(const CELTMode *m, int len, int isTransient, + int *tf_res, int lambda, celt_norm *X, int N0, int LM, + opus_val16 tf_estimate, int tf_chan, int *importance) +{ + int i; + VARDECL(int, metric); + int cost0; + int cost1; + VARDECL(int, path0); + VARDECL(int, path1); + VARDECL(celt_norm, tmp); + VARDECL(celt_norm, tmp_1); + int sel; + int selcost[2]; + int tf_select=0; + opus_val16 bias; + + SAVE_STACK; + bias = MULT16_16_Q14(QCONST16(.04f,15), MAX16(-QCONST16(.25f,14), QCONST16(.5f,14)-tf_estimate)); + /*printf("%f ", bias);*/ + + ALLOC(metric, len, int); + ALLOC(tmp, (m->eBands[len]-m->eBands[len-1])<<LM, celt_norm); + ALLOC(tmp_1, (m->eBands[len]-m->eBands[len-1])<<LM, celt_norm); + ALLOC(path0, len, int); + ALLOC(path1, len, int); + + for (i=0;i<len;i++) + { + int k, N; + int narrow; + opus_val32 L1, best_L1; + int best_level=0; + N = (m->eBands[i+1]-m->eBands[i])<<LM; + /* band is too narrow to be split down to LM=-1 */ + narrow = (m->eBands[i+1]-m->eBands[i])==1; + OPUS_COPY(tmp, &X[tf_chan*N0 + (m->eBands[i]<<LM)], N); + /* Just add the right channel if we're in stereo */ + /*if (C==2) + for (j=0;j<N;j++) + tmp[j] = ADD16(SHR16(tmp[j], 1),SHR16(X[N0+j+(m->eBands[i]<<LM)], 1));*/ + L1 = l1_metric(tmp, N, isTransient ? LM : 0, bias); + best_L1 = L1; + /* Check the -1 case for transients */ + if (isTransient && !narrow) + { + OPUS_COPY(tmp_1, tmp, N); + haar1(tmp_1, N>>LM, 1<<LM); + L1 = l1_metric(tmp_1, N, LM+1, bias); + if (L1<best_L1) + { + best_L1 = L1; + best_level = -1; + } + } + /*printf ("%f ", L1);*/ + for (k=0;k<LM+!(isTransient||narrow);k++) + { + int B; + + if (isTransient) + B = (LM-k-1); + else + B = k+1; + + haar1(tmp, N>>k, 1<<k); + + L1 = l1_metric(tmp, N, B, bias); + + if (L1 < best_L1) + { + best_L1 = L1; + best_level = k+1; + } + } + /*printf ("%d ", isTransient ? LM-best_level : best_level);*/ + /* metric is in Q1 to be able to select the mid-point (-0.5) for narrower bands */ + if (isTransient) + metric[i] = 2*best_level; + else + metric[i] = -2*best_level; + /* For bands that can't be split to -1, set the metric to the half-way point to avoid + biasing the decision */ + if (narrow && (metric[i]==0 || metric[i]==-2*LM)) + metric[i]-=1; + /*printf("%d ", metric[i]/2 + (!isTransient)*LM);*/ + } + /*printf("\n");*/ + /* Search for the optimal tf resolution, including tf_select */ + tf_select = 0; + for (sel=0;sel<2;sel++) + { + cost0 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*sel+0]); + cost1 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*sel+1]) + (isTransient ? 0 : lambda); + for (i=1;i<len;i++) + { + int curr0, curr1; + curr0 = IMIN(cost0, cost1 + lambda); + curr1 = IMIN(cost0 + lambda, cost1); + cost0 = curr0 + importance[i]*abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*sel+0]); + cost1 = curr1 + importance[i]*abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*sel+1]); + } + cost0 = IMIN(cost0, cost1); + selcost[sel]=cost0; + } + /* For now, we're conservative and only allow tf_select=1 for transients. + * If tests confirm it's useful for non-transients, we could allow it. */ + if (selcost[1]<selcost[0] && isTransient) + tf_select=1; + cost0 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*tf_select+0]); + cost1 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*tf_select+1]) + (isTransient ? 0 : lambda); + /* Viterbi forward pass */ + for (i=1;i<len;i++) + { + int curr0, curr1; + int from0, from1; + + from0 = cost0; + from1 = cost1 + lambda; + if (from0 < from1) + { + curr0 = from0; + path0[i]= 0; + } else { + curr0 = from1; + path0[i]= 1; + } + + from0 = cost0 + lambda; + from1 = cost1; + if (from0 < from1) + { + curr1 = from0; + path1[i]= 0; + } else { + curr1 = from1; + path1[i]= 1; + } + cost0 = curr0 + importance[i]*abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*tf_select+0]); + cost1 = curr1 + importance[i]*abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*tf_select+1]); + } + tf_res[len-1] = cost0 < cost1 ? 0 : 1; + /* Viterbi backward pass to check the decisions */ + for (i=len-2;i>=0;i--) + { + if (tf_res[i+1] == 1) + tf_res[i] = path1[i+1]; + else + tf_res[i] = path0[i+1]; + } + /*printf("%d %f\n", *tf_sum, tf_estimate);*/ + RESTORE_STACK; +#ifdef FUZZING + tf_select = rand()&0x1; + tf_res[0] = rand()&0x1; + for (i=1;i<len;i++) + tf_res[i] = tf_res[i-1] ^ ((rand()&0xF) == 0); +#endif + return tf_select; +} + +static void tf_encode(int start, int end, int isTransient, int *tf_res, int LM, int tf_select, ec_enc *enc) +{ + int curr, i; + int tf_select_rsv; + int tf_changed; + int logp; + opus_uint32 budget; + opus_uint32 tell; + budget = enc->storage*8; + tell = ec_tell(enc); + logp = isTransient ? 2 : 4; + /* Reserve space to code the tf_select decision. */ + tf_select_rsv = LM>0 && tell+logp+1 <= budget; + budget -= tf_select_rsv; + curr = tf_changed = 0; + for (i=start;i<end;i++) + { + if (tell+logp<=budget) + { + ec_enc_bit_logp(enc, tf_res[i] ^ curr, logp); + tell = ec_tell(enc); + curr = tf_res[i]; + tf_changed |= curr; + } + else + tf_res[i] = curr; + logp = isTransient ? 4 : 5; + } + /* Only code tf_select if it would actually make a difference. */ + if (tf_select_rsv && + tf_select_table[LM][4*isTransient+0+tf_changed]!= + tf_select_table[LM][4*isTransient+2+tf_changed]) + ec_enc_bit_logp(enc, tf_select, 1); + else + tf_select = 0; + for (i=start;i<end;i++) + tf_res[i] = tf_select_table[LM][4*isTransient+2*tf_select+tf_res[i]]; + /*for(i=0;i<end;i++)printf("%d ", isTransient ? tf_res[i] : LM+tf_res[i]);printf("\n");*/ +} + + +static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X, + const opus_val16 *bandLogE, int end, int LM, int C, int N0, + AnalysisInfo *analysis, opus_val16 *stereo_saving, opus_val16 tf_estimate, + int intensity, opus_val16 surround_trim, opus_int32 equiv_rate, int arch) +{ + int i; + opus_val32 diff=0; + int c; + int trim_index; + opus_val16 trim = QCONST16(5.f, 8); + opus_val16 logXC, logXC2; + /* At low bitrate, reducing the trim seems to help. At higher bitrates, it's less + clear what's best, so we're keeping it as it was before, at least for now. */ + if (equiv_rate < 64000) { + trim = QCONST16(4.f, 8); + } else if (equiv_rate < 80000) { + opus_int32 frac = (equiv_rate-64000) >> 10; + trim = QCONST16(4.f, 8) + QCONST16(1.f/16.f, 8)*frac; + } + if (C==2) + { + opus_val16 sum = 0; /* Q10 */ + opus_val16 minXC; /* Q10 */ + /* Compute inter-channel correlation for low frequencies */ + for (i=0;i<8;i++) + { + opus_val32 partial; + partial = celt_inner_prod(&X[m->eBands[i]<<LM], &X[N0+(m->eBands[i]<<LM)], + (m->eBands[i+1]-m->eBands[i])<<LM, arch); + sum = ADD16(sum, EXTRACT16(SHR32(partial, 18))); + } + sum = MULT16_16_Q15(QCONST16(1.f/8, 15), sum); + sum = MIN16(QCONST16(1.f, 10), ABS16(sum)); + minXC = sum; + for (i=8;i<intensity;i++) + { + opus_val32 partial; + partial = celt_inner_prod(&X[m->eBands[i]<<LM], &X[N0+(m->eBands[i]<<LM)], + (m->eBands[i+1]-m->eBands[i])<<LM, arch); + minXC = MIN16(minXC, ABS16(EXTRACT16(SHR32(partial, 18)))); + } + minXC = MIN16(QCONST16(1.f, 10), ABS16(minXC)); + /*printf ("%f\n", sum);*/ + /* mid-side savings estimations based on the LF average*/ + logXC = celt_log2(QCONST32(1.001f, 20)-MULT16_16(sum, sum)); + /* mid-side savings estimations based on min correlation */ + logXC2 = MAX16(HALF16(logXC), celt_log2(QCONST32(1.001f, 20)-MULT16_16(minXC, minXC))); +#ifdef FIXED_POINT + /* Compensate for Q20 vs Q14 input and convert output to Q8 */ + logXC = PSHR32(logXC-QCONST16(6.f, DB_SHIFT),DB_SHIFT-8); + logXC2 = PSHR32(logXC2-QCONST16(6.f, DB_SHIFT),DB_SHIFT-8); +#endif + + trim += MAX16(-QCONST16(4.f, 8), MULT16_16_Q15(QCONST16(.75f,15),logXC)); + *stereo_saving = MIN16(*stereo_saving + QCONST16(0.25f, 8), -HALF16(logXC2)); + } + + /* Estimate spectral tilt */ + c=0; do { + for (i=0;i<end-1;i++) + { + diff += bandLogE[i+c*m->nbEBands]*(opus_int32)(2+2*i-end); + } + } while (++c<C); + diff /= C*(end-1); + /*printf("%f\n", diff);*/ + trim -= MAX32(-QCONST16(2.f, 8), MIN32(QCONST16(2.f, 8), SHR32(diff+QCONST16(1.f, DB_SHIFT),DB_SHIFT-8)/6 )); + trim -= SHR16(surround_trim, DB_SHIFT-8); + trim -= 2*SHR16(tf_estimate, 14-8); +#ifndef DISABLE_FLOAT_API + if (analysis->valid) + { + trim -= MAX16(-QCONST16(2.f, 8), MIN16(QCONST16(2.f, 8), + (opus_val16)(QCONST16(2.f, 8)*(analysis->tonality_slope+.05f)))); + } +#else + (void)analysis; +#endif + +#ifdef FIXED_POINT + trim_index = PSHR32(trim, 8); +#else + trim_index = (int)floor(.5f+trim); +#endif + trim_index = IMAX(0, IMIN(10, trim_index)); + /*printf("%d\n", trim_index);*/ +#ifdef FUZZING + trim_index = rand()%11; +#endif + return trim_index; +} + +static int stereo_analysis(const CELTMode *m, const celt_norm *X, + int LM, int N0) +{ + int i; + int thetas; + opus_val32 sumLR = EPSILON, sumMS = EPSILON; + + /* Use the L1 norm to model the entropy of the L/R signal vs the M/S signal */ + for (i=0;i<13;i++) + { + int j; + for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++) + { + opus_val32 L, R, M, S; + /* We cast to 32-bit first because of the -32768 case */ + L = EXTEND32(X[j]); + R = EXTEND32(X[N0+j]); + M = ADD32(L, R); + S = SUB32(L, R); + sumLR = ADD32(sumLR, ADD32(ABS32(L), ABS32(R))); + sumMS = ADD32(sumMS, ADD32(ABS32(M), ABS32(S))); + } + } + sumMS = MULT16_32_Q15(QCONST16(0.707107f, 15), sumMS); + thetas = 13; + /* We don't need thetas for lower bands with LM<=1 */ + if (LM<=1) + thetas -= 8; + return MULT16_32_Q15((m->eBands[13]<<(LM+1))+thetas, sumMS) + > MULT16_32_Q15(m->eBands[13]<<(LM+1), sumLR); +} + +#define MSWAP(a,b) do {opus_val16 tmp = a;a=b;b=tmp;} while(0) +static opus_val16 median_of_5(const opus_val16 *x) +{ + opus_val16 t0, t1, t2, t3, t4; + t2 = x[2]; + if (x[0] > x[1]) + { + t0 = x[1]; + t1 = x[0]; + } else { + t0 = x[0]; + t1 = x[1]; + } + if (x[3] > x[4]) + { + t3 = x[4]; + t4 = x[3]; + } else { + t3 = x[3]; + t4 = x[4]; + } + if (t0 > t3) + { + MSWAP(t0, t3); + MSWAP(t1, t4); + } + if (t2 > t1) + { + if (t1 < t3) + return MIN16(t2, t3); + else + return MIN16(t4, t1); + } else { + if (t2 < t3) + return MIN16(t1, t3); + else + return MIN16(t2, t4); + } +} + +static opus_val16 median_of_3(const opus_val16 *x) +{ + opus_val16 t0, t1, t2; + if (x[0] > x[1]) + { + t0 = x[1]; + t1 = x[0]; + } else { + t0 = x[0]; + t1 = x[1]; + } + t2 = x[2]; + if (t1 < t2) + return t1; + else if (t0 < t2) + return t2; + else + return t0; +} + +static opus_val16 dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16 *bandLogE2, + int nbEBands, int start, int end, int C, int *offsets, int lsb_depth, const opus_int16 *logN, + int isTransient, int vbr, int constrained_vbr, const opus_int16 *eBands, int LM, + int effectiveBytes, opus_int32 *tot_boost_, int lfe, opus_val16 *surround_dynalloc, + AnalysisInfo *analysis, int *importance, int *spread_weight) +{ + int i, c; + opus_int32 tot_boost=0; + opus_val16 maxDepth; + VARDECL(opus_val16, follower); + VARDECL(opus_val16, noise_floor); + SAVE_STACK; + ALLOC(follower, C*nbEBands, opus_val16); + ALLOC(noise_floor, C*nbEBands, opus_val16); + OPUS_CLEAR(offsets, nbEBands); + /* Dynamic allocation code */ + maxDepth=-QCONST16(31.9f, DB_SHIFT); + for (i=0;i<end;i++) + { + /* Noise floor must take into account eMeans, the depth, the width of the bands + and the preemphasis filter (approx. square of bark band ID) */ + noise_floor[i] = MULT16_16(QCONST16(0.0625f, DB_SHIFT),logN[i]) + +QCONST16(.5f,DB_SHIFT)+SHL16(9-lsb_depth,DB_SHIFT)-SHL16(eMeans[i],6) + +MULT16_16(QCONST16(.0062,DB_SHIFT),(i+5)*(i+5)); + } + c=0;do + { + for (i=0;i<end;i++) + maxDepth = MAX16(maxDepth, bandLogE[c*nbEBands+i]-noise_floor[i]); + } while (++c<C); + { + /* Compute a really simple masking model to avoid taking into account completely masked + bands when computing the spreading decision. */ + VARDECL(opus_val16, mask); + VARDECL(opus_val16, sig); + ALLOC(mask, nbEBands, opus_val16); + ALLOC(sig, nbEBands, opus_val16); + for (i=0;i<end;i++) + mask[i] = bandLogE[i]-noise_floor[i]; + if (C==2) + { + for (i=0;i<end;i++) + mask[i] = MAX16(mask[i], bandLogE[nbEBands+i]-noise_floor[i]); + } + OPUS_COPY(sig, mask, end); + for (i=1;i<end;i++) + mask[i] = MAX16(mask[i], mask[i-1] - QCONST16(2.f, DB_SHIFT)); + for (i=end-2;i>=0;i--) + mask[i] = MAX16(mask[i], mask[i+1] - QCONST16(3.f, DB_SHIFT)); + for (i=0;i<end;i++) + { + /* Compute SMR: Mask is never more than 72 dB below the peak and never below the noise floor.*/ + opus_val16 smr = sig[i]-MAX16(MAX16(0, maxDepth-QCONST16(12.f, DB_SHIFT)), mask[i]); + /* Clamp SMR to make sure we're not shifting by something negative or too large. */ +#ifdef FIXED_POINT + /* FIXME: Use PSHR16() instead */ + int shift = -PSHR32(MAX16(-QCONST16(5.f, DB_SHIFT), MIN16(0, smr)), DB_SHIFT); +#else + int shift = IMIN(5, IMAX(0, -(int)floor(.5f + smr))); +#endif + spread_weight[i] = 32 >> shift; + } + /*for (i=0;i<end;i++) + printf("%d ", spread_weight[i]); + printf("\n");*/ + } + /* Make sure that dynamic allocation can't make us bust the budget */ + if (effectiveBytes > 50 && LM>=1 && !lfe) + { + int last=0; + c=0;do + { + opus_val16 offset; + opus_val16 tmp; + opus_val16 *f; + f = &follower[c*nbEBands]; + f[0] = bandLogE2[c*nbEBands]; + for (i=1;i<end;i++) + { + /* The last band to be at least 3 dB higher than the previous one + is the last we'll consider. Otherwise, we run into problems on + bandlimited signals. */ + if (bandLogE2[c*nbEBands+i] > bandLogE2[c*nbEBands+i-1]+QCONST16(.5f,DB_SHIFT)) + last=i; + f[i] = MIN16(f[i-1]+QCONST16(1.5f,DB_SHIFT), bandLogE2[c*nbEBands+i]); + } + for (i=last-1;i>=0;i--) + f[i] = MIN16(f[i], MIN16(f[i+1]+QCONST16(2.f,DB_SHIFT), bandLogE2[c*nbEBands+i])); + + /* Combine with a median filter to avoid dynalloc triggering unnecessarily. + The "offset" value controls how conservative we are -- a higher offset + reduces the impact of the median filter and makes dynalloc use more bits. */ + offset = QCONST16(1.f, DB_SHIFT); + for (i=2;i<end-2;i++) + f[i] = MAX16(f[i], median_of_5(&bandLogE2[c*nbEBands+i-2])-offset); + tmp = median_of_3(&bandLogE2[c*nbEBands])-offset; + f[0] = MAX16(f[0], tmp); + f[1] = MAX16(f[1], tmp); + tmp = median_of_3(&bandLogE2[c*nbEBands+end-3])-offset; + f[end-2] = MAX16(f[end-2], tmp); + f[end-1] = MAX16(f[end-1], tmp); + + for (i=0;i<end;i++) + f[i] = MAX16(f[i], noise_floor[i]); + } while (++c<C); + if (C==2) + { + for (i=start;i<end;i++) + { + /* Consider 24 dB "cross-talk" */ + follower[nbEBands+i] = MAX16(follower[nbEBands+i], follower[ i]-QCONST16(4.f,DB_SHIFT)); + follower[ i] = MAX16(follower[ i], follower[nbEBands+i]-QCONST16(4.f,DB_SHIFT)); + follower[i] = HALF16(MAX16(0, bandLogE[i]-follower[i]) + MAX16(0, bandLogE[nbEBands+i]-follower[nbEBands+i])); + } + } else { + for (i=start;i<end;i++) + { + follower[i] = MAX16(0, bandLogE[i]-follower[i]); + } + } + for (i=start;i<end;i++) + follower[i] = MAX16(follower[i], surround_dynalloc[i]); + for (i=start;i<end;i++) + { +#ifdef FIXED_POINT + importance[i] = PSHR32(13*celt_exp2(MIN16(follower[i], QCONST16(4.f, DB_SHIFT))), 16); +#else + importance[i] = (int)floor(.5f+13*celt_exp2(MIN16(follower[i], QCONST16(4.f, DB_SHIFT)))); +#endif + } + /* For non-transient CBR/CVBR frames, halve the dynalloc contribution */ + if ((!vbr || constrained_vbr)&&!isTransient) + { + for (i=start;i<end;i++) + follower[i] = HALF16(follower[i]); + } + for (i=start;i<end;i++) + { + if (i<8) + follower[i] *= 2; + if (i>=12) + follower[i] = HALF16(follower[i]); + } +#ifdef DISABLE_FLOAT_API + (void)analysis; +#else + if (analysis->valid) + { + for (i=start;i<IMIN(LEAK_BANDS, end);i++) + follower[i] = follower[i] + QCONST16(1.f/64.f, DB_SHIFT)*analysis->leak_boost[i]; + } +#endif + for (i=start;i<end;i++) + { + int width; + int boost; + int boost_bits; + + follower[i] = MIN16(follower[i], QCONST16(4, DB_SHIFT)); + + width = C*(eBands[i+1]-eBands[i])<<LM; + if (width<6) + { + boost = (int)SHR32(EXTEND32(follower[i]),DB_SHIFT); + boost_bits = boost*width<<BITRES; + } else if (width > 48) { + boost = (int)SHR32(EXTEND32(follower[i])*8,DB_SHIFT); + boost_bits = (boost*width<<BITRES)/8; + } else { + boost = (int)SHR32(EXTEND32(follower[i])*width/6,DB_SHIFT); + boost_bits = boost*6<<BITRES; + } + /* For CBR and non-transient CVBR frames, limit dynalloc to 2/3 of the bits */ + if ((!vbr || (constrained_vbr&&!isTransient)) + && (tot_boost+boost_bits)>>BITRES>>3 > 2*effectiveBytes/3) + { + opus_int32 cap = ((2*effectiveBytes/3)<<BITRES<<3); + offsets[i] = cap-tot_boost; + tot_boost = cap; + break; + } else { + offsets[i] = boost; + tot_boost += boost_bits; + } + } + } else { + for (i=start;i<end;i++) + importance[i] = 13; + } + *tot_boost_ = tot_boost; + RESTORE_STACK; + return maxDepth; +} + + +static int run_prefilter(CELTEncoder *st, celt_sig *in, celt_sig *prefilter_mem, int CC, int N, + int prefilter_tapset, int *pitch, opus_val16 *gain, int *qgain, int enabled, int nbAvailableBytes, AnalysisInfo *analysis) +{ + int c; + VARDECL(celt_sig, _pre); + celt_sig *pre[2]; + const CELTMode *mode; + int pitch_index; + opus_val16 gain1; + opus_val16 pf_threshold; + int pf_on; + int qg; + int overlap; + SAVE_STACK; + + mode = st->mode; + overlap = mode->overlap; + ALLOC(_pre, CC*(N+COMBFILTER_MAXPERIOD), celt_sig); + + pre[0] = _pre; + pre[1] = _pre + (N+COMBFILTER_MAXPERIOD); + + + c=0; do { + OPUS_COPY(pre[c], prefilter_mem+c*COMBFILTER_MAXPERIOD, COMBFILTER_MAXPERIOD); + OPUS_COPY(pre[c]+COMBFILTER_MAXPERIOD, in+c*(N+overlap)+overlap, N); + } while (++c<CC); + + if (enabled) + { + VARDECL(opus_val16, pitch_buf); + ALLOC(pitch_buf, (COMBFILTER_MAXPERIOD+N)>>1, opus_val16); + + pitch_downsample(pre, pitch_buf, COMBFILTER_MAXPERIOD+N, CC, st->arch); + /* Don't search for the fir last 1.5 octave of the range because + there's too many false-positives due to short-term correlation */ + pitch_search(pitch_buf+(COMBFILTER_MAXPERIOD>>1), pitch_buf, N, + COMBFILTER_MAXPERIOD-3*COMBFILTER_MINPERIOD, &pitch_index, + st->arch); + pitch_index = COMBFILTER_MAXPERIOD-pitch_index; + + gain1 = remove_doubling(pitch_buf, COMBFILTER_MAXPERIOD, COMBFILTER_MINPERIOD, + N, &pitch_index, st->prefilter_period, st->prefilter_gain, st->arch); + if (pitch_index > COMBFILTER_MAXPERIOD-2) + pitch_index = COMBFILTER_MAXPERIOD-2; + gain1 = MULT16_16_Q15(QCONST16(.7f,15),gain1); + /*printf("%d %d %f %f\n", pitch_change, pitch_index, gain1, st->analysis.tonality);*/ + if (st->loss_rate>2) + gain1 = HALF32(gain1); + if (st->loss_rate>4) + gain1 = HALF32(gain1); + if (st->loss_rate>8) + gain1 = 0; + } else { + gain1 = 0; + pitch_index = COMBFILTER_MINPERIOD; + } +#ifndef DISABLE_FLOAT_API + if (analysis->valid) + gain1 = (opus_val16)(gain1 * analysis->max_pitch_ratio); +#else + (void)analysis; +#endif + /* Gain threshold for enabling the prefilter/postfilter */ + pf_threshold = QCONST16(.2f,15); + + /* Adjusting the threshold based on rate and continuity */ + if (abs(pitch_index-st->prefilter_period)*10>pitch_index) + pf_threshold += QCONST16(.2f,15); + if (nbAvailableBytes<25) + pf_threshold += QCONST16(.1f,15); + if (nbAvailableBytes<35) + pf_threshold += QCONST16(.1f,15); + if (st->prefilter_gain > QCONST16(.4f,15)) + pf_threshold -= QCONST16(.1f,15); + if (st->prefilter_gain > QCONST16(.55f,15)) + pf_threshold -= QCONST16(.1f,15); + + /* Hard threshold at 0.2 */ + pf_threshold = MAX16(pf_threshold, QCONST16(.2f,15)); + if (gain1<pf_threshold) + { + gain1 = 0; + pf_on = 0; + qg = 0; + } else { + /*This block is not gated by a total bits check only because + of the nbAvailableBytes check above.*/ + if (ABS16(gain1-st->prefilter_gain)<QCONST16(.1f,15)) + gain1=st->prefilter_gain; + +#ifdef FIXED_POINT + qg = ((gain1+1536)>>10)/3-1; +#else + qg = (int)floor(.5f+gain1*32/3)-1; +#endif + qg = IMAX(0, IMIN(7, qg)); + gain1 = QCONST16(0.09375f,15)*(qg+1); + pf_on = 1; + } + /*printf("%d %f\n", pitch_index, gain1);*/ + + c=0; do { + int offset = mode->shortMdctSize-overlap; + st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); + OPUS_COPY(in+c*(N+overlap), st->in_mem+c*(overlap), overlap); + if (offset) + comb_filter(in+c*(N+overlap)+overlap, pre[c]+COMBFILTER_MAXPERIOD, + st->prefilter_period, st->prefilter_period, offset, -st->prefilter_gain, -st->prefilter_gain, + st->prefilter_tapset, st->prefilter_tapset, NULL, 0, st->arch); + + comb_filter(in+c*(N+overlap)+overlap+offset, pre[c]+COMBFILTER_MAXPERIOD+offset, + st->prefilter_period, pitch_index, N-offset, -st->prefilter_gain, -gain1, + st->prefilter_tapset, prefilter_tapset, mode->window, overlap, st->arch); + OPUS_COPY(st->in_mem+c*(overlap), in+c*(N+overlap)+N, overlap); + + if (N>COMBFILTER_MAXPERIOD) + { + OPUS_COPY(prefilter_mem+c*COMBFILTER_MAXPERIOD, pre[c]+N, COMBFILTER_MAXPERIOD); + } else { + OPUS_MOVE(prefilter_mem+c*COMBFILTER_MAXPERIOD, prefilter_mem+c*COMBFILTER_MAXPERIOD+N, COMBFILTER_MAXPERIOD-N); + OPUS_COPY(prefilter_mem+c*COMBFILTER_MAXPERIOD+COMBFILTER_MAXPERIOD-N, pre[c]+COMBFILTER_MAXPERIOD, N); + } + } while (++c<CC); + + RESTORE_STACK; + *gain = gain1; + *pitch = pitch_index; + *qgain = qg; + return pf_on; +} + +static int compute_vbr(const CELTMode *mode, AnalysisInfo *analysis, opus_int32 base_target, + int LM, opus_int32 bitrate, int lastCodedBands, int C, int intensity, + int constrained_vbr, opus_val16 stereo_saving, int tot_boost, + opus_val16 tf_estimate, int pitch_change, opus_val16 maxDepth, + int lfe, int has_surround_mask, opus_val16 surround_masking, + opus_val16 temporal_vbr) +{ + /* The target rate in 8th bits per frame */ + opus_int32 target; + int coded_bins; + int coded_bands; + opus_val16 tf_calibration; + int nbEBands; + const opus_int16 *eBands; + + nbEBands = mode->nbEBands; + eBands = mode->eBands; + + coded_bands = lastCodedBands ? lastCodedBands : nbEBands; + coded_bins = eBands[coded_bands]<<LM; + if (C==2) + coded_bins += eBands[IMIN(intensity, coded_bands)]<<LM; + + target = base_target; + + /*printf("%f %f %f %f %d %d ", st->analysis.activity, st->analysis.tonality, tf_estimate, st->stereo_saving, tot_boost, coded_bands);*/ +#ifndef DISABLE_FLOAT_API + if (analysis->valid && analysis->activity<.4) + target -= (opus_int32)((coded_bins<<BITRES)*(.4f-analysis->activity)); +#endif + /* Stereo savings */ + if (C==2) + { + int coded_stereo_bands; + int coded_stereo_dof; + opus_val16 max_frac; + coded_stereo_bands = IMIN(intensity, coded_bands); + coded_stereo_dof = (eBands[coded_stereo_bands]<<LM)-coded_stereo_bands; + /* Maximum fraction of the bits we can save if the signal is mono. */ + max_frac = DIV32_16(MULT16_16(QCONST16(0.8f, 15), coded_stereo_dof), coded_bins); + stereo_saving = MIN16(stereo_saving, QCONST16(1.f, 8)); + /*printf("%d %d %d ", coded_stereo_dof, coded_bins, tot_boost);*/ + target -= (opus_int32)MIN32(MULT16_32_Q15(max_frac,target), + SHR32(MULT16_16(stereo_saving-QCONST16(0.1f,8),(coded_stereo_dof<<BITRES)),8)); + } + /* Boost the rate according to dynalloc (minus the dynalloc average for calibration). */ + target += tot_boost-(19<<LM); + /* Apply transient boost, compensating for average boost. */ + tf_calibration = QCONST16(0.044f,14); + target += (opus_int32)SHL32(MULT16_32_Q15(tf_estimate-tf_calibration, target),1); + +#ifndef DISABLE_FLOAT_API + /* Apply tonality boost */ + if (analysis->valid && !lfe) + { + opus_int32 tonal_target; + float tonal; + + /* Tonality boost (compensating for the average). */ + tonal = MAX16(0.f,analysis->tonality-.15f)-0.12f; + tonal_target = target + (opus_int32)((coded_bins<<BITRES)*1.2f*tonal); + if (pitch_change) + tonal_target += (opus_int32)((coded_bins<<BITRES)*.8f); + /*printf("%f %f ", analysis->tonality, tonal);*/ + target = tonal_target; + } +#else + (void)analysis; + (void)pitch_change; +#endif + + if (has_surround_mask&&!lfe) + { + opus_int32 surround_target = target + (opus_int32)SHR32(MULT16_16(surround_masking,coded_bins<<BITRES), DB_SHIFT); + /*printf("%f %d %d %d %d %d %d ", surround_masking, coded_bins, st->end, st->intensity, surround_target, target, st->bitrate);*/ + target = IMAX(target/4, surround_target); + } + + { + opus_int32 floor_depth; + int bins; + bins = eBands[nbEBands-2]<<LM; + /*floor_depth = SHR32(MULT16_16((C*bins<<BITRES),celt_log2(SHL32(MAX16(1,sample_max),13))), DB_SHIFT);*/ + floor_depth = (opus_int32)SHR32(MULT16_16((C*bins<<BITRES),maxDepth), DB_SHIFT); + floor_depth = IMAX(floor_depth, target>>2); + target = IMIN(target, floor_depth); + /*printf("%f %d\n", maxDepth, floor_depth);*/ + } + + /* Make VBR less aggressive for constrained VBR because we can't keep a higher bitrate + for long. Needs tuning. */ + if ((!has_surround_mask||lfe) && constrained_vbr) + { + target = base_target + (opus_int32)MULT16_32_Q15(QCONST16(0.67f, 15), target-base_target); + } + + if (!has_surround_mask && tf_estimate < QCONST16(.2f, 14)) + { + opus_val16 amount; + opus_val16 tvbr_factor; + amount = MULT16_16_Q15(QCONST16(.0000031f, 30), IMAX(0, IMIN(32000, 96000-bitrate))); + tvbr_factor = SHR32(MULT16_16(temporal_vbr, amount), DB_SHIFT); + target += (opus_int32)MULT16_32_Q15(tvbr_factor, target); + } + + /* Don't allow more than doubling the rate */ + target = IMIN(2*base_target, target); + + return target; +} + +int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes, ec_enc *enc) +{ + int i, c, N; + opus_int32 bits; + ec_enc _enc; + VARDECL(celt_sig, in); + VARDECL(celt_sig, freq); + VARDECL(celt_norm, X); + VARDECL(celt_ener, bandE); + VARDECL(opus_val16, bandLogE); + VARDECL(opus_val16, bandLogE2); + VARDECL(int, fine_quant); + VARDECL(opus_val16, error); + VARDECL(int, pulses); + VARDECL(int, cap); + VARDECL(int, offsets); + VARDECL(int, importance); + VARDECL(int, spread_weight); + VARDECL(int, fine_priority); + VARDECL(int, tf_res); + VARDECL(unsigned char, collapse_masks); + celt_sig *prefilter_mem; + opus_val16 *oldBandE, *oldLogE, *oldLogE2, *energyError; + int shortBlocks=0; + int isTransient=0; + const int CC = st->channels; + const int C = st->stream_channels; + int LM, M; + int tf_select; + int nbFilledBytes, nbAvailableBytes; + int start; + int end; + int effEnd; + int codedBands; + int alloc_trim; + int pitch_index=COMBFILTER_MINPERIOD; + opus_val16 gain1 = 0; + int dual_stereo=0; + int effectiveBytes; + int dynalloc_logp; + opus_int32 vbr_rate; + opus_int32 total_bits; + opus_int32 total_boost; + opus_int32 balance; + opus_int32 tell; + opus_int32 tell0_frac; + int prefilter_tapset=0; + int pf_on; + int anti_collapse_rsv; + int anti_collapse_on=0; + int silence=0; + int tf_chan = 0; + opus_val16 tf_estimate; + int pitch_change=0; + opus_int32 tot_boost; + opus_val32 sample_max; + opus_val16 maxDepth; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + const opus_int16 *eBands; + int secondMdct; + int signalBandwidth; + int transient_got_disabled=0; + opus_val16 surround_masking=0; + opus_val16 temporal_vbr=0; + opus_val16 surround_trim = 0; + opus_int32 equiv_rate; + int hybrid; + int weak_transient = 0; + int enable_tf_analysis; + VARDECL(opus_val16, surround_dynalloc); + ALLOC_STACK; + + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + start = st->start; + end = st->end; + hybrid = start != 0; + tf_estimate = 0; + if (nbCompressedBytes<2 || pcm==NULL) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + frame_size *= st->upsample; + for (LM=0;LM<=mode->maxLM;LM++) + if (mode->shortMdctSize<<LM==frame_size) + break; + if (LM>mode->maxLM) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + M=1<<LM; + N = M*mode->shortMdctSize; + + prefilter_mem = st->in_mem+CC*(overlap); + oldBandE = (opus_val16*)(st->in_mem+CC*(overlap+COMBFILTER_MAXPERIOD)); + oldLogE = oldBandE + CC*nbEBands; + oldLogE2 = oldLogE + CC*nbEBands; + energyError = oldLogE2 + CC*nbEBands; + + if (enc==NULL) + { + tell0_frac=tell=1; + nbFilledBytes=0; + } else { + tell0_frac=ec_tell_frac(enc); + tell=ec_tell(enc); + nbFilledBytes=(tell+4)>>3; + } + +#ifdef CUSTOM_MODES + if (st->signalling && enc==NULL) + { + int tmp = (mode->effEBands-end)>>1; + end = st->end = IMAX(1, mode->effEBands-tmp); + compressed[0] = tmp<<5; + compressed[0] |= LM<<3; + compressed[0] |= (C==2)<<2; + /* Convert "standard mode" to Opus header */ + if (mode->Fs==48000 && mode->shortMdctSize==120) + { + int c0 = toOpus(compressed[0]); + if (c0<0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + compressed[0] = c0; + } + compressed++; + nbCompressedBytes--; + } +#else + celt_assert(st->signalling==0); +#endif + + /* Can't produce more than 1275 output bytes */ + nbCompressedBytes = IMIN(nbCompressedBytes,1275); + nbAvailableBytes = nbCompressedBytes - nbFilledBytes; + + if (st->vbr && st->bitrate!=OPUS_BITRATE_MAX) + { + opus_int32 den=mode->Fs>>BITRES; + vbr_rate=(st->bitrate*frame_size+(den>>1))/den; +#ifdef CUSTOM_MODES + if (st->signalling) + vbr_rate -= 8<<BITRES; +#endif + effectiveBytes = vbr_rate>>(3+BITRES); + } else { + opus_int32 tmp; + vbr_rate = 0; + tmp = st->bitrate*frame_size; + if (tell>1) + tmp += tell; + if (st->bitrate!=OPUS_BITRATE_MAX) + nbCompressedBytes = IMAX(2, IMIN(nbCompressedBytes, + (tmp+4*mode->Fs)/(8*mode->Fs)-!!st->signalling)); + effectiveBytes = nbCompressedBytes - nbFilledBytes; + } + equiv_rate = ((opus_int32)nbCompressedBytes*8*50 >> (3-LM)) - (40*C+20)*((400>>LM) - 50); + if (st->bitrate != OPUS_BITRATE_MAX) + equiv_rate = IMIN(equiv_rate, st->bitrate - (40*C+20)*((400>>LM) - 50)); + + if (enc==NULL) + { + ec_enc_init(&_enc, compressed, nbCompressedBytes); + enc = &_enc; + } + + if (vbr_rate>0) + { + /* Computes the max bit-rate allowed in VBR mode to avoid violating the + target rate and buffering. + We must do this up front so that bust-prevention logic triggers + correctly if we don't have enough bits. */ + if (st->constrained_vbr) + { + opus_int32 vbr_bound; + opus_int32 max_allowed; + /* We could use any multiple of vbr_rate as bound (depending on the + delay). + This is clamped to ensure we use at least two bytes if the encoder + was entirely empty, but to allow 0 in hybrid mode. */ + vbr_bound = vbr_rate; + max_allowed = IMIN(IMAX(tell==1?2:0, + (vbr_rate+vbr_bound-st->vbr_reservoir)>>(BITRES+3)), + nbAvailableBytes); + if(max_allowed < nbAvailableBytes) + { + nbCompressedBytes = nbFilledBytes+max_allowed; + nbAvailableBytes = max_allowed; + ec_enc_shrink(enc, nbCompressedBytes); + } + } + } + total_bits = nbCompressedBytes*8; + + effEnd = end; + if (effEnd > mode->effEBands) + effEnd = mode->effEBands; + + ALLOC(in, CC*(N+overlap), celt_sig); + + sample_max=MAX32(st->overlap_max, celt_maxabs16(pcm, C*(N-overlap)/st->upsample)); + st->overlap_max=celt_maxabs16(pcm+C*(N-overlap)/st->upsample, C*overlap/st->upsample); + sample_max=MAX32(sample_max, st->overlap_max); +#ifdef FIXED_POINT + silence = (sample_max==0); +#else + silence = (sample_max <= (opus_val16)1/(1<<st->lsb_depth)); +#endif +#ifdef FUZZING + if ((rand()&0x3F)==0) + silence = 1; +#endif + if (tell==1) + ec_enc_bit_logp(enc, silence, 15); + else + silence=0; + if (silence) + { + /*In VBR mode there is no need to send more than the minimum. */ + if (vbr_rate>0) + { + effectiveBytes=nbCompressedBytes=IMIN(nbCompressedBytes, nbFilledBytes+2); + total_bits=nbCompressedBytes*8; + nbAvailableBytes=2; + ec_enc_shrink(enc, nbCompressedBytes); + } + /* Pretend we've filled all the remaining bits with zeros + (that's what the initialiser did anyway) */ + tell = nbCompressedBytes*8; + enc->nbits_total+=tell-ec_tell(enc); + } + c=0; do { + int need_clip=0; +#ifndef FIXED_POINT + need_clip = st->clip && sample_max>65536.f; +#endif + celt_preemphasis(pcm+c, in+c*(N+overlap)+overlap, N, CC, st->upsample, + mode->preemph, st->preemph_memE+c, need_clip); + } while (++c<CC); + + + + /* Find pitch period and gain */ + { + int enabled; + int qg; + enabled = ((st->lfe&&nbAvailableBytes>3) || nbAvailableBytes>12*C) && !hybrid && !silence && !st->disable_pf + && st->complexity >= 5; + + prefilter_tapset = st->tapset_decision; + pf_on = run_prefilter(st, in, prefilter_mem, CC, N, prefilter_tapset, &pitch_index, &gain1, &qg, enabled, nbAvailableBytes, &st->analysis); + if ((gain1 > QCONST16(.4f,15) || st->prefilter_gain > QCONST16(.4f,15)) && (!st->analysis.valid || st->analysis.tonality > .3) + && (pitch_index > 1.26*st->prefilter_period || pitch_index < .79*st->prefilter_period)) + pitch_change = 1; + if (pf_on==0) + { + if(!hybrid && tell+16<=total_bits) + ec_enc_bit_logp(enc, 0, 1); + } else { + /*This block is not gated by a total bits check only because + of the nbAvailableBytes check above.*/ + int octave; + ec_enc_bit_logp(enc, 1, 1); + pitch_index += 1; + octave = EC_ILOG(pitch_index)-5; + ec_enc_uint(enc, octave, 6); + ec_enc_bits(enc, pitch_index-(16<<octave), 4+octave); + pitch_index -= 1; + ec_enc_bits(enc, qg, 3); + ec_enc_icdf(enc, prefilter_tapset, tapset_icdf, 2); + } + } + + isTransient = 0; + shortBlocks = 0; + if (st->complexity >= 1 && !st->lfe) + { + /* Reduces the likelihood of energy instability on fricatives at low bitrate + in hybrid mode. It seems like we still want to have real transients on vowels + though (small SILK quantization offset value). */ + int allow_weak_transients = hybrid && effectiveBytes<15 && st->silk_info.signalType != 2; + isTransient = transient_analysis(in, N+overlap, CC, + &tf_estimate, &tf_chan, allow_weak_transients, &weak_transient); + } + if (LM>0 && ec_tell(enc)+3<=total_bits) + { + if (isTransient) + shortBlocks = M; + } else { + isTransient = 0; + transient_got_disabled=1; + } + + ALLOC(freq, CC*N, celt_sig); /**< Interleaved signal MDCTs */ + ALLOC(bandE,nbEBands*CC, celt_ener); + ALLOC(bandLogE,nbEBands*CC, opus_val16); + + secondMdct = shortBlocks && st->complexity>=8; + ALLOC(bandLogE2, C*nbEBands, opus_val16); + if (secondMdct) + { + compute_mdcts(mode, 0, in, freq, C, CC, LM, st->upsample, st->arch); + compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); + amp2Log2(mode, effEnd, end, bandE, bandLogE2, C); + for (i=0;i<C*nbEBands;i++) + bandLogE2[i] += HALF16(SHL16(LM, DB_SHIFT)); + } + + compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample, st->arch); + /* This should catch any NaN in the CELT input. Since we're not supposed to see any (they're filtered + at the Opus layer), just abort. */ + celt_assert(!celt_isnan(freq[0]) && (C==1 || !celt_isnan(freq[N]))); + if (CC==2&&C==1) + tf_chan = 0; + compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); + + if (st->lfe) + { + for (i=2;i<end;i++) + { + bandE[i] = IMIN(bandE[i], MULT16_32_Q15(QCONST16(1e-4f,15),bandE[0])); + bandE[i] = MAX32(bandE[i], EPSILON); + } + } + amp2Log2(mode, effEnd, end, bandE, bandLogE, C); + + ALLOC(surround_dynalloc, C*nbEBands, opus_val16); + OPUS_CLEAR(surround_dynalloc, end); + /* This computes how much masking takes place between surround channels */ + if (!hybrid&&st->energy_mask&&!st->lfe) + { + int mask_end; + int midband; + int count_dynalloc; + opus_val32 mask_avg=0; + opus_val32 diff=0; + int count=0; + mask_end = IMAX(2,st->lastCodedBands); + for (c=0;c<C;c++) + { + for(i=0;i<mask_end;i++) + { + opus_val16 mask; + mask = MAX16(MIN16(st->energy_mask[nbEBands*c+i], + QCONST16(.25f, DB_SHIFT)), -QCONST16(2.0f, DB_SHIFT)); + if (mask > 0) + mask = HALF16(mask); + mask_avg += MULT16_16(mask, eBands[i+1]-eBands[i]); + count += eBands[i+1]-eBands[i]; + diff += MULT16_16(mask, 1+2*i-mask_end); + } + } + celt_assert(count>0); + mask_avg = DIV32_16(mask_avg,count); + mask_avg += QCONST16(.2f, DB_SHIFT); + diff = diff*6/(C*(mask_end-1)*(mask_end+1)*mask_end); + /* Again, being conservative */ + diff = HALF32(diff); + diff = MAX32(MIN32(diff, QCONST32(.031f, DB_SHIFT)), -QCONST32(.031f, DB_SHIFT)); + /* Find the band that's in the middle of the coded spectrum */ + for (midband=0;eBands[midband+1] < eBands[mask_end]/2;midband++); + count_dynalloc=0; + for(i=0;i<mask_end;i++) + { + opus_val32 lin; + opus_val16 unmask; + lin = mask_avg + diff*(i-midband); + if (C==2) + unmask = MAX16(st->energy_mask[i], st->energy_mask[nbEBands+i]); + else + unmask = st->energy_mask[i]; + unmask = MIN16(unmask, QCONST16(.0f, DB_SHIFT)); + unmask -= lin; + if (unmask > QCONST16(.25f, DB_SHIFT)) + { + surround_dynalloc[i] = unmask - QCONST16(.25f, DB_SHIFT); + count_dynalloc++; + } + } + if (count_dynalloc>=3) + { + /* If we need dynalloc in many bands, it's probably because our + initial masking rate was too low. */ + mask_avg += QCONST16(.25f, DB_SHIFT); + if (mask_avg>0) + { + /* Something went really wrong in the original calculations, + disabling masking. */ + mask_avg = 0; + diff = 0; + OPUS_CLEAR(surround_dynalloc, mask_end); + } else { + for(i=0;i<mask_end;i++) + surround_dynalloc[i] = MAX16(0, surround_dynalloc[i]-QCONST16(.25f, DB_SHIFT)); + } + } + mask_avg += QCONST16(.2f, DB_SHIFT); + /* Convert to 1/64th units used for the trim */ + surround_trim = 64*diff; + /*printf("%d %d ", mask_avg, surround_trim);*/ + surround_masking = mask_avg; + } + /* Temporal VBR (but not for LFE) */ + if (!st->lfe) + { + opus_val16 follow=-QCONST16(10.0f,DB_SHIFT); + opus_val32 frame_avg=0; + opus_val16 offset = shortBlocks?HALF16(SHL16(LM, DB_SHIFT)):0; + for(i=start;i<end;i++) + { + follow = MAX16(follow-QCONST16(1.f, DB_SHIFT), bandLogE[i]-offset); + if (C==2) + follow = MAX16(follow, bandLogE[i+nbEBands]-offset); + frame_avg += follow; + } + frame_avg /= (end-start); + temporal_vbr = SUB16(frame_avg,st->spec_avg); + temporal_vbr = MIN16(QCONST16(3.f, DB_SHIFT), MAX16(-QCONST16(1.5f, DB_SHIFT), temporal_vbr)); + st->spec_avg += MULT16_16_Q15(QCONST16(.02f, 15), temporal_vbr); + } + /*for (i=0;i<21;i++) + printf("%f ", bandLogE[i]); + printf("\n");*/ + + if (!secondMdct) + { + OPUS_COPY(bandLogE2, bandLogE, C*nbEBands); + } + + /* Last chance to catch any transient we might have missed in the + time-domain analysis */ + if (LM>0 && ec_tell(enc)+3<=total_bits && !isTransient && st->complexity>=5 && !st->lfe && !hybrid) + { + if (patch_transient_decision(bandLogE, oldBandE, nbEBands, start, end, C)) + { + isTransient = 1; + shortBlocks = M; + compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample, st->arch); + compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); + amp2Log2(mode, effEnd, end, bandE, bandLogE, C); + /* Compensate for the scaling of short vs long mdcts */ + for (i=0;i<C*nbEBands;i++) + bandLogE2[i] += HALF16(SHL16(LM, DB_SHIFT)); + tf_estimate = QCONST16(.2f,14); + } + } + + if (LM>0 && ec_tell(enc)+3<=total_bits) + ec_enc_bit_logp(enc, isTransient, 3); + + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ + + /* Band normalisation */ + normalise_bands(mode, freq, X, bandE, effEnd, C, M); + + enable_tf_analysis = effectiveBytes>=15*C && !hybrid && st->complexity>=2 && !st->lfe; + + ALLOC(offsets, nbEBands, int); + ALLOC(importance, nbEBands, int); + ALLOC(spread_weight, nbEBands, int); + + maxDepth = dynalloc_analysis(bandLogE, bandLogE2, nbEBands, start, end, C, offsets, + st->lsb_depth, mode->logN, isTransient, st->vbr, st->constrained_vbr, + eBands, LM, effectiveBytes, &tot_boost, st->lfe, surround_dynalloc, &st->analysis, importance, spread_weight); + + ALLOC(tf_res, nbEBands, int); + /* Disable variable tf resolution for hybrid and at very low bitrate */ + if (enable_tf_analysis) + { + int lambda; + lambda = IMAX(80, 20480/effectiveBytes + 2); + tf_select = tf_analysis(mode, effEnd, isTransient, tf_res, lambda, X, N, LM, tf_estimate, tf_chan, importance); + for (i=effEnd;i<end;i++) + tf_res[i] = tf_res[effEnd-1]; + } else if (hybrid && weak_transient) + { + /* For weak transients, we rely on the fact that improving time resolution using + TF on a long window is imperfect and will not result in an energy collapse at + low bitrate. */ + for (i=0;i<end;i++) + tf_res[i] = 1; + tf_select=0; + } else if (hybrid && effectiveBytes<15 && st->silk_info.signalType != 2) + { + /* For low bitrate hybrid, we force temporal resolution to 5 ms rather than 2.5 ms. */ + for (i=0;i<end;i++) + tf_res[i] = 0; + tf_select=isTransient; + } else { + for (i=0;i<end;i++) + tf_res[i] = isTransient; + tf_select=0; + } + + ALLOC(error, C*nbEBands, opus_val16); + c=0; + do { + for (i=start;i<end;i++) + { + /* When the energy is stable, slightly bias energy quantization towards + the previous error to make the gain more stable (a constant offset is + better than fluctuations). */ + if (ABS32(SUB32(bandLogE[i+c*nbEBands], oldBandE[i+c*nbEBands])) < QCONST16(2.f, DB_SHIFT)) + { + bandLogE[i+c*nbEBands] -= MULT16_16_Q15(energyError[i+c*nbEBands], QCONST16(0.25f, 15)); + } + } + } while (++c < C); + quant_coarse_energy(mode, start, end, effEnd, bandLogE, + oldBandE, total_bits, error, enc, + C, LM, nbAvailableBytes, st->force_intra, + &st->delayedIntra, st->complexity >= 4, st->loss_rate, st->lfe); + + tf_encode(start, end, isTransient, tf_res, LM, tf_select, enc); + + if (ec_tell(enc)+4<=total_bits) + { + if (st->lfe) + { + st->tapset_decision = 0; + st->spread_decision = SPREAD_NORMAL; + } else if (hybrid) + { + if (st->complexity == 0) + st->spread_decision = SPREAD_NONE; + else if (isTransient) + st->spread_decision = SPREAD_NORMAL; + else + st->spread_decision = SPREAD_AGGRESSIVE; + } else if (shortBlocks || st->complexity < 3 || nbAvailableBytes < 10*C) + { + if (st->complexity == 0) + st->spread_decision = SPREAD_NONE; + else + st->spread_decision = SPREAD_NORMAL; + } else { + /* Disable new spreading+tapset estimator until we can show it works + better than the old one. So far it seems like spreading_decision() + works best. */ +#if 0 + if (st->analysis.valid) + { + static const opus_val16 spread_thresholds[3] = {-QCONST16(.6f, 15), -QCONST16(.2f, 15), -QCONST16(.07f, 15)}; + static const opus_val16 spread_histeresis[3] = {QCONST16(.15f, 15), QCONST16(.07f, 15), QCONST16(.02f, 15)}; + static const opus_val16 tapset_thresholds[2] = {QCONST16(.0f, 15), QCONST16(.15f, 15)}; + static const opus_val16 tapset_histeresis[2] = {QCONST16(.1f, 15), QCONST16(.05f, 15)}; + st->spread_decision = hysteresis_decision(-st->analysis.tonality, spread_thresholds, spread_histeresis, 3, st->spread_decision); + st->tapset_decision = hysteresis_decision(st->analysis.tonality_slope, tapset_thresholds, tapset_histeresis, 2, st->tapset_decision); + } else +#endif + { + st->spread_decision = spreading_decision(mode, X, + &st->tonal_average, st->spread_decision, &st->hf_average, + &st->tapset_decision, pf_on&&!shortBlocks, effEnd, C, M, spread_weight); + } + /*printf("%d %d\n", st->tapset_decision, st->spread_decision);*/ + /*printf("%f %d %f %d\n\n", st->analysis.tonality, st->spread_decision, st->analysis.tonality_slope, st->tapset_decision);*/ + } + ec_enc_icdf(enc, st->spread_decision, spread_icdf, 5); + } + + /* For LFE, everything interesting is in the first band */ + if (st->lfe) + offsets[0] = IMIN(8, effectiveBytes/3); + ALLOC(cap, nbEBands, int); + init_caps(mode,cap,LM,C); + + dynalloc_logp = 6; + total_bits<<=BITRES; + total_boost = 0; + tell = ec_tell_frac(enc); + for (i=start;i<end;i++) + { + int width, quanta; + int dynalloc_loop_logp; + int boost; + int j; + width = C*(eBands[i+1]-eBands[i])<<LM; + /* quanta is 6 bits, but no more than 1 bit/sample + and no less than 1/8 bit/sample */ + quanta = IMIN(width<<BITRES, IMAX(6<<BITRES, width)); + dynalloc_loop_logp = dynalloc_logp; + boost = 0; + for (j = 0; tell+(dynalloc_loop_logp<<BITRES) < total_bits-total_boost + && boost < cap[i]; j++) + { + int flag; + flag = j<offsets[i]; + ec_enc_bit_logp(enc, flag, dynalloc_loop_logp); + tell = ec_tell_frac(enc); + if (!flag) + break; + boost += quanta; + total_boost += quanta; + dynalloc_loop_logp = 1; + } + /* Making dynalloc more likely */ + if (j) + dynalloc_logp = IMAX(2, dynalloc_logp-1); + offsets[i] = boost; + } + + if (C==2) + { + static const opus_val16 intensity_thresholds[21]= + /* 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 off*/ + { 1, 2, 3, 4, 5, 6, 7, 8,16,24,36,44,50,56,62,67,72,79,88,106,134}; + static const opus_val16 intensity_histeresis[21]= + { 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 3, 3, 4, 5, 6, 8, 8}; + + /* Always use MS for 2.5 ms frames until we can do a better analysis */ + if (LM!=0) + dual_stereo = stereo_analysis(mode, X, LM, N); + + st->intensity = hysteresis_decision((opus_val16)(equiv_rate/1000), + intensity_thresholds, intensity_histeresis, 21, st->intensity); + st->intensity = IMIN(end,IMAX(start, st->intensity)); + } + + alloc_trim = 5; + if (tell+(6<<BITRES) <= total_bits - total_boost) + { + if (start > 0 || st->lfe) + { + st->stereo_saving = 0; + alloc_trim = 5; + } else { + alloc_trim = alloc_trim_analysis(mode, X, bandLogE, + end, LM, C, N, &st->analysis, &st->stereo_saving, tf_estimate, + st->intensity, surround_trim, equiv_rate, st->arch); + } + ec_enc_icdf(enc, alloc_trim, trim_icdf, 7); + tell = ec_tell_frac(enc); + } + + /* Variable bitrate */ + if (vbr_rate>0) + { + opus_val16 alpha; + opus_int32 delta; + /* The target rate in 8th bits per frame */ + opus_int32 target, base_target; + opus_int32 min_allowed; + int lm_diff = mode->maxLM - LM; + + /* Don't attempt to use more than 510 kb/s, even for frames smaller than 20 ms. + The CELT allocator will just not be able to use more than that anyway. */ + nbCompressedBytes = IMIN(nbCompressedBytes,1275>>(3-LM)); + if (!hybrid) + { + base_target = vbr_rate - ((40*C+20)<<BITRES); + } else { + base_target = IMAX(0, vbr_rate - ((9*C+4)<<BITRES)); + } + + if (st->constrained_vbr) + base_target += (st->vbr_offset>>lm_diff); + + if (!hybrid) + { + target = compute_vbr(mode, &st->analysis, base_target, LM, equiv_rate, + st->lastCodedBands, C, st->intensity, st->constrained_vbr, + st->stereo_saving, tot_boost, tf_estimate, pitch_change, maxDepth, + st->lfe, st->energy_mask!=NULL, surround_masking, + temporal_vbr); + } else { + target = base_target; + /* Tonal frames (offset<100) need more bits than noisy (offset>100) ones. */ + if (st->silk_info.offset < 100) target += 12 << BITRES >> (3-LM); + if (st->silk_info.offset > 100) target -= 18 << BITRES >> (3-LM); + /* Boosting bitrate on transients and vowels with significant temporal + spikes. */ + target += (opus_int32)MULT16_16_Q14(tf_estimate-QCONST16(.25f,14), (50<<BITRES)); + /* If we have a strong transient, let's make sure it has enough bits to code + the first two bands, so that it can use folding rather than noise. */ + if (tf_estimate > QCONST16(.7f,14)) + target = IMAX(target, 50<<BITRES); + } + /* The current offset is removed from the target and the space used + so far is added*/ + target=target+tell; + /* In VBR mode the frame size must not be reduced so much that it would + result in the encoder running out of bits. + The margin of 2 bytes ensures that none of the bust-prevention logic + in the decoder will have triggered so far. */ + min_allowed = ((tell+total_boost+(1<<(BITRES+3))-1)>>(BITRES+3)) + 2; + /* Take into account the 37 bits we need to have left in the packet to + signal a redundant frame in hybrid mode. Creating a shorter packet would + create an entropy coder desync. */ + if (hybrid) + min_allowed = IMAX(min_allowed, (tell0_frac+(37<<BITRES)+total_boost+(1<<(BITRES+3))-1)>>(BITRES+3)); + + nbAvailableBytes = (target+(1<<(BITRES+2)))>>(BITRES+3); + nbAvailableBytes = IMAX(min_allowed,nbAvailableBytes); + nbAvailableBytes = IMIN(nbCompressedBytes,nbAvailableBytes); + + /* By how much did we "miss" the target on that frame */ + delta = target - vbr_rate; + + target=nbAvailableBytes<<(BITRES+3); + + /*If the frame is silent we don't adjust our drift, otherwise + the encoder will shoot to very high rates after hitting a + span of silence, but we do allow the bitres to refill. + This means that we'll undershoot our target in CVBR/VBR modes + on files with lots of silence. */ + if(silence) + { + nbAvailableBytes = 2; + target = 2*8<<BITRES; + delta = 0; + } + + if (st->vbr_count < 970) + { + st->vbr_count++; + alpha = celt_rcp(SHL32(EXTEND32(st->vbr_count+20),16)); + } else + alpha = QCONST16(.001f,15); + /* How many bits have we used in excess of what we're allowed */ + if (st->constrained_vbr) + st->vbr_reservoir += target - vbr_rate; + /*printf ("%d\n", st->vbr_reservoir);*/ + + /* Compute the offset we need to apply in order to reach the target */ + if (st->constrained_vbr) + { + st->vbr_drift += (opus_int32)MULT16_32_Q15(alpha,(delta*(1<<lm_diff))-st->vbr_offset-st->vbr_drift); + st->vbr_offset = -st->vbr_drift; + } + /*printf ("%d\n", st->vbr_drift);*/ + + if (st->constrained_vbr && st->vbr_reservoir < 0) + { + /* We're under the min value -- increase rate */ + int adjust = (-st->vbr_reservoir)/(8<<BITRES); + /* Unless we're just coding silence */ + nbAvailableBytes += silence?0:adjust; + st->vbr_reservoir = 0; + /*printf ("+%d\n", adjust);*/ + } + nbCompressedBytes = IMIN(nbCompressedBytes,nbAvailableBytes); + /*printf("%d\n", nbCompressedBytes*50*8);*/ + /* This moves the raw bits to take into account the new compressed size */ + ec_enc_shrink(enc, nbCompressedBytes); + } + + /* Bit allocation */ + ALLOC(fine_quant, nbEBands, int); + ALLOC(pulses, nbEBands, int); + ALLOC(fine_priority, nbEBands, int); + + /* bits = packet size - where we are - safety*/ + bits = (((opus_int32)nbCompressedBytes*8)<<BITRES) - ec_tell_frac(enc) - 1; + anti_collapse_rsv = isTransient&&LM>=2&&bits>=((LM+2)<<BITRES) ? (1<<BITRES) : 0; + bits -= anti_collapse_rsv; + signalBandwidth = end-1; +#ifndef DISABLE_FLOAT_API + if (st->analysis.valid) + { + int min_bandwidth; + if (equiv_rate < (opus_int32)32000*C) + min_bandwidth = 13; + else if (equiv_rate < (opus_int32)48000*C) + min_bandwidth = 16; + else if (equiv_rate < (opus_int32)60000*C) + min_bandwidth = 18; + else if (equiv_rate < (opus_int32)80000*C) + min_bandwidth = 19; + else + min_bandwidth = 20; + signalBandwidth = IMAX(st->analysis.bandwidth, min_bandwidth); + } +#endif + if (st->lfe) + signalBandwidth = 1; + codedBands = clt_compute_allocation(mode, start, end, offsets, cap, + alloc_trim, &st->intensity, &dual_stereo, bits, &balance, pulses, + fine_quant, fine_priority, C, LM, enc, 1, st->lastCodedBands, signalBandwidth); + if (st->lastCodedBands) + st->lastCodedBands = IMIN(st->lastCodedBands+1,IMAX(st->lastCodedBands-1,codedBands)); + else + st->lastCodedBands = codedBands; + + quant_fine_energy(mode, start, end, oldBandE, error, fine_quant, enc, C); + + /* Residual quantisation */ + ALLOC(collapse_masks, C*nbEBands, unsigned char); + quant_all_bands(1, mode, start, end, X, C==2 ? X+N : NULL, collapse_masks, + bandE, pulses, shortBlocks, st->spread_decision, + dual_stereo, st->intensity, tf_res, nbCompressedBytes*(8<<BITRES)-anti_collapse_rsv, + balance, enc, LM, codedBands, &st->rng, st->complexity, st->arch, st->disable_inv); + + if (anti_collapse_rsv > 0) + { + anti_collapse_on = st->consec_transient<2; +#ifdef FUZZING + anti_collapse_on = rand()&0x1; +#endif + ec_enc_bits(enc, anti_collapse_on, 1); + } + quant_energy_finalise(mode, start, end, oldBandE, error, fine_quant, fine_priority, nbCompressedBytes*8-ec_tell(enc), enc, C); + OPUS_CLEAR(energyError, nbEBands*CC); + c=0; + do { + for (i=start;i<end;i++) + { + energyError[i+c*nbEBands] = MAX16(-QCONST16(0.5f, 15), MIN16(QCONST16(0.5f, 15), error[i+c*nbEBands])); + } + } while (++c < C); + + if (silence) + { + for (i=0;i<C*nbEBands;i++) + oldBandE[i] = -QCONST16(28.f,DB_SHIFT); + } + +#ifdef RESYNTH + /* Re-synthesis of the coded audio if required */ + { + celt_sig *out_mem[2]; + + if (anti_collapse_on) + { + anti_collapse(mode, X, collapse_masks, LM, C, N, + start, end, oldBandE, oldLogE, oldLogE2, pulses, st->rng); + } + + c=0; do { + OPUS_MOVE(st->syn_mem[c], st->syn_mem[c]+N, 2*MAX_PERIOD-N+overlap/2); + } while (++c<CC); + + c=0; do { + out_mem[c] = st->syn_mem[c]+2*MAX_PERIOD-N; + } while (++c<CC); + + celt_synthesis(mode, X, out_mem, oldBandE, start, effEnd, + C, CC, isTransient, LM, st->upsample, silence, st->arch); + + c=0; do { + st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); + st->prefilter_period_old=IMAX(st->prefilter_period_old, COMBFILTER_MINPERIOD); + comb_filter(out_mem[c], out_mem[c], st->prefilter_period_old, st->prefilter_period, mode->shortMdctSize, + st->prefilter_gain_old, st->prefilter_gain, st->prefilter_tapset_old, st->prefilter_tapset, + mode->window, overlap); + if (LM!=0) + comb_filter(out_mem[c]+mode->shortMdctSize, out_mem[c]+mode->shortMdctSize, st->prefilter_period, pitch_index, N-mode->shortMdctSize, + st->prefilter_gain, gain1, st->prefilter_tapset, prefilter_tapset, + mode->window, overlap); + } while (++c<CC); + + /* We reuse freq[] as scratch space for the de-emphasis */ + deemphasis(out_mem, (opus_val16*)pcm, N, CC, st->upsample, mode->preemph, st->preemph_memD); + st->prefilter_period_old = st->prefilter_period; + st->prefilter_gain_old = st->prefilter_gain; + st->prefilter_tapset_old = st->prefilter_tapset; + } +#endif + + st->prefilter_period = pitch_index; + st->prefilter_gain = gain1; + st->prefilter_tapset = prefilter_tapset; +#ifdef RESYNTH + if (LM!=0) + { + st->prefilter_period_old = st->prefilter_period; + st->prefilter_gain_old = st->prefilter_gain; + st->prefilter_tapset_old = st->prefilter_tapset; + } +#endif + + if (CC==2&&C==1) { + OPUS_COPY(&oldBandE[nbEBands], oldBandE, nbEBands); + } + + if (!isTransient) + { + OPUS_COPY(oldLogE2, oldLogE, CC*nbEBands); + OPUS_COPY(oldLogE, oldBandE, CC*nbEBands); + } else { + for (i=0;i<CC*nbEBands;i++) + oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]); + } + /* In case start or end were to change */ + c=0; do + { + for (i=0;i<start;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + for (i=end;i<nbEBands;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + } while (++c<CC); + + if (isTransient || transient_got_disabled) + st->consec_transient++; + else + st->consec_transient=0; + st->rng = enc->rng; + + /* If there's any room left (can only happen for very high rates), + it's already filled with zeros */ + ec_enc_done(enc); + +#ifdef CUSTOM_MODES + if (st->signalling) + nbCompressedBytes++; +#endif + + RESTORE_STACK; + if (ec_get_error(enc)) + return OPUS_INTERNAL_ERROR; + else + return nbCompressedBytes; +} + + +#ifdef CUSTOM_MODES + +#ifdef FIXED_POINT +int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + return celt_encode_with_ec(st, pcm, frame_size, compressed, nbCompressedBytes, NULL); +} + +#ifndef DISABLE_FLOAT_API +int opus_custom_encode_float(CELTEncoder * OPUS_RESTRICT st, const float * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + int j, ret, C, N; + VARDECL(opus_int16, in); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + ALLOC(in, C*N, opus_int16); + + for (j=0;j<C*N;j++) + in[j] = FLOAT2INT16(pcm[j]); + + ret=celt_encode_with_ec(st,in,frame_size,compressed,nbCompressedBytes, NULL); +#ifdef RESYNTH + for (j=0;j<C*N;j++) + ((float*)pcm)[j]=in[j]*(1.f/32768.f); +#endif + RESTORE_STACK; + return ret; +} +#endif /* DISABLE_FLOAT_API */ +#else + +int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + int j, ret, C, N; + VARDECL(celt_sig, in); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C=st->channels; + N=frame_size; + ALLOC(in, C*N, celt_sig); + for (j=0;j<C*N;j++) { + in[j] = SCALEOUT(pcm[j]); + } + + ret = celt_encode_with_ec(st,in,frame_size,compressed,nbCompressedBytes, NULL); +#ifdef RESYNTH + for (j=0;j<C*N;j++) + ((opus_int16*)pcm)[j] = FLOAT2INT16(in[j]); +#endif + RESTORE_STACK; + return ret; +} + +int opus_custom_encode_float(CELTEncoder * OPUS_RESTRICT st, const float * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + return celt_encode_with_ec(st, pcm, frame_size, compressed, nbCompressedBytes, NULL); +} + +#endif + +#endif /* CUSTOM_MODES */ + +int opus_custom_encoder_ctl(CELTEncoder * OPUS_RESTRICT st, int request, ...) +{ + va_list ap; + + va_start(ap, request); + switch (request) + { + case OPUS_SET_COMPLEXITY_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>10) + goto bad_arg; + st->complexity = value; + } + break; + case CELT_SET_START_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<0 || value>=st->mode->nbEBands) + goto bad_arg; + st->start = value; + } + break; + case CELT_SET_END_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>st->mode->nbEBands) + goto bad_arg; + st->end = value; + } + break; + case CELT_SET_PREDICTION_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>2) + goto bad_arg; + st->disable_pf = value<=1; + st->force_intra = value==0; + } + break; + case OPUS_SET_PACKET_LOSS_PERC_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>100) + goto bad_arg; + st->loss_rate = value; + } + break; + case OPUS_SET_VBR_CONSTRAINT_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->constrained_vbr = value; + } + break; + case OPUS_SET_VBR_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->vbr = value; + } + break; + case OPUS_SET_BITRATE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<=500 && value!=OPUS_BITRATE_MAX) + goto bad_arg; + value = IMIN(value, 260000*st->channels); + st->bitrate = value; + } + break; + case CELT_SET_CHANNELS_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>2) + goto bad_arg; + st->stream_channels = value; + } + break; + case OPUS_SET_LSB_DEPTH_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<8 || value>24) + goto bad_arg; + st->lsb_depth=value; + } + break; + case OPUS_GET_LSB_DEPTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + *value=st->lsb_depth; + } + break; + case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->disable_inv = value; + } + break; + case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->disable_inv; + } + break; + case OPUS_RESET_STATE: + { + int i; + opus_val16 *oldBandE, *oldLogE, *oldLogE2; + oldBandE = (opus_val16*)(st->in_mem+st->channels*(st->mode->overlap+COMBFILTER_MAXPERIOD)); + oldLogE = oldBandE + st->channels*st->mode->nbEBands; + oldLogE2 = oldLogE + st->channels*st->mode->nbEBands; + OPUS_CLEAR((char*)&st->ENCODER_RESET_START, + opus_custom_encoder_get_size(st->mode, st->channels)- + ((char*)&st->ENCODER_RESET_START - (char*)st)); + for (i=0;i<st->channels*st->mode->nbEBands;i++) + oldLogE[i]=oldLogE2[i]=-QCONST16(28.f,DB_SHIFT); + st->vbr_offset = 0; + st->delayedIntra = 1; + st->spread_decision = SPREAD_NORMAL; + st->tonal_average = 256; + st->hf_average = 0; + st->tapset_decision = 0; + } + break; +#ifdef CUSTOM_MODES + case CELT_SET_INPUT_CLIPPING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->clip = value; + } + break; +#endif + case CELT_SET_SIGNALLING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->signalling = value; + } + break; + case CELT_SET_ANALYSIS_REQUEST: + { + AnalysisInfo *info = va_arg(ap, AnalysisInfo *); + if (info) + OPUS_COPY(&st->analysis, info, 1); + } + break; + case CELT_SET_SILK_INFO_REQUEST: + { + SILKInfo *info = va_arg(ap, SILKInfo *); + if (info) + OPUS_COPY(&st->silk_info, info, 1); + } + break; + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (value==0) + goto bad_arg; + *value=st->mode; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 * value = va_arg(ap, opus_uint32 *); + if (value==0) + goto bad_arg; + *value=st->rng; + } + break; + case OPUS_SET_LFE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->lfe = value; + } + break; + case OPUS_SET_ENERGY_MASK_REQUEST: + { + opus_val16 *value = va_arg(ap, opus_val16*); + st->energy_mask = value; + } + break; + default: + goto bad_request; + } + va_end(ap); + return OPUS_OK; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +bad_request: + va_end(ap); + return OPUS_UNIMPLEMENTED; +} diff --git a/libopus/celt/celt_lpc.c b/libopus/celt/celt_lpc.c new file mode 100644 index 0000000..8ecb693 --- /dev/null +++ b/libopus/celt/celt_lpc.c @@ -0,0 +1,296 @@ +/* Copyright (c) 2009-2010 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" + +void _celt_lpc( + opus_val16 *_lpc, /* out: [0...p-1] LPC coefficients */ +const opus_val32 *ac, /* in: [0...p] autocorrelation values */ +int p +) +{ + int i, j; + opus_val32 r; + opus_val32 error = ac[0]; +#ifdef FIXED_POINT + opus_val32 lpc[LPC_ORDER]; +#else + float *lpc = _lpc; +#endif + + OPUS_CLEAR(lpc, p); + if (ac[0] != 0) + { + for (i = 0; i < p; i++) { + /* Sum up this iteration's reflection coefficient */ + opus_val32 rr = 0; + for (j = 0; j < i; j++) + rr += MULT32_32_Q31(lpc[j],ac[i - j]); + rr += SHR32(ac[i + 1],3); + r = -frac_div32(SHL32(rr,3), error); + /* Update LPC coefficients and total error */ + lpc[i] = SHR32(r,3); + for (j = 0; j < (i+1)>>1; j++) + { + opus_val32 tmp1, tmp2; + tmp1 = lpc[j]; + tmp2 = lpc[i-1-j]; + lpc[j] = tmp1 + MULT32_32_Q31(r,tmp2); + lpc[i-1-j] = tmp2 + MULT32_32_Q31(r,tmp1); + } + + error = error - MULT32_32_Q31(MULT32_32_Q31(r,r),error); + /* Bail out once we get 30 dB gain */ +#ifdef FIXED_POINT + if (error<SHR32(ac[0],10)) + break; +#else + if (error<.001f*ac[0]) + break; +#endif + } + } +#ifdef FIXED_POINT + for (i=0;i<p;i++) + _lpc[i] = ROUND16(lpc[i],16); +#endif +} + + +void celt_fir_c( + const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch) +{ + int i,j; + VARDECL(opus_val16, rnum); + SAVE_STACK; + celt_assert(x != y); + ALLOC(rnum, ord, opus_val16); + for(i=0;i<ord;i++) + rnum[i] = num[ord-i-1]; + for (i=0;i<N-3;i+=4) + { + opus_val32 sum[4]; + sum[0] = SHL32(EXTEND32(x[i ]), SIG_SHIFT); + sum[1] = SHL32(EXTEND32(x[i+1]), SIG_SHIFT); + sum[2] = SHL32(EXTEND32(x[i+2]), SIG_SHIFT); + sum[3] = SHL32(EXTEND32(x[i+3]), SIG_SHIFT); + xcorr_kernel(rnum, x+i-ord, sum, ord, arch); + y[i ] = ROUND16(sum[0], SIG_SHIFT); + y[i+1] = ROUND16(sum[1], SIG_SHIFT); + y[i+2] = ROUND16(sum[2], SIG_SHIFT); + y[i+3] = ROUND16(sum[3], SIG_SHIFT); + } + for (;i<N;i++) + { + opus_val32 sum = SHL32(EXTEND32(x[i]), SIG_SHIFT); + for (j=0;j<ord;j++) + sum = MAC16_16(sum,rnum[j],x[i+j-ord]); + y[i] = ROUND16(sum, SIG_SHIFT); + } + RESTORE_STACK; +} + +void celt_iir(const opus_val32 *_x, + const opus_val16 *den, + opus_val32 *_y, + int N, + int ord, + opus_val16 *mem, + int arch) +{ +#ifdef SMALL_FOOTPRINT + int i,j; + (void)arch; + for (i=0;i<N;i++) + { + opus_val32 sum = _x[i]; + for (j=0;j<ord;j++) + { + sum -= MULT16_16(den[j],mem[j]); + } + for (j=ord-1;j>=1;j--) + { + mem[j]=mem[j-1]; + } + mem[0] = SROUND16(sum, SIG_SHIFT); + _y[i] = sum; + } +#else + int i,j; + VARDECL(opus_val16, rden); + VARDECL(opus_val16, y); + SAVE_STACK; + + celt_assert((ord&3)==0); + ALLOC(rden, ord, opus_val16); + ALLOC(y, N+ord, opus_val16); + for(i=0;i<ord;i++) + rden[i] = den[ord-i-1]; + for(i=0;i<ord;i++) + y[i] = -mem[ord-i-1]; + for(;i<N+ord;i++) + y[i]=0; + for (i=0;i<N-3;i+=4) + { + /* Unroll by 4 as if it were an FIR filter */ + opus_val32 sum[4]; + sum[0]=_x[i]; + sum[1]=_x[i+1]; + sum[2]=_x[i+2]; + sum[3]=_x[i+3]; + xcorr_kernel(rden, y+i, sum, ord, arch); + + /* Patch up the result to compensate for the fact that this is an IIR */ + y[i+ord ] = -SROUND16(sum[0],SIG_SHIFT); + _y[i ] = sum[0]; + sum[1] = MAC16_16(sum[1], y[i+ord ], den[0]); + y[i+ord+1] = -SROUND16(sum[1],SIG_SHIFT); + _y[i+1] = sum[1]; + sum[2] = MAC16_16(sum[2], y[i+ord+1], den[0]); + sum[2] = MAC16_16(sum[2], y[i+ord ], den[1]); + y[i+ord+2] = -SROUND16(sum[2],SIG_SHIFT); + _y[i+2] = sum[2]; + + sum[3] = MAC16_16(sum[3], y[i+ord+2], den[0]); + sum[3] = MAC16_16(sum[3], y[i+ord+1], den[1]); + sum[3] = MAC16_16(sum[3], y[i+ord ], den[2]); + y[i+ord+3] = -SROUND16(sum[3],SIG_SHIFT); + _y[i+3] = sum[3]; + } + for (;i<N;i++) + { + opus_val32 sum = _x[i]; + for (j=0;j<ord;j++) + sum -= MULT16_16(rden[j],y[i+j]); + y[i+ord] = SROUND16(sum,SIG_SHIFT); + _y[i] = sum; + } + for(i=0;i<ord;i++) + mem[i] = _y[N-i-1]; + RESTORE_STACK; +#endif +} + +int _celt_autocorr( + const opus_val16 *x, /* in: [0...n-1] samples x */ + opus_val32 *ac, /* out: [0...lag-1] ac values */ + const opus_val16 *window, + int overlap, + int lag, + int n, + int arch + ) +{ + opus_val32 d; + int i, k; + int fastN=n-lag; + int shift; + const opus_val16 *xptr; + VARDECL(opus_val16, xx); + SAVE_STACK; + ALLOC(xx, n, opus_val16); + celt_assert(n>0); + celt_assert(overlap>=0); + if (overlap == 0) + { + xptr = x; + } else { + for (i=0;i<n;i++) + xx[i] = x[i]; + for (i=0;i<overlap;i++) + { + xx[i] = MULT16_16_Q15(x[i],window[i]); + xx[n-i-1] = MULT16_16_Q15(x[n-i-1],window[i]); + } + xptr = xx; + } + shift=0; +#ifdef FIXED_POINT + { + opus_val32 ac0; + ac0 = 1+(n<<7); + if (n&1) ac0 += SHR32(MULT16_16(xptr[0],xptr[0]),9); + for(i=(n&1);i<n;i+=2) + { + ac0 += SHR32(MULT16_16(xptr[i],xptr[i]),9); + ac0 += SHR32(MULT16_16(xptr[i+1],xptr[i+1]),9); + } + + shift = celt_ilog2(ac0)-30+10; + shift = (shift)/2; + if (shift>0) + { + for(i=0;i<n;i++) + xx[i] = PSHR32(xptr[i], shift); + xptr = xx; + } else + shift = 0; + } +#endif + celt_pitch_xcorr(xptr, xptr, ac, fastN, lag+1, arch); + for (k=0;k<=lag;k++) + { + for (i = k+fastN, d = 0; i < n; i++) + d = MAC16_16(d, xptr[i], xptr[i-k]); + ac[k] += d; + } +#ifdef FIXED_POINT + shift = 2*shift; + if (shift<=0) + ac[0] += SHL32((opus_int32)1, -shift); + if (ac[0] < 268435456) + { + int shift2 = 29 - EC_ILOG(ac[0]); + for (i=0;i<=lag;i++) + ac[i] = SHL32(ac[i], shift2); + shift -= shift2; + } else if (ac[0] >= 536870912) + { + int shift2=1; + if (ac[0] >= 1073741824) + shift2++; + for (i=0;i<=lag;i++) + ac[i] = SHR32(ac[i], shift2); + shift += shift2; + } +#endif + + RESTORE_STACK; + return shift; +} diff --git a/libopus/celt/celt_lpc.h b/libopus/celt/celt_lpc.h new file mode 100644 index 0000000..a4c5fd6 --- /dev/null +++ b/libopus/celt/celt_lpc.h @@ -0,0 +1,66 @@ +/* Copyright (c) 2009-2010 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PLC_H +#define PLC_H + +#include "arch.h" +#include "cpu_support.h" + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) +#include "x86/celt_lpc_sse.h" +#endif + +#define LPC_ORDER 24 + +void _celt_lpc(opus_val16 *_lpc, const opus_val32 *ac, int p); + +void celt_fir_c( + const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch); + +#if !defined(OVERRIDE_CELT_FIR) +#define celt_fir(x, num, y, N, ord, arch) \ + (celt_fir_c(x, num, y, N, ord, arch)) +#endif + +void celt_iir(const opus_val32 *x, + const opus_val16 *den, + opus_val32 *y, + int N, + int ord, + opus_val16 *mem, + int arch); + +int _celt_autocorr(const opus_val16 *x, opus_val32 *ac, + const opus_val16 *window, int overlap, int lag, int n, int arch); + +#endif /* PLC_H */ diff --git a/libopus/celt/cpu_support.h b/libopus/celt/cpu_support.h new file mode 100644 index 0000000..68fc606 --- /dev/null +++ b/libopus/celt/cpu_support.h @@ -0,0 +1,70 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CPU_SUPPORT_H +#define CPU_SUPPORT_H + +#include "opus_types.h" +#include "opus_defines.h" + +#if defined(OPUS_HAVE_RTCD) && \ + (defined(OPUS_ARM_ASM) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)) +#include "arm/armcpu.h" + +/* We currently support 4 ARM variants: + * arch[0] -> ARMv4 + * arch[1] -> ARMv5E + * arch[2] -> ARMv6 + * arch[3] -> NEON + */ +#define OPUS_ARCHMASK 3 + +#elif (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \ + (defined(OPUS_X86_MAY_HAVE_AVX) && !defined(OPUS_X86_PRESUME_AVX)) + +#include "x86/x86cpu.h" +/* We currently support 5 x86 variants: + * arch[0] -> non-sse + * arch[1] -> sse + * arch[2] -> sse2 + * arch[3] -> sse4.1 + * arch[4] -> avx + */ +#define OPUS_ARCHMASK 7 +int opus_select_arch(void); + +#else +#define OPUS_ARCHMASK 0 + +static OPUS_INLINE int opus_select_arch(void) +{ + return 0; +} +#endif +#endif diff --git a/libopus/celt/cwrs.c b/libopus/celt/cwrs.c new file mode 100644 index 0000000..a552e4f --- /dev/null +++ b/libopus/celt/cwrs.c @@ -0,0 +1,715 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2007-2009 Timothy B. Terriberry + Written by Timothy B. Terriberry and Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "os_support.h" +#include "cwrs.h" +#include "mathops.h" +#include "arch.h" + +#ifdef CUSTOM_MODES + +/*Guaranteed to return a conservatively large estimate of the binary logarithm + with frac bits of fractional precision. + Tested for all possible 32-bit inputs with frac=4, where the maximum + overestimation is 0.06254243 bits.*/ +int log2_frac(opus_uint32 val, int frac) +{ + int l; + l=EC_ILOG(val); + if(val&(val-1)){ + /*This is (val>>l-16), but guaranteed to round up, even if adding a bias + before the shift would cause overflow (e.g., for 0xFFFFxxxx). + Doesn't work for val=0, but that case fails the test above.*/ + if(l>16)val=((val-1)>>(l-16))+1; + else val<<=16-l; + l=(l-1)<<frac; + /*Note that we always need one iteration, since the rounding up above means + that we might need to adjust the integer part of the logarithm.*/ + do{ + int b; + b=(int)(val>>16); + l+=b<<frac; + val=(val+b)>>b; + val=(val*val+0x7FFF)>>15; + } + while(frac-->0); + /*If val is not exactly 0x8000, then we have to round up the remainder.*/ + return l+(val>0x8000); + } + /*Exact powers of two require no rounding.*/ + else return (l-1)<<frac; +} +#endif + +/*Although derived separately, the pulse vector coding scheme is equivalent to + a Pyramid Vector Quantizer \cite{Fis86}. + Some additional notes about an early version appear at + https://people.xiph.org/~tterribe/notes/cwrs.html, but the codebook ordering + and the definitions of some terms have evolved since that was written. + + The conversion from a pulse vector to an integer index (encoding) and back + (decoding) is governed by two related functions, V(N,K) and U(N,K). + + V(N,K) = the number of combinations, with replacement, of N items, taken K + at a time, when a sign bit is added to each item taken at least once (i.e., + the number of N-dimensional unit pulse vectors with K pulses). + One way to compute this is via + V(N,K) = K>0 ? sum(k=1...K,2**k*choose(N,k)*choose(K-1,k-1)) : 1, + where choose() is the binomial function. + A table of values for N<10 and K<10 looks like: + V[10][10] = { + {1, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + {1, 2, 2, 2, 2, 2, 2, 2, 2, 2}, + {1, 4, 8, 12, 16, 20, 24, 28, 32, 36}, + {1, 6, 18, 38, 66, 102, 146, 198, 258, 326}, + {1, 8, 32, 88, 192, 360, 608, 952, 1408, 1992}, + {1, 10, 50, 170, 450, 1002, 1970, 3530, 5890, 9290}, + {1, 12, 72, 292, 912, 2364, 5336, 10836, 20256, 35436}, + {1, 14, 98, 462, 1666, 4942, 12642, 28814, 59906, 115598}, + {1, 16, 128, 688, 2816, 9424, 27008, 68464, 157184, 332688}, + {1, 18, 162, 978, 4482, 16722, 53154, 148626, 374274, 864146} + }; + + U(N,K) = the number of such combinations wherein N-1 objects are taken at + most K-1 at a time. + This is given by + U(N,K) = sum(k=0...K-1,V(N-1,k)) + = K>0 ? (V(N-1,K-1) + V(N,K-1))/2 : 0. + The latter expression also makes clear that U(N,K) is half the number of such + combinations wherein the first object is taken at least once. + Although it may not be clear from either of these definitions, U(N,K) is the + natural function to work with when enumerating the pulse vector codebooks, + not V(N,K). + U(N,K) is not well-defined for N=0, but with the extension + U(0,K) = K>0 ? 0 : 1, + the function becomes symmetric: U(N,K) = U(K,N), with a similar table: + U[10][10] = { + {1, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + {0, 1, 1, 1, 1, 1, 1, 1, 1, 1}, + {0, 1, 3, 5, 7, 9, 11, 13, 15, 17}, + {0, 1, 5, 13, 25, 41, 61, 85, 113, 145}, + {0, 1, 7, 25, 63, 129, 231, 377, 575, 833}, + {0, 1, 9, 41, 129, 321, 681, 1289, 2241, 3649}, + {0, 1, 11, 61, 231, 681, 1683, 3653, 7183, 13073}, + {0, 1, 13, 85, 377, 1289, 3653, 8989, 19825, 40081}, + {0, 1, 15, 113, 575, 2241, 7183, 19825, 48639, 108545}, + {0, 1, 17, 145, 833, 3649, 13073, 40081, 108545, 265729} + }; + + With this extension, V(N,K) may be written in terms of U(N,K): + V(N,K) = U(N,K) + U(N,K+1) + for all N>=0, K>=0. + Thus U(N,K+1) represents the number of combinations where the first element + is positive or zero, and U(N,K) represents the number of combinations where + it is negative. + With a large enough table of U(N,K) values, we could write O(N) encoding + and O(min(N*log(K),N+K)) decoding routines, but such a table would be + prohibitively large for small embedded devices (K may be as large as 32767 + for small N, and N may be as large as 200). + + Both functions obey the same recurrence relation: + V(N,K) = V(N-1,K) + V(N,K-1) + V(N-1,K-1), + U(N,K) = U(N-1,K) + U(N,K-1) + U(N-1,K-1), + for all N>0, K>0, with different initial conditions at N=0 or K=0. + This allows us to construct a row of one of the tables above given the + previous row or the next row. + Thus we can derive O(NK) encoding and decoding routines with O(K) memory + using only addition and subtraction. + + When encoding, we build up from the U(2,K) row and work our way forwards. + When decoding, we need to start at the U(N,K) row and work our way backwards, + which requires a means of computing U(N,K). + U(N,K) may be computed from two previous values with the same N: + U(N,K) = ((2*N-1)*U(N,K-1) - U(N,K-2))/(K-1) + U(N,K-2) + for all N>1, and since U(N,K) is symmetric, a similar relation holds for two + previous values with the same K: + U(N,K>1) = ((2*K-1)*U(N-1,K) - U(N-2,K))/(N-1) + U(N-2,K) + for all K>1. + This allows us to construct an arbitrary row of the U(N,K) table by starting + with the first two values, which are constants. + This saves roughly 2/3 the work in our O(NK) decoding routine, but costs O(K) + multiplications. + Similar relations can be derived for V(N,K), but are not used here. + + For N>0 and K>0, U(N,K) and V(N,K) take on the form of an (N-1)-degree + polynomial for fixed N. + The first few are + U(1,K) = 1, + U(2,K) = 2*K-1, + U(3,K) = (2*K-2)*K+1, + U(4,K) = (((4*K-6)*K+8)*K-3)/3, + U(5,K) = ((((2*K-4)*K+10)*K-8)*K+3)/3, + and + V(1,K) = 2, + V(2,K) = 4*K, + V(3,K) = 4*K*K+2, + V(4,K) = 8*(K*K+2)*K/3, + V(5,K) = ((4*K*K+20)*K*K+6)/3, + for all K>0. + This allows us to derive O(N) encoding and O(N*log(K)) decoding routines for + small N (and indeed decoding is also O(N) for N<3). + + @ARTICLE{Fis86, + author="Thomas R. Fischer", + title="A Pyramid Vector Quantizer", + journal="IEEE Transactions on Information Theory", + volume="IT-32", + number=4, + pages="568--583", + month=Jul, + year=1986 + }*/ + +#if !defined(SMALL_FOOTPRINT) + +/*U(N,K) = U(K,N) := N>0?K>0?U(N-1,K)+U(N,K-1)+U(N-1,K-1):0:K>0?1:0*/ +# define CELT_PVQ_U(_n,_k) (CELT_PVQ_U_ROW[IMIN(_n,_k)][IMAX(_n,_k)]) +/*V(N,K) := U(N,K)+U(N,K+1) = the number of PVQ codewords for a band of size N + with K pulses allocated to it.*/ +# define CELT_PVQ_V(_n,_k) (CELT_PVQ_U(_n,_k)+CELT_PVQ_U(_n,(_k)+1)) + +/*For each V(N,K) supported, we will access element U(min(N,K+1),max(N,K+1)). + Thus, the number of entries in row I is the larger of the maximum number of + pulses we will ever allocate for a given N=I (K=128, or however many fit in + 32 bits, whichever is smaller), plus one, and the maximum N for which + K=I-1 pulses fit in 32 bits. + The largest band size in an Opus Custom mode is 208. + Otherwise, we can limit things to the set of N which can be achieved by + splitting a band from a standard Opus mode: 176, 144, 96, 88, 72, 64, 48, + 44, 36, 32, 24, 22, 18, 16, 8, 4, 2).*/ +#if defined(CUSTOM_MODES) +static const opus_uint32 CELT_PVQ_U_DATA[1488]={ +#else +static const opus_uint32 CELT_PVQ_U_DATA[1272]={ +#endif + /*N=0, K=0...176:*/ + 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +#if defined(CUSTOM_MODES) + /*...208:*/ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, +#endif + /*N=1, K=1...176:*/ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, +#if defined(CUSTOM_MODES) + /*...208:*/ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, +#endif + /*N=2, K=2...176:*/ + 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31, 33, 35, 37, 39, 41, + 43, 45, 47, 49, 51, 53, 55, 57, 59, 61, 63, 65, 67, 69, 71, 73, 75, 77, 79, + 81, 83, 85, 87, 89, 91, 93, 95, 97, 99, 101, 103, 105, 107, 109, 111, 113, + 115, 117, 119, 121, 123, 125, 127, 129, 131, 133, 135, 137, 139, 141, 143, + 145, 147, 149, 151, 153, 155, 157, 159, 161, 163, 165, 167, 169, 171, 173, + 175, 177, 179, 181, 183, 185, 187, 189, 191, 193, 195, 197, 199, 201, 203, + 205, 207, 209, 211, 213, 215, 217, 219, 221, 223, 225, 227, 229, 231, 233, + 235, 237, 239, 241, 243, 245, 247, 249, 251, 253, 255, 257, 259, 261, 263, + 265, 267, 269, 271, 273, 275, 277, 279, 281, 283, 285, 287, 289, 291, 293, + 295, 297, 299, 301, 303, 305, 307, 309, 311, 313, 315, 317, 319, 321, 323, + 325, 327, 329, 331, 333, 335, 337, 339, 341, 343, 345, 347, 349, 351, +#if defined(CUSTOM_MODES) + /*...208:*/ + 353, 355, 357, 359, 361, 363, 365, 367, 369, 371, 373, 375, 377, 379, 381, + 383, 385, 387, 389, 391, 393, 395, 397, 399, 401, 403, 405, 407, 409, 411, + 413, 415, +#endif + /*N=3, K=3...176:*/ + 13, 25, 41, 61, 85, 113, 145, 181, 221, 265, 313, 365, 421, 481, 545, 613, + 685, 761, 841, 925, 1013, 1105, 1201, 1301, 1405, 1513, 1625, 1741, 1861, + 1985, 2113, 2245, 2381, 2521, 2665, 2813, 2965, 3121, 3281, 3445, 3613, 3785, + 3961, 4141, 4325, 4513, 4705, 4901, 5101, 5305, 5513, 5725, 5941, 6161, 6385, + 6613, 6845, 7081, 7321, 7565, 7813, 8065, 8321, 8581, 8845, 9113, 9385, 9661, + 9941, 10225, 10513, 10805, 11101, 11401, 11705, 12013, 12325, 12641, 12961, + 13285, 13613, 13945, 14281, 14621, 14965, 15313, 15665, 16021, 16381, 16745, + 17113, 17485, 17861, 18241, 18625, 19013, 19405, 19801, 20201, 20605, 21013, + 21425, 21841, 22261, 22685, 23113, 23545, 23981, 24421, 24865, 25313, 25765, + 26221, 26681, 27145, 27613, 28085, 28561, 29041, 29525, 30013, 30505, 31001, + 31501, 32005, 32513, 33025, 33541, 34061, 34585, 35113, 35645, 36181, 36721, + 37265, 37813, 38365, 38921, 39481, 40045, 40613, 41185, 41761, 42341, 42925, + 43513, 44105, 44701, 45301, 45905, 46513, 47125, 47741, 48361, 48985, 49613, + 50245, 50881, 51521, 52165, 52813, 53465, 54121, 54781, 55445, 56113, 56785, + 57461, 58141, 58825, 59513, 60205, 60901, 61601, +#if defined(CUSTOM_MODES) + /*...208:*/ + 62305, 63013, 63725, 64441, 65161, 65885, 66613, 67345, 68081, 68821, 69565, + 70313, 71065, 71821, 72581, 73345, 74113, 74885, 75661, 76441, 77225, 78013, + 78805, 79601, 80401, 81205, 82013, 82825, 83641, 84461, 85285, 86113, +#endif + /*N=4, K=4...176:*/ + 63, 129, 231, 377, 575, 833, 1159, 1561, 2047, 2625, 3303, 4089, 4991, 6017, + 7175, 8473, 9919, 11521, 13287, 15225, 17343, 19649, 22151, 24857, 27775, + 30913, 34279, 37881, 41727, 45825, 50183, 54809, 59711, 64897, 70375, 76153, + 82239, 88641, 95367, 102425, 109823, 117569, 125671, 134137, 142975, 152193, + 161799, 171801, 182207, 193025, 204263, 215929, 228031, 240577, 253575, + 267033, 280959, 295361, 310247, 325625, 341503, 357889, 374791, 392217, + 410175, 428673, 447719, 467321, 487487, 508225, 529543, 551449, 573951, + 597057, 620775, 645113, 670079, 695681, 721927, 748825, 776383, 804609, + 833511, 863097, 893375, 924353, 956039, 988441, 1021567, 1055425, 1090023, + 1125369, 1161471, 1198337, 1235975, 1274393, 1313599, 1353601, 1394407, + 1436025, 1478463, 1521729, 1565831, 1610777, 1656575, 1703233, 1750759, + 1799161, 1848447, 1898625, 1949703, 2001689, 2054591, 2108417, 2163175, + 2218873, 2275519, 2333121, 2391687, 2451225, 2511743, 2573249, 2635751, + 2699257, 2763775, 2829313, 2895879, 2963481, 3032127, 3101825, 3172583, + 3244409, 3317311, 3391297, 3466375, 3542553, 3619839, 3698241, 3777767, + 3858425, 3940223, 4023169, 4107271, 4192537, 4278975, 4366593, 4455399, + 4545401, 4636607, 4729025, 4822663, 4917529, 5013631, 5110977, 5209575, + 5309433, 5410559, 5512961, 5616647, 5721625, 5827903, 5935489, 6044391, + 6154617, 6266175, 6379073, 6493319, 6608921, 6725887, 6844225, 6963943, + 7085049, 7207551, +#if defined(CUSTOM_MODES) + /*...208:*/ + 7331457, 7456775, 7583513, 7711679, 7841281, 7972327, 8104825, 8238783, + 8374209, 8511111, 8649497, 8789375, 8930753, 9073639, 9218041, 9363967, + 9511425, 9660423, 9810969, 9963071, 10116737, 10271975, 10428793, 10587199, + 10747201, 10908807, 11072025, 11236863, 11403329, 11571431, 11741177, + 11912575, +#endif + /*N=5, K=5...176:*/ + 321, 681, 1289, 2241, 3649, 5641, 8361, 11969, 16641, 22569, 29961, 39041, + 50049, 63241, 78889, 97281, 118721, 143529, 172041, 204609, 241601, 283401, + 330409, 383041, 441729, 506921, 579081, 658689, 746241, 842249, 947241, + 1061761, 1186369, 1321641, 1468169, 1626561, 1797441, 1981449, 2179241, + 2391489, 2618881, 2862121, 3121929, 3399041, 3694209, 4008201, 4341801, + 4695809, 5071041, 5468329, 5888521, 6332481, 6801089, 7295241, 7815849, + 8363841, 8940161, 9545769, 10181641, 10848769, 11548161, 12280841, 13047849, + 13850241, 14689089, 15565481, 16480521, 17435329, 18431041, 19468809, + 20549801, 21675201, 22846209, 24064041, 25329929, 26645121, 28010881, + 29428489, 30899241, 32424449, 34005441, 35643561, 37340169, 39096641, + 40914369, 42794761, 44739241, 46749249, 48826241, 50971689, 53187081, + 55473921, 57833729, 60268041, 62778409, 65366401, 68033601, 70781609, + 73612041, 76526529, 79526721, 82614281, 85790889, 89058241, 92418049, + 95872041, 99421961, 103069569, 106816641, 110664969, 114616361, 118672641, + 122835649, 127107241, 131489289, 135983681, 140592321, 145317129, 150160041, + 155123009, 160208001, 165417001, 170752009, 176215041, 181808129, 187533321, + 193392681, 199388289, 205522241, 211796649, 218213641, 224775361, 231483969, + 238341641, 245350569, 252512961, 259831041, 267307049, 274943241, 282741889, + 290705281, 298835721, 307135529, 315607041, 324252609, 333074601, 342075401, + 351257409, 360623041, 370174729, 379914921, 389846081, 399970689, 410291241, + 420810249, 431530241, 442453761, 453583369, 464921641, 476471169, 488234561, + 500214441, 512413449, 524834241, 537479489, 550351881, 563454121, 576788929, + 590359041, 604167209, 618216201, 632508801, +#if defined(CUSTOM_MODES) + /*...208:*/ + 647047809, 661836041, 676876329, 692171521, 707724481, 723538089, 739615241, + 755958849, 772571841, 789457161, 806617769, 824056641, 841776769, 859781161, + 878072841, 896654849, 915530241, 934702089, 954173481, 973947521, 994027329, + 1014416041, 1035116809, 1056132801, 1077467201, 1099123209, 1121104041, + 1143412929, 1166053121, 1189027881, 1212340489, 1235994241, +#endif + /*N=6, K=6...96:*/ + 1683, 3653, 7183, 13073, 22363, 36365, 56695, 85305, 124515, 177045, 246047, + 335137, 448427, 590557, 766727, 982729, 1244979, 1560549, 1937199, 2383409, + 2908411, 3522221, 4235671, 5060441, 6009091, 7095093, 8332863, 9737793, + 11326283, 13115773, 15124775, 17372905, 19880915, 22670725, 25765455, + 29189457, 32968347, 37129037, 41699767, 46710137, 52191139, 58175189, + 64696159, 71789409, 79491819, 87841821, 96879431, 106646281, 117185651, + 128542501, 140763503, 153897073, 167993403, 183104493, 199284183, 216588185, + 235074115, 254801525, 275831935, 298228865, 322057867, 347386557, 374284647, + 402823977, 433078547, 465124549, 499040399, 534906769, 572806619, 612825229, + 655050231, 699571641, 746481891, 795875861, 847850911, 902506913, 959946283, + 1020274013, 1083597703, 1150027593, 1219676595, 1292660325, 1369097135, + 1449108145, 1532817275, 1620351277, 1711839767, 1807415257, 1907213187, + 2011371957, 2120032959, +#if defined(CUSTOM_MODES) + /*...109:*/ + 2233340609U, 2351442379U, 2474488829U, 2602633639U, 2736033641U, 2874848851U, + 3019242501U, 3169381071U, 3325434321U, 3487575323U, 3655980493U, 3830829623U, + 4012305913U, +#endif + /*N=7, K=7...54*/ + 8989, 19825, 40081, 75517, 134245, 227305, 369305, 579125, 880685, 1303777, + 1884961, 2668525, 3707509, 5064793, 6814249, 9041957, 11847485, 15345233, + 19665841, 24957661, 31388293, 39146185, 48442297, 59511829, 72616013, + 88043969, 106114625, 127178701, 151620757, 179861305, 212358985, 249612805, + 292164445, 340600625, 395555537, 457713341, 527810725, 606639529, 695049433, + 793950709, 904317037, 1027188385, 1163673953, 1314955181, 1482288821, + 1667010073, 1870535785, 2094367717, +#if defined(CUSTOM_MODES) + /*...60:*/ + 2340095869U, 2609401873U, 2904062449U, 3225952925U, 3577050821U, 3959439497U, +#endif + /*N=8, K=8...37*/ + 48639, 108545, 224143, 433905, 795455, 1392065, 2340495, 3800305, 5984767, + 9173505, 13726991, 20103025, 28875327, 40754369, 56610575, 77500017, + 104692735, 139703809, 184327311, 240673265, 311207743, 398796225, 506750351, + 638878193, 799538175, 993696769, 1226990095, 1505789553, 1837271615, + 2229491905U, +#if defined(CUSTOM_MODES) + /*...40:*/ + 2691463695U, 3233240945U, 3866006015U, +#endif + /*N=9, K=9...28:*/ + 265729, 598417, 1256465, 2485825, 4673345, 8405905, 14546705, 24331777, + 39490049, 62390545, 96220561, 145198913, 214828609, 312193553, 446304145, + 628496897, 872893441, 1196924561, 1621925137, 2173806145U, +#if defined(CUSTOM_MODES) + /*...29:*/ + 2883810113U, +#endif + /*N=10, K=10...24:*/ + 1462563, 3317445, 7059735, 14218905, 27298155, 50250765, 89129247, 152951073, + 254831667, 413442773, 654862247, 1014889769, 1541911931, 2300409629U, + 3375210671U, + /*N=11, K=11...19:*/ + 8097453, 18474633, 39753273, 81270333, 158819253, 298199265, 540279585, + 948062325, 1616336765, +#if defined(CUSTOM_MODES) + /*...20:*/ + 2684641785U, +#endif + /*N=12, K=12...18:*/ + 45046719, 103274625, 224298231, 464387817, 921406335, 1759885185, + 3248227095U, + /*N=13, K=13...16:*/ + 251595969, 579168825, 1267854873, 2653649025U, + /*N=14, K=14:*/ + 1409933619 +}; + +#if defined(CUSTOM_MODES) +static const opus_uint32 *const CELT_PVQ_U_ROW[15]={ + CELT_PVQ_U_DATA+ 0,CELT_PVQ_U_DATA+ 208,CELT_PVQ_U_DATA+ 415, + CELT_PVQ_U_DATA+ 621,CELT_PVQ_U_DATA+ 826,CELT_PVQ_U_DATA+1030, + CELT_PVQ_U_DATA+1233,CELT_PVQ_U_DATA+1336,CELT_PVQ_U_DATA+1389, + CELT_PVQ_U_DATA+1421,CELT_PVQ_U_DATA+1441,CELT_PVQ_U_DATA+1455, + CELT_PVQ_U_DATA+1464,CELT_PVQ_U_DATA+1470,CELT_PVQ_U_DATA+1473 +}; +#else +static const opus_uint32 *const CELT_PVQ_U_ROW[15]={ + CELT_PVQ_U_DATA+ 0,CELT_PVQ_U_DATA+ 176,CELT_PVQ_U_DATA+ 351, + CELT_PVQ_U_DATA+ 525,CELT_PVQ_U_DATA+ 698,CELT_PVQ_U_DATA+ 870, + CELT_PVQ_U_DATA+1041,CELT_PVQ_U_DATA+1131,CELT_PVQ_U_DATA+1178, + CELT_PVQ_U_DATA+1207,CELT_PVQ_U_DATA+1226,CELT_PVQ_U_DATA+1240, + CELT_PVQ_U_DATA+1248,CELT_PVQ_U_DATA+1254,CELT_PVQ_U_DATA+1257 +}; +#endif + +#if defined(CUSTOM_MODES) +void get_required_bits(opus_int16 *_bits,int _n,int _maxk,int _frac){ + int k; + /*_maxk==0 => there's nothing to do.*/ + celt_assert(_maxk>0); + _bits[0]=0; + for(k=1;k<=_maxk;k++)_bits[k]=log2_frac(CELT_PVQ_V(_n,k),_frac); +} +#endif + +static opus_uint32 icwrs(int _n,const int *_y){ + opus_uint32 i; + int j; + int k; + celt_assert(_n>=2); + j=_n-1; + i=_y[j]<0; + k=abs(_y[j]); + do{ + j--; + i+=CELT_PVQ_U(_n-j,k); + k+=abs(_y[j]); + if(_y[j]<0)i+=CELT_PVQ_U(_n-j,k+1); + } + while(j>0); + return i; +} + +void encode_pulses(const int *_y,int _n,int _k,ec_enc *_enc){ + celt_assert(_k>0); + ec_enc_uint(_enc,icwrs(_n,_y),CELT_PVQ_V(_n,_k)); +} + +static opus_val32 cwrsi(int _n,int _k,opus_uint32 _i,int *_y){ + opus_uint32 p; + int s; + int k0; + opus_int16 val; + opus_val32 yy=0; + celt_assert(_k>0); + celt_assert(_n>1); + while(_n>2){ + opus_uint32 q; + /*Lots of pulses case:*/ + if(_k>=_n){ + const opus_uint32 *row; + row=CELT_PVQ_U_ROW[_n]; + /*Are the pulses in this dimension negative?*/ + p=row[_k+1]; + s=-(_i>=p); + _i-=p&s; + /*Count how many pulses were placed in this dimension.*/ + k0=_k; + q=row[_n]; + if(q>_i){ + celt_sig_assert(p>q); + _k=_n; + do p=CELT_PVQ_U_ROW[--_k][_n]; + while(p>_i); + } + else for(p=row[_k];p>_i;p=row[_k])_k--; + _i-=p; + val=(k0-_k+s)^s; + *_y++=val; + yy=MAC16_16(yy,val,val); + } + /*Lots of dimensions case:*/ + else{ + /*Are there any pulses in this dimension at all?*/ + p=CELT_PVQ_U_ROW[_k][_n]; + q=CELT_PVQ_U_ROW[_k+1][_n]; + if(p<=_i&&_i<q){ + _i-=p; + *_y++=0; + } + else{ + /*Are the pulses in this dimension negative?*/ + s=-(_i>=q); + _i-=q&s; + /*Count how many pulses were placed in this dimension.*/ + k0=_k; + do p=CELT_PVQ_U_ROW[--_k][_n]; + while(p>_i); + _i-=p; + val=(k0-_k+s)^s; + *_y++=val; + yy=MAC16_16(yy,val,val); + } + } + _n--; + } + /*_n==2*/ + p=2*_k+1; + s=-(_i>=p); + _i-=p&s; + k0=_k; + _k=(_i+1)>>1; + if(_k)_i-=2*_k-1; + val=(k0-_k+s)^s; + *_y++=val; + yy=MAC16_16(yy,val,val); + /*_n==1*/ + s=-(int)_i; + val=(_k+s)^s; + *_y=val; + yy=MAC16_16(yy,val,val); + return yy; +} + +opus_val32 decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){ + return cwrsi(_n,_k,ec_dec_uint(_dec,CELT_PVQ_V(_n,_k)),_y); +} + +#else /* SMALL_FOOTPRINT */ + +/*Computes the next row/column of any recurrence that obeys the relation + u[i][j]=u[i-1][j]+u[i][j-1]+u[i-1][j-1]. + _ui0 is the base case for the new row/column.*/ +static OPUS_INLINE void unext(opus_uint32 *_ui,unsigned _len,opus_uint32 _ui0){ + opus_uint32 ui1; + unsigned j; + /*This do-while will overrun the array if we don't have storage for at least + 2 values.*/ + j=1; do { + ui1=UADD32(UADD32(_ui[j],_ui[j-1]),_ui0); + _ui[j-1]=_ui0; + _ui0=ui1; + } while (++j<_len); + _ui[j-1]=_ui0; +} + +/*Computes the previous row/column of any recurrence that obeys the relation + u[i-1][j]=u[i][j]-u[i][j-1]-u[i-1][j-1]. + _ui0 is the base case for the new row/column.*/ +static OPUS_INLINE void uprev(opus_uint32 *_ui,unsigned _n,opus_uint32 _ui0){ + opus_uint32 ui1; + unsigned j; + /*This do-while will overrun the array if we don't have storage for at least + 2 values.*/ + j=1; do { + ui1=USUB32(USUB32(_ui[j],_ui[j-1]),_ui0); + _ui[j-1]=_ui0; + _ui0=ui1; + } while (++j<_n); + _ui[j-1]=_ui0; +} + +/*Compute V(_n,_k), as well as U(_n,0..._k+1). + _u: On exit, _u[i] contains U(_n,i) for i in [0..._k+1].*/ +static opus_uint32 ncwrs_urow(unsigned _n,unsigned _k,opus_uint32 *_u){ + opus_uint32 um2; + unsigned len; + unsigned k; + len=_k+2; + /*We require storage at least 3 values (e.g., _k>0).*/ + celt_assert(len>=3); + _u[0]=0; + _u[1]=um2=1; + /*If _n==0, _u[0] should be 1 and the rest should be 0.*/ + /*If _n==1, _u[i] should be 1 for i>1.*/ + celt_assert(_n>=2); + /*If _k==0, the following do-while loop will overflow the buffer.*/ + celt_assert(_k>0); + k=2; + do _u[k]=(k<<1)-1; + while(++k<len); + for(k=2;k<_n;k++)unext(_u+1,_k+1,1); + return _u[_k]+_u[_k+1]; +} + +/*Returns the _i'th combination of _k elements chosen from a set of size _n + with associated sign bits. + _y: Returns the vector of pulses. + _u: Must contain entries [0..._k+1] of row _n of U() on input. + Its contents will be destructively modified.*/ +static opus_val32 cwrsi(int _n,int _k,opus_uint32 _i,int *_y,opus_uint32 *_u){ + int j; + opus_int16 val; + opus_val32 yy=0; + celt_assert(_n>0); + j=0; + do{ + opus_uint32 p; + int s; + int yj; + p=_u[_k+1]; + s=-(_i>=p); + _i-=p&s; + yj=_k; + p=_u[_k]; + while(p>_i)p=_u[--_k]; + _i-=p; + yj-=_k; + val=(yj+s)^s; + _y[j]=val; + yy=MAC16_16(yy,val,val); + uprev(_u,_k+2,0); + } + while(++j<_n); + return yy; +} + +/*Returns the index of the given combination of K elements chosen from a set + of size 1 with associated sign bits. + _y: The vector of pulses, whose sum of absolute values is K. + _k: Returns K.*/ +static OPUS_INLINE opus_uint32 icwrs1(const int *_y,int *_k){ + *_k=abs(_y[0]); + return _y[0]<0; +} + +/*Returns the index of the given combination of K elements chosen from a set + of size _n with associated sign bits. + _y: The vector of pulses, whose sum of absolute values must be _k. + _nc: Returns V(_n,_k).*/ +static OPUS_INLINE opus_uint32 icwrs(int _n,int _k,opus_uint32 *_nc,const int *_y, + opus_uint32 *_u){ + opus_uint32 i; + int j; + int k; + /*We can't unroll the first two iterations of the loop unless _n>=2.*/ + celt_assert(_n>=2); + _u[0]=0; + for(k=1;k<=_k+1;k++)_u[k]=(k<<1)-1; + i=icwrs1(_y+_n-1,&k); + j=_n-2; + i+=_u[k]; + k+=abs(_y[j]); + if(_y[j]<0)i+=_u[k+1]; + while(j-->0){ + unext(_u,_k+2,0); + i+=_u[k]; + k+=abs(_y[j]); + if(_y[j]<0)i+=_u[k+1]; + } + *_nc=_u[k]+_u[k+1]; + return i; +} + +#ifdef CUSTOM_MODES +void get_required_bits(opus_int16 *_bits,int _n,int _maxk,int _frac){ + int k; + /*_maxk==0 => there's nothing to do.*/ + celt_assert(_maxk>0); + _bits[0]=0; + if (_n==1) + { + for (k=1;k<=_maxk;k++) + _bits[k] = 1<<_frac; + } + else { + VARDECL(opus_uint32,u); + SAVE_STACK; + ALLOC(u,_maxk+2U,opus_uint32); + ncwrs_urow(_n,_maxk,u); + for(k=1;k<=_maxk;k++) + _bits[k]=log2_frac(u[k]+u[k+1],_frac); + RESTORE_STACK; + } +} +#endif /* CUSTOM_MODES */ + +void encode_pulses(const int *_y,int _n,int _k,ec_enc *_enc){ + opus_uint32 i; + VARDECL(opus_uint32,u); + opus_uint32 nc; + SAVE_STACK; + celt_assert(_k>0); + ALLOC(u,_k+2U,opus_uint32); + i=icwrs(_n,_k,&nc,_y,u); + ec_enc_uint(_enc,i,nc); + RESTORE_STACK; +} + +opus_val32 decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){ + VARDECL(opus_uint32,u); + int ret; + SAVE_STACK; + celt_assert(_k>0); + ALLOC(u,_k+2U,opus_uint32); + ret = cwrsi(_n,_k,ec_dec_uint(_dec,ncwrs_urow(_n,_k,u)),_y,u); + RESTORE_STACK; + return ret; +} + +#endif /* SMALL_FOOTPRINT */ diff --git a/libopus/celt/cwrs.h b/libopus/celt/cwrs.h new file mode 100644 index 0000000..7cd4717 --- /dev/null +++ b/libopus/celt/cwrs.h @@ -0,0 +1,48 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2007-2009 Timothy B. Terriberry + Written by Timothy B. Terriberry and Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CWRS_H +#define CWRS_H + +#include "arch.h" +#include "stack_alloc.h" +#include "entenc.h" +#include "entdec.h" + +#ifdef CUSTOM_MODES +int log2_frac(opus_uint32 val, int frac); +#endif + +void get_required_bits(opus_int16 *bits, int N, int K, int frac); + +void encode_pulses(const int *_y, int N, int K, ec_enc *enc); + +opus_val32 decode_pulses(int *_y, int N, int K, ec_dec *dec); + +#endif /* CWRS_H */ diff --git a/libopus/celt/ecintrin.h b/libopus/celt/ecintrin.h new file mode 100644 index 0000000..2263cff --- /dev/null +++ b/libopus/celt/ecintrin.h @@ -0,0 +1,87 @@ +/* Copyright (c) 2003-2008 Timothy B. Terriberry + Copyright (c) 2008 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/*Some common macros for potential platform-specific optimization.*/ +#include "opus_types.h" +#include <math.h> +#include <limits.h> +#include "arch.h" +#if !defined(_ecintrin_H) +# define _ecintrin_H (1) + +/*Some specific platforms may have optimized intrinsic or OPUS_INLINE assembly + versions of these functions which can substantially improve performance. + We define macros for them to allow easy incorporation of these non-ANSI + features.*/ + +/*Modern gcc (4.x) can compile the naive versions of min and max with cmov if + given an appropriate architecture, but the branchless bit-twiddling versions + are just as fast, and do not require any special target architecture. + Earlier gcc versions (3.x) compiled both code to the same assembly + instructions, because of the way they represented ((_b)>(_a)) internally.*/ +# define EC_MINI(_a,_b) ((_a)+(((_b)-(_a))&-((_b)<(_a)))) + +/*Count leading zeros. + This macro should only be used for implementing ec_ilog(), if it is defined. + All other code should use EC_ILOG() instead.*/ +#if defined(_MSC_VER) && (_MSC_VER >= 1400) +# include <intrin.h> +/*In _DEBUG mode this is not an intrinsic by default.*/ +# pragma intrinsic(_BitScanReverse) + +static __inline int ec_bsr(unsigned long _x){ + unsigned long ret; + _BitScanReverse(&ret,_x); + return (int)ret; +} +# define EC_CLZ0 (1) +# define EC_CLZ(_x) (-ec_bsr(_x)) +#elif defined(ENABLE_TI_DSPLIB) +# include "dsplib.h" +# define EC_CLZ0 (31) +# define EC_CLZ(_x) (_lnorm(_x)) +#elif __GNUC_PREREQ(3,4) +# if INT_MAX>=2147483647 +# define EC_CLZ0 ((int)sizeof(unsigned)*CHAR_BIT) +# define EC_CLZ(_x) (__builtin_clz(_x)) +# elif LONG_MAX>=2147483647L +# define EC_CLZ0 ((int)sizeof(unsigned long)*CHAR_BIT) +# define EC_CLZ(_x) (__builtin_clzl(_x)) +# endif +#endif + +#if defined(EC_CLZ) +/*Note that __builtin_clz is not defined when _x==0, according to the gcc + documentation (and that of the BSR instruction that implements it on x86). + The majority of the time we can never pass it zero. + When we need to, it can be special cased.*/ +# define EC_ILOG(_x) (EC_CLZ0-EC_CLZ(_x)) +#else +int ec_ilog(opus_uint32 _v); +# define EC_ILOG(_x) (ec_ilog(_x)) +#endif +#endif diff --git a/libopus/celt/entcode.c b/libopus/celt/entcode.c new file mode 100644 index 0000000..70f3201 --- /dev/null +++ b/libopus/celt/entcode.c @@ -0,0 +1,153 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "entcode.h" +#include "arch.h" + +#if !defined(EC_CLZ) +/*This is a fallback for systems where we don't know how to access + a BSR or CLZ instruction (see ecintrin.h). + If you are optimizing Opus on a new platform and it has a native CLZ or + BZR (e.g. cell, MIPS, x86, etc) then making it available to Opus will be + an easy performance win.*/ +int ec_ilog(opus_uint32 _v){ + /*On a Pentium M, this branchless version tested as the fastest on + 1,000,000,000 random 32-bit integers, edging out a similar version with + branches, and a 256-entry LUT version.*/ + int ret; + int m; + ret=!!_v; + m=!!(_v&0xFFFF0000)<<4; + _v>>=m; + ret|=m; + m=!!(_v&0xFF00)<<3; + _v>>=m; + ret|=m; + m=!!(_v&0xF0)<<2; + _v>>=m; + ret|=m; + m=!!(_v&0xC)<<1; + _v>>=m; + ret|=m; + ret+=!!(_v&0x2); + return ret; +} +#endif + +#if 1 +/* This is a faster version of ec_tell_frac() that takes advantage + of the low (1/8 bit) resolution to use just a linear function + followed by a lookup to determine the exact transition thresholds. */ +opus_uint32 ec_tell_frac(ec_ctx *_this){ + static const unsigned correction[8] = + {35733, 38967, 42495, 46340, + 50535, 55109, 60097, 65535}; + opus_uint32 nbits; + opus_uint32 r; + int l; + unsigned b; + nbits=_this->nbits_total<<BITRES; + l=EC_ILOG(_this->rng); + r=_this->rng>>(l-16); + b = (r>>12)-8; + b += r>correction[b]; + l = (l<<3)+b; + return nbits-l; +} +#else +opus_uint32 ec_tell_frac(ec_ctx *_this){ + opus_uint32 nbits; + opus_uint32 r; + int l; + int i; + /*To handle the non-integral number of bits still left in the encoder/decoder + state, we compute the worst-case number of bits of val that must be + encoded to ensure that the value is inside the range for any possible + subsequent bits. + The computation here is independent of val itself (the decoder does not + even track that value), even though the real number of bits used after + ec_enc_done() may be 1 smaller if rng is a power of two and the + corresponding trailing bits of val are all zeros. + If we did try to track that special case, then coding a value with a + probability of 1/(1<<n) might sometimes appear to use more than n bits. + This may help explain the surprising result that a newly initialized + encoder or decoder claims to have used 1 bit.*/ + nbits=_this->nbits_total<<BITRES; + l=EC_ILOG(_this->rng); + r=_this->rng>>(l-16); + for(i=BITRES;i-->0;){ + int b; + r=r*r>>15; + b=(int)(r>>16); + l=l<<1|b; + r>>=b; + } + return nbits-l; +} +#endif + +#ifdef USE_SMALL_DIV_TABLE +/* Result of 2^32/(2*i+1), except for i=0. */ +const opus_uint32 SMALL_DIV_TABLE[129] = { + 0xFFFFFFFF, 0x55555555, 0x33333333, 0x24924924, + 0x1C71C71C, 0x1745D174, 0x13B13B13, 0x11111111, + 0x0F0F0F0F, 0x0D79435E, 0x0C30C30C, 0x0B21642C, + 0x0A3D70A3, 0x097B425E, 0x08D3DCB0, 0x08421084, + 0x07C1F07C, 0x07507507, 0x06EB3E45, 0x06906906, + 0x063E7063, 0x05F417D0, 0x05B05B05, 0x0572620A, + 0x05397829, 0x05050505, 0x04D4873E, 0x04A7904A, + 0x047DC11F, 0x0456C797, 0x04325C53, 0x04104104, + 0x03F03F03, 0x03D22635, 0x03B5CC0E, 0x039B0AD1, + 0x0381C0E0, 0x0369D036, 0x03531DEC, 0x033D91D2, + 0x0329161F, 0x03159721, 0x03030303, 0x02F14990, + 0x02E05C0B, 0x02D02D02, 0x02C0B02C, 0x02B1DA46, + 0x02A3A0FD, 0x0295FAD4, 0x0288DF0C, 0x027C4597, + 0x02702702, 0x02647C69, 0x02593F69, 0x024E6A17, + 0x0243F6F0, 0x0239E0D5, 0x02302302, 0x0226B902, + 0x021D9EAD, 0x0214D021, 0x020C49BA, 0x02040810, + 0x01FC07F0, 0x01F44659, 0x01ECC07B, 0x01E573AC, + 0x01DE5D6E, 0x01D77B65, 0x01D0CB58, 0x01CA4B30, + 0x01C3F8F0, 0x01BDD2B8, 0x01B7D6C3, 0x01B20364, + 0x01AC5701, 0x01A6D01A, 0x01A16D3F, 0x019C2D14, + 0x01970E4F, 0x01920FB4, 0x018D3018, 0x01886E5F, + 0x0183C977, 0x017F405F, 0x017AD220, 0x01767DCE, + 0x01724287, 0x016E1F76, 0x016A13CD, 0x01661EC6, + 0x01623FA7, 0x015E75BB, 0x015AC056, 0x01571ED3, + 0x01539094, 0x01501501, 0x014CAB88, 0x0149539E, + 0x01460CBC, 0x0142D662, 0x013FB013, 0x013C995A, + 0x013991C2, 0x013698DF, 0x0133AE45, 0x0130D190, + 0x012E025C, 0x012B404A, 0x01288B01, 0x0125E227, + 0x01234567, 0x0120B470, 0x011E2EF3, 0x011BB4A4, + 0x01194538, 0x0116E068, 0x011485F0, 0x0112358E, + 0x010FEF01, 0x010DB20A, 0x010B7E6E, 0x010953F3, + 0x01073260, 0x0105197F, 0x0103091B, 0x01010101 +}; +#endif diff --git a/libopus/celt/entcode.h b/libopus/celt/entcode.h new file mode 100644 index 0000000..3763e3f --- /dev/null +++ b/libopus/celt/entcode.h @@ -0,0 +1,152 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include "opus_types.h" +#include "opus_defines.h" + +#if !defined(_entcode_H) +# define _entcode_H (1) +# include <limits.h> +# include <stddef.h> +# include "ecintrin.h" + +extern const opus_uint32 SMALL_DIV_TABLE[129]; + +#ifdef OPUS_ARM_ASM +#define USE_SMALL_DIV_TABLE +#endif + +/*OPT: ec_window must be at least 32 bits, but if you have fast arithmetic on a + larger type, you can speed up the decoder by using it here.*/ +typedef opus_uint32 ec_window; +typedef struct ec_ctx ec_ctx; +typedef struct ec_ctx ec_enc; +typedef struct ec_ctx ec_dec; + +# define EC_WINDOW_SIZE ((int)sizeof(ec_window)*CHAR_BIT) + +/*The number of bits to use for the range-coded part of unsigned integers.*/ +# define EC_UINT_BITS (8) + +/*The resolution of fractional-precision bit usage measurements, i.e., + 3 => 1/8th bits.*/ +# define BITRES 3 + +/*The entropy encoder/decoder context. + We use the same structure for both, so that common functions like ec_tell() + can be used on either one.*/ +struct ec_ctx{ + /*Buffered input/output.*/ + unsigned char *buf; + /*The size of the buffer.*/ + opus_uint32 storage; + /*The offset at which the last byte containing raw bits was read/written.*/ + opus_uint32 end_offs; + /*Bits that will be read from/written at the end.*/ + ec_window end_window; + /*Number of valid bits in end_window.*/ + int nend_bits; + /*The total number of whole bits read/written. + This does not include partial bits currently in the range coder.*/ + int nbits_total; + /*The offset at which the next range coder byte will be read/written.*/ + opus_uint32 offs; + /*The number of values in the current range.*/ + opus_uint32 rng; + /*In the decoder: the difference between the top of the current range and + the input value, minus one. + In the encoder: the low end of the current range.*/ + opus_uint32 val; + /*In the decoder: the saved normalization factor from ec_decode(). + In the encoder: the number of oustanding carry propagating symbols.*/ + opus_uint32 ext; + /*A buffered input/output symbol, awaiting carry propagation.*/ + int rem; + /*Nonzero if an error occurred.*/ + int error; +}; + +static OPUS_INLINE opus_uint32 ec_range_bytes(ec_ctx *_this){ + return _this->offs; +} + +static OPUS_INLINE unsigned char *ec_get_buffer(ec_ctx *_this){ + return _this->buf; +} + +static OPUS_INLINE int ec_get_error(ec_ctx *_this){ + return _this->error; +} + +/*Returns the number of bits "used" by the encoded or decoded symbols so far. + This same number can be computed in either the encoder or the decoder, and is + suitable for making coding decisions. + Return: The number of bits. + This will always be slightly larger than the exact value (e.g., all + rounding error is in the positive direction).*/ +static OPUS_INLINE int ec_tell(ec_ctx *_this){ + return _this->nbits_total-EC_ILOG(_this->rng); +} + +/*Returns the number of bits "used" by the encoded or decoded symbols so far. + This same number can be computed in either the encoder or the decoder, and is + suitable for making coding decisions. + Return: The number of bits scaled by 2**BITRES. + This will always be slightly larger than the exact value (e.g., all + rounding error is in the positive direction).*/ +opus_uint32 ec_tell_frac(ec_ctx *_this); + +/* Tested exhaustively for all n and for 1<=d<=256 */ +static OPUS_INLINE opus_uint32 celt_udiv(opus_uint32 n, opus_uint32 d) { + celt_sig_assert(d>0); +#ifdef USE_SMALL_DIV_TABLE + if (d>256) + return n/d; + else { + opus_uint32 t, q; + t = EC_ILOG(d&-d); + q = (opus_uint64)SMALL_DIV_TABLE[d>>t]*(n>>(t-1))>>32; + return q+(n-q*d >= d); + } +#else + return n/d; +#endif +} + +static OPUS_INLINE opus_int32 celt_sudiv(opus_int32 n, opus_int32 d) { + celt_sig_assert(d>0); +#ifdef USE_SMALL_DIV_TABLE + if (n<0) + return -(opus_int32)celt_udiv(-n, d); + else + return celt_udiv(n, d); +#else + return n/d; +#endif +} + +#endif diff --git a/libopus/celt/entdec.c b/libopus/celt/entdec.c new file mode 100644 index 0000000..0b3433e --- /dev/null +++ b/libopus/celt/entdec.c @@ -0,0 +1,245 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stddef.h> +#include "os_support.h" +#include "arch.h" +#include "entdec.h" +#include "mfrngcod.h" + +/*A range decoder. + This is an entropy decoder based upon \cite{Mar79}, which is itself a + rediscovery of the FIFO arithmetic code introduced by \cite{Pas76}. + It is very similar to arithmetic encoding, except that encoding is done with + digits in any base, instead of with bits, and so it is faster when using + larger bases (i.e.: a byte). + The author claims an average waste of $\frac{1}{2}\log_b(2b)$ bits, where $b$ + is the base, longer than the theoretical optimum, but to my knowledge there + is no published justification for this claim. + This only seems true when using near-infinite precision arithmetic so that + the process is carried out with no rounding errors. + + An excellent description of implementation details is available at + http://www.arturocampos.com/ac_range.html + A recent work \cite{MNW98} which proposes several changes to arithmetic + encoding for efficiency actually re-discovers many of the principles + behind range encoding, and presents a good theoretical analysis of them. + + End of stream is handled by writing out the smallest number of bits that + ensures that the stream will be correctly decoded regardless of the value of + any subsequent bits. + ec_tell() can be used to determine how many bits were needed to decode + all the symbols thus far; other data can be packed in the remaining bits of + the input buffer. + @PHDTHESIS{Pas76, + author="Richard Clark Pasco", + title="Source coding algorithms for fast data compression", + school="Dept. of Electrical Engineering, Stanford University", + address="Stanford, CA", + month=May, + year=1976 + } + @INPROCEEDINGS{Mar79, + author="Martin, G.N.N.", + title="Range encoding: an algorithm for removing redundancy from a digitised + message", + booktitle="Video & Data Recording Conference", + year=1979, + address="Southampton", + month=Jul + } + @ARTICLE{MNW98, + author="Alistair Moffat and Radford Neal and Ian H. Witten", + title="Arithmetic Coding Revisited", + journal="{ACM} Transactions on Information Systems", + year=1998, + volume=16, + number=3, + pages="256--294", + month=Jul, + URL="http://www.stanford.edu/class/ee398a/handouts/papers/Moffat98ArithmCoding.pdf" + }*/ + +static int ec_read_byte(ec_dec *_this){ + return _this->offs<_this->storage?_this->buf[_this->offs++]:0; +} + +static int ec_read_byte_from_end(ec_dec *_this){ + return _this->end_offs<_this->storage? + _this->buf[_this->storage-++(_this->end_offs)]:0; +} + +/*Normalizes the contents of val and rng so that rng lies entirely in the + high-order symbol.*/ +static void ec_dec_normalize(ec_dec *_this){ + /*If the range is too small, rescale it and input some bits.*/ + while(_this->rng<=EC_CODE_BOT){ + int sym; + _this->nbits_total+=EC_SYM_BITS; + _this->rng<<=EC_SYM_BITS; + /*Use up the remaining bits from our last symbol.*/ + sym=_this->rem; + /*Read the next value from the input.*/ + _this->rem=ec_read_byte(_this); + /*Take the rest of the bits we need from this new symbol.*/ + sym=(sym<<EC_SYM_BITS|_this->rem)>>(EC_SYM_BITS-EC_CODE_EXTRA); + /*And subtract them from val, capped to be less than EC_CODE_TOP.*/ + _this->val=((_this->val<<EC_SYM_BITS)+(EC_SYM_MAX&~sym))&(EC_CODE_TOP-1); + } +} + +void ec_dec_init(ec_dec *_this,unsigned char *_buf,opus_uint32 _storage){ + _this->buf=_buf; + _this->storage=_storage; + _this->end_offs=0; + _this->end_window=0; + _this->nend_bits=0; + /*This is the offset from which ec_tell() will subtract partial bits. + The final value after the ec_dec_normalize() call will be the same as in + the encoder, but we have to compensate for the bits that are added there.*/ + _this->nbits_total=EC_CODE_BITS+1 + -((EC_CODE_BITS-EC_CODE_EXTRA)/EC_SYM_BITS)*EC_SYM_BITS; + _this->offs=0; + _this->rng=1U<<EC_CODE_EXTRA; + _this->rem=ec_read_byte(_this); + _this->val=_this->rng-1-(_this->rem>>(EC_SYM_BITS-EC_CODE_EXTRA)); + _this->error=0; + /*Normalize the interval.*/ + ec_dec_normalize(_this); +} + +unsigned ec_decode(ec_dec *_this,unsigned _ft){ + unsigned s; + _this->ext=celt_udiv(_this->rng,_ft); + s=(unsigned)(_this->val/_this->ext); + return _ft-EC_MINI(s+1,_ft); +} + +unsigned ec_decode_bin(ec_dec *_this,unsigned _bits){ + unsigned s; + _this->ext=_this->rng>>_bits; + s=(unsigned)(_this->val/_this->ext); + return (1U<<_bits)-EC_MINI(s+1U,1U<<_bits); +} + +void ec_dec_update(ec_dec *_this,unsigned _fl,unsigned _fh,unsigned _ft){ + opus_uint32 s; + s=IMUL32(_this->ext,_ft-_fh); + _this->val-=s; + _this->rng=_fl>0?IMUL32(_this->ext,_fh-_fl):_this->rng-s; + ec_dec_normalize(_this); +} + +/*The probability of having a "one" is 1/(1<<_logp).*/ +int ec_dec_bit_logp(ec_dec *_this,unsigned _logp){ + opus_uint32 r; + opus_uint32 d; + opus_uint32 s; + int ret; + r=_this->rng; + d=_this->val; + s=r>>_logp; + ret=d<s; + if(!ret)_this->val=d-s; + _this->rng=ret?s:r-s; + ec_dec_normalize(_this); + return ret; +} + +int ec_dec_icdf(ec_dec *_this,const unsigned char *_icdf,unsigned _ftb){ + opus_uint32 r; + opus_uint32 d; + opus_uint32 s; + opus_uint32 t; + int ret; + s=_this->rng; + d=_this->val; + r=s>>_ftb; + ret=-1; + do{ + t=s; + s=IMUL32(r,_icdf[++ret]); + } + while(d<s); + _this->val=d-s; + _this->rng=t-s; + ec_dec_normalize(_this); + return ret; +} + +opus_uint32 ec_dec_uint(ec_dec *_this,opus_uint32 _ft){ + unsigned ft; + unsigned s; + int ftb; + /*In order to optimize EC_ILOG(), it is undefined for the value 0.*/ + celt_assert(_ft>1); + _ft--; + ftb=EC_ILOG(_ft); + if(ftb>EC_UINT_BITS){ + opus_uint32 t; + ftb-=EC_UINT_BITS; + ft=(unsigned)(_ft>>ftb)+1; + s=ec_decode(_this,ft); + ec_dec_update(_this,s,s+1,ft); + t=(opus_uint32)s<<ftb|ec_dec_bits(_this,ftb); + if(t<=_ft)return t; + _this->error=1; + return _ft; + } + else{ + _ft++; + s=ec_decode(_this,(unsigned)_ft); + ec_dec_update(_this,s,s+1,(unsigned)_ft); + return s; + } +} + +opus_uint32 ec_dec_bits(ec_dec *_this,unsigned _bits){ + ec_window window; + int available; + opus_uint32 ret; + window=_this->end_window; + available=_this->nend_bits; + if((unsigned)available<_bits){ + do{ + window|=(ec_window)ec_read_byte_from_end(_this)<<available; + available+=EC_SYM_BITS; + } + while(available<=EC_WINDOW_SIZE-EC_SYM_BITS); + } + ret=(opus_uint32)window&(((opus_uint32)1<<_bits)-1U); + window>>=_bits; + available-=_bits; + _this->end_window=window; + _this->nend_bits=available; + _this->nbits_total+=_bits; + return ret; +} diff --git a/libopus/celt/entdec.h b/libopus/celt/entdec.h new file mode 100644 index 0000000..025fc18 --- /dev/null +++ b/libopus/celt/entdec.h @@ -0,0 +1,100 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(_entdec_H) +# define _entdec_H (1) +# include <limits.h> +# include "entcode.h" + +/*Initializes the decoder. + _buf: The input buffer to use. + Return: 0 on success, or a negative value on error.*/ +void ec_dec_init(ec_dec *_this,unsigned char *_buf,opus_uint32 _storage); + +/*Calculates the cumulative frequency for the next symbol. + This can then be fed into the probability model to determine what that + symbol is, and the additional frequency information required to advance to + the next symbol. + This function cannot be called more than once without a corresponding call to + ec_dec_update(), or decoding will not proceed correctly. + _ft: The total frequency of the symbols in the alphabet the next symbol was + encoded with. + Return: A cumulative frequency representing the encoded symbol. + If the cumulative frequency of all the symbols before the one that + was encoded was fl, and the cumulative frequency of all the symbols + up to and including the one encoded is fh, then the returned value + will fall in the range [fl,fh).*/ +unsigned ec_decode(ec_dec *_this,unsigned _ft); + +/*Equivalent to ec_decode() with _ft==1<<_bits.*/ +unsigned ec_decode_bin(ec_dec *_this,unsigned _bits); + +/*Advance the decoder past the next symbol using the frequency information the + symbol was encoded with. + Exactly one call to ec_decode() must have been made so that all necessary + intermediate calculations are performed. + _fl: The cumulative frequency of all symbols that come before the symbol + decoded. + _fh: The cumulative frequency of all symbols up to and including the symbol + decoded. + Together with _fl, this defines the range [_fl,_fh) in which the value + returned above must fall. + _ft: The total frequency of the symbols in the alphabet the symbol decoded + was encoded in. + This must be the same as passed to the preceding call to ec_decode().*/ +void ec_dec_update(ec_dec *_this,unsigned _fl,unsigned _fh,unsigned _ft); + +/* Decode a bit that has a 1/(1<<_logp) probability of being a one */ +int ec_dec_bit_logp(ec_dec *_this,unsigned _logp); + +/*Decodes a symbol given an "inverse" CDF table. + No call to ec_dec_update() is necessary after this call. + _icdf: The "inverse" CDF, such that symbol s falls in the range + [s>0?ft-_icdf[s-1]:0,ft-_icdf[s]), where ft=1<<_ftb. + The values must be monotonically non-increasing, and the last value + must be 0. + _ftb: The number of bits of precision in the cumulative distribution. + Return: The decoded symbol s.*/ +int ec_dec_icdf(ec_dec *_this,const unsigned char *_icdf,unsigned _ftb); + +/*Extracts a raw unsigned integer with a non-power-of-2 range from the stream. + The bits must have been encoded with ec_enc_uint(). + No call to ec_dec_update() is necessary after this call. + _ft: The number of integers that can be decoded (one more than the max). + This must be at least 2, and no more than 2**32-1. + Return: The decoded bits.*/ +opus_uint32 ec_dec_uint(ec_dec *_this,opus_uint32 _ft); + +/*Extracts a sequence of raw bits from the stream. + The bits must have been encoded with ec_enc_bits(). + No call to ec_dec_update() is necessary after this call. + _ftb: The number of bits to extract. + This must be between 0 and 25, inclusive. + Return: The decoded bits.*/ +opus_uint32 ec_dec_bits(ec_dec *_this,unsigned _ftb); + +#endif diff --git a/libopus/celt/entenc.c b/libopus/celt/entenc.c new file mode 100644 index 0000000..f1750d2 --- /dev/null +++ b/libopus/celt/entenc.c @@ -0,0 +1,294 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if defined(HAVE_CONFIG_H) +# include "config.h" +#endif +#include "os_support.h" +#include "arch.h" +#include "entenc.h" +#include "mfrngcod.h" + +/*A range encoder. + See entdec.c and the references for implementation details \cite{Mar79,MNW98}. + + @INPROCEEDINGS{Mar79, + author="Martin, G.N.N.", + title="Range encoding: an algorithm for removing redundancy from a digitised + message", + booktitle="Video \& Data Recording Conference", + year=1979, + address="Southampton", + month=Jul + } + @ARTICLE{MNW98, + author="Alistair Moffat and Radford Neal and Ian H. Witten", + title="Arithmetic Coding Revisited", + journal="{ACM} Transactions on Information Systems", + year=1998, + volume=16, + number=3, + pages="256--294", + month=Jul, + URL="http://www.stanford.edu/class/ee398/handouts/papers/Moffat98ArithmCoding.pdf" + }*/ + +static int ec_write_byte(ec_enc *_this,unsigned _value){ + if(_this->offs+_this->end_offs>=_this->storage)return -1; + _this->buf[_this->offs++]=(unsigned char)_value; + return 0; +} + +static int ec_write_byte_at_end(ec_enc *_this,unsigned _value){ + if(_this->offs+_this->end_offs>=_this->storage)return -1; + _this->buf[_this->storage-++(_this->end_offs)]=(unsigned char)_value; + return 0; +} + +/*Outputs a symbol, with a carry bit. + If there is a potential to propagate a carry over several symbols, they are + buffered until it can be determined whether or not an actual carry will + occur. + If the counter for the buffered symbols overflows, then the stream becomes + undecodable. + This gives a theoretical limit of a few billion symbols in a single packet on + 32-bit systems. + The alternative is to truncate the range in order to force a carry, but + requires similar carry tracking in the decoder, needlessly slowing it down.*/ +static void ec_enc_carry_out(ec_enc *_this,int _c){ + if(_c!=EC_SYM_MAX){ + /*No further carry propagation possible, flush buffer.*/ + int carry; + carry=_c>>EC_SYM_BITS; + /*Don't output a byte on the first write. + This compare should be taken care of by branch-prediction thereafter.*/ + if(_this->rem>=0)_this->error|=ec_write_byte(_this,_this->rem+carry); + if(_this->ext>0){ + unsigned sym; + sym=(EC_SYM_MAX+carry)&EC_SYM_MAX; + do _this->error|=ec_write_byte(_this,sym); + while(--(_this->ext)>0); + } + _this->rem=_c&EC_SYM_MAX; + } + else _this->ext++; +} + +static OPUS_INLINE void ec_enc_normalize(ec_enc *_this){ + /*If the range is too small, output some bits and rescale it.*/ + while(_this->rng<=EC_CODE_BOT){ + ec_enc_carry_out(_this,(int)(_this->val>>EC_CODE_SHIFT)); + /*Move the next-to-high-order symbol into the high-order position.*/ + _this->val=(_this->val<<EC_SYM_BITS)&(EC_CODE_TOP-1); + _this->rng<<=EC_SYM_BITS; + _this->nbits_total+=EC_SYM_BITS; + } +} + +void ec_enc_init(ec_enc *_this,unsigned char *_buf,opus_uint32 _size){ + _this->buf=_buf; + _this->end_offs=0; + _this->end_window=0; + _this->nend_bits=0; + /*This is the offset from which ec_tell() will subtract partial bits.*/ + _this->nbits_total=EC_CODE_BITS+1; + _this->offs=0; + _this->rng=EC_CODE_TOP; + _this->rem=-1; + _this->val=0; + _this->ext=0; + _this->storage=_size; + _this->error=0; +} + +void ec_encode(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _ft){ + opus_uint32 r; + r=celt_udiv(_this->rng,_ft); + if(_fl>0){ + _this->val+=_this->rng-IMUL32(r,(_ft-_fl)); + _this->rng=IMUL32(r,(_fh-_fl)); + } + else _this->rng-=IMUL32(r,(_ft-_fh)); + ec_enc_normalize(_this); +} + +void ec_encode_bin(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _bits){ + opus_uint32 r; + r=_this->rng>>_bits; + if(_fl>0){ + _this->val+=_this->rng-IMUL32(r,((1U<<_bits)-_fl)); + _this->rng=IMUL32(r,(_fh-_fl)); + } + else _this->rng-=IMUL32(r,((1U<<_bits)-_fh)); + ec_enc_normalize(_this); +} + +/*The probability of having a "one" is 1/(1<<_logp).*/ +void ec_enc_bit_logp(ec_enc *_this,int _val,unsigned _logp){ + opus_uint32 r; + opus_uint32 s; + opus_uint32 l; + r=_this->rng; + l=_this->val; + s=r>>_logp; + r-=s; + if(_val)_this->val=l+r; + _this->rng=_val?s:r; + ec_enc_normalize(_this); +} + +void ec_enc_icdf(ec_enc *_this,int _s,const unsigned char *_icdf,unsigned _ftb){ + opus_uint32 r; + r=_this->rng>>_ftb; + if(_s>0){ + _this->val+=_this->rng-IMUL32(r,_icdf[_s-1]); + _this->rng=IMUL32(r,_icdf[_s-1]-_icdf[_s]); + } + else _this->rng-=IMUL32(r,_icdf[_s]); + ec_enc_normalize(_this); +} + +void ec_enc_uint(ec_enc *_this,opus_uint32 _fl,opus_uint32 _ft){ + unsigned ft; + unsigned fl; + int ftb; + /*In order to optimize EC_ILOG(), it is undefined for the value 0.*/ + celt_assert(_ft>1); + _ft--; + ftb=EC_ILOG(_ft); + if(ftb>EC_UINT_BITS){ + ftb-=EC_UINT_BITS; + ft=(_ft>>ftb)+1; + fl=(unsigned)(_fl>>ftb); + ec_encode(_this,fl,fl+1,ft); + ec_enc_bits(_this,_fl&(((opus_uint32)1<<ftb)-1U),ftb); + } + else ec_encode(_this,_fl,_fl+1,_ft+1); +} + +void ec_enc_bits(ec_enc *_this,opus_uint32 _fl,unsigned _bits){ + ec_window window; + int used; + window=_this->end_window; + used=_this->nend_bits; + celt_assert(_bits>0); + if(used+_bits>EC_WINDOW_SIZE){ + do{ + _this->error|=ec_write_byte_at_end(_this,(unsigned)window&EC_SYM_MAX); + window>>=EC_SYM_BITS; + used-=EC_SYM_BITS; + } + while(used>=EC_SYM_BITS); + } + window|=(ec_window)_fl<<used; + used+=_bits; + _this->end_window=window; + _this->nend_bits=used; + _this->nbits_total+=_bits; +} + +void ec_enc_patch_initial_bits(ec_enc *_this,unsigned _val,unsigned _nbits){ + int shift; + unsigned mask; + celt_assert(_nbits<=EC_SYM_BITS); + shift=EC_SYM_BITS-_nbits; + mask=((1<<_nbits)-1)<<shift; + if(_this->offs>0){ + /*The first byte has been finalized.*/ + _this->buf[0]=(unsigned char)((_this->buf[0]&~mask)|_val<<shift); + } + else if(_this->rem>=0){ + /*The first byte is still awaiting carry propagation.*/ + _this->rem=(_this->rem&~mask)|_val<<shift; + } + else if(_this->rng<=(EC_CODE_TOP>>_nbits)){ + /*The renormalization loop has never been run.*/ + _this->val=(_this->val&~((opus_uint32)mask<<EC_CODE_SHIFT))| + (opus_uint32)_val<<(EC_CODE_SHIFT+shift); + } + /*The encoder hasn't even encoded _nbits of data yet.*/ + else _this->error=-1; +} + +void ec_enc_shrink(ec_enc *_this,opus_uint32 _size){ + celt_assert(_this->offs+_this->end_offs<=_size); + OPUS_MOVE(_this->buf+_size-_this->end_offs, + _this->buf+_this->storage-_this->end_offs,_this->end_offs); + _this->storage=_size; +} + +void ec_enc_done(ec_enc *_this){ + ec_window window; + int used; + opus_uint32 msk; + opus_uint32 end; + int l; + /*We output the minimum number of bits that ensures that the symbols encoded + thus far will be decoded correctly regardless of the bits that follow.*/ + l=EC_CODE_BITS-EC_ILOG(_this->rng); + msk=(EC_CODE_TOP-1)>>l; + end=(_this->val+msk)&~msk; + if((end|msk)>=_this->val+_this->rng){ + l++; + msk>>=1; + end=(_this->val+msk)&~msk; + } + while(l>0){ + ec_enc_carry_out(_this,(int)(end>>EC_CODE_SHIFT)); + end=(end<<EC_SYM_BITS)&(EC_CODE_TOP-1); + l-=EC_SYM_BITS; + } + /*If we have a buffered byte flush it into the output buffer.*/ + if(_this->rem>=0||_this->ext>0)ec_enc_carry_out(_this,0); + /*If we have buffered extra bits, flush them as well.*/ + window=_this->end_window; + used=_this->nend_bits; + while(used>=EC_SYM_BITS){ + _this->error|=ec_write_byte_at_end(_this,(unsigned)window&EC_SYM_MAX); + window>>=EC_SYM_BITS; + used-=EC_SYM_BITS; + } + /*Clear any excess space and add any remaining extra bits to the last byte.*/ + if(!_this->error){ + OPUS_CLEAR(_this->buf+_this->offs, + _this->storage-_this->offs-_this->end_offs); + if(used>0){ + /*If there's no range coder data at all, give up.*/ + if(_this->end_offs>=_this->storage)_this->error=-1; + else{ + l=-l; + /*If we've busted, don't add too many extra bits to the last byte; it + would corrupt the range coder data, and that's more important.*/ + if(_this->offs+_this->end_offs>=_this->storage&&l<used){ + window&=(1<<l)-1; + _this->error=-1; + } + _this->buf[_this->storage-_this->end_offs-1]|=(unsigned char)window; + } + } + } +} diff --git a/libopus/celt/entenc.h b/libopus/celt/entenc.h new file mode 100644 index 0000000..f502eaf --- /dev/null +++ b/libopus/celt/entenc.h @@ -0,0 +1,110 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(_entenc_H) +# define _entenc_H (1) +# include <stddef.h> +# include "entcode.h" + +/*Initializes the encoder. + _buf: The buffer to store output bytes in. + _size: The size of the buffer, in chars.*/ +void ec_enc_init(ec_enc *_this,unsigned char *_buf,opus_uint32 _size); +/*Encodes a symbol given its frequency information. + The frequency information must be discernable by the decoder, assuming it + has read only the previous symbols from the stream. + It is allowable to change the frequency information, or even the entire + source alphabet, so long as the decoder can tell from the context of the + previously encoded information that it is supposed to do so as well. + _fl: The cumulative frequency of all symbols that come before the one to be + encoded. + _fh: The cumulative frequency of all symbols up to and including the one to + be encoded. + Together with _fl, this defines the range [_fl,_fh) in which the + decoded value will fall. + _ft: The sum of the frequencies of all the symbols*/ +void ec_encode(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _ft); + +/*Equivalent to ec_encode() with _ft==1<<_bits.*/ +void ec_encode_bin(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _bits); + +/* Encode a bit that has a 1/(1<<_logp) probability of being a one */ +void ec_enc_bit_logp(ec_enc *_this,int _val,unsigned _logp); + +/*Encodes a symbol given an "inverse" CDF table. + _s: The index of the symbol to encode. + _icdf: The "inverse" CDF, such that symbol _s falls in the range + [_s>0?ft-_icdf[_s-1]:0,ft-_icdf[_s]), where ft=1<<_ftb. + The values must be monotonically non-increasing, and the last value + must be 0. + _ftb: The number of bits of precision in the cumulative distribution.*/ +void ec_enc_icdf(ec_enc *_this,int _s,const unsigned char *_icdf,unsigned _ftb); + +/*Encodes a raw unsigned integer in the stream. + _fl: The integer to encode. + _ft: The number of integers that can be encoded (one more than the max). + This must be at least 2, and no more than 2**32-1.*/ +void ec_enc_uint(ec_enc *_this,opus_uint32 _fl,opus_uint32 _ft); + +/*Encodes a sequence of raw bits in the stream. + _fl: The bits to encode. + _ftb: The number of bits to encode. + This must be between 1 and 25, inclusive.*/ +void ec_enc_bits(ec_enc *_this,opus_uint32 _fl,unsigned _ftb); + +/*Overwrites a few bits at the very start of an existing stream, after they + have already been encoded. + This makes it possible to have a few flags up front, where it is easy for + decoders to access them without parsing the whole stream, even if their + values are not determined until late in the encoding process, without having + to buffer all the intermediate symbols in the encoder. + In order for this to work, at least _nbits bits must have already been + encoded using probabilities that are an exact power of two. + The encoder can verify the number of encoded bits is sufficient, but cannot + check this latter condition. + _val: The bits to encode (in the least _nbits significant bits). + They will be decoded in order from most-significant to least. + _nbits: The number of bits to overwrite. + This must be no more than 8.*/ +void ec_enc_patch_initial_bits(ec_enc *_this,unsigned _val,unsigned _nbits); + +/*Compacts the data to fit in the target size. + This moves up the raw bits at the end of the current buffer so they are at + the end of the new buffer size. + The caller must ensure that the amount of data that's already been written + will fit in the new size. + _size: The number of bytes in the new buffer. + This must be large enough to contain the bits already written, and + must be no larger than the existing size.*/ +void ec_enc_shrink(ec_enc *_this,opus_uint32 _size); + +/*Indicates that there are no more symbols to encode. + All reamining output bytes are flushed to the output buffer. + ec_enc_init() must be called before the encoder can be used again.*/ +void ec_enc_done(ec_enc *_this); + +#endif diff --git a/libopus/celt/fixed_debug.h b/libopus/celt/fixed_debug.h new file mode 100644 index 0000000..f435295 --- /dev/null +++ b/libopus/celt/fixed_debug.h @@ -0,0 +1,791 @@ +/* Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2012 Xiph.Org Foundation */ +/** + @file fixed_debug.h + @brief Fixed-point operations with debugging +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_DEBUG_H +#define FIXED_DEBUG_H + +#include <stdio.h> +#include "opus_defines.h" + +#ifdef CELT_C +OPUS_EXPORT opus_int64 celt_mips=0; +#else +extern opus_int64 celt_mips; +#endif + +#define MULT16_16SU(a,b) ((opus_val32)(opus_val16)(a)*(opus_val32)(opus_uint16)(b)) +#define MULT32_32_Q31(a,b) ADD32(ADD32(SHL32(MULT16_16(SHR32((a),16),SHR((b),16)),1), SHR32(MULT16_16SU(SHR32((a),16),((b)&0x0000ffff)),15)), SHR32(MULT16_16SU(SHR32((b),16),((a)&0x0000ffff)),15)) + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#define MULT16_32_Q16(a,b) ADD32(MULT16_16((a),SHR32((b),16)), SHR32(MULT16_16SU((a),((b)&0x0000ffff)),16)) + +#define MULT16_32_P16(a,b) MULT16_32_PX(a,b,16) + +#define QCONST16(x,bits) ((opus_val16)(.5+(x)*(((opus_val32)1)<<(bits)))) +#define QCONST32(x,bits) ((opus_val32)(.5+(x)*(((opus_val32)1)<<(bits)))) + +#define VERIFY_SHORT(x) ((x)<=32767&&(x)>=-32768) +#define VERIFY_INT(x) ((x)<=2147483647LL&&(x)>=-2147483648LL) +#define VERIFY_UINT(x) ((x)<=(2147483647LLU<<1)) + +#define SHR(a,b) SHR32(a,b) +#define PSHR(a,b) PSHR32(a,b) + +/** Add two 32-bit values, ignore any overflows */ +#define ADD32_ovflw(a,b) (celt_mips+=2,(opus_val32)((opus_uint32)(a)+(opus_uint32)(b))) +/** Subtract two 32-bit values, ignore any overflows */ +#define SUB32_ovflw(a,b) (celt_mips+=2,(opus_val32)((opus_uint32)(a)-(opus_uint32)(b))) +/* Avoid MSVC warning C4146: unary minus operator applied to unsigned type */ +/** Negate 32-bit value, ignore any overflows */ +#define NEG32_ovflw(a) (celt_mips+=2,(opus_val32)(0-(opus_uint32)(a))) + +static OPUS_INLINE short NEG16(int x) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "NEG16: input is not short: %d\n", (int)x); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = -x; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "NEG16: output is not short: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} +static OPUS_INLINE int NEG32(opus_int64 x) +{ + opus_int64 res; + if (!VERIFY_INT(x)) + { + fprintf (stderr, "NEG16: input is not int: %d\n", (int)x); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = -x; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "NEG16: output is not int: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define EXTRACT16(x) EXTRACT16_(x, __FILE__, __LINE__) +static OPUS_INLINE short EXTRACT16_(int x, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "EXTRACT16: input is not short: %d in %s: line %d\n", x, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = x; + celt_mips++; + return res; +} + +#define EXTEND32(x) EXTEND32_(x, __FILE__, __LINE__) +static OPUS_INLINE int EXTEND32_(int x, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "EXTEND32: input is not short: %d in %s: line %d\n", x, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = x; + celt_mips++; + return res; +} + +#define SHR16(a, shift) SHR16_(a, shift, __FILE__, __LINE__) +static OPUS_INLINE short SHR16_(int a, int shift, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHR16: inputs are not short: %d >> %d in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a>>shift; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "SHR16: output is not short: %d in %s: line %d\n", res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} +#define SHL16(a, shift) SHL16_(a, shift, __FILE__, __LINE__) +static OPUS_INLINE short SHL16_(int a, int shift, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHL16: inputs are not short: %d %d in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a<<shift; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "SHL16: output is not short: %d in %s: line %d\n", res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +static OPUS_INLINE int SHR32(opus_int64 a, int shift) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHR32: inputs are not int: %d %d\n", (int)a, shift); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a>>shift; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "SHR32: output is not int: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} +#define SHL32(a, shift) SHL32_(a, shift, __FILE__, __LINE__) +static OPUS_INLINE int SHL32_(opus_int64 a, int shift, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHL32: inputs are not int: %lld %d in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a<<shift; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "SHL32: output is not int: %lld<<%d = %lld in %s: line %d\n", a, shift, res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define PSHR32(a,shift) (celt_mips--,SHR32(ADD32((a),(((opus_val32)(1)<<((shift))>>1))),shift)) +#define VSHR32(a, shift) (((shift)>0) ? SHR32(a, shift) : SHL32(a, -(shift))) + +#define ROUND16(x,a) (celt_mips--,EXTRACT16(PSHR32((x),(a)))) +#define SROUND16(x,a) (celt_mips--,EXTRACT16(SATURATE(PSHR32(x,a), 32767))); + +#define HALF16(x) (SHR16(x,1)) +#define HALF32(x) (SHR32(x,1)) + +#define ADD16(a, b) ADD16_(a, b, __FILE__, __LINE__) +static OPUS_INLINE short ADD16_(int a, int b, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "ADD16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a+b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "ADD16: output is not short: %d+%d=%d in %s: line %d\n", a,b,res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +#define SUB16(a, b) SUB16_(a, b, __FILE__, __LINE__) +static OPUS_INLINE short SUB16_(int a, int b, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "SUB16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a-b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "SUB16: output is not short: %d in %s: line %d\n", res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +#define ADD32(a, b) ADD32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int ADD32_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "ADD32: inputs are not int: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a+b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "ADD32: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define SUB32(a, b) SUB32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int SUB32_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "SUB32: inputs are not int: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a-b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "SUB32: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#undef UADD32 +#define UADD32(a, b) UADD32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE unsigned int UADD32_(opus_uint64 a, opus_uint64 b, char *file, int line) +{ + opus_uint64 res; + if (!VERIFY_UINT(a) || !VERIFY_UINT(b)) + { + fprintf (stderr, "UADD32: inputs are not uint32: %llu %llu in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a+b; + if (!VERIFY_UINT(res)) + { + fprintf (stderr, "UADD32: output is not uint32: %llu in %s: line %d\n", res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#undef USUB32 +#define USUB32(a, b) USUB32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE unsigned int USUB32_(opus_uint64 a, opus_uint64 b, char *file, int line) +{ + opus_uint64 res; + if (!VERIFY_UINT(a) || !VERIFY_UINT(b)) + { + fprintf (stderr, "USUB32: inputs are not uint32: %llu %llu in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (a<b) + { + fprintf (stderr, "USUB32: inputs underflow: %llu < %llu in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a-b; + if (!VERIFY_UINT(res)) + { + fprintf (stderr, "USUB32: output is not uint32: %llu - %llu = %llu in %s: line %d\n", a, b, res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +/* result fits in 16 bits */ +static OPUS_INLINE short MULT16_16_16(int a, int b) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_16: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a*b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_16: output is not short: %d\n", res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +#define MULT16_16(a, b) MULT16_16_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int MULT16_16_(int a, int b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +#define MAC16_16(c,a,b) (celt_mips-=2,ADD32((c),MULT16_16((a),(b)))) + +#define MULT16_32_QX(a, b, Q) MULT16_32_QX_(a, b, Q, __FILE__, __LINE__) +static OPUS_INLINE int MULT16_32_QX_(int a, opus_int64 b, int Q, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "MULT16_32_Q%d: inputs are not short+int: %d %d in %s: line %d\n", Q, (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (ABS32(b)>=((opus_val32)(1)<<(15+Q))) + { + fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d in %s: line %d\n", Q, (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = (((opus_int64)a)*(opus_int64)b) >> Q; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_32_Q%d: output is not int: %d*%d=%d in %s: line %d\n", Q, (int)a, (int)b,(int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (Q==15) + celt_mips+=3; + else + celt_mips+=4; + return res; +} + +#define MULT16_32_PX(a, b, Q) MULT16_32_PX_(a, b, Q, __FILE__, __LINE__) +static OPUS_INLINE int MULT16_32_PX_(int a, opus_int64 b, int Q, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "MULT16_32_P%d: inputs are not short+int: %d %d in %s: line %d\n\n", Q, (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (ABS32(b)>=((opus_int64)(1)<<(15+Q))) + { + fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d in %s: line %d\n\n", Q, (int)a, (int)b,file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((((opus_int64)a)*(opus_int64)b) + (((opus_val32)(1)<<Q)>>1))>> Q; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_32_P%d: output is not int: %d*%d=%d in %s: line %d\n\n", Q, (int)a, (int)b,(int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (Q==15) + celt_mips+=4; + else + celt_mips+=5; + return res; +} + +#define MULT16_32_Q15(a,b) MULT16_32_QX(a,b,15) +#define MAC16_32_Q15(c,a,b) (celt_mips-=2,ADD32((c),MULT16_32_Q15((a),(b)))) +#define MAC16_32_Q16(c,a,b) (celt_mips-=2,ADD32((c),MULT16_32_Q16((a),(b)))) + +static OPUS_INLINE int SATURATE(int a, int b) +{ + if (a>b) + a=b; + if (a<-b) + a = -b; + celt_mips+=3; + return a; +} + +static OPUS_INLINE opus_int16 SATURATE16(opus_int32 a) +{ + celt_mips+=3; + if (a>32767) + return 32767; + else if (a<-32768) + return -32768; + else return a; +} + +static OPUS_INLINE int MULT16_16_Q11_32(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q11: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 11; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_Q11: output is not short: %d*%d=%d\n", (int)a, (int)b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=3; + return res; +} +static OPUS_INLINE short MULT16_16_Q13(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q13: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 13; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_Q13: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=3; + return res; +} +static OPUS_INLINE short MULT16_16_Q14(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q14: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 14; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_Q14: output is not short: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=3; + return res; +} + +#define MULT16_16_Q15(a, b) MULT16_16_Q15_(a, b, __FILE__, __LINE__) +static OPUS_INLINE short MULT16_16_Q15_(int a, int b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q15: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 15; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_Q15: output is not short: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=1; + return res; +} + +static OPUS_INLINE short MULT16_16_P13(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P13: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res += 4096; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_P13: overflow: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res >>= 13; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_P13: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=4; + return res; +} +static OPUS_INLINE short MULT16_16_P14(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P14: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res += 8192; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_P14: overflow: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res >>= 14; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_P14: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=4; + return res; +} +static OPUS_INLINE short MULT16_16_P15(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P15: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res += 16384; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_P15: overflow: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res >>= 15; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_P15: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define DIV32_16(a, b) DIV32_16_(a, b, __FILE__, __LINE__) + +static OPUS_INLINE int DIV32_16_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (b==0) + { + fprintf(stderr, "DIV32_16: divide by zero: %d/%d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + return 0; + } + if (!VERIFY_INT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "DIV32_16: inputs are not int/short: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a/b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "DIV32_16: output is not short: %d / %d = %d in %s: line %d\n", (int)a,(int)b,(int)res, file, line); + if (res>32767) + res = 32767; + if (res<-32768) + res = -32768; +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=35; + return res; +} + +#define DIV32(a, b) DIV32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int DIV32_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (b==0) + { + fprintf(stderr, "DIV32: divide by zero: %d/%d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + return 0; + } + + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "DIV32: inputs are not int/short: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a/b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "DIV32: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=70; + return res; +} + +static OPUS_INLINE opus_val16 SIG2WORD16_generic(celt_sig x) +{ + x = PSHR32(x, SIG_SHIFT); + x = MAX32(x, -32768); + x = MIN32(x, 32767); + return EXTRACT16(x); +} +#define SIG2WORD16(x) (SIG2WORD16_generic(x)) + + +#undef PRINT_MIPS +#define PRINT_MIPS(file) do {fprintf (file, "total complexity = %llu MIPS\n", celt_mips);} while (0); + +#endif diff --git a/libopus/celt/fixed_generic.h b/libopus/celt/fixed_generic.h new file mode 100644 index 0000000..5f4abda --- /dev/null +++ b/libopus/celt/fixed_generic.h @@ -0,0 +1,178 @@ +/* Copyright (C) 2007-2009 Xiph.Org Foundation + Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2008 CSIRO */ +/** + @file fixed_generic.h + @brief Generic fixed-point operations +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_GENERIC_H +#define FIXED_GENERIC_H + +/** Multiply a 16-bit signed value by a 16-bit unsigned value. The result is a 32-bit signed value */ +#define MULT16_16SU(a,b) ((opus_val32)(opus_val16)(a)*(opus_val32)(opus_uint16)(b)) + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#if OPUS_FAST_INT64 +#define MULT16_32_Q16(a,b) ((opus_val32)SHR((opus_int64)((opus_val16)(a))*(b),16)) +#else +#define MULT16_32_Q16(a,b) ADD32(MULT16_16((a),SHR((b),16)), SHR(MULT16_16SU((a),((b)&0x0000ffff)),16)) +#endif + +/** 16x32 multiplication, followed by a 16-bit shift right (round-to-nearest). Results fits in 32 bits */ +#if OPUS_FAST_INT64 +#define MULT16_32_P16(a,b) ((opus_val32)PSHR((opus_int64)((opus_val16)(a))*(b),16)) +#else +#define MULT16_32_P16(a,b) ADD32(MULT16_16((a),SHR((b),16)), PSHR(MULT16_16SU((a),((b)&0x0000ffff)),16)) +#endif + +/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ +#if OPUS_FAST_INT64 +#define MULT16_32_Q15(a,b) ((opus_val32)SHR((opus_int64)((opus_val16)(a))*(b),15)) +#else +#define MULT16_32_Q15(a,b) ADD32(SHL(MULT16_16((a),SHR((b),16)),1), SHR(MULT16_16SU((a),((b)&0x0000ffff)),15)) +#endif + +/** 32x32 multiplication, followed by a 31-bit shift right. Results fits in 32 bits */ +#if OPUS_FAST_INT64 +#define MULT32_32_Q31(a,b) ((opus_val32)SHR((opus_int64)(a)*(opus_int64)(b),31)) +#else +#define MULT32_32_Q31(a,b) ADD32(ADD32(SHL(MULT16_16(SHR((a),16),SHR((b),16)),1), SHR(MULT16_16SU(SHR((a),16),((b)&0x0000ffff)),15)), SHR(MULT16_16SU(SHR((b),16),((a)&0x0000ffff)),15)) +#endif + +/** Compile-time conversion of float constant to 16-bit value */ +#define QCONST16(x,bits) ((opus_val16)(.5+(x)*(((opus_val32)1)<<(bits)))) + +/** Compile-time conversion of float constant to 32-bit value */ +#define QCONST32(x,bits) ((opus_val32)(.5+(x)*(((opus_val32)1)<<(bits)))) + +/** Negate a 16-bit value */ +#define NEG16(x) (-(x)) +/** Negate a 32-bit value */ +#define NEG32(x) (-(x)) + +/** Change a 32-bit value into a 16-bit value. The value is assumed to fit in 16-bit, otherwise the result is undefined */ +#define EXTRACT16(x) ((opus_val16)(x)) +/** Change a 16-bit value into a 32-bit value */ +#define EXTEND32(x) ((opus_val32)(x)) + +/** Arithmetic shift-right of a 16-bit value */ +#define SHR16(a,shift) ((a) >> (shift)) +/** Arithmetic shift-left of a 16-bit value */ +#define SHL16(a,shift) ((opus_int16)((opus_uint16)(a)<<(shift))) +/** Arithmetic shift-right of a 32-bit value */ +#define SHR32(a,shift) ((a) >> (shift)) +/** Arithmetic shift-left of a 32-bit value */ +#define SHL32(a,shift) ((opus_int32)((opus_uint32)(a)<<(shift))) + +/** 32-bit arithmetic shift right with rounding-to-nearest instead of rounding down */ +#define PSHR32(a,shift) (SHR32((a)+((EXTEND32(1)<<((shift))>>1)),shift)) +/** 32-bit arithmetic shift right where the argument can be negative */ +#define VSHR32(a, shift) (((shift)>0) ? SHR32(a, shift) : SHL32(a, -(shift))) + +/** "RAW" macros, should not be used outside of this header file */ +#define SHR(a,shift) ((a) >> (shift)) +#define SHL(a,shift) SHL32(a,shift) +#define PSHR(a,shift) (SHR((a)+((EXTEND32(1)<<((shift))>>1)),shift)) +#define SATURATE(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x))) + +#define SATURATE16(x) (EXTRACT16((x)>32767 ? 32767 : (x)<-32768 ? -32768 : (x))) + +/** Shift by a and round-to-neareast 32-bit value. Result is a 16-bit value */ +#define ROUND16(x,a) (EXTRACT16(PSHR32((x),(a)))) +/** Shift by a and round-to-neareast 32-bit value. Result is a saturated 16-bit value */ +#define SROUND16(x,a) EXTRACT16(SATURATE(PSHR32(x,a), 32767)); + +/** Divide by two */ +#define HALF16(x) (SHR16(x,1)) +#define HALF32(x) (SHR32(x,1)) + +/** Add two 16-bit values */ +#define ADD16(a,b) ((opus_val16)((opus_val16)(a)+(opus_val16)(b))) +/** Subtract two 16-bit values */ +#define SUB16(a,b) ((opus_val16)(a)-(opus_val16)(b)) +/** Add two 32-bit values */ +#define ADD32(a,b) ((opus_val32)(a)+(opus_val32)(b)) +/** Subtract two 32-bit values */ +#define SUB32(a,b) ((opus_val32)(a)-(opus_val32)(b)) + +/** Add two 32-bit values, ignore any overflows */ +#define ADD32_ovflw(a,b) ((opus_val32)((opus_uint32)(a)+(opus_uint32)(b))) +/** Subtract two 32-bit values, ignore any overflows */ +#define SUB32_ovflw(a,b) ((opus_val32)((opus_uint32)(a)-(opus_uint32)(b))) +/* Avoid MSVC warning C4146: unary minus operator applied to unsigned type */ +/** Negate 32-bit value, ignore any overflows */ +#define NEG32_ovflw(a) ((opus_val32)(0-(opus_uint32)(a))) + +/** 16x16 multiplication where the result fits in 16 bits */ +#define MULT16_16_16(a,b) ((((opus_val16)(a))*((opus_val16)(b)))) + +/* (opus_val32)(opus_val16) gives TI compiler a hint that it's 16x16->32 multiply */ +/** 16x16 multiplication where the result fits in 32 bits */ +#define MULT16_16(a,b) (((opus_val32)(opus_val16)(a))*((opus_val32)(opus_val16)(b))) + +/** 16x16 multiply-add where the result fits in 32 bits */ +#define MAC16_16(c,a,b) (ADD32((c),MULT16_16((a),(b)))) +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ +#define MAC16_32_Q15(c,a,b) ADD32((c),ADD32(MULT16_16((a),SHR((b),15)), SHR(MULT16_16((a),((b)&0x00007fff)),15))) + +/** 16x32 multiplication, followed by a 16-bit shift right and 32-bit add. + Results fits in 32 bits */ +#define MAC16_32_Q16(c,a,b) ADD32((c),ADD32(MULT16_16((a),SHR((b),16)), SHR(MULT16_16SU((a),((b)&0x0000ffff)),16))) + +#define MULT16_16_Q11_32(a,b) (SHR(MULT16_16((a),(b)),11)) +#define MULT16_16_Q11(a,b) (SHR(MULT16_16((a),(b)),11)) +#define MULT16_16_Q13(a,b) (SHR(MULT16_16((a),(b)),13)) +#define MULT16_16_Q14(a,b) (SHR(MULT16_16((a),(b)),14)) +#define MULT16_16_Q15(a,b) (SHR(MULT16_16((a),(b)),15)) + +#define MULT16_16_P13(a,b) (SHR(ADD32(4096,MULT16_16((a),(b))),13)) +#define MULT16_16_P14(a,b) (SHR(ADD32(8192,MULT16_16((a),(b))),14)) +#define MULT16_16_P15(a,b) (SHR(ADD32(16384,MULT16_16((a),(b))),15)) + +/** Divide a 32-bit value by a 16-bit value. Result fits in 16 bits */ +#define DIV32_16(a,b) ((opus_val16)(((opus_val32)(a))/((opus_val16)(b)))) + +/** Divide a 32-bit value by a 32-bit value. Result fits in 32 bits */ +#define DIV32(a,b) (((opus_val32)(a))/((opus_val32)(b))) + +#if defined(MIPSr1_ASM) +#include "mips/fixed_generic_mipsr1.h" +#endif + +static OPUS_INLINE opus_val16 SIG2WORD16_generic(celt_sig x) +{ + x = PSHR32(x, SIG_SHIFT); + x = MAX32(x, -32768); + x = MIN32(x, 32767); + return EXTRACT16(x); +} +#define SIG2WORD16(x) (SIG2WORD16_generic(x)) + +#endif diff --git a/libopus/celt/float_cast.h b/libopus/celt/float_cast.h new file mode 100644 index 0000000..d33ed11 --- /dev/null +++ b/libopus/celt/float_cast.h @@ -0,0 +1,146 @@ +/* Copyright (C) 2001 Erik de Castro Lopo <erikd AT mega-nerd DOT com> */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* Version 1.1 */ + +#ifndef FLOAT_CAST_H +#define FLOAT_CAST_H + + +#include "arch.h" + +/*============================================================================ +** On Intel Pentium processors (especially PIII and probably P4), converting +** from float to int is very slow. To meet the C specs, the code produced by +** most C compilers targeting Pentium needs to change the FPU rounding mode +** before the float to int conversion is performed. +** +** Changing the FPU rounding mode causes the FPU pipeline to be flushed. It +** is this flushing of the pipeline which is so slow. +** +** Fortunately the ISO C99 specifications define the functions lrint, lrintf, +** llrint and llrintf which fix this problem as a side effect. +** +** On Unix-like systems, the configure process should have detected the +** presence of these functions. If they weren't found we have to replace them +** here with a standard C cast. +*/ + +/* +** The C99 prototypes for lrint and lrintf are as follows: +** +** long int lrintf (float x) ; +** long int lrint (double x) ; +*/ + +/* The presence of the required functions are detected during the configure +** process and the values HAVE_LRINT and HAVE_LRINTF are set accordingly in +** the config.h file. +*/ + +/* With GCC, when SSE is available, the fastest conversion is cvtss2si. */ +#if defined(__GNUC__) && defined(__SSE__) + +#include <xmmintrin.h> +static OPUS_INLINE opus_int32 float2int(float x) {return _mm_cvt_ss2si(_mm_set_ss(x));} + +#elif defined(HAVE_LRINTF) + +/* These defines enable functionality introduced with the 1999 ISO C +** standard. They must be defined before the inclusion of math.h to +** engage them. If optimisation is enabled, these functions will be +** inlined. With optimisation switched off, you have to link in the +** maths library using -lm. +*/ + +#define _ISOC9X_SOURCE 1 +#define _ISOC99_SOURCE 1 + +#define __USE_ISOC9X 1 +#define __USE_ISOC99 1 + +#include <math.h> +#define float2int(x) lrintf(x) + +#elif (defined(HAVE_LRINT)) + +#define _ISOC9X_SOURCE 1 +#define _ISOC99_SOURCE 1 + +#define __USE_ISOC9X 1 +#define __USE_ISOC99 1 + +#include <math.h> +#define float2int(x) lrint(x) + +#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && (defined(_M_X64) || (defined(_M_IX86_FP) && _M_IX86_FP >= 1)) + #include <xmmintrin.h> + + static __inline long int float2int(float value) + { + return _mm_cvtss_si32(_mm_load_ss(&value)); + } +#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && defined (_M_IX86) + #include <math.h> + + /* Win32 doesn't seem to have these functions. + ** Therefore implement OPUS_INLINE versions of these functions here. + */ + + static __inline long int + float2int (float flt) + { int intgr; + + _asm + { fld flt + fistp intgr + } ; + + return intgr ; + } + +#else + +#if (defined(__GNUC__) && defined(__STDC__) && __STDC__ && __STDC_VERSION__ >= 199901L) + /* supported by gcc in C99 mode, but not by all other compilers */ + //#warning "Don't have the functions lrint() and lrintf ()." + //#warning "Replacing these functions with a standard C cast." +#endif /* __STDC_VERSION__ >= 199901L */ + #include <math.h> + #define float2int(flt) ((int)(floor(.5+flt))) +#endif + +#ifndef DISABLE_FLOAT_API +static OPUS_INLINE opus_int16 FLOAT2INT16(float x) +{ + x = x*CELT_SIG_SCALE; + x = MAX32(x, -32768); + x = MIN32(x, 32767); + return (opus_int16)float2int(x); +} +#endif /* DISABLE_FLOAT_API */ + +#endif /* FLOAT_CAST_H */ diff --git a/libopus/celt/kiss_fft.c b/libopus/celt/kiss_fft.c new file mode 100644 index 0000000..8377516 --- /dev/null +++ b/libopus/celt/kiss_fft.c @@ -0,0 +1,604 @@ +/*Copyright (c) 2003-2004, Mark Borgerding + Lots of modifications by Jean-Marc Valin + Copyright (c) 2005-2007, Xiph.Org Foundation + Copyright (c) 2008, Xiph.Org Foundation, CSIRO + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +/* This code is originally from Mark Borgerding's KISS-FFT but has been + heavily modified to better suit Opus */ + +#ifndef SKIP_CONFIG_H +# ifdef HAVE_CONFIG_H +# include "config.h" +# endif +#endif + +#include "_kiss_fft_guts.h" +#include "arch.h" +#include "os_support.h" +#include "mathops.h" +#include "stack_alloc.h" + +/* The guts header contains all the multiplication and addition macros that are defined for + complex numbers. It also delares the kf_ internal functions. +*/ + +static void kf_bfly2( + kiss_fft_cpx * Fout, + int m, + int N + ) +{ + kiss_fft_cpx * Fout2; + int i; + (void)m; +#ifdef CUSTOM_MODES + if (m==1) + { + celt_assert(m==1); + for (i=0;i<N;i++) + { + kiss_fft_cpx t; + Fout2 = Fout + 1; + t = *Fout2; + C_SUB( *Fout2 , *Fout , t ); + C_ADDTO( *Fout , t ); + Fout += 2; + } + } else +#endif + { + opus_val16 tw; + tw = QCONST16(0.7071067812f, 15); + /* We know that m==4 here because the radix-2 is just after a radix-4 */ + celt_assert(m==4); + for (i=0;i<N;i++) + { + kiss_fft_cpx t; + Fout2 = Fout + 4; + t = Fout2[0]; + C_SUB( Fout2[0] , Fout[0] , t ); + C_ADDTO( Fout[0] , t ); + + t.r = S_MUL(ADD32_ovflw(Fout2[1].r, Fout2[1].i), tw); + t.i = S_MUL(SUB32_ovflw(Fout2[1].i, Fout2[1].r), tw); + C_SUB( Fout2[1] , Fout[1] , t ); + C_ADDTO( Fout[1] , t ); + + t.r = Fout2[2].i; + t.i = -Fout2[2].r; + C_SUB( Fout2[2] , Fout[2] , t ); + C_ADDTO( Fout[2] , t ); + + t.r = S_MUL(SUB32_ovflw(Fout2[3].i, Fout2[3].r), tw); + t.i = S_MUL(NEG32_ovflw(ADD32_ovflw(Fout2[3].i, Fout2[3].r)), tw); + C_SUB( Fout2[3] , Fout[3] , t ); + C_ADDTO( Fout[3] , t ); + Fout += 8; + } + } +} + +static void kf_bfly4( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + int i; + + if (m==1) + { + /* Degenerate case where all the twiddles are 1. */ + for (i=0;i<N;i++) + { + kiss_fft_cpx scratch0, scratch1; + + C_SUB( scratch0 , *Fout, Fout[2] ); + C_ADDTO(*Fout, Fout[2]); + C_ADD( scratch1 , Fout[1] , Fout[3] ); + C_SUB( Fout[2], *Fout, scratch1 ); + C_ADDTO( *Fout , scratch1 ); + C_SUB( scratch1 , Fout[1] , Fout[3] ); + + Fout[1].r = ADD32_ovflw(scratch0.r, scratch1.i); + Fout[1].i = SUB32_ovflw(scratch0.i, scratch1.r); + Fout[3].r = SUB32_ovflw(scratch0.r, scratch1.i); + Fout[3].i = ADD32_ovflw(scratch0.i, scratch1.r); + Fout+=4; + } + } else { + int j; + kiss_fft_cpx scratch[6]; + const kiss_twiddle_cpx *tw1,*tw2,*tw3; + const int m2=2*m; + const int m3=3*m; + kiss_fft_cpx * Fout_beg = Fout; + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + tw3 = tw2 = tw1 = st->twiddles; + /* m is guaranteed to be a multiple of 4. */ + for (j=0;j<m;j++) + { + C_MUL(scratch[0],Fout[m] , *tw1 ); + C_MUL(scratch[1],Fout[m2] , *tw2 ); + C_MUL(scratch[2],Fout[m3] , *tw3 ); + + C_SUB( scratch[5] , *Fout, scratch[1] ); + C_ADDTO(*Fout, scratch[1]); + C_ADD( scratch[3] , scratch[0] , scratch[2] ); + C_SUB( scratch[4] , scratch[0] , scratch[2] ); + C_SUB( Fout[m2], *Fout, scratch[3] ); + tw1 += fstride; + tw2 += fstride*2; + tw3 += fstride*3; + C_ADDTO( *Fout , scratch[3] ); + + Fout[m].r = ADD32_ovflw(scratch[5].r, scratch[4].i); + Fout[m].i = SUB32_ovflw(scratch[5].i, scratch[4].r); + Fout[m3].r = SUB32_ovflw(scratch[5].r, scratch[4].i); + Fout[m3].i = ADD32_ovflw(scratch[5].i, scratch[4].r); + ++Fout; + } + } + } +} + + +#ifndef RADIX_TWO_ONLY + +static void kf_bfly3( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + int i; + size_t k; + const size_t m2 = 2*m; + const kiss_twiddle_cpx *tw1,*tw2; + kiss_fft_cpx scratch[5]; + kiss_twiddle_cpx epi3; + + kiss_fft_cpx * Fout_beg = Fout; +#ifdef FIXED_POINT + /*epi3.r = -16384;*/ /* Unused */ + epi3.i = -28378; +#else + epi3 = st->twiddles[fstride*m]; +#endif + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + tw1=tw2=st->twiddles; + /* For non-custom modes, m is guaranteed to be a multiple of 4. */ + k=m; + do { + + C_MUL(scratch[1],Fout[m] , *tw1); + C_MUL(scratch[2],Fout[m2] , *tw2); + + C_ADD(scratch[3],scratch[1],scratch[2]); + C_SUB(scratch[0],scratch[1],scratch[2]); + tw1 += fstride; + tw2 += fstride*2; + + Fout[m].r = SUB32_ovflw(Fout->r, HALF_OF(scratch[3].r)); + Fout[m].i = SUB32_ovflw(Fout->i, HALF_OF(scratch[3].i)); + + C_MULBYSCALAR( scratch[0] , epi3.i ); + + C_ADDTO(*Fout,scratch[3]); + + Fout[m2].r = ADD32_ovflw(Fout[m].r, scratch[0].i); + Fout[m2].i = SUB32_ovflw(Fout[m].i, scratch[0].r); + + Fout[m].r = SUB32_ovflw(Fout[m].r, scratch[0].i); + Fout[m].i = ADD32_ovflw(Fout[m].i, scratch[0].r); + + ++Fout; + } while(--k); + } +} + + +#ifndef OVERRIDE_kf_bfly5 +static void kf_bfly5( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4; + int i, u; + kiss_fft_cpx scratch[13]; + const kiss_twiddle_cpx *tw; + kiss_twiddle_cpx ya,yb; + kiss_fft_cpx * Fout_beg = Fout; + +#ifdef FIXED_POINT + ya.r = 10126; + ya.i = -31164; + yb.r = -26510; + yb.i = -19261; +#else + ya = st->twiddles[fstride*m]; + yb = st->twiddles[fstride*2*m]; +#endif + tw=st->twiddles; + + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + Fout0=Fout; + Fout1=Fout0+m; + Fout2=Fout0+2*m; + Fout3=Fout0+3*m; + Fout4=Fout0+4*m; + + /* For non-custom modes, m is guaranteed to be a multiple of 4. */ + for ( u=0; u<m; ++u ) { + scratch[0] = *Fout0; + + C_MUL(scratch[1] ,*Fout1, tw[u*fstride]); + C_MUL(scratch[2] ,*Fout2, tw[2*u*fstride]); + C_MUL(scratch[3] ,*Fout3, tw[3*u*fstride]); + C_MUL(scratch[4] ,*Fout4, tw[4*u*fstride]); + + C_ADD( scratch[7],scratch[1],scratch[4]); + C_SUB( scratch[10],scratch[1],scratch[4]); + C_ADD( scratch[8],scratch[2],scratch[3]); + C_SUB( scratch[9],scratch[2],scratch[3]); + + Fout0->r = ADD32_ovflw(Fout0->r, ADD32_ovflw(scratch[7].r, scratch[8].r)); + Fout0->i = ADD32_ovflw(Fout0->i, ADD32_ovflw(scratch[7].i, scratch[8].i)); + + scratch[5].r = ADD32_ovflw(scratch[0].r, ADD32_ovflw(S_MUL(scratch[7].r,ya.r), S_MUL(scratch[8].r,yb.r))); + scratch[5].i = ADD32_ovflw(scratch[0].i, ADD32_ovflw(S_MUL(scratch[7].i,ya.r), S_MUL(scratch[8].i,yb.r))); + + scratch[6].r = ADD32_ovflw(S_MUL(scratch[10].i,ya.i), S_MUL(scratch[9].i,yb.i)); + scratch[6].i = NEG32_ovflw(ADD32_ovflw(S_MUL(scratch[10].r,ya.i), S_MUL(scratch[9].r,yb.i))); + + C_SUB(*Fout1,scratch[5],scratch[6]); + C_ADD(*Fout4,scratch[5],scratch[6]); + + scratch[11].r = ADD32_ovflw(scratch[0].r, ADD32_ovflw(S_MUL(scratch[7].r,yb.r), S_MUL(scratch[8].r,ya.r))); + scratch[11].i = ADD32_ovflw(scratch[0].i, ADD32_ovflw(S_MUL(scratch[7].i,yb.r), S_MUL(scratch[8].i,ya.r))); + scratch[12].r = SUB32_ovflw(S_MUL(scratch[9].i,ya.i), S_MUL(scratch[10].i,yb.i)); + scratch[12].i = SUB32_ovflw(S_MUL(scratch[10].r,yb.i), S_MUL(scratch[9].r,ya.i)); + + C_ADD(*Fout2,scratch[11],scratch[12]); + C_SUB(*Fout3,scratch[11],scratch[12]); + + ++Fout0;++Fout1;++Fout2;++Fout3;++Fout4; + } + } +} +#endif /* OVERRIDE_kf_bfly5 */ + + +#endif + + +#ifdef CUSTOM_MODES + +static +void compute_bitrev_table( + int Fout, + opus_int16 *f, + const size_t fstride, + int in_stride, + opus_int16 * factors, + const kiss_fft_state *st + ) +{ + const int p=*factors++; /* the radix */ + const int m=*factors++; /* stage's fft length/p */ + + /*printf ("fft %d %d %d %d %d %d\n", p*m, m, p, s2, fstride*in_stride, N);*/ + if (m==1) + { + int j; + for (j=0;j<p;j++) + { + *f = Fout+j; + f += fstride*in_stride; + } + } else { + int j; + for (j=0;j<p;j++) + { + compute_bitrev_table( Fout , f, fstride*p, in_stride, factors,st); + f += fstride*in_stride; + Fout += m; + } + } +} + +/* facbuf is populated by p1,m1,p2,m2, ... + where + p[i] * m[i] = m[i-1] + m0 = n */ +static +int kf_factor(int n,opus_int16 * facbuf) +{ + int p=4; + int i; + int stages=0; + int nbak = n; + + /*factor out powers of 4, powers of 2, then any remaining primes */ + do { + while (n % p) { + switch (p) { + case 4: p = 2; break; + case 2: p = 3; break; + default: p += 2; break; + } + if (p>32000 || (opus_int32)p*(opus_int32)p > n) + p = n; /* no more factors, skip to end */ + } + n /= p; +#ifdef RADIX_TWO_ONLY + if (p!=2 && p != 4) +#else + if (p>5) +#endif + { + return 0; + } + facbuf[2*stages] = p; + if (p==2 && stages > 1) + { + facbuf[2*stages] = 4; + facbuf[2] = 2; + } + stages++; + } while (n > 1); + n = nbak; + /* Reverse the order to get the radix 4 at the end, so we can use the + fast degenerate case. It turns out that reversing the order also + improves the noise behaviour. */ + for (i=0;i<stages/2;i++) + { + int tmp; + tmp = facbuf[2*i]; + facbuf[2*i] = facbuf[2*(stages-i-1)]; + facbuf[2*(stages-i-1)] = tmp; + } + for (i=0;i<stages;i++) + { + n /= facbuf[2*i]; + facbuf[2*i+1] = n; + } + return 1; +} + +static void compute_twiddles(kiss_twiddle_cpx *twiddles, int nfft) +{ + int i; +#ifdef FIXED_POINT + for (i=0;i<nfft;++i) { + opus_val32 phase = -i; + kf_cexp2(twiddles+i, DIV32(SHL32(phase,17),nfft)); + } +#else + for (i=0;i<nfft;++i) { + const double pi=3.14159265358979323846264338327; + double phase = ( -2*pi /nfft ) * i; + kf_cexp(twiddles+i, phase ); + } +#endif +} + +int opus_fft_alloc_arch_c(kiss_fft_state *st) { + (void)st; + return 0; +} + +/* + * + * Allocates all necessary storage space for the fft and ifft. + * The return value is a contiguous block of memory. As such, + * It can be freed with free(). + * */ +kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, + const kiss_fft_state *base, int arch) +{ + kiss_fft_state *st=NULL; + size_t memneeded = sizeof(struct kiss_fft_state); /* twiddle factors*/ + + if ( lenmem==NULL ) { + st = ( kiss_fft_state*)KISS_FFT_MALLOC( memneeded ); + }else{ + if (mem != NULL && *lenmem >= memneeded) + st = (kiss_fft_state*)mem; + *lenmem = memneeded; + } + if (st) { + opus_int16 *bitrev; + kiss_twiddle_cpx *twiddles; + + st->nfft=nfft; +#ifdef FIXED_POINT + st->scale_shift = celt_ilog2(st->nfft); + if (st->nfft == 1<<st->scale_shift) + st->scale = Q15ONE; + else + st->scale = (1073741824+st->nfft/2)/st->nfft>>(15-st->scale_shift); +#else + st->scale = 1.f/nfft; +#endif + if (base != NULL) + { + st->twiddles = base->twiddles; + st->shift = 0; + while (st->shift < 32 && nfft<<st->shift != base->nfft) + st->shift++; + if (st->shift>=32) + goto fail; + } else { + st->twiddles = twiddles = (kiss_twiddle_cpx*)KISS_FFT_MALLOC(sizeof(kiss_twiddle_cpx)*nfft); + compute_twiddles(twiddles, nfft); + st->shift = -1; + } + if (!kf_factor(nfft,st->factors)) + { + goto fail; + } + + /* bitrev */ + st->bitrev = bitrev = (opus_int16*)KISS_FFT_MALLOC(sizeof(opus_int16)*nfft); + if (st->bitrev==NULL) + goto fail; + compute_bitrev_table(0, bitrev, 1,1, st->factors,st); + + /* Initialize architecture specific fft parameters */ + if (opus_fft_alloc_arch(st, arch)) + goto fail; + } + return st; +fail: + opus_fft_free(st, arch); + return NULL; +} + +kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem, int arch) +{ + return opus_fft_alloc_twiddles(nfft, mem, lenmem, NULL, arch); +} + +void opus_fft_free_arch_c(kiss_fft_state *st) { + (void)st; +} + +void opus_fft_free(const kiss_fft_state *cfg, int arch) +{ + if (cfg) + { + opus_fft_free_arch((kiss_fft_state *)cfg, arch); + opus_free((opus_int16*)cfg->bitrev); + if (cfg->shift < 0) + opus_free((kiss_twiddle_cpx*)cfg->twiddles); + opus_free((kiss_fft_state*)cfg); + } +} + +#endif /* CUSTOM_MODES */ + +void opus_fft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout) +{ + int m2, m; + int p; + int L; + int fstride[MAXFACTORS]; + int i; + int shift; + + /* st->shift can be -1 */ + shift = st->shift>0 ? st->shift : 0; + + fstride[0] = 1; + L=0; + do { + p = st->factors[2*L]; + m = st->factors[2*L+1]; + fstride[L+1] = fstride[L]*p; + L++; + } while(m!=1); + m = st->factors[2*L-1]; + for (i=L-1;i>=0;i--) + { + if (i!=0) + m2 = st->factors[2*i-1]; + else + m2 = 1; + switch (st->factors[2*i]) + { + case 2: + kf_bfly2(fout, m, fstride[i]); + break; + case 4: + kf_bfly4(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + #ifndef RADIX_TWO_ONLY + case 3: + kf_bfly3(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + case 5: + kf_bfly5(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + #endif + } + m = m2; + } +} + +void opus_fft_c(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout) +{ + int i; + opus_val16 scale; +#ifdef FIXED_POINT + /* Allows us to scale with MULT16_32_Q16(), which is faster than + MULT16_32_Q15() on ARM. */ + int scale_shift = st->scale_shift-1; +#endif + scale = st->scale; + + celt_assert2 (fin != fout, "In-place FFT not supported"); + /* Bit-reverse the input */ + for (i=0;i<st->nfft;i++) + { + kiss_fft_cpx x = fin[i]; + fout[st->bitrev[i]].r = SHR32(MULT16_32_Q16(scale, x.r), scale_shift); + fout[st->bitrev[i]].i = SHR32(MULT16_32_Q16(scale, x.i), scale_shift); + } + opus_fft_impl(st, fout); +} + + +void opus_ifft_c(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout) +{ + int i; + celt_assert2 (fin != fout, "In-place FFT not supported"); + /* Bit-reverse the input */ + for (i=0;i<st->nfft;i++) + fout[st->bitrev[i]] = fin[i]; + for (i=0;i<st->nfft;i++) + fout[i].i = -fout[i].i; + opus_fft_impl(st, fout); + for (i=0;i<st->nfft;i++) + fout[i].i = -fout[i].i; +} diff --git a/libopus/celt/kiss_fft.h b/libopus/celt/kiss_fft.h new file mode 100644 index 0000000..bffa2bf --- /dev/null +++ b/libopus/celt/kiss_fft.h @@ -0,0 +1,200 @@ +/*Copyright (c) 2003-2004, Mark Borgerding + Lots of modifications by Jean-Marc Valin + Copyright (c) 2005-2007, Xiph.Org Foundation + Copyright (c) 2008, Xiph.Org Foundation, CSIRO + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_H +#define KISS_FFT_H + +#include <stdlib.h> +#include <math.h> +#include "arch.h" +#include "cpu_support.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#ifdef USE_SIMD +# include <xmmintrin.h> +# define kiss_fft_scalar __m128 +#define KISS_FFT_MALLOC(nbytes) memalign(16,nbytes) +#else +#define KISS_FFT_MALLOC opus_alloc +#endif + +#ifdef FIXED_POINT +#include "arch.h" + +# define kiss_fft_scalar opus_int32 +# define kiss_twiddle_scalar opus_int16 + + +#else +# ifndef kiss_fft_scalar +/* default is float */ +# define kiss_fft_scalar float +# define kiss_twiddle_scalar float +# define KF_SUFFIX _celt_single +# endif +#endif + +typedef struct { + kiss_fft_scalar r; + kiss_fft_scalar i; +}kiss_fft_cpx; + +typedef struct { + kiss_twiddle_scalar r; + kiss_twiddle_scalar i; +}kiss_twiddle_cpx; + +#define MAXFACTORS 8 +/* e.g. an fft of length 128 has 4 factors + as far as kissfft is concerned + 4*4*4*2 + */ + +typedef struct arch_fft_state{ + int is_supported; + void *priv; +} arch_fft_state; + +typedef struct kiss_fft_state{ + int nfft; + opus_val16 scale; +#ifdef FIXED_POINT + int scale_shift; +#endif + int shift; + opus_int16 factors[2*MAXFACTORS]; + const opus_int16 *bitrev; + const kiss_twiddle_cpx *twiddles; + arch_fft_state *arch_fft; +} kiss_fft_state; + +#if defined(HAVE_ARM_NE10) +#include "arm/fft_arm.h" +#endif + +/*typedef struct kiss_fft_state* kiss_fft_cfg;*/ + +/** + * opus_fft_alloc + * + * Initialize a FFT (or IFFT) algorithm's cfg/state buffer. + * + * typical usage: kiss_fft_cfg mycfg=opus_fft_alloc(1024,0,NULL,NULL); + * + * The return value from fft_alloc is a cfg buffer used internally + * by the fft routine or NULL. + * + * If lenmem is NULL, then opus_fft_alloc will allocate a cfg buffer using malloc. + * The returned value should be free()d when done to avoid memory leaks. + * + * The state can be placed in a user supplied buffer 'mem': + * If lenmem is not NULL and mem is not NULL and *lenmem is large enough, + * then the function places the cfg in mem and the size used in *lenmem + * and returns mem. + * + * If lenmem is not NULL and ( mem is NULL or *lenmem is not large enough), + * then the function returns NULL and places the minimum cfg + * buffer size in *lenmem. + * */ + +kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, const kiss_fft_state *base, int arch); + +kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem, int arch); + +/** + * opus_fft(cfg,in_out_buf) + * + * Perform an FFT on a complex input buffer. + * for a forward FFT, + * fin should be f[0] , f[1] , ... ,f[nfft-1] + * fout will be F[0] , F[1] , ... ,F[nfft-1] + * Note that each element is complex and can be accessed like + f[k].r and f[k].i + * */ +void opus_fft_c(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout); +void opus_ifft_c(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout); + +void opus_fft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout); +void opus_ifft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout); + +void opus_fft_free(const kiss_fft_state *cfg, int arch); + + +void opus_fft_free_arch_c(kiss_fft_state *st); +int opus_fft_alloc_arch_c(kiss_fft_state *st); + +#if !defined(OVERRIDE_OPUS_FFT) +/* Is run-time CPU detection enabled on this platform? */ +#if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) + +extern int (*const OPUS_FFT_ALLOC_ARCH_IMPL[OPUS_ARCHMASK+1])( + kiss_fft_state *st); + +#define opus_fft_alloc_arch(_st, arch) \ + ((*OPUS_FFT_ALLOC_ARCH_IMPL[(arch)&OPUS_ARCHMASK])(_st)) + +extern void (*const OPUS_FFT_FREE_ARCH_IMPL[OPUS_ARCHMASK+1])( + kiss_fft_state *st); +#define opus_fft_free_arch(_st, arch) \ + ((*OPUS_FFT_FREE_ARCH_IMPL[(arch)&OPUS_ARCHMASK])(_st)) + +extern void (*const OPUS_FFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg, + const kiss_fft_cpx *fin, kiss_fft_cpx *fout); +#define opus_fft(_cfg, _fin, _fout, arch) \ + ((*OPUS_FFT[(arch)&OPUS_ARCHMASK])(_cfg, _fin, _fout)) + +extern void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg, + const kiss_fft_cpx *fin, kiss_fft_cpx *fout); +#define opus_ifft(_cfg, _fin, _fout, arch) \ + ((*OPUS_IFFT[(arch)&OPUS_ARCHMASK])(_cfg, _fin, _fout)) + +#else /* else for if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) */ + +#define opus_fft_alloc_arch(_st, arch) \ + ((void)(arch), opus_fft_alloc_arch_c(_st)) + +#define opus_fft_free_arch(_st, arch) \ + ((void)(arch), opus_fft_free_arch_c(_st)) + +#define opus_fft(_cfg, _fin, _fout, arch) \ + ((void)(arch), opus_fft_c(_cfg, _fin, _fout)) + +#define opus_ifft(_cfg, _fin, _fout, arch) \ + ((void)(arch), opus_ifft_c(_cfg, _fin, _fout)) + +#endif /* end if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) */ +#endif /* end if !defined(OVERRIDE_OPUS_FFT) */ + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/libopus/celt/laplace.c b/libopus/celt/laplace.c new file mode 100644 index 0000000..a7bca87 --- /dev/null +++ b/libopus/celt/laplace.c @@ -0,0 +1,134 @@ +/* Copyright (c) 2007 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "laplace.h" +#include "mathops.h" + +/* The minimum probability of an energy delta (out of 32768). */ +#define LAPLACE_LOG_MINP (0) +#define LAPLACE_MINP (1<<LAPLACE_LOG_MINP) +/* The minimum number of guaranteed representable energy deltas (in one + direction). */ +#define LAPLACE_NMIN (16) + +/* When called, decay is positive and at most 11456. */ +static unsigned ec_laplace_get_freq1(unsigned fs0, int decay) +{ + unsigned ft; + ft = 32768 - LAPLACE_MINP*(2*LAPLACE_NMIN) - fs0; + return ft*(opus_int32)(16384-decay)>>15; +} + +void ec_laplace_encode(ec_enc *enc, int *value, unsigned fs, int decay) +{ + unsigned fl; + int val = *value; + fl = 0; + if (val) + { + int s; + int i; + s = -(val<0); + val = (val+s)^s; + fl = fs; + fs = ec_laplace_get_freq1(fs, decay); + /* Search the decaying part of the PDF.*/ + for (i=1; fs > 0 && i < val; i++) + { + fs *= 2; + fl += fs+2*LAPLACE_MINP; + fs = (fs*(opus_int32)decay)>>15; + } + /* Everything beyond that has probability LAPLACE_MINP. */ + if (!fs) + { + int di; + int ndi_max; + ndi_max = (32768-fl+LAPLACE_MINP-1)>>LAPLACE_LOG_MINP; + ndi_max = (ndi_max-s)>>1; + di = IMIN(val - i, ndi_max - 1); + fl += (2*di+1+s)*LAPLACE_MINP; + fs = IMIN(LAPLACE_MINP, 32768-fl); + *value = (i+di+s)^s; + } + else + { + fs += LAPLACE_MINP; + fl += fs&~s; + } + celt_assert(fl+fs<=32768); + celt_assert(fs>0); + } + ec_encode_bin(enc, fl, fl+fs, 15); +} + +int ec_laplace_decode(ec_dec *dec, unsigned fs, int decay) +{ + int val=0; + unsigned fl; + unsigned fm; + fm = ec_decode_bin(dec, 15); + fl = 0; + if (fm >= fs) + { + val++; + fl = fs; + fs = ec_laplace_get_freq1(fs, decay)+LAPLACE_MINP; + /* Search the decaying part of the PDF.*/ + while(fs > LAPLACE_MINP && fm >= fl+2*fs) + { + fs *= 2; + fl += fs; + fs = ((fs-2*LAPLACE_MINP)*(opus_int32)decay)>>15; + fs += LAPLACE_MINP; + val++; + } + /* Everything beyond that has probability LAPLACE_MINP. */ + if (fs <= LAPLACE_MINP) + { + int di; + di = (fm-fl)>>(LAPLACE_LOG_MINP+1); + val += di; + fl += 2*di*LAPLACE_MINP; + } + if (fm < fl+fs) + val = -val; + else + fl += fs; + } + celt_assert(fl<32768); + celt_assert(fs>0); + celt_assert(fl<=fm); + celt_assert(fm<IMIN(fl+fs,32768)); + ec_dec_update(dec, fl, IMIN(fl+fs,32768), 32768); + return val; +} diff --git a/libopus/celt/laplace.h b/libopus/celt/laplace.h new file mode 100644 index 0000000..46c14b5 --- /dev/null +++ b/libopus/celt/laplace.h @@ -0,0 +1,48 @@ +/* Copyright (c) 2007 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include "entenc.h" +#include "entdec.h" + +/** Encode a value that is assumed to be the realisation of a + Laplace-distributed random process + @param enc Entropy encoder state + @param value Value to encode + @param fs Probability of 0, multiplied by 32768 + @param decay Probability of the value +/- 1, multiplied by 16384 +*/ +void ec_laplace_encode(ec_enc *enc, int *value, unsigned fs, int decay); + +/** Decode a value that is assumed to be the realisation of a + Laplace-distributed random process + @param dec Entropy decoder state + @param fs Probability of 0, multiplied by 32768 + @param decay Probability of the value +/- 1, multiplied by 16384 + @return Value decoded + */ +int ec_laplace_decode(ec_dec *dec, unsigned fs, int decay); diff --git a/libopus/celt/mathops.c b/libopus/celt/mathops.c new file mode 100644 index 0000000..6ee9b9e --- /dev/null +++ b/libopus/celt/mathops.c @@ -0,0 +1,209 @@ +/* Copyright (c) 2002-2008 Jean-Marc Valin + Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file mathops.h + @brief Various math functions +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mathops.h" + +/*Compute floor(sqrt(_val)) with exact arithmetic. + _val must be greater than 0. + This has been tested on all possible 32-bit inputs greater than 0.*/ +unsigned isqrt32(opus_uint32 _val){ + unsigned b; + unsigned g; + int bshift; + /*Uses the second method from + http://www.azillionmonkeys.com/qed/sqroot.html + The main idea is to search for the largest binary digit b such that + (g+b)*(g+b) <= _val, and add it to the solution g.*/ + g=0; + bshift=(EC_ILOG(_val)-1)>>1; + b=1U<<bshift; + do{ + opus_uint32 t; + t=(((opus_uint32)g<<1)+b)<<bshift; + if(t<=_val){ + g+=b; + _val-=t; + } + b>>=1; + bshift--; + } + while(bshift>=0); + return g; +} + +#ifdef FIXED_POINT + +opus_val32 frac_div32(opus_val32 a, opus_val32 b) +{ + opus_val16 rcp; + opus_val32 result, rem; + int shift = celt_ilog2(b)-29; + a = VSHR32(a,shift); + b = VSHR32(b,shift); + /* 16-bit reciprocal */ + rcp = ROUND16(celt_rcp(ROUND16(b,16)),3); + result = MULT16_32_Q15(rcp, a); + rem = PSHR32(a,2)-MULT32_32_Q31(result, b); + result = ADD32(result, SHL32(MULT16_32_Q15(rcp, rem),2)); + if (result >= 536870912) /* 2^29 */ + return 2147483647; /* 2^31 - 1 */ + else if (result <= -536870912) /* -2^29 */ + return -2147483647; /* -2^31 */ + else + return SHL32(result, 2); +} + +/** Reciprocal sqrt approximation in the range [0.25,1) (Q16 in, Q14 out) */ +opus_val16 celt_rsqrt_norm(opus_val32 x) +{ + opus_val16 n; + opus_val16 r; + opus_val16 r2; + opus_val16 y; + /* Range of n is [-16384,32767] ([-0.5,1) in Q15). */ + n = x-32768; + /* Get a rough initial guess for the root. + The optimal minimax quadratic approximation (using relative error) is + r = 1.437799046117536+n*(-0.823394375837328+n*0.4096419668459485). + Coefficients here, and the final result r, are Q14.*/ + r = ADD16(23557, MULT16_16_Q15(n, ADD16(-13490, MULT16_16_Q15(n, 6713)))); + /* We want y = x*r*r-1 in Q15, but x is 32-bit Q16 and r is Q14. + We can compute the result from n and r using Q15 multiplies with some + adjustment, carefully done to avoid overflow. + Range of y is [-1564,1594]. */ + r2 = MULT16_16_Q15(r, r); + y = SHL16(SUB16(ADD16(MULT16_16_Q15(r2, n), r2), 16384), 1); + /* Apply a 2nd-order Householder iteration: r += r*y*(y*0.375-0.5). + This yields the Q14 reciprocal square root of the Q16 x, with a maximum + relative error of 1.04956E-4, a (relative) RMSE of 2.80979E-5, and a + peak absolute error of 2.26591/16384. */ + return ADD16(r, MULT16_16_Q15(r, MULT16_16_Q15(y, + SUB16(MULT16_16_Q15(y, 12288), 16384)))); +} + +/** Sqrt approximation (QX input, QX/2 output) */ +opus_val32 celt_sqrt(opus_val32 x) +{ + int k; + opus_val16 n; + opus_val32 rt; + static const opus_val16 C[5] = {23175, 11561, -3011, 1699, -664}; + if (x==0) + return 0; + else if (x>=1073741824) + return 32767; + k = (celt_ilog2(x)>>1)-7; + x = VSHR32(x, 2*k); + n = x-32768; + rt = ADD16(C[0], MULT16_16_Q15(n, ADD16(C[1], MULT16_16_Q15(n, ADD16(C[2], + MULT16_16_Q15(n, ADD16(C[3], MULT16_16_Q15(n, (C[4]))))))))); + rt = VSHR32(rt,7-k); + return rt; +} + +#define L1 32767 +#define L2 -7651 +#define L3 8277 +#define L4 -626 + +static OPUS_INLINE opus_val16 _celt_cos_pi_2(opus_val16 x) +{ + opus_val16 x2; + + x2 = MULT16_16_P15(x,x); + return ADD16(1,MIN16(32766,ADD32(SUB16(L1,x2), MULT16_16_P15(x2, ADD32(L2, MULT16_16_P15(x2, ADD32(L3, MULT16_16_P15(L4, x2 + )))))))); +} + +#undef L1 +#undef L2 +#undef L3 +#undef L4 + +opus_val16 celt_cos_norm(opus_val32 x) +{ + x = x&0x0001ffff; + if (x>SHL32(EXTEND32(1), 16)) + x = SUB32(SHL32(EXTEND32(1), 17),x); + if (x&0x00007fff) + { + if (x<SHL32(EXTEND32(1), 15)) + { + return _celt_cos_pi_2(EXTRACT16(x)); + } else { + return NEG16(_celt_cos_pi_2(EXTRACT16(65536-x))); + } + } else { + if (x&0x0000ffff) + return 0; + else if (x&0x0001ffff) + return -32767; + else + return 32767; + } +} + +/** Reciprocal approximation (Q15 input, Q16 output) */ +opus_val32 celt_rcp(opus_val32 x) +{ + int i; + opus_val16 n; + opus_val16 r; + celt_sig_assert(x>0); + i = celt_ilog2(x); + /* n is Q15 with range [0,1). */ + n = VSHR32(x,i-15)-32768; + /* Start with a linear approximation: + r = 1.8823529411764706-0.9411764705882353*n. + The coefficients and the result are Q14 in the range [15420,30840].*/ + r = ADD16(30840, MULT16_16_Q15(-15420, n)); + /* Perform two Newton iterations: + r -= r*((r*n)-1.Q15) + = r*((r*n)+(r-1.Q15)). */ + r = SUB16(r, MULT16_16_Q15(r, + ADD16(MULT16_16_Q15(r, n), ADD16(r, -32768)))); + /* We subtract an extra 1 in the second iteration to avoid overflow; it also + neatly compensates for truncation error in the rest of the process. */ + r = SUB16(r, ADD16(1, MULT16_16_Q15(r, + ADD16(MULT16_16_Q15(r, n), ADD16(r, -32768))))); + /* r is now the Q15 solution to 2/(n+1), with a maximum relative error + of 7.05346E-5, a (relative) RMSE of 2.14418E-5, and a peak absolute + error of 1.24665/32768. */ + return VSHR32(EXTEND32(r),i-16); +} + +#endif diff --git a/libopus/celt/mathops.h b/libopus/celt/mathops.h new file mode 100644 index 0000000..5e86ff0 --- /dev/null +++ b/libopus/celt/mathops.h @@ -0,0 +1,290 @@ +/* Copyright (c) 2002-2008 Jean-Marc Valin + Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file mathops.h + @brief Various math functions +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef MATHOPS_H +#define MATHOPS_H + +#include "arch.h" +#include "entcode.h" +#include "os_support.h" + +#define PI 3.141592653f + +/* Multiplies two 16-bit fractional values. Bit-exactness of this macro is important */ +#define FRAC_MUL16(a,b) ((16384+((opus_int32)(opus_int16)(a)*(opus_int16)(b)))>>15) + +unsigned isqrt32(opus_uint32 _val); + +/* CELT doesn't need it for fixed-point, by analysis.c does. */ +#if !defined(FIXED_POINT) || defined(ANALYSIS_C) +#define cA 0.43157974f +#define cB 0.67848403f +#define cC 0.08595542f +#define cE ((float)PI/2) +static OPUS_INLINE float fast_atan2f(float y, float x) { + float x2, y2; + x2 = x*x; + y2 = y*y; + /* For very small values, we don't care about the answer, so + we can just return 0. */ + if (x2 + y2 < 1e-18f) + { + return 0; + } + if(x2<y2){ + float den = (y2 + cB*x2) * (y2 + cC*x2); + return -x*y*(y2 + cA*x2) / den + (y<0 ? -cE : cE); + }else{ + float den = (x2 + cB*y2) * (x2 + cC*y2); + return x*y*(x2 + cA*y2) / den + (y<0 ? -cE : cE) - (x*y<0 ? -cE : cE); + } +} +#undef cA +#undef cB +#undef cC +#undef cE +#endif + + +#ifndef OVERRIDE_CELT_MAXABS16 +static OPUS_INLINE opus_val32 celt_maxabs16(const opus_val16 *x, int len) +{ + int i; + opus_val16 maxval = 0; + opus_val16 minval = 0; + for (i=0;i<len;i++) + { + maxval = MAX16(maxval, x[i]); + minval = MIN16(minval, x[i]); + } + return MAX32(EXTEND32(maxval),-EXTEND32(minval)); +} +#endif + +#ifndef OVERRIDE_CELT_MAXABS32 +#ifdef FIXED_POINT +static OPUS_INLINE opus_val32 celt_maxabs32(const opus_val32 *x, int len) +{ + int i; + opus_val32 maxval = 0; + opus_val32 minval = 0; + for (i=0;i<len;i++) + { + maxval = MAX32(maxval, x[i]); + minval = MIN32(minval, x[i]); + } + return MAX32(maxval, -minval); +} +#else +#define celt_maxabs32(x,len) celt_maxabs16(x,len) +#endif +#endif + + +#ifndef FIXED_POINT + +#define celt_sqrt(x) ((float)sqrt(x)) +#define celt_rsqrt(x) (1.f/celt_sqrt(x)) +#define celt_rsqrt_norm(x) (celt_rsqrt(x)) +#define celt_cos_norm(x) ((float)cos((.5f*PI)*(x))) +#define celt_rcp(x) (1.f/(x)) +#define celt_div(a,b) ((a)/(b)) +#define frac_div32(a,b) ((float)(a)/(b)) + +#ifdef FLOAT_APPROX + +/* Note: This assumes radix-2 floating point with the exponent at bits 23..30 and an offset of 127 + denorm, +/- inf and NaN are *not* handled */ + +/** Base-2 log approximation (log2(x)). */ +static OPUS_INLINE float celt_log2(float x) +{ + int integer; + float frac; + union { + float f; + opus_uint32 i; + } in; + in.f = x; + integer = (in.i>>23)-127; + in.i -= integer<<23; + frac = in.f - 1.5f; + frac = -0.41445418f + frac*(0.95909232f + + frac*(-0.33951290f + frac*0.16541097f)); + return 1+integer+frac; +} + +/** Base-2 exponential approximation (2^x). */ +static OPUS_INLINE float celt_exp2(float x) +{ + int integer; + float frac; + union { + float f; + opus_uint32 i; + } res; + integer = floor(x); + if (integer < -50) + return 0; + frac = x-integer; + /* K0 = 1, K1 = log(2), K2 = 3-4*log(2), K3 = 3*log(2) - 2 */ + res.f = 0.99992522f + frac * (0.69583354f + + frac * (0.22606716f + 0.078024523f*frac)); + res.i = (res.i + (integer<<23)) & 0x7fffffff; + return res.f; +} + +#else +#define celt_log2(x) ((float)(1.442695040888963387*log(x))) +#define celt_exp2(x) ((float)exp(0.6931471805599453094*(x))) +#endif + +#endif + +#ifdef FIXED_POINT + +#include "os_support.h" + +#ifndef OVERRIDE_CELT_ILOG2 +/** Integer log in base2. Undefined for zero and negative numbers */ +static OPUS_INLINE opus_int16 celt_ilog2(opus_int32 x) +{ + celt_sig_assert(x>0); + return EC_ILOG(x)-1; +} +#endif + + +/** Integer log in base2. Defined for zero, but not for negative numbers */ +static OPUS_INLINE opus_int16 celt_zlog2(opus_val32 x) +{ + return x <= 0 ? 0 : celt_ilog2(x); +} + +opus_val16 celt_rsqrt_norm(opus_val32 x); + +opus_val32 celt_sqrt(opus_val32 x); + +opus_val16 celt_cos_norm(opus_val32 x); + +/** Base-2 logarithm approximation (log2(x)). (Q14 input, Q10 output) */ +static OPUS_INLINE opus_val16 celt_log2(opus_val32 x) +{ + int i; + opus_val16 n, frac; + /* -0.41509302963303146, 0.9609890551383969, -0.31836011537636605, + 0.15530808010959576, -0.08556153059057618 */ + static const opus_val16 C[5] = {-6801+(1<<(13-DB_SHIFT)), 15746, -5217, 2545, -1401}; + if (x==0) + return -32767; + i = celt_ilog2(x); + n = VSHR32(x,i-15)-32768-16384; + frac = ADD16(C[0], MULT16_16_Q15(n, ADD16(C[1], MULT16_16_Q15(n, ADD16(C[2], MULT16_16_Q15(n, ADD16(C[3], MULT16_16_Q15(n, C[4])))))))); + return SHL16(i-13,DB_SHIFT)+SHR16(frac,14-DB_SHIFT); +} + +/* + K0 = 1 + K1 = log(2) + K2 = 3-4*log(2) + K3 = 3*log(2) - 2 +*/ +#define D0 16383 +#define D1 22804 +#define D2 14819 +#define D3 10204 + +static OPUS_INLINE opus_val32 celt_exp2_frac(opus_val16 x) +{ + opus_val16 frac; + frac = SHL16(x, 4); + return ADD16(D0, MULT16_16_Q15(frac, ADD16(D1, MULT16_16_Q15(frac, ADD16(D2 , MULT16_16_Q15(D3,frac)))))); +} +/** Base-2 exponential approximation (2^x). (Q10 input, Q16 output) */ +static OPUS_INLINE opus_val32 celt_exp2(opus_val16 x) +{ + int integer; + opus_val16 frac; + integer = SHR16(x,10); + if (integer>14) + return 0x7f000000; + else if (integer < -15) + return 0; + frac = celt_exp2_frac(x-SHL16(integer,10)); + return VSHR32(EXTEND32(frac), -integer-2); +} + +opus_val32 celt_rcp(opus_val32 x); + +#define celt_div(a,b) MULT32_32_Q31((opus_val32)(a),celt_rcp(b)) + +opus_val32 frac_div32(opus_val32 a, opus_val32 b); + +#define M1 32767 +#define M2 -21 +#define M3 -11943 +#define M4 4936 + +/* Atan approximation using a 4th order polynomial. Input is in Q15 format + and normalized by pi/4. Output is in Q15 format */ +static OPUS_INLINE opus_val16 celt_atan01(opus_val16 x) +{ + return MULT16_16_P15(x, ADD32(M1, MULT16_16_P15(x, ADD32(M2, MULT16_16_P15(x, ADD32(M3, MULT16_16_P15(M4, x))))))); +} + +#undef M1 +#undef M2 +#undef M3 +#undef M4 + +/* atan2() approximation valid for positive input values */ +static OPUS_INLINE opus_val16 celt_atan2p(opus_val16 y, opus_val16 x) +{ + if (y < x) + { + opus_val32 arg; + arg = celt_div(SHL32(EXTEND32(y),15),x); + if (arg >= 32767) + arg = 32767; + return SHR16(celt_atan01(EXTRACT16(arg)),1); + } else { + opus_val32 arg; + arg = celt_div(SHL32(EXTEND32(x),15),y); + if (arg >= 32767) + arg = 32767; + return 25736-SHR16(celt_atan01(EXTRACT16(arg)),1); + } +} + +#endif /* FIXED_POINT */ +#endif /* MATHOPS_H */ diff --git a/libopus/celt/mdct.c b/libopus/celt/mdct.c new file mode 100644 index 0000000..5c6dab5 --- /dev/null +++ b/libopus/celt/mdct.c @@ -0,0 +1,343 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2008 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* This is a simple MDCT implementation that uses a N/4 complex FFT + to do most of the work. It should be relatively straightforward to + plug in pretty much and FFT here. + + This replaces the Vorbis FFT (and uses the exact same API), which + was a bit too messy and that was ending up duplicating code + (might as well use the same FFT everywhere). + + The algorithm is similar to (and inspired from) Fabrice Bellard's + MDCT implementation in FFMPEG, but has differences in signs, ordering + and scaling in many places. +*/ + +#ifndef SKIP_CONFIG_H +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#endif + +#include "mdct.h" +#include "kiss_fft.h" +#include "_kiss_fft_guts.h" +#include <math.h> +#include "os_support.h" +#include "mathops.h" +#include "stack_alloc.h" + +#if defined(MIPSr1_ASM) +#include "mips/mdct_mipsr1.h" +#endif + + +#ifdef CUSTOM_MODES + +int clt_mdct_init(mdct_lookup *l,int N, int maxshift, int arch) +{ + int i; + kiss_twiddle_scalar *trig; + int shift; + int N2=N>>1; + l->n = N; + l->maxshift = maxshift; + for (i=0;i<=maxshift;i++) + { + if (i==0) + l->kfft[i] = opus_fft_alloc(N>>2>>i, 0, 0, arch); + else + l->kfft[i] = opus_fft_alloc_twiddles(N>>2>>i, 0, 0, l->kfft[0], arch); +#ifndef ENABLE_TI_DSPLIB55 + if (l->kfft[i]==NULL) + return 0; +#endif + } + l->trig = trig = (kiss_twiddle_scalar*)opus_alloc((N-(N2>>maxshift))*sizeof(kiss_twiddle_scalar)); + if (l->trig==NULL) + return 0; + for (shift=0;shift<=maxshift;shift++) + { + /* We have enough points that sine isn't necessary */ +#if defined(FIXED_POINT) +#if 1 + for (i=0;i<N2;i++) + trig[i] = TRIG_UPSCALE*celt_cos_norm(DIV32(ADD32(SHL32(EXTEND32(i),17),N2+16384),N)); +#else + for (i=0;i<N2;i++) + trig[i] = (kiss_twiddle_scalar)MAX32(-32767,MIN32(32767,floor(.5+32768*cos(2*M_PI*(i+.125)/N)))); +#endif +#else + for (i=0;i<N2;i++) + trig[i] = (kiss_twiddle_scalar)cos(2*PI*(i+.125)/N); +#endif + trig += N2; + N2 >>= 1; + N >>= 1; + } + return 1; +} + +void clt_mdct_clear(mdct_lookup *l, int arch) +{ + int i; + for (i=0;i<=l->maxshift;i++) + opus_fft_free(l->kfft[i], arch); + opus_free((kiss_twiddle_scalar*)l->trig); +} + +#endif /* CUSTOM_MODES */ + +/* Forward MDCT trashes the input array */ +#ifndef OVERRIDE_clt_mdct_forward +void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + VARDECL(kiss_fft_scalar, f); + VARDECL(kiss_fft_cpx, f2); + const kiss_fft_state *st = l->kfft[shift]; + const kiss_twiddle_scalar *trig; + opus_val16 scale; +#ifdef FIXED_POINT + /* Allows us to scale with MULT16_32_Q16(), which is faster than + MULT16_32_Q15() on ARM. */ + int scale_shift = st->scale_shift-1; +#endif + SAVE_STACK; + (void)arch; + scale = st->scale; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + ALLOC(f, N2, kiss_fft_scalar); + ALLOC(f2, N4, kiss_fft_cpx); + + /* Consider the input to be composed of four blocks: [a, b, c, d] */ + /* Window, shuffle, fold */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1); + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1); + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1); + const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1; + for(i=0;i<((overlap+3)>>2);i++) + { + /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/ + *yp++ = MULT16_32_Q15(*wp2, xp1[N2]) + MULT16_32_Q15(*wp1,*xp2); + *yp++ = MULT16_32_Q15(*wp1, *xp1) - MULT16_32_Q15(*wp2, xp2[-N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + wp1 = window; + wp2 = window+overlap-1; + for(;i<N4-((overlap+3)>>2);i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = *xp2; + *yp++ = *xp1; + xp1+=2; + xp2-=2; + } + for(;i<N4;i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = -MULT16_32_Q15(*wp1, xp1[-N2]) + MULT16_32_Q15(*wp2, *xp2); + *yp++ = MULT16_32_Q15(*wp2, *xp1) + MULT16_32_Q15(*wp1, xp2[N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + } + /* Pre-rotation */ + { + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const kiss_twiddle_scalar *t = &trig[0]; + for(i=0;i<N4;i++) + { + kiss_fft_cpx yc; + kiss_twiddle_scalar t0, t1; + kiss_fft_scalar re, im, yr, yi; + t0 = t[i]; + t1 = t[N4+i]; + re = *yp++; + im = *yp++; + yr = S_MUL(re,t0) - S_MUL(im,t1); + yi = S_MUL(im,t0) + S_MUL(re,t1); + yc.r = yr; + yc.i = yi; + yc.r = PSHR32(MULT16_32_Q16(scale, yc.r), scale_shift); + yc.i = PSHR32(MULT16_32_Q16(scale, yc.i), scale_shift); + f2[st->bitrev[i]] = yc; + } + } + + /* N/4 complex FFT, does not downscale anymore */ + opus_fft_impl(st, f2); + + /* Post-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_cpx * OPUS_RESTRICT fp = f2; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1); + const kiss_twiddle_scalar *t = &trig[0]; + /* Temp pointers to make it really clear to the compiler what we're doing */ + for(i=0;i<N4;i++) + { + kiss_fft_scalar yr, yi; + yr = S_MUL(fp->i,t[N4+i]) - S_MUL(fp->r,t[i]); + yi = S_MUL(fp->r,t[N4+i]) + S_MUL(fp->i,t[i]); + *yp1 = yr; + *yp2 = yi; + fp++; + yp1 += 2*stride; + yp2 -= 2*stride; + } + } + RESTORE_STACK; +} +#endif /* OVERRIDE_clt_mdct_forward */ + +#ifndef OVERRIDE_clt_mdct_backward +void clt_mdct_backward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + const kiss_twiddle_scalar *trig; + (void) arch; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + /* Pre-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in; + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1); + kiss_fft_scalar * OPUS_RESTRICT yp = out+(overlap>>1); + const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0]; + const opus_int16 * OPUS_RESTRICT bitrev = l->kfft[shift]->bitrev; + for(i=0;i<N4;i++) + { + int rev; + kiss_fft_scalar yr, yi; + rev = *bitrev++; + yr = ADD32_ovflw(S_MUL(*xp2, t[i]), S_MUL(*xp1, t[N4+i])); + yi = SUB32_ovflw(S_MUL(*xp1, t[i]), S_MUL(*xp2, t[N4+i])); + /* We swap real and imag because we use an FFT instead of an IFFT. */ + yp[2*rev+1] = yr; + yp[2*rev] = yi; + /* Storing the pre-rotation directly in the bitrev order. */ + xp1+=2*stride; + xp2-=2*stride; + } + } + + opus_fft_impl(l->kfft[shift], (kiss_fft_cpx*)(out+(overlap>>1))); + + /* Post-rotate and de-shuffle from both ends of the buffer at once to make + it in-place. */ + { + kiss_fft_scalar * yp0 = out+(overlap>>1); + kiss_fft_scalar * yp1 = out+(overlap>>1)+N2-2; + const kiss_twiddle_scalar *t = &trig[0]; + /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the + middle pair will be computed twice. */ + for(i=0;i<(N4+1)>>1;i++) + { + kiss_fft_scalar re, im, yr, yi; + kiss_twiddle_scalar t0, t1; + /* We swap real and imag because we're using an FFT instead of an IFFT. */ + re = yp0[1]; + im = yp0[0]; + t0 = t[i]; + t1 = t[N4+i]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = ADD32_ovflw(S_MUL(re,t0), S_MUL(im,t1)); + yi = SUB32_ovflw(S_MUL(re,t1), S_MUL(im,t0)); + /* We swap real and imag because we're using an FFT instead of an IFFT. */ + re = yp1[1]; + im = yp1[0]; + yp0[0] = yr; + yp1[1] = yi; + + t0 = t[(N4-i-1)]; + t1 = t[(N2-i-1)]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = ADD32_ovflw(S_MUL(re,t0), S_MUL(im,t1)); + yi = SUB32_ovflw(S_MUL(re,t1), S_MUL(im,t0)); + yp1[0] = yr; + yp0[1] = yi; + yp0 += 2; + yp1 -= 2; + } + } + + /* Mirror on both sides for TDAC */ + { + kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + const opus_val16 * OPUS_RESTRICT wp1 = window; + const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1; + + for(i = 0; i < overlap/2; i++) + { + kiss_fft_scalar x1, x2; + x1 = *xp1; + x2 = *yp1; + *yp1++ = SUB32_ovflw(MULT16_32_Q15(*wp2, x2), MULT16_32_Q15(*wp1, x1)); + *xp1-- = ADD32_ovflw(MULT16_32_Q15(*wp1, x2), MULT16_32_Q15(*wp2, x1)); + wp1++; + wp2--; + } + } +} +#endif /* OVERRIDE_clt_mdct_backward */ diff --git a/libopus/celt/mdct.h b/libopus/celt/mdct.h new file mode 100644 index 0000000..160ae4e --- /dev/null +++ b/libopus/celt/mdct.h @@ -0,0 +1,112 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2008 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* This is a simple MDCT implementation that uses a N/4 complex FFT + to do most of the work. It should be relatively straightforward to + plug in pretty much and FFT here. + + This replaces the Vorbis FFT (and uses the exact same API), which + was a bit too messy and that was ending up duplicating code + (might as well use the same FFT everywhere). + + The algorithm is similar to (and inspired from) Fabrice Bellard's + MDCT implementation in FFMPEG, but has differences in signs, ordering + and scaling in many places. +*/ + +#ifndef MDCT_H +#define MDCT_H + +#include "opus_defines.h" +#include "kiss_fft.h" +#include "arch.h" + +typedef struct { + int n; + int maxshift; + const kiss_fft_state *kfft[4]; + const kiss_twiddle_scalar * OPUS_RESTRICT trig; +} mdct_lookup; + +#if defined(HAVE_ARM_NE10) +#include "arm/mdct_arm.h" +#endif + + +int clt_mdct_init(mdct_lookup *l,int N, int maxshift, int arch); +void clt_mdct_clear(mdct_lookup *l, int arch); + +/** Compute a forward MDCT and scale by 4/N, trashes the input array */ +void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, + int shift, int stride, int arch); + +/** Compute a backward MDCT (no scaling) and performs weighted overlap-add + (scales implicitly by 1/2) */ +void clt_mdct_backward_c(const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 * OPUS_RESTRICT window, + int overlap, int shift, int stride, int arch); + +#if !defined(OVERRIDE_OPUS_MDCT) +/* Is run-time CPU detection enabled on this platform? */ +#if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10) + +extern void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])( + const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, const opus_val16 *window, + int overlap, int shift, int stride, int arch); + +#define clt_mdct_forward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \ + ((*CLT_MDCT_FORWARD_IMPL[(arch)&OPUS_ARCHMASK])(_l, _in, _out, \ + _window, _overlap, _shift, \ + _stride, _arch)) + +extern void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])( + const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, const opus_val16 *window, + int overlap, int shift, int stride, int arch); + +#define clt_mdct_backward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \ + (*CLT_MDCT_BACKWARD_IMPL[(arch)&OPUS_ARCHMASK])(_l, _in, _out, \ + _window, _overlap, _shift, \ + _stride, _arch) + +#else /* if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10) */ + +#define clt_mdct_forward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \ + clt_mdct_forward_c(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) + +#define clt_mdct_backward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \ + clt_mdct_backward_c(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) + +#endif /* end if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10) && !defined(FIXED_POINT) */ +#endif /* end if !defined(OVERRIDE_OPUS_MDCT) */ + +#endif diff --git a/libopus/celt/mfrngcod.h b/libopus/celt/mfrngcod.h new file mode 100644 index 0000000..809152a --- /dev/null +++ b/libopus/celt/mfrngcod.h @@ -0,0 +1,48 @@ +/* Copyright (c) 2001-2008 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(_mfrngcode_H) +# define _mfrngcode_H (1) +# include "entcode.h" + +/*Constants used by the entropy encoder/decoder.*/ + +/*The number of bits to output at a time.*/ +# define EC_SYM_BITS (8) +/*The total number of bits in each of the state registers.*/ +# define EC_CODE_BITS (32) +/*The maximum symbol value.*/ +# define EC_SYM_MAX ((1U<<EC_SYM_BITS)-1) +/*Bits to shift by to move a symbol into the high-order position.*/ +# define EC_CODE_SHIFT (EC_CODE_BITS-EC_SYM_BITS-1) +/*Carry bit of the high-order range symbol.*/ +# define EC_CODE_TOP (((opus_uint32)1U)<<(EC_CODE_BITS-1)) +/*Low-order bit of the high-order range symbol.*/ +# define EC_CODE_BOT (EC_CODE_TOP>>EC_SYM_BITS) +/*The number of bits available for the last, partial symbol in the code field.*/ +# define EC_CODE_EXTRA ((EC_CODE_BITS-2)%EC_SYM_BITS+1) +#endif diff --git a/libopus/celt/mips/celt_mipsr1.h b/libopus/celt/mips/celt_mipsr1.h new file mode 100644 index 0000000..c332fe0 --- /dev/null +++ b/libopus/celt/mips/celt_mipsr1.h @@ -0,0 +1,152 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef __CELT_MIPSR1_H__ +#define __CELT_MIPSR1_H__ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_C + +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + +#define OVERRIDE_COMB_FILTER_CONST +#define OVERRIDE_comb_filter +void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, + opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, + const opus_val16 *window, int overlap, int arch) +{ + int i; + opus_val32 x0, x1, x2, x3, x4; + + (void)arch; + + /* printf ("%d %d %f %f\n", T0, T1, g0, g1); */ + opus_val16 g00, g01, g02, g10, g11, g12; + static const opus_val16 gains[3][3] = { + {QCONST16(0.3066406250f, 15), QCONST16(0.2170410156f, 15), QCONST16(0.1296386719f, 15)}, + {QCONST16(0.4638671875f, 15), QCONST16(0.2680664062f, 15), QCONST16(0.f, 15)}, + {QCONST16(0.7998046875f, 15), QCONST16(0.1000976562f, 15), QCONST16(0.f, 15)}}; + + if (g0==0 && g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y, x, N); + return; + } + + g00 = MULT16_16_P15(g0, gains[tapset0][0]); + g01 = MULT16_16_P15(g0, gains[tapset0][1]); + g02 = MULT16_16_P15(g0, gains[tapset0][2]); + g10 = MULT16_16_P15(g1, gains[tapset1][0]); + g11 = MULT16_16_P15(g1, gains[tapset1][1]); + g12 = MULT16_16_P15(g1, gains[tapset1][2]); + x1 = x[-T1+1]; + x2 = x[-T1 ]; + x3 = x[-T1-1]; + x4 = x[-T1-2]; + /* If the filter didn't change, we don't need the overlap */ + if (g0==g1 && T0==T1 && tapset0==tapset1) + overlap=0; + + for (i=0;i<overlap;i++) + { + opus_val16 f; + opus_val32 res; + f = MULT16_16_Q15(window[i],window[i]); + x0= x[i-T1+2]; + + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g00)), "r" ((int)x[i-T0])); + + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g01)), "r" ((int)ADD32(x[i-T0-1],x[i-T0+1]))); + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g02)), "r" ((int)ADD32(x[i-T0-2],x[i-T0+2]))); + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g10)), "r" ((int)x2)); + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g11)), "r" ((int)ADD32(x3,x1))); + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g12)), "r" ((int)ADD32(x4,x0))); + + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (res): "i" (15)); + + y[i] = x[i] + res; + + x4=x3; + x3=x2; + x2=x1; + x1=x0; + } + + x4 = x[i-T1-2]; + x3 = x[i-T1-1]; + x2 = x[i-T1]; + x1 = x[i-T1+1]; + + if (g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y+overlap, x+overlap, N-overlap); + return; + } + + for (i=overlap;i<N;i++) + { + opus_val32 res; + x0=x[i-T1+2]; + + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)g10), "r" ((int)x2)); + + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)g11), "r" ((int)ADD32(x3,x1))); + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)g12), "r" ((int)ADD32(x4,x0))); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (res): "i" (15)); + y[i] = x[i] + res; + x4=x3; + x3=x2; + x2=x1; + x1=x0; + } +} + +#endif /* __CELT_MIPSR1_H__ */ diff --git a/libopus/celt/mips/fixed_generic_mipsr1.h b/libopus/celt/mips/fixed_generic_mipsr1.h new file mode 100644 index 0000000..4a05efb --- /dev/null +++ b/libopus/celt/mips/fixed_generic_mipsr1.h @@ -0,0 +1,126 @@ +/* Copyright (C) 2007-2009 Xiph.Org Foundation + Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2008 CSIRO */ +/** + @file fixed_generic.h + @brief Generic fixed-point operations +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CELT_FIXED_GENERIC_MIPSR1_H +#define CELT_FIXED_GENERIC_MIPSR1_H + +#undef MULT16_32_Q15_ADD +static inline int MULT16_32_Q15_ADD(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + +#undef MULT16_32_Q15_SUB +static inline int MULT16_32_Q15_SUB(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + +#undef MULT16_16_Q15_ADD +static inline int MULT16_16_Q15_ADD(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + +#undef MULT16_16_Q15_SUB +static inline int MULT16_16_Q15_SUB(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + + +#undef MULT16_32_Q16 +static inline int MULT16_32_Q16(int a, int b) +{ + int c; + asm volatile("MULT $ac1,%0, %1" : : "r" (a), "r" (b)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (c): "i" (16)); + return c; +} + +#undef MULT16_32_P16 +static inline int MULT16_32_P16(int a, int b) +{ + int c; + asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b)); + asm volatile("EXTR_R.W %0,$ac1, %1" : "=r" (c): "i" (16)); + return c; +} + +#undef MULT16_32_Q15 +static inline int MULT16_32_Q15(int a, int b) +{ + int c; + asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (c): "i" (15)); + return c; +} + +#undef MULT32_32_Q31 +static inline int MULT32_32_Q31(int a, int b) +{ + int r; + asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (r): "i" (31)); + return r; +} + +#undef PSHR32 +static inline int PSHR32(int a, int shift) +{ + int r; + asm volatile ("SHRAV_R.W %0, %1, %2" :"=r" (r): "r" (a), "r" (shift)); + return r; +} + +#undef MULT16_16_P15 +static inline int MULT16_16_P15(int a, int b) +{ + int r; + asm volatile ("mul %0, %1, %2" :"=r" (r): "r" (a), "r" (b)); + asm volatile ("SHRA_R.W %0, %1, %2" : "+r" (r): "0" (r), "i"(15)); + return r; +} + +#endif /* CELT_FIXED_GENERIC_MIPSR1_H */ diff --git a/libopus/celt/mips/kiss_fft_mipsr1.h b/libopus/celt/mips/kiss_fft_mipsr1.h new file mode 100644 index 0000000..400ca4d --- /dev/null +++ b/libopus/celt/mips/kiss_fft_mipsr1.h @@ -0,0 +1,167 @@ +/*Copyright (c) 2013, Xiph.Org Foundation and contributors. + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_MIPSR1_H +#define KISS_FFT_MIPSR1_H + +#if !defined(KISS_FFT_GUTS_H) +#error "This file should only be included from _kiss_fft_guts.h" +#endif + +#ifdef FIXED_POINT + +#define S_MUL_ADD(a, b, c, d) (S_MUL(a,b)+S_MUL(c,d)) +#define S_MUL_SUB(a, b, c, d) (S_MUL(a,b)-S_MUL(c,d)) + +#undef S_MUL_ADD +static inline int S_MUL_ADD(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + +#undef S_MUL_SUB +static inline int S_MUL_SUB(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + +#undef C_MUL +# define C_MUL(m,a,b) (m=C_MUL_fun(a,b)) +static inline kiss_fft_cpx C_MUL_fun(kiss_fft_cpx a, kiss_twiddle_cpx b) { + kiss_fft_cpx m; + + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.r)); + asm volatile("msub $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.i)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.r): "i" (15)); + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.i)); + asm volatile("madd $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.r)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.i): "i" (15)); + + return m; +} +#undef C_MULC +# define C_MULC(m,a,b) (m=C_MULC_fun(a,b)) +static inline kiss_fft_cpx C_MULC_fun(kiss_fft_cpx a, kiss_twiddle_cpx b) { + kiss_fft_cpx m; + + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.r)); + asm volatile("madd $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.i)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.r): "i" (15)); + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.r)); + asm volatile("msub $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.i)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.i): "i" (15)); + + return m; +} + +#endif /* FIXED_POINT */ + +#define OVERRIDE_kf_bfly5 +static void kf_bfly5( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4; + int i, u; + kiss_fft_cpx scratch[13]; + + const kiss_twiddle_cpx *tw; + kiss_twiddle_cpx ya,yb; + kiss_fft_cpx * Fout_beg = Fout; + +#ifdef FIXED_POINT + ya.r = 10126; + ya.i = -31164; + yb.r = -26510; + yb.i = -19261; +#else + ya = st->twiddles[fstride*m]; + yb = st->twiddles[fstride*2*m]; +#endif + + tw=st->twiddles; + + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + Fout0=Fout; + Fout1=Fout0+m; + Fout2=Fout0+2*m; + Fout3=Fout0+3*m; + Fout4=Fout0+4*m; + + /* For non-custom modes, m is guaranteed to be a multiple of 4. */ + for ( u=0; u<m; ++u ) { + scratch[0] = *Fout0; + + + C_MUL(scratch[1] ,*Fout1, tw[u*fstride]); + C_MUL(scratch[2] ,*Fout2, tw[2*u*fstride]); + C_MUL(scratch[3] ,*Fout3, tw[3*u*fstride]); + C_MUL(scratch[4] ,*Fout4, tw[4*u*fstride]); + + C_ADD( scratch[7],scratch[1],scratch[4]); + C_SUB( scratch[10],scratch[1],scratch[4]); + C_ADD( scratch[8],scratch[2],scratch[3]); + C_SUB( scratch[9],scratch[2],scratch[3]); + + Fout0->r += scratch[7].r + scratch[8].r; + Fout0->i += scratch[7].i + scratch[8].i; + scratch[5].r = scratch[0].r + S_MUL_ADD(scratch[7].r,ya.r,scratch[8].r,yb.r); + scratch[5].i = scratch[0].i + S_MUL_ADD(scratch[7].i,ya.r,scratch[8].i,yb.r); + + scratch[6].r = S_MUL_ADD(scratch[10].i,ya.i,scratch[9].i,yb.i); + scratch[6].i = -S_MUL_ADD(scratch[10].r,ya.i,scratch[9].r,yb.i); + + C_SUB(*Fout1,scratch[5],scratch[6]); + C_ADD(*Fout4,scratch[5],scratch[6]); + + scratch[11].r = scratch[0].r + S_MUL_ADD(scratch[7].r,yb.r,scratch[8].r,ya.r); + scratch[11].i = scratch[0].i + S_MUL_ADD(scratch[7].i,yb.r,scratch[8].i,ya.r); + + scratch[12].r = S_MUL_SUB(scratch[9].i,ya.i,scratch[10].i,yb.i); + scratch[12].i = S_MUL_SUB(scratch[10].r,yb.i,scratch[9].r,ya.i); + + C_ADD(*Fout2,scratch[11],scratch[12]); + C_SUB(*Fout3,scratch[11],scratch[12]); + + ++Fout0;++Fout1;++Fout2;++Fout3;++Fout4; + } + } +} + + +#endif /* KISS_FFT_MIPSR1_H */ diff --git a/libopus/celt/mips/mdct_mipsr1.h b/libopus/celt/mips/mdct_mipsr1.h new file mode 100644 index 0000000..2934dab --- /dev/null +++ b/libopus/celt/mips/mdct_mipsr1.h @@ -0,0 +1,288 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2008 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* This is a simple MDCT implementation that uses a N/4 complex FFT + to do most of the work. It should be relatively straightforward to + plug in pretty much and FFT here. + + This replaces the Vorbis FFT (and uses the exact same API), which + was a bit too messy and that was ending up duplicating code + (might as well use the same FFT everywhere). + + The algorithm is similar to (and inspired from) Fabrice Bellard's + MDCT implementation in FFMPEG, but has differences in signs, ordering + and scaling in many places. +*/ +#ifndef __MDCT_MIPSR1_H__ +#define __MDCT_MIPSR1_H__ + +#ifndef SKIP_CONFIG_H +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#endif + +#include "mdct.h" +#include "kiss_fft.h" +#include "_kiss_fft_guts.h" +#include <math.h> +#include "os_support.h" +#include "mathops.h" +#include "stack_alloc.h" + +/* Forward MDCT trashes the input array */ +#define OVERRIDE_clt_mdct_forward +void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + VARDECL(kiss_fft_scalar, f); + VARDECL(kiss_fft_cpx, f2); + const kiss_fft_state *st = l->kfft[shift]; + const kiss_twiddle_scalar *trig; + opus_val16 scale; +#ifdef FIXED_POINT + /* Allows us to scale with MULT16_32_Q16(), which is faster than + MULT16_32_Q15() on ARM. */ + int scale_shift = st->scale_shift-1; +#endif + + (void)arch; + + SAVE_STACK; + scale = st->scale; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + ALLOC(f, N2, kiss_fft_scalar); + ALLOC(f2, N4, kiss_fft_cpx); + + /* Consider the input to be composed of four blocks: [a, b, c, d] */ + /* Window, shuffle, fold */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1); + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1); + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1); + const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1; + for(i=0;i<((overlap+3)>>2);i++) + { + /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/ + *yp++ = S_MUL_ADD(*wp2, xp1[N2],*wp1,*xp2); + *yp++ = S_MUL_SUB(*wp1, *xp1,*wp2, xp2[-N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + wp1 = window; + wp2 = window+overlap-1; + for(;i<N4-((overlap+3)>>2);i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = *xp2; + *yp++ = *xp1; + xp1+=2; + xp2-=2; + } + for(;i<N4;i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = S_MUL_SUB(*wp2, *xp2, *wp1, xp1[-N2]); + *yp++ = S_MUL_ADD(*wp2, *xp1, *wp1, xp2[N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + } + /* Pre-rotation */ + { + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const kiss_twiddle_scalar *t = &trig[0]; + for(i=0;i<N4;i++) + { + kiss_fft_cpx yc; + kiss_twiddle_scalar t0, t1; + kiss_fft_scalar re, im, yr, yi; + t0 = t[i]; + t1 = t[N4+i]; + re = *yp++; + im = *yp++; + + yr = S_MUL_SUB(re,t0,im,t1); + yi = S_MUL_ADD(im,t0,re,t1); + + yc.r = yr; + yc.i = yi; + yc.r = PSHR32(MULT16_32_Q16(scale, yc.r), scale_shift); + yc.i = PSHR32(MULT16_32_Q16(scale, yc.i), scale_shift); + f2[st->bitrev[i]] = yc; + } + } + + /* N/4 complex FFT, does not downscale anymore */ + opus_fft_impl(st, f2); + + /* Post-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_cpx * OPUS_RESTRICT fp = f2; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1); + const kiss_twiddle_scalar *t = &trig[0]; + /* Temp pointers to make it really clear to the compiler what we're doing */ + for(i=0;i<N4;i++) + { + kiss_fft_scalar yr, yi; + yr = S_MUL_SUB(fp->i,t[N4+i] , fp->r,t[i]); + yi = S_MUL_ADD(fp->r,t[N4+i] ,fp->i,t[i]); + *yp1 = yr; + *yp2 = yi; + fp++; + yp1 += 2*stride; + yp2 -= 2*stride; + } + } + RESTORE_STACK; +} + +#define OVERRIDE_clt_mdct_backward +void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + const kiss_twiddle_scalar *trig; + + (void)arch; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + /* Pre-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in; + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1); + kiss_fft_scalar * OPUS_RESTRICT yp = out+(overlap>>1); + const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0]; + const opus_int16 * OPUS_RESTRICT bitrev = l->kfft[shift]->bitrev; + for(i=0;i<N4;i++) + { + int rev; + kiss_fft_scalar yr, yi; + rev = *bitrev++; + yr = S_MUL_ADD(*xp2, t[i] , *xp1, t[N4+i]); + yi = S_MUL_SUB(*xp1, t[i] , *xp2, t[N4+i]); + /* We swap real and imag because we use an FFT instead of an IFFT. */ + yp[2*rev+1] = yr; + yp[2*rev] = yi; + /* Storing the pre-rotation directly in the bitrev order. */ + xp1+=2*stride; + xp2-=2*stride; + } + } + + opus_fft_impl(l->kfft[shift], (kiss_fft_cpx*)(out+(overlap>>1))); + + /* Post-rotate and de-shuffle from both ends of the buffer at once to make + it in-place. */ + { + kiss_fft_scalar * OPUS_RESTRICT yp0 = out+(overlap>>1); + kiss_fft_scalar * OPUS_RESTRICT yp1 = out+(overlap>>1)+N2-2; + const kiss_twiddle_scalar *t = &trig[0]; + /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the + middle pair will be computed twice. */ + for(i=0;i<(N4+1)>>1;i++) + { + kiss_fft_scalar re, im, yr, yi; + kiss_twiddle_scalar t0, t1; + /* We swap real and imag because we're using an FFT instead of an IFFT. */ + re = yp0[1]; + im = yp0[0]; + t0 = t[i]; + t1 = t[N4+i]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = S_MUL_ADD(re,t0 , im,t1); + yi = S_MUL_SUB(re,t1 , im,t0); + /* We swap real and imag because we're using an FFT instead of an IFFT. */ + re = yp1[1]; + im = yp1[0]; + yp0[0] = yr; + yp1[1] = yi; + + t0 = t[(N4-i-1)]; + t1 = t[(N2-i-1)]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = S_MUL_ADD(re,t0,im,t1); + yi = S_MUL_SUB(re,t1,im,t0); + yp1[0] = yr; + yp0[1] = yi; + yp0 += 2; + yp1 -= 2; + } + } + + /* Mirror on both sides for TDAC */ + { + kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + const opus_val16 * OPUS_RESTRICT wp1 = window; + const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1; + + for(i = 0; i < overlap/2; i++) + { + kiss_fft_scalar x1, x2; + x1 = *xp1; + x2 = *yp1; + *yp1++ = MULT16_32_Q15(*wp2, x2) - MULT16_32_Q15(*wp1, x1); + *xp1-- = MULT16_32_Q15(*wp1, x2) + MULT16_32_Q15(*wp2, x1); + wp1++; + wp2--; + } + } +} +#endif /* __MDCT_MIPSR1_H__ */ diff --git a/libopus/celt/mips/pitch_mipsr1.h b/libopus/celt/mips/pitch_mipsr1.h new file mode 100644 index 0000000..a9500af --- /dev/null +++ b/libopus/celt/mips/pitch_mipsr1.h @@ -0,0 +1,161 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file pitch.h + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PITCH_MIPSR1_H +#define PITCH_MIPSR1_H + +#define OVERRIDE_DUAL_INNER_PROD +static inline void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2, int arch) +{ + int j; + opus_val32 xy01=0; + opus_val32 xy02=0; + + (void)arch; + + asm volatile("MULT $ac1, $0, $0"); + asm volatile("MULT $ac2, $0, $0"); + /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */ + for (j=0;j<N;j++) + { + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j])); + asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j])); + ++j; + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j])); + asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j])); + } + asm volatile ("mflo %0, $ac1": "=r"(xy01)); + asm volatile ("mflo %0, $ac2": "=r"(xy02)); + *xy1 = xy01; + *xy2 = xy02; +} + +static inline void xcorr_kernel_mips(const opus_val16 * x, + const opus_val16 * y, opus_val32 sum[4], int len) +{ + int j; + opus_val16 y_0, y_1, y_2, y_3; + + opus_int64 sum_0, sum_1, sum_2, sum_3; + sum_0 = (opus_int64)sum[0]; + sum_1 = (opus_int64)sum[1]; + sum_2 = (opus_int64)sum[2]; + sum_3 = (opus_int64)sum[3]; + + y_3=0; /* gcc doesn't realize that y_3 can't be used uninitialized */ + y_0=*y++; + y_1=*y++; + y_2=*y++; + for (j=0;j<len-3;j+=4) + { + opus_val16 tmp; + tmp = *x++; + y_3=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_0); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_1); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_2); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_3); + + tmp=*x++; + y_0=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_3); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_0); + + tmp=*x++; + y_1=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_0); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_1); + + + tmp=*x++; + y_2=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_3 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_0 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_1); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_2); + + } + if (j++<len) + { + opus_val16 tmp = *x++; + y_3=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_0 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_1 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_2); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_3); + } + + if (j++<len) + { + opus_val16 tmp=*x++; + y_0=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_3); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_0); + } + + if (j<len) + { + opus_val16 tmp=*x++; + y_1=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_0); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_1); + + } + + sum[0] = (opus_val32)sum_0; + sum[1] = (opus_val32)sum_1; + sum[2] = (opus_val32)sum_2; + sum[3] = (opus_val32)sum_3; +} + +#define OVERRIDE_XCORR_KERNEL +#define xcorr_kernel(x, y, sum, len, arch) \ + ((void)(arch), xcorr_kernel_mips(x, y, sum, len)) + +#endif /* PITCH_MIPSR1_H */ diff --git a/libopus/celt/mips/vq_mipsr1.h b/libopus/celt/mips/vq_mipsr1.h new file mode 100644 index 0000000..f26a33e --- /dev/null +++ b/libopus/celt/mips/vq_mipsr1.h @@ -0,0 +1,116 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef __VQ_MIPSR1_H__ +#define __VQ_MIPSR1_H__ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mathops.h" +#include "arch.h" + +#define OVERRIDE_vq_exp_rotation1 +static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s) +{ + int i; + opus_val16 ms; + celt_norm *Xptr; + Xptr = X; + ms = NEG16(s); + for (i=0;i<len-stride;i++) + { + celt_norm x1, x2; + x1 = Xptr[0]; + x2 = Xptr[stride]; + Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2), s, x1), 15)); + *Xptr++ = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15)); + } + Xptr = &X[len-2*stride-1]; + for (i=len-2*stride-1;i>=0;i--) + { + celt_norm x1, x2; + x1 = Xptr[0]; + x2 = Xptr[stride]; + Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2), s, x1), 15)); + *Xptr-- = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15)); + } +} + +#define OVERRIDE_renormalise_vector +void renormalise_vector(celt_norm *X, int N, opus_val16 gain, int arch) +{ + int i; +#ifdef FIXED_POINT + int k; +#endif + opus_val32 E = EPSILON; + opus_val16 g; + opus_val32 t; + celt_norm *xptr = X; + int X0, X1; + + (void)arch; + + asm volatile("mult $ac1, $0, $0"); + asm volatile("MTLO %0, $ac1" : :"r" (E)); + /*if(N %4) + printf("error");*/ + for (i=0;i<N-2;i+=2) + { + X0 = (int)*xptr++; + asm volatile("MADD $ac1, %0, %1" : : "r" (X0), "r" (X0)); + + X1 = (int)*xptr++; + asm volatile("MADD $ac1, %0, %1" : : "r" (X1), "r" (X1)); + } + + for (;i<N;i++) + { + X0 = (int)*xptr++; + asm volatile("MADD $ac1, %0, %1" : : "r" (X0), "r" (X0)); + } + + asm volatile("MFLO %0, $ac1" : "=r" (E)); +#ifdef FIXED_POINT + k = celt_ilog2(E)>>1; +#endif + t = VSHR32(E, 2*(k-7)); + g = MULT16_16_P15(celt_rsqrt_norm(t),gain); + + xptr = X; + for (i=0;i<N;i++) + { + *xptr = EXTRACT16(PSHR32(MULT16_16(g, *xptr), k+1)); + xptr++; + } + /*return celt_sqrt(E);*/ +} + +#endif /* __VQ_MIPSR1_H__ */ diff --git a/libopus/celt/modes.c b/libopus/celt/modes.c new file mode 100644 index 0000000..390c5e8 --- /dev/null +++ b/libopus/celt/modes.c @@ -0,0 +1,442 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "celt.h" +#include "modes.h" +#include "rate.h" +#include "os_support.h" +#include "stack_alloc.h" +#include "quant_bands.h" +#include "cpu_support.h" + +static const opus_int16 eband5ms[] = { +/*0 200 400 600 800 1k 1.2 1.4 1.6 2k 2.4 2.8 3.2 4k 4.8 5.6 6.8 8k 9.6 12k 15.6 */ + 0, 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20, 24, 28, 34, 40, 48, 60, 78, 100 +}; + +/* Alternate tuning (partially derived from Vorbis) */ +#define BITALLOC_SIZE 11 +/* Bit allocation table in units of 1/32 bit/sample (0.1875 dB SNR) */ +static const unsigned char band_allocation[] = { +/*0 200 400 600 800 1k 1.2 1.4 1.6 2k 2.4 2.8 3.2 4k 4.8 5.6 6.8 8k 9.6 12k 15.6 */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 90, 80, 75, 69, 63, 56, 49, 40, 34, 29, 20, 18, 10, 0, 0, 0, 0, 0, 0, 0, 0, +110,100, 90, 84, 78, 71, 65, 58, 51, 45, 39, 32, 26, 20, 12, 0, 0, 0, 0, 0, 0, +118,110,103, 93, 86, 80, 75, 70, 65, 59, 53, 47, 40, 31, 23, 15, 4, 0, 0, 0, 0, +126,119,112,104, 95, 89, 83, 78, 72, 66, 60, 54, 47, 39, 32, 25, 17, 12, 1, 0, 0, +134,127,120,114,103, 97, 91, 85, 78, 72, 66, 60, 54, 47, 41, 35, 29, 23, 16, 10, 1, +144,137,130,124,113,107,101, 95, 88, 82, 76, 70, 64, 57, 51, 45, 39, 33, 26, 15, 1, +152,145,138,132,123,117,111,105, 98, 92, 86, 80, 74, 67, 61, 55, 49, 43, 36, 20, 1, +162,155,148,142,133,127,121,115,108,102, 96, 90, 84, 77, 71, 65, 59, 53, 46, 30, 1, +172,165,158,152,143,137,131,125,118,112,106,100, 94, 87, 81, 75, 69, 63, 56, 45, 20, +200,200,200,200,200,200,200,200,198,193,188,183,178,173,168,163,158,153,148,129,104, +}; + +#ifndef CUSTOM_MODES_ONLY + #ifdef FIXED_POINT + #include "static_modes_fixed.h" + #else + #include "static_modes_float.h" + #endif +#endif /* CUSTOM_MODES_ONLY */ + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +#ifdef CUSTOM_MODES + +/* Defining 25 critical bands for the full 0-20 kHz audio bandwidth + Taken from http://ccrma.stanford.edu/~jos/bbt/Bark_Frequency_Scale.html */ +#define BARK_BANDS 25 +static const opus_int16 bark_freq[BARK_BANDS+1] = { + 0, 100, 200, 300, 400, + 510, 630, 770, 920, 1080, + 1270, 1480, 1720, 2000, 2320, + 2700, 3150, 3700, 4400, 5300, + 6400, 7700, 9500, 12000, 15500, + 20000}; + +static opus_int16 *compute_ebands(opus_int32 Fs, int frame_size, int res, int *nbEBands) +{ + opus_int16 *eBands; + int i, j, lin, low, high, nBark, offset=0; + + /* All modes that have 2.5 ms short blocks use the same definition */ + if (Fs == 400*(opus_int32)frame_size) + { + *nbEBands = sizeof(eband5ms)/sizeof(eband5ms[0])-1; + eBands = opus_alloc(sizeof(opus_int16)*(*nbEBands+1)); + for (i=0;i<*nbEBands+1;i++) + eBands[i] = eband5ms[i]; + return eBands; + } + /* Find the number of critical bands supported by our sampling rate */ + for (nBark=1;nBark<BARK_BANDS;nBark++) + if (bark_freq[nBark+1]*2 >= Fs) + break; + + /* Find where the linear part ends (i.e. where the spacing is more than min_width */ + for (lin=0;lin<nBark;lin++) + if (bark_freq[lin+1]-bark_freq[lin] >= res) + break; + + low = (bark_freq[lin]+res/2)/res; + high = nBark-lin; + *nbEBands = low+high; + eBands = opus_alloc(sizeof(opus_int16)*(*nbEBands+2)); + + if (eBands==NULL) + return NULL; + + /* Linear spacing (min_width) */ + for (i=0;i<low;i++) + eBands[i] = i; + if (low>0) + offset = eBands[low-1]*res - bark_freq[lin-1]; + /* Spacing follows critical bands */ + for (i=0;i<high;i++) + { + int target = bark_freq[lin+i]; + /* Round to an even value */ + eBands[i+low] = (target+offset/2+res)/(2*res)*2; + offset = eBands[i+low]*res - target; + } + /* Enforce the minimum spacing at the boundary */ + for (i=0;i<*nbEBands;i++) + if (eBands[i] < i) + eBands[i] = i; + /* Round to an even value */ + eBands[*nbEBands] = (bark_freq[nBark]+res)/(2*res)*2; + if (eBands[*nbEBands] > frame_size) + eBands[*nbEBands] = frame_size; + for (i=1;i<*nbEBands-1;i++) + { + if (eBands[i+1]-eBands[i] < eBands[i]-eBands[i-1]) + { + eBands[i] -= (2*eBands[i]-eBands[i-1]-eBands[i+1])/2; + } + } + /* Remove any empty bands. */ + for (i=j=0;i<*nbEBands;i++) + if(eBands[i+1]>eBands[j]) + eBands[++j]=eBands[i+1]; + *nbEBands=j; + + for (i=1;i<*nbEBands;i++) + { + /* Every band must be smaller than the last band. */ + celt_assert(eBands[i]-eBands[i-1]<=eBands[*nbEBands]-eBands[*nbEBands-1]); + /* Each band must be no larger than twice the size of the previous one. */ + celt_assert(eBands[i+1]-eBands[i]<=2*(eBands[i]-eBands[i-1])); + } + + return eBands; +} + +static void compute_allocation_table(CELTMode *mode) +{ + int i, j; + unsigned char *allocVectors; + int maxBands = sizeof(eband5ms)/sizeof(eband5ms[0])-1; + + mode->nbAllocVectors = BITALLOC_SIZE; + allocVectors = opus_alloc(sizeof(unsigned char)*(BITALLOC_SIZE*mode->nbEBands)); + if (allocVectors==NULL) + return; + + /* Check for standard mode */ + if (mode->Fs == 400*(opus_int32)mode->shortMdctSize) + { + for (i=0;i<BITALLOC_SIZE*mode->nbEBands;i++) + allocVectors[i] = band_allocation[i]; + mode->allocVectors = allocVectors; + return; + } + /* If not the standard mode, interpolate */ + /* Compute per-codec-band allocation from per-critical-band matrix */ + for (i=0;i<BITALLOC_SIZE;i++) + { + for (j=0;j<mode->nbEBands;j++) + { + int k; + for (k=0;k<maxBands;k++) + { + if (400*(opus_int32)eband5ms[k] > mode->eBands[j]*(opus_int32)mode->Fs/mode->shortMdctSize) + break; + } + if (k>maxBands-1) + allocVectors[i*mode->nbEBands+j] = band_allocation[i*maxBands + maxBands-1]; + else { + opus_int32 a0, a1; + a1 = mode->eBands[j]*(opus_int32)mode->Fs/mode->shortMdctSize - 400*(opus_int32)eband5ms[k-1]; + a0 = 400*(opus_int32)eband5ms[k] - mode->eBands[j]*(opus_int32)mode->Fs/mode->shortMdctSize; + allocVectors[i*mode->nbEBands+j] = (a0*band_allocation[i*maxBands+k-1] + + a1*band_allocation[i*maxBands+k])/(a0+a1); + } + } + } + + /*printf ("\n"); + for (i=0;i<BITALLOC_SIZE;i++) + { + for (j=0;j<mode->nbEBands;j++) + printf ("%d ", allocVectors[i*mode->nbEBands+j]); + printf ("\n"); + } + exit(0);*/ + + mode->allocVectors = allocVectors; +} + +#endif /* CUSTOM_MODES */ + +CELTMode *opus_custom_mode_create(opus_int32 Fs, int frame_size, int *error) +{ + int i; +#ifdef CUSTOM_MODES + CELTMode *mode=NULL; + int res; + opus_val16 *window; + opus_int16 *logN; + int LM; + int arch = opus_select_arch(); + ALLOC_STACK; +#if !defined(VAR_ARRAYS) && !defined(USE_ALLOCA) + if (global_stack==NULL) + goto failure; +#endif +#endif + +#ifndef CUSTOM_MODES_ONLY + for (i=0;i<TOTAL_MODES;i++) + { + int j; + for (j=0;j<4;j++) + { + if (Fs == static_mode_list[i]->Fs && + (frame_size<<j) == static_mode_list[i]->shortMdctSize*static_mode_list[i]->nbShortMdcts) + { + if (error) + *error = OPUS_OK; + return (CELTMode*)static_mode_list[i]; + } + } + } +#endif /* CUSTOM_MODES_ONLY */ + +#ifndef CUSTOM_MODES + if (error) + *error = OPUS_BAD_ARG; + return NULL; +#else + + /* The good thing here is that permutation of the arguments will automatically be invalid */ + + if (Fs < 8000 || Fs > 96000) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + if (frame_size < 40 || frame_size > 1024 || frame_size%2!=0) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + /* Frames of less than 1ms are not supported. */ + if ((opus_int32)frame_size*1000 < Fs) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + + if ((opus_int32)frame_size*75 >= Fs && (frame_size%16)==0) + { + LM = 3; + } else if ((opus_int32)frame_size*150 >= Fs && (frame_size%8)==0) + { + LM = 2; + } else if ((opus_int32)frame_size*300 >= Fs && (frame_size%4)==0) + { + LM = 1; + } else + { + LM = 0; + } + + /* Shorts longer than 3.3ms are not supported. */ + if ((opus_int32)(frame_size>>LM)*300 > Fs) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + + mode = opus_alloc(sizeof(CELTMode)); + if (mode==NULL) + goto failure; + mode->Fs = Fs; + + /* Pre/de-emphasis depends on sampling rate. The "standard" pre-emphasis + is defined as A(z) = 1 - 0.85*z^-1 at 48 kHz. Other rates should + approximate that. */ + if(Fs < 12000) /* 8 kHz */ + { + mode->preemph[0] = QCONST16(0.3500061035f, 15); + mode->preemph[1] = -QCONST16(0.1799926758f, 15); + mode->preemph[2] = QCONST16(0.2719968125f, SIG_SHIFT); /* exact 1/preemph[3] */ + mode->preemph[3] = QCONST16(3.6765136719f, 13); + } else if(Fs < 24000) /* 16 kHz */ + { + mode->preemph[0] = QCONST16(0.6000061035f, 15); + mode->preemph[1] = -QCONST16(0.1799926758f, 15); + mode->preemph[2] = QCONST16(0.4424998650f, SIG_SHIFT); /* exact 1/preemph[3] */ + mode->preemph[3] = QCONST16(2.2598876953f, 13); + } else if(Fs < 40000) /* 32 kHz */ + { + mode->preemph[0] = QCONST16(0.7799987793f, 15); + mode->preemph[1] = -QCONST16(0.1000061035f, 15); + mode->preemph[2] = QCONST16(0.7499771125f, SIG_SHIFT); /* exact 1/preemph[3] */ + mode->preemph[3] = QCONST16(1.3333740234f, 13); + } else /* 48 kHz */ + { + mode->preemph[0] = QCONST16(0.8500061035f, 15); + mode->preemph[1] = QCONST16(0.0f, 15); + mode->preemph[2] = QCONST16(1.f, SIG_SHIFT); + mode->preemph[3] = QCONST16(1.f, 13); + } + + mode->maxLM = LM; + mode->nbShortMdcts = 1<<LM; + mode->shortMdctSize = frame_size/mode->nbShortMdcts; + res = (mode->Fs+mode->shortMdctSize)/(2*mode->shortMdctSize); + + mode->eBands = compute_ebands(Fs, mode->shortMdctSize, res, &mode->nbEBands); + if (mode->eBands==NULL) + goto failure; +#if !defined(SMALL_FOOTPRINT) + /* Make sure we don't allocate a band larger than our PVQ table. + 208 should be enough, but let's be paranoid. */ + if ((mode->eBands[mode->nbEBands] - mode->eBands[mode->nbEBands-1])<<LM > + 208) { + goto failure; + } +#endif + + mode->effEBands = mode->nbEBands; + while (mode->eBands[mode->effEBands] > mode->shortMdctSize) + mode->effEBands--; + + /* Overlap must be divisible by 4 */ + mode->overlap = ((mode->shortMdctSize>>2)<<2); + + compute_allocation_table(mode); + if (mode->allocVectors==NULL) + goto failure; + + window = (opus_val16*)opus_alloc(mode->overlap*sizeof(opus_val16)); + if (window==NULL) + goto failure; + +#ifndef FIXED_POINT + for (i=0;i<mode->overlap;i++) + window[i] = Q15ONE*sin(.5*M_PI* sin(.5*M_PI*(i+.5)/mode->overlap) * sin(.5*M_PI*(i+.5)/mode->overlap)); +#else + for (i=0;i<mode->overlap;i++) + window[i] = MIN32(32767,floor(.5+32768.*sin(.5*M_PI* sin(.5*M_PI*(i+.5)/mode->overlap) * sin(.5*M_PI*(i+.5)/mode->overlap)))); +#endif + mode->window = window; + + logN = (opus_int16*)opus_alloc(mode->nbEBands*sizeof(opus_int16)); + if (logN==NULL) + goto failure; + + for (i=0;i<mode->nbEBands;i++) + logN[i] = log2_frac(mode->eBands[i+1]-mode->eBands[i], BITRES); + mode->logN = logN; + + compute_pulse_cache(mode, mode->maxLM); + + if (clt_mdct_init(&mode->mdct, 2*mode->shortMdctSize*mode->nbShortMdcts, + mode->maxLM, arch) == 0) + goto failure; + + if (error) + *error = OPUS_OK; + + return mode; +failure: + if (error) + *error = OPUS_ALLOC_FAIL; + if (mode!=NULL) + opus_custom_mode_destroy(mode); + return NULL; +#endif /* !CUSTOM_MODES */ +} + +#ifdef CUSTOM_MODES +void opus_custom_mode_destroy(CELTMode *mode) +{ + int arch = opus_select_arch(); + + if (mode == NULL) + return; +#ifndef CUSTOM_MODES_ONLY + { + int i; + for (i=0;i<TOTAL_MODES;i++) + { + if (mode == static_mode_list[i]) + { + return; + } + } + } +#endif /* CUSTOM_MODES_ONLY */ + opus_free((opus_int16*)mode->eBands); + opus_free((unsigned char*)mode->allocVectors); + + opus_free((opus_val16*)mode->window); + opus_free((opus_int16*)mode->logN); + + opus_free((opus_int16*)mode->cache.index); + opus_free((unsigned char*)mode->cache.bits); + opus_free((unsigned char*)mode->cache.caps); + clt_mdct_clear(&mode->mdct, arch); + + opus_free((CELTMode *)mode); +} +#endif diff --git a/libopus/celt/modes.h b/libopus/celt/modes.h new file mode 100644 index 0000000..be813cc --- /dev/null +++ b/libopus/celt/modes.h @@ -0,0 +1,75 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef MODES_H +#define MODES_H + +#include "opus_types.h" +#include "celt.h" +#include "arch.h" +#include "mdct.h" +#include "entenc.h" +#include "entdec.h" + +#define MAX_PERIOD 1024 + +typedef struct { + int size; + const opus_int16 *index; + const unsigned char *bits; + const unsigned char *caps; +} PulseCache; + +/** Mode definition (opaque) + @brief Mode definition + */ +struct OpusCustomMode { + opus_int32 Fs; + int overlap; + + int nbEBands; + int effEBands; + opus_val16 preemph[4]; + const opus_int16 *eBands; /**< Definition for each "pseudo-critical band" */ + + int maxLM; + int nbShortMdcts; + int shortMdctSize; + + int nbAllocVectors; /**< Number of lines in the matrix below */ + const unsigned char *allocVectors; /**< Number of bits in each band for several rates */ + const opus_int16 *logN; + + const opus_val16 *window; + mdct_lookup mdct; + PulseCache cache; +}; + + +#endif diff --git a/libopus/celt/opus_custom_demo.c b/libopus/celt/opus_custom_demo.c new file mode 100644 index 0000000..ae41c0d --- /dev/null +++ b/libopus/celt/opus_custom_demo.c @@ -0,0 +1,210 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus_custom.h" +#include "arch.h" +#include <stdio.h> +#include <stdlib.h> +#include <math.h> +#include <string.h> + +#define MAX_PACKET 1275 + +int main(int argc, char *argv[]) +{ + int err; + char *inFile, *outFile; + FILE *fin, *fout; + OpusCustomMode *mode=NULL; + OpusCustomEncoder *enc; + OpusCustomDecoder *dec; + int len; + opus_int32 frame_size, channels, rate; + int bytes_per_packet; + unsigned char data[MAX_PACKET]; + int complexity; +#if !(defined (FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH) + int i; + double rmsd = 0; +#endif + int count = 0; + opus_int32 skip; + opus_int16 *in, *out; + if (argc != 9 && argc != 8 && argc != 7) + { + fprintf (stderr, "Usage: test_opus_custom <rate> <channels> <frame size> " + " <bytes per packet> [<complexity> [packet loss rate]] " + "<input> <output>\n"); + return 1; + } + + rate = (opus_int32)atol(argv[1]); + channels = atoi(argv[2]); + frame_size = atoi(argv[3]); + mode = opus_custom_mode_create(rate, frame_size, NULL); + if (mode == NULL) + { + fprintf(stderr, "failed to create a mode\n"); + return 1; + } + + bytes_per_packet = atoi(argv[4]); + if (bytes_per_packet < 0 || bytes_per_packet > MAX_PACKET) + { + fprintf (stderr, "bytes per packet must be between 0 and %d\n", + MAX_PACKET); + return 1; + } + + inFile = argv[argc-2]; + fin = fopen(inFile, "rb"); + if (!fin) + { + fprintf (stderr, "Could not open input file %s\n", argv[argc-2]); + return 1; + } + outFile = argv[argc-1]; + fout = fopen(outFile, "wb+"); + if (!fout) + { + fprintf (stderr, "Could not open output file %s\n", argv[argc-1]); + fclose(fin); + return 1; + } + + enc = opus_custom_encoder_create(mode, channels, &err); + if (err != 0) + { + fprintf(stderr, "Failed to create the encoder: %s\n", opus_strerror(err)); + fclose(fin); + fclose(fout); + return 1; + } + dec = opus_custom_decoder_create(mode, channels, &err); + if (err != 0) + { + fprintf(stderr, "Failed to create the decoder: %s\n", opus_strerror(err)); + fclose(fin); + fclose(fout); + return 1; + } + opus_custom_decoder_ctl(dec, OPUS_GET_LOOKAHEAD(&skip)); + + if (argc>7) + { + complexity=atoi(argv[5]); + opus_custom_encoder_ctl(enc,OPUS_SET_COMPLEXITY(complexity)); + } + + in = (opus_int16*)malloc(frame_size*channels*sizeof(opus_int16)); + out = (opus_int16*)malloc(frame_size*channels*sizeof(opus_int16)); + + while (!feof(fin)) + { + int ret; + err = fread(in, sizeof(short), frame_size*channels, fin); + if (feof(fin)) + break; + len = opus_custom_encode(enc, in, frame_size, data, bytes_per_packet); + if (len <= 0) + fprintf (stderr, "opus_custom_encode() failed: %s\n", opus_strerror(len)); + + /* This is for simulating bit errors */ +#if 0 + int errors = 0; + int eid = 0; + /* This simulates random bit error */ + for (i=0;i<len*8;i++) + { + if (rand()%atoi(argv[8])==0) + { + if (i<64) + { + errors++; + eid = i; + } + data[i/8] ^= 1<<(7-(i%8)); + } + } + if (errors == 1) + data[eid/8] ^= 1<<(7-(eid%8)); + else if (errors%2 == 1) + data[rand()%8] ^= 1<<rand()%8; +#endif + +#if 1 /* Set to zero to use the encoder's output instead */ + /* This is to simulate packet loss */ + if (argc==9 && rand()%1000<atoi(argv[argc-3])) + /*if (errors && (errors%2==0))*/ + ret = opus_custom_decode(dec, NULL, len, out, frame_size); + else + ret = opus_custom_decode(dec, data, len, out, frame_size); + if (ret < 0) + fprintf(stderr, "opus_custom_decode() failed: %s\n", opus_strerror(ret)); +#else + for (i=0;i<ret*channels;i++) + out[i] = in[i]; +#endif +#if !(defined (FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH) + for (i=0;i<ret*channels;i++) + { + rmsd += (in[i]-out[i])*1.0*(in[i]-out[i]); + /*out[i] -= in[i];*/ + } +#endif + count++; + fwrite(out+skip*channels, sizeof(short), (ret-skip)*channels, fout); + skip = 0; + } + PRINT_MIPS(stderr); + + opus_custom_encoder_destroy(enc); + opus_custom_decoder_destroy(dec); + fclose(fin); + fclose(fout); + opus_custom_mode_destroy(mode); + free(in); + free(out); +#if !(defined (FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH) + if (rmsd > 0) + { + rmsd = sqrt(rmsd/(1.0*frame_size*channels*count)); + fprintf (stderr, "Error: encoder doesn't match decoder\n"); + fprintf (stderr, "RMS mismatch is %f\n", rmsd); + return 1; + } else { + fprintf (stderr, "Encoder matches decoder!!\n"); + } +#endif + return 0; +} + diff --git a/libopus/celt/os_support.h b/libopus/celt/os_support.h new file mode 100644 index 0000000..a217197 --- /dev/null +++ b/libopus/celt/os_support.h @@ -0,0 +1,92 @@ +/* Copyright (C) 2007 Jean-Marc Valin + + File: os_support.h + This is the (tiny) OS abstraction layer. Aside from math.h, this is the + only place where system headers are allowed. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are + met: + + 1. Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + + 2. Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR + IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES + OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, + INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN + ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef OS_SUPPORT_H +#define OS_SUPPORT_H + +#ifdef CUSTOM_SUPPORT +# include "custom_support.h" +#endif + +#include "opus_types.h" +#include "opus_defines.h" + +#include <string.h> +#include <stdio.h> +#include <stdlib.h> + +/** Opus wrapper for malloc(). To do your own dynamic allocation, all you need to do is replace this function and opus_free */ +#ifndef OVERRIDE_OPUS_ALLOC +static OPUS_INLINE void *opus_alloc (size_t size) +{ + return malloc(size); +} +#endif + +/** Same as celt_alloc(), except that the area is only needed inside a CELT call (might cause problem with wideband though) */ +#ifndef OVERRIDE_OPUS_ALLOC_SCRATCH +static OPUS_INLINE void *opus_alloc_scratch (size_t size) +{ + /* Scratch space doesn't need to be cleared */ + return opus_alloc(size); +} +#endif + +/** Opus wrapper for free(). To do your own dynamic allocation, all you need to do is replace this function and opus_alloc */ +#ifndef OVERRIDE_OPUS_FREE +static OPUS_INLINE void opus_free (void *ptr) +{ + free(ptr); +} +#endif + +/** Copy n elements from src to dst. The 0* term provides compile-time type checking */ +#ifndef OVERRIDE_OPUS_COPY +#define OPUS_COPY(dst, src, n) (memcpy((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) )) +#endif + +/** Copy n elements from src to dst, allowing overlapping regions. The 0* term + provides compile-time type checking */ +#ifndef OVERRIDE_OPUS_MOVE +#define OPUS_MOVE(dst, src, n) (memmove((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) )) +#endif + +/** Set n elements of dst to zero */ +#ifndef OVERRIDE_OPUS_CLEAR +#define OPUS_CLEAR(dst, n) (memset((dst), 0, (n)*sizeof(*(dst)))) +#endif + +/*#ifdef __GNUC__ +#pragma GCC poison printf sprintf +#pragma GCC poison malloc free realloc calloc +#endif*/ + +#endif /* OS_SUPPORT_H */ + diff --git a/libopus/celt/pitch.c b/libopus/celt/pitch.c new file mode 100644 index 0000000..872582a --- /dev/null +++ b/libopus/celt/pitch.c @@ -0,0 +1,537 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file pitch.c + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "pitch.h" +#include "os_support.h" +#include "modes.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "celt_lpc.h" + +static void find_best_pitch(opus_val32 *xcorr, opus_val16 *y, int len, + int max_pitch, int *best_pitch +#ifdef FIXED_POINT + , int yshift, opus_val32 maxcorr +#endif + ) +{ + int i, j; + opus_val32 Syy=1; + opus_val16 best_num[2]; + opus_val32 best_den[2]; +#ifdef FIXED_POINT + int xshift; + + xshift = celt_ilog2(maxcorr)-14; +#endif + + best_num[0] = -1; + best_num[1] = -1; + best_den[0] = 0; + best_den[1] = 0; + best_pitch[0] = 0; + best_pitch[1] = 1; + for (j=0;j<len;j++) + Syy = ADD32(Syy, SHR32(MULT16_16(y[j],y[j]), yshift)); + for (i=0;i<max_pitch;i++) + { + if (xcorr[i]>0) + { + opus_val16 num; + opus_val32 xcorr16; + xcorr16 = EXTRACT16(VSHR32(xcorr[i], xshift)); +#ifndef FIXED_POINT + /* Considering the range of xcorr16, this should avoid both underflows + and overflows (inf) when squaring xcorr16 */ + xcorr16 *= 1e-12f; +#endif + num = MULT16_16_Q15(xcorr16,xcorr16); + if (MULT16_32_Q15(num,best_den[1]) > MULT16_32_Q15(best_num[1],Syy)) + { + if (MULT16_32_Q15(num,best_den[0]) > MULT16_32_Q15(best_num[0],Syy)) + { + best_num[1] = best_num[0]; + best_den[1] = best_den[0]; + best_pitch[1] = best_pitch[0]; + best_num[0] = num; + best_den[0] = Syy; + best_pitch[0] = i; + } else { + best_num[1] = num; + best_den[1] = Syy; + best_pitch[1] = i; + } + } + } + Syy += SHR32(MULT16_16(y[i+len],y[i+len]),yshift) - SHR32(MULT16_16(y[i],y[i]),yshift); + Syy = MAX32(1, Syy); + } +} + +static void celt_fir5(opus_val16 *x, + const opus_val16 *num, + int N) +{ + int i; + opus_val16 num0, num1, num2, num3, num4; + opus_val32 mem0, mem1, mem2, mem3, mem4; + num0=num[0]; + num1=num[1]; + num2=num[2]; + num3=num[3]; + num4=num[4]; + mem0=0; + mem1=0; + mem2=0; + mem3=0; + mem4=0; + for (i=0;i<N;i++) + { + opus_val32 sum = SHL32(EXTEND32(x[i]), SIG_SHIFT); + sum = MAC16_16(sum,num0,mem0); + sum = MAC16_16(sum,num1,mem1); + sum = MAC16_16(sum,num2,mem2); + sum = MAC16_16(sum,num3,mem3); + sum = MAC16_16(sum,num4,mem4); + mem4 = mem3; + mem3 = mem2; + mem2 = mem1; + mem1 = mem0; + mem0 = x[i]; + x[i] = ROUND16(sum, SIG_SHIFT); + } +} + + +void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp, + int len, int C, int arch) +{ + int i; + opus_val32 ac[5]; + opus_val16 tmp=Q15ONE; + opus_val16 lpc[4]; + opus_val16 lpc2[5]; + opus_val16 c1 = QCONST16(.8f,15); +#ifdef FIXED_POINT + int shift; + opus_val32 maxabs = celt_maxabs32(x[0], len); + if (C==2) + { + opus_val32 maxabs_1 = celt_maxabs32(x[1], len); + maxabs = MAX32(maxabs, maxabs_1); + } + if (maxabs<1) + maxabs=1; + shift = celt_ilog2(maxabs)-10; + if (shift<0) + shift=0; + if (C==2) + shift++; +#endif + for (i=1;i<len>>1;i++) + x_lp[i] = SHR32(HALF32(HALF32(x[0][(2*i-1)]+x[0][(2*i+1)])+x[0][2*i]), shift); + x_lp[0] = SHR32(HALF32(HALF32(x[0][1])+x[0][0]), shift); + if (C==2) + { + for (i=1;i<len>>1;i++) + x_lp[i] += SHR32(HALF32(HALF32(x[1][(2*i-1)]+x[1][(2*i+1)])+x[1][2*i]), shift); + x_lp[0] += SHR32(HALF32(HALF32(x[1][1])+x[1][0]), shift); + } + + _celt_autocorr(x_lp, ac, NULL, 0, + 4, len>>1, arch); + + /* Noise floor -40 dB */ +#ifdef FIXED_POINT + ac[0] += SHR32(ac[0],13); +#else + ac[0] *= 1.0001f; +#endif + /* Lag windowing */ + for (i=1;i<=4;i++) + { + /*ac[i] *= exp(-.5*(2*M_PI*.002*i)*(2*M_PI*.002*i));*/ +#ifdef FIXED_POINT + ac[i] -= MULT16_32_Q15(2*i*i, ac[i]); +#else + ac[i] -= ac[i]*(.008f*i)*(.008f*i); +#endif + } + + _celt_lpc(lpc, ac, 4); + for (i=0;i<4;i++) + { + tmp = MULT16_16_Q15(QCONST16(.9f,15), tmp); + lpc[i] = MULT16_16_Q15(lpc[i], tmp); + } + /* Add a zero */ + lpc2[0] = lpc[0] + QCONST16(.8f,SIG_SHIFT); + lpc2[1] = lpc[1] + MULT16_16_Q15(c1,lpc[0]); + lpc2[2] = lpc[2] + MULT16_16_Q15(c1,lpc[1]); + lpc2[3] = lpc[3] + MULT16_16_Q15(c1,lpc[2]); + lpc2[4] = MULT16_16_Q15(c1,lpc[3]); + celt_fir5(x_lp, lpc2, len>>1); +} + +/* Pure C implementation. */ +#ifdef FIXED_POINT +opus_val32 +#else +void +#endif +celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch) +{ + +#if 0 /* This is a simple version of the pitch correlation that should work + well on DSPs like Blackfin and TI C5x/C6x */ + int i, j; +#ifdef FIXED_POINT + opus_val32 maxcorr=1; +#endif +#if !defined(OVERRIDE_PITCH_XCORR) + (void)arch; +#endif + for (i=0;i<max_pitch;i++) + { + opus_val32 sum = 0; + for (j=0;j<len;j++) + sum = MAC16_16(sum, _x[j], _y[i+j]); + xcorr[i] = sum; +#ifdef FIXED_POINT + maxcorr = MAX32(maxcorr, sum); +#endif + } +#ifdef FIXED_POINT + return maxcorr; +#endif + +#else /* Unrolled version of the pitch correlation -- runs faster on x86 and ARM */ + int i; + /*The EDSP version requires that max_pitch is at least 1, and that _x is + 32-bit aligned. + Since it's hard to put asserts in assembly, put them here.*/ +#ifdef FIXED_POINT + opus_val32 maxcorr=1; +#endif + celt_assert(max_pitch>0); + celt_sig_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0); + for (i=0;i<max_pitch-3;i+=4) + { + opus_val32 sum[4]={0,0,0,0}; + xcorr_kernel(_x, _y+i, sum, len, arch); + xcorr[i]=sum[0]; + xcorr[i+1]=sum[1]; + xcorr[i+2]=sum[2]; + xcorr[i+3]=sum[3]; +#ifdef FIXED_POINT + sum[0] = MAX32(sum[0], sum[1]); + sum[2] = MAX32(sum[2], sum[3]); + sum[0] = MAX32(sum[0], sum[2]); + maxcorr = MAX32(maxcorr, sum[0]); +#endif + } + /* In case max_pitch isn't a multiple of 4, do non-unrolled version. */ + for (;i<max_pitch;i++) + { + opus_val32 sum; + sum = celt_inner_prod(_x, _y+i, len, arch); + xcorr[i] = sum; +#ifdef FIXED_POINT + maxcorr = MAX32(maxcorr, sum); +#endif + } +#ifdef FIXED_POINT + return maxcorr; +#endif +#endif +} + +void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y, + int len, int max_pitch, int *pitch, int arch) +{ + int i, j; + int lag; + int best_pitch[2]={0,0}; + VARDECL(opus_val16, x_lp4); + VARDECL(opus_val16, y_lp4); + VARDECL(opus_val32, xcorr); +#ifdef FIXED_POINT + opus_val32 maxcorr; + opus_val32 xmax, ymax; + int shift=0; +#endif + int offset; + + SAVE_STACK; + + celt_assert(len>0); + celt_assert(max_pitch>0); + lag = len+max_pitch; + + ALLOC(x_lp4, len>>2, opus_val16); + ALLOC(y_lp4, lag>>2, opus_val16); + ALLOC(xcorr, max_pitch>>1, opus_val32); + + /* Downsample by 2 again */ + for (j=0;j<len>>2;j++) + x_lp4[j] = x_lp[2*j]; + for (j=0;j<lag>>2;j++) + y_lp4[j] = y[2*j]; + +#ifdef FIXED_POINT + xmax = celt_maxabs16(x_lp4, len>>2); + ymax = celt_maxabs16(y_lp4, lag>>2); + shift = celt_ilog2(MAX32(1, MAX32(xmax, ymax)))-11; + if (shift>0) + { + for (j=0;j<len>>2;j++) + x_lp4[j] = SHR16(x_lp4[j], shift); + for (j=0;j<lag>>2;j++) + y_lp4[j] = SHR16(y_lp4[j], shift); + /* Use double the shift for a MAC */ + shift *= 2; + } else { + shift = 0; + } +#endif + + /* Coarse search with 4x decimation */ + +#ifdef FIXED_POINT + maxcorr = +#endif + celt_pitch_xcorr(x_lp4, y_lp4, xcorr, len>>2, max_pitch>>2, arch); + + find_best_pitch(xcorr, y_lp4, len>>2, max_pitch>>2, best_pitch +#ifdef FIXED_POINT + , 0, maxcorr +#endif + ); + + /* Finer search with 2x decimation */ +#ifdef FIXED_POINT + maxcorr=1; +#endif + for (i=0;i<max_pitch>>1;i++) + { + opus_val32 sum; + xcorr[i] = 0; + if (abs(i-2*best_pitch[0])>2 && abs(i-2*best_pitch[1])>2) + continue; +#ifdef FIXED_POINT + sum = 0; + for (j=0;j<len>>1;j++) + sum += SHR32(MULT16_16(x_lp[j],y[i+j]), shift); +#else + sum = celt_inner_prod(x_lp, y+i, len>>1, arch); +#endif + xcorr[i] = MAX32(-1, sum); +#ifdef FIXED_POINT + maxcorr = MAX32(maxcorr, sum); +#endif + } + find_best_pitch(xcorr, y, len>>1, max_pitch>>1, best_pitch +#ifdef FIXED_POINT + , shift+1, maxcorr +#endif + ); + + /* Refine by pseudo-interpolation */ + if (best_pitch[0]>0 && best_pitch[0]<(max_pitch>>1)-1) + { + opus_val32 a, b, c; + a = xcorr[best_pitch[0]-1]; + b = xcorr[best_pitch[0]]; + c = xcorr[best_pitch[0]+1]; + if ((c-a) > MULT16_32_Q15(QCONST16(.7f,15),b-a)) + offset = 1; + else if ((a-c) > MULT16_32_Q15(QCONST16(.7f,15),b-c)) + offset = -1; + else + offset = 0; + } else { + offset = 0; + } + *pitch = 2*best_pitch[0]-offset; + + RESTORE_STACK; +} + +#ifdef FIXED_POINT +static opus_val16 compute_pitch_gain(opus_val32 xy, opus_val32 xx, opus_val32 yy) +{ + opus_val32 x2y2; + int sx, sy, shift; + opus_val32 g; + opus_val16 den; + if (xy == 0 || xx == 0 || yy == 0) + return 0; + sx = celt_ilog2(xx)-14; + sy = celt_ilog2(yy)-14; + shift = sx + sy; + x2y2 = SHR32(MULT16_16(VSHR32(xx, sx), VSHR32(yy, sy)), 14); + if (shift & 1) { + if (x2y2 < 32768) + { + x2y2 <<= 1; + shift--; + } else { + x2y2 >>= 1; + shift++; + } + } + den = celt_rsqrt_norm(x2y2); + g = MULT16_32_Q15(den, xy); + g = VSHR32(g, (shift>>1)-1); + return EXTRACT16(MIN32(g, Q15ONE)); +} +#else +static opus_val16 compute_pitch_gain(opus_val32 xy, opus_val32 xx, opus_val32 yy) +{ + return xy/celt_sqrt(1+xx*yy); +} +#endif + +static const int second_check[16] = {0, 0, 3, 2, 3, 2, 5, 2, 3, 2, 3, 2, 5, 2, 3, 2}; +opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, + int N, int *T0_, int prev_period, opus_val16 prev_gain, int arch) +{ + int k, i, T, T0; + opus_val16 g, g0; + opus_val16 pg; + opus_val32 xy,xx,yy,xy2; + opus_val32 xcorr[3]; + opus_val32 best_xy, best_yy; + int offset; + int minperiod0; + VARDECL(opus_val32, yy_lookup); + SAVE_STACK; + + minperiod0 = minperiod; + maxperiod /= 2; + minperiod /= 2; + *T0_ /= 2; + prev_period /= 2; + N /= 2; + x += maxperiod; + if (*T0_>=maxperiod) + *T0_=maxperiod-1; + + T = T0 = *T0_; + ALLOC(yy_lookup, maxperiod+1, opus_val32); + dual_inner_prod(x, x, x-T0, N, &xx, &xy, arch); + yy_lookup[0] = xx; + yy=xx; + for (i=1;i<=maxperiod;i++) + { + yy = yy+MULT16_16(x[-i],x[-i])-MULT16_16(x[N-i],x[N-i]); + yy_lookup[i] = MAX32(0, yy); + } + yy = yy_lookup[T0]; + best_xy = xy; + best_yy = yy; + g = g0 = compute_pitch_gain(xy, xx, yy); + /* Look for any pitch at T/k */ + for (k=2;k<=15;k++) + { + int T1, T1b; + opus_val16 g1; + opus_val16 cont=0; + opus_val16 thresh; + T1 = celt_udiv(2*T0+k, 2*k); + if (T1 < minperiod) + break; + /* Look for another strong correlation at T1b */ + if (k==2) + { + if (T1+T0>maxperiod) + T1b = T0; + else + T1b = T0+T1; + } else + { + T1b = celt_udiv(2*second_check[k]*T0+k, 2*k); + } + dual_inner_prod(x, &x[-T1], &x[-T1b], N, &xy, &xy2, arch); + xy = HALF32(xy + xy2); + yy = HALF32(yy_lookup[T1] + yy_lookup[T1b]); + g1 = compute_pitch_gain(xy, xx, yy); + if (abs(T1-prev_period)<=1) + cont = prev_gain; + else if (abs(T1-prev_period)<=2 && 5*k*k < T0) + cont = HALF16(prev_gain); + else + cont = 0; + thresh = MAX16(QCONST16(.3f,15), MULT16_16_Q15(QCONST16(.7f,15),g0)-cont); + /* Bias against very high pitch (very short period) to avoid false-positives + due to short-term correlation */ + if (T1<3*minperiod) + thresh = MAX16(QCONST16(.4f,15), MULT16_16_Q15(QCONST16(.85f,15),g0)-cont); + else if (T1<2*minperiod) + thresh = MAX16(QCONST16(.5f,15), MULT16_16_Q15(QCONST16(.9f,15),g0)-cont); + if (g1 > thresh) + { + best_xy = xy; + best_yy = yy; + T = T1; + g = g1; + } + } + best_xy = MAX32(0, best_xy); + if (best_yy <= best_xy) + pg = Q15ONE; + else + pg = SHR32(frac_div32(best_xy,best_yy+1),16); + + for (k=0;k<3;k++) + xcorr[k] = celt_inner_prod(x, x-(T+k-1), N, arch); + if ((xcorr[2]-xcorr[0]) > MULT16_32_Q15(QCONST16(.7f,15),xcorr[1]-xcorr[0])) + offset = 1; + else if ((xcorr[0]-xcorr[2]) > MULT16_32_Q15(QCONST16(.7f,15),xcorr[1]-xcorr[2])) + offset = -1; + else + offset = 0; + if (pg > g) + pg = g; + *T0_ = 2*T+offset; + + if (*T0_<minperiod0) + *T0_=minperiod0; + RESTORE_STACK; + return pg; +} diff --git a/libopus/celt/pitch.h b/libopus/celt/pitch.h new file mode 100644 index 0000000..e425f56 --- /dev/null +++ b/libopus/celt/pitch.h @@ -0,0 +1,192 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file pitch.h + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PITCH_H +#define PITCH_H + +#include "modes.h" +#include "cpu_support.h" + +#if (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)) \ + || ((defined(OPUS_X86_MAY_HAVE_SSE4_1) || defined(OPUS_X86_MAY_HAVE_SSE2)) && defined(FIXED_POINT)) +#include "x86/pitch_sse.h" +#endif + +#if defined(MIPSr1_ASM) +#include "mips/pitch_mipsr1.h" +#endif + +#if (defined(OPUS_ARM_ASM) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)) +# include "arm/pitch_arm.h" +#endif + +void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp, + int len, int C, int arch); + +void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y, + int len, int max_pitch, int *pitch, int arch); + +opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, + int N, int *T0, int prev_period, opus_val16 prev_gain, int arch); + + +/* OPT: This is the kernel you really want to optimize. It gets used a lot + by the prefilter and by the PLC. */ +static OPUS_INLINE void xcorr_kernel_c(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len) +{ + int j; + opus_val16 y_0, y_1, y_2, y_3; + celt_assert(len>=3); + y_3=0; /* gcc doesn't realize that y_3 can't be used uninitialized */ + y_0=*y++; + y_1=*y++; + y_2=*y++; + for (j=0;j<len-3;j+=4) + { + opus_val16 tmp; + tmp = *x++; + y_3=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_0); + sum[1] = MAC16_16(sum[1],tmp,y_1); + sum[2] = MAC16_16(sum[2],tmp,y_2); + sum[3] = MAC16_16(sum[3],tmp,y_3); + tmp=*x++; + y_0=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_1); + sum[1] = MAC16_16(sum[1],tmp,y_2); + sum[2] = MAC16_16(sum[2],tmp,y_3); + sum[3] = MAC16_16(sum[3],tmp,y_0); + tmp=*x++; + y_1=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_2); + sum[1] = MAC16_16(sum[1],tmp,y_3); + sum[2] = MAC16_16(sum[2],tmp,y_0); + sum[3] = MAC16_16(sum[3],tmp,y_1); + tmp=*x++; + y_2=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_3); + sum[1] = MAC16_16(sum[1],tmp,y_0); + sum[2] = MAC16_16(sum[2],tmp,y_1); + sum[3] = MAC16_16(sum[3],tmp,y_2); + } + if (j++<len) + { + opus_val16 tmp = *x++; + y_3=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_0); + sum[1] = MAC16_16(sum[1],tmp,y_1); + sum[2] = MAC16_16(sum[2],tmp,y_2); + sum[3] = MAC16_16(sum[3],tmp,y_3); + } + if (j++<len) + { + opus_val16 tmp=*x++; + y_0=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_1); + sum[1] = MAC16_16(sum[1],tmp,y_2); + sum[2] = MAC16_16(sum[2],tmp,y_3); + sum[3] = MAC16_16(sum[3],tmp,y_0); + } + if (j<len) + { + opus_val16 tmp=*x++; + y_1=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_2); + sum[1] = MAC16_16(sum[1],tmp,y_3); + sum[2] = MAC16_16(sum[2],tmp,y_0); + sum[3] = MAC16_16(sum[3],tmp,y_1); + } +} + +#ifndef OVERRIDE_XCORR_KERNEL +#define xcorr_kernel(x, y, sum, len, arch) \ + ((void)(arch),xcorr_kernel_c(x, y, sum, len)) +#endif /* OVERRIDE_XCORR_KERNEL */ + + +static OPUS_INLINE void dual_inner_prod_c(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + opus_val32 xy01=0; + opus_val32 xy02=0; + for (i=0;i<N;i++) + { + xy01 = MAC16_16(xy01, x[i], y01[i]); + xy02 = MAC16_16(xy02, x[i], y02[i]); + } + *xy1 = xy01; + *xy2 = xy02; +} + +#ifndef OVERRIDE_DUAL_INNER_PROD +# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \ + ((void)(arch),dual_inner_prod_c(x, y01, y02, N, xy1, xy2)) +#endif + +/*We make sure a C version is always available for cases where the overhead of + vectorization and passing around an arch flag aren't worth it.*/ +static OPUS_INLINE opus_val32 celt_inner_prod_c(const opus_val16 *x, + const opus_val16 *y, int N) +{ + int i; + opus_val32 xy=0; + for (i=0;i<N;i++) + xy = MAC16_16(xy, x[i], y[i]); + return xy; +} + +#if !defined(OVERRIDE_CELT_INNER_PROD) +# define celt_inner_prod(x, y, N, arch) \ + ((void)(arch),celt_inner_prod_c(x, y, N)) +#endif + +#ifdef NON_STATIC_COMB_FILTER_CONST_C +void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12); +#endif + + +#ifdef FIXED_POINT +opus_val32 +#else +void +#endif +celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch); + +#ifndef OVERRIDE_PITCH_XCORR +# define celt_pitch_xcorr celt_pitch_xcorr_c +#endif + +#endif diff --git a/libopus/celt/quant_bands.c b/libopus/celt/quant_bands.c new file mode 100644 index 0000000..39a221e --- /dev/null +++ b/libopus/celt/quant_bands.c @@ -0,0 +1,563 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "quant_bands.h" +#include "laplace.h" +#include <math.h> +#include "os_support.h" +#include "arch.h" +#include "mathops.h" +#include "stack_alloc.h" +#include "rate.h" + +#ifdef FIXED_POINT +/* Mean energy in each band quantized in Q4 */ +const signed char eMeans[25] = { + 103,100, 92, 85, 81, + 77, 72, 70, 78, 75, + 73, 71, 78, 74, 69, + 72, 70, 74, 76, 71, + 60, 60, 60, 60, 60 +}; +#else +/* Mean energy in each band quantized in Q4 and converted back to float */ +const opus_val16 eMeans[25] = { + 6.437500f, 6.250000f, 5.750000f, 5.312500f, 5.062500f, + 4.812500f, 4.500000f, 4.375000f, 4.875000f, 4.687500f, + 4.562500f, 4.437500f, 4.875000f, 4.625000f, 4.312500f, + 4.500000f, 4.375000f, 4.625000f, 4.750000f, 4.437500f, + 3.750000f, 3.750000f, 3.750000f, 3.750000f, 3.750000f +}; +#endif +/* prediction coefficients: 0.9, 0.8, 0.65, 0.5 */ +#ifdef FIXED_POINT +static const opus_val16 pred_coef[4] = {29440, 26112, 21248, 16384}; +static const opus_val16 beta_coef[4] = {30147, 22282, 12124, 6554}; +static const opus_val16 beta_intra = 4915; +#else +static const opus_val16 pred_coef[4] = {29440/32768., 26112/32768., 21248/32768., 16384/32768.}; +static const opus_val16 beta_coef[4] = {30147/32768., 22282/32768., 12124/32768., 6554/32768.}; +static const opus_val16 beta_intra = 4915/32768.; +#endif + +/*Parameters of the Laplace-like probability models used for the coarse energy. + There is one pair of parameters for each frame size, prediction type + (inter/intra), and band number. + The first number of each pair is the probability of 0, and the second is the + decay rate, both in Q8 precision.*/ +static const unsigned char e_prob_model[4][2][42] = { + /*120 sample frames.*/ + { + /*Inter*/ + { + 72, 127, 65, 129, 66, 128, 65, 128, 64, 128, 62, 128, 64, 128, + 64, 128, 92, 78, 92, 79, 92, 78, 90, 79, 116, 41, 115, 40, + 114, 40, 132, 26, 132, 26, 145, 17, 161, 12, 176, 10, 177, 11 + }, + /*Intra*/ + { + 24, 179, 48, 138, 54, 135, 54, 132, 53, 134, 56, 133, 55, 132, + 55, 132, 61, 114, 70, 96, 74, 88, 75, 88, 87, 74, 89, 66, + 91, 67, 100, 59, 108, 50, 120, 40, 122, 37, 97, 43, 78, 50 + } + }, + /*240 sample frames.*/ + { + /*Inter*/ + { + 83, 78, 84, 81, 88, 75, 86, 74, 87, 71, 90, 73, 93, 74, + 93, 74, 109, 40, 114, 36, 117, 34, 117, 34, 143, 17, 145, 18, + 146, 19, 162, 12, 165, 10, 178, 7, 189, 6, 190, 8, 177, 9 + }, + /*Intra*/ + { + 23, 178, 54, 115, 63, 102, 66, 98, 69, 99, 74, 89, 71, 91, + 73, 91, 78, 89, 86, 80, 92, 66, 93, 64, 102, 59, 103, 60, + 104, 60, 117, 52, 123, 44, 138, 35, 133, 31, 97, 38, 77, 45 + } + }, + /*480 sample frames.*/ + { + /*Inter*/ + { + 61, 90, 93, 60, 105, 42, 107, 41, 110, 45, 116, 38, 113, 38, + 112, 38, 124, 26, 132, 27, 136, 19, 140, 20, 155, 14, 159, 16, + 158, 18, 170, 13, 177, 10, 187, 8, 192, 6, 175, 9, 159, 10 + }, + /*Intra*/ + { + 21, 178, 59, 110, 71, 86, 75, 85, 84, 83, 91, 66, 88, 73, + 87, 72, 92, 75, 98, 72, 105, 58, 107, 54, 115, 52, 114, 55, + 112, 56, 129, 51, 132, 40, 150, 33, 140, 29, 98, 35, 77, 42 + } + }, + /*960 sample frames.*/ + { + /*Inter*/ + { + 42, 121, 96, 66, 108, 43, 111, 40, 117, 44, 123, 32, 120, 36, + 119, 33, 127, 33, 134, 34, 139, 21, 147, 23, 152, 20, 158, 25, + 154, 26, 166, 21, 173, 16, 184, 13, 184, 10, 150, 13, 139, 15 + }, + /*Intra*/ + { + 22, 178, 63, 114, 74, 82, 84, 83, 92, 82, 103, 62, 96, 72, + 96, 67, 101, 73, 107, 72, 113, 55, 118, 52, 125, 52, 118, 52, + 117, 55, 135, 49, 137, 39, 157, 32, 145, 29, 97, 33, 77, 40 + } + } +}; + +static const unsigned char small_energy_icdf[3]={2,1,0}; + +static opus_val32 loss_distortion(const opus_val16 *eBands, opus_val16 *oldEBands, int start, int end, int len, int C) +{ + int c, i; + opus_val32 dist = 0; + c=0; do { + for (i=start;i<end;i++) + { + opus_val16 d = SUB16(SHR16(eBands[i+c*len], 3), SHR16(oldEBands[i+c*len], 3)); + dist = MAC16_16(dist, d,d); + } + } while (++c<C); + return MIN32(200,SHR32(dist,2*DB_SHIFT-6)); +} + +static int quant_coarse_energy_impl(const CELTMode *m, int start, int end, + const opus_val16 *eBands, opus_val16 *oldEBands, + opus_int32 budget, opus_int32 tell, + const unsigned char *prob_model, opus_val16 *error, ec_enc *enc, + int C, int LM, int intra, opus_val16 max_decay, int lfe) +{ + int i, c; + int badness = 0; + opus_val32 prev[2] = {0,0}; + opus_val16 coef; + opus_val16 beta; + + if (tell+3 <= budget) + ec_enc_bit_logp(enc, intra, 3); + if (intra) + { + coef = 0; + beta = beta_intra; + } else { + beta = beta_coef[LM]; + coef = pred_coef[LM]; + } + + /* Encode at a fixed coarse resolution */ + for (i=start;i<end;i++) + { + c=0; + do { + int bits_left; + int qi, qi0; + opus_val32 q; + opus_val16 x; + opus_val32 f, tmp; + opus_val16 oldE; + opus_val16 decay_bound; + x = eBands[i+c*m->nbEBands]; + oldE = MAX16(-QCONST16(9.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]); +#ifdef FIXED_POINT + f = SHL32(EXTEND32(x),7) - PSHR32(MULT16_16(coef,oldE), 8) - prev[c]; + /* Rounding to nearest integer here is really important! */ + qi = (f+QCONST32(.5f,DB_SHIFT+7))>>(DB_SHIFT+7); + decay_bound = EXTRACT16(MAX32(-QCONST16(28.f,DB_SHIFT), + SUB32((opus_val32)oldEBands[i+c*m->nbEBands],max_decay))); +#else + f = x-coef*oldE-prev[c]; + /* Rounding to nearest integer here is really important! */ + qi = (int)floor(.5f+f); + decay_bound = MAX16(-QCONST16(28.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]) - max_decay; +#endif + /* Prevent the energy from going down too quickly (e.g. for bands + that have just one bin) */ + if (qi < 0 && x < decay_bound) + { + qi += (int)SHR16(SUB16(decay_bound,x), DB_SHIFT); + if (qi > 0) + qi = 0; + } + qi0 = qi; + /* If we don't have enough bits to encode all the energy, just assume + something safe. */ + tell = ec_tell(enc); + bits_left = budget-tell-3*C*(end-i); + if (i!=start && bits_left < 30) + { + if (bits_left < 24) + qi = IMIN(1, qi); + if (bits_left < 16) + qi = IMAX(-1, qi); + } + if (lfe && i>=2) + qi = IMIN(qi, 0); + if (budget-tell >= 15) + { + int pi; + pi = 2*IMIN(i,20); + ec_laplace_encode(enc, &qi, + prob_model[pi]<<7, prob_model[pi+1]<<6); + } + else if(budget-tell >= 2) + { + qi = IMAX(-1, IMIN(qi, 1)); + ec_enc_icdf(enc, 2*qi^-(qi<0), small_energy_icdf, 2); + } + else if(budget-tell >= 1) + { + qi = IMIN(0, qi); + ec_enc_bit_logp(enc, -qi, 1); + } + else + qi = -1; + error[i+c*m->nbEBands] = PSHR32(f,7) - SHL16(qi,DB_SHIFT); + badness += abs(qi0-qi); + q = (opus_val32)SHL32(EXTEND32(qi),DB_SHIFT); + + tmp = PSHR32(MULT16_16(coef,oldE),8) + prev[c] + SHL32(q,7); +#ifdef FIXED_POINT + tmp = MAX32(-QCONST32(28.f, DB_SHIFT+7), tmp); +#endif + oldEBands[i+c*m->nbEBands] = PSHR32(tmp, 7); + prev[c] = prev[c] + SHL32(q,7) - MULT16_16(beta,PSHR32(q,8)); + } while (++c < C); + } + return lfe ? 0 : badness; +} + +void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, + const opus_val16 *eBands, opus_val16 *oldEBands, opus_uint32 budget, + opus_val16 *error, ec_enc *enc, int C, int LM, int nbAvailableBytes, + int force_intra, opus_val32 *delayedIntra, int two_pass, int loss_rate, int lfe) +{ + int intra; + opus_val16 max_decay; + VARDECL(opus_val16, oldEBands_intra); + VARDECL(opus_val16, error_intra); + ec_enc enc_start_state; + opus_uint32 tell; + int badness1=0; + opus_int32 intra_bias; + opus_val32 new_distortion; + SAVE_STACK; + + intra = force_intra || (!two_pass && *delayedIntra>2*C*(end-start) && nbAvailableBytes > (end-start)*C); + intra_bias = (opus_int32)((budget**delayedIntra*loss_rate)/(C*512)); + new_distortion = loss_distortion(eBands, oldEBands, start, effEnd, m->nbEBands, C); + + tell = ec_tell(enc); + if (tell+3 > budget) + two_pass = intra = 0; + + max_decay = QCONST16(16.f,DB_SHIFT); + if (end-start>10) + { +#ifdef FIXED_POINT + max_decay = MIN32(max_decay, SHL32(EXTEND32(nbAvailableBytes),DB_SHIFT-3)); +#else + max_decay = MIN32(max_decay, .125f*nbAvailableBytes); +#endif + } + if (lfe) + max_decay = QCONST16(3.f,DB_SHIFT); + enc_start_state = *enc; + + ALLOC(oldEBands_intra, C*m->nbEBands, opus_val16); + ALLOC(error_intra, C*m->nbEBands, opus_val16); + OPUS_COPY(oldEBands_intra, oldEBands, C*m->nbEBands); + + if (two_pass || intra) + { + badness1 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands_intra, budget, + tell, e_prob_model[LM][1], error_intra, enc, C, LM, 1, max_decay, lfe); + } + + if (!intra) + { + unsigned char *intra_buf; + ec_enc enc_intra_state; + opus_int32 tell_intra; + opus_uint32 nstart_bytes; + opus_uint32 nintra_bytes; + opus_uint32 save_bytes; + int badness2; + VARDECL(unsigned char, intra_bits); + + tell_intra = ec_tell_frac(enc); + + enc_intra_state = *enc; + + nstart_bytes = ec_range_bytes(&enc_start_state); + nintra_bytes = ec_range_bytes(&enc_intra_state); + intra_buf = ec_get_buffer(&enc_intra_state) + nstart_bytes; + save_bytes = nintra_bytes-nstart_bytes; + if (save_bytes == 0) + save_bytes = ALLOC_NONE; + ALLOC(intra_bits, save_bytes, unsigned char); + /* Copy bits from intra bit-stream */ + OPUS_COPY(intra_bits, intra_buf, nintra_bytes - nstart_bytes); + + *enc = enc_start_state; + + badness2 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands, budget, + tell, e_prob_model[LM][intra], error, enc, C, LM, 0, max_decay, lfe); + + if (two_pass && (badness1 < badness2 || (badness1 == badness2 && ((opus_int32)ec_tell_frac(enc))+intra_bias > tell_intra))) + { + *enc = enc_intra_state; + /* Copy intra bits to bit-stream */ + OPUS_COPY(intra_buf, intra_bits, nintra_bytes - nstart_bytes); + OPUS_COPY(oldEBands, oldEBands_intra, C*m->nbEBands); + OPUS_COPY(error, error_intra, C*m->nbEBands); + intra = 1; + } + } else { + OPUS_COPY(oldEBands, oldEBands_intra, C*m->nbEBands); + OPUS_COPY(error, error_intra, C*m->nbEBands); + } + + if (intra) + *delayedIntra = new_distortion; + else + *delayedIntra = ADD32(MULT16_32_Q15(MULT16_16_Q15(pred_coef[LM], pred_coef[LM]),*delayedIntra), + new_distortion); + + RESTORE_STACK; +} + +void quant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, ec_enc *enc, int C) +{ + int i, c; + + /* Encode finer resolution */ + for (i=start;i<end;i++) + { + opus_int16 frac = 1<<fine_quant[i]; + if (fine_quant[i] <= 0) + continue; + c=0; + do { + int q2; + opus_val16 offset; +#ifdef FIXED_POINT + /* Has to be without rounding */ + q2 = (error[i+c*m->nbEBands]+QCONST16(.5f,DB_SHIFT))>>(DB_SHIFT-fine_quant[i]); +#else + q2 = (int)floor((error[i+c*m->nbEBands]+.5f)*frac); +#endif + if (q2 > frac-1) + q2 = frac-1; + if (q2<0) + q2 = 0; + ec_enc_bits(enc, q2, fine_quant[i]); +#ifdef FIXED_POINT + offset = SUB16(SHR32(SHL32(EXTEND32(q2),DB_SHIFT)+QCONST16(.5f,DB_SHIFT),fine_quant[i]),QCONST16(.5f,DB_SHIFT)); +#else + offset = (q2+.5f)*(1<<(14-fine_quant[i]))*(1.f/16384) - .5f; +#endif + oldEBands[i+c*m->nbEBands] += offset; + error[i+c*m->nbEBands] -= offset; + /*printf ("%f ", error[i] - offset);*/ + } while (++c < C); + } +} + +void quant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, int *fine_priority, int bits_left, ec_enc *enc, int C) +{ + int i, prio, c; + + /* Use up the remaining bits */ + for (prio=0;prio<2;prio++) + { + for (i=start;i<end && bits_left>=C ;i++) + { + if (fine_quant[i] >= MAX_FINE_BITS || fine_priority[i]!=prio) + continue; + c=0; + do { + int q2; + opus_val16 offset; + q2 = error[i+c*m->nbEBands]<0 ? 0 : 1; + ec_enc_bits(enc, q2, 1); +#ifdef FIXED_POINT + offset = SHR16(SHL16(q2,DB_SHIFT)-QCONST16(.5f,DB_SHIFT),fine_quant[i]+1); +#else + offset = (q2-.5f)*(1<<(14-fine_quant[i]-1))*(1.f/16384); +#endif + oldEBands[i+c*m->nbEBands] += offset; + error[i+c*m->nbEBands] -= offset; + bits_left--; + } while (++c < C); + } + } +} + +void unquant_coarse_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int intra, ec_dec *dec, int C, int LM) +{ + const unsigned char *prob_model = e_prob_model[LM][intra]; + int i, c; + opus_val32 prev[2] = {0, 0}; + opus_val16 coef; + opus_val16 beta; + opus_int32 budget; + opus_int32 tell; + + if (intra) + { + coef = 0; + beta = beta_intra; + } else { + beta = beta_coef[LM]; + coef = pred_coef[LM]; + } + + budget = dec->storage*8; + + /* Decode at a fixed coarse resolution */ + for (i=start;i<end;i++) + { + c=0; + do { + int qi; + opus_val32 q; + opus_val32 tmp; + /* It would be better to express this invariant as a + test on C at function entry, but that isn't enough + to make the static analyzer happy. */ + celt_sig_assert(c<2); + tell = ec_tell(dec); + if(budget-tell>=15) + { + int pi; + pi = 2*IMIN(i,20); + qi = ec_laplace_decode(dec, + prob_model[pi]<<7, prob_model[pi+1]<<6); + } + else if(budget-tell>=2) + { + qi = ec_dec_icdf(dec, small_energy_icdf, 2); + qi = (qi>>1)^-(qi&1); + } + else if(budget-tell>=1) + { + qi = -ec_dec_bit_logp(dec, 1); + } + else + qi = -1; + q = (opus_val32)SHL32(EXTEND32(qi),DB_SHIFT); + + oldEBands[i+c*m->nbEBands] = MAX16(-QCONST16(9.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]); + tmp = PSHR32(MULT16_16(coef,oldEBands[i+c*m->nbEBands]),8) + prev[c] + SHL32(q,7); +#ifdef FIXED_POINT + tmp = MAX32(-QCONST32(28.f, DB_SHIFT+7), tmp); +#endif + oldEBands[i+c*m->nbEBands] = PSHR32(tmp, 7); + prev[c] = prev[c] + SHL32(q,7) - MULT16_16(beta,PSHR32(q,8)); + } while (++c < C); + } +} + +void unquant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, ec_dec *dec, int C) +{ + int i, c; + /* Decode finer resolution */ + for (i=start;i<end;i++) + { + if (fine_quant[i] <= 0) + continue; + c=0; + do { + int q2; + opus_val16 offset; + q2 = ec_dec_bits(dec, fine_quant[i]); +#ifdef FIXED_POINT + offset = SUB16(SHR32(SHL32(EXTEND32(q2),DB_SHIFT)+QCONST16(.5f,DB_SHIFT),fine_quant[i]),QCONST16(.5f,DB_SHIFT)); +#else + offset = (q2+.5f)*(1<<(14-fine_quant[i]))*(1.f/16384) - .5f; +#endif + oldEBands[i+c*m->nbEBands] += offset; + } while (++c < C); + } +} + +void unquant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, int *fine_priority, int bits_left, ec_dec *dec, int C) +{ + int i, prio, c; + + /* Use up the remaining bits */ + for (prio=0;prio<2;prio++) + { + for (i=start;i<end && bits_left>=C ;i++) + { + if (fine_quant[i] >= MAX_FINE_BITS || fine_priority[i]!=prio) + continue; + c=0; + do { + int q2; + opus_val16 offset; + q2 = ec_dec_bits(dec, 1); +#ifdef FIXED_POINT + offset = SHR16(SHL16(q2,DB_SHIFT)-QCONST16(.5f,DB_SHIFT),fine_quant[i]+1); +#else + offset = (q2-.5f)*(1<<(14-fine_quant[i]-1))*(1.f/16384); +#endif + oldEBands[i+c*m->nbEBands] += offset; + bits_left--; + } while (++c < C); + } + } +} + +void amp2Log2(const CELTMode *m, int effEnd, int end, + celt_ener *bandE, opus_val16 *bandLogE, int C) +{ + int c, i; + c=0; + do { + for (i=0;i<effEnd;i++) + { + bandLogE[i+c*m->nbEBands] = + celt_log2(bandE[i+c*m->nbEBands]) + - SHL16((opus_val16)eMeans[i],6); +#ifdef FIXED_POINT + /* Compensate for bandE[] being Q12 but celt_log2() taking a Q14 input. */ + bandLogE[i+c*m->nbEBands] += QCONST16(2.f, DB_SHIFT); +#endif + } + for (i=effEnd;i<end;i++) + bandLogE[c*m->nbEBands+i] = -QCONST16(14.f,DB_SHIFT); + } while (++c < C); +} diff --git a/libopus/celt/quant_bands.h b/libopus/celt/quant_bands.h new file mode 100644 index 0000000..0490bca --- /dev/null +++ b/libopus/celt/quant_bands.h @@ -0,0 +1,66 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef QUANT_BANDS +#define QUANT_BANDS + +#include "arch.h" +#include "modes.h" +#include "entenc.h" +#include "entdec.h" +#include "mathops.h" + +#ifdef FIXED_POINT +extern const signed char eMeans[25]; +#else +extern const opus_val16 eMeans[25]; +#endif + +void amp2Log2(const CELTMode *m, int effEnd, int end, + celt_ener *bandE, opus_val16 *bandLogE, int C); + +void log2Amp(const CELTMode *m, int start, int end, + celt_ener *eBands, const opus_val16 *oldEBands, int C); + +void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, + const opus_val16 *eBands, opus_val16 *oldEBands, opus_uint32 budget, + opus_val16 *error, ec_enc *enc, int C, int LM, + int nbAvailableBytes, int force_intra, opus_val32 *delayedIntra, + int two_pass, int loss_rate, int lfe); + +void quant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, ec_enc *enc, int C); + +void quant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, int *fine_priority, int bits_left, ec_enc *enc, int C); + +void unquant_coarse_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int intra, ec_dec *dec, int C, int LM); + +void unquant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, ec_dec *dec, int C); + +void unquant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, int *fine_priority, int bits_left, ec_dec *dec, int C); + +#endif /* QUANT_BANDS */ diff --git a/libopus/celt/rate.c b/libopus/celt/rate.c new file mode 100644 index 0000000..465e1ba --- /dev/null +++ b/libopus/celt/rate.c @@ -0,0 +1,644 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <math.h> +#include "modes.h" +#include "cwrs.h" +#include "arch.h" +#include "os_support.h" + +#include "entcode.h" +#include "rate.h" + +static const unsigned char LOG2_FRAC_TABLE[24]={ + 0, + 8,13, + 16,19,21,23, + 24,26,27,28,29,30,31,32, + 32,33,34,34,35,36,36,37,37 +}; + +#ifdef CUSTOM_MODES + +/*Determines if V(N,K) fits in a 32-bit unsigned integer. + N and K are themselves limited to 15 bits.*/ +static int fits_in32(int _n, int _k) +{ + static const opus_int16 maxN[15] = { + 32767, 32767, 32767, 1476, 283, 109, 60, 40, + 29, 24, 20, 18, 16, 14, 13}; + static const opus_int16 maxK[15] = { + 32767, 32767, 32767, 32767, 1172, 238, 95, 53, + 36, 27, 22, 18, 16, 15, 13}; + if (_n>=14) + { + if (_k>=14) + return 0; + else + return _n <= maxN[_k]; + } else { + return _k <= maxK[_n]; + } +} + +void compute_pulse_cache(CELTMode *m, int LM) +{ + int C; + int i; + int j; + int curr=0; + int nbEntries=0; + int entryN[100], entryK[100], entryI[100]; + const opus_int16 *eBands = m->eBands; + PulseCache *cache = &m->cache; + opus_int16 *cindex; + unsigned char *bits; + unsigned char *cap; + + cindex = (opus_int16 *)opus_alloc(sizeof(cache->index[0])*m->nbEBands*(LM+2)); + cache->index = cindex; + + /* Scan for all unique band sizes */ + for (i=0;i<=LM+1;i++) + { + for (j=0;j<m->nbEBands;j++) + { + int k; + int N = (eBands[j+1]-eBands[j])<<i>>1; + cindex[i*m->nbEBands+j] = -1; + /* Find other bands that have the same size */ + for (k=0;k<=i;k++) + { + int n; + for (n=0;n<m->nbEBands && (k!=i || n<j);n++) + { + if (N == (eBands[n+1]-eBands[n])<<k>>1) + { + cindex[i*m->nbEBands+j] = cindex[k*m->nbEBands+n]; + break; + } + } + } + if (cache->index[i*m->nbEBands+j] == -1 && N!=0) + { + int K; + entryN[nbEntries] = N; + K = 0; + while (fits_in32(N,get_pulses(K+1)) && K<MAX_PSEUDO) + K++; + entryK[nbEntries] = K; + cindex[i*m->nbEBands+j] = curr; + entryI[nbEntries] = curr; + + curr += K+1; + nbEntries++; + } + } + } + bits = (unsigned char *)opus_alloc(sizeof(unsigned char)*curr); + cache->bits = bits; + cache->size = curr; + /* Compute the cache for all unique sizes */ + for (i=0;i<nbEntries;i++) + { + unsigned char *ptr = bits+entryI[i]; + opus_int16 tmp[CELT_MAX_PULSES+1]; + get_required_bits(tmp, entryN[i], get_pulses(entryK[i]), BITRES); + for (j=1;j<=entryK[i];j++) + ptr[j] = tmp[get_pulses(j)]-1; + ptr[0] = entryK[i]; + } + + /* Compute the maximum rate for each band at which we'll reliably use as + many bits as we ask for. */ + cache->caps = cap = (unsigned char *)opus_alloc(sizeof(cache->caps[0])*(LM+1)*2*m->nbEBands); + for (i=0;i<=LM;i++) + { + for (C=1;C<=2;C++) + { + for (j=0;j<m->nbEBands;j++) + { + int N0; + int max_bits; + N0 = m->eBands[j+1]-m->eBands[j]; + /* N=1 bands only have a sign bit and fine bits. */ + if (N0<<i == 1) + max_bits = C*(1+MAX_FINE_BITS)<<BITRES; + else + { + const unsigned char *pcache; + opus_int32 num; + opus_int32 den; + int LM0; + int N; + int offset; + int ndof; + int qb; + int k; + LM0 = 0; + /* Even-sized bands bigger than N=2 can be split one more time. + As of commit 44203907 all bands >1 are even, including custom modes.*/ + if (N0 > 2) + { + N0>>=1; + LM0--; + } + /* N0=1 bands can't be split down to N<2. */ + else if (N0 <= 1) + { + LM0=IMIN(i,1); + N0<<=LM0; + } + /* Compute the cost for the lowest-level PVQ of a fully split + band. */ + pcache = bits + cindex[(LM0+1)*m->nbEBands+j]; + max_bits = pcache[pcache[0]]+1; + /* Add in the cost of coding regular splits. */ + N = N0; + for(k=0;k<i-LM0;k++){ + max_bits <<= 1; + /* Offset the number of qtheta bits by log2(N)/2 + + QTHETA_OFFSET compared to their "fair share" of + total/N */ + offset = ((m->logN[j]+((LM0+k)<<BITRES))>>1)-QTHETA_OFFSET; + /* The number of qtheta bits we'll allocate if the remainder + is to be max_bits. + The average measured cost for theta is 0.89701 times qb, + approximated here as 459/512. */ + num=459*(opus_int32)((2*N-1)*offset+max_bits); + den=((opus_int32)(2*N-1)<<9)-459; + qb = IMIN((num+(den>>1))/den, 57); + celt_assert(qb >= 0); + max_bits += qb; + N <<= 1; + } + /* Add in the cost of a stereo split, if necessary. */ + if (C==2) + { + max_bits <<= 1; + offset = ((m->logN[j]+(i<<BITRES))>>1)-(N==2?QTHETA_OFFSET_TWOPHASE:QTHETA_OFFSET); + ndof = 2*N-1-(N==2); + /* The average measured cost for theta with the step PDF is + 0.95164 times qb, approximated here as 487/512. */ + num = (N==2?512:487)*(opus_int32)(max_bits+ndof*offset); + den = ((opus_int32)ndof<<9)-(N==2?512:487); + qb = IMIN((num+(den>>1))/den, (N==2?64:61)); + celt_assert(qb >= 0); + max_bits += qb; + } + /* Add the fine bits we'll use. */ + /* Compensate for the extra DoF in stereo */ + ndof = C*N + ((C==2 && N>2) ? 1 : 0); + /* Offset the number of fine bits by log2(N)/2 + FINE_OFFSET + compared to their "fair share" of total/N */ + offset = ((m->logN[j] + (i<<BITRES))>>1)-FINE_OFFSET; + /* N=2 is the only point that doesn't match the curve */ + if (N==2) + offset += 1<<BITRES>>2; + /* The number of fine bits we'll allocate if the remainder is + to be max_bits. */ + num = max_bits+ndof*offset; + den = (ndof-1)<<BITRES; + qb = IMIN((num+(den>>1))/den, MAX_FINE_BITS); + celt_assert(qb >= 0); + max_bits += C*qb<<BITRES; + } + max_bits = (4*max_bits/(C*((m->eBands[j+1]-m->eBands[j])<<i)))-64; + celt_assert(max_bits >= 0); + celt_assert(max_bits < 256); + *cap++ = (unsigned char)max_bits; + } + } + } +} + +#endif /* CUSTOM_MODES */ + +#define ALLOC_STEPS 6 + +static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end, int skip_start, + const int *bits1, const int *bits2, const int *thresh, const int *cap, opus_int32 total, opus_int32 *_balance, + int skip_rsv, int *intensity, int intensity_rsv, int *dual_stereo, int dual_stereo_rsv, int *bits, + int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth) +{ + opus_int32 psum; + int lo, hi; + int i, j; + int logM; + int stereo; + int codedBands=-1; + int alloc_floor; + opus_int32 left, percoeff; + int done; + opus_int32 balance; + SAVE_STACK; + + alloc_floor = C<<BITRES; + stereo = C>1; + + logM = LM<<BITRES; + lo = 0; + hi = 1<<ALLOC_STEPS; + for (i=0;i<ALLOC_STEPS;i++) + { + int mid = (lo+hi)>>1; + psum = 0; + done = 0; + for (j=end;j-->start;) + { + int tmp = bits1[j] + (mid*(opus_int32)bits2[j]>>ALLOC_STEPS); + if (tmp >= thresh[j] || done) + { + done = 1; + /* Don't allocate more than we can actually use */ + psum += IMIN(tmp, cap[j]); + } else { + if (tmp >= alloc_floor) + psum += alloc_floor; + } + } + if (psum > total) + hi = mid; + else + lo = mid; + } + psum = 0; + /*printf ("interp bisection gave %d\n", lo);*/ + done = 0; + for (j=end;j-->start;) + { + int tmp = bits1[j] + ((opus_int32)lo*bits2[j]>>ALLOC_STEPS); + if (tmp < thresh[j] && !done) + { + if (tmp >= alloc_floor) + tmp = alloc_floor; + else + tmp = 0; + } else + done = 1; + /* Don't allocate more than we can actually use */ + tmp = IMIN(tmp, cap[j]); + bits[j] = tmp; + psum += tmp; + } + + /* Decide which bands to skip, working backwards from the end. */ + for (codedBands=end;;codedBands--) + { + int band_width; + int band_bits; + int rem; + j = codedBands-1; + /* Never skip the first band, nor a band that has been boosted by + dynalloc. + In the first case, we'd be coding a bit to signal we're going to waste + all the other bits. + In the second case, we'd be coding a bit to redistribute all the bits + we just signaled should be cocentrated in this band. */ + if (j<=skip_start) + { + /* Give the bit we reserved to end skipping back. */ + total += skip_rsv; + break; + } + /*Figure out how many left-over bits we would be adding to this band. + This can include bits we've stolen back from higher, skipped bands.*/ + left = total-psum; + percoeff = celt_udiv(left, m->eBands[codedBands]-m->eBands[start]); + left -= (m->eBands[codedBands]-m->eBands[start])*percoeff; + rem = IMAX(left-(m->eBands[j]-m->eBands[start]),0); + band_width = m->eBands[codedBands]-m->eBands[j]; + band_bits = (int)(bits[j] + percoeff*band_width + rem); + /*Only code a skip decision if we're above the threshold for this band. + Otherwise it is force-skipped. + This ensures that we have enough bits to code the skip flag.*/ + if (band_bits >= IMAX(thresh[j], alloc_floor+(1<<BITRES))) + { + if (encode) + { + /*This if() block is the only part of the allocation function that + is not a mandatory part of the bitstream: any bands we choose to + skip here must be explicitly signaled.*/ + int depth_threshold; + /*We choose a threshold with some hysteresis to keep bands from + fluctuating in and out, but we try not to fold below a certain point. */ + if (codedBands > 17) + depth_threshold = j<prev ? 7 : 9; + else + depth_threshold = 0; +#ifdef FUZZING + if ((rand()&0x1) == 0) +#else + if (codedBands<=start+2 || (band_bits > (depth_threshold*band_width<<LM<<BITRES)>>4 && j<=signalBandwidth)) +#endif + { + ec_enc_bit_logp(ec, 1, 1); + break; + } + ec_enc_bit_logp(ec, 0, 1); + } else if (ec_dec_bit_logp(ec, 1)) { + break; + } + /*We used a bit to skip this band.*/ + psum += 1<<BITRES; + band_bits -= 1<<BITRES; + } + /*Reclaim the bits originally allocated to this band.*/ + psum -= bits[j]+intensity_rsv; + if (intensity_rsv > 0) + intensity_rsv = LOG2_FRAC_TABLE[j-start]; + psum += intensity_rsv; + if (band_bits >= alloc_floor) + { + /*If we have enough for a fine energy bit per channel, use it.*/ + psum += alloc_floor; + bits[j] = alloc_floor; + } else { + /*Otherwise this band gets nothing at all.*/ + bits[j] = 0; + } + } + + celt_assert(codedBands > start); + /* Code the intensity and dual stereo parameters. */ + if (intensity_rsv > 0) + { + if (encode) + { + *intensity = IMIN(*intensity, codedBands); + ec_enc_uint(ec, *intensity-start, codedBands+1-start); + } + else + *intensity = start+ec_dec_uint(ec, codedBands+1-start); + } + else + *intensity = 0; + if (*intensity <= start) + { + total += dual_stereo_rsv; + dual_stereo_rsv = 0; + } + if (dual_stereo_rsv > 0) + { + if (encode) + ec_enc_bit_logp(ec, *dual_stereo, 1); + else + *dual_stereo = ec_dec_bit_logp(ec, 1); + } + else + *dual_stereo = 0; + + /* Allocate the remaining bits */ + left = total-psum; + percoeff = celt_udiv(left, m->eBands[codedBands]-m->eBands[start]); + left -= (m->eBands[codedBands]-m->eBands[start])*percoeff; + for (j=start;j<codedBands;j++) + bits[j] += ((int)percoeff*(m->eBands[j+1]-m->eBands[j])); + for (j=start;j<codedBands;j++) + { + int tmp = (int)IMIN(left, m->eBands[j+1]-m->eBands[j]); + bits[j] += tmp; + left -= tmp; + } + /*for (j=0;j<end;j++)printf("%d ", bits[j]);printf("\n");*/ + + balance = 0; + for (j=start;j<codedBands;j++) + { + int N0, N, den; + int offset; + int NClogN; + opus_int32 excess, bit; + + celt_assert(bits[j] >= 0); + N0 = m->eBands[j+1]-m->eBands[j]; + N=N0<<LM; + bit = (opus_int32)bits[j]+balance; + + if (N>1) + { + excess = MAX32(bit-cap[j],0); + bits[j] = bit-excess; + + /* Compensate for the extra DoF in stereo */ + den=(C*N+ ((C==2 && N>2 && !*dual_stereo && j<*intensity) ? 1 : 0)); + + NClogN = den*(m->logN[j] + logM); + + /* Offset for the number of fine bits by log2(N)/2 + FINE_OFFSET + compared to their "fair share" of total/N */ + offset = (NClogN>>1)-den*FINE_OFFSET; + + /* N=2 is the only point that doesn't match the curve */ + if (N==2) + offset += den<<BITRES>>2; + + /* Changing the offset for allocating the second and third + fine energy bit */ + if (bits[j] + offset < den*2<<BITRES) + offset += NClogN>>2; + else if (bits[j] + offset < den*3<<BITRES) + offset += NClogN>>3; + + /* Divide with rounding */ + ebits[j] = IMAX(0, (bits[j] + offset + (den<<(BITRES-1)))); + ebits[j] = celt_udiv(ebits[j], den)>>BITRES; + + /* Make sure not to bust */ + if (C*ebits[j] > (bits[j]>>BITRES)) + ebits[j] = bits[j] >> stereo >> BITRES; + + /* More than that is useless because that's about as far as PVQ can go */ + ebits[j] = IMIN(ebits[j], MAX_FINE_BITS); + + /* If we rounded down or capped this band, make it a candidate for the + final fine energy pass */ + fine_priority[j] = ebits[j]*(den<<BITRES) >= bits[j]+offset; + + /* Remove the allocated fine bits; the rest are assigned to PVQ */ + bits[j] -= C*ebits[j]<<BITRES; + + } else { + /* For N=1, all bits go to fine energy except for a single sign bit */ + excess = MAX32(0,bit-(C<<BITRES)); + bits[j] = bit-excess; + ebits[j] = 0; + fine_priority[j] = 1; + } + + /* Fine energy can't take advantage of the re-balancing in + quant_all_bands(). + Instead, do the re-balancing here.*/ + if(excess > 0) + { + int extra_fine; + int extra_bits; + extra_fine = IMIN(excess>>(stereo+BITRES),MAX_FINE_BITS-ebits[j]); + ebits[j] += extra_fine; + extra_bits = extra_fine*C<<BITRES; + fine_priority[j] = extra_bits >= excess-balance; + excess -= extra_bits; + } + balance = excess; + + celt_assert(bits[j] >= 0); + celt_assert(ebits[j] >= 0); + } + /* Save any remaining bits over the cap for the rebalancing in + quant_all_bands(). */ + *_balance = balance; + + /* The skipped bands use all their bits for fine energy. */ + for (;j<end;j++) + { + ebits[j] = bits[j] >> stereo >> BITRES; + celt_assert(C*ebits[j]<<BITRES == bits[j]); + bits[j] = 0; + fine_priority[j] = ebits[j]<1; + } + RESTORE_STACK; + return codedBands; +} + +int clt_compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stereo, + opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth) +{ + int lo, hi, len, j; + int codedBands; + int skip_start; + int skip_rsv; + int intensity_rsv; + int dual_stereo_rsv; + VARDECL(int, bits1); + VARDECL(int, bits2); + VARDECL(int, thresh); + VARDECL(int, trim_offset); + SAVE_STACK; + + total = IMAX(total, 0); + len = m->nbEBands; + skip_start = start; + /* Reserve a bit to signal the end of manually skipped bands. */ + skip_rsv = total >= 1<<BITRES ? 1<<BITRES : 0; + total -= skip_rsv; + /* Reserve bits for the intensity and dual stereo parameters. */ + intensity_rsv = dual_stereo_rsv = 0; + if (C==2) + { + intensity_rsv = LOG2_FRAC_TABLE[end-start]; + if (intensity_rsv>total) + intensity_rsv = 0; + else + { + total -= intensity_rsv; + dual_stereo_rsv = total>=1<<BITRES ? 1<<BITRES : 0; + total -= dual_stereo_rsv; + } + } + ALLOC(bits1, len, int); + ALLOC(bits2, len, int); + ALLOC(thresh, len, int); + ALLOC(trim_offset, len, int); + + for (j=start;j<end;j++) + { + /* Below this threshold, we're sure not to allocate any PVQ bits */ + thresh[j] = IMAX((C)<<BITRES, (3*(m->eBands[j+1]-m->eBands[j])<<LM<<BITRES)>>4); + /* Tilt of the allocation curve */ + trim_offset[j] = C*(m->eBands[j+1]-m->eBands[j])*(alloc_trim-5-LM)*(end-j-1) + *(1<<(LM+BITRES))>>6; + /* Giving less resolution to single-coefficient bands because they get + more benefit from having one coarse value per coefficient*/ + if ((m->eBands[j+1]-m->eBands[j])<<LM==1) + trim_offset[j] -= C<<BITRES; + } + lo = 1; + hi = m->nbAllocVectors - 1; + do + { + int done = 0; + int psum = 0; + int mid = (lo+hi) >> 1; + for (j=end;j-->start;) + { + int bitsj; + int N = m->eBands[j+1]-m->eBands[j]; + bitsj = C*N*m->allocVectors[mid*len+j]<<LM>>2; + if (bitsj > 0) + bitsj = IMAX(0, bitsj + trim_offset[j]); + bitsj += offsets[j]; + if (bitsj >= thresh[j] || done) + { + done = 1; + /* Don't allocate more than we can actually use */ + psum += IMIN(bitsj, cap[j]); + } else { + if (bitsj >= C<<BITRES) + psum += C<<BITRES; + } + } + if (psum > total) + hi = mid - 1; + else + lo = mid + 1; + /*printf ("lo = %d, hi = %d\n", lo, hi);*/ + } + while (lo <= hi); + hi = lo--; + /*printf ("interp between %d and %d\n", lo, hi);*/ + for (j=start;j<end;j++) + { + int bits1j, bits2j; + int N = m->eBands[j+1]-m->eBands[j]; + bits1j = C*N*m->allocVectors[lo*len+j]<<LM>>2; + bits2j = hi>=m->nbAllocVectors ? + cap[j] : C*N*m->allocVectors[hi*len+j]<<LM>>2; + if (bits1j > 0) + bits1j = IMAX(0, bits1j + trim_offset[j]); + if (bits2j > 0) + bits2j = IMAX(0, bits2j + trim_offset[j]); + if (lo > 0) + bits1j += offsets[j]; + bits2j += offsets[j]; + if (offsets[j]>0) + skip_start = j; + bits2j = IMAX(0,bits2j-bits1j); + bits1[j] = bits1j; + bits2[j] = bits2j; + } + codedBands = interp_bits2pulses(m, start, end, skip_start, bits1, bits2, thresh, cap, + total, balance, skip_rsv, intensity, intensity_rsv, dual_stereo, dual_stereo_rsv, + pulses, ebits, fine_priority, C, LM, ec, encode, prev, signalBandwidth); + RESTORE_STACK; + return codedBands; +} + diff --git a/libopus/celt/rate.h b/libopus/celt/rate.h new file mode 100644 index 0000000..fad5e41 --- /dev/null +++ b/libopus/celt/rate.h @@ -0,0 +1,101 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef RATE_H +#define RATE_H + +#define MAX_PSEUDO 40 +#define LOG_MAX_PSEUDO 6 + +#define CELT_MAX_PULSES 128 + +#define MAX_FINE_BITS 8 + +#define FINE_OFFSET 21 +#define QTHETA_OFFSET 4 +#define QTHETA_OFFSET_TWOPHASE 16 + +#include "cwrs.h" +#include "modes.h" + +void compute_pulse_cache(CELTMode *m, int LM); + +static OPUS_INLINE int get_pulses(int i) +{ + return i<8 ? i : (8 + (i&7)) << ((i>>3)-1); +} + +static OPUS_INLINE int bits2pulses(const CELTMode *m, int band, int LM, int bits) +{ + int i; + int lo, hi; + const unsigned char *cache; + + LM++; + cache = m->cache.bits + m->cache.index[LM*m->nbEBands+band]; + + lo = 0; + hi = cache[0]; + bits--; + for (i=0;i<LOG_MAX_PSEUDO;i++) + { + int mid = (lo+hi+1)>>1; + /* OPT: Make sure this is implemented with a conditional move */ + if ((int)cache[mid] >= bits) + hi = mid; + else + lo = mid; + } + if (bits- (lo == 0 ? -1 : (int)cache[lo]) <= (int)cache[hi]-bits) + return lo; + else + return hi; +} + +static OPUS_INLINE int pulses2bits(const CELTMode *m, int band, int LM, int pulses) +{ + const unsigned char *cache; + + LM++; + cache = m->cache.bits + m->cache.index[LM*m->nbEBands+band]; + return pulses == 0 ? 0 : cache[pulses]+1; +} + +/** Compute the pulse allocation, i.e. how many pulses will go in each + * band. + @param m mode + @param offsets Requested increase or decrease in the number of bits for + each band + @param total Number of bands + @param pulses Number of pulses per band (returned) + @return Total number of bits allocated +*/ +int clt_compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stereo, + opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth); + +#endif diff --git a/libopus/celt/stack_alloc.h b/libopus/celt/stack_alloc.h new file mode 100644 index 0000000..2b51c8d --- /dev/null +++ b/libopus/celt/stack_alloc.h @@ -0,0 +1,184 @@ +/* Copyright (C) 2002-2003 Jean-Marc Valin + Copyright (C) 2007-2009 Xiph.Org Foundation */ +/** + @file stack_alloc.h + @brief Temporary memory allocation on stack +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef STACK_ALLOC_H +#define STACK_ALLOC_H + +#include "opus_types.h" +#include "opus_defines.h" + +#if (!defined (VAR_ARRAYS) && !defined (USE_ALLOCA) && !defined (NONTHREADSAFE_PSEUDOSTACK)) +#error "Opus requires one of VAR_ARRAYS, USE_ALLOCA, or NONTHREADSAFE_PSEUDOSTACK be defined to select the temporary allocation mode." +#endif + +#ifdef USE_ALLOCA +# ifdef WIN32 +# include <malloc.h> +# else +# ifdef HAVE_ALLOCA_H +# include <alloca.h> +# else +# include <stdlib.h> +# endif +# endif +#endif + +/** + * @def ALIGN(stack, size) + * + * Aligns the stack to a 'size' boundary + * + * @param stack Stack + * @param size New size boundary + */ + +/** + * @def PUSH(stack, size, type) + * + * Allocates 'size' elements of type 'type' on the stack + * + * @param stack Stack + * @param size Number of elements + * @param type Type of element + */ + +/** + * @def VARDECL(var) + * + * Declare variable on stack + * + * @param var Variable to declare + */ + +/** + * @def ALLOC(var, size, type) + * + * Allocate 'size' elements of 'type' on stack + * + * @param var Name of variable to allocate + * @param size Number of elements + * @param type Type of element + */ + +#if defined(VAR_ARRAYS) + +#define VARDECL(type, var) +#define ALLOC(var, size, type) type var[size] +#define SAVE_STACK +#define RESTORE_STACK +#define ALLOC_STACK +/* C99 does not allow VLAs of size zero */ +#define ALLOC_NONE 1 + +#elif defined(USE_ALLOCA) + +#define VARDECL(type, var) type *var + +# ifdef WIN32 +# define ALLOC(var, size, type) var = ((type*)_alloca(sizeof(type)*(size))) +# else +# define ALLOC(var, size, type) var = ((type*)alloca(sizeof(type)*(size))) +# endif + +#define SAVE_STACK +#define RESTORE_STACK +#define ALLOC_STACK +#define ALLOC_NONE 0 + +#else + +#ifdef CELT_C +char *scratch_ptr=0; +char *global_stack=0; +#else +extern char *global_stack; +extern char *scratch_ptr; +#endif /* CELT_C */ + +#ifdef ENABLE_VALGRIND + +#include <valgrind/memcheck.h> + +#ifdef CELT_C +char *global_stack_top=0; +#else +extern char *global_stack_top; +#endif /* CELT_C */ + +#define ALIGN(stack, size) ((stack) += ((size) - (long)(stack)) & ((size) - 1)) +#define PUSH(stack, size, type) (VALGRIND_MAKE_MEM_NOACCESS(stack, global_stack_top-stack),ALIGN((stack),sizeof(type)/sizeof(char)),VALGRIND_MAKE_MEM_UNDEFINED(stack, ((size)*sizeof(type)/sizeof(char))),(stack)+=(2*(size)*sizeof(type)/sizeof(char)),(type*)((stack)-(2*(size)*sizeof(type)/sizeof(char)))) +#define RESTORE_STACK ((global_stack = _saved_stack),VALGRIND_MAKE_MEM_NOACCESS(global_stack, global_stack_top-global_stack)) +#define ALLOC_STACK char *_saved_stack; ((global_stack = (global_stack==0) ? ((global_stack_top=opus_alloc_scratch(GLOBAL_STACK_SIZE*2)+(GLOBAL_STACK_SIZE*2))-(GLOBAL_STACK_SIZE*2)) : global_stack),VALGRIND_MAKE_MEM_NOACCESS(global_stack, global_stack_top-global_stack)); _saved_stack = global_stack; + +#else + +#define ALIGN(stack, size) ((stack) += ((size) - (long)(stack)) & ((size) - 1)) +#define PUSH(stack, size, type) (ALIGN((stack),sizeof(type)/sizeof(char)),(stack)+=(size)*(sizeof(type)/sizeof(char)),(type*)((stack)-(size)*(sizeof(type)/sizeof(char)))) +#if 0 /* Set this to 1 to instrument pseudostack usage */ +#define RESTORE_STACK (printf("%ld %s:%d\n", global_stack-scratch_ptr, __FILE__, __LINE__),global_stack = _saved_stack) +#else +#define RESTORE_STACK (global_stack = _saved_stack) +#endif +#define ALLOC_STACK char *_saved_stack; (global_stack = (global_stack==0) ? (scratch_ptr=opus_alloc_scratch(GLOBAL_STACK_SIZE)) : global_stack); _saved_stack = global_stack; + +#endif /* ENABLE_VALGRIND */ + +#include "os_support.h" +#define VARDECL(type, var) type *var +#define ALLOC(var, size, type) var = PUSH(global_stack, size, type) +#define SAVE_STACK char *_saved_stack = global_stack; +#define ALLOC_NONE 0 + +#endif /* VAR_ARRAYS */ + + +#ifdef ENABLE_VALGRIND + +#include <valgrind/memcheck.h> +#define OPUS_CHECK_ARRAY(ptr, len) VALGRIND_CHECK_MEM_IS_DEFINED(ptr, len*sizeof(*ptr)) +#define OPUS_CHECK_VALUE(value) VALGRIND_CHECK_VALUE_IS_DEFINED(value) +#define OPUS_CHECK_ARRAY_COND(ptr, len) VALGRIND_CHECK_MEM_IS_DEFINED(ptr, len*sizeof(*ptr)) +#define OPUS_CHECK_VALUE_COND(value) VALGRIND_CHECK_VALUE_IS_DEFINED(value) +#define OPUS_PRINT_INT(value) do {fprintf(stderr, #value " = %d at %s:%d\n", value, __FILE__, __LINE__);}while(0) +#define OPUS_FPRINTF fprintf + +#else + +static OPUS_INLINE int _opus_false(void) {return 0;} +#define OPUS_CHECK_ARRAY(ptr, len) _opus_false() +#define OPUS_CHECK_VALUE(value) _opus_false() +#define OPUS_PRINT_INT(value) do{}while(0) +#define OPUS_FPRINTF (void) + +#endif + + +#endif /* STACK_ALLOC_H */ diff --git a/libopus/celt/static_modes_fixed.h b/libopus/celt/static_modes_fixed.h new file mode 100644 index 0000000..8717d62 --- /dev/null +++ b/libopus/celt/static_modes_fixed.h @@ -0,0 +1,892 @@ +/* The contents of this file was automatically generated by dump_modes.c + with arguments: 48000 960 + It contains static definitions for some pre-defined modes. */ +#include "modes.h" +#include "rate.h" + +#ifdef HAVE_ARM_NE10 +#define OVERRIDE_FFT 1 +#include "static_modes_fixed_arm_ne10.h" +#endif + +#ifndef DEF_WINDOW120 +#define DEF_WINDOW120 +static const opus_val16 window120[120] = { +2, 20, 55, 108, 178, +266, 372, 494, 635, 792, +966, 1157, 1365, 1590, 1831, +2089, 2362, 2651, 2956, 3276, +3611, 3961, 4325, 4703, 5094, +5499, 5916, 6346, 6788, 7241, +7705, 8179, 8663, 9156, 9657, +10167, 10684, 11207, 11736, 12271, +12810, 13353, 13899, 14447, 14997, +15547, 16098, 16648, 17197, 17744, +18287, 18827, 19363, 19893, 20418, +20936, 21447, 21950, 22445, 22931, +23407, 23874, 24330, 24774, 25208, +25629, 26039, 26435, 26819, 27190, +27548, 27893, 28224, 28541, 28845, +29135, 29411, 29674, 29924, 30160, +30384, 30594, 30792, 30977, 31151, +31313, 31463, 31602, 31731, 31849, +31958, 32057, 32148, 32229, 32303, +32370, 32429, 32481, 32528, 32568, +32604, 32634, 32661, 32683, 32701, +32717, 32729, 32740, 32748, 32754, +32758, 32762, 32764, 32766, 32767, +32767, 32767, 32767, 32767, 32767, +}; +#endif + +#ifndef DEF_LOGN400 +#define DEF_LOGN400 +static const opus_int16 logN400[21] = { +0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, 16, 16, 16, 21, 21, 24, 29, 34, 36, }; +#endif + +#ifndef DEF_PULSE_CACHE50 +#define DEF_PULSE_CACHE50 +static const opus_int16 cache_index50[105] = { +-1, -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 41, 41, 41, +82, 82, 123, 164, 200, 222, 0, 0, 0, 0, 0, 0, 0, 0, 41, +41, 41, 41, 123, 123, 123, 164, 164, 240, 266, 283, 295, 41, 41, 41, +41, 41, 41, 41, 41, 123, 123, 123, 123, 240, 240, 240, 266, 266, 305, +318, 328, 336, 123, 123, 123, 123, 123, 123, 123, 123, 240, 240, 240, 240, +305, 305, 305, 318, 318, 343, 351, 358, 364, 240, 240, 240, 240, 240, 240, +240, 240, 305, 305, 305, 305, 343, 343, 343, 351, 351, 370, 376, 382, 387, +}; +static const unsigned char cache_bits50[392] = { +40, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 40, 15, 23, 28, +31, 34, 36, 38, 39, 41, 42, 43, 44, 45, 46, 47, 47, 49, 50, +51, 52, 53, 54, 55, 55, 57, 58, 59, 60, 61, 62, 63, 63, 65, +66, 67, 68, 69, 70, 71, 71, 40, 20, 33, 41, 48, 53, 57, 61, +64, 66, 69, 71, 73, 75, 76, 78, 80, 82, 85, 87, 89, 91, 92, +94, 96, 98, 101, 103, 105, 107, 108, 110, 112, 114, 117, 119, 121, 123, +124, 126, 128, 40, 23, 39, 51, 60, 67, 73, 79, 83, 87, 91, 94, +97, 100, 102, 105, 107, 111, 115, 118, 121, 124, 126, 129, 131, 135, 139, +142, 145, 148, 150, 153, 155, 159, 163, 166, 169, 172, 174, 177, 179, 35, +28, 49, 65, 78, 89, 99, 107, 114, 120, 126, 132, 136, 141, 145, 149, +153, 159, 165, 171, 176, 180, 185, 189, 192, 199, 205, 211, 216, 220, 225, +229, 232, 239, 245, 251, 21, 33, 58, 79, 97, 112, 125, 137, 148, 157, +166, 174, 182, 189, 195, 201, 207, 217, 227, 235, 243, 251, 17, 35, 63, +86, 106, 123, 139, 152, 165, 177, 187, 197, 206, 214, 222, 230, 237, 250, +25, 31, 55, 75, 91, 105, 117, 128, 138, 146, 154, 161, 168, 174, 180, +185, 190, 200, 208, 215, 222, 229, 235, 240, 245, 255, 16, 36, 65, 89, +110, 128, 144, 159, 173, 185, 196, 207, 217, 226, 234, 242, 250, 11, 41, +74, 103, 128, 151, 172, 191, 209, 225, 241, 255, 9, 43, 79, 110, 138, +163, 186, 207, 227, 246, 12, 39, 71, 99, 123, 144, 164, 182, 198, 214, +228, 241, 253, 9, 44, 81, 113, 142, 168, 192, 214, 235, 255, 7, 49, +90, 127, 160, 191, 220, 247, 6, 51, 95, 134, 170, 203, 234, 7, 47, +87, 123, 155, 184, 212, 237, 6, 52, 97, 137, 174, 208, 240, 5, 57, +106, 151, 192, 231, 5, 59, 111, 158, 202, 243, 5, 55, 103, 147, 187, +224, 5, 60, 113, 161, 206, 248, 4, 65, 122, 175, 224, 4, 67, 127, +182, 234, }; +static const unsigned char cache_caps50[168] = { +224, 224, 224, 224, 224, 224, 224, 224, 160, 160, 160, 160, 185, 185, 185, +178, 178, 168, 134, 61, 37, 224, 224, 224, 224, 224, 224, 224, 224, 240, +240, 240, 240, 207, 207, 207, 198, 198, 183, 144, 66, 40, 160, 160, 160, +160, 160, 160, 160, 160, 185, 185, 185, 185, 193, 193, 193, 183, 183, 172, +138, 64, 38, 240, 240, 240, 240, 240, 240, 240, 240, 207, 207, 207, 207, +204, 204, 204, 193, 193, 180, 143, 66, 40, 185, 185, 185, 185, 185, 185, +185, 185, 193, 193, 193, 193, 193, 193, 193, 183, 183, 172, 138, 65, 39, +207, 207, 207, 207, 207, 207, 207, 207, 204, 204, 204, 204, 201, 201, 201, +188, 188, 176, 141, 66, 40, 193, 193, 193, 193, 193, 193, 193, 193, 193, +193, 193, 193, 194, 194, 194, 184, 184, 173, 139, 65, 39, 204, 204, 204, +204, 204, 204, 204, 204, 201, 201, 201, 201, 198, 198, 198, 187, 187, 175, +140, 66, 40, }; +#endif + +#ifndef FFT_TWIDDLES48000_960 +#define FFT_TWIDDLES48000_960 +static const kiss_twiddle_cpx fft_twiddles48000_960[480] = { +{32767, 0}, {32766, -429}, +{32757, -858}, {32743, -1287}, +{32724, -1715}, {32698, -2143}, +{32667, -2570}, {32631, -2998}, +{32588, -3425}, {32541, -3851}, +{32488, -4277}, {32429, -4701}, +{32364, -5125}, {32295, -5548}, +{32219, -5971}, {32138, -6393}, +{32051, -6813}, {31960, -7231}, +{31863, -7650}, {31760, -8067}, +{31652, -8481}, {31539, -8895}, +{31419, -9306}, {31294, -9716}, +{31165, -10126}, {31030, -10532}, +{30889, -10937}, {30743, -11340}, +{30592, -11741}, {30436, -12141}, +{30274, -12540}, {30107, -12935}, +{29936, -13328}, {29758, -13718}, +{29577, -14107}, {29390, -14493}, +{29197, -14875}, {29000, -15257}, +{28797, -15635}, {28590, -16010}, +{28379, -16384}, {28162, -16753}, +{27940, -17119}, {27714, -17484}, +{27482, -17845}, {27246, -18205}, +{27006, -18560}, {26760, -18911}, +{26510, -19260}, {26257, -19606}, +{25997, -19947}, {25734, -20286}, +{25466, -20621}, {25194, -20952}, +{24918, -21281}, {24637, -21605}, +{24353, -21926}, {24063, -22242}, +{23770, -22555}, {23473, -22865}, +{23171, -23171}, {22866, -23472}, +{22557, -23769}, {22244, -24063}, +{21927, -24352}, {21606, -24636}, +{21282, -24917}, {20954, -25194}, +{20622, -25465}, {20288, -25733}, +{19949, -25997}, {19607, -26255}, +{19261, -26509}, {18914, -26760}, +{18561, -27004}, {18205, -27246}, +{17846, -27481}, {17485, -27713}, +{17122, -27940}, {16755, -28162}, +{16385, -28378}, {16012, -28590}, +{15636, -28797}, {15258, -28999}, +{14878, -29197}, {14494, -29389}, +{14108, -29576}, {13720, -29757}, +{13329, -29934}, {12937, -30107}, +{12540, -30274}, {12142, -30435}, +{11744, -30592}, {11342, -30743}, +{10939, -30889}, {10534, -31030}, +{10127, -31164}, {9718, -31294}, +{9307, -31418}, {8895, -31537}, +{8482, -31652}, {8067, -31759}, +{7650, -31862}, {7233, -31960}, +{6815, -32051}, {6393, -32138}, +{5973, -32219}, {5549, -32294}, +{5127, -32364}, {4703, -32429}, +{4278, -32487}, {3852, -32541}, +{3426, -32588}, {2999, -32630}, +{2572, -32667}, {2144, -32698}, +{1716, -32724}, {1287, -32742}, +{860, -32757}, {430, -32766}, +{0, -32767}, {-429, -32766}, +{-858, -32757}, {-1287, -32743}, +{-1715, -32724}, {-2143, -32698}, +{-2570, -32667}, {-2998, -32631}, +{-3425, -32588}, {-3851, -32541}, +{-4277, -32488}, {-4701, -32429}, +{-5125, -32364}, {-5548, -32295}, +{-5971, -32219}, {-6393, -32138}, +{-6813, -32051}, {-7231, -31960}, +{-7650, -31863}, {-8067, -31760}, +{-8481, -31652}, {-8895, -31539}, +{-9306, -31419}, {-9716, -31294}, +{-10126, -31165}, {-10532, -31030}, +{-10937, -30889}, {-11340, -30743}, +{-11741, -30592}, {-12141, -30436}, +{-12540, -30274}, {-12935, -30107}, +{-13328, -29936}, {-13718, -29758}, +{-14107, -29577}, {-14493, -29390}, +{-14875, -29197}, {-15257, -29000}, +{-15635, -28797}, {-16010, -28590}, +{-16384, -28379}, {-16753, -28162}, +{-17119, -27940}, {-17484, -27714}, +{-17845, -27482}, {-18205, -27246}, +{-18560, -27006}, {-18911, -26760}, +{-19260, -26510}, {-19606, -26257}, +{-19947, -25997}, {-20286, -25734}, +{-20621, -25466}, {-20952, -25194}, +{-21281, -24918}, {-21605, -24637}, +{-21926, -24353}, {-22242, -24063}, +{-22555, -23770}, {-22865, -23473}, +{-23171, -23171}, {-23472, -22866}, +{-23769, -22557}, {-24063, -22244}, +{-24352, -21927}, {-24636, -21606}, +{-24917, -21282}, {-25194, -20954}, +{-25465, -20622}, {-25733, -20288}, +{-25997, -19949}, {-26255, -19607}, +{-26509, -19261}, {-26760, -18914}, +{-27004, -18561}, {-27246, -18205}, +{-27481, -17846}, {-27713, -17485}, +{-27940, -17122}, {-28162, -16755}, +{-28378, -16385}, {-28590, -16012}, +{-28797, -15636}, {-28999, -15258}, +{-29197, -14878}, {-29389, -14494}, +{-29576, -14108}, {-29757, -13720}, +{-29934, -13329}, {-30107, -12937}, +{-30274, -12540}, {-30435, -12142}, +{-30592, -11744}, {-30743, -11342}, +{-30889, -10939}, {-31030, -10534}, +{-31164, -10127}, {-31294, -9718}, +{-31418, -9307}, {-31537, -8895}, +{-31652, -8482}, {-31759, -8067}, +{-31862, -7650}, {-31960, -7233}, +{-32051, -6815}, {-32138, -6393}, +{-32219, -5973}, {-32294, -5549}, +{-32364, -5127}, {-32429, -4703}, +{-32487, -4278}, {-32541, -3852}, +{-32588, -3426}, {-32630, -2999}, +{-32667, -2572}, {-32698, -2144}, +{-32724, -1716}, {-32742, -1287}, +{-32757, -860}, {-32766, -430}, +{-32767, 0}, {-32766, 429}, +{-32757, 858}, {-32743, 1287}, +{-32724, 1715}, {-32698, 2143}, +{-32667, 2570}, {-32631, 2998}, +{-32588, 3425}, {-32541, 3851}, +{-32488, 4277}, {-32429, 4701}, +{-32364, 5125}, {-32295, 5548}, +{-32219, 5971}, {-32138, 6393}, +{-32051, 6813}, {-31960, 7231}, +{-31863, 7650}, {-31760, 8067}, +{-31652, 8481}, {-31539, 8895}, +{-31419, 9306}, {-31294, 9716}, +{-31165, 10126}, {-31030, 10532}, +{-30889, 10937}, {-30743, 11340}, +{-30592, 11741}, {-30436, 12141}, +{-30274, 12540}, {-30107, 12935}, +{-29936, 13328}, {-29758, 13718}, +{-29577, 14107}, {-29390, 14493}, +{-29197, 14875}, {-29000, 15257}, +{-28797, 15635}, {-28590, 16010}, +{-28379, 16384}, {-28162, 16753}, +{-27940, 17119}, {-27714, 17484}, +{-27482, 17845}, {-27246, 18205}, +{-27006, 18560}, {-26760, 18911}, +{-26510, 19260}, {-26257, 19606}, +{-25997, 19947}, {-25734, 20286}, +{-25466, 20621}, {-25194, 20952}, +{-24918, 21281}, {-24637, 21605}, +{-24353, 21926}, {-24063, 22242}, +{-23770, 22555}, {-23473, 22865}, +{-23171, 23171}, {-22866, 23472}, +{-22557, 23769}, {-22244, 24063}, +{-21927, 24352}, {-21606, 24636}, +{-21282, 24917}, {-20954, 25194}, +{-20622, 25465}, {-20288, 25733}, +{-19949, 25997}, {-19607, 26255}, +{-19261, 26509}, {-18914, 26760}, +{-18561, 27004}, {-18205, 27246}, +{-17846, 27481}, {-17485, 27713}, +{-17122, 27940}, {-16755, 28162}, +{-16385, 28378}, {-16012, 28590}, +{-15636, 28797}, {-15258, 28999}, +{-14878, 29197}, {-14494, 29389}, +{-14108, 29576}, {-13720, 29757}, +{-13329, 29934}, {-12937, 30107}, +{-12540, 30274}, {-12142, 30435}, +{-11744, 30592}, {-11342, 30743}, +{-10939, 30889}, {-10534, 31030}, +{-10127, 31164}, {-9718, 31294}, +{-9307, 31418}, {-8895, 31537}, +{-8482, 31652}, {-8067, 31759}, +{-7650, 31862}, {-7233, 31960}, +{-6815, 32051}, {-6393, 32138}, +{-5973, 32219}, {-5549, 32294}, +{-5127, 32364}, {-4703, 32429}, +{-4278, 32487}, {-3852, 32541}, +{-3426, 32588}, {-2999, 32630}, +{-2572, 32667}, {-2144, 32698}, +{-1716, 32724}, {-1287, 32742}, +{-860, 32757}, {-430, 32766}, +{0, 32767}, {429, 32766}, +{858, 32757}, {1287, 32743}, +{1715, 32724}, {2143, 32698}, +{2570, 32667}, {2998, 32631}, +{3425, 32588}, {3851, 32541}, +{4277, 32488}, {4701, 32429}, +{5125, 32364}, {5548, 32295}, +{5971, 32219}, {6393, 32138}, +{6813, 32051}, {7231, 31960}, +{7650, 31863}, {8067, 31760}, +{8481, 31652}, {8895, 31539}, +{9306, 31419}, {9716, 31294}, +{10126, 31165}, {10532, 31030}, +{10937, 30889}, {11340, 30743}, +{11741, 30592}, {12141, 30436}, +{12540, 30274}, {12935, 30107}, +{13328, 29936}, {13718, 29758}, +{14107, 29577}, {14493, 29390}, +{14875, 29197}, {15257, 29000}, +{15635, 28797}, {16010, 28590}, +{16384, 28379}, {16753, 28162}, +{17119, 27940}, {17484, 27714}, +{17845, 27482}, {18205, 27246}, +{18560, 27006}, {18911, 26760}, +{19260, 26510}, {19606, 26257}, +{19947, 25997}, {20286, 25734}, +{20621, 25466}, {20952, 25194}, +{21281, 24918}, {21605, 24637}, +{21926, 24353}, {22242, 24063}, +{22555, 23770}, {22865, 23473}, +{23171, 23171}, {23472, 22866}, +{23769, 22557}, {24063, 22244}, +{24352, 21927}, {24636, 21606}, +{24917, 21282}, {25194, 20954}, +{25465, 20622}, {25733, 20288}, +{25997, 19949}, {26255, 19607}, +{26509, 19261}, {26760, 18914}, +{27004, 18561}, {27246, 18205}, +{27481, 17846}, {27713, 17485}, +{27940, 17122}, {28162, 16755}, +{28378, 16385}, {28590, 16012}, +{28797, 15636}, {28999, 15258}, +{29197, 14878}, {29389, 14494}, +{29576, 14108}, {29757, 13720}, +{29934, 13329}, {30107, 12937}, +{30274, 12540}, {30435, 12142}, +{30592, 11744}, {30743, 11342}, +{30889, 10939}, {31030, 10534}, +{31164, 10127}, {31294, 9718}, +{31418, 9307}, {31537, 8895}, +{31652, 8482}, {31759, 8067}, +{31862, 7650}, {31960, 7233}, +{32051, 6815}, {32138, 6393}, +{32219, 5973}, {32294, 5549}, +{32364, 5127}, {32429, 4703}, +{32487, 4278}, {32541, 3852}, +{32588, 3426}, {32630, 2999}, +{32667, 2572}, {32698, 2144}, +{32724, 1716}, {32742, 1287}, +{32757, 860}, {32766, 430}, +}; +#ifndef FFT_BITREV480 +#define FFT_BITREV480 +static const opus_int16 fft_bitrev480[480] = { +0, 96, 192, 288, 384, 32, 128, 224, 320, 416, 64, 160, 256, 352, 448, +8, 104, 200, 296, 392, 40, 136, 232, 328, 424, 72, 168, 264, 360, 456, +16, 112, 208, 304, 400, 48, 144, 240, 336, 432, 80, 176, 272, 368, 464, +24, 120, 216, 312, 408, 56, 152, 248, 344, 440, 88, 184, 280, 376, 472, +4, 100, 196, 292, 388, 36, 132, 228, 324, 420, 68, 164, 260, 356, 452, +12, 108, 204, 300, 396, 44, 140, 236, 332, 428, 76, 172, 268, 364, 460, +20, 116, 212, 308, 404, 52, 148, 244, 340, 436, 84, 180, 276, 372, 468, +28, 124, 220, 316, 412, 60, 156, 252, 348, 444, 92, 188, 284, 380, 476, +1, 97, 193, 289, 385, 33, 129, 225, 321, 417, 65, 161, 257, 353, 449, +9, 105, 201, 297, 393, 41, 137, 233, 329, 425, 73, 169, 265, 361, 457, +17, 113, 209, 305, 401, 49, 145, 241, 337, 433, 81, 177, 273, 369, 465, +25, 121, 217, 313, 409, 57, 153, 249, 345, 441, 89, 185, 281, 377, 473, +5, 101, 197, 293, 389, 37, 133, 229, 325, 421, 69, 165, 261, 357, 453, +13, 109, 205, 301, 397, 45, 141, 237, 333, 429, 77, 173, 269, 365, 461, +21, 117, 213, 309, 405, 53, 149, 245, 341, 437, 85, 181, 277, 373, 469, +29, 125, 221, 317, 413, 61, 157, 253, 349, 445, 93, 189, 285, 381, 477, +2, 98, 194, 290, 386, 34, 130, 226, 322, 418, 66, 162, 258, 354, 450, +10, 106, 202, 298, 394, 42, 138, 234, 330, 426, 74, 170, 266, 362, 458, +18, 114, 210, 306, 402, 50, 146, 242, 338, 434, 82, 178, 274, 370, 466, +26, 122, 218, 314, 410, 58, 154, 250, 346, 442, 90, 186, 282, 378, 474, +6, 102, 198, 294, 390, 38, 134, 230, 326, 422, 70, 166, 262, 358, 454, +14, 110, 206, 302, 398, 46, 142, 238, 334, 430, 78, 174, 270, 366, 462, +22, 118, 214, 310, 406, 54, 150, 246, 342, 438, 86, 182, 278, 374, 470, +30, 126, 222, 318, 414, 62, 158, 254, 350, 446, 94, 190, 286, 382, 478, +3, 99, 195, 291, 387, 35, 131, 227, 323, 419, 67, 163, 259, 355, 451, +11, 107, 203, 299, 395, 43, 139, 235, 331, 427, 75, 171, 267, 363, 459, +19, 115, 211, 307, 403, 51, 147, 243, 339, 435, 83, 179, 275, 371, 467, +27, 123, 219, 315, 411, 59, 155, 251, 347, 443, 91, 187, 283, 379, 475, +7, 103, 199, 295, 391, 39, 135, 231, 327, 423, 71, 167, 263, 359, 455, +15, 111, 207, 303, 399, 47, 143, 239, 335, 431, 79, 175, 271, 367, 463, +23, 119, 215, 311, 407, 55, 151, 247, 343, 439, 87, 183, 279, 375, 471, +31, 127, 223, 319, 415, 63, 159, 255, 351, 447, 95, 191, 287, 383, 479, +}; +#endif + +#ifndef FFT_BITREV240 +#define FFT_BITREV240 +static const opus_int16 fft_bitrev240[240] = { +0, 48, 96, 144, 192, 16, 64, 112, 160, 208, 32, 80, 128, 176, 224, +4, 52, 100, 148, 196, 20, 68, 116, 164, 212, 36, 84, 132, 180, 228, +8, 56, 104, 152, 200, 24, 72, 120, 168, 216, 40, 88, 136, 184, 232, +12, 60, 108, 156, 204, 28, 76, 124, 172, 220, 44, 92, 140, 188, 236, +1, 49, 97, 145, 193, 17, 65, 113, 161, 209, 33, 81, 129, 177, 225, +5, 53, 101, 149, 197, 21, 69, 117, 165, 213, 37, 85, 133, 181, 229, +9, 57, 105, 153, 201, 25, 73, 121, 169, 217, 41, 89, 137, 185, 233, +13, 61, 109, 157, 205, 29, 77, 125, 173, 221, 45, 93, 141, 189, 237, +2, 50, 98, 146, 194, 18, 66, 114, 162, 210, 34, 82, 130, 178, 226, +6, 54, 102, 150, 198, 22, 70, 118, 166, 214, 38, 86, 134, 182, 230, +10, 58, 106, 154, 202, 26, 74, 122, 170, 218, 42, 90, 138, 186, 234, +14, 62, 110, 158, 206, 30, 78, 126, 174, 222, 46, 94, 142, 190, 238, +3, 51, 99, 147, 195, 19, 67, 115, 163, 211, 35, 83, 131, 179, 227, +7, 55, 103, 151, 199, 23, 71, 119, 167, 215, 39, 87, 135, 183, 231, +11, 59, 107, 155, 203, 27, 75, 123, 171, 219, 43, 91, 139, 187, 235, +15, 63, 111, 159, 207, 31, 79, 127, 175, 223, 47, 95, 143, 191, 239, +}; +#endif + +#ifndef FFT_BITREV120 +#define FFT_BITREV120 +static const opus_int16 fft_bitrev120[120] = { +0, 24, 48, 72, 96, 8, 32, 56, 80, 104, 16, 40, 64, 88, 112, +4, 28, 52, 76, 100, 12, 36, 60, 84, 108, 20, 44, 68, 92, 116, +1, 25, 49, 73, 97, 9, 33, 57, 81, 105, 17, 41, 65, 89, 113, +5, 29, 53, 77, 101, 13, 37, 61, 85, 109, 21, 45, 69, 93, 117, +2, 26, 50, 74, 98, 10, 34, 58, 82, 106, 18, 42, 66, 90, 114, +6, 30, 54, 78, 102, 14, 38, 62, 86, 110, 22, 46, 70, 94, 118, +3, 27, 51, 75, 99, 11, 35, 59, 83, 107, 19, 43, 67, 91, 115, +7, 31, 55, 79, 103, 15, 39, 63, 87, 111, 23, 47, 71, 95, 119, +}; +#endif + +#ifndef FFT_BITREV60 +#define FFT_BITREV60 +static const opus_int16 fft_bitrev60[60] = { +0, 12, 24, 36, 48, 4, 16, 28, 40, 52, 8, 20, 32, 44, 56, +1, 13, 25, 37, 49, 5, 17, 29, 41, 53, 9, 21, 33, 45, 57, +2, 14, 26, 38, 50, 6, 18, 30, 42, 54, 10, 22, 34, 46, 58, +3, 15, 27, 39, 51, 7, 19, 31, 43, 55, 11, 23, 35, 47, 59, +}; +#endif + +#ifndef FFT_STATE48000_960_0 +#define FFT_STATE48000_960_0 +static const kiss_fft_state fft_state48000_960_0 = { +480, /* nfft */ +17476, /* scale */ +8, /* scale_shift */ +-1, /* shift */ +{5, 96, 3, 32, 4, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev480, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_480, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_1 +#define FFT_STATE48000_960_1 +static const kiss_fft_state fft_state48000_960_1 = { +240, /* nfft */ +17476, /* scale */ +7, /* scale_shift */ +1, /* shift */ +{5, 48, 3, 16, 4, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev240, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_240, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_2 +#define FFT_STATE48000_960_2 +static const kiss_fft_state fft_state48000_960_2 = { +120, /* nfft */ +17476, /* scale */ +6, /* scale_shift */ +2, /* shift */ +{5, 24, 3, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev120, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_120, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_3 +#define FFT_STATE48000_960_3 +static const kiss_fft_state fft_state48000_960_3 = { +60, /* nfft */ +17476, /* scale */ +5, /* scale_shift */ +3, /* shift */ +{5, 12, 3, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev60, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_60, +#else +NULL, +#endif +}; +#endif + +#endif + +#ifndef MDCT_TWIDDLES960 +#define MDCT_TWIDDLES960 +static const opus_val16 mdct_twiddles960[1800] = { +32767, 32767, 32767, 32766, 32765, +32763, 32761, 32759, 32756, 32753, +32750, 32746, 32742, 32738, 32733, +32728, 32722, 32717, 32710, 32704, +32697, 32690, 32682, 32674, 32666, +32657, 32648, 32639, 32629, 32619, +32609, 32598, 32587, 32576, 32564, +32552, 32539, 32526, 32513, 32500, +32486, 32472, 32457, 32442, 32427, +32411, 32395, 32379, 32362, 32345, +32328, 32310, 32292, 32274, 32255, +32236, 32217, 32197, 32177, 32157, +32136, 32115, 32093, 32071, 32049, +32027, 32004, 31981, 31957, 31933, +31909, 31884, 31859, 31834, 31809, +31783, 31756, 31730, 31703, 31676, +31648, 31620, 31592, 31563, 31534, +31505, 31475, 31445, 31415, 31384, +31353, 31322, 31290, 31258, 31226, +31193, 31160, 31127, 31093, 31059, +31025, 30990, 30955, 30920, 30884, +30848, 30812, 30775, 30738, 30701, +30663, 30625, 30587, 30548, 30509, +30470, 30430, 30390, 30350, 30309, +30269, 30227, 30186, 30144, 30102, +30059, 30016, 29973, 29930, 29886, +29842, 29797, 29752, 29707, 29662, +29616, 29570, 29524, 29477, 29430, +29383, 29335, 29287, 29239, 29190, +29142, 29092, 29043, 28993, 28943, +28892, 28842, 28791, 28739, 28688, +28636, 28583, 28531, 28478, 28425, +28371, 28317, 28263, 28209, 28154, +28099, 28044, 27988, 27932, 27876, +27820, 27763, 27706, 27648, 27591, +27533, 27474, 27416, 27357, 27298, +27238, 27178, 27118, 27058, 26997, +26936, 26875, 26814, 26752, 26690, +26628, 26565, 26502, 26439, 26375, +26312, 26247, 26183, 26119, 26054, +25988, 25923, 25857, 25791, 25725, +25658, 25592, 25524, 25457, 25389, +25322, 25253, 25185, 25116, 25047, +24978, 24908, 24838, 24768, 24698, +24627, 24557, 24485, 24414, 24342, +24270, 24198, 24126, 24053, 23980, +23907, 23834, 23760, 23686, 23612, +23537, 23462, 23387, 23312, 23237, +23161, 23085, 23009, 22932, 22856, +22779, 22701, 22624, 22546, 22468, +22390, 22312, 22233, 22154, 22075, +21996, 21916, 21836, 21756, 21676, +21595, 21515, 21434, 21352, 21271, +21189, 21107, 21025, 20943, 20860, +20777, 20694, 20611, 20528, 20444, +20360, 20276, 20192, 20107, 20022, +19937, 19852, 19767, 19681, 19595, +19509, 19423, 19336, 19250, 19163, +19076, 18988, 18901, 18813, 18725, +18637, 18549, 18460, 18372, 18283, +18194, 18104, 18015, 17925, 17835, +17745, 17655, 17565, 17474, 17383, +17292, 17201, 17110, 17018, 16927, +16835, 16743, 16650, 16558, 16465, +16372, 16279, 16186, 16093, 15999, +15906, 15812, 15718, 15624, 15529, +15435, 15340, 15245, 15150, 15055, +14960, 14864, 14769, 14673, 14577, +14481, 14385, 14288, 14192, 14095, +13998, 13901, 13804, 13706, 13609, +13511, 13414, 13316, 13218, 13119, +13021, 12923, 12824, 12725, 12626, +12527, 12428, 12329, 12230, 12130, +12030, 11930, 11831, 11730, 11630, +11530, 11430, 11329, 11228, 11128, +11027, 10926, 10824, 10723, 10622, +10520, 10419, 10317, 10215, 10113, +10011, 9909, 9807, 9704, 9602, +9499, 9397, 9294, 9191, 9088, +8985, 8882, 8778, 8675, 8572, +8468, 8364, 8261, 8157, 8053, +7949, 7845, 7741, 7637, 7532, +7428, 7323, 7219, 7114, 7009, +6905, 6800, 6695, 6590, 6485, +6380, 6274, 6169, 6064, 5958, +5853, 5747, 5642, 5536, 5430, +5325, 5219, 5113, 5007, 4901, +4795, 4689, 4583, 4476, 4370, +4264, 4157, 4051, 3945, 3838, +3732, 3625, 3518, 3412, 3305, +3198, 3092, 2985, 2878, 2771, +2664, 2558, 2451, 2344, 2237, +2130, 2023, 1916, 1809, 1702, +1594, 1487, 1380, 1273, 1166, +1059, 952, 844, 737, 630, +523, 416, 308, 201, 94, +-13, -121, -228, -335, -442, +-550, -657, -764, -871, -978, +-1086, -1193, -1300, -1407, -1514, +-1621, -1728, -1835, -1942, -2049, +-2157, -2263, -2370, -2477, -2584, +-2691, -2798, -2905, -3012, -3118, +-3225, -3332, -3439, -3545, -3652, +-3758, -3865, -3971, -4078, -4184, +-4290, -4397, -4503, -4609, -4715, +-4821, -4927, -5033, -5139, -5245, +-5351, -5457, -5562, -5668, -5774, +-5879, -5985, -6090, -6195, -6301, +-6406, -6511, -6616, -6721, -6826, +-6931, -7036, -7140, -7245, -7349, +-7454, -7558, -7663, -7767, -7871, +-7975, -8079, -8183, -8287, -8390, +-8494, -8597, -8701, -8804, -8907, +-9011, -9114, -9217, -9319, -9422, +-9525, -9627, -9730, -9832, -9934, +-10037, -10139, -10241, -10342, -10444, +-10546, -10647, -10748, -10850, -10951, +-11052, -11153, -11253, -11354, -11455, +-11555, -11655, -11756, -11856, -11955, +-12055, -12155, -12254, -12354, -12453, +-12552, -12651, -12750, -12849, -12947, +-13046, -13144, -13242, -13340, -13438, +-13536, -13633, -13731, -13828, -13925, +-14022, -14119, -14216, -14312, -14409, +-14505, -14601, -14697, -14793, -14888, +-14984, -15079, -15174, -15269, -15364, +-15459, -15553, -15647, -15741, -15835, +-15929, -16023, -16116, -16210, -16303, +-16396, -16488, -16581, -16673, -16766, +-16858, -16949, -17041, -17133, -17224, +-17315, -17406, -17497, -17587, -17678, +-17768, -17858, -17948, -18037, -18127, +-18216, -18305, -18394, -18483, -18571, +-18659, -18747, -18835, -18923, -19010, +-19098, -19185, -19271, -19358, -19444, +-19531, -19617, -19702, -19788, -19873, +-19959, -20043, -20128, -20213, -20297, +-20381, -20465, -20549, -20632, -20715, +-20798, -20881, -20963, -21046, -21128, +-21210, -21291, -21373, -21454, -21535, +-21616, -21696, -21776, -21856, -21936, +-22016, -22095, -22174, -22253, -22331, +-22410, -22488, -22566, -22643, -22721, +-22798, -22875, -22951, -23028, -23104, +-23180, -23256, -23331, -23406, -23481, +-23556, -23630, -23704, -23778, -23852, +-23925, -23998, -24071, -24144, -24216, +-24288, -24360, -24432, -24503, -24574, +-24645, -24716, -24786, -24856, -24926, +-24995, -25064, -25133, -25202, -25270, +-25339, -25406, -25474, -25541, -25608, +-25675, -25742, -25808, -25874, -25939, +-26005, -26070, -26135, -26199, -26264, +-26327, -26391, -26455, -26518, -26581, +-26643, -26705, -26767, -26829, -26891, +-26952, -27013, -27073, -27133, -27193, +-27253, -27312, -27372, -27430, -27489, +-27547, -27605, -27663, -27720, -27777, +-27834, -27890, -27946, -28002, -28058, +-28113, -28168, -28223, -28277, -28331, +-28385, -28438, -28491, -28544, -28596, +-28649, -28701, -28752, -28803, -28854, +-28905, -28955, -29006, -29055, -29105, +-29154, -29203, -29251, -29299, -29347, +-29395, -29442, -29489, -29535, -29582, +-29628, -29673, -29719, -29764, -29808, +-29853, -29897, -29941, -29984, -30027, +-30070, -30112, -30154, -30196, -30238, +-30279, -30320, -30360, -30400, -30440, +-30480, -30519, -30558, -30596, -30635, +-30672, -30710, -30747, -30784, -30821, +-30857, -30893, -30929, -30964, -30999, +-31033, -31068, -31102, -31135, -31168, +-31201, -31234, -31266, -31298, -31330, +-31361, -31392, -31422, -31453, -31483, +-31512, -31541, -31570, -31599, -31627, +-31655, -31682, -31710, -31737, -31763, +-31789, -31815, -31841, -31866, -31891, +-31915, -31939, -31963, -31986, -32010, +-32032, -32055, -32077, -32099, -32120, +-32141, -32162, -32182, -32202, -32222, +-32241, -32260, -32279, -32297, -32315, +-32333, -32350, -32367, -32383, -32399, +-32415, -32431, -32446, -32461, -32475, +-32489, -32503, -32517, -32530, -32542, +-32555, -32567, -32579, -32590, -32601, +-32612, -32622, -32632, -32641, -32651, +-32659, -32668, -32676, -32684, -32692, +-32699, -32706, -32712, -32718, -32724, +-32729, -32734, -32739, -32743, -32747, +-32751, -32754, -32757, -32760, -32762, +-32764, -32765, -32767, -32767, -32767, +32767, 32767, 32765, 32761, 32756, +32750, 32742, 32732, 32722, 32710, +32696, 32681, 32665, 32647, 32628, +32608, 32586, 32562, 32538, 32512, +32484, 32455, 32425, 32393, 32360, +32326, 32290, 32253, 32214, 32174, +32133, 32090, 32046, 32001, 31954, +31906, 31856, 31805, 31753, 31700, +31645, 31588, 31530, 31471, 31411, +31349, 31286, 31222, 31156, 31089, +31020, 30951, 30880, 30807, 30733, +30658, 30582, 30504, 30425, 30345, +30263, 30181, 30096, 30011, 29924, +29836, 29747, 29656, 29564, 29471, +29377, 29281, 29184, 29086, 28987, +28886, 28784, 28681, 28577, 28471, +28365, 28257, 28147, 28037, 27925, +27812, 27698, 27583, 27467, 27349, +27231, 27111, 26990, 26868, 26744, +26620, 26494, 26367, 26239, 26110, +25980, 25849, 25717, 25583, 25449, +25313, 25176, 25038, 24900, 24760, +24619, 24477, 24333, 24189, 24044, +23898, 23751, 23602, 23453, 23303, +23152, 22999, 22846, 22692, 22537, +22380, 22223, 22065, 21906, 21746, +21585, 21423, 21261, 21097, 20933, +20767, 20601, 20434, 20265, 20096, +19927, 19756, 19584, 19412, 19239, +19065, 18890, 18714, 18538, 18361, +18183, 18004, 17824, 17644, 17463, +17281, 17098, 16915, 16731, 16546, +16361, 16175, 15988, 15800, 15612, +15423, 15234, 15043, 14852, 14661, +14469, 14276, 14083, 13889, 13694, +13499, 13303, 13107, 12910, 12713, +12515, 12317, 12118, 11918, 11718, +11517, 11316, 11115, 10913, 10710, +10508, 10304, 10100, 9896, 9691, +9486, 9281, 9075, 8869, 8662, +8455, 8248, 8040, 7832, 7623, +7415, 7206, 6996, 6787, 6577, +6366, 6156, 5945, 5734, 5523, +5311, 5100, 4888, 4675, 4463, +4251, 4038, 3825, 3612, 3399, +3185, 2972, 2758, 2544, 2330, +2116, 1902, 1688, 1474, 1260, +1045, 831, 617, 402, 188, +-27, -241, -456, -670, -885, +-1099, -1313, -1528, -1742, -1956, +-2170, -2384, -2598, -2811, -3025, +-3239, -3452, -3665, -3878, -4091, +-4304, -4516, -4728, -4941, -5153, +-5364, -5576, -5787, -5998, -6209, +-6419, -6629, -6839, -7049, -7258, +-7467, -7676, -7884, -8092, -8300, +-8507, -8714, -8920, -9127, -9332, +-9538, -9743, -9947, -10151, -10355, +-10558, -10761, -10963, -11165, -11367, +-11568, -11768, -11968, -12167, -12366, +-12565, -12762, -12960, -13156, -13352, +-13548, -13743, -13937, -14131, -14324, +-14517, -14709, -14900, -15091, -15281, +-15470, -15659, -15847, -16035, -16221, +-16407, -16593, -16777, -16961, -17144, +-17326, -17508, -17689, -17869, -18049, +-18227, -18405, -18582, -18758, -18934, +-19108, -19282, -19455, -19627, -19799, +-19969, -20139, -20308, -20475, -20642, +-20809, -20974, -21138, -21301, -21464, +-21626, -21786, -21946, -22105, -22263, +-22420, -22575, -22730, -22884, -23037, +-23189, -23340, -23490, -23640, -23788, +-23935, -24080, -24225, -24369, -24512, +-24654, -24795, -24934, -25073, -25211, +-25347, -25482, -25617, -25750, -25882, +-26013, -26143, -26272, -26399, -26526, +-26651, -26775, -26898, -27020, -27141, +-27260, -27379, -27496, -27612, -27727, +-27841, -27953, -28065, -28175, -28284, +-28391, -28498, -28603, -28707, -28810, +-28911, -29012, -29111, -29209, -29305, +-29401, -29495, -29587, -29679, -29769, +-29858, -29946, -30032, -30118, -30201, +-30284, -30365, -30445, -30524, -30601, +-30677, -30752, -30825, -30897, -30968, +-31038, -31106, -31172, -31238, -31302, +-31365, -31426, -31486, -31545, -31602, +-31658, -31713, -31766, -31818, -31869, +-31918, -31966, -32012, -32058, -32101, +-32144, -32185, -32224, -32262, -32299, +-32335, -32369, -32401, -32433, -32463, +-32491, -32518, -32544, -32568, -32591, +-32613, -32633, -32652, -32669, -32685, +-32700, -32713, -32724, -32735, -32744, +-32751, -32757, -32762, -32766, -32767, +32767, 32764, 32755, 32741, 32720, +32694, 32663, 32626, 32583, 32535, +32481, 32421, 32356, 32286, 32209, +32128, 32041, 31948, 31850, 31747, +31638, 31523, 31403, 31278, 31148, +31012, 30871, 30724, 30572, 30415, +30253, 30086, 29913, 29736, 29553, +29365, 29172, 28974, 28771, 28564, +28351, 28134, 27911, 27684, 27452, +27216, 26975, 26729, 26478, 26223, +25964, 25700, 25432, 25159, 24882, +24601, 24315, 24026, 23732, 23434, +23133, 22827, 22517, 22204, 21886, +21565, 21240, 20912, 20580, 20244, +19905, 19563, 19217, 18868, 18516, +18160, 17802, 17440, 17075, 16708, +16338, 15964, 15588, 15210, 14829, +14445, 14059, 13670, 13279, 12886, +12490, 12093, 11693, 11291, 10888, +10482, 10075, 9666, 9255, 8843, +8429, 8014, 7597, 7180, 6760, +6340, 5919, 5496, 5073, 4649, +4224, 3798, 3372, 2945, 2517, +2090, 1661, 1233, 804, 375, +-54, -483, -911, -1340, -1768, +-2197, -2624, -3052, -3479, -3905, +-4330, -4755, -5179, -5602, -6024, +-6445, -6865, -7284, -7702, -8118, +-8533, -8946, -9358, -9768, -10177, +-10584, -10989, -11392, -11793, -12192, +-12589, -12984, -13377, -13767, -14155, +-14541, -14924, -15305, -15683, -16058, +-16430, -16800, -17167, -17531, -17892, +-18249, -18604, -18956, -19304, -19649, +-19990, -20329, -20663, -20994, -21322, +-21646, -21966, -22282, -22595, -22904, +-23208, -23509, -23806, -24099, -24387, +-24672, -24952, -25228, -25499, -25766, +-26029, -26288, -26541, -26791, -27035, +-27275, -27511, -27741, -27967, -28188, +-28405, -28616, -28823, -29024, -29221, +-29412, -29599, -29780, -29957, -30128, +-30294, -30455, -30611, -30761, -30906, +-31046, -31181, -31310, -31434, -31552, +-31665, -31773, -31875, -31972, -32063, +-32149, -32229, -32304, -32373, -32437, +-32495, -32547, -32594, -32635, -32671, +-32701, -32726, -32745, -32758, -32766, +32767, 32754, 32717, 32658, 32577, +32473, 32348, 32200, 32029, 31837, +31624, 31388, 31131, 30853, 30553, +30232, 29891, 29530, 29148, 28746, +28324, 27883, 27423, 26944, 26447, +25931, 25398, 24847, 24279, 23695, +23095, 22478, 21846, 21199, 20538, +19863, 19174, 18472, 17757, 17030, +16291, 15541, 14781, 14010, 13230, +12441, 11643, 10837, 10024, 9204, +8377, 7545, 6708, 5866, 5020, +4171, 3319, 2464, 1608, 751, +-107, -965, -1822, -2678, -3532, +-4383, -5232, -6077, -6918, -7754, +-8585, -9409, -10228, -11039, -11843, +-12639, -13426, -14204, -14972, -15730, +-16477, -17213, -17937, -18648, -19347, +-20033, -20705, -21363, -22006, -22634, +-23246, -23843, -24423, -24986, -25533, +-26062, -26573, -27066, -27540, -27995, +-28431, -28848, -29245, -29622, -29979, +-30315, -30630, -30924, -31197, -31449, +-31679, -31887, -32074, -32239, -32381, +-32501, -32600, -32675, -32729, -32759, +}; +#endif + +static const CELTMode mode48000_960_120 = { +48000, /* Fs */ +120, /* overlap */ +21, /* nbEBands */ +21, /* effEBands */ +{27853, 0, 4096, 8192, }, /* preemph */ +eband5ms, /* eBands */ +3, /* maxLM */ +8, /* nbShortMdcts */ +120, /* shortMdctSize */ +11, /* nbAllocVectors */ +band_allocation, /* allocVectors */ +logN400, /* logN */ +window120, /* window */ +{1920, 3, {&fft_state48000_960_0, &fft_state48000_960_1, &fft_state48000_960_2, &fft_state48000_960_3, }, mdct_twiddles960}, /* mdct */ +{392, cache_index50, cache_bits50, cache_caps50}, /* cache */ +}; + +/* List of all the available modes */ +#define TOTAL_MODES 1 +static const CELTMode * const static_mode_list[TOTAL_MODES] = { +&mode48000_960_120, +}; diff --git a/libopus/celt/static_modes_fixed_arm_ne10.h b/libopus/celt/static_modes_fixed_arm_ne10.h new file mode 100644 index 0000000..7623092 --- /dev/null +++ b/libopus/celt/static_modes_fixed_arm_ne10.h @@ -0,0 +1,388 @@ +/* The contents of this file was automatically generated by + * dump_mode_arm_ne10.c with arguments: 48000 960 + * It contains static definitions for some pre-defined modes. */ +#include <NE10_types.h> + +#ifndef NE10_FFT_PARAMS48000_960 +#define NE10_FFT_PARAMS48000_960 +static const ne10_int32_t ne10_factors_480[64] = { +4, 40, 4, 30, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_240[64] = { +3, 20, 4, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_120[64] = { +3, 10, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_60[64] = { +2, 5, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_fft_cpx_int32_t ne10_twiddles_480[480] = { +{0,0}, {2147483647,0}, {2147483647,0}, +{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394}, +{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496}, +{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096}, +{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152}, +{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313}, +{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424}, +{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496}, +{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268}, +{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785}, +{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172}, +{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682}, +{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313}, +{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450}, +{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067}, +{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277}, +{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424}, +{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771}, +{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994}, +{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593}, +{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968}, +{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851}, +{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394}, +{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959}, +{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516}, +{-94,-2147483647}, {-224473265,-2135719496}, {-446487060,-2100555955}, +{-663609049,-2042378281}, {-873460398,-1961823883}, {-1073741932,-1859775330}, +{-1262259116,-1737350839}, {-1436947137,-1595891268}, {-1595891628,-1436946738}, +{-1737350854,-1262259096}, {-1859775343,-1073741910}, {-1961823997,-873460141}, +{-2042378447,-663608538}, {-2100556013,-446486785}, {-2135719499,-224473240}, +{2147483647,0}, {2121044558,-335940465}, {2042378310,-663608960}, +{1913421927,-974937199}, {1737350743,-1262259248}, {1518500216,-1518500282}, +{1262259172,-1737350799}, {974937230,-1913421912}, {663608871,-2042378339}, +{335940246,-2121044593}, {-94,-2147483647}, {-335940431,-2121044564}, +{-663609049,-2042378281}, {-974937397,-1913421827}, {-1262259116,-1737350839}, +{-1518500258,-1518500240}, {-1737350854,-1262259096}, {-1913422071,-974936918}, +{-2042378447,-663608538}, {-2121044568,-335940406}, {-2147483647,188}, +{-2121044509,335940777}, {-2042378331,663608895}, {-1913421900,974937252}, +{-1737350633,1262259400}, {-1518499993,1518500506}, {-1262258813,1737351059}, +{-974936606,1913422229}, {-663609179,2042378239}, {-335940566,2121044542}, +{2147483647,0}, {2147299667,-28109693}, {2146747758,-56214570}, +{2145828015,-84309815}, {2144540595,-112390613}, {2142885719,-140452154}, +{2140863671,-168489630}, {2138474797,-196498235}, {2135719506,-224473172}, +{2132598271,-252409646}, {2129111626,-280302871}, {2125260168,-308148068}, +{2121044558,-335940465}, {2116465518,-363675300}, {2111523833,-391347822}, +{2106220349,-418953288}, {2100555974,-446486968}, {2094531681,-473944146}, +{2088148500,-501320115}, {2081407525,-528610186}, {2074309912,-555809682}, +{2066856885,-582913912}, {2059049696,-609918325}, {2050889698,-636818231}, +{2042378310,-663608960}, {2033516972,-690285983}, {2024307180,-716844791}, +{2014750533,-743280770}, {2004848691,-769589332}, {1994603329,-795766029}, +{1984016179,-821806435}, {1973089077,-847706028}, {1961823921,-873460313}, +{1950222618,-899064934}, {1938287127,-924515564}, {1926019520,-949807783}, +{1913421927,-974937199}, {1900496481,-999899565}, {1887245364,-1024690661}, +{1873670877,-1049306180}, {1859775377,-1073741851}, {1845561215,-1097993541}, +{1831030826,-1122057097}, {1816186632,-1145928502}, {1801031311,-1169603450}, +{1785567394,-1193077993}, {1769797456,-1216348214}, {1753724345,-1239409914}, +{1737350743,-1262259248}, {1720679456,-1284892300}, {1703713340,-1307305194}, +{1686455222,-1329494189}, {1668908218,-1351455280}, {1651075255,-1373184807}, +{1632959307,-1394679144}, {1614563642,-1415934412}, {1595891331,-1436947067}, +{1576945572,-1457713510}, {1557729613,-1478230181}, {1538246655,-1498493658}, +{1518500216,-1518500282}, {1498493590,-1538246721}, {1478230113,-1557729677}, +{1457713441,-1576945636}, {1436946998,-1595891394}, {1415934341,-1614563704}, +{1394679073,-1632959368}, {1373184735,-1651075315}, {1351455207,-1668908277}, +{1329494115,-1686455280}, {1307305120,-1703713397}, {1284892225,-1720679512}, +{1262259172,-1737350799}, {1239409837,-1753724400}, {1216348136,-1769797510}, +{1193077915,-1785567446}, {1169603371,-1801031362}, {1145928423,-1816186682}, +{1122057017,-1831030875}, {1097993571,-1845561197}, {1073741769,-1859775424}, +{1049305987,-1873670985}, {1024690635,-1887245378}, {999899482,-1900496524}, +{974937230,-1913421912}, {949807699,-1926019561}, {924515422,-1938287195}, +{899064965,-1950222603}, {873460227,-1961823959}, {847705824,-1973089164}, +{821806407,-1984016190}, {795765941,-1994603364}, {769589125,-2004848771}, +{743280682,-2014750566}, {716844642,-2024307233}, {690286016,-2033516961}, +{663608871,-2042378339}, {636818019,-2050889764}, {609918296,-2059049705}, +{582913822,-2066856911}, {555809715,-2074309903}, {528610126,-2081407540}, +{501319962,-2088148536}, {473944148,-2094531680}, {446486876,-2100555994}, +{418953102,-2106220386}, {391347792,-2111523838}, {363675176,-2116465540}, +{335940246,-2121044593}, {308148006,-2125260177}, {280302715,-2129111646}, +{252409648,-2132598271}, {224473078,-2135719516}, {196498046,-2138474814}, +{168489600,-2140863674}, {140452029,-2142885728}, {112390647,-2144540593}, +{84309753,-2145828017}, {56214412,-2146747762}, {28109695,-2147299667}, +{2147483647,0}, {2146747758,-56214570}, {2144540595,-112390613}, +{2140863671,-168489630}, {2135719506,-224473172}, {2129111626,-280302871}, +{2121044558,-335940465}, {2111523833,-391347822}, {2100555974,-446486968}, +{2088148500,-501320115}, {2074309912,-555809682}, {2059049696,-609918325}, +{2042378310,-663608960}, {2024307180,-716844791}, {2004848691,-769589332}, +{1984016179,-821806435}, {1961823921,-873460313}, {1938287127,-924515564}, +{1913421927,-974937199}, {1887245364,-1024690661}, {1859775377,-1073741851}, +{1831030826,-1122057097}, {1801031311,-1169603450}, {1769797456,-1216348214}, +{1737350743,-1262259248}, {1703713340,-1307305194}, {1668908218,-1351455280}, +{1632959307,-1394679144}, {1595891331,-1436947067}, {1557729613,-1478230181}, +{1518500216,-1518500282}, {1478230113,-1557729677}, {1436946998,-1595891394}, +{1394679073,-1632959368}, {1351455207,-1668908277}, {1307305120,-1703713397}, +{1262259172,-1737350799}, {1216348136,-1769797510}, {1169603371,-1801031362}, +{1122057017,-1831030875}, {1073741769,-1859775424}, {1024690635,-1887245378}, +{974937230,-1913421912}, {924515422,-1938287195}, {873460227,-1961823959}, +{821806407,-1984016190}, {769589125,-2004848771}, {716844642,-2024307233}, +{663608871,-2042378339}, {609918296,-2059049705}, {555809715,-2074309903}, +{501319962,-2088148536}, {446486876,-2100555994}, {391347792,-2111523838}, +{335940246,-2121044593}, {280302715,-2129111646}, {224473078,-2135719516}, +{168489600,-2140863674}, {112390647,-2144540593}, {56214412,-2146747762}, +{-94,-2147483647}, {-56214600,-2146747757}, {-112390835,-2144540584}, +{-168489787,-2140863659}, {-224473265,-2135719496}, {-280302901,-2129111622}, +{-335940431,-2121044564}, {-391347977,-2111523804}, {-446487060,-2100555955}, +{-501320144,-2088148493}, {-555809896,-2074309855}, {-609918476,-2059049651}, +{-663609049,-2042378281}, {-716844819,-2024307170}, {-769589300,-2004848703}, +{-821806581,-1984016118}, {-873460398,-1961823883}, {-924515591,-1938287114}, +{-974937397,-1913421827}, {-1024690575,-1887245411}, {-1073741932,-1859775330}, +{-1122057395,-1831030643}, {-1169603421,-1801031330}, {-1216348291,-1769797403}, +{-1262259116,-1737350839}, {-1307305268,-1703713283}, {-1351455453,-1668908078}, +{-1394679021,-1632959413}, {-1436947137,-1595891268}, {-1478230435,-1557729372}, +{-1518500258,-1518500240}, {-1557729742,-1478230045}, {-1595891628,-1436946738}, +{-1632959429,-1394679001}, {-1668908417,-1351455035}, {-1703713298,-1307305248}, +{-1737350854,-1262259096}, {-1769797708,-1216347848}, {-1801031344,-1169603400}, +{-1831030924,-1122056937}, {-1859775343,-1073741910}, {-1887245423,-1024690552}, +{-1913422071,-974936918}, {-1938287125,-924515568}, {-1961823997,-873460141}, +{-1984016324,-821806084}, {-2004848713,-769589276}, {-2024307264,-716844553}, +{-2042378447,-663608538}, {-2059049731,-609918206}, {-2074309994,-555809377}, +{-2088148499,-501320119}, {-2100556013,-446486785}, {-2111523902,-391347448}, +{-2121044568,-335940406}, {-2129111659,-280302621}, {-2135719499,-224473240}, +{-2140863681,-168489506}, {-2144540612,-112390298}, {-2146747758,-56214574}, +{2147483647,0}, {2145828015,-84309815}, {2140863671,-168489630}, +{2132598271,-252409646}, {2121044558,-335940465}, {2106220349,-418953288}, +{2088148500,-501320115}, {2066856885,-582913912}, {2042378310,-663608960}, +{2014750533,-743280770}, {1984016179,-821806435}, {1950222618,-899064934}, +{1913421927,-974937199}, {1873670877,-1049306180}, {1831030826,-1122057097}, +{1785567394,-1193077993}, {1737350743,-1262259248}, {1686455222,-1329494189}, +{1632959307,-1394679144}, {1576945572,-1457713510}, {1518500216,-1518500282}, +{1457713441,-1576945636}, {1394679073,-1632959368}, {1329494115,-1686455280}, +{1262259172,-1737350799}, {1193077915,-1785567446}, {1122057017,-1831030875}, +{1049305987,-1873670985}, {974937230,-1913421912}, {899064965,-1950222603}, +{821806407,-1984016190}, {743280682,-2014750566}, {663608871,-2042378339}, +{582913822,-2066856911}, {501319962,-2088148536}, {418953102,-2106220386}, +{335940246,-2121044593}, {252409648,-2132598271}, {168489600,-2140863674}, +{84309753,-2145828017}, {-94,-2147483647}, {-84309940,-2145828010}, +{-168489787,-2140863659}, {-252409834,-2132598249}, {-335940431,-2121044564}, +{-418953286,-2106220349}, {-501320144,-2088148493}, {-582914003,-2066856860}, +{-663609049,-2042378281}, {-743280858,-2014750501}, {-821806581,-1984016118}, +{-899065136,-1950222525}, {-974937397,-1913421827}, {-1049306374,-1873670768}, +{-1122057395,-1831030643}, {-1193078284,-1785567199}, {-1262259116,-1737350839}, +{-1329494061,-1686455323}, {-1394679021,-1632959413}, {-1457713485,-1576945595}, +{-1518500258,-1518500240}, {-1576945613,-1457713466}, {-1632959429,-1394679001}, +{-1686455338,-1329494041}, {-1737350854,-1262259096}, {-1785567498,-1193077837}, +{-1831030924,-1122056937}, {-1873671031,-1049305905}, {-1913422071,-974936918}, +{-1950222750,-899064648}, {-1984016324,-821806084}, {-2014750687,-743280354}, +{-2042378447,-663608538}, {-2066856867,-582913978}, {-2088148499,-501320119}, +{-2106220354,-418953261}, {-2121044568,-335940406}, {-2132598282,-252409555}, +{-2140863681,-168489506}, {-2145828021,-84309659}, {-2147483647,188}, +{-2145828006,84310034}, {-2140863651,168489881}, {-2132598237,252409928}, +{-2121044509,335940777}, {-2106220281,418953629}, {-2088148411,501320484}, +{-2066856765,582914339}, {-2042378331,663608895}, {-2014750557,743280706}, +{-1984016181,821806431}, {-1950222593,899064989}, {-1913421900,974937252}, +{-1873670848,1049306232}, {-1831030728,1122057257}, {-1785567289,1193078149}, +{-1737350633,1262259400}, {-1686455106,1329494336}, {-1632959185,1394679287}, +{-1576945358,1457713742}, {-1518499993,1518500506}, {-1457713209,1576945850}, +{-1394678735,1632959656}, {-1329493766,1686455555}, {-1262258813,1737351059}, +{-1193077546,1785567692}, {-1122056638,1831031107}, {-1049305599,1873671202}, +{-974936606,1913422229}, {-899064330,1950222896}, {-821805761,1984016458}, +{-743280025,2014750808}, {-663609179,2042378239}, {-582914134,2066856823}, +{-501320277,2088148461}, {-418953420,2106220322}, {-335940566,2121044542}, +{-252409716,2132598263}, {-168489668,2140863668}, {-84309821,2145828015}, +}; +static const ne10_fft_cpx_int32_t ne10_twiddles_240[240] = { +{0,0}, {2147483647,0}, {2147483647,0}, +{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394}, +{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496}, +{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096}, +{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152}, +{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968}, +{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851}, +{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394}, +{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959}, +{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516}, +{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313}, +{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424}, +{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496}, +{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268}, +{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785}, +{2147483647,0}, {2042378310,-663608960}, {1737350743,-1262259248}, +{1262259172,-1737350799}, {663608871,-2042378339}, {-94,-2147483647}, +{-663609049,-2042378281}, {-1262259116,-1737350839}, {-1737350854,-1262259096}, +{-2042378447,-663608538}, {-2147483647,188}, {-2042378331,663608895}, +{-1737350633,1262259400}, {-1262258813,1737351059}, {-663609179,2042378239}, +{2147483647,0}, {2146747758,-56214570}, {2144540595,-112390613}, +{2140863671,-168489630}, {2135719506,-224473172}, {2129111626,-280302871}, +{2121044558,-335940465}, {2111523833,-391347822}, {2100555974,-446486968}, +{2088148500,-501320115}, {2074309912,-555809682}, {2059049696,-609918325}, +{2042378310,-663608960}, {2024307180,-716844791}, {2004848691,-769589332}, +{1984016179,-821806435}, {1961823921,-873460313}, {1938287127,-924515564}, +{1913421927,-974937199}, {1887245364,-1024690661}, {1859775377,-1073741851}, +{1831030826,-1122057097}, {1801031311,-1169603450}, {1769797456,-1216348214}, +{1737350743,-1262259248}, {1703713340,-1307305194}, {1668908218,-1351455280}, +{1632959307,-1394679144}, {1595891331,-1436947067}, {1557729613,-1478230181}, +{1518500216,-1518500282}, {1478230113,-1557729677}, {1436946998,-1595891394}, +{1394679073,-1632959368}, {1351455207,-1668908277}, {1307305120,-1703713397}, +{1262259172,-1737350799}, {1216348136,-1769797510}, {1169603371,-1801031362}, +{1122057017,-1831030875}, {1073741769,-1859775424}, {1024690635,-1887245378}, +{974937230,-1913421912}, {924515422,-1938287195}, {873460227,-1961823959}, +{821806407,-1984016190}, {769589125,-2004848771}, {716844642,-2024307233}, +{663608871,-2042378339}, {609918296,-2059049705}, {555809715,-2074309903}, +{501319962,-2088148536}, {446486876,-2100555994}, {391347792,-2111523838}, +{335940246,-2121044593}, {280302715,-2129111646}, {224473078,-2135719516}, +{168489600,-2140863674}, {112390647,-2144540593}, {56214412,-2146747762}, +{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172}, +{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682}, +{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313}, +{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450}, +{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067}, +{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277}, +{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424}, +{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771}, +{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994}, +{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593}, +{-94,-2147483647}, {-112390835,-2144540584}, {-224473265,-2135719496}, +{-335940431,-2121044564}, {-446487060,-2100555955}, {-555809896,-2074309855}, +{-663609049,-2042378281}, {-769589300,-2004848703}, {-873460398,-1961823883}, +{-974937397,-1913421827}, {-1073741932,-1859775330}, {-1169603421,-1801031330}, +{-1262259116,-1737350839}, {-1351455453,-1668908078}, {-1436947137,-1595891268}, +{-1518500258,-1518500240}, {-1595891628,-1436946738}, {-1668908417,-1351455035}, +{-1737350854,-1262259096}, {-1801031344,-1169603400}, {-1859775343,-1073741910}, +{-1913422071,-974936918}, {-1961823997,-873460141}, {-2004848713,-769589276}, +{-2042378447,-663608538}, {-2074309994,-555809377}, {-2100556013,-446486785}, +{-2121044568,-335940406}, {-2135719499,-224473240}, {-2144540612,-112390298}, +{2147483647,0}, {2140863671,-168489630}, {2121044558,-335940465}, +{2088148500,-501320115}, {2042378310,-663608960}, {1984016179,-821806435}, +{1913421927,-974937199}, {1831030826,-1122057097}, {1737350743,-1262259248}, +{1632959307,-1394679144}, {1518500216,-1518500282}, {1394679073,-1632959368}, +{1262259172,-1737350799}, {1122057017,-1831030875}, {974937230,-1913421912}, +{821806407,-1984016190}, {663608871,-2042378339}, {501319962,-2088148536}, +{335940246,-2121044593}, {168489600,-2140863674}, {-94,-2147483647}, +{-168489787,-2140863659}, {-335940431,-2121044564}, {-501320144,-2088148493}, +{-663609049,-2042378281}, {-821806581,-1984016118}, {-974937397,-1913421827}, +{-1122057395,-1831030643}, {-1262259116,-1737350839}, {-1394679021,-1632959413}, +{-1518500258,-1518500240}, {-1632959429,-1394679001}, {-1737350854,-1262259096}, +{-1831030924,-1122056937}, {-1913422071,-974936918}, {-1984016324,-821806084}, +{-2042378447,-663608538}, {-2088148499,-501320119}, {-2121044568,-335940406}, +{-2140863681,-168489506}, {-2147483647,188}, {-2140863651,168489881}, +{-2121044509,335940777}, {-2088148411,501320484}, {-2042378331,663608895}, +{-1984016181,821806431}, {-1913421900,974937252}, {-1831030728,1122057257}, +{-1737350633,1262259400}, {-1632959185,1394679287}, {-1518499993,1518500506}, +{-1394678735,1632959656}, {-1262258813,1737351059}, {-1122056638,1831031107}, +{-974936606,1913422229}, {-821805761,1984016458}, {-663609179,2042378239}, +{-501320277,2088148461}, {-335940566,2121044542}, {-168489668,2140863668}, +}; +static const ne10_fft_cpx_int32_t ne10_twiddles_120[120] = { +{0,0}, {2147483647,0}, {2147483647,0}, +{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394}, +{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496}, +{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096}, +{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152}, +{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313}, +{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424}, +{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496}, +{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268}, +{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785}, +{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172}, +{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682}, +{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313}, +{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450}, +{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067}, +{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277}, +{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424}, +{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771}, +{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994}, +{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593}, +{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968}, +{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851}, +{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394}, +{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959}, +{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516}, +{-94,-2147483647}, {-224473265,-2135719496}, {-446487060,-2100555955}, +{-663609049,-2042378281}, {-873460398,-1961823883}, {-1073741932,-1859775330}, +{-1262259116,-1737350839}, {-1436947137,-1595891268}, {-1595891628,-1436946738}, +{-1737350854,-1262259096}, {-1859775343,-1073741910}, {-1961823997,-873460141}, +{-2042378447,-663608538}, {-2100556013,-446486785}, {-2135719499,-224473240}, +{2147483647,0}, {2121044558,-335940465}, {2042378310,-663608960}, +{1913421927,-974937199}, {1737350743,-1262259248}, {1518500216,-1518500282}, +{1262259172,-1737350799}, {974937230,-1913421912}, {663608871,-2042378339}, +{335940246,-2121044593}, {-94,-2147483647}, {-335940431,-2121044564}, +{-663609049,-2042378281}, {-974937397,-1913421827}, {-1262259116,-1737350839}, +{-1518500258,-1518500240}, {-1737350854,-1262259096}, {-1913422071,-974936918}, +{-2042378447,-663608538}, {-2121044568,-335940406}, {-2147483647,188}, +{-2121044509,335940777}, {-2042378331,663608895}, {-1913421900,974937252}, +{-1737350633,1262259400}, {-1518499993,1518500506}, {-1262258813,1737351059}, +{-974936606,1913422229}, {-663609179,2042378239}, {-335940566,2121044542}, +}; +static const ne10_fft_cpx_int32_t ne10_twiddles_60[60] = { +{0,0}, {2147483647,0}, {2147483647,0}, +{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394}, +{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496}, +{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096}, +{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152}, +{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968}, +{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851}, +{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394}, +{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959}, +{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516}, +{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313}, +{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424}, +{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496}, +{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268}, +{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785}, +{2147483647,0}, {2042378310,-663608960}, {1737350743,-1262259248}, +{1262259172,-1737350799}, {663608871,-2042378339}, {-94,-2147483647}, +{-663609049,-2042378281}, {-1262259116,-1737350839}, {-1737350854,-1262259096}, +{-2042378447,-663608538}, {-2147483647,188}, {-2042378331,663608895}, +{-1737350633,1262259400}, {-1262258813,1737351059}, {-663609179,2042378239}, +}; +static const ne10_fft_state_int32_t ne10_fft_state_int32_t_480 = { +120, +(ne10_int32_t *)ne10_factors_480, +(ne10_fft_cpx_int32_t *)ne10_twiddles_480, +NULL, +(ne10_fft_cpx_int32_t *)&ne10_twiddles_480[120], +}; +static const arch_fft_state cfg_arch_480 = { +1, +(void *)&ne10_fft_state_int32_t_480, +}; + +static const ne10_fft_state_int32_t ne10_fft_state_int32_t_240 = { +60, +(ne10_int32_t *)ne10_factors_240, +(ne10_fft_cpx_int32_t *)ne10_twiddles_240, +NULL, +(ne10_fft_cpx_int32_t *)&ne10_twiddles_240[60], +}; +static const arch_fft_state cfg_arch_240 = { +1, +(void *)&ne10_fft_state_int32_t_240, +}; + +static const ne10_fft_state_int32_t ne10_fft_state_int32_t_120 = { +30, +(ne10_int32_t *)ne10_factors_120, +(ne10_fft_cpx_int32_t *)ne10_twiddles_120, +NULL, +(ne10_fft_cpx_int32_t *)&ne10_twiddles_120[30], +}; +static const arch_fft_state cfg_arch_120 = { +1, +(void *)&ne10_fft_state_int32_t_120, +}; + +static const ne10_fft_state_int32_t ne10_fft_state_int32_t_60 = { +15, +(ne10_int32_t *)ne10_factors_60, +(ne10_fft_cpx_int32_t *)ne10_twiddles_60, +NULL, +(ne10_fft_cpx_int32_t *)&ne10_twiddles_60[15], +}; +static const arch_fft_state cfg_arch_60 = { +1, +(void *)&ne10_fft_state_int32_t_60, +}; + +#endif /* end NE10_FFT_PARAMS48000_960 */ diff --git a/libopus/celt/static_modes_float.h b/libopus/celt/static_modes_float.h new file mode 100644 index 0000000..e102a38 --- /dev/null +++ b/libopus/celt/static_modes_float.h @@ -0,0 +1,888 @@ +/* The contents of this file was automatically generated by dump_modes.c + with arguments: 48000 960 + It contains static definitions for some pre-defined modes. */ +#include "modes.h" +#include "rate.h" + +#ifdef HAVE_ARM_NE10 +#define OVERRIDE_FFT 1 +#include "static_modes_float_arm_ne10.h" +#endif + +#ifndef DEF_WINDOW120 +#define DEF_WINDOW120 +static const opus_val16 window120[120] = { +6.7286966e-05f, 0.00060551348f, 0.0016815970f, 0.0032947962f, 0.0054439943f, +0.0081276923f, 0.011344001f, 0.015090633f, 0.019364886f, 0.024163635f, +0.029483315f, 0.035319905f, 0.041668911f, 0.048525347f, 0.055883718f, +0.063737999f, 0.072081616f, 0.080907428f, 0.090207705f, 0.099974111f, +0.11019769f, 0.12086883f, 0.13197729f, 0.14351214f, 0.15546177f, +0.16781389f, 0.18055550f, 0.19367290f, 0.20715171f, 0.22097682f, +0.23513243f, 0.24960208f, 0.26436860f, 0.27941419f, 0.29472040f, +0.31026818f, 0.32603788f, 0.34200931f, 0.35816177f, 0.37447407f, +0.39092462f, 0.40749142f, 0.42415215f, 0.44088423f, 0.45766484f, +0.47447104f, 0.49127978f, 0.50806798f, 0.52481261f, 0.54149077f, +0.55807973f, 0.57455701f, 0.59090049f, 0.60708841f, 0.62309951f, +0.63891306f, 0.65450896f, 0.66986776f, 0.68497077f, 0.69980010f, +0.71433873f, 0.72857055f, 0.74248043f, 0.75605424f, 0.76927895f, +0.78214257f, 0.79463430f, 0.80674445f, 0.81846456f, 0.82978733f, +0.84070669f, 0.85121779f, 0.86131698f, 0.87100183f, 0.88027111f, +0.88912479f, 0.89756398f, 0.90559094f, 0.91320904f, 0.92042270f, +0.92723738f, 0.93365955f, 0.93969656f, 0.94535671f, 0.95064907f, +0.95558353f, 0.96017067f, 0.96442171f, 0.96834849f, 0.97196334f, +0.97527906f, 0.97830883f, 0.98106616f, 0.98356480f, 0.98581869f, +0.98784191f, 0.98964856f, 0.99125274f, 0.99266849f, 0.99390969f, +0.99499004f, 0.99592297f, 0.99672162f, 0.99739874f, 0.99796667f, +0.99843728f, 0.99882195f, 0.99913147f, 0.99937606f, 0.99956527f, +0.99970802f, 0.99981248f, 0.99988613f, 0.99993565f, 0.99996697f, +0.99998518f, 0.99999457f, 0.99999859f, 0.99999982f, 1.0000000f, +}; +#endif + +#ifndef DEF_LOGN400 +#define DEF_LOGN400 +static const opus_int16 logN400[21] = { +0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, 16, 16, 16, 21, 21, 24, 29, 34, 36, }; +#endif + +#ifndef DEF_PULSE_CACHE50 +#define DEF_PULSE_CACHE50 +static const opus_int16 cache_index50[105] = { +-1, -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 41, 41, 41, +82, 82, 123, 164, 200, 222, 0, 0, 0, 0, 0, 0, 0, 0, 41, +41, 41, 41, 123, 123, 123, 164, 164, 240, 266, 283, 295, 41, 41, 41, +41, 41, 41, 41, 41, 123, 123, 123, 123, 240, 240, 240, 266, 266, 305, +318, 328, 336, 123, 123, 123, 123, 123, 123, 123, 123, 240, 240, 240, 240, +305, 305, 305, 318, 318, 343, 351, 358, 364, 240, 240, 240, 240, 240, 240, +240, 240, 305, 305, 305, 305, 343, 343, 343, 351, 351, 370, 376, 382, 387, +}; +static const unsigned char cache_bits50[392] = { +40, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 40, 15, 23, 28, +31, 34, 36, 38, 39, 41, 42, 43, 44, 45, 46, 47, 47, 49, 50, +51, 52, 53, 54, 55, 55, 57, 58, 59, 60, 61, 62, 63, 63, 65, +66, 67, 68, 69, 70, 71, 71, 40, 20, 33, 41, 48, 53, 57, 61, +64, 66, 69, 71, 73, 75, 76, 78, 80, 82, 85, 87, 89, 91, 92, +94, 96, 98, 101, 103, 105, 107, 108, 110, 112, 114, 117, 119, 121, 123, +124, 126, 128, 40, 23, 39, 51, 60, 67, 73, 79, 83, 87, 91, 94, +97, 100, 102, 105, 107, 111, 115, 118, 121, 124, 126, 129, 131, 135, 139, +142, 145, 148, 150, 153, 155, 159, 163, 166, 169, 172, 174, 177, 179, 35, +28, 49, 65, 78, 89, 99, 107, 114, 120, 126, 132, 136, 141, 145, 149, +153, 159, 165, 171, 176, 180, 185, 189, 192, 199, 205, 211, 216, 220, 225, +229, 232, 239, 245, 251, 21, 33, 58, 79, 97, 112, 125, 137, 148, 157, +166, 174, 182, 189, 195, 201, 207, 217, 227, 235, 243, 251, 17, 35, 63, +86, 106, 123, 139, 152, 165, 177, 187, 197, 206, 214, 222, 230, 237, 250, +25, 31, 55, 75, 91, 105, 117, 128, 138, 146, 154, 161, 168, 174, 180, +185, 190, 200, 208, 215, 222, 229, 235, 240, 245, 255, 16, 36, 65, 89, +110, 128, 144, 159, 173, 185, 196, 207, 217, 226, 234, 242, 250, 11, 41, +74, 103, 128, 151, 172, 191, 209, 225, 241, 255, 9, 43, 79, 110, 138, +163, 186, 207, 227, 246, 12, 39, 71, 99, 123, 144, 164, 182, 198, 214, +228, 241, 253, 9, 44, 81, 113, 142, 168, 192, 214, 235, 255, 7, 49, +90, 127, 160, 191, 220, 247, 6, 51, 95, 134, 170, 203, 234, 7, 47, +87, 123, 155, 184, 212, 237, 6, 52, 97, 137, 174, 208, 240, 5, 57, +106, 151, 192, 231, 5, 59, 111, 158, 202, 243, 5, 55, 103, 147, 187, +224, 5, 60, 113, 161, 206, 248, 4, 65, 122, 175, 224, 4, 67, 127, +182, 234, }; +static const unsigned char cache_caps50[168] = { +224, 224, 224, 224, 224, 224, 224, 224, 160, 160, 160, 160, 185, 185, 185, +178, 178, 168, 134, 61, 37, 224, 224, 224, 224, 224, 224, 224, 224, 240, +240, 240, 240, 207, 207, 207, 198, 198, 183, 144, 66, 40, 160, 160, 160, +160, 160, 160, 160, 160, 185, 185, 185, 185, 193, 193, 193, 183, 183, 172, +138, 64, 38, 240, 240, 240, 240, 240, 240, 240, 240, 207, 207, 207, 207, +204, 204, 204, 193, 193, 180, 143, 66, 40, 185, 185, 185, 185, 185, 185, +185, 185, 193, 193, 193, 193, 193, 193, 193, 183, 183, 172, 138, 65, 39, +207, 207, 207, 207, 207, 207, 207, 207, 204, 204, 204, 204, 201, 201, 201, +188, 188, 176, 141, 66, 40, 193, 193, 193, 193, 193, 193, 193, 193, 193, +193, 193, 193, 194, 194, 194, 184, 184, 173, 139, 65, 39, 204, 204, 204, +204, 204, 204, 204, 204, 201, 201, 201, 201, 198, 198, 198, 187, 187, 175, +140, 66, 40, }; +#endif + +#ifndef FFT_TWIDDLES48000_960 +#define FFT_TWIDDLES48000_960 +static const kiss_twiddle_cpx fft_twiddles48000_960[480] = { +{1.0000000f, -0.0000000f}, {0.99991433f, -0.013089596f}, +{0.99965732f, -0.026176948f}, {0.99922904f, -0.039259816f}, +{0.99862953f, -0.052335956f}, {0.99785892f, -0.065403129f}, +{0.99691733f, -0.078459096f}, {0.99580493f, -0.091501619f}, +{0.99452190f, -0.10452846f}, {0.99306846f, -0.11753740f}, +{0.99144486f, -0.13052619f}, {0.98965139f, -0.14349262f}, +{0.98768834f, -0.15643447f}, {0.98555606f, -0.16934950f}, +{0.98325491f, -0.18223553f}, {0.98078528f, -0.19509032f}, +{0.97814760f, -0.20791169f}, {0.97534232f, -0.22069744f}, +{0.97236992f, -0.23344536f}, {0.96923091f, -0.24615329f}, +{0.96592583f, -0.25881905f}, {0.96245524f, -0.27144045f}, +{0.95881973f, -0.28401534f}, {0.95501994f, -0.29654157f}, +{0.95105652f, -0.30901699f}, {0.94693013f, -0.32143947f}, +{0.94264149f, -0.33380686f}, {0.93819134f, -0.34611706f}, +{0.93358043f, -0.35836795f}, {0.92880955f, -0.37055744f}, +{0.92387953f, -0.38268343f}, {0.91879121f, -0.39474386f}, +{0.91354546f, -0.40673664f}, {0.90814317f, -0.41865974f}, +{0.90258528f, -0.43051110f}, {0.89687274f, -0.44228869f}, +{0.89100652f, -0.45399050f}, {0.88498764f, -0.46561452f}, +{0.87881711f, -0.47715876f}, {0.87249601f, -0.48862124f}, +{0.86602540f, -0.50000000f}, {0.85940641f, -0.51129309f}, +{0.85264016f, -0.52249856f}, {0.84572782f, -0.53361452f}, +{0.83867057f, -0.54463904f}, {0.83146961f, -0.55557023f}, +{0.82412619f, -0.56640624f}, {0.81664156f, -0.57714519f}, +{0.80901699f, -0.58778525f}, {0.80125381f, -0.59832460f}, +{0.79335334f, -0.60876143f}, {0.78531693f, -0.61909395f}, +{0.77714596f, -0.62932039f}, {0.76884183f, -0.63943900f}, +{0.76040597f, -0.64944805f}, {0.75183981f, -0.65934582f}, +{0.74314483f, -0.66913061f}, {0.73432251f, -0.67880075f}, +{0.72537437f, -0.68835458f}, {0.71630194f, -0.69779046f}, +{0.70710678f, -0.70710678f}, {0.69779046f, -0.71630194f}, +{0.68835458f, -0.72537437f}, {0.67880075f, -0.73432251f}, +{0.66913061f, -0.74314483f}, {0.65934582f, -0.75183981f}, +{0.64944805f, -0.76040597f}, {0.63943900f, -0.76884183f}, +{0.62932039f, -0.77714596f}, {0.61909395f, -0.78531693f}, +{0.60876143f, -0.79335334f}, {0.59832460f, -0.80125381f}, +{0.58778525f, -0.80901699f}, {0.57714519f, -0.81664156f}, +{0.56640624f, -0.82412619f}, {0.55557023f, -0.83146961f}, +{0.54463904f, -0.83867057f}, {0.53361452f, -0.84572782f}, +{0.52249856f, -0.85264016f}, {0.51129309f, -0.85940641f}, +{0.50000000f, -0.86602540f}, {0.48862124f, -0.87249601f}, +{0.47715876f, -0.87881711f}, {0.46561452f, -0.88498764f}, +{0.45399050f, -0.89100652f}, {0.44228869f, -0.89687274f}, +{0.43051110f, -0.90258528f}, {0.41865974f, -0.90814317f}, +{0.40673664f, -0.91354546f}, {0.39474386f, -0.91879121f}, +{0.38268343f, -0.92387953f}, {0.37055744f, -0.92880955f}, +{0.35836795f, -0.93358043f}, {0.34611706f, -0.93819134f}, +{0.33380686f, -0.94264149f}, {0.32143947f, -0.94693013f}, +{0.30901699f, -0.95105652f}, {0.29654157f, -0.95501994f}, +{0.28401534f, -0.95881973f}, {0.27144045f, -0.96245524f}, +{0.25881905f, -0.96592583f}, {0.24615329f, -0.96923091f}, +{0.23344536f, -0.97236992f}, {0.22069744f, -0.97534232f}, +{0.20791169f, -0.97814760f}, {0.19509032f, -0.98078528f}, +{0.18223553f, -0.98325491f}, {0.16934950f, -0.98555606f}, +{0.15643447f, -0.98768834f}, {0.14349262f, -0.98965139f}, +{0.13052619f, -0.99144486f}, {0.11753740f, -0.99306846f}, +{0.10452846f, -0.99452190f}, {0.091501619f, -0.99580493f}, +{0.078459096f, -0.99691733f}, {0.065403129f, -0.99785892f}, +{0.052335956f, -0.99862953f}, {0.039259816f, -0.99922904f}, +{0.026176948f, -0.99965732f}, {0.013089596f, -0.99991433f}, +{6.1230318e-17f, -1.0000000f}, {-0.013089596f, -0.99991433f}, +{-0.026176948f, -0.99965732f}, {-0.039259816f, -0.99922904f}, +{-0.052335956f, -0.99862953f}, {-0.065403129f, -0.99785892f}, +{-0.078459096f, -0.99691733f}, {-0.091501619f, -0.99580493f}, +{-0.10452846f, -0.99452190f}, {-0.11753740f, -0.99306846f}, +{-0.13052619f, -0.99144486f}, {-0.14349262f, -0.98965139f}, +{-0.15643447f, -0.98768834f}, {-0.16934950f, -0.98555606f}, +{-0.18223553f, -0.98325491f}, {-0.19509032f, -0.98078528f}, +{-0.20791169f, -0.97814760f}, {-0.22069744f, -0.97534232f}, +{-0.23344536f, -0.97236992f}, {-0.24615329f, -0.96923091f}, +{-0.25881905f, -0.96592583f}, {-0.27144045f, -0.96245524f}, +{-0.28401534f, -0.95881973f}, {-0.29654157f, -0.95501994f}, +{-0.30901699f, -0.95105652f}, {-0.32143947f, -0.94693013f}, +{-0.33380686f, -0.94264149f}, {-0.34611706f, -0.93819134f}, +{-0.35836795f, -0.93358043f}, {-0.37055744f, -0.92880955f}, +{-0.38268343f, -0.92387953f}, {-0.39474386f, -0.91879121f}, +{-0.40673664f, -0.91354546f}, {-0.41865974f, -0.90814317f}, +{-0.43051110f, -0.90258528f}, {-0.44228869f, -0.89687274f}, +{-0.45399050f, -0.89100652f}, {-0.46561452f, -0.88498764f}, +{-0.47715876f, -0.87881711f}, {-0.48862124f, -0.87249601f}, +{-0.50000000f, -0.86602540f}, {-0.51129309f, -0.85940641f}, +{-0.52249856f, -0.85264016f}, {-0.53361452f, -0.84572782f}, +{-0.54463904f, -0.83867057f}, {-0.55557023f, -0.83146961f}, +{-0.56640624f, -0.82412619f}, {-0.57714519f, -0.81664156f}, +{-0.58778525f, -0.80901699f}, {-0.59832460f, -0.80125381f}, +{-0.60876143f, -0.79335334f}, {-0.61909395f, -0.78531693f}, +{-0.62932039f, -0.77714596f}, {-0.63943900f, -0.76884183f}, +{-0.64944805f, -0.76040597f}, {-0.65934582f, -0.75183981f}, +{-0.66913061f, -0.74314483f}, {-0.67880075f, -0.73432251f}, +{-0.68835458f, -0.72537437f}, {-0.69779046f, -0.71630194f}, +{-0.70710678f, -0.70710678f}, {-0.71630194f, -0.69779046f}, +{-0.72537437f, -0.68835458f}, {-0.73432251f, -0.67880075f}, +{-0.74314483f, -0.66913061f}, {-0.75183981f, -0.65934582f}, +{-0.76040597f, -0.64944805f}, {-0.76884183f, -0.63943900f}, +{-0.77714596f, -0.62932039f}, {-0.78531693f, -0.61909395f}, +{-0.79335334f, -0.60876143f}, {-0.80125381f, -0.59832460f}, +{-0.80901699f, -0.58778525f}, {-0.81664156f, -0.57714519f}, +{-0.82412619f, -0.56640624f}, {-0.83146961f, -0.55557023f}, +{-0.83867057f, -0.54463904f}, {-0.84572782f, -0.53361452f}, +{-0.85264016f, -0.52249856f}, {-0.85940641f, -0.51129309f}, +{-0.86602540f, -0.50000000f}, {-0.87249601f, -0.48862124f}, +{-0.87881711f, -0.47715876f}, {-0.88498764f, -0.46561452f}, +{-0.89100652f, -0.45399050f}, {-0.89687274f, -0.44228869f}, +{-0.90258528f, -0.43051110f}, {-0.90814317f, -0.41865974f}, +{-0.91354546f, -0.40673664f}, {-0.91879121f, -0.39474386f}, +{-0.92387953f, -0.38268343f}, {-0.92880955f, -0.37055744f}, +{-0.93358043f, -0.35836795f}, {-0.93819134f, -0.34611706f}, +{-0.94264149f, -0.33380686f}, {-0.94693013f, -0.32143947f}, +{-0.95105652f, -0.30901699f}, {-0.95501994f, -0.29654157f}, +{-0.95881973f, -0.28401534f}, {-0.96245524f, -0.27144045f}, +{-0.96592583f, -0.25881905f}, {-0.96923091f, -0.24615329f}, +{-0.97236992f, -0.23344536f}, {-0.97534232f, -0.22069744f}, +{-0.97814760f, -0.20791169f}, {-0.98078528f, -0.19509032f}, +{-0.98325491f, -0.18223553f}, {-0.98555606f, -0.16934950f}, +{-0.98768834f, -0.15643447f}, {-0.98965139f, -0.14349262f}, +{-0.99144486f, -0.13052619f}, {-0.99306846f, -0.11753740f}, +{-0.99452190f, -0.10452846f}, {-0.99580493f, -0.091501619f}, +{-0.99691733f, -0.078459096f}, {-0.99785892f, -0.065403129f}, +{-0.99862953f, -0.052335956f}, {-0.99922904f, -0.039259816f}, +{-0.99965732f, -0.026176948f}, {-0.99991433f, -0.013089596f}, +{-1.0000000f, -1.2246064e-16f}, {-0.99991433f, 0.013089596f}, +{-0.99965732f, 0.026176948f}, {-0.99922904f, 0.039259816f}, +{-0.99862953f, 0.052335956f}, {-0.99785892f, 0.065403129f}, +{-0.99691733f, 0.078459096f}, {-0.99580493f, 0.091501619f}, +{-0.99452190f, 0.10452846f}, {-0.99306846f, 0.11753740f}, +{-0.99144486f, 0.13052619f}, {-0.98965139f, 0.14349262f}, +{-0.98768834f, 0.15643447f}, {-0.98555606f, 0.16934950f}, +{-0.98325491f, 0.18223553f}, {-0.98078528f, 0.19509032f}, +{-0.97814760f, 0.20791169f}, {-0.97534232f, 0.22069744f}, +{-0.97236992f, 0.23344536f}, {-0.96923091f, 0.24615329f}, +{-0.96592583f, 0.25881905f}, {-0.96245524f, 0.27144045f}, +{-0.95881973f, 0.28401534f}, {-0.95501994f, 0.29654157f}, +{-0.95105652f, 0.30901699f}, {-0.94693013f, 0.32143947f}, +{-0.94264149f, 0.33380686f}, {-0.93819134f, 0.34611706f}, +{-0.93358043f, 0.35836795f}, {-0.92880955f, 0.37055744f}, +{-0.92387953f, 0.38268343f}, {-0.91879121f, 0.39474386f}, +{-0.91354546f, 0.40673664f}, {-0.90814317f, 0.41865974f}, +{-0.90258528f, 0.43051110f}, {-0.89687274f, 0.44228869f}, +{-0.89100652f, 0.45399050f}, {-0.88498764f, 0.46561452f}, +{-0.87881711f, 0.47715876f}, {-0.87249601f, 0.48862124f}, +{-0.86602540f, 0.50000000f}, {-0.85940641f, 0.51129309f}, +{-0.85264016f, 0.52249856f}, {-0.84572782f, 0.53361452f}, +{-0.83867057f, 0.54463904f}, {-0.83146961f, 0.55557023f}, +{-0.82412619f, 0.56640624f}, {-0.81664156f, 0.57714519f}, +{-0.80901699f, 0.58778525f}, {-0.80125381f, 0.59832460f}, +{-0.79335334f, 0.60876143f}, {-0.78531693f, 0.61909395f}, +{-0.77714596f, 0.62932039f}, {-0.76884183f, 0.63943900f}, +{-0.76040597f, 0.64944805f}, {-0.75183981f, 0.65934582f}, +{-0.74314483f, 0.66913061f}, {-0.73432251f, 0.67880075f}, +{-0.72537437f, 0.68835458f}, {-0.71630194f, 0.69779046f}, +{-0.70710678f, 0.70710678f}, {-0.69779046f, 0.71630194f}, +{-0.68835458f, 0.72537437f}, {-0.67880075f, 0.73432251f}, +{-0.66913061f, 0.74314483f}, {-0.65934582f, 0.75183981f}, +{-0.64944805f, 0.76040597f}, {-0.63943900f, 0.76884183f}, +{-0.62932039f, 0.77714596f}, {-0.61909395f, 0.78531693f}, +{-0.60876143f, 0.79335334f}, {-0.59832460f, 0.80125381f}, +{-0.58778525f, 0.80901699f}, {-0.57714519f, 0.81664156f}, +{-0.56640624f, 0.82412619f}, {-0.55557023f, 0.83146961f}, +{-0.54463904f, 0.83867057f}, {-0.53361452f, 0.84572782f}, +{-0.52249856f, 0.85264016f}, {-0.51129309f, 0.85940641f}, +{-0.50000000f, 0.86602540f}, {-0.48862124f, 0.87249601f}, +{-0.47715876f, 0.87881711f}, {-0.46561452f, 0.88498764f}, +{-0.45399050f, 0.89100652f}, {-0.44228869f, 0.89687274f}, +{-0.43051110f, 0.90258528f}, {-0.41865974f, 0.90814317f}, +{-0.40673664f, 0.91354546f}, {-0.39474386f, 0.91879121f}, +{-0.38268343f, 0.92387953f}, {-0.37055744f, 0.92880955f}, +{-0.35836795f, 0.93358043f}, {-0.34611706f, 0.93819134f}, +{-0.33380686f, 0.94264149f}, {-0.32143947f, 0.94693013f}, +{-0.30901699f, 0.95105652f}, {-0.29654157f, 0.95501994f}, +{-0.28401534f, 0.95881973f}, {-0.27144045f, 0.96245524f}, +{-0.25881905f, 0.96592583f}, {-0.24615329f, 0.96923091f}, +{-0.23344536f, 0.97236992f}, {-0.22069744f, 0.97534232f}, +{-0.20791169f, 0.97814760f}, {-0.19509032f, 0.98078528f}, +{-0.18223553f, 0.98325491f}, {-0.16934950f, 0.98555606f}, +{-0.15643447f, 0.98768834f}, {-0.14349262f, 0.98965139f}, +{-0.13052619f, 0.99144486f}, {-0.11753740f, 0.99306846f}, +{-0.10452846f, 0.99452190f}, {-0.091501619f, 0.99580493f}, +{-0.078459096f, 0.99691733f}, {-0.065403129f, 0.99785892f}, +{-0.052335956f, 0.99862953f}, {-0.039259816f, 0.99922904f}, +{-0.026176948f, 0.99965732f}, {-0.013089596f, 0.99991433f}, +{-1.8369095e-16f, 1.0000000f}, {0.013089596f, 0.99991433f}, +{0.026176948f, 0.99965732f}, {0.039259816f, 0.99922904f}, +{0.052335956f, 0.99862953f}, {0.065403129f, 0.99785892f}, +{0.078459096f, 0.99691733f}, {0.091501619f, 0.99580493f}, +{0.10452846f, 0.99452190f}, {0.11753740f, 0.99306846f}, +{0.13052619f, 0.99144486f}, {0.14349262f, 0.98965139f}, +{0.15643447f, 0.98768834f}, {0.16934950f, 0.98555606f}, +{0.18223553f, 0.98325491f}, {0.19509032f, 0.98078528f}, +{0.20791169f, 0.97814760f}, {0.22069744f, 0.97534232f}, +{0.23344536f, 0.97236992f}, {0.24615329f, 0.96923091f}, +{0.25881905f, 0.96592583f}, {0.27144045f, 0.96245524f}, +{0.28401534f, 0.95881973f}, {0.29654157f, 0.95501994f}, +{0.30901699f, 0.95105652f}, {0.32143947f, 0.94693013f}, +{0.33380686f, 0.94264149f}, {0.34611706f, 0.93819134f}, +{0.35836795f, 0.93358043f}, {0.37055744f, 0.92880955f}, +{0.38268343f, 0.92387953f}, {0.39474386f, 0.91879121f}, +{0.40673664f, 0.91354546f}, {0.41865974f, 0.90814317f}, +{0.43051110f, 0.90258528f}, {0.44228869f, 0.89687274f}, +{0.45399050f, 0.89100652f}, {0.46561452f, 0.88498764f}, +{0.47715876f, 0.87881711f}, {0.48862124f, 0.87249601f}, +{0.50000000f, 0.86602540f}, {0.51129309f, 0.85940641f}, +{0.52249856f, 0.85264016f}, {0.53361452f, 0.84572782f}, +{0.54463904f, 0.83867057f}, {0.55557023f, 0.83146961f}, +{0.56640624f, 0.82412619f}, {0.57714519f, 0.81664156f}, +{0.58778525f, 0.80901699f}, {0.59832460f, 0.80125381f}, +{0.60876143f, 0.79335334f}, {0.61909395f, 0.78531693f}, +{0.62932039f, 0.77714596f}, {0.63943900f, 0.76884183f}, +{0.64944805f, 0.76040597f}, {0.65934582f, 0.75183981f}, +{0.66913061f, 0.74314483f}, {0.67880075f, 0.73432251f}, +{0.68835458f, 0.72537437f}, {0.69779046f, 0.71630194f}, +{0.70710678f, 0.70710678f}, {0.71630194f, 0.69779046f}, +{0.72537437f, 0.68835458f}, {0.73432251f, 0.67880075f}, +{0.74314483f, 0.66913061f}, {0.75183981f, 0.65934582f}, +{0.76040597f, 0.64944805f}, {0.76884183f, 0.63943900f}, +{0.77714596f, 0.62932039f}, {0.78531693f, 0.61909395f}, +{0.79335334f, 0.60876143f}, {0.80125381f, 0.59832460f}, +{0.80901699f, 0.58778525f}, {0.81664156f, 0.57714519f}, +{0.82412619f, 0.56640624f}, {0.83146961f, 0.55557023f}, +{0.83867057f, 0.54463904f}, {0.84572782f, 0.53361452f}, +{0.85264016f, 0.52249856f}, {0.85940641f, 0.51129309f}, +{0.86602540f, 0.50000000f}, {0.87249601f, 0.48862124f}, +{0.87881711f, 0.47715876f}, {0.88498764f, 0.46561452f}, +{0.89100652f, 0.45399050f}, {0.89687274f, 0.44228869f}, +{0.90258528f, 0.43051110f}, {0.90814317f, 0.41865974f}, +{0.91354546f, 0.40673664f}, {0.91879121f, 0.39474386f}, +{0.92387953f, 0.38268343f}, {0.92880955f, 0.37055744f}, +{0.93358043f, 0.35836795f}, {0.93819134f, 0.34611706f}, +{0.94264149f, 0.33380686f}, {0.94693013f, 0.32143947f}, +{0.95105652f, 0.30901699f}, {0.95501994f, 0.29654157f}, +{0.95881973f, 0.28401534f}, {0.96245524f, 0.27144045f}, +{0.96592583f, 0.25881905f}, {0.96923091f, 0.24615329f}, +{0.97236992f, 0.23344536f}, {0.97534232f, 0.22069744f}, +{0.97814760f, 0.20791169f}, {0.98078528f, 0.19509032f}, +{0.98325491f, 0.18223553f}, {0.98555606f, 0.16934950f}, +{0.98768834f, 0.15643447f}, {0.98965139f, 0.14349262f}, +{0.99144486f, 0.13052619f}, {0.99306846f, 0.11753740f}, +{0.99452190f, 0.10452846f}, {0.99580493f, 0.091501619f}, +{0.99691733f, 0.078459096f}, {0.99785892f, 0.065403129f}, +{0.99862953f, 0.052335956f}, {0.99922904f, 0.039259816f}, +{0.99965732f, 0.026176948f}, {0.99991433f, 0.013089596f}, +}; +#ifndef FFT_BITREV480 +#define FFT_BITREV480 +static const opus_int16 fft_bitrev480[480] = { +0, 96, 192, 288, 384, 32, 128, 224, 320, 416, 64, 160, 256, 352, 448, +8, 104, 200, 296, 392, 40, 136, 232, 328, 424, 72, 168, 264, 360, 456, +16, 112, 208, 304, 400, 48, 144, 240, 336, 432, 80, 176, 272, 368, 464, +24, 120, 216, 312, 408, 56, 152, 248, 344, 440, 88, 184, 280, 376, 472, +4, 100, 196, 292, 388, 36, 132, 228, 324, 420, 68, 164, 260, 356, 452, +12, 108, 204, 300, 396, 44, 140, 236, 332, 428, 76, 172, 268, 364, 460, +20, 116, 212, 308, 404, 52, 148, 244, 340, 436, 84, 180, 276, 372, 468, +28, 124, 220, 316, 412, 60, 156, 252, 348, 444, 92, 188, 284, 380, 476, +1, 97, 193, 289, 385, 33, 129, 225, 321, 417, 65, 161, 257, 353, 449, +9, 105, 201, 297, 393, 41, 137, 233, 329, 425, 73, 169, 265, 361, 457, +17, 113, 209, 305, 401, 49, 145, 241, 337, 433, 81, 177, 273, 369, 465, +25, 121, 217, 313, 409, 57, 153, 249, 345, 441, 89, 185, 281, 377, 473, +5, 101, 197, 293, 389, 37, 133, 229, 325, 421, 69, 165, 261, 357, 453, +13, 109, 205, 301, 397, 45, 141, 237, 333, 429, 77, 173, 269, 365, 461, +21, 117, 213, 309, 405, 53, 149, 245, 341, 437, 85, 181, 277, 373, 469, +29, 125, 221, 317, 413, 61, 157, 253, 349, 445, 93, 189, 285, 381, 477, +2, 98, 194, 290, 386, 34, 130, 226, 322, 418, 66, 162, 258, 354, 450, +10, 106, 202, 298, 394, 42, 138, 234, 330, 426, 74, 170, 266, 362, 458, +18, 114, 210, 306, 402, 50, 146, 242, 338, 434, 82, 178, 274, 370, 466, +26, 122, 218, 314, 410, 58, 154, 250, 346, 442, 90, 186, 282, 378, 474, +6, 102, 198, 294, 390, 38, 134, 230, 326, 422, 70, 166, 262, 358, 454, +14, 110, 206, 302, 398, 46, 142, 238, 334, 430, 78, 174, 270, 366, 462, +22, 118, 214, 310, 406, 54, 150, 246, 342, 438, 86, 182, 278, 374, 470, +30, 126, 222, 318, 414, 62, 158, 254, 350, 446, 94, 190, 286, 382, 478, +3, 99, 195, 291, 387, 35, 131, 227, 323, 419, 67, 163, 259, 355, 451, +11, 107, 203, 299, 395, 43, 139, 235, 331, 427, 75, 171, 267, 363, 459, +19, 115, 211, 307, 403, 51, 147, 243, 339, 435, 83, 179, 275, 371, 467, +27, 123, 219, 315, 411, 59, 155, 251, 347, 443, 91, 187, 283, 379, 475, +7, 103, 199, 295, 391, 39, 135, 231, 327, 423, 71, 167, 263, 359, 455, +15, 111, 207, 303, 399, 47, 143, 239, 335, 431, 79, 175, 271, 367, 463, +23, 119, 215, 311, 407, 55, 151, 247, 343, 439, 87, 183, 279, 375, 471, +31, 127, 223, 319, 415, 63, 159, 255, 351, 447, 95, 191, 287, 383, 479, +}; +#endif + +#ifndef FFT_BITREV240 +#define FFT_BITREV240 +static const opus_int16 fft_bitrev240[240] = { +0, 48, 96, 144, 192, 16, 64, 112, 160, 208, 32, 80, 128, 176, 224, +4, 52, 100, 148, 196, 20, 68, 116, 164, 212, 36, 84, 132, 180, 228, +8, 56, 104, 152, 200, 24, 72, 120, 168, 216, 40, 88, 136, 184, 232, +12, 60, 108, 156, 204, 28, 76, 124, 172, 220, 44, 92, 140, 188, 236, +1, 49, 97, 145, 193, 17, 65, 113, 161, 209, 33, 81, 129, 177, 225, +5, 53, 101, 149, 197, 21, 69, 117, 165, 213, 37, 85, 133, 181, 229, +9, 57, 105, 153, 201, 25, 73, 121, 169, 217, 41, 89, 137, 185, 233, +13, 61, 109, 157, 205, 29, 77, 125, 173, 221, 45, 93, 141, 189, 237, +2, 50, 98, 146, 194, 18, 66, 114, 162, 210, 34, 82, 130, 178, 226, +6, 54, 102, 150, 198, 22, 70, 118, 166, 214, 38, 86, 134, 182, 230, +10, 58, 106, 154, 202, 26, 74, 122, 170, 218, 42, 90, 138, 186, 234, +14, 62, 110, 158, 206, 30, 78, 126, 174, 222, 46, 94, 142, 190, 238, +3, 51, 99, 147, 195, 19, 67, 115, 163, 211, 35, 83, 131, 179, 227, +7, 55, 103, 151, 199, 23, 71, 119, 167, 215, 39, 87, 135, 183, 231, +11, 59, 107, 155, 203, 27, 75, 123, 171, 219, 43, 91, 139, 187, 235, +15, 63, 111, 159, 207, 31, 79, 127, 175, 223, 47, 95, 143, 191, 239, +}; +#endif + +#ifndef FFT_BITREV120 +#define FFT_BITREV120 +static const opus_int16 fft_bitrev120[120] = { +0, 24, 48, 72, 96, 8, 32, 56, 80, 104, 16, 40, 64, 88, 112, +4, 28, 52, 76, 100, 12, 36, 60, 84, 108, 20, 44, 68, 92, 116, +1, 25, 49, 73, 97, 9, 33, 57, 81, 105, 17, 41, 65, 89, 113, +5, 29, 53, 77, 101, 13, 37, 61, 85, 109, 21, 45, 69, 93, 117, +2, 26, 50, 74, 98, 10, 34, 58, 82, 106, 18, 42, 66, 90, 114, +6, 30, 54, 78, 102, 14, 38, 62, 86, 110, 22, 46, 70, 94, 118, +3, 27, 51, 75, 99, 11, 35, 59, 83, 107, 19, 43, 67, 91, 115, +7, 31, 55, 79, 103, 15, 39, 63, 87, 111, 23, 47, 71, 95, 119, +}; +#endif + +#ifndef FFT_BITREV60 +#define FFT_BITREV60 +static const opus_int16 fft_bitrev60[60] = { +0, 12, 24, 36, 48, 4, 16, 28, 40, 52, 8, 20, 32, 44, 56, +1, 13, 25, 37, 49, 5, 17, 29, 41, 53, 9, 21, 33, 45, 57, +2, 14, 26, 38, 50, 6, 18, 30, 42, 54, 10, 22, 34, 46, 58, +3, 15, 27, 39, 51, 7, 19, 31, 43, 55, 11, 23, 35, 47, 59, +}; +#endif + +#ifndef FFT_STATE48000_960_0 +#define FFT_STATE48000_960_0 +static const kiss_fft_state fft_state48000_960_0 = { +480, /* nfft */ +0.002083333f, /* scale */ +-1, /* shift */ +{5, 96, 3, 32, 4, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev480, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_480, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_1 +#define FFT_STATE48000_960_1 +static const kiss_fft_state fft_state48000_960_1 = { +240, /* nfft */ +0.004166667f, /* scale */ +1, /* shift */ +{5, 48, 3, 16, 4, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev240, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_240, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_2 +#define FFT_STATE48000_960_2 +static const kiss_fft_state fft_state48000_960_2 = { +120, /* nfft */ +0.008333333f, /* scale */ +2, /* shift */ +{5, 24, 3, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev120, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_120, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_3 +#define FFT_STATE48000_960_3 +static const kiss_fft_state fft_state48000_960_3 = { +60, /* nfft */ +0.016666667f, /* scale */ +3, /* shift */ +{5, 12, 3, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev60, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_60, +#else +NULL, +#endif +}; +#endif + +#endif + +#ifndef MDCT_TWIDDLES960 +#define MDCT_TWIDDLES960 +static const opus_val16 mdct_twiddles960[1800] = { +0.99999994f, 0.99999321f, 0.99997580f, 0.99994773f, 0.99990886f, +0.99985933f, 0.99979913f, 0.99972820f, 0.99964654f, 0.99955416f, +0.99945110f, 0.99933738f, 0.99921292f, 0.99907774f, 0.99893188f, +0.99877530f, 0.99860805f, 0.99843007f, 0.99824142f, 0.99804211f, +0.99783206f, 0.99761140f, 0.99737996f, 0.99713790f, 0.99688518f, +0.99662173f, 0.99634761f, 0.99606287f, 0.99576741f, 0.99546129f, +0.99514455f, 0.99481714f, 0.99447906f, 0.99413031f, 0.99377096f, +0.99340093f, 0.99302030f, 0.99262899f, 0.99222708f, 0.99181455f, +0.99139136f, 0.99095762f, 0.99051321f, 0.99005818f, 0.98959261f, +0.98911643f, 0.98862964f, 0.98813224f, 0.98762429f, 0.98710573f, +0.98657662f, 0.98603696f, 0.98548669f, 0.98492593f, 0.98435456f, +0.98377270f, 0.98318028f, 0.98257732f, 0.98196387f, 0.98133987f, +0.98070538f, 0.98006040f, 0.97940493f, 0.97873890f, 0.97806245f, +0.97737551f, 0.97667813f, 0.97597027f, 0.97525197f, 0.97452319f, +0.97378403f, 0.97303438f, 0.97227436f, 0.97150391f, 0.97072303f, +0.96993178f, 0.96913016f, 0.96831810f, 0.96749574f, 0.96666300f, +0.96581990f, 0.96496642f, 0.96410263f, 0.96322852f, 0.96234411f, +0.96144938f, 0.96054435f, 0.95962906f, 0.95870346f, 0.95776761f, +0.95682150f, 0.95586514f, 0.95489854f, 0.95392174f, 0.95293468f, +0.95193744f, 0.95093000f, 0.94991243f, 0.94888461f, 0.94784665f, +0.94679856f, 0.94574034f, 0.94467193f, 0.94359344f, 0.94250488f, +0.94140619f, 0.94029742f, 0.93917859f, 0.93804967f, 0.93691075f, +0.93576175f, 0.93460274f, 0.93343377f, 0.93225473f, 0.93106574f, +0.92986679f, 0.92865789f, 0.92743903f, 0.92621022f, 0.92497152f, +0.92372292f, 0.92246443f, 0.92119598f, 0.91991776f, 0.91862965f, +0.91733170f, 0.91602397f, 0.91470635f, 0.91337901f, 0.91204184f, +0.91069490f, 0.90933824f, 0.90797186f, 0.90659571f, 0.90520984f, +0.90381432f, 0.90240908f, 0.90099424f, 0.89956969f, 0.89813554f, +0.89669174f, 0.89523834f, 0.89377540f, 0.89230281f, 0.89082074f, +0.88932908f, 0.88782793f, 0.88631725f, 0.88479710f, 0.88326746f, +0.88172835f, 0.88017982f, 0.87862182f, 0.87705445f, 0.87547767f, +0.87389153f, 0.87229604f, 0.87069118f, 0.86907703f, 0.86745358f, +0.86582077f, 0.86417878f, 0.86252749f, 0.86086690f, 0.85919720f, +0.85751826f, 0.85583007f, 0.85413277f, 0.85242635f, 0.85071075f, +0.84898609f, 0.84725231f, 0.84550947f, 0.84375757f, 0.84199661f, +0.84022665f, 0.83844769f, 0.83665979f, 0.83486289f, 0.83305705f, +0.83124226f, 0.82941860f, 0.82758605f, 0.82574469f, 0.82389444f, +0.82203537f, 0.82016748f, 0.81829083f, 0.81640542f, 0.81451124f, +0.81260836f, 0.81069672f, 0.80877650f, 0.80684757f, 0.80490994f, +0.80296379f, 0.80100900f, 0.79904562f, 0.79707366f, 0.79509324f, +0.79310423f, 0.79110676f, 0.78910083f, 0.78708643f, 0.78506362f, +0.78303236f, 0.78099275f, 0.77894479f, 0.77688843f, 0.77482378f, +0.77275085f, 0.77066964f, 0.76858020f, 0.76648247f, 0.76437658f, +0.76226246f, 0.76014024f, 0.75800985f, 0.75587130f, 0.75372469f, +0.75157005f, 0.74940729f, 0.74723655f, 0.74505776f, 0.74287105f, +0.74067634f, 0.73847371f, 0.73626316f, 0.73404479f, 0.73181850f, +0.72958434f, 0.72734243f, 0.72509271f, 0.72283524f, 0.72057003f, +0.71829706f, 0.71601641f, 0.71372813f, 0.71143216f, 0.70912862f, +0.70681745f, 0.70449871f, 0.70217246f, 0.69983864f, 0.69749737f, +0.69514859f, 0.69279242f, 0.69042879f, 0.68805778f, 0.68567938f, +0.68329364f, 0.68090063f, 0.67850029f, 0.67609268f, 0.67367786f, +0.67125577f, 0.66882652f, 0.66639012f, 0.66394657f, 0.66149592f, +0.65903819f, 0.65657341f, 0.65410155f, 0.65162271f, 0.64913690f, +0.64664418f, 0.64414448f, 0.64163786f, 0.63912445f, 0.63660413f, +0.63407701f, 0.63154310f, 0.62900239f, 0.62645501f, 0.62390089f, +0.62134010f, 0.61877263f, 0.61619854f, 0.61361790f, 0.61103064f, +0.60843682f, 0.60583651f, 0.60322970f, 0.60061646f, 0.59799677f, +0.59537065f, 0.59273821f, 0.59009939f, 0.58745426f, 0.58480281f, +0.58214509f, 0.57948118f, 0.57681108f, 0.57413477f, 0.57145232f, +0.56876373f, 0.56606907f, 0.56336832f, 0.56066155f, 0.55794877f, +0.55523002f, 0.55250537f, 0.54977477f, 0.54703826f, 0.54429591f, +0.54154772f, 0.53879374f, 0.53603399f, 0.53326851f, 0.53049731f, +0.52772039f, 0.52493787f, 0.52214974f, 0.51935595f, 0.51655668f, +0.51375180f, 0.51094145f, 0.50812566f, 0.50530440f, 0.50247771f, +0.49964568f, 0.49680826f, 0.49396557f, 0.49111754f, 0.48826426f, +0.48540577f, 0.48254207f, 0.47967321f, 0.47679919f, 0.47392011f, +0.47103590f, 0.46814668f, 0.46525243f, 0.46235323f, 0.45944905f, +0.45653993f, 0.45362595f, 0.45070711f, 0.44778344f, 0.44485497f, +0.44192174f, 0.43898380f, 0.43604112f, 0.43309379f, 0.43014181f, +0.42718524f, 0.42422408f, 0.42125839f, 0.41828820f, 0.41531351f, +0.41233435f, 0.40935081f, 0.40636289f, 0.40337059f, 0.40037400f, +0.39737311f, 0.39436796f, 0.39135858f, 0.38834500f, 0.38532731f, +0.38230544f, 0.37927949f, 0.37624949f, 0.37321547f, 0.37017745f, +0.36713544f, 0.36408952f, 0.36103970f, 0.35798600f, 0.35492846f, +0.35186714f, 0.34880206f, 0.34573323f, 0.34266070f, 0.33958447f, +0.33650464f, 0.33342120f, 0.33033419f, 0.32724363f, 0.32414958f, +0.32105204f, 0.31795108f, 0.31484672f, 0.31173897f, 0.30862790f, +0.30551350f, 0.30239585f, 0.29927495f, 0.29615086f, 0.29302359f, +0.28989318f, 0.28675964f, 0.28362307f, 0.28048345f, 0.27734083f, +0.27419522f, 0.27104670f, 0.26789525f, 0.26474094f, 0.26158381f, +0.25842386f, 0.25526115f, 0.25209570f, 0.24892756f, 0.24575676f, +0.24258332f, 0.23940729f, 0.23622867f, 0.23304754f, 0.22986393f, +0.22667783f, 0.22348931f, 0.22029841f, 0.21710514f, 0.21390954f, +0.21071166f, 0.20751151f, 0.20430915f, 0.20110460f, 0.19789790f, +0.19468907f, 0.19147816f, 0.18826519f, 0.18505022f, 0.18183327f, +0.17861435f, 0.17539354f, 0.17217083f, 0.16894630f, 0.16571994f, +0.16249183f, 0.15926196f, 0.15603039f, 0.15279715f, 0.14956227f, +0.14632578f, 0.14308774f, 0.13984816f, 0.13660708f, 0.13336454f, +0.13012058f, 0.12687522f, 0.12362850f, 0.12038045f, 0.11713112f, +0.11388054f, 0.11062872f, 0.10737573f, 0.10412160f, 0.10086634f, +0.097609997f, 0.094352618f, 0.091094226f, 0.087834857f, 0.084574550f, +0.081313334f, 0.078051247f, 0.074788325f, 0.071524605f, 0.068260118f, +0.064994894f, 0.061728980f, 0.058462404f, 0.055195201f, 0.051927410f, +0.048659060f, 0.045390189f, 0.042120833f, 0.038851023f, 0.035580799f, +0.032310195f, 0.029039243f, 0.025767982f, 0.022496443f, 0.019224664f, +0.015952680f, 0.012680525f, 0.0094082337f, 0.0061358409f, 0.0028633832f, +-0.00040910527f, -0.0036815894f, -0.0069540343f, -0.010226404f, -0.013498665f, +-0.016770782f, -0.020042717f, -0.023314439f, -0.026585912f, -0.029857099f, +-0.033127967f, -0.036398482f, -0.039668605f, -0.042938303f, -0.046207540f, +-0.049476285f, -0.052744497f, -0.056012146f, -0.059279196f, -0.062545612f, +-0.065811358f, -0.069076397f, -0.072340697f, -0.075604223f, -0.078866936f, +-0.082128808f, -0.085389800f, -0.088649876f, -0.091909006f, -0.095167145f, +-0.098424271f, -0.10168034f, -0.10493532f, -0.10818918f, -0.11144188f, +-0.11469338f, -0.11794366f, -0.12119267f, -0.12444039f, -0.12768677f, +-0.13093179f, -0.13417540f, -0.13741758f, -0.14065829f, -0.14389749f, +-0.14713514f, -0.15037122f, -0.15360570f, -0.15683852f, -0.16006967f, +-0.16329910f, -0.16652679f, -0.16975269f, -0.17297678f, -0.17619900f, +-0.17941935f, -0.18263777f, -0.18585424f, -0.18906870f, -0.19228116f, +-0.19549155f, -0.19869985f, -0.20190603f, -0.20511003f, -0.20831184f, +-0.21151142f, -0.21470875f, -0.21790376f, -0.22109644f, -0.22428675f, +-0.22747467f, -0.23066014f, -0.23384315f, -0.23702365f, -0.24020162f, +-0.24337701f, -0.24654980f, -0.24971995f, -0.25288740f, -0.25605217f, +-0.25921419f, -0.26237345f, -0.26552987f, -0.26868346f, -0.27183419f, +-0.27498198f, -0.27812684f, -0.28126872f, -0.28440759f, -0.28754342f, +-0.29067615f, -0.29380578f, -0.29693225f, -0.30005556f, -0.30317566f, +-0.30629250f, -0.30940607f, -0.31251630f, -0.31562322f, -0.31872672f, +-0.32182685f, -0.32492352f, -0.32801670f, -0.33110636f, -0.33419248f, +-0.33727503f, -0.34035397f, -0.34342924f, -0.34650084f, -0.34956875f, +-0.35263291f, -0.35569328f, -0.35874987f, -0.36180258f, -0.36485144f, +-0.36789638f, -0.37093741f, -0.37397444f, -0.37700745f, -0.38003644f, +-0.38306138f, -0.38608220f, -0.38909888f, -0.39211139f, -0.39511973f, +-0.39812380f, -0.40112361f, -0.40411916f, -0.40711036f, -0.41009718f, +-0.41307965f, -0.41605768f, -0.41903123f, -0.42200032f, -0.42496487f, +-0.42792490f, -0.43088034f, -0.43383113f, -0.43677729f, -0.43971881f, +-0.44265559f, -0.44558764f, -0.44851488f, -0.45143735f, -0.45435500f, +-0.45726776f, -0.46017563f, -0.46307856f, -0.46597654f, -0.46886954f, +-0.47175750f, -0.47464043f, -0.47751826f, -0.48039100f, -0.48325855f, +-0.48612097f, -0.48897815f, -0.49183011f, -0.49467680f, -0.49751821f, +-0.50035429f, -0.50318497f, -0.50601029f, -0.50883019f, -0.51164466f, +-0.51445359f, -0.51725709f, -0.52005500f, -0.52284735f, -0.52563411f, +-0.52841520f, -0.53119069f, -0.53396046f, -0.53672451f, -0.53948283f, +-0.54223537f, -0.54498214f, -0.54772300f, -0.55045801f, -0.55318713f, +-0.55591035f, -0.55862761f, -0.56133890f, -0.56404412f, -0.56674337f, +-0.56943649f, -0.57212353f, -0.57480448f, -0.57747924f, -0.58014780f, +-0.58281022f, -0.58546633f, -0.58811617f, -0.59075975f, -0.59339696f, +-0.59602785f, -0.59865236f, -0.60127044f, -0.60388207f, -0.60648727f, +-0.60908598f, -0.61167812f, -0.61426371f, -0.61684275f, -0.61941516f, +-0.62198097f, -0.62454009f, -0.62709254f, -0.62963831f, -0.63217729f, +-0.63470948f, -0.63723493f, -0.63975352f, -0.64226526f, -0.64477009f, +-0.64726806f, -0.64975911f, -0.65224314f, -0.65472025f, -0.65719032f, +-0.65965337f, -0.66210932f, -0.66455823f, -0.66700000f, -0.66943461f, +-0.67186207f, -0.67428231f, -0.67669535f, -0.67910111f, -0.68149966f, +-0.68389088f, -0.68627477f, -0.68865126f, -0.69102043f, -0.69338220f, +-0.69573659f, -0.69808346f, -0.70042288f, -0.70275480f, -0.70507920f, +-0.70739603f, -0.70970529f, -0.71200693f, -0.71430099f, -0.71658736f, +-0.71886611f, -0.72113711f, -0.72340041f, -0.72565591f, -0.72790372f, +-0.73014367f, -0.73237586f, -0.73460019f, -0.73681659f, -0.73902518f, +-0.74122584f, -0.74341851f, -0.74560326f, -0.74778003f, -0.74994880f, +-0.75210953f, -0.75426215f, -0.75640678f, -0.75854325f, -0.76067162f, +-0.76279181f, -0.76490390f, -0.76700771f, -0.76910341f, -0.77119076f, +-0.77326995f, -0.77534080f, -0.77740335f, -0.77945763f, -0.78150350f, +-0.78354102f, -0.78557014f, -0.78759086f, -0.78960317f, -0.79160696f, +-0.79360235f, -0.79558921f, -0.79756755f, -0.79953730f, -0.80149853f, +-0.80345118f, -0.80539525f, -0.80733067f, -0.80925739f, -0.81117553f, +-0.81308490f, -0.81498563f, -0.81687760f, -0.81876087f, -0.82063532f, +-0.82250100f, -0.82435787f, -0.82620591f, -0.82804507f, -0.82987541f, +-0.83169687f, -0.83350939f, -0.83531296f, -0.83710766f, -0.83889335f, +-0.84067005f, -0.84243774f, -0.84419644f, -0.84594607f, -0.84768665f, +-0.84941816f, -0.85114056f, -0.85285389f, -0.85455805f, -0.85625303f, +-0.85793889f, -0.85961550f, -0.86128294f, -0.86294121f, -0.86459017f, +-0.86622989f, -0.86786032f, -0.86948150f, -0.87109333f, -0.87269586f, +-0.87428904f, -0.87587279f, -0.87744725f, -0.87901229f, -0.88056785f, +-0.88211405f, -0.88365078f, -0.88517809f, -0.88669586f, -0.88820416f, +-0.88970292f, -0.89119220f, -0.89267188f, -0.89414203f, -0.89560264f, +-0.89705360f, -0.89849502f, -0.89992678f, -0.90134889f, -0.90276134f, +-0.90416414f, -0.90555727f, -0.90694070f, -0.90831441f, -0.90967834f, +-0.91103262f, -0.91237706f, -0.91371179f, -0.91503674f, -0.91635185f, +-0.91765714f, -0.91895264f, -0.92023826f, -0.92151409f, -0.92277998f, +-0.92403603f, -0.92528218f, -0.92651838f, -0.92774469f, -0.92896110f, +-0.93016750f, -0.93136400f, -0.93255049f, -0.93372697f, -0.93489349f, +-0.93604994f, -0.93719643f, -0.93833286f, -0.93945926f, -0.94057560f, +-0.94168180f, -0.94277799f, -0.94386405f, -0.94494003f, -0.94600588f, +-0.94706154f, -0.94810712f, -0.94914252f, -0.95016778f, -0.95118284f, +-0.95218778f, -0.95318246f, -0.95416695f, -0.95514119f, -0.95610523f, +-0.95705903f, -0.95800257f, -0.95893586f, -0.95985889f, -0.96077162f, +-0.96167403f, -0.96256620f, -0.96344805f, -0.96431959f, -0.96518075f, +-0.96603161f, -0.96687216f, -0.96770233f, -0.96852213f, -0.96933156f, +-0.97013056f, -0.97091925f, -0.97169751f, -0.97246534f, -0.97322279f, +-0.97396982f, -0.97470641f, -0.97543252f, -0.97614825f, -0.97685349f, +-0.97754824f, -0.97823256f, -0.97890645f, -0.97956979f, -0.98022264f, +-0.98086500f, -0.98149687f, -0.98211825f, -0.98272908f, -0.98332942f, +-0.98391914f, -0.98449844f, -0.98506713f, -0.98562527f, -0.98617285f, +-0.98670989f, -0.98723638f, -0.98775226f, -0.98825759f, -0.98875231f, +-0.98923647f, -0.98971003f, -0.99017298f, -0.99062532f, -0.99106705f, +-0.99149817f, -0.99191868f, -0.99232858f, -0.99272782f, -0.99311644f, +-0.99349445f, -0.99386179f, -0.99421853f, -0.99456459f, -0.99489999f, +-0.99522477f, -0.99553883f, -0.99584228f, -0.99613506f, -0.99641716f, +-0.99668860f, -0.99694937f, -0.99719942f, -0.99743885f, -0.99766755f, +-0.99788558f, -0.99809295f, -0.99828959f, -0.99847561f, -0.99865085f, +-0.99881548f, -0.99896932f, -0.99911255f, -0.99924499f, -0.99936682f, +-0.99947786f, -0.99957830f, -0.99966794f, -0.99974692f, -0.99981517f, +-0.99987274f, -0.99991959f, -0.99995571f, -0.99998116f, -0.99999589f, +0.99999964f, 0.99997288f, 0.99990326f, 0.99979085f, 0.99963558f, +0.99943751f, 0.99919659f, 0.99891287f, 0.99858636f, 0.99821711f, +0.99780506f, 0.99735034f, 0.99685282f, 0.99631262f, 0.99572974f, +0.99510419f, 0.99443603f, 0.99372530f, 0.99297196f, 0.99217612f, +0.99133772f, 0.99045694f, 0.98953366f, 0.98856801f, 0.98756003f, +0.98650974f, 0.98541719f, 0.98428243f, 0.98310548f, 0.98188645f, +0.98062533f, 0.97932225f, 0.97797716f, 0.97659022f, 0.97516143f, +0.97369087f, 0.97217858f, 0.97062469f, 0.96902919f, 0.96739221f, +0.96571374f, 0.96399397f, 0.96223283f, 0.96043050f, 0.95858705f, +0.95670253f, 0.95477700f, 0.95281059f, 0.95080340f, 0.94875544f, +0.94666684f, 0.94453770f, 0.94236809f, 0.94015813f, 0.93790787f, +0.93561745f, 0.93328691f, 0.93091643f, 0.92850608f, 0.92605597f, +0.92356616f, 0.92103678f, 0.91846794f, 0.91585976f, 0.91321236f, +0.91052586f, 0.90780038f, 0.90503591f, 0.90223277f, 0.89939094f, +0.89651060f, 0.89359182f, 0.89063478f, 0.88763964f, 0.88460642f, +0.88153529f, 0.87842643f, 0.87527996f, 0.87209594f, 0.86887461f, +0.86561602f, 0.86232042f, 0.85898781f, 0.85561842f, 0.85221243f, +0.84876984f, 0.84529096f, 0.84177583f, 0.83822471f, 0.83463764f, +0.83101481f, 0.82735640f, 0.82366252f, 0.81993335f, 0.81616908f, +0.81236988f, 0.80853581f, 0.80466717f, 0.80076402f, 0.79682660f, +0.79285502f, 0.78884947f, 0.78481019f, 0.78073722f, 0.77663082f, +0.77249116f, 0.76831841f, 0.76411277f, 0.75987434f, 0.75560343f, +0.75130010f, 0.74696463f, 0.74259710f, 0.73819780f, 0.73376691f, +0.72930455f, 0.72481096f, 0.72028631f, 0.71573079f, 0.71114463f, +0.70652801f, 0.70188117f, 0.69720417f, 0.69249737f, 0.68776089f, +0.68299496f, 0.67819971f, 0.67337549f, 0.66852236f, 0.66364062f, +0.65873051f, 0.65379208f, 0.64882571f, 0.64383155f, 0.63880974f, +0.63376063f, 0.62868434f, 0.62358117f, 0.61845124f, 0.61329484f, +0.60811216f, 0.60290343f, 0.59766883f, 0.59240872f, 0.58712316f, +0.58181250f, 0.57647687f, 0.57111657f, 0.56573176f, 0.56032276f, +0.55488980f, 0.54943299f, 0.54395270f, 0.53844911f, 0.53292239f, +0.52737290f, 0.52180082f, 0.51620632f, 0.51058978f, 0.50495136f, +0.49929130f, 0.49360985f, 0.48790723f, 0.48218375f, 0.47643960f, +0.47067502f, 0.46489030f, 0.45908567f, 0.45326138f, 0.44741765f, +0.44155475f, 0.43567297f, 0.42977250f, 0.42385364f, 0.41791660f, +0.41196167f, 0.40598908f, 0.39999911f, 0.39399201f, 0.38796803f, +0.38192743f, 0.37587047f, 0.36979741f, 0.36370850f, 0.35760403f, +0.35148421f, 0.34534934f, 0.33919969f, 0.33303553f, 0.32685706f, +0.32066461f, 0.31445843f, 0.30823877f, 0.30200592f, 0.29576012f, +0.28950164f, 0.28323078f, 0.27694780f, 0.27065292f, 0.26434645f, +0.25802869f, 0.25169984f, 0.24536023f, 0.23901010f, 0.23264973f, +0.22627939f, 0.21989937f, 0.21350993f, 0.20711134f, 0.20070387f, +0.19428782f, 0.18786344f, 0.18143101f, 0.17499080f, 0.16854310f, +0.16208819f, 0.15562633f, 0.14915779f, 0.14268288f, 0.13620184f, +0.12971498f, 0.12322257f, 0.11672486f, 0.11022217f, 0.10371475f, +0.097202882f, 0.090686858f, 0.084166944f, 0.077643424f, 0.071116582f, +0.064586692f, 0.058054037f, 0.051518895f, 0.044981543f, 0.038442269f, +0.031901345f, 0.025359053f, 0.018815678f, 0.012271495f, 0.0057267868f, +-0.00081816671f, -0.0073630852f, -0.013907688f, -0.020451695f, -0.026994826f, +-0.033536803f, -0.040077340f, -0.046616159f, -0.053152986f, -0.059687532f, +-0.066219524f, -0.072748676f, -0.079274714f, -0.085797355f, -0.092316322f, +-0.098831341f, -0.10534211f, -0.11184838f, -0.11834986f, -0.12484626f, +-0.13133731f, -0.13782275f, -0.14430228f, -0.15077563f, -0.15724251f, +-0.16370267f, -0.17015581f, -0.17660165f, -0.18303993f, -0.18947038f, +-0.19589271f, -0.20230664f, -0.20871192f, -0.21510825f, -0.22149536f, +-0.22787298f, -0.23424086f, -0.24059868f, -0.24694622f, -0.25328314f, +-0.25960925f, -0.26592422f, -0.27222782f, -0.27851975f, -0.28479972f, +-0.29106751f, -0.29732284f, -0.30356544f, -0.30979502f, -0.31601134f, +-0.32221413f, -0.32840309f, -0.33457801f, -0.34073856f, -0.34688455f, +-0.35301566f, -0.35913166f, -0.36523229f, -0.37131724f, -0.37738630f, +-0.38343921f, -0.38947567f, -0.39549544f, -0.40149832f, -0.40748394f, +-0.41345215f, -0.41940263f, -0.42533514f, -0.43124944f, -0.43714526f, +-0.44302234f, -0.44888046f, -0.45471936f, -0.46053877f, -0.46633846f, +-0.47211814f, -0.47787762f, -0.48361665f, -0.48933494f, -0.49503228f, +-0.50070840f, -0.50636309f, -0.51199609f, -0.51760709f, -0.52319598f, +-0.52876246f, -0.53430629f, -0.53982723f, -0.54532504f, -0.55079949f, +-0.55625033f, -0.56167740f, -0.56708032f, -0.57245898f, -0.57781315f, +-0.58314258f, -0.58844697f, -0.59372622f, -0.59897995f, -0.60420811f, +-0.60941035f, -0.61458647f, -0.61973625f, -0.62485951f, -0.62995601f, +-0.63502556f, -0.64006782f, -0.64508271f, -0.65007001f, -0.65502942f, +-0.65996075f, -0.66486382f, -0.66973841f, -0.67458433f, -0.67940134f, +-0.68418926f, -0.68894786f, -0.69367695f, -0.69837630f, -0.70304573f, +-0.70768511f, -0.71229410f, -0.71687263f, -0.72142041f, -0.72593731f, +-0.73042315f, -0.73487765f, -0.73930067f, -0.74369204f, -0.74805158f, +-0.75237900f, -0.75667429f, -0.76093709f, -0.76516730f, -0.76936477f, +-0.77352923f, -0.77766061f, -0.78175867f, -0.78582323f, -0.78985411f, +-0.79385114f, -0.79781419f, -0.80174309f, -0.80563760f, -0.80949765f, +-0.81332302f, -0.81711352f, -0.82086903f, -0.82458937f, -0.82827437f, +-0.83192390f, -0.83553779f, -0.83911592f, -0.84265804f, -0.84616417f, +-0.84963393f, -0.85306740f, -0.85646427f, -0.85982448f, -0.86314780f, +-0.86643422f, -0.86968350f, -0.87289548f, -0.87607014f, -0.87920725f, +-0.88230664f, -0.88536829f, -0.88839203f, -0.89137769f, -0.89432514f, +-0.89723432f, -0.90010506f, -0.90293723f, -0.90573072f, -0.90848541f, +-0.91120118f, -0.91387796f, -0.91651553f, -0.91911387f, -0.92167282f, +-0.92419231f, -0.92667222f, -0.92911243f, -0.93151283f, -0.93387336f, +-0.93619382f, -0.93847424f, -0.94071442f, -0.94291431f, -0.94507378f, +-0.94719279f, -0.94927126f, -0.95130903f, -0.95330608f, -0.95526224f, +-0.95717752f, -0.95905179f, -0.96088499f, -0.96267700f, -0.96442777f, +-0.96613729f, -0.96780539f, -0.96943200f, -0.97101706f, -0.97256058f, +-0.97406244f, -0.97552258f, -0.97694093f, -0.97831738f, -0.97965199f, +-0.98094457f, -0.98219514f, -0.98340368f, -0.98457009f, -0.98569429f, +-0.98677629f, -0.98781598f, -0.98881340f, -0.98976845f, -0.99068111f, +-0.99155134f, -0.99237907f, -0.99316430f, -0.99390697f, -0.99460709f, +-0.99526459f, -0.99587947f, -0.99645168f, -0.99698120f, -0.99746799f, +-0.99791211f, -0.99831343f, -0.99867201f, -0.99898779f, -0.99926084f, +-0.99949104f, -0.99967843f, -0.99982297f, -0.99992472f, -0.99998361f, +0.99999869f, 0.99989158f, 0.99961317f, 0.99916345f, 0.99854255f, +0.99775058f, 0.99678761f, 0.99565387f, 0.99434954f, 0.99287480f, +0.99122995f, 0.98941529f, 0.98743105f, 0.98527765f, 0.98295540f, +0.98046476f, 0.97780609f, 0.97497988f, 0.97198665f, 0.96882683f, +0.96550101f, 0.96200979f, 0.95835376f, 0.95453346f, 0.95054960f, +0.94640291f, 0.94209403f, 0.93762374f, 0.93299282f, 0.92820197f, +0.92325211f, 0.91814411f, 0.91287869f, 0.90745693f, 0.90187967f, +0.89614785f, 0.89026248f, 0.88422459f, 0.87803519f, 0.87169534f, +0.86520612f, 0.85856867f, 0.85178405f, 0.84485358f, 0.83777827f, +0.83055943f, 0.82319832f, 0.81569612f, 0.80805415f, 0.80027372f, +0.79235619f, 0.78430289f, 0.77611518f, 0.76779449f, 0.75934225f, +0.75075996f, 0.74204898f, 0.73321080f, 0.72424710f, 0.71515924f, +0.70594883f, 0.69661748f, 0.68716675f, 0.67759830f, 0.66791373f, +0.65811473f, 0.64820296f, 0.63818014f, 0.62804794f, 0.61780810f, +0.60746247f, 0.59701276f, 0.58646071f, 0.57580817f, 0.56505698f, +0.55420899f, 0.54326600f, 0.53222996f, 0.52110273f, 0.50988621f, +0.49858227f, 0.48719296f, 0.47572014f, 0.46416581f, 0.45253196f, +0.44082057f, 0.42903364f, 0.41717321f, 0.40524128f, 0.39323992f, +0.38117120f, 0.36903715f, 0.35683987f, 0.34458145f, 0.33226398f, +0.31988961f, 0.30746040f, 0.29497850f, 0.28244606f, 0.26986524f, +0.25723818f, 0.24456702f, 0.23185398f, 0.21910121f, 0.20631088f, +0.19348522f, 0.18062639f, 0.16773662f, 0.15481812f, 0.14187308f, +0.12890373f, 0.11591230f, 0.10290100f, 0.089872077f, 0.076827750f, +0.063770257f, 0.050701842f, 0.037624735f, 0.024541186f, 0.011453429f, +-0.0016362892f, -0.014725727f, -0.027812643f, -0.040894791f, -0.053969935f, +-0.067035832f, -0.080090240f, -0.093130924f, -0.10615565f, -0.11916219f, +-0.13214831f, -0.14511178f, -0.15805040f, -0.17096193f, -0.18384418f, +-0.19669491f, -0.20951195f, -0.22229309f, -0.23503613f, -0.24773891f, +-0.26039925f, -0.27301496f, -0.28558388f, -0.29810387f, -0.31057280f, +-0.32298848f, -0.33534884f, -0.34765175f, -0.35989508f, -0.37207675f, +-0.38419467f, -0.39624676f, -0.40823093f, -0.42014518f, -0.43198743f, +-0.44375566f, -0.45544785f, -0.46706200f, -0.47859612f, -0.49004826f, +-0.50141639f, -0.51269865f, -0.52389306f, -0.53499764f, -0.54601061f, +-0.55693001f, -0.56775403f, -0.57848072f, -0.58910829f, -0.59963489f, +-0.61005878f, -0.62037814f, -0.63059121f, -0.64069623f, -0.65069145f, +-0.66057515f, -0.67034572f, -0.68000144f, -0.68954057f, -0.69896162f, +-0.70826286f, -0.71744281f, -0.72649974f, -0.73543227f, -0.74423873f, +-0.75291771f, -0.76146764f, -0.76988715f, -0.77817470f, -0.78632891f, +-0.79434842f, -0.80223179f, -0.80997771f, -0.81758487f, -0.82505190f, +-0.83237761f, -0.83956063f, -0.84659988f, -0.85349399f, -0.86024189f, +-0.86684239f, -0.87329435f, -0.87959671f, -0.88574833f, -0.89174819f, +-0.89759529f, -0.90328854f, -0.90882701f, -0.91420978f, -0.91943592f, +-0.92450452f, -0.92941469f, -0.93416560f, -0.93875647f, -0.94318646f, +-0.94745487f, -0.95156091f, -0.95550388f, -0.95928317f, -0.96289814f, +-0.96634805f, -0.96963239f, -0.97275060f, -0.97570217f, -0.97848648f, +-0.98110318f, -0.98355180f, -0.98583186f, -0.98794299f, -0.98988485f, +-0.99165714f, -0.99325943f, -0.99469161f, -0.99595332f, -0.99704438f, +-0.99796462f, -0.99871385f, -0.99929196f, -0.99969882f, -0.99993443f, +0.99999464f, 0.99956632f, 0.99845290f, 0.99665523f, 0.99417448f, +0.99101239f, 0.98717111f, 0.98265326f, 0.97746199f, 0.97160077f, +0.96507365f, 0.95788515f, 0.95004016f, 0.94154406f, 0.93240267f, +0.92262226f, 0.91220951f, 0.90117162f, 0.88951606f, 0.87725091f, +0.86438453f, 0.85092574f, 0.83688372f, 0.82226819f, 0.80708915f, +0.79135692f, 0.77508235f, 0.75827658f, 0.74095112f, 0.72311783f, +0.70478898f, 0.68597710f, 0.66669506f, 0.64695615f, 0.62677377f, +0.60616189f, 0.58513457f, 0.56370622f, 0.54189157f, 0.51970547f, +0.49716324f, 0.47428027f, 0.45107225f, 0.42755505f, 0.40374488f, +0.37965798f, 0.35531086f, 0.33072025f, 0.30590299f, 0.28087607f, +0.25565663f, 0.23026201f, 0.20470956f, 0.17901683f, 0.15320139f, +0.12728097f, 0.10127331f, 0.075196236f, 0.049067631f, 0.022905400f, +-0.0032725304f, -0.029448219f, -0.055603724f, -0.081721120f, -0.10778251f, +-0.13377003f, -0.15966587f, -0.18545228f, -0.21111161f, -0.23662624f, +-0.26197869f, -0.28715160f, -0.31212771f, -0.33688989f, -0.36142120f, +-0.38570482f, -0.40972409f, -0.43346253f, -0.45690393f, -0.48003218f, +-0.50283146f, -0.52528608f, -0.54738069f, -0.56910020f, -0.59042966f, +-0.61135447f, -0.63186026f, -0.65193301f, -0.67155898f, -0.69072473f, +-0.70941705f, -0.72762316f, -0.74533063f, -0.76252723f, -0.77920127f, +-0.79534131f, -0.81093621f, -0.82597536f, -0.84044844f, -0.85434550f, +-0.86765707f, -0.88037395f, -0.89248747f, -0.90398932f, -0.91487163f, +-0.92512697f, -0.93474823f, -0.94372886f, -0.95206273f, -0.95974404f, +-0.96676767f, -0.97312868f, -0.97882277f, -0.98384601f, -0.98819500f, +-0.99186671f, -0.99485862f, -0.99716878f, -0.99879545f, -0.99973762f, +}; +#endif + +static const CELTMode mode48000_960_120 = { +48000, /* Fs */ +120, /* overlap */ +21, /* nbEBands */ +21, /* effEBands */ +{0.85000610f, 0.0000000f, 1.0000000f, 1.0000000f, }, /* preemph */ +eband5ms, /* eBands */ +3, /* maxLM */ +8, /* nbShortMdcts */ +120, /* shortMdctSize */ +11, /* nbAllocVectors */ +band_allocation, /* allocVectors */ +logN400, /* logN */ +window120, /* window */ +{1920, 3, {&fft_state48000_960_0, &fft_state48000_960_1, &fft_state48000_960_2, &fft_state48000_960_3, }, mdct_twiddles960}, /* mdct */ +{392, cache_index50, cache_bits50, cache_caps50}, /* cache */ +}; + +/* List of all the available modes */ +#define TOTAL_MODES 1 +static const CELTMode * const static_mode_list[TOTAL_MODES] = { +&mode48000_960_120, +}; diff --git a/libopus/celt/static_modes_float_arm_ne10.h b/libopus/celt/static_modes_float_arm_ne10.h new file mode 100644 index 0000000..66e1abb --- /dev/null +++ b/libopus/celt/static_modes_float_arm_ne10.h @@ -0,0 +1,404 @@ +/* The contents of this file was automatically generated by + * dump_mode_arm_ne10.c with arguments: 48000 960 + * It contains static definitions for some pre-defined modes. */ +#include <NE10_types.h> + +#ifndef NE10_FFT_PARAMS48000_960 +#define NE10_FFT_PARAMS48000_960 +static const ne10_int32_t ne10_factors_480[64] = { +4, 40, 4, 30, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_240[64] = { +3, 20, 4, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_120[64] = { +3, 10, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_60[64] = { +2, 5, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_fft_cpx_float32_t ne10_twiddles_480[480] = { +{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f}, +{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f}, +{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f}, +{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f}, +{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f}, +{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f}, +{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f}, +{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f}, +{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f}, +{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f}, +{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f}, +{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f}, +{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f}, +{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f}, +{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f}, +{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f}, +{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f}, +{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f}, +{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f}, +{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f}, +{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f}, +{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f}, +{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f}, +{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f}, +{-4.3711388e-08f,-1.0000000f}, {-0.10452851f,-0.99452192f}, {-0.20791174f,-0.97814757f}, +{-0.30901703f,-0.95105648f}, {-0.40673670f,-0.91354543f}, {-0.50000006f,-0.86602533f}, +{-0.58778518f,-0.80901700f}, {-0.66913068f,-0.74314475f}, {-0.74314493f,-0.66913044f}, +{-0.80901700f,-0.58778518f}, {-0.86602539f,-0.50000006f}, {-0.91354549f,-0.40673658f}, +{-0.95105654f,-0.30901679f}, {-0.97814763f,-0.20791161f}, {-0.99452192f,-0.10452849f}, +{1.0000000f,-0.0000000f}, {0.98768836f,-0.15643448f}, {0.95105648f,-0.30901700f}, +{0.89100653f,-0.45399052f}, {0.80901700f,-0.58778524f}, {0.70710677f,-0.70710683f}, +{0.58778524f,-0.80901700f}, {0.45399052f,-0.89100653f}, {0.30901697f,-0.95105654f}, +{0.15643437f,-0.98768836f}, {-4.3711388e-08f,-1.0000000f}, {-0.15643445f,-0.98768836f}, +{-0.30901703f,-0.95105648f}, {-0.45399061f,-0.89100647f}, {-0.58778518f,-0.80901700f}, +{-0.70710677f,-0.70710677f}, {-0.80901700f,-0.58778518f}, {-0.89100659f,-0.45399037f}, +{-0.95105654f,-0.30901679f}, {-0.98768836f,-0.15643445f}, {-1.0000000f,8.7422777e-08f}, +{-0.98768830f,0.15643461f}, {-0.95105654f,0.30901697f}, {-0.89100653f,0.45399055f}, +{-0.80901694f,0.58778536f}, {-0.70710665f,0.70710689f}, {-0.58778507f,0.80901712f}, +{-0.45399022f,0.89100665f}, {-0.30901709f,0.95105648f}, {-0.15643452f,0.98768830f}, +{1.0000000f,-0.0000000f}, {0.99991435f,-0.013089596f}, {0.99965733f,-0.026176950f}, +{0.99922901f,-0.039259817f}, {0.99862951f,-0.052335959f}, {0.99785894f,-0.065403134f}, +{0.99691731f,-0.078459099f}, {0.99580491f,-0.091501623f}, {0.99452192f,-0.10452846f}, +{0.99306846f,-0.11753740f}, {0.99144489f,-0.13052620f}, {0.98965138f,-0.14349262f}, +{0.98768836f,-0.15643448f}, {0.98555607f,-0.16934951f}, {0.98325491f,-0.18223552f}, +{0.98078525f,-0.19509032f}, {0.97814763f,-0.20791170f}, {0.97534233f,-0.22069745f}, +{0.97236991f,-0.23344538f}, {0.96923089f,-0.24615330f}, {0.96592581f,-0.25881904f}, +{0.96245521f,-0.27144045f}, {0.95881975f,-0.28401536f}, {0.95501995f,-0.29654160f}, +{0.95105648f,-0.30901700f}, {0.94693011f,-0.32143945f}, {0.94264150f,-0.33380687f}, +{0.93819129f,-0.34611708f}, {0.93358040f,-0.35836795f}, {0.92880952f,-0.37055743f}, +{0.92387956f,-0.38268346f}, {0.91879117f,-0.39474389f}, {0.91354543f,-0.40673664f}, +{0.90814316f,-0.41865975f}, {0.90258527f,-0.43051112f}, {0.89687270f,-0.44228873f}, +{0.89100653f,-0.45399052f}, {0.88498765f,-0.46561453f}, {0.87881708f,-0.47715878f}, +{0.87249601f,-0.48862126f}, {0.86602545f,-0.50000000f}, {0.85940641f,-0.51129311f}, +{0.85264015f,-0.52249855f}, {0.84572786f,-0.53361452f}, {0.83867055f,-0.54463905f}, +{0.83146960f,-0.55557024f}, {0.82412618f,-0.56640625f}, {0.81664151f,-0.57714522f}, +{0.80901700f,-0.58778524f}, {0.80125380f,-0.59832460f}, {0.79335332f,-0.60876143f}, +{0.78531694f,-0.61909395f}, {0.77714598f,-0.62932038f}, {0.76884180f,-0.63943899f}, +{0.76040596f,-0.64944810f}, {0.75183982f,-0.65934587f}, {0.74314475f,-0.66913062f}, +{0.73432249f,-0.67880076f}, {0.72537434f,-0.68835455f}, {0.71630192f,-0.69779050f}, +{0.70710677f,-0.70710683f}, {0.69779044f,-0.71630198f}, {0.68835455f,-0.72537440f}, +{0.67880070f,-0.73432255f}, {0.66913056f,-0.74314487f}, {0.65934581f,-0.75183982f}, +{0.64944804f,-0.76040596f}, {0.63943899f,-0.76884186f}, {0.62932038f,-0.77714598f}, +{0.61909395f,-0.78531694f}, {0.60876137f,-0.79335338f}, {0.59832460f,-0.80125386f}, +{0.58778524f,-0.80901700f}, {0.57714516f,-0.81664151f}, {0.56640625f,-0.82412618f}, +{0.55557019f,-0.83146960f}, {0.54463899f,-0.83867055f}, {0.53361452f,-0.84572786f}, +{0.52249849f,-0.85264015f}, {0.51129311f,-0.85940641f}, {0.49999997f,-0.86602545f}, +{0.48862118f,-0.87249601f}, {0.47715876f,-0.87881708f}, {0.46561447f,-0.88498765f}, +{0.45399052f,-0.89100653f}, {0.44228867f,-0.89687276f}, {0.43051103f,-0.90258533f}, +{0.41865975f,-0.90814316f}, {0.40673661f,-0.91354549f}, {0.39474380f,-0.91879129f}, +{0.38268343f,-0.92387956f}, {0.37055740f,-0.92880958f}, {0.35836786f,-0.93358046f}, +{0.34611705f,-0.93819135f}, {0.33380681f,-0.94264150f}, {0.32143947f,-0.94693011f}, +{0.30901697f,-0.95105654f}, {0.29654151f,-0.95501995f}, {0.28401533f,-0.95881975f}, +{0.27144039f,-0.96245527f}, {0.25881907f,-0.96592581f}, {0.24615327f,-0.96923089f}, +{0.23344530f,-0.97236991f}, {0.22069745f,-0.97534233f}, {0.20791166f,-0.97814763f}, +{0.19509023f,-0.98078531f}, {0.18223552f,-0.98325491f}, {0.16934945f,-0.98555607f}, +{0.15643437f,-0.98768836f}, {0.14349259f,-0.98965138f}, {0.13052613f,-0.99144489f}, +{0.11753740f,-0.99306846f}, {0.10452842f,-0.99452192f}, {0.091501534f,-0.99580491f}, +{0.078459084f,-0.99691731f}, {0.065403074f,-0.99785894f}, {0.052335974f,-0.99862951f}, +{0.039259788f,-0.99922901f}, {0.026176875f,-0.99965733f}, {0.013089597f,-0.99991435f}, +{1.0000000f,-0.0000000f}, {0.99965733f,-0.026176950f}, {0.99862951f,-0.052335959f}, +{0.99691731f,-0.078459099f}, {0.99452192f,-0.10452846f}, {0.99144489f,-0.13052620f}, +{0.98768836f,-0.15643448f}, {0.98325491f,-0.18223552f}, {0.97814763f,-0.20791170f}, +{0.97236991f,-0.23344538f}, {0.96592581f,-0.25881904f}, {0.95881975f,-0.28401536f}, +{0.95105648f,-0.30901700f}, {0.94264150f,-0.33380687f}, {0.93358040f,-0.35836795f}, +{0.92387956f,-0.38268346f}, {0.91354543f,-0.40673664f}, {0.90258527f,-0.43051112f}, +{0.89100653f,-0.45399052f}, {0.87881708f,-0.47715878f}, {0.86602545f,-0.50000000f}, +{0.85264015f,-0.52249855f}, {0.83867055f,-0.54463905f}, {0.82412618f,-0.56640625f}, +{0.80901700f,-0.58778524f}, {0.79335332f,-0.60876143f}, {0.77714598f,-0.62932038f}, +{0.76040596f,-0.64944810f}, {0.74314475f,-0.66913062f}, {0.72537434f,-0.68835455f}, +{0.70710677f,-0.70710683f}, {0.68835455f,-0.72537440f}, {0.66913056f,-0.74314487f}, +{0.64944804f,-0.76040596f}, {0.62932038f,-0.77714598f}, {0.60876137f,-0.79335338f}, +{0.58778524f,-0.80901700f}, {0.56640625f,-0.82412618f}, {0.54463899f,-0.83867055f}, +{0.52249849f,-0.85264015f}, {0.49999997f,-0.86602545f}, {0.47715876f,-0.87881708f}, +{0.45399052f,-0.89100653f}, {0.43051103f,-0.90258533f}, {0.40673661f,-0.91354549f}, +{0.38268343f,-0.92387956f}, {0.35836786f,-0.93358046f}, {0.33380681f,-0.94264150f}, +{0.30901697f,-0.95105654f}, {0.28401533f,-0.95881975f}, {0.25881907f,-0.96592581f}, +{0.23344530f,-0.97236991f}, {0.20791166f,-0.97814763f}, {0.18223552f,-0.98325491f}, +{0.15643437f,-0.98768836f}, {0.13052613f,-0.99144489f}, {0.10452842f,-0.99452192f}, +{0.078459084f,-0.99691731f}, {0.052335974f,-0.99862951f}, {0.026176875f,-0.99965733f}, +{-4.3711388e-08f,-1.0000000f}, {-0.026176963f,-0.99965733f}, {-0.052336060f,-0.99862951f}, +{-0.078459173f,-0.99691731f}, {-0.10452851f,-0.99452192f}, {-0.13052621f,-0.99144489f}, +{-0.15643445f,-0.98768836f}, {-0.18223560f,-0.98325491f}, {-0.20791174f,-0.97814757f}, +{-0.23344538f,-0.97236991f}, {-0.25881916f,-0.96592581f}, {-0.28401542f,-0.95881969f}, +{-0.30901703f,-0.95105648f}, {-0.33380687f,-0.94264150f}, {-0.35836795f,-0.93358040f}, +{-0.38268352f,-0.92387950f}, {-0.40673670f,-0.91354543f}, {-0.43051112f,-0.90258527f}, +{-0.45399061f,-0.89100647f}, {-0.47715873f,-0.87881708f}, {-0.50000006f,-0.86602533f}, +{-0.52249867f,-0.85264009f}, {-0.54463905f,-0.83867055f}, {-0.56640631f,-0.82412612f}, +{-0.58778518f,-0.80901700f}, {-0.60876143f,-0.79335332f}, {-0.62932050f,-0.77714586f}, +{-0.64944804f,-0.76040596f}, {-0.66913068f,-0.74314475f}, {-0.68835467f,-0.72537428f}, +{-0.70710677f,-0.70710677f}, {-0.72537446f,-0.68835449f}, {-0.74314493f,-0.66913044f}, +{-0.76040596f,-0.64944804f}, {-0.77714604f,-0.62932026f}, {-0.79335332f,-0.60876143f}, +{-0.80901700f,-0.58778518f}, {-0.82412624f,-0.56640613f}, {-0.83867055f,-0.54463899f}, +{-0.85264021f,-0.52249849f}, {-0.86602539f,-0.50000006f}, {-0.87881714f,-0.47715873f}, +{-0.89100659f,-0.45399037f}, {-0.90258527f,-0.43051112f}, {-0.91354549f,-0.40673658f}, +{-0.92387956f,-0.38268328f}, {-0.93358040f,-0.35836792f}, {-0.94264150f,-0.33380675f}, +{-0.95105654f,-0.30901679f}, {-0.95881975f,-0.28401530f}, {-0.96592587f,-0.25881892f}, +{-0.97236991f,-0.23344538f}, {-0.97814763f,-0.20791161f}, {-0.98325491f,-0.18223536f}, +{-0.98768836f,-0.15643445f}, {-0.99144489f,-0.13052608f}, {-0.99452192f,-0.10452849f}, +{-0.99691737f,-0.078459039f}, {-0.99862957f,-0.052335810f}, {-0.99965733f,-0.026176952f}, +{1.0000000f,-0.0000000f}, {0.99922901f,-0.039259817f}, {0.99691731f,-0.078459099f}, +{0.99306846f,-0.11753740f}, {0.98768836f,-0.15643448f}, {0.98078525f,-0.19509032f}, +{0.97236991f,-0.23344538f}, {0.96245521f,-0.27144045f}, {0.95105648f,-0.30901700f}, +{0.93819129f,-0.34611708f}, {0.92387956f,-0.38268346f}, {0.90814316f,-0.41865975f}, +{0.89100653f,-0.45399052f}, {0.87249601f,-0.48862126f}, {0.85264015f,-0.52249855f}, +{0.83146960f,-0.55557024f}, {0.80901700f,-0.58778524f}, {0.78531694f,-0.61909395f}, +{0.76040596f,-0.64944810f}, {0.73432249f,-0.67880076f}, {0.70710677f,-0.70710683f}, +{0.67880070f,-0.73432255f}, {0.64944804f,-0.76040596f}, {0.61909395f,-0.78531694f}, +{0.58778524f,-0.80901700f}, {0.55557019f,-0.83146960f}, {0.52249849f,-0.85264015f}, +{0.48862118f,-0.87249601f}, {0.45399052f,-0.89100653f}, {0.41865975f,-0.90814316f}, +{0.38268343f,-0.92387956f}, {0.34611705f,-0.93819135f}, {0.30901697f,-0.95105654f}, +{0.27144039f,-0.96245527f}, {0.23344530f,-0.97236991f}, {0.19509023f,-0.98078531f}, +{0.15643437f,-0.98768836f}, {0.11753740f,-0.99306846f}, {0.078459084f,-0.99691731f}, +{0.039259788f,-0.99922901f}, {-4.3711388e-08f,-1.0000000f}, {-0.039259877f,-0.99922901f}, +{-0.078459173f,-0.99691731f}, {-0.11753749f,-0.99306846f}, {-0.15643445f,-0.98768836f}, +{-0.19509032f,-0.98078525f}, {-0.23344538f,-0.97236991f}, {-0.27144048f,-0.96245521f}, +{-0.30901703f,-0.95105648f}, {-0.34611711f,-0.93819129f}, {-0.38268352f,-0.92387950f}, +{-0.41865984f,-0.90814310f}, {-0.45399061f,-0.89100647f}, {-0.48862135f,-0.87249595f}, +{-0.52249867f,-0.85264009f}, {-0.55557036f,-0.83146954f}, {-0.58778518f,-0.80901700f}, +{-0.61909389f,-0.78531694f}, {-0.64944804f,-0.76040596f}, {-0.67880076f,-0.73432249f}, +{-0.70710677f,-0.70710677f}, {-0.73432249f,-0.67880070f}, {-0.76040596f,-0.64944804f}, +{-0.78531694f,-0.61909389f}, {-0.80901700f,-0.58778518f}, {-0.83146966f,-0.55557019f}, +{-0.85264021f,-0.52249849f}, {-0.87249607f,-0.48862115f}, {-0.89100659f,-0.45399037f}, +{-0.90814322f,-0.41865960f}, {-0.92387956f,-0.38268328f}, {-0.93819135f,-0.34611690f}, +{-0.95105654f,-0.30901679f}, {-0.96245521f,-0.27144048f}, {-0.97236991f,-0.23344538f}, +{-0.98078531f,-0.19509031f}, {-0.98768836f,-0.15643445f}, {-0.99306846f,-0.11753736f}, +{-0.99691737f,-0.078459039f}, {-0.99922901f,-0.039259743f}, {-1.0000000f,8.7422777e-08f}, +{-0.99922901f,0.039259918f}, {-0.99691731f,0.078459218f}, {-0.99306846f,0.11753753f}, +{-0.98768830f,0.15643461f}, {-0.98078525f,0.19509049f}, {-0.97236985f,0.23344554f}, +{-0.96245515f,0.27144065f}, {-0.95105654f,0.30901697f}, {-0.93819135f,0.34611705f}, +{-0.92387956f,0.38268346f}, {-0.90814316f,0.41865975f}, {-0.89100653f,0.45399055f}, +{-0.87249601f,0.48862129f}, {-0.85264015f,0.52249861f}, {-0.83146960f,0.55557030f}, +{-0.80901694f,0.58778536f}, {-0.78531688f,0.61909401f}, {-0.76040590f,0.64944816f}, +{-0.73432243f,0.67880082f}, {-0.70710665f,0.70710689f}, {-0.67880058f,0.73432261f}, +{-0.64944792f,0.76040608f}, {-0.61909378f,0.78531706f}, {-0.58778507f,0.80901712f}, +{-0.55557001f,0.83146977f}, {-0.52249837f,0.85264033f}, {-0.48862100f,0.87249613f}, +{-0.45399022f,0.89100665f}, {-0.41865945f,0.90814328f}, {-0.38268313f,0.92387968f}, +{-0.34611672f,0.93819147f}, {-0.30901709f,0.95105648f}, {-0.27144054f,0.96245521f}, +{-0.23344545f,0.97236991f}, {-0.19509038f,0.98078525f}, {-0.15643452f,0.98768830f}, +{-0.11753743f,0.99306846f}, {-0.078459114f,0.99691731f}, {-0.039259821f,0.99922901f}, +}; +static const ne10_fft_cpx_float32_t ne10_twiddles_240[240] = { +{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f}, +{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f}, +{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f}, +{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f}, +{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f}, +{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f}, +{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f}, +{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f}, +{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f}, +{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f}, +{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f}, +{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f}, +{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f}, +{1.0000000f,-0.0000000f}, {0.95105648f,-0.30901700f}, {0.80901700f,-0.58778524f}, +{0.58778524f,-0.80901700f}, {0.30901697f,-0.95105654f}, {-4.3711388e-08f,-1.0000000f}, +{-0.30901703f,-0.95105648f}, {-0.58778518f,-0.80901700f}, {-0.80901700f,-0.58778518f}, +{-0.95105654f,-0.30901679f}, {-1.0000000f,8.7422777e-08f}, {-0.95105654f,0.30901697f}, +{-0.80901694f,0.58778536f}, {-0.58778507f,0.80901712f}, {-0.30901709f,0.95105648f}, +{1.0000000f,-0.0000000f}, {0.99965733f,-0.026176950f}, {0.99862951f,-0.052335959f}, +{0.99691731f,-0.078459099f}, {0.99452192f,-0.10452846f}, {0.99144489f,-0.13052620f}, +{0.98768836f,-0.15643448f}, {0.98325491f,-0.18223552f}, {0.97814763f,-0.20791170f}, +{0.97236991f,-0.23344538f}, {0.96592581f,-0.25881904f}, {0.95881975f,-0.28401536f}, +{0.95105648f,-0.30901700f}, {0.94264150f,-0.33380687f}, {0.93358040f,-0.35836795f}, +{0.92387956f,-0.38268346f}, {0.91354543f,-0.40673664f}, {0.90258527f,-0.43051112f}, +{0.89100653f,-0.45399052f}, {0.87881708f,-0.47715878f}, {0.86602545f,-0.50000000f}, +{0.85264015f,-0.52249855f}, {0.83867055f,-0.54463905f}, {0.82412618f,-0.56640625f}, +{0.80901700f,-0.58778524f}, {0.79335332f,-0.60876143f}, {0.77714598f,-0.62932038f}, +{0.76040596f,-0.64944810f}, {0.74314475f,-0.66913062f}, {0.72537434f,-0.68835455f}, +{0.70710677f,-0.70710683f}, {0.68835455f,-0.72537440f}, {0.66913056f,-0.74314487f}, +{0.64944804f,-0.76040596f}, {0.62932038f,-0.77714598f}, {0.60876137f,-0.79335338f}, +{0.58778524f,-0.80901700f}, {0.56640625f,-0.82412618f}, {0.54463899f,-0.83867055f}, +{0.52249849f,-0.85264015f}, {0.49999997f,-0.86602545f}, {0.47715876f,-0.87881708f}, +{0.45399052f,-0.89100653f}, {0.43051103f,-0.90258533f}, {0.40673661f,-0.91354549f}, +{0.38268343f,-0.92387956f}, {0.35836786f,-0.93358046f}, {0.33380681f,-0.94264150f}, +{0.30901697f,-0.95105654f}, {0.28401533f,-0.95881975f}, {0.25881907f,-0.96592581f}, +{0.23344530f,-0.97236991f}, {0.20791166f,-0.97814763f}, {0.18223552f,-0.98325491f}, +{0.15643437f,-0.98768836f}, {0.13052613f,-0.99144489f}, {0.10452842f,-0.99452192f}, +{0.078459084f,-0.99691731f}, {0.052335974f,-0.99862951f}, {0.026176875f,-0.99965733f}, +{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f}, +{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f}, +{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f}, +{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f}, +{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f}, +{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f}, +{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f}, +{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f}, +{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f}, +{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f}, +{-4.3711388e-08f,-1.0000000f}, {-0.052336060f,-0.99862951f}, {-0.10452851f,-0.99452192f}, +{-0.15643445f,-0.98768836f}, {-0.20791174f,-0.97814757f}, {-0.25881916f,-0.96592581f}, +{-0.30901703f,-0.95105648f}, {-0.35836795f,-0.93358040f}, {-0.40673670f,-0.91354543f}, +{-0.45399061f,-0.89100647f}, {-0.50000006f,-0.86602533f}, {-0.54463905f,-0.83867055f}, +{-0.58778518f,-0.80901700f}, {-0.62932050f,-0.77714586f}, {-0.66913068f,-0.74314475f}, +{-0.70710677f,-0.70710677f}, {-0.74314493f,-0.66913044f}, {-0.77714604f,-0.62932026f}, +{-0.80901700f,-0.58778518f}, {-0.83867055f,-0.54463899f}, {-0.86602539f,-0.50000006f}, +{-0.89100659f,-0.45399037f}, {-0.91354549f,-0.40673658f}, {-0.93358040f,-0.35836792f}, +{-0.95105654f,-0.30901679f}, {-0.96592587f,-0.25881892f}, {-0.97814763f,-0.20791161f}, +{-0.98768836f,-0.15643445f}, {-0.99452192f,-0.10452849f}, {-0.99862957f,-0.052335810f}, +{1.0000000f,-0.0000000f}, {0.99691731f,-0.078459099f}, {0.98768836f,-0.15643448f}, +{0.97236991f,-0.23344538f}, {0.95105648f,-0.30901700f}, {0.92387956f,-0.38268346f}, +{0.89100653f,-0.45399052f}, {0.85264015f,-0.52249855f}, {0.80901700f,-0.58778524f}, +{0.76040596f,-0.64944810f}, {0.70710677f,-0.70710683f}, {0.64944804f,-0.76040596f}, +{0.58778524f,-0.80901700f}, {0.52249849f,-0.85264015f}, {0.45399052f,-0.89100653f}, +{0.38268343f,-0.92387956f}, {0.30901697f,-0.95105654f}, {0.23344530f,-0.97236991f}, +{0.15643437f,-0.98768836f}, {0.078459084f,-0.99691731f}, {-4.3711388e-08f,-1.0000000f}, +{-0.078459173f,-0.99691731f}, {-0.15643445f,-0.98768836f}, {-0.23344538f,-0.97236991f}, +{-0.30901703f,-0.95105648f}, {-0.38268352f,-0.92387950f}, {-0.45399061f,-0.89100647f}, +{-0.52249867f,-0.85264009f}, {-0.58778518f,-0.80901700f}, {-0.64944804f,-0.76040596f}, +{-0.70710677f,-0.70710677f}, {-0.76040596f,-0.64944804f}, {-0.80901700f,-0.58778518f}, +{-0.85264021f,-0.52249849f}, {-0.89100659f,-0.45399037f}, {-0.92387956f,-0.38268328f}, +{-0.95105654f,-0.30901679f}, {-0.97236991f,-0.23344538f}, {-0.98768836f,-0.15643445f}, +{-0.99691737f,-0.078459039f}, {-1.0000000f,8.7422777e-08f}, {-0.99691731f,0.078459218f}, +{-0.98768830f,0.15643461f}, {-0.97236985f,0.23344554f}, {-0.95105654f,0.30901697f}, +{-0.92387956f,0.38268346f}, {-0.89100653f,0.45399055f}, {-0.85264015f,0.52249861f}, +{-0.80901694f,0.58778536f}, {-0.76040590f,0.64944816f}, {-0.70710665f,0.70710689f}, +{-0.64944792f,0.76040608f}, {-0.58778507f,0.80901712f}, {-0.52249837f,0.85264033f}, +{-0.45399022f,0.89100665f}, {-0.38268313f,0.92387968f}, {-0.30901709f,0.95105648f}, +{-0.23344545f,0.97236991f}, {-0.15643452f,0.98768830f}, {-0.078459114f,0.99691731f}, +}; +static const ne10_fft_cpx_float32_t ne10_twiddles_120[120] = { +{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f}, +{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f}, +{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f}, +{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f}, +{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f}, +{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f}, +{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f}, +{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f}, +{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f}, +{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f}, +{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f}, +{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f}, +{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f}, +{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f}, +{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f}, +{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f}, +{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f}, +{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f}, +{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f}, +{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f}, +{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f}, +{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f}, +{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f}, +{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f}, +{-4.3711388e-08f,-1.0000000f}, {-0.10452851f,-0.99452192f}, {-0.20791174f,-0.97814757f}, +{-0.30901703f,-0.95105648f}, {-0.40673670f,-0.91354543f}, {-0.50000006f,-0.86602533f}, +{-0.58778518f,-0.80901700f}, {-0.66913068f,-0.74314475f}, {-0.74314493f,-0.66913044f}, +{-0.80901700f,-0.58778518f}, {-0.86602539f,-0.50000006f}, {-0.91354549f,-0.40673658f}, +{-0.95105654f,-0.30901679f}, {-0.97814763f,-0.20791161f}, {-0.99452192f,-0.10452849f}, +{1.0000000f,-0.0000000f}, {0.98768836f,-0.15643448f}, {0.95105648f,-0.30901700f}, +{0.89100653f,-0.45399052f}, {0.80901700f,-0.58778524f}, {0.70710677f,-0.70710683f}, +{0.58778524f,-0.80901700f}, {0.45399052f,-0.89100653f}, {0.30901697f,-0.95105654f}, +{0.15643437f,-0.98768836f}, {-4.3711388e-08f,-1.0000000f}, {-0.15643445f,-0.98768836f}, +{-0.30901703f,-0.95105648f}, {-0.45399061f,-0.89100647f}, {-0.58778518f,-0.80901700f}, +{-0.70710677f,-0.70710677f}, {-0.80901700f,-0.58778518f}, {-0.89100659f,-0.45399037f}, +{-0.95105654f,-0.30901679f}, {-0.98768836f,-0.15643445f}, {-1.0000000f,8.7422777e-08f}, +{-0.98768830f,0.15643461f}, {-0.95105654f,0.30901697f}, {-0.89100653f,0.45399055f}, +{-0.80901694f,0.58778536f}, {-0.70710665f,0.70710689f}, {-0.58778507f,0.80901712f}, +{-0.45399022f,0.89100665f}, {-0.30901709f,0.95105648f}, {-0.15643452f,0.98768830f}, +}; +static const ne10_fft_cpx_float32_t ne10_twiddles_60[60] = { +{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f}, +{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f}, +{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f}, +{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f}, +{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f}, +{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f}, +{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f}, +{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f}, +{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f}, +{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f}, +{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f}, +{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f}, +{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f}, +{1.0000000f,-0.0000000f}, {0.95105648f,-0.30901700f}, {0.80901700f,-0.58778524f}, +{0.58778524f,-0.80901700f}, {0.30901697f,-0.95105654f}, {-4.3711388e-08f,-1.0000000f}, +{-0.30901703f,-0.95105648f}, {-0.58778518f,-0.80901700f}, {-0.80901700f,-0.58778518f}, +{-0.95105654f,-0.30901679f}, {-1.0000000f,8.7422777e-08f}, {-0.95105654f,0.30901697f}, +{-0.80901694f,0.58778536f}, {-0.58778507f,0.80901712f}, {-0.30901709f,0.95105648f}, +}; +static const ne10_fft_state_float32_t ne10_fft_state_float32_t_480 = { +120, +(ne10_int32_t *)ne10_factors_480, +(ne10_fft_cpx_float32_t *)ne10_twiddles_480, +NULL, +(ne10_fft_cpx_float32_t *)&ne10_twiddles_480[120], +/* is_forward_scaled = true */ +(ne10_int32_t) 1, +/* is_backward_scaled = false */ +(ne10_int32_t) 0, +}; +static const arch_fft_state cfg_arch_480 = { +1, +(void *)&ne10_fft_state_float32_t_480, +}; + +static const ne10_fft_state_float32_t ne10_fft_state_float32_t_240 = { +60, +(ne10_int32_t *)ne10_factors_240, +(ne10_fft_cpx_float32_t *)ne10_twiddles_240, +NULL, +(ne10_fft_cpx_float32_t *)&ne10_twiddles_240[60], +/* is_forward_scaled = true */ +(ne10_int32_t) 1, +/* is_backward_scaled = false */ +(ne10_int32_t) 0, +}; +static const arch_fft_state cfg_arch_240 = { +1, +(void *)&ne10_fft_state_float32_t_240, +}; + +static const ne10_fft_state_float32_t ne10_fft_state_float32_t_120 = { +30, +(ne10_int32_t *)ne10_factors_120, +(ne10_fft_cpx_float32_t *)ne10_twiddles_120, +NULL, +(ne10_fft_cpx_float32_t *)&ne10_twiddles_120[30], +/* is_forward_scaled = true */ +(ne10_int32_t) 1, +/* is_backward_scaled = false */ +(ne10_int32_t) 0, +}; +static const arch_fft_state cfg_arch_120 = { +1, +(void *)&ne10_fft_state_float32_t_120, +}; + +static const ne10_fft_state_float32_t ne10_fft_state_float32_t_60 = { +15, +(ne10_int32_t *)ne10_factors_60, +(ne10_fft_cpx_float32_t *)ne10_twiddles_60, +NULL, +(ne10_fft_cpx_float32_t *)&ne10_twiddles_60[15], +/* is_forward_scaled = true */ +(ne10_int32_t) 1, +/* is_backward_scaled = false */ +(ne10_int32_t) 0, +}; +static const arch_fft_state cfg_arch_60 = { +1, +(void *)&ne10_fft_state_float32_t_60, +}; + +#endif /* end NE10_FFT_PARAMS48000_960 */ diff --git a/libopus/celt/tests/test_unit_cwrs32.c b/libopus/celt/tests/test_unit_cwrs32.c new file mode 100644 index 0000000..36dd8af --- /dev/null +++ b/libopus/celt/tests/test_unit_cwrs32.c @@ -0,0 +1,161 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation, Mozilla Corporation, + Gregory Maxwell + Written by Jean-Marc Valin, Gregory Maxwell, and Timothy B. Terriberry */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stdio.h> +#include <string.h> + +#ifndef CUSTOM_MODES +#define CUSTOM_MODES +#else +#define TEST_CUSTOM_MODES +#endif + +#define CELT_C +#include "stack_alloc.h" +#include "entenc.c" +#include "entdec.c" +#include "entcode.c" +#include "cwrs.c" +#include "mathops.c" +#include "rate.h" + +#define NMAX (240) +#define KMAX (128) + +#ifdef TEST_CUSTOM_MODES + +#define NDIMS (44) +static const int pn[NDIMS]={ + 2, 3, 4, 5, 6, 7, 8, 9, 10, + 11, 12, 13, 14, 15, 16, 18, 20, 22, + 24, 26, 28, 30, 32, 36, 40, 44, 48, + 52, 56, 60, 64, 72, 80, 88, 96, 104, + 112, 120, 128, 144, 160, 176, 192, 208 +}; +static const int pkmax[NDIMS]={ + 128, 128, 128, 128, 88, 52, 36, 26, 22, + 18, 16, 15, 13, 12, 12, 11, 10, 9, + 9, 8, 8, 7, 7, 7, 7, 6, 6, + 6, 6, 6, 5, 5, 5, 5, 5, 5, + 4, 4, 4, 4, 4, 4, 4, 4 +}; + +#else /* TEST_CUSTOM_MODES */ + +#define NDIMS (22) +static const int pn[NDIMS]={ + 2, 3, 4, 6, 8, 9, 11, 12, 16, + 18, 22, 24, 32, 36, 44, 48, 64, 72, + 88, 96, 144, 176 +}; +static const int pkmax[NDIMS]={ + 128, 128, 128, 88, 36, 26, 18, 16, 12, + 11, 9, 9, 7, 7, 6, 6, 5, 5, + 5, 5, 4, 4 +}; + +#endif + +int main(void){ + int t; + int n; + ALLOC_STACK; + for(t=0;t<NDIMS;t++){ + int pseudo; + n=pn[t]; + for(pseudo=1;pseudo<41;pseudo++) + { + int k; +#if defined(SMALL_FOOTPRINT) + opus_uint32 uu[KMAX+2U]; +#endif + opus_uint32 inc; + opus_uint32 nc; + opus_uint32 i; + k=get_pulses(pseudo); + if (k>pkmax[t])break; + printf("Testing CWRS with N=%i, K=%i...\n",n,k); +#if defined(SMALL_FOOTPRINT) + nc=ncwrs_urow(n,k,uu); +#else + nc=CELT_PVQ_V(n,k); +#endif + inc=nc/20000; + if(inc<1)inc=1; + for(i=0;i<nc;i+=inc){ +#if defined(SMALL_FOOTPRINT) + opus_uint32 u[KMAX+2U]; +#endif + int y[NMAX]; + int sy; + opus_uint32 v; + opus_uint32 ii; + int j; +#if defined(SMALL_FOOTPRINT) + memcpy(u,uu,(k+2U)*sizeof(*u)); + cwrsi(n,k,i,y,u); +#else + cwrsi(n,k,i,y); +#endif + sy=0; + for(j=0;j<n;j++)sy+=abs(y[j]); + if(sy!=k){ + fprintf(stderr,"N=%d Pulse count mismatch in cwrsi (%d!=%d).\n", + n,sy,k); + return 99; + } + /*printf("%6u of %u:",i,nc); + for(j=0;j<n;j++)printf(" %+3i",y[j]); + printf(" ->");*/ +#if defined(SMALL_FOOTPRINT) + ii=icwrs(n,k,&v,y,u); +#else + ii=icwrs(n,y); + v=CELT_PVQ_V(n,k); +#endif + if(ii!=i){ + fprintf(stderr,"Combination-index mismatch (%lu!=%lu).\n", + (long)ii,(long)i); + return 1; + } + if(v!=nc){ + fprintf(stderr,"Combination count mismatch (%lu!=%lu).\n", + (long)v,(long)nc); + return 2; + } + /*printf(" %6u\n",i);*/ + } + /*printf("\n");*/ + } + } + return 0; +} diff --git a/libopus/celt/tests/test_unit_dft.c b/libopus/celt/tests/test_unit_dft.c new file mode 100644 index 0000000..70f8f49 --- /dev/null +++ b/libopus/celt/tests/test_unit_dft.c @@ -0,0 +1,179 @@ +/* Copyright (c) 2008 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stdio.h> + +#include "stack_alloc.h" +#include "kiss_fft.h" +#include "mathops.h" +#include "modes.h" + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +int ret = 0; + +void check(kiss_fft_cpx * in,kiss_fft_cpx * out,int nfft,int isinverse) +{ + int bin,k; + double errpow=0,sigpow=0, snr; + + for (bin=0;bin<nfft;++bin) { + double ansr = 0; + double ansi = 0; + double difr; + double difi; + + for (k=0;k<nfft;++k) { + double phase = -2*M_PI*bin*k/nfft; + double re = cos(phase); + double im = sin(phase); + if (isinverse) + im = -im; + + if (!isinverse) + { + re /= nfft; + im /= nfft; + } + + ansr += in[k].r * re - in[k].i * im; + ansi += in[k].r * im + in[k].i * re; + } + /*printf ("%d %d ", (int)ansr, (int)ansi);*/ + difr = ansr - out[bin].r; + difi = ansi - out[bin].i; + errpow += difr*difr + difi*difi; + sigpow += ansr*ansr+ansi*ansi; + } + snr = 10*log10(sigpow/errpow); + printf("nfft=%d inverse=%d,snr = %f\n",nfft,isinverse,snr ); + if (snr<60) { + printf( "** poor snr: %f ** \n", snr); + ret = 1; + } +} + +void test1d(int nfft,int isinverse,int arch) +{ + size_t buflen = sizeof(kiss_fft_cpx)*nfft; + kiss_fft_cpx *in; + kiss_fft_cpx *out; + int k; +#ifdef CUSTOM_MODES + kiss_fft_state *cfg = opus_fft_alloc(nfft,0,0,arch); +#else + int id; + const kiss_fft_state *cfg; + CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); + if (nfft == 480) id = 0; + else if (nfft == 240) id = 1; + else if (nfft == 120) id = 2; + else if (nfft == 60) id = 3; + else return; + cfg = mode->mdct.kfft[id]; +#endif + + in = (kiss_fft_cpx*)malloc(buflen); + out = (kiss_fft_cpx*)malloc(buflen); + + for (k=0;k<nfft;++k) { + in[k].r = (rand() % 32767) - 16384; + in[k].i = (rand() % 32767) - 16384; + } + + for (k=0;k<nfft;++k) { + in[k].r *= 32768; + in[k].i *= 32768; + } + + if (isinverse) + { + for (k=0;k<nfft;++k) { + in[k].r /= nfft; + in[k].i /= nfft; + } + } + + /*for (k=0;k<nfft;++k) printf("%d %d ", in[k].r, in[k].i);printf("\n");*/ + + if (isinverse) + opus_ifft(cfg,in,out, arch); + else + opus_fft(cfg,in,out, arch); + + /*for (k=0;k<nfft;++k) printf("%d %d ", out[k].r, out[k].i);printf("\n");*/ + + check(in,out,nfft,isinverse); + + free(in); + free(out); +#ifdef CUSTOM_MODES + opus_fft_free(cfg, arch); +#endif +} + +int main(int argc,char ** argv) +{ + ALLOC_STACK; + int arch = opus_select_arch(); + + if (argc>1) { + int k; + for (k=1;k<argc;++k) { + test1d(atoi(argv[k]),0,arch); + test1d(atoi(argv[k]),1,arch); + } + }else{ + test1d(32,0,arch); + test1d(32,1,arch); + test1d(128,0,arch); + test1d(128,1,arch); + test1d(256,0,arch); + test1d(256,1,arch); +#ifndef RADIX_TWO_ONLY + test1d(36,0,arch); + test1d(36,1,arch); + test1d(50,0,arch); + test1d(50,1,arch); + test1d(60,0,arch); + test1d(60,1,arch); + test1d(120,0,arch); + test1d(120,1,arch); + test1d(240,0,arch); + test1d(240,1,arch); + test1d(480,0,arch); + test1d(480,1,arch); +#endif + } + return ret; +} diff --git a/libopus/celt/tests/test_unit_entropy.c b/libopus/celt/tests/test_unit_entropy.c new file mode 100644 index 0000000..7f67452 --- /dev/null +++ b/libopus/celt/tests/test_unit_entropy.c @@ -0,0 +1,383 @@ +/* Copyright (c) 2007-2011 Xiph.Org Foundation, Mozilla Corporation, + Gregory Maxwell + Written by Jean-Marc Valin, Gregory Maxwell, and Timothy B. Terriberry */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stdlib.h> +#include <stdio.h> +#include <math.h> +#include <time.h> +#define CELT_C +#include "entcode.h" +#include "entenc.h" +#include "entdec.h" +#include <string.h> + +#include "entenc.c" +#include "entdec.c" +#include "entcode.c" + +#ifndef M_LOG2E +# define M_LOG2E 1.4426950408889634074 +#endif +#define DATA_SIZE 10000000 +#define DATA_SIZE2 10000 + +int main(int _argc,char **_argv){ + ec_enc enc; + ec_dec dec; + long nbits; + long nbits2; + double entropy; + int ft; + int ftb; + int sz; + int i; + int ret; + unsigned int sym; + unsigned int seed; + unsigned char *ptr; + const char *env_seed; + ret=0; + entropy=0; + if (_argc > 2) { + fprintf(stderr, "Usage: %s [<seed>]\n", _argv[0]); + return 1; + } + env_seed = getenv("SEED"); + if (_argc > 1) + seed = atoi(_argv[1]); + else if (env_seed) + seed = atoi(env_seed); + else + seed = time(NULL); + /*Testing encoding of raw bit values.*/ + ptr = (unsigned char *)malloc(DATA_SIZE); + ec_enc_init(&enc,ptr, DATA_SIZE); + for(ft=2;ft<1024;ft++){ + for(i=0;i<ft;i++){ + entropy+=log(ft)*M_LOG2E; + ec_enc_uint(&enc,i,ft); + } + } + /*Testing encoding of raw bit values.*/ + for(ftb=1;ftb<16;ftb++){ + for(i=0;i<(1<<ftb);i++){ + entropy+=ftb; + nbits=ec_tell(&enc); + ec_enc_bits(&enc,i,ftb); + nbits2=ec_tell(&enc); + if(nbits2-nbits!=ftb){ + fprintf(stderr,"Used %li bits to encode %i bits directly.\n", + nbits2-nbits,ftb); + ret=-1; + } + } + } + nbits=ec_tell_frac(&enc); + ec_enc_done(&enc); + fprintf(stderr, + "Encoded %0.2lf bits of entropy to %0.2lf bits (%0.3lf%% wasted).\n", + entropy,ldexp(nbits,-3),100*(nbits-ldexp(entropy,3))/nbits); + fprintf(stderr,"Packed to %li bytes.\n",(long)ec_range_bytes(&enc)); + ec_dec_init(&dec,ptr,DATA_SIZE); + for(ft=2;ft<1024;ft++){ + for(i=0;i<ft;i++){ + sym=ec_dec_uint(&dec,ft); + if(sym!=(unsigned)i){ + fprintf(stderr,"Decoded %i instead of %i with ft of %i.\n",sym,i,ft); + ret=-1; + } + } + } + for(ftb=1;ftb<16;ftb++){ + for(i=0;i<(1<<ftb);i++){ + sym=ec_dec_bits(&dec,ftb); + if(sym!=(unsigned)i){ + fprintf(stderr,"Decoded %i instead of %i with ftb of %i.\n",sym,i,ftb); + ret=-1; + } + } + } + nbits2=ec_tell_frac(&dec); + if(nbits!=nbits2){ + fprintf(stderr, + "Reported number of bits used was %0.2lf, should be %0.2lf.\n", + ldexp(nbits2,-3),ldexp(nbits,-3)); + ret=-1; + } + /*Testing an encoder bust prefers range coder data over raw bits. + This isn't a general guarantee, will only work for data that is buffered in + the encoder state and not yet stored in the user buffer, and should never + get used in practice. + It's mostly here for code coverage completeness.*/ + /*Start with a 16-bit buffer.*/ + ec_enc_init(&enc,ptr,2); + /*Write 7 raw bits.*/ + ec_enc_bits(&enc,0x55,7); + /*Write 12.3 bits of range coder data.*/ + ec_enc_uint(&enc,1,2); + ec_enc_uint(&enc,1,3); + ec_enc_uint(&enc,1,4); + ec_enc_uint(&enc,1,5); + ec_enc_uint(&enc,2,6); + ec_enc_uint(&enc,6,7); + ec_enc_done(&enc); + ec_dec_init(&dec,ptr,2); + if(!enc.error + /*The raw bits should have been overwritten by the range coder data.*/ + ||ec_dec_bits(&dec,7)!=0x05 + /*And all the range coder data should have been encoded correctly.*/ + ||ec_dec_uint(&dec,2)!=1 + ||ec_dec_uint(&dec,3)!=1 + ||ec_dec_uint(&dec,4)!=1 + ||ec_dec_uint(&dec,5)!=1 + ||ec_dec_uint(&dec,6)!=2 + ||ec_dec_uint(&dec,7)!=6){ + fprintf(stderr,"Encoder bust overwrote range coder data with raw bits.\n"); + ret=-1; + } + srand(seed); + fprintf(stderr,"Testing random streams... Random seed: %u (%.4X)\n", seed, rand() % 65536); + for(i=0;i<409600;i++){ + unsigned *data; + unsigned *tell; + unsigned tell_bits; + int j; + int zeros; + ft=rand()/((RAND_MAX>>(rand()%11U))+1U)+10; + sz=rand()/((RAND_MAX>>(rand()%9U))+1U); + data=(unsigned *)malloc(sz*sizeof(*data)); + tell=(unsigned *)malloc((sz+1)*sizeof(*tell)); + ec_enc_init(&enc,ptr,DATA_SIZE2); + zeros = rand()%13==0; + tell[0]=ec_tell_frac(&enc); + for(j=0;j<sz;j++){ + if (zeros) + data[j]=0; + else + data[j]=rand()%ft; + ec_enc_uint(&enc,data[j],ft); + tell[j+1]=ec_tell_frac(&enc); + } + if (rand()%2==0) + while(ec_tell(&enc)%8 != 0) + ec_enc_uint(&enc, rand()%2, 2); + tell_bits = ec_tell(&enc); + ec_enc_done(&enc); + if(tell_bits!=(unsigned)ec_tell(&enc)){ + fprintf(stderr,"ec_tell() changed after ec_enc_done(): %i instead of %i (Random seed: %u)\n", + ec_tell(&enc),tell_bits,seed); + ret=-1; + } + if ((tell_bits+7)/8 < ec_range_bytes(&enc)) + { + fprintf (stderr, "ec_tell() lied, there's %i bytes instead of %d (Random seed: %u)\n", + ec_range_bytes(&enc), (tell_bits+7)/8,seed); + ret=-1; + } + ec_dec_init(&dec,ptr,DATA_SIZE2); + if(ec_tell_frac(&dec)!=tell[0]){ + fprintf(stderr, + "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n", + 0,ec_tell_frac(&dec),tell[0],seed); + } + for(j=0;j<sz;j++){ + sym=ec_dec_uint(&dec,ft); + if(sym!=data[j]){ + fprintf(stderr, + "Decoded %i instead of %i with ft of %i at position %i of %i (Random seed: %u).\n", + sym,data[j],ft,j,sz,seed); + ret=-1; + } + if(ec_tell_frac(&dec)!=tell[j+1]){ + fprintf(stderr, + "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n", + j+1,ec_tell_frac(&dec),tell[j+1],seed); + } + } + free(tell); + free(data); + } + /*Test compatibility between multiple different encode/decode routines.*/ + for(i=0;i<409600;i++){ + unsigned *logp1; + unsigned *data; + unsigned *tell; + unsigned *enc_method; + int j; + sz=rand()/((RAND_MAX>>(rand()%9U))+1U); + logp1=(unsigned *)malloc(sz*sizeof(*logp1)); + data=(unsigned *)malloc(sz*sizeof(*data)); + tell=(unsigned *)malloc((sz+1)*sizeof(*tell)); + enc_method=(unsigned *)malloc(sz*sizeof(*enc_method)); + ec_enc_init(&enc,ptr,DATA_SIZE2); + tell[0]=ec_tell_frac(&enc); + for(j=0;j<sz;j++){ + data[j]=rand()/((RAND_MAX>>1)+1); + logp1[j]=(rand()%15)+1; + enc_method[j]=rand()/((RAND_MAX>>2)+1); + switch(enc_method[j]){ + case 0:{ + ec_encode(&enc,data[j]?(1<<logp1[j])-1:0, + (1<<logp1[j])-(data[j]?0:1),1<<logp1[j]); + }break; + case 1:{ + ec_encode_bin(&enc,data[j]?(1<<logp1[j])-1:0, + (1<<logp1[j])-(data[j]?0:1),logp1[j]); + }break; + case 2:{ + ec_enc_bit_logp(&enc,data[j],logp1[j]); + }break; + case 3:{ + unsigned char icdf[2]; + icdf[0]=1; + icdf[1]=0; + ec_enc_icdf(&enc,data[j],icdf,logp1[j]); + }break; + } + tell[j+1]=ec_tell_frac(&enc); + } + ec_enc_done(&enc); + if((ec_tell(&enc)+7U)/8U<ec_range_bytes(&enc)){ + fprintf(stderr,"tell() lied, there's %i bytes instead of %d (Random seed: %u)\n", + ec_range_bytes(&enc),(ec_tell(&enc)+7)/8,seed); + ret=-1; + } + ec_dec_init(&dec,ptr,DATA_SIZE2); + if(ec_tell_frac(&dec)!=tell[0]){ + fprintf(stderr, + "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n", + 0,ec_tell_frac(&dec),tell[0],seed); + } + for(j=0;j<sz;j++){ + int fs; + int dec_method; + dec_method=rand()/((RAND_MAX>>2)+1); + switch(dec_method){ + case 0:{ + fs=ec_decode(&dec,1<<logp1[j]); + sym=fs>=(1<<logp1[j])-1; + ec_dec_update(&dec,sym?(1<<logp1[j])-1:0, + (1<<logp1[j])-(sym?0:1),1<<logp1[j]); + }break; + case 1:{ + fs=ec_decode_bin(&dec,logp1[j]); + sym=fs>=(1<<logp1[j])-1; + ec_dec_update(&dec,sym?(1<<logp1[j])-1:0, + (1<<logp1[j])-(sym?0:1),1<<logp1[j]); + }break; + case 2:{ + sym=ec_dec_bit_logp(&dec,logp1[j]); + }break; + case 3:{ + unsigned char icdf[2]; + icdf[0]=1; + icdf[1]=0; + sym=ec_dec_icdf(&dec,icdf,logp1[j]); + }break; + } + if(sym!=data[j]){ + fprintf(stderr, + "Decoded %i instead of %i with logp1 of %i at position %i of %i (Random seed: %u).\n", + sym,data[j],logp1[j],j,sz,seed); + fprintf(stderr,"Encoding method: %i, decoding method: %i\n", + enc_method[j],dec_method); + ret=-1; + } + if(ec_tell_frac(&dec)!=tell[j+1]){ + fprintf(stderr, + "Tell mismatch between encoder and decoder at symbol %i: %i instead of %i (Random seed: %u).\n", + j+1,ec_tell_frac(&dec),tell[j+1],seed); + } + } + free(enc_method); + free(tell); + free(data); + free(logp1); + } + ec_enc_init(&enc,ptr,DATA_SIZE2); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,0,2); + ec_enc_patch_initial_bits(&enc,3,2); + if(enc.error){ + fprintf(stderr,"patch_initial_bits failed"); + ret=-1; + } + ec_enc_patch_initial_bits(&enc,0,5); + if(!enc.error){ + fprintf(stderr,"patch_initial_bits didn't fail when it should have"); + ret=-1; + } + ec_enc_done(&enc); + if(ec_range_bytes(&enc)!=1||ptr[0]!=192){ + fprintf(stderr,"Got %d when expecting 192 for patch_initial_bits",ptr[0]); + ret=-1; + } + ec_enc_init(&enc,ptr,DATA_SIZE2); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,0,1); + ec_enc_bit_logp(&enc,1,6); + ec_enc_bit_logp(&enc,0,2); + ec_enc_patch_initial_bits(&enc,0,2); + if(enc.error){ + fprintf(stderr,"patch_initial_bits failed"); + ret=-1; + } + ec_enc_done(&enc); + if(ec_range_bytes(&enc)!=2||ptr[0]!=63){ + fprintf(stderr,"Got %d when expecting 63 for patch_initial_bits",ptr[0]); + ret=-1; + } + ec_enc_init(&enc,ptr,2); + ec_enc_bit_logp(&enc,0,2); + for(i=0;i<48;i++){ + ec_enc_bits(&enc,0,1); + } + ec_enc_done(&enc); + if(!enc.error){ + fprintf(stderr,"Raw bits overfill didn't fail when it should have"); + ret=-1; + } + ec_enc_init(&enc,ptr,2); + for(i=0;i<17;i++){ + ec_enc_bits(&enc,0,1); + } + ec_enc_done(&enc); + if(!enc.error){ + fprintf(stderr,"17 raw bits encoded in two bytes"); + ret=-1; + } + free(ptr); + return ret; +} diff --git a/libopus/celt/tests/test_unit_laplace.c b/libopus/celt/tests/test_unit_laplace.c new file mode 100644 index 0000000..727bf01 --- /dev/null +++ b/libopus/celt/tests/test_unit_laplace.c @@ -0,0 +1,93 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation, Mozilla Corporation + Written by Jean-Marc Valin and Timothy B. Terriberry */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stdio.h> +#include <stdlib.h> +#define CELT_C +#include "laplace.h" +#include "stack_alloc.h" + +#include "entenc.c" +#include "entdec.c" +#include "entcode.c" +#include "laplace.c" + +#define DATA_SIZE 40000 + +int ec_laplace_get_start_freq(int decay) +{ + opus_uint32 ft = 32768 - LAPLACE_MINP*(2*LAPLACE_NMIN+1); + int fs = (ft*(16384-decay))/(16384+decay); + return fs+LAPLACE_MINP; +} + +int main(void) +{ + int i; + int ret = 0; + ec_enc enc; + ec_dec dec; + unsigned char *ptr; + int val[10000], decay[10000]; + ALLOC_STACK; + ptr = (unsigned char *)malloc(DATA_SIZE); + ec_enc_init(&enc,ptr,DATA_SIZE); + + val[0] = 3; decay[0] = 6000; + val[1] = 0; decay[1] = 5800; + val[2] = -1; decay[2] = 5600; + for (i=3;i<10000;i++) + { + val[i] = rand()%15-7; + decay[i] = rand()%11000+5000; + } + for (i=0;i<10000;i++) + ec_laplace_encode(&enc, &val[i], + ec_laplace_get_start_freq(decay[i]), decay[i]); + + ec_enc_done(&enc); + + ec_dec_init(&dec,ec_get_buffer(&enc),ec_range_bytes(&enc)); + + for (i=0;i<10000;i++) + { + int d = ec_laplace_decode(&dec, + ec_laplace_get_start_freq(decay[i]), decay[i]); + if (d != val[i]) + { + fprintf (stderr, "Got %d instead of %d\n", d, val[i]); + ret = 1; + } + } + + free(ptr); + return ret; +} diff --git a/libopus/celt/tests/test_unit_mathops.c b/libopus/celt/tests/test_unit_mathops.c new file mode 100644 index 0000000..874e9ad --- /dev/null +++ b/libopus/celt/tests/test_unit_mathops.c @@ -0,0 +1,266 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation, Mozilla Corporation, + Gregory Maxwell + Written by Jean-Marc Valin, Gregory Maxwell, and Timothy B. Terriberry */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifndef CUSTOM_MODES +#define CUSTOM_MODES +#endif + +#include <stdio.h> +#include <math.h> +#include "mathops.h" +#include "bands.h" + +#ifdef FIXED_POINT +#define WORD "%d" +#else +#define WORD "%f" +#endif + +int ret = 0; + +void testdiv(void) +{ + opus_int32 i; + for (i=1;i<=327670;i++) + { + double prod; + opus_val32 val; + val = celt_rcp(i); +#ifdef FIXED_POINT + prod = (1./32768./65526.)*val*i; +#else + prod = val*i; +#endif + if (fabs(prod-1) > .00025) + { + fprintf (stderr, "div failed: 1/%d="WORD" (product = %f)\n", i, val, prod); + ret = 1; + } + } +} + +void testsqrt(void) +{ + opus_int32 i; + for (i=1;i<=1000000000;i++) + { + double ratio; + opus_val16 val; + val = celt_sqrt(i); + ratio = val/sqrt(i); + if (fabs(ratio - 1) > .0005 && fabs(val-sqrt(i)) > 2) + { + fprintf (stderr, "sqrt failed: sqrt(%d)="WORD" (ratio = %f)\n", i, val, ratio); + ret = 1; + } + i+= i>>10; + } +} + +void testbitexactcos(void) +{ + int i; + opus_int32 min_d,max_d,last,chk; + chk=max_d=0; + last=min_d=32767; + for(i=64;i<=16320;i++) + { + opus_int32 d; + opus_int32 q=bitexact_cos(i); + chk ^= q*i; + d = last - q; + if (d>max_d)max_d=d; + if (d<min_d)min_d=d; + last = q; + } + if ((chk!=89408644)||(max_d!=5)||(min_d!=0)||(bitexact_cos(64)!=32767)|| + (bitexact_cos(16320)!=200)||(bitexact_cos(8192)!=23171)) + { + fprintf (stderr, "bitexact_cos failed\n"); + ret = 1; + } +} + +void testbitexactlog2tan(void) +{ + int i,fail; + opus_int32 min_d,max_d,last,chk; + fail=chk=max_d=0; + last=min_d=15059; + for(i=64;i<8193;i++) + { + opus_int32 d; + opus_int32 mid=bitexact_cos(i); + opus_int32 side=bitexact_cos(16384-i); + opus_int32 q=bitexact_log2tan(mid,side); + chk ^= q*i; + d = last - q; + if (q!=-1*bitexact_log2tan(side,mid)) + fail = 1; + if (d>max_d)max_d=d; + if (d<min_d)min_d=d; + last = q; + } + if ((chk!=15821257)||(max_d!=61)||(min_d!=-2)||fail|| + (bitexact_log2tan(32767,200)!=15059)||(bitexact_log2tan(30274,12540)!=2611)|| + (bitexact_log2tan(23171,23171)!=0)) + { + fprintf (stderr, "bitexact_log2tan failed\n"); + ret = 1; + } +} + +#ifndef FIXED_POINT +void testlog2(void) +{ + float x; + for (x=0.001;x<1677700.0;x+=(x/8.0)) + { + float error = fabs((1.442695040888963387*log(x))-celt_log2(x)); + if (error>0.0009) + { + fprintf (stderr, "celt_log2 failed: fabs((1.442695040888963387*log(x))-celt_log2(x))>0.001 (x = %f, error = %f)\n", x,error); + ret = 1; + } + } +} + +void testexp2(void) +{ + float x; + for (x=-11.0;x<24.0;x+=0.0007) + { + float error = fabs(x-(1.442695040888963387*log(celt_exp2(x)))); + if (error>0.0002) + { + fprintf (stderr, "celt_exp2 failed: fabs(x-(1.442695040888963387*log(celt_exp2(x))))>0.0005 (x = %f, error = %f)\n", x,error); + ret = 1; + } + } +} + +void testexp2log2(void) +{ + float x; + for (x=-11.0;x<24.0;x+=0.0007) + { + float error = fabs(x-(celt_log2(celt_exp2(x)))); + if (error>0.001) + { + fprintf (stderr, "celt_log2/celt_exp2 failed: fabs(x-(celt_log2(celt_exp2(x))))>0.001 (x = %f, error = %f)\n", x,error); + ret = 1; + } + } +} +#else +void testlog2(void) +{ + opus_val32 x; + for (x=8;x<1073741824;x+=(x>>3)) + { + float error = fabs((1.442695040888963387*log(x/16384.0))-celt_log2(x)/1024.0); + if (error>0.003) + { + fprintf (stderr, "celt_log2 failed: x = %ld, error = %f\n", (long)x,error); + ret = 1; + } + } +} + +void testexp2(void) +{ + opus_val16 x; + for (x=-32768;x<15360;x++) + { + float error1 = fabs(x/1024.0-(1.442695040888963387*log(celt_exp2(x)/65536.0))); + float error2 = fabs(exp(0.6931471805599453094*x/1024.0)-celt_exp2(x)/65536.0); + if (error1>0.0002&&error2>0.00004) + { + fprintf (stderr, "celt_exp2 failed: x = "WORD", error1 = %f, error2 = %f\n", x,error1,error2); + ret = 1; + } + } +} + +void testexp2log2(void) +{ + opus_val32 x; + for (x=8;x<65536;x+=(x>>3)) + { + float error = fabs(x-0.25*celt_exp2(celt_log2(x)))/16384; + if (error>0.004) + { + fprintf (stderr, "celt_log2/celt_exp2 failed: fabs(x-(celt_exp2(celt_log2(x))))>0.001 (x = %ld, error = %f)\n", (long)x,error); + ret = 1; + } + } +} + +void testilog2(void) +{ + opus_val32 x; + for (x=1;x<=268435455;x+=127) + { + opus_val32 lg; + opus_val32 y; + + lg = celt_ilog2(x); + if (lg<0 || lg>=31) + { + printf("celt_ilog2 failed: 0<=celt_ilog2(x)<31 (x = %d, celt_ilog2(x) = %d)\n",x,lg); + ret = 1; + } + y = 1<<lg; + + if (x<y || (x>>1)>=y) + { + printf("celt_ilog2 failed: 2**celt_ilog2(x)<=x<2**(celt_ilog2(x)+1) (x = %d, 2**celt_ilog2(x) = %d)\n",x,y); + ret = 1; + } + } +} +#endif + +int main(void) +{ + testbitexactcos(); + testbitexactlog2tan(); + testdiv(); + testsqrt(); + testlog2(); + testexp2(); + testexp2log2(); +#ifdef FIXED_POINT + testilog2(); +#endif + return ret; +} diff --git a/libopus/celt/tests/test_unit_mdct.c b/libopus/celt/tests/test_unit_mdct.c new file mode 100644 index 0000000..4a563cc --- /dev/null +++ b/libopus/celt/tests/test_unit_mdct.c @@ -0,0 +1,227 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stdio.h> + +#include "mdct.h" +#include "stack_alloc.h" +#include "kiss_fft.h" +#include "mdct.h" +#include "modes.h" + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +int ret = 0; +void check(kiss_fft_scalar * in,kiss_fft_scalar * out,int nfft,int isinverse) +{ + int bin,k; + double errpow=0,sigpow=0; + double snr; + for (bin=0;bin<nfft/2;++bin) { + double ansr = 0; + double difr; + + for (k=0;k<nfft;++k) { + double phase = 2*M_PI*(k+.5+.25*nfft)*(bin+.5)/nfft; + double re = cos(phase); + + re /= nfft/4; + + ansr += in[k] * re; + } + /*printf ("%f %f\n", ansr, out[bin]);*/ + difr = ansr - out[bin]; + errpow += difr*difr; + sigpow += ansr*ansr; + } + snr = 10*log10(sigpow/errpow); + printf("nfft=%d inverse=%d,snr = %f\n",nfft,isinverse,snr ); + if (snr<60) { + printf( "** poor snr: %f **\n", snr); + ret = 1; + } +} + +void check_inv(kiss_fft_scalar * in,kiss_fft_scalar * out,int nfft,int isinverse) +{ + int bin,k; + double errpow=0,sigpow=0; + double snr; + for (bin=0;bin<nfft;++bin) { + double ansr = 0; + double difr; + + for (k=0;k<nfft/2;++k) { + double phase = 2*M_PI*(bin+.5+.25*nfft)*(k+.5)/nfft; + double re = cos(phase); + + /*re *= 2;*/ + + ansr += in[k] * re; + } + /*printf ("%f %f\n", ansr, out[bin]);*/ + difr = ansr - out[bin]; + errpow += difr*difr; + sigpow += ansr*ansr; + } + snr = 10*log10(sigpow/errpow); + printf("nfft=%d inverse=%d,snr = %f\n",nfft,isinverse,snr ); + if (snr<60) { + printf( "** poor snr: %f **\n", snr); + ret = 1; + } +} + + +void test1d(int nfft,int isinverse,int arch) +{ + size_t buflen = sizeof(kiss_fft_scalar)*nfft; + kiss_fft_scalar *in; + kiss_fft_scalar *in_copy; + kiss_fft_scalar *out; + opus_val16 *window; + int k; + +#ifdef CUSTOM_MODES + int shift = 0; + const mdct_lookup *cfg; + mdct_lookup _cfg; + clt_mdct_init(&_cfg, nfft, 0, arch); + cfg = &_cfg; +#else + int shift; + const mdct_lookup *cfg; + CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); + if (nfft == 1920) shift = 0; + else if (nfft == 960) shift = 1; + else if (nfft == 480) shift = 2; + else if (nfft == 240) shift = 3; + else return; + cfg = &mode->mdct; +#endif + + in = (kiss_fft_scalar*)malloc(buflen); + in_copy = (kiss_fft_scalar*)malloc(buflen); + out = (kiss_fft_scalar*)malloc(buflen); + window = (opus_val16*)malloc(sizeof(opus_val16)*nfft/2); + + for (k=0;k<nfft;++k) { + in[k] = (rand() % 32768) - 16384; + } + + for (k=0;k<nfft/2;++k) { + window[k] = Q15ONE; + } + for (k=0;k<nfft;++k) { + in[k] *= 32768; + } + + if (isinverse) + { + for (k=0;k<nfft;++k) { + in[k] /= nfft; + } + } + + for (k=0;k<nfft;++k) + in_copy[k] = in[k]; + /*for (k=0;k<nfft;++k) printf("%d %d ", in[k].r, in[k].i);printf("\n");*/ + + if (isinverse) + { + for (k=0;k<nfft;++k) + out[k] = 0; + clt_mdct_backward(cfg,in,out, window, nfft/2, shift, 1, arch); + /* apply TDAC because clt_mdct_backward() no longer does that */ + for (k=0;k<nfft/4;++k) + out[nfft-k-1] = out[nfft/2+k]; + check_inv(in,out,nfft,isinverse); + } else { + clt_mdct_forward(cfg,in,out,window, nfft/2, shift, 1, arch); + check(in_copy,out,nfft,isinverse); + } + /*for (k=0;k<nfft;++k) printf("%d %d ", out[k].r, out[k].i);printf("\n");*/ + + + free(in); + free(in_copy); + free(out); + free(window); +#ifdef CUSTOM_MODES + clt_mdct_clear(&_cfg, arch); +#endif +} + +int main(int argc,char ** argv) +{ + ALLOC_STACK; + int arch = opus_select_arch(); + + if (argc>1) { + int k; + for (k=1;k<argc;++k) { + test1d(atoi(argv[k]),0,arch); + test1d(atoi(argv[k]),1,arch); + } + }else{ + test1d(32,0,arch); + test1d(32,1,arch); + test1d(256,0,arch); + test1d(256,1,arch); + test1d(512,0,arch); + test1d(512,1,arch); + test1d(1024,0,arch); + test1d(1024,1,arch); + test1d(2048,0,arch); + test1d(2048,1,arch); +#ifndef RADIX_TWO_ONLY + test1d(36,0,arch); + test1d(36,1,arch); + test1d(40,0,arch); + test1d(40,1,arch); + test1d(60,0,arch); + test1d(60,1,arch); + test1d(120,0,arch); + test1d(120,1,arch); + test1d(240,0,arch); + test1d(240,1,arch); + test1d(480,0,arch); + test1d(480,1,arch); + test1d(960,0,arch); + test1d(960,1,arch); + test1d(1920,0,arch); + test1d(1920,1,arch); +#endif + } + return ret; +} diff --git a/libopus/celt/tests/test_unit_rotation.c b/libopus/celt/tests/test_unit_rotation.c new file mode 100644 index 0000000..8a31b3f --- /dev/null +++ b/libopus/celt/tests/test_unit_rotation.c @@ -0,0 +1,86 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifndef CUSTOM_MODES +#define CUSTOM_MODES +#endif + +#include <stdio.h> +#include <stdlib.h> +#include "vq.h" +#include "bands.h" +#include "stack_alloc.h" +#include <math.h> + + +#define MAX_SIZE 100 + +int ret=0; +void test_rotation(int N, int K) +{ + int i; + double err = 0, ener = 0, snr, snr0; + opus_val16 x0[MAX_SIZE]; + opus_val16 x1[MAX_SIZE]; + for (i=0;i<N;i++) + x1[i] = x0[i] = rand()%32767-16384; + exp_rotation(x1, N, 1, 1, K, SPREAD_NORMAL); + for (i=0;i<N;i++) + { + err += (x0[i]-(double)x1[i])*(x0[i]-(double)x1[i]); + ener += x0[i]*(double)x0[i]; + } + snr0 = 20*log10(ener/err); + err = ener = 0; + exp_rotation(x1, N, -1, 1, K, SPREAD_NORMAL); + for (i=0;i<N;i++) + { + err += (x0[i]-(double)x1[i])*(x0[i]-(double)x1[i]); + ener += x0[i]*(double)x0[i]; + } + snr = 20*log10(ener/err); + printf ("SNR for size %d (%d pulses) is %f (was %f without inverse)\n", N, K, snr, snr0); + if (snr < 60 || snr0 > 20) + { + fprintf(stderr, "FAIL!\n"); + ret = 1; + } +} + +int main(void) +{ + ALLOC_STACK; + test_rotation(15, 3); + test_rotation(23, 5); + test_rotation(50, 3); + test_rotation(80, 1); + return ret; +} diff --git a/libopus/celt/tests/test_unit_types.c b/libopus/celt/tests/test_unit_types.c new file mode 100644 index 0000000..67a0fb8 --- /dev/null +++ b/libopus/celt/tests/test_unit_types.c @@ -0,0 +1,50 @@ +/* Copyright (c) 2008-2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus_types.h" +#include <stdio.h> + +int main(void) +{ + opus_int16 i = 1; + i <<= 14; + if (i>>14 != 1) + { + fprintf(stderr, "opus_int16 isn't 16 bits\n"); + return 1; + } + if (sizeof(opus_int16)*2 != sizeof(opus_int32)) + { + fprintf(stderr, "16*2 != 32\n"); + return 1; + } + return 0; +} diff --git a/libopus/celt/vq.c b/libopus/celt/vq.c new file mode 100644 index 0000000..8011e22 --- /dev/null +++ b/libopus/celt/vq.c @@ -0,0 +1,442 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mathops.h" +#include "cwrs.h" +#include "vq.h" +#include "arch.h" +#include "os_support.h" +#include "bands.h" +#include "rate.h" +#include "pitch.h" + +#if defined(MIPSr1_ASM) +#include "mips/vq_mipsr1.h" +#endif + +#ifndef OVERRIDE_vq_exp_rotation1 +static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s) +{ + int i; + opus_val16 ms; + celt_norm *Xptr; + Xptr = X; + ms = NEG16(s); + for (i=0;i<len-stride;i++) + { + celt_norm x1, x2; + x1 = Xptr[0]; + x2 = Xptr[stride]; + Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2), s, x1), 15)); + *Xptr++ = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15)); + } + Xptr = &X[len-2*stride-1]; + for (i=len-2*stride-1;i>=0;i--) + { + celt_norm x1, x2; + x1 = Xptr[0]; + x2 = Xptr[stride]; + Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2), s, x1), 15)); + *Xptr-- = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15)); + } +} +#endif /* OVERRIDE_vq_exp_rotation1 */ + +void exp_rotation(celt_norm *X, int len, int dir, int stride, int K, int spread) +{ + static const int SPREAD_FACTOR[3]={15,10,5}; + int i; + opus_val16 c, s; + opus_val16 gain, theta; + int stride2=0; + int factor; + + if (2*K>=len || spread==SPREAD_NONE) + return; + factor = SPREAD_FACTOR[spread-1]; + + gain = celt_div((opus_val32)MULT16_16(Q15_ONE,len),(opus_val32)(len+factor*K)); + theta = HALF16(MULT16_16_Q15(gain,gain)); + + c = celt_cos_norm(EXTEND32(theta)); + s = celt_cos_norm(EXTEND32(SUB16(Q15ONE,theta))); /* sin(theta) */ + + if (len>=8*stride) + { + stride2 = 1; + /* This is just a simple (equivalent) way of computing sqrt(len/stride) with rounding. + It's basically incrementing long as (stride2+0.5)^2 < len/stride. */ + while ((stride2*stride2+stride2)*stride + (stride>>2) < len) + stride2++; + } + /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for + extract_collapse_mask().*/ + len = celt_udiv(len, stride); + for (i=0;i<stride;i++) + { + if (dir < 0) + { + if (stride2) + exp_rotation1(X+i*len, len, stride2, s, c); + exp_rotation1(X+i*len, len, 1, c, s); + } else { + exp_rotation1(X+i*len, len, 1, c, -s); + if (stride2) + exp_rotation1(X+i*len, len, stride2, s, -c); + } + } +} + +/** Takes the pitch vector and the decoded residual vector, computes the gain + that will give ||p+g*y||=1 and mixes the residual with the pitch. */ +static void normalise_residual(int * OPUS_RESTRICT iy, celt_norm * OPUS_RESTRICT X, + int N, opus_val32 Ryy, opus_val16 gain) +{ + int i; +#ifdef FIXED_POINT + int k; +#endif + opus_val32 t; + opus_val16 g; + +#ifdef FIXED_POINT + k = celt_ilog2(Ryy)>>1; +#endif + t = VSHR32(Ryy, 2*(k-7)); + g = MULT16_16_P15(celt_rsqrt_norm(t),gain); + + i=0; + do + X[i] = EXTRACT16(PSHR32(MULT16_16(g, iy[i]), k+1)); + while (++i < N); +} + +static unsigned extract_collapse_mask(int *iy, int N, int B) +{ + unsigned collapse_mask; + int N0; + int i; + if (B<=1) + return 1; + /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for + exp_rotation().*/ + N0 = celt_udiv(N, B); + collapse_mask = 0; + i=0; do { + int j; + unsigned tmp=0; + j=0; do { + tmp |= iy[i*N0+j]; + } while (++j<N0); + collapse_mask |= (tmp!=0)<<i; + } while (++i<B); + return collapse_mask; +} + +opus_val16 op_pvq_search_c(celt_norm *X, int *iy, int K, int N, int arch) +{ + VARDECL(celt_norm, y); + VARDECL(int, signx); + int i, j; + int pulsesLeft; + opus_val32 sum; + opus_val32 xy; + opus_val16 yy; + SAVE_STACK; + + (void)arch; + ALLOC(y, N, celt_norm); + ALLOC(signx, N, int); + + /* Get rid of the sign */ + sum = 0; + j=0; do { + signx[j] = X[j]<0; + /* OPT: Make sure the compiler doesn't use a branch on ABS16(). */ + X[j] = ABS16(X[j]); + iy[j] = 0; + y[j] = 0; + } while (++j<N); + + xy = yy = 0; + + pulsesLeft = K; + + /* Do a pre-search by projecting on the pyramid */ + if (K > (N>>1)) + { + opus_val16 rcp; + j=0; do { + sum += X[j]; + } while (++j<N); + + /* If X is too small, just replace it with a pulse at 0 */ +#ifdef FIXED_POINT + if (sum <= K) +#else + /* Prevents infinities and NaNs from causing too many pulses + to be allocated. 64 is an approximation of infinity here. */ + if (!(sum > EPSILON && sum < 64)) +#endif + { + X[0] = QCONST16(1.f,14); + j=1; do + X[j]=0; + while (++j<N); + sum = QCONST16(1.f,14); + } +#ifdef FIXED_POINT + rcp = EXTRACT16(MULT16_32_Q16(K, celt_rcp(sum))); +#else + /* Using K+e with e < 1 guarantees we cannot get more than K pulses. */ + rcp = EXTRACT16(MULT16_32_Q16(K+0.8f, celt_rcp(sum))); +#endif + j=0; do { +#ifdef FIXED_POINT + /* It's really important to round *towards zero* here */ + iy[j] = MULT16_16_Q15(X[j],rcp); +#else + iy[j] = (int)floor(rcp*X[j]); +#endif + y[j] = (celt_norm)iy[j]; + yy = MAC16_16(yy, y[j],y[j]); + xy = MAC16_16(xy, X[j],y[j]); + y[j] *= 2; + pulsesLeft -= iy[j]; + } while (++j<N); + } + celt_sig_assert(pulsesLeft>=0); + + /* This should never happen, but just in case it does (e.g. on silence) + we fill the first bin with pulses. */ +#ifdef FIXED_POINT_DEBUG + celt_sig_assert(pulsesLeft<=N+3); +#endif + if (pulsesLeft > N+3) + { + opus_val16 tmp = (opus_val16)pulsesLeft; + yy = MAC16_16(yy, tmp, tmp); + yy = MAC16_16(yy, tmp, y[0]); + iy[0] += pulsesLeft; + pulsesLeft=0; + } + + for (i=0;i<pulsesLeft;i++) + { + opus_val16 Rxy, Ryy; + int best_id; + opus_val32 best_num; + opus_val16 best_den; +#ifdef FIXED_POINT + int rshift; +#endif +#ifdef FIXED_POINT + rshift = 1+celt_ilog2(K-pulsesLeft+i+1); +#endif + best_id = 0; + /* The squared magnitude term gets added anyway, so we might as well + add it outside the loop */ + yy = ADD16(yy, 1); + + /* Calculations for position 0 are out of the loop, in part to reduce + mispredicted branches (since the if condition is usually false) + in the loop. */ + /* Temporary sums of the new pulse(s) */ + Rxy = EXTRACT16(SHR32(ADD32(xy, EXTEND32(X[0])),rshift)); + /* We're multiplying y[j] by two so we don't have to do it here */ + Ryy = ADD16(yy, y[0]); + + /* Approximate score: we maximise Rxy/sqrt(Ryy) (we're guaranteed that + Rxy is positive because the sign is pre-computed) */ + Rxy = MULT16_16_Q15(Rxy,Rxy); + best_den = Ryy; + best_num = Rxy; + j=1; + do { + /* Temporary sums of the new pulse(s) */ + Rxy = EXTRACT16(SHR32(ADD32(xy, EXTEND32(X[j])),rshift)); + /* We're multiplying y[j] by two so we don't have to do it here */ + Ryy = ADD16(yy, y[j]); + + /* Approximate score: we maximise Rxy/sqrt(Ryy) (we're guaranteed that + Rxy is positive because the sign is pre-computed) */ + Rxy = MULT16_16_Q15(Rxy,Rxy); + /* The idea is to check for num/den >= best_num/best_den, but that way + we can do it without any division */ + /* OPT: It's not clear whether a cmov is faster than a branch here + since the condition is more often false than true and using + a cmov introduces data dependencies across iterations. The optimal + choice may be architecture-dependent. */ + if (opus_unlikely(MULT16_16(best_den, Rxy) > MULT16_16(Ryy, best_num))) + { + best_den = Ryy; + best_num = Rxy; + best_id = j; + } + } while (++j<N); + + /* Updating the sums of the new pulse(s) */ + xy = ADD32(xy, EXTEND32(X[best_id])); + /* We're multiplying y[j] by two so we don't have to do it here */ + yy = ADD16(yy, y[best_id]); + + /* Only now that we've made the final choice, update y/iy */ + /* Multiplying y[j] by 2 so we don't have to do it everywhere else */ + y[best_id] += 2; + iy[best_id]++; + } + + /* Put the original sign back */ + j=0; + do { + /*iy[j] = signx[j] ? -iy[j] : iy[j];*/ + /* OPT: The is more likely to be compiled without a branch than the code above + but has the same performance otherwise. */ + iy[j] = (iy[j]^-signx[j]) + signx[j]; + } while (++j<N); + RESTORE_STACK; + return yy; +} + +unsigned alg_quant(celt_norm *X, int N, int K, int spread, int B, ec_enc *enc, + opus_val16 gain, int resynth, int arch) +{ + VARDECL(int, iy); + opus_val16 yy; + unsigned collapse_mask; + SAVE_STACK; + + celt_assert2(K>0, "alg_quant() needs at least one pulse"); + celt_assert2(N>1, "alg_quant() needs at least two dimensions"); + + /* Covers vectorization by up to 4. */ + ALLOC(iy, N+3, int); + + exp_rotation(X, N, 1, B, K, spread); + + yy = op_pvq_search(X, iy, K, N, arch); + + encode_pulses(iy, N, K, enc); + + if (resynth) + { + normalise_residual(iy, X, N, yy, gain); + exp_rotation(X, N, -1, B, K, spread); + } + + collapse_mask = extract_collapse_mask(iy, N, B); + RESTORE_STACK; + return collapse_mask; +} + +/** Decode pulse vector and combine the result with the pitch vector to produce + the final normalised signal in the current band. */ +unsigned alg_unquant(celt_norm *X, int N, int K, int spread, int B, + ec_dec *dec, opus_val16 gain) +{ + opus_val32 Ryy; + unsigned collapse_mask; + VARDECL(int, iy); + SAVE_STACK; + + celt_assert2(K>0, "alg_unquant() needs at least one pulse"); + celt_assert2(N>1, "alg_unquant() needs at least two dimensions"); + ALLOC(iy, N, int); + Ryy = decode_pulses(iy, N, K, dec); + normalise_residual(iy, X, N, Ryy, gain); + exp_rotation(X, N, -1, B, K, spread); + collapse_mask = extract_collapse_mask(iy, N, B); + RESTORE_STACK; + return collapse_mask; +} + +#ifndef OVERRIDE_renormalise_vector +void renormalise_vector(celt_norm *X, int N, opus_val16 gain, int arch) +{ + int i; +#ifdef FIXED_POINT + int k; +#endif + opus_val32 E; + opus_val16 g; + opus_val32 t; + celt_norm *xptr; + E = EPSILON + celt_inner_prod(X, X, N, arch); +#ifdef FIXED_POINT + k = celt_ilog2(E)>>1; +#endif + t = VSHR32(E, 2*(k-7)); + g = MULT16_16_P15(celt_rsqrt_norm(t),gain); + + xptr = X; + for (i=0;i<N;i++) + { + *xptr = EXTRACT16(PSHR32(MULT16_16(g, *xptr), k+1)); + xptr++; + } + /*return celt_sqrt(E);*/ +} +#endif /* OVERRIDE_renormalise_vector */ + +int stereo_itheta(const celt_norm *X, const celt_norm *Y, int stereo, int N, int arch) +{ + int i; + int itheta; + opus_val16 mid, side; + opus_val32 Emid, Eside; + + Emid = Eside = EPSILON; + if (stereo) + { + for (i=0;i<N;i++) + { + celt_norm m, s; + m = ADD16(SHR16(X[i],1),SHR16(Y[i],1)); + s = SUB16(SHR16(X[i],1),SHR16(Y[i],1)); + Emid = MAC16_16(Emid, m, m); + Eside = MAC16_16(Eside, s, s); + } + } else { + Emid += celt_inner_prod(X, X, N, arch); + Eside += celt_inner_prod(Y, Y, N, arch); + } + mid = celt_sqrt(Emid); + side = celt_sqrt(Eside); +#ifdef FIXED_POINT + /* 0.63662 = 2/pi */ + itheta = MULT16_16_Q15(QCONST16(0.63662f,15),celt_atan2p(side, mid)); +#else + itheta = (int)floor(.5f+16384*0.63662f*fast_atan2f(side,mid)); +#endif + + return itheta; +} diff --git a/libopus/celt/vq.h b/libopus/celt/vq.h new file mode 100644 index 0000000..45ec559 --- /dev/null +++ b/libopus/celt/vq.h @@ -0,0 +1,79 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file vq.h + @brief Vector quantisation of the residual + */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef VQ_H +#define VQ_H + +#include "entenc.h" +#include "entdec.h" +#include "modes.h" + +#if (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(FIXED_POINT)) +#include "x86/vq_sse.h" +#endif + +void exp_rotation(celt_norm *X, int len, int dir, int stride, int K, int spread); + +opus_val16 op_pvq_search_c(celt_norm *X, int *iy, int K, int N, int arch); + +#if !defined(OVERRIDE_OP_PVQ_SEARCH) +#define op_pvq_search(x, iy, K, N, arch) \ + (op_pvq_search_c(x, iy, K, N, arch)) +#endif + +/** Algebraic pulse-vector quantiser. The signal x is replaced by the sum of + * the pitch and a combination of pulses such that its norm is still equal + * to 1. This is the function that will typically require the most CPU. + * @param X Residual signal to quantise/encode (returns quantised version) + * @param N Number of samples to encode + * @param K Number of pulses to use + * @param enc Entropy encoder state + * @ret A mask indicating which blocks in the band received pulses +*/ +unsigned alg_quant(celt_norm *X, int N, int K, int spread, int B, ec_enc *enc, + opus_val16 gain, int resynth, int arch); + +/** Algebraic pulse decoder + * @param X Decoded normalised spectrum (returned) + * @param N Number of samples to decode + * @param K Number of pulses to use + * @param dec Entropy decoder state + * @ret A mask indicating which blocks in the band received pulses + */ +unsigned alg_unquant(celt_norm *X, int N, int K, int spread, int B, + ec_dec *dec, opus_val16 gain); + +void renormalise_vector(celt_norm *X, int N, opus_val16 gain, int arch); + +int stereo_itheta(const celt_norm *X, const celt_norm *Y, int stereo, int N, int arch); + +#endif /* VQ_H */ diff --git a/libopus/celt/x86/celt_lpc_sse.h b/libopus/celt/x86/celt_lpc_sse.h new file mode 100644 index 0000000..7d1ecf7 --- /dev/null +++ b/libopus/celt/x86/celt_lpc_sse.h @@ -0,0 +1,66 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CELT_LPC_SSE_H +#define CELT_LPC_SSE_H + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT) +#define OVERRIDE_CELT_FIR + +void celt_fir_sse4_1( + const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch); + +#if defined(OPUS_X86_PRESUME_SSE4_1) +#define celt_fir(x, num, y, N, ord, arch) \ + ((void)arch, celt_fir_sse4_1(x, num, y, N, ord, arch)) + +#else + +extern void (*const CELT_FIR_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch); + +# define celt_fir(x, num, y, N, ord, arch) \ + ((*CELT_FIR_IMPL[(arch) & OPUS_ARCHMASK])(x, num, y, N, ord, arch)) + +#endif +#endif + +#endif diff --git a/libopus/celt/x86/celt_lpc_sse4_1.c b/libopus/celt/x86/celt_lpc_sse4_1.c new file mode 100644 index 0000000..5478568 --- /dev/null +++ b/libopus/celt/x86/celt_lpc_sse4_1.c @@ -0,0 +1,89 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> +#include <smmintrin.h> +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" +#include "x86cpu.h" + +#if defined(FIXED_POINT) + +void celt_fir_sse4_1(const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch) +{ + int i,j; + VARDECL(opus_val16, rnum); + + __m128i vecNoA; + opus_int32 noA ; + SAVE_STACK; + + ALLOC(rnum, ord, opus_val16); + for(i=0;i<ord;i++) + rnum[i] = num[ord-i-1]; + noA = EXTEND32(1) << SIG_SHIFT >> 1; + vecNoA = _mm_set_epi32(noA, noA, noA, noA); + + for (i=0;i<N-3;i+=4) + { + opus_val32 sums[4] = {0}; + __m128i vecSum, vecX; + + xcorr_kernel(rnum, x+i-ord, sums, ord, arch); + + vecSum = _mm_loadu_si128((__m128i *)sums); + vecSum = _mm_add_epi32(vecSum, vecNoA); + vecSum = _mm_srai_epi32(vecSum, SIG_SHIFT); + vecX = OP_CVTEPI16_EPI32_M64(x + i); + vecSum = _mm_add_epi32(vecSum, vecX); + vecSum = _mm_packs_epi32(vecSum, vecSum); + _mm_storel_epi64((__m128i *)(y + i), vecSum); + } + for (;i<N;i++) + { + opus_val32 sum = 0; + for (j=0;j<ord;j++) + sum = MAC16_16(sum, rnum[j], x[i+j-ord]); + y[i] = SATURATE16(ADD32(EXTEND32(x[i]), PSHR32(sum, SIG_SHIFT))); + } + + RESTORE_STACK; +} + +#endif diff --git a/libopus/celt/x86/pitch_sse.c b/libopus/celt/x86/pitch_sse.c new file mode 100644 index 0000000..20e7312 --- /dev/null +++ b/libopus/celt/x86/pitch_sse.c @@ -0,0 +1,185 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "macros.h" +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" + +#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT) + +#include <xmmintrin.h> +#include "arch.h" + +void xcorr_kernel_sse(const opus_val16 *x, const opus_val16 *y, opus_val32 sum[4], int len) +{ + int j; + __m128 xsum1, xsum2; + xsum1 = _mm_loadu_ps(sum); + xsum2 = _mm_setzero_ps(); + + for (j = 0; j < len-3; j += 4) + { + __m128 x0 = _mm_loadu_ps(x+j); + __m128 yj = _mm_loadu_ps(y+j); + __m128 y3 = _mm_loadu_ps(y+j+3); + + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x00),yj)); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x55), + _mm_shuffle_ps(yj,y3,0x49))); + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xaa), + _mm_shuffle_ps(yj,y3,0x9e))); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xff),y3)); + } + if (j < len) + { + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + if (++j < len) + { + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + if (++j < len) + { + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + } + } + } + _mm_storeu_ps(sum,_mm_add_ps(xsum1,xsum2)); +} + + +void dual_inner_prod_sse(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + __m128 xsum1, xsum2; + xsum1 = _mm_setzero_ps(); + xsum2 = _mm_setzero_ps(); + for (i=0;i<N-3;i+=4) + { + __m128 xi = _mm_loadu_ps(x+i); + __m128 y1i = _mm_loadu_ps(y01+i); + __m128 y2i = _mm_loadu_ps(y02+i); + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(xi, y1i)); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(xi, y2i)); + } + /* Horizontal sum */ + xsum1 = _mm_add_ps(xsum1, _mm_movehl_ps(xsum1, xsum1)); + xsum1 = _mm_add_ss(xsum1, _mm_shuffle_ps(xsum1, xsum1, 0x55)); + _mm_store_ss(xy1, xsum1); + xsum2 = _mm_add_ps(xsum2, _mm_movehl_ps(xsum2, xsum2)); + xsum2 = _mm_add_ss(xsum2, _mm_shuffle_ps(xsum2, xsum2, 0x55)); + _mm_store_ss(xy2, xsum2); + for (;i<N;i++) + { + *xy1 = MAC16_16(*xy1, x[i], y01[i]); + *xy2 = MAC16_16(*xy2, x[i], y02[i]); + } +} + +opus_val32 celt_inner_prod_sse(const opus_val16 *x, const opus_val16 *y, + int N) +{ + int i; + float xy; + __m128 sum; + sum = _mm_setzero_ps(); + /* FIXME: We should probably go 8-way and use 2 sums. */ + for (i=0;i<N-3;i+=4) + { + __m128 xi = _mm_loadu_ps(x+i); + __m128 yi = _mm_loadu_ps(y+i); + sum = _mm_add_ps(sum,_mm_mul_ps(xi, yi)); + } + /* Horizontal sum */ + sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum)); + sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55)); + _mm_store_ss(&xy, sum); + for (;i<N;i++) + { + xy = MAC16_16(xy, x[i], y[i]); + } + return xy; +} + +void comb_filter_const_sse(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12) +{ + int i; + __m128 x0v; + __m128 g10v, g11v, g12v; + g10v = _mm_load1_ps(&g10); + g11v = _mm_load1_ps(&g11); + g12v = _mm_load1_ps(&g12); + x0v = _mm_loadu_ps(&x[-T-2]); + for (i=0;i<N-3;i+=4) + { + __m128 yi, yi2, x1v, x2v, x3v, x4v; + const opus_val32 *xp = &x[i-T-2]; + yi = _mm_loadu_ps(x+i); + x4v = _mm_loadu_ps(xp+4); +#if 0 + /* Slower version with all loads */ + x1v = _mm_loadu_ps(xp+1); + x2v = _mm_loadu_ps(xp+2); + x3v = _mm_loadu_ps(xp+3); +#else + x2v = _mm_shuffle_ps(x0v, x4v, 0x4e); + x1v = _mm_shuffle_ps(x0v, x2v, 0x99); + x3v = _mm_shuffle_ps(x2v, x4v, 0x99); +#endif + + yi = _mm_add_ps(yi, _mm_mul_ps(g10v,x2v)); +#if 0 /* Set to 1 to make it bit-exact with the non-SSE version */ + yi = _mm_add_ps(yi, _mm_mul_ps(g11v,_mm_add_ps(x3v,x1v))); + yi = _mm_add_ps(yi, _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v))); +#else + /* Use partial sums */ + yi2 = _mm_add_ps(_mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)), + _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v))); + yi = _mm_add_ps(yi, yi2); +#endif + x0v=x4v; + _mm_storeu_ps(y+i, yi); + } +#ifdef CUSTOM_MODES + for (;i<N;i++) + { + y[i] = x[i] + + MULT16_32_Q15(g10,x[i-T]) + + MULT16_32_Q15(g11,ADD32(x[i-T+1],x[i-T-1])) + + MULT16_32_Q15(g12,ADD32(x[i-T+2],x[i-T-2])); + } +#endif +} + + +#endif diff --git a/libopus/celt/x86/pitch_sse.h b/libopus/celt/x86/pitch_sse.h new file mode 100644 index 0000000..e5f87ab --- /dev/null +++ b/libopus/celt/x86/pitch_sse.h @@ -0,0 +1,192 @@ +/* Copyright (c) 2013 Jean-Marc Valin and John Ridges + Copyright (c) 2014, Cisco Systems, INC MingXiang WeiZhou MinPeng YanWang*/ +/** + @file pitch_sse.h + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PITCH_SSE_H +#define PITCH_SSE_H + +#if defined(HAVE_CONFIG_H) +#include "config.h" +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT) +void xcorr_kernel_sse4_1( + const opus_int16 *x, + const opus_int16 *y, + opus_val32 sum[4], + int len); +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT) +void xcorr_kernel_sse( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len); +#endif + +#if defined(OPUS_X86_PRESUME_SSE4_1) && defined(FIXED_POINT) +#define OVERRIDE_XCORR_KERNEL +#define xcorr_kernel(x, y, sum, len, arch) \ + ((void)arch, xcorr_kernel_sse4_1(x, y, sum, len)) + +#elif defined(OPUS_X86_PRESUME_SSE) && !defined(FIXED_POINT) +#define OVERRIDE_XCORR_KERNEL +#define xcorr_kernel(x, y, sum, len, arch) \ + ((void)arch, xcorr_kernel_sse(x, y, sum, len)) + +#elif (defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT)) || (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)) + +extern void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len); + +#define OVERRIDE_XCORR_KERNEL +#define xcorr_kernel(x, y, sum, len, arch) \ + ((*XCORR_KERNEL_IMPL[(arch) & OPUS_ARCHMASK])(x, y, sum, len)) + +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT) +opus_val32 celt_inner_prod_sse4_1( + const opus_int16 *x, + const opus_int16 *y, + int N); +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE2) && defined(FIXED_POINT) +opus_val32 celt_inner_prod_sse2( + const opus_int16 *x, + const opus_int16 *y, + int N); +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(FIXED_POINT) +opus_val32 celt_inner_prod_sse( + const opus_val16 *x, + const opus_val16 *y, + int N); +#endif + + +#if defined(OPUS_X86_PRESUME_SSE4_1) && defined(FIXED_POINT) +#define OVERRIDE_CELT_INNER_PROD +#define celt_inner_prod(x, y, N, arch) \ + ((void)arch, celt_inner_prod_sse4_1(x, y, N)) + +#elif defined(OPUS_X86_PRESUME_SSE2) && defined(FIXED_POINT) && !defined(OPUS_X86_MAY_HAVE_SSE4_1) +#define OVERRIDE_CELT_INNER_PROD +#define celt_inner_prod(x, y, N, arch) \ + ((void)arch, celt_inner_prod_sse2(x, y, N)) + +#elif defined(OPUS_X86_PRESUME_SSE) && !defined(FIXED_POINT) +#define OVERRIDE_CELT_INNER_PROD +#define celt_inner_prod(x, y, N, arch) \ + ((void)arch, celt_inner_prod_sse(x, y, N)) + + +#elif ((defined(OPUS_X86_MAY_HAVE_SSE4_1) || defined(OPUS_X86_MAY_HAVE_SSE2)) && defined(FIXED_POINT)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)) + +extern opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + int N); + +#define OVERRIDE_CELT_INNER_PROD +#define celt_inner_prod(x, y, N, arch) \ + ((*CELT_INNER_PROD_IMPL[(arch) & OPUS_ARCHMASK])(x, y, N)) + +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT) + +#define OVERRIDE_DUAL_INNER_PROD +#define OVERRIDE_COMB_FILTER_CONST + +#undef dual_inner_prod +#undef comb_filter_const + +void dual_inner_prod_sse(const opus_val16 *x, + const opus_val16 *y01, + const opus_val16 *y02, + int N, + opus_val32 *xy1, + opus_val32 *xy2); + +void comb_filter_const_sse(opus_val32 *y, + opus_val32 *x, + int T, + int N, + opus_val16 g10, + opus_val16 g11, + opus_val16 g12); + + +#if defined(OPUS_X86_PRESUME_SSE) +# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \ + ((void)(arch),dual_inner_prod_sse(x, y01, y02, N, xy1, xy2)) + +# define comb_filter_const(y, x, T, N, g10, g11, g12, arch) \ + ((void)(arch),comb_filter_const_sse(y, x, T, N, g10, g11, g12)) +#else + +extern void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y01, + const opus_val16 *y02, + int N, + opus_val32 *xy1, + opus_val32 *xy2); + +#define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \ + ((*DUAL_INNER_PROD_IMPL[(arch) & OPUS_ARCHMASK])(x, y01, y02, N, xy1, xy2)) + +extern void (*const COMB_FILTER_CONST_IMPL[OPUS_ARCHMASK + 1])( + opus_val32 *y, + opus_val32 *x, + int T, + int N, + opus_val16 g10, + opus_val16 g11, + opus_val16 g12); + +#define comb_filter_const(y, x, T, N, g10, g11, g12, arch) \ + ((*COMB_FILTER_CONST_IMPL[(arch) & OPUS_ARCHMASK])(y, x, T, N, g10, g11, g12)) + +#define NON_STATIC_COMB_FILTER_CONST_C + +#endif +#endif + +#endif diff --git a/libopus/celt/x86/pitch_sse2.c b/libopus/celt/x86/pitch_sse2.c new file mode 100644 index 0000000..a0e7d1b --- /dev/null +++ b/libopus/celt/x86/pitch_sse2.c @@ -0,0 +1,95 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> + +#include "macros.h" +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" + +#if defined(OPUS_X86_MAY_HAVE_SSE2) && defined(FIXED_POINT) +opus_val32 celt_inner_prod_sse2(const opus_val16 *x, const opus_val16 *y, + int N) +{ + opus_int i, dataSize16; + opus_int32 sum; + + __m128i inVec1_76543210, inVec1_FEDCBA98, acc1; + __m128i inVec2_76543210, inVec2_FEDCBA98, acc2; + + sum = 0; + dataSize16 = N & ~15; + + acc1 = _mm_setzero_si128(); + acc2 = _mm_setzero_si128(); + + for (i=0;i<dataSize16;i+=16) + { + inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0])); + inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0])); + + inVec1_FEDCBA98 = _mm_loadu_si128((__m128i *)(&x[i + 8])); + inVec2_FEDCBA98 = _mm_loadu_si128((__m128i *)(&y[i + 8])); + + inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210); + inVec1_FEDCBA98 = _mm_madd_epi16(inVec1_FEDCBA98, inVec2_FEDCBA98); + + acc1 = _mm_add_epi32(acc1, inVec1_76543210); + acc2 = _mm_add_epi32(acc2, inVec1_FEDCBA98); + } + + acc1 = _mm_add_epi32( acc1, acc2 ); + + if (N - i >= 8) + { + inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0])); + inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0])); + + inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210); + + acc1 = _mm_add_epi32(acc1, inVec1_76543210); + i += 8; + } + + acc1 = _mm_add_epi32(acc1, _mm_unpackhi_epi64( acc1, acc1)); + acc1 = _mm_add_epi32(acc1, _mm_shufflelo_epi16( acc1, 0x0E)); + sum += _mm_cvtsi128_si32(acc1); + + for (;i<N;i++) { + sum = silk_SMLABB(sum, x[i], y[i]); + } + + return sum; +} +#endif diff --git a/libopus/celt/x86/pitch_sse4_1.c b/libopus/celt/x86/pitch_sse4_1.c new file mode 100644 index 0000000..a092c68 --- /dev/null +++ b/libopus/celt/x86/pitch_sse4_1.c @@ -0,0 +1,195 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> + +#include "macros.h" +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT) +#include <smmintrin.h> +#include "x86cpu.h" + +opus_val32 celt_inner_prod_sse4_1(const opus_val16 *x, const opus_val16 *y, + int N) +{ + opus_int i, dataSize16; + opus_int32 sum; + __m128i inVec1_76543210, inVec1_FEDCBA98, acc1; + __m128i inVec2_76543210, inVec2_FEDCBA98, acc2; + __m128i inVec1_3210, inVec2_3210; + + sum = 0; + dataSize16 = N & ~15; + + acc1 = _mm_setzero_si128(); + acc2 = _mm_setzero_si128(); + + for (i=0;i<dataSize16;i+=16) { + inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0])); + inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0])); + + inVec1_FEDCBA98 = _mm_loadu_si128((__m128i *)(&x[i + 8])); + inVec2_FEDCBA98 = _mm_loadu_si128((__m128i *)(&y[i + 8])); + + inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210); + inVec1_FEDCBA98 = _mm_madd_epi16(inVec1_FEDCBA98, inVec2_FEDCBA98); + + acc1 = _mm_add_epi32(acc1, inVec1_76543210); + acc2 = _mm_add_epi32(acc2, inVec1_FEDCBA98); + } + + acc1 = _mm_add_epi32(acc1, acc2); + + if (N - i >= 8) + { + inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0])); + inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0])); + + inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210); + + acc1 = _mm_add_epi32(acc1, inVec1_76543210); + i += 8; + } + + if (N - i >= 4) + { + inVec1_3210 = OP_CVTEPI16_EPI32_M64(&x[i + 0]); + inVec2_3210 = OP_CVTEPI16_EPI32_M64(&y[i + 0]); + + inVec1_3210 = _mm_mullo_epi32(inVec1_3210, inVec2_3210); + + acc1 = _mm_add_epi32(acc1, inVec1_3210); + i += 4; + } + + acc1 = _mm_add_epi32(acc1, _mm_unpackhi_epi64(acc1, acc1)); + acc1 = _mm_add_epi32(acc1, _mm_shufflelo_epi16(acc1, 0x0E)); + + sum += _mm_cvtsi128_si32(acc1); + + for (;i<N;i++) + { + sum = silk_SMLABB(sum, x[i], y[i]); + } + + return sum; +} + +void xcorr_kernel_sse4_1(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[ 4 ], int len) +{ + int j; + + __m128i vecX, vecX0, vecX1, vecX2, vecX3; + __m128i vecY0, vecY1, vecY2, vecY3; + __m128i sum0, sum1, sum2, sum3, vecSum; + __m128i initSum; + + celt_assert(len >= 3); + + sum0 = _mm_setzero_si128(); + sum1 = _mm_setzero_si128(); + sum2 = _mm_setzero_si128(); + sum3 = _mm_setzero_si128(); + + for (j=0;j<(len-7);j+=8) + { + vecX = _mm_loadu_si128((__m128i *)(&x[j + 0])); + vecY0 = _mm_loadu_si128((__m128i *)(&y[j + 0])); + vecY1 = _mm_loadu_si128((__m128i *)(&y[j + 1])); + vecY2 = _mm_loadu_si128((__m128i *)(&y[j + 2])); + vecY3 = _mm_loadu_si128((__m128i *)(&y[j + 3])); + + sum0 = _mm_add_epi32(sum0, _mm_madd_epi16(vecX, vecY0)); + sum1 = _mm_add_epi32(sum1, _mm_madd_epi16(vecX, vecY1)); + sum2 = _mm_add_epi32(sum2, _mm_madd_epi16(vecX, vecY2)); + sum3 = _mm_add_epi32(sum3, _mm_madd_epi16(vecX, vecY3)); + } + + sum0 = _mm_add_epi32(sum0, _mm_unpackhi_epi64( sum0, sum0)); + sum0 = _mm_add_epi32(sum0, _mm_shufflelo_epi16( sum0, 0x0E)); + + sum1 = _mm_add_epi32(sum1, _mm_unpackhi_epi64( sum1, sum1)); + sum1 = _mm_add_epi32(sum1, _mm_shufflelo_epi16( sum1, 0x0E)); + + sum2 = _mm_add_epi32(sum2, _mm_unpackhi_epi64( sum2, sum2)); + sum2 = _mm_add_epi32(sum2, _mm_shufflelo_epi16( sum2, 0x0E)); + + sum3 = _mm_add_epi32(sum3, _mm_unpackhi_epi64( sum3, sum3)); + sum3 = _mm_add_epi32(sum3, _mm_shufflelo_epi16( sum3, 0x0E)); + + vecSum = _mm_unpacklo_epi64(_mm_unpacklo_epi32(sum0, sum1), + _mm_unpacklo_epi32(sum2, sum3)); + + for (;j<(len-3);j+=4) + { + vecX = OP_CVTEPI16_EPI32_M64(&x[j + 0]); + vecX0 = _mm_shuffle_epi32(vecX, 0x00); + vecX1 = _mm_shuffle_epi32(vecX, 0x55); + vecX2 = _mm_shuffle_epi32(vecX, 0xaa); + vecX3 = _mm_shuffle_epi32(vecX, 0xff); + + vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]); + vecY1 = OP_CVTEPI16_EPI32_M64(&y[j + 1]); + vecY2 = OP_CVTEPI16_EPI32_M64(&y[j + 2]); + vecY3 = OP_CVTEPI16_EPI32_M64(&y[j + 3]); + + sum0 = _mm_mullo_epi32(vecX0, vecY0); + sum1 = _mm_mullo_epi32(vecX1, vecY1); + sum2 = _mm_mullo_epi32(vecX2, vecY2); + sum3 = _mm_mullo_epi32(vecX3, vecY3); + + sum0 = _mm_add_epi32(sum0, sum1); + sum2 = _mm_add_epi32(sum2, sum3); + vecSum = _mm_add_epi32(vecSum, sum0); + vecSum = _mm_add_epi32(vecSum, sum2); + } + + for (;j<len;j++) + { + vecX = OP_CVTEPI16_EPI32_M64(&x[j + 0]); + vecX0 = _mm_shuffle_epi32(vecX, 0x00); + + vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]); + + sum0 = _mm_mullo_epi32(vecX0, vecY0); + vecSum = _mm_add_epi32(vecSum, sum0); + } + + initSum = _mm_loadu_si128((__m128i *)(&sum[0])); + initSum = _mm_add_epi32(initSum, vecSum); + _mm_storeu_si128((__m128i *)sum, initSum); +} +#endif diff --git a/libopus/celt/x86/vq_sse.h b/libopus/celt/x86/vq_sse.h new file mode 100644 index 0000000..b4efe8f --- /dev/null +++ b/libopus/celt/x86/vq_sse.h @@ -0,0 +1,50 @@ +/* Copyright (c) 2016 Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef VQ_SSE_H +#define VQ_SSE_H + +#if defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(FIXED_POINT) +#define OVERRIDE_OP_PVQ_SEARCH + +opus_val16 op_pvq_search_sse2(celt_norm *_X, int *iy, int K, int N, int arch); + +#if defined(OPUS_X86_PRESUME_SSE2) +#define op_pvq_search(x, iy, K, N, arch) \ + (op_pvq_search_sse2(x, iy, K, N, arch)) + +#else + +extern opus_val16 (*const OP_PVQ_SEARCH_IMPL[OPUS_ARCHMASK + 1])( + celt_norm *_X, int *iy, int K, int N, int arch); + +# define op_pvq_search(X, iy, K, N, arch) \ + ((*OP_PVQ_SEARCH_IMPL[(arch) & OPUS_ARCHMASK])(X, iy, K, N, arch)) + +#endif +#endif + +#endif diff --git a/libopus/celt/x86/vq_sse2.c b/libopus/celt/x86/vq_sse2.c new file mode 100644 index 0000000..7750428 --- /dev/null +++ b/libopus/celt/x86/vq_sse2.c @@ -0,0 +1,217 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2007-2016 Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "vq.h" +#include "x86cpu.h" + + +#ifndef FIXED_POINT + +opus_val16 op_pvq_search_sse2(celt_norm *_X, int *iy, int K, int N, int arch) +{ + int i, j; + int pulsesLeft; + float xy, yy; + VARDECL(celt_norm, y); + VARDECL(celt_norm, X); + VARDECL(float, signy); + __m128 signmask; + __m128 sums; + __m128i fours; + SAVE_STACK; + + (void)arch; + /* All bits set to zero, except for the sign bit. */ + signmask = _mm_set_ps1(-0.f); + fours = _mm_set_epi32(4, 4, 4, 4); + ALLOC(y, N+3, celt_norm); + ALLOC(X, N+3, celt_norm); + ALLOC(signy, N+3, float); + + OPUS_COPY(X, _X, N); + X[N] = X[N+1] = X[N+2] = 0; + sums = _mm_setzero_ps(); + for (j=0;j<N;j+=4) + { + __m128 x4, s4; + x4 = _mm_loadu_ps(&X[j]); + s4 = _mm_cmplt_ps(x4, _mm_setzero_ps()); + /* Get rid of the sign */ + x4 = _mm_andnot_ps(signmask, x4); + sums = _mm_add_ps(sums, x4); + /* Clear y and iy in case we don't do the projection. */ + _mm_storeu_ps(&y[j], _mm_setzero_ps()); + _mm_storeu_si128((__m128i*)&iy[j], _mm_setzero_si128()); + _mm_storeu_ps(&X[j], x4); + _mm_storeu_ps(&signy[j], s4); + } + sums = _mm_add_ps(sums, _mm_shuffle_ps(sums, sums, _MM_SHUFFLE(1, 0, 3, 2))); + sums = _mm_add_ps(sums, _mm_shuffle_ps(sums, sums, _MM_SHUFFLE(2, 3, 0, 1))); + + xy = yy = 0; + + pulsesLeft = K; + + /* Do a pre-search by projecting on the pyramid */ + if (K > (N>>1)) + { + __m128i pulses_sum; + __m128 yy4, xy4; + __m128 rcp4; + opus_val32 sum = _mm_cvtss_f32(sums); + /* If X is too small, just replace it with a pulse at 0 */ + /* Prevents infinities and NaNs from causing too many pulses + to be allocated. 64 is an approximation of infinity here. */ + if (!(sum > EPSILON && sum < 64)) + { + X[0] = QCONST16(1.f,14); + j=1; do + X[j]=0; + while (++j<N); + sums = _mm_set_ps1(1.f); + } + /* Using K+e with e < 1 guarantees we cannot get more than K pulses. */ + rcp4 = _mm_mul_ps(_mm_set_ps1((float)(K+.8)), _mm_rcp_ps(sums)); + xy4 = yy4 = _mm_setzero_ps(); + pulses_sum = _mm_setzero_si128(); + for (j=0;j<N;j+=4) + { + __m128 rx4, x4, y4; + __m128i iy4; + x4 = _mm_loadu_ps(&X[j]); + rx4 = _mm_mul_ps(x4, rcp4); + iy4 = _mm_cvttps_epi32(rx4); + pulses_sum = _mm_add_epi32(pulses_sum, iy4); + _mm_storeu_si128((__m128i*)&iy[j], iy4); + y4 = _mm_cvtepi32_ps(iy4); + xy4 = _mm_add_ps(xy4, _mm_mul_ps(x4, y4)); + yy4 = _mm_add_ps(yy4, _mm_mul_ps(y4, y4)); + /* double the y[] vector so we don't have to do it in the search loop. */ + _mm_storeu_ps(&y[j], _mm_add_ps(y4, y4)); + } + pulses_sum = _mm_add_epi32(pulses_sum, _mm_shuffle_epi32(pulses_sum, _MM_SHUFFLE(1, 0, 3, 2))); + pulses_sum = _mm_add_epi32(pulses_sum, _mm_shuffle_epi32(pulses_sum, _MM_SHUFFLE(2, 3, 0, 1))); + pulsesLeft -= _mm_cvtsi128_si32(pulses_sum); + xy4 = _mm_add_ps(xy4, _mm_shuffle_ps(xy4, xy4, _MM_SHUFFLE(1, 0, 3, 2))); + xy4 = _mm_add_ps(xy4, _mm_shuffle_ps(xy4, xy4, _MM_SHUFFLE(2, 3, 0, 1))); + xy = _mm_cvtss_f32(xy4); + yy4 = _mm_add_ps(yy4, _mm_shuffle_ps(yy4, yy4, _MM_SHUFFLE(1, 0, 3, 2))); + yy4 = _mm_add_ps(yy4, _mm_shuffle_ps(yy4, yy4, _MM_SHUFFLE(2, 3, 0, 1))); + yy = _mm_cvtss_f32(yy4); + } + X[N] = X[N+1] = X[N+2] = -100; + y[N] = y[N+1] = y[N+2] = 100; + celt_sig_assert(pulsesLeft>=0); + + /* This should never happen, but just in case it does (e.g. on silence) + we fill the first bin with pulses. */ + if (pulsesLeft > N+3) + { + opus_val16 tmp = (opus_val16)pulsesLeft; + yy = MAC16_16(yy, tmp, tmp); + yy = MAC16_16(yy, tmp, y[0]); + iy[0] += pulsesLeft; + pulsesLeft=0; + } + + for (i=0;i<pulsesLeft;i++) + { + int best_id; + __m128 xy4, yy4; + __m128 max, max2; + __m128i count; + __m128i pos; + /* The squared magnitude term gets added anyway, so we might as well + add it outside the loop */ + yy = ADD16(yy, 1); + xy4 = _mm_load1_ps(&xy); + yy4 = _mm_load1_ps(&yy); + max = _mm_setzero_ps(); + pos = _mm_setzero_si128(); + count = _mm_set_epi32(3, 2, 1, 0); + for (j=0;j<N;j+=4) + { + __m128 x4, y4, r4; + x4 = _mm_loadu_ps(&X[j]); + y4 = _mm_loadu_ps(&y[j]); + x4 = _mm_add_ps(x4, xy4); + y4 = _mm_add_ps(y4, yy4); + y4 = _mm_rsqrt_ps(y4); + r4 = _mm_mul_ps(x4, y4); + /* Update the index of the max. */ + pos = _mm_max_epi16(pos, _mm_and_si128(count, _mm_castps_si128(_mm_cmpgt_ps(r4, max)))); + /* Update the max. */ + max = _mm_max_ps(max, r4); + /* Update the indices (+4) */ + count = _mm_add_epi32(count, fours); + } + /* Horizontal max */ + max2 = _mm_max_ps(max, _mm_shuffle_ps(max, max, _MM_SHUFFLE(1, 0, 3, 2))); + max2 = _mm_max_ps(max2, _mm_shuffle_ps(max2, max2, _MM_SHUFFLE(2, 3, 0, 1))); + /* Now that max2 contains the max at all positions, look at which value(s) of the + partial max is equal to the global max. */ + pos = _mm_and_si128(pos, _mm_castps_si128(_mm_cmpeq_ps(max, max2))); + pos = _mm_max_epi16(pos, _mm_unpackhi_epi64(pos, pos)); + pos = _mm_max_epi16(pos, _mm_shufflelo_epi16(pos, _MM_SHUFFLE(1, 0, 3, 2))); + best_id = _mm_cvtsi128_si32(pos); + + /* Updating the sums of the new pulse(s) */ + xy = ADD32(xy, EXTEND32(X[best_id])); + /* We're multiplying y[j] by two so we don't have to do it here */ + yy = ADD16(yy, y[best_id]); + + /* Only now that we've made the final choice, update y/iy */ + /* Multiplying y[j] by 2 so we don't have to do it everywhere else */ + y[best_id] += 2; + iy[best_id]++; + } + + /* Put the original sign back */ + for (j=0;j<N;j+=4) + { + __m128i y4; + __m128i s4; + y4 = _mm_loadu_si128((__m128i*)&iy[j]); + s4 = _mm_castps_si128(_mm_loadu_ps(&signy[j])); + y4 = _mm_xor_si128(_mm_add_epi32(y4, s4), s4); + _mm_storeu_si128((__m128i*)&iy[j], y4); + } + RESTORE_STACK; + return yy; +} + +#endif diff --git a/libopus/celt/x86/x86_celt_map.c b/libopus/celt/x86/x86_celt_map.c new file mode 100644 index 0000000..d39d88e --- /dev/null +++ b/libopus/celt/x86/x86_celt_map.c @@ -0,0 +1,167 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if defined(HAVE_CONFIG_H) +#include "config.h" +#endif + +#include "x86/x86cpu.h" +#include "celt_lpc.h" +#include "pitch.h" +#include "pitch_sse.h" +#include "vq.h" + +#if defined(OPUS_HAVE_RTCD) + +# if defined(FIXED_POINT) + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1) + +void (*const CELT_FIR_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch +) = { + celt_fir_c, /* non-sse */ + celt_fir_c, + celt_fir_c, + MAY_HAVE_SSE4_1(celt_fir), /* sse4.1 */ + MAY_HAVE_SSE4_1(celt_fir) /* avx */ +}; + +void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len +) = { + xcorr_kernel_c, /* non-sse */ + xcorr_kernel_c, + xcorr_kernel_c, + MAY_HAVE_SSE4_1(xcorr_kernel), /* sse4.1 */ + MAY_HAVE_SSE4_1(xcorr_kernel) /* avx */ +}; + +#endif + +#if (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \ + (!defined(OPUS_X86_MAY_HAVE_SSE_4_1) && defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) + +opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + int N +) = { + celt_inner_prod_c, /* non-sse */ + celt_inner_prod_c, + MAY_HAVE_SSE2(celt_inner_prod), + MAY_HAVE_SSE4_1(celt_inner_prod), /* sse4.1 */ + MAY_HAVE_SSE4_1(celt_inner_prod) /* avx */ +}; + +#endif + +# else + +#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE) + +void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len +) = { + xcorr_kernel_c, /* non-sse */ + MAY_HAVE_SSE(xcorr_kernel), + MAY_HAVE_SSE(xcorr_kernel), + MAY_HAVE_SSE(xcorr_kernel), + MAY_HAVE_SSE(xcorr_kernel) +}; + +opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + int N +) = { + celt_inner_prod_c, /* non-sse */ + MAY_HAVE_SSE(celt_inner_prod), + MAY_HAVE_SSE(celt_inner_prod), + MAY_HAVE_SSE(celt_inner_prod), + MAY_HAVE_SSE(celt_inner_prod) +}; + +void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y01, + const opus_val16 *y02, + int N, + opus_val32 *xy1, + opus_val32 *xy2 +) = { + dual_inner_prod_c, /* non-sse */ + MAY_HAVE_SSE(dual_inner_prod), + MAY_HAVE_SSE(dual_inner_prod), + MAY_HAVE_SSE(dual_inner_prod), + MAY_HAVE_SSE(dual_inner_prod) +}; + +void (*const COMB_FILTER_CONST_IMPL[OPUS_ARCHMASK + 1])( + opus_val32 *y, + opus_val32 *x, + int T, + int N, + opus_val16 g10, + opus_val16 g11, + opus_val16 g12 +) = { + comb_filter_const_c, /* non-sse */ + MAY_HAVE_SSE(comb_filter_const), + MAY_HAVE_SSE(comb_filter_const), + MAY_HAVE_SSE(comb_filter_const), + MAY_HAVE_SSE(comb_filter_const) +}; + + +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2) +opus_val16 (*const OP_PVQ_SEARCH_IMPL[OPUS_ARCHMASK + 1])( + celt_norm *_X, int *iy, int K, int N, int arch +) = { + op_pvq_search_c, /* non-sse */ + op_pvq_search_c, + MAY_HAVE_SSE2(op_pvq_search), + MAY_HAVE_SSE2(op_pvq_search), + MAY_HAVE_SSE2(op_pvq_search) +}; +#endif + +#endif +#endif diff --git a/libopus/celt/x86/x86cpu.c b/libopus/celt/x86/x86cpu.c new file mode 100644 index 0000000..080eb25 --- /dev/null +++ b/libopus/celt/x86/x86cpu.c @@ -0,0 +1,157 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "cpu_support.h" +#include "macros.h" +#include "main.h" +#include "pitch.h" +#include "x86cpu.h" + +#if (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \ + (defined(OPUS_X86_MAY_HAVE_AVX) && !defined(OPUS_X86_PRESUME_AVX)) + + +#if defined(_MSC_VER) + +#include <intrin.h> +static _inline void cpuid(unsigned int CPUInfo[4], unsigned int InfoType) +{ + __cpuid((int*)CPUInfo, InfoType); +} + +#else + +#if defined(CPU_INFO_BY_C) +#include <cpuid.h> +#endif + +static void cpuid(unsigned int CPUInfo[4], unsigned int InfoType) +{ +#if defined(CPU_INFO_BY_ASM) +#if defined(__i386__) && defined(__PIC__) +/* %ebx is PIC register in 32-bit, so mustn't clobber it. */ + __asm__ __volatile__ ( + "xchg %%ebx, %1\n" + "cpuid\n" + "xchg %%ebx, %1\n": + "=a" (CPUInfo[0]), + "=r" (CPUInfo[1]), + "=c" (CPUInfo[2]), + "=d" (CPUInfo[3]) : + "0" (InfoType) + ); +#else + __asm__ __volatile__ ( + "cpuid": + "=a" (CPUInfo[0]), + "=b" (CPUInfo[1]), + "=c" (CPUInfo[2]), + "=d" (CPUInfo[3]) : + "0" (InfoType) + ); +#endif +#elif defined(CPU_INFO_BY_C) + __get_cpuid(InfoType, &(CPUInfo[0]), &(CPUInfo[1]), &(CPUInfo[2]), &(CPUInfo[3])); +#endif +} + +#endif + +typedef struct CPU_Feature{ + /* SIMD: 128-bit */ + int HW_SSE; + int HW_SSE2; + int HW_SSE41; + /* SIMD: 256-bit */ + int HW_AVX; +} CPU_Feature; + +static void opus_cpu_feature_check(CPU_Feature *cpu_feature) +{ + unsigned int info[4] = {0}; + unsigned int nIds = 0; + + cpuid(info, 0); + nIds = info[0]; + + if (nIds >= 1){ + cpuid(info, 1); + cpu_feature->HW_SSE = (info[3] & (1 << 25)) != 0; + cpu_feature->HW_SSE2 = (info[3] & (1 << 26)) != 0; + cpu_feature->HW_SSE41 = (info[2] & (1 << 19)) != 0; + cpu_feature->HW_AVX = (info[2] & (1 << 28)) != 0; + } + else { + cpu_feature->HW_SSE = 0; + cpu_feature->HW_SSE2 = 0; + cpu_feature->HW_SSE41 = 0; + cpu_feature->HW_AVX = 0; + } +} + +int opus_select_arch(void) +{ + CPU_Feature cpu_feature; + int arch; + + opus_cpu_feature_check(&cpu_feature); + + arch = 0; + if (!cpu_feature.HW_SSE) + { + return arch; + } + arch++; + + if (!cpu_feature.HW_SSE2) + { + return arch; + } + arch++; + + if (!cpu_feature.HW_SSE41) + { + return arch; + } + arch++; + + if (!cpu_feature.HW_AVX) + { + return arch; + } + arch++; + + return arch; +} + +#endif diff --git a/libopus/celt/x86/x86cpu.h b/libopus/celt/x86/x86cpu.h new file mode 100644 index 0000000..1e2bf17 --- /dev/null +++ b/libopus/celt/x86/x86cpu.h @@ -0,0 +1,95 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(X86CPU_H) +# define X86CPU_H + +# if defined(OPUS_X86_MAY_HAVE_SSE) +# define MAY_HAVE_SSE(name) name ## _sse +# else +# define MAY_HAVE_SSE(name) name ## _c +# endif + +# if defined(OPUS_X86_MAY_HAVE_SSE2) +# define MAY_HAVE_SSE2(name) name ## _sse2 +# else +# define MAY_HAVE_SSE2(name) name ## _c +# endif + +# if defined(OPUS_X86_MAY_HAVE_SSE4_1) +# define MAY_HAVE_SSE4_1(name) name ## _sse4_1 +# else +# define MAY_HAVE_SSE4_1(name) name ## _c +# endif + +# if defined(OPUS_X86_MAY_HAVE_AVX) +# define MAY_HAVE_AVX(name) name ## _avx +# else +# define MAY_HAVE_AVX(name) name ## _c +# endif + +# if defined(OPUS_HAVE_RTCD) +int opus_select_arch(void); +# endif + +/*gcc appears to emit MOVDQA's to load the argument of an _mm_cvtepi8_epi32() + or _mm_cvtepi16_epi32() when optimizations are disabled, even though the + actual PMOVSXWD instruction takes an m32 or m64. Unlike a normal memory + reference, these require 16-byte alignment and load a full 16 bytes (instead + of 4 or 8), possibly reading out of bounds. + + We can insert an explicit MOVD or MOVQ using _mm_cvtsi32_si128() or + _mm_loadl_epi64(), which should have the same semantics as an m32 or m64 + reference in the PMOVSXWD instruction itself, but gcc is not smart enough to + optimize this out when optimizations ARE enabled. + + Clang, in contrast, requires us to do this always for _mm_cvtepi8_epi32 + (which is fair, since technically the compiler is always allowed to do the + dereference before invoking the function implementing the intrinsic). + However, it is smart enough to eliminate the extra MOVD instruction. + For _mm_cvtepi16_epi32, it does the right thing, though does *not* optimize out + the extra MOVQ if it's specified explicitly */ + +# if defined(__clang__) || !defined(__OPTIMIZE__) +# define OP_CVTEPI8_EPI32_M32(x) \ + (_mm_cvtepi8_epi32(_mm_cvtsi32_si128(*(int *)(x)))) +# else +# define OP_CVTEPI8_EPI32_M32(x) \ + (_mm_cvtepi8_epi32(*(__m128i *)(x))) +#endif + +/* similar reasoning about the instruction sequence as in the 32-bit macro above, + */ +# if defined(__clang__) || !defined(__OPTIMIZE__) +# define OP_CVTEPI16_EPI32_M64(x) \ + (_mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i *)(x)))) +# else +# define OP_CVTEPI16_EPI32_M64(x) \ + (_mm_cvtepi16_epi32(*(__m128i *)(x))) +# endif + +#endif |