summaryrefslogtreecommitdiff
path: root/make/readme.aix
blob: 6ea582d87c8fe820b275689d95dee9cfaa92d4b3 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
General Information for BLAST Developers on IBM-AIX Systems
===========================================================

 To compile the toolkit under AIX follow the steps in readme.unx
 SPECIAL NOTE:
 To build the 64-bit version type:
 $ env AIX_MODE=64 ./ncbi/make/makedis.csh or
 $ env AIX_MODE=64 ./ncbi/make/makedis.csh 2> log 1>&2    ! Korn shell users
 % env AIX_MODE=64 ./ncbi/make/makedis.csh >& log         ! C shell users

 AIX Specific information
 ------------------------

 The following files are the main platform dependent files

 1.  makedis.csh in ./ncbi/make

 2.  ibm64_auto.ncbi.mk and ibm_auto.ncbi.mk in ./ncbi/platform

 3.  ncbilcl.ibm in ./ncbi/corelib


makedis.csh:

     This file identifies the machine model via uname -M and assigns
     the corresponding architecture.  Currently, AIX Power3 is the
     only architecture explicitly targeted.  All other architectures
     are identified as auto.  auto refers to the compiler flag that
     will check the architecture and build corresponding executables.
     The auto flag assumes that the execution environment is the same
     as the compilation environment.

ibm_xxxx.ncbi.mk:

     Currently xxxx can be either auto64 or auto.  auto64 builds the
     64-bit version.
     This file contains compiler and loader flags specific for IBM.

ncbilcl.ibm:

     Pre-processor defines / platform specific.



Multi-threading
---------------


     When running multi-threaded it may be beneficial to set the
     following environment variables:

                 AIXTHREAD_SCOPE=S
                 MALLOCMULTIHEAP=true

     Addional benefit is possibly achieved by experimenting with 
     the value of the environment variables SPINLOOPTIME and 
     YIELDLOOPTIME. The user might initially try values of 500 
     for each of these and vary them to see the effect.  These 
     variables affect the behavior of a thread at the end of a 
     parallel region. Generally, best application (though not
     necessarily system throughput) performance is achieved by 
     putting an idle thread into a busy wait rather than letting 
     it sleep.  Setting SPINLOOPTIME and YIELDLOOPTIME to a suitable 
     "large" number encourages the former behavior.