floraachy/OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #1971 from martin-frbg/trsm-threshold

Browse Source 
Shift transition to multithreading towards larger matrix sizes
This commit is contained in:
Martin Kroeker
2019-01-24 09:17:48 +01:00
committed by GitHub
parent 16494cb7c4 8533aca964
commit 3f7bb87a2a
1 changed files with 12 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
interface/trsm.c
View File

@@ -81,6 +81,12 @@
#endif
#endif
#ifndef COMPLEX
#define SMP_FACTOR 256
#else
#define SMP_FACTOR 128
#endif
static int (*trsm[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) = {
#ifndef TRMM
TRSM_LNUU, TRSM_LNUN, TRSM_LNLU, TRSM_LNLN,
@@ -366,11 +372,15 @@ void CNAME(enum CBLAS_ORDER order,
mode |= (trans << BLAS_TRANSA_SHIFT);
mode |= (side << BLAS_RSIDE_SHIFT);
if ( args.m < 2*GEMM_MULTITHREAD_THRESHOLD )
/*
if ( args.m < 2 * GEMM_MULTITHREAD_THRESHOLD )
args.nthreads = 1;
else
if ( args.n < 2*GEMM_MULTITHREAD_THRESHOLD )
if ( args.n < 2 * GEMM_MULTITHREAD_THRESHOLD )
args.nthreads = 1;
*/
if ( args.m * args.n < SMP_FACTOR * GEMM_MULTITHREAD_THRESHOLD)
args.nthreads = 1;
else
args.nthreads = num_cpu_avail(3);