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

Add multithreading threshold

This commit is contained in:
Martin Kroeker 2023-03-26 00:25:28 +01:00 committed by GitHub
parent a495ffc554
commit f2d6b1c70e
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 18 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
interface/symm.c
View File

@ -44,6 +44,7 @@
#endif #endif
#ifndef COMPLEX #ifndef COMPLEX
#define SMP_THRESHOLD_MIN 65536.
#ifdef XDOUBLE #ifdef XDOUBLE
#define ERROR_NAME "QSYMM " #define ERROR_NAME "QSYMM "
#elif defined(DOUBLE) #elif defined(DOUBLE)
@ -52,6 +53,7 @@
#define ERROR_NAME "SSYMM " #define ERROR_NAME "SSYMM "
#endif #endif
#else #else
#define SMP_THRESHOLD_MIN 8192.
#ifndef GEMM3M #ifndef GEMM3M
#ifndef HEMM #ifndef HEMM
#ifdef XDOUBLE #ifdef XDOUBLE
@ -91,6 +93,10 @@
#endif #endif
#endif #endif
#ifndef GEMM_MULTITHREAD_THRESHOLD
#define GEMM_MULTITHREAD_THRESHOLD 4
#endif
#ifdef SMP #ifdef SMP
#ifndef COMPLEX #ifndef COMPLEX
@ -159,7 +165,9 @@ void NAME(char *SIDE, char *UPLO,
#if defined(SMP) && !defined(NO_AFFINITY) #if defined(SMP) && !defined(NO_AFFINITY)
int nodes; int nodes;
#endif #endif
# if defined(SMP)
int MN;
#endif
blasint info; blasint info;
int side; int side;
int uplo; int uplo;
@ -255,6 +263,9 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_SIDE Side, enum CBLAS_UPLO Uplo,
#if defined(SMP) && !defined(NO_AFFINITY) #if defined(SMP) && !defined(NO_AFFINITY)
int nodes; int nodes;
#endif #endif
#if defined(SMP)
int MN;
#endif
PRINT_DEBUG_CNAME; PRINT_DEBUG_CNAME;
@ -375,15 +386,18 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_SIDE Side, enum CBLAS_UPLO Uplo,
#ifdef SMP #ifdef SMP
args.common = NULL; args.common = NULL;
MN = 2.* (double) args.m * (double)args.m * (double) args.n;
if (MN <= (SMP_THRESHOLD_MIN * (double) GEMM_MULTITHREAD_THRESHOLD) ) {
args.nthreads = 1;
} else {
args.nthreads = num_cpu_avail(3); args.nthreads = num_cpu_avail(3);
}
if (args.nthreads == 1) { if (args.nthreads == 1) {
#endif #endif
(symm[(side << 1) | uplo ])(&args, NULL, NULL, sa, sb, 0); (symm[(side << 1) | uplo ])(&args, NULL, NULL, sa, sb, 0);
#ifdef SMP #ifdef SMP
} else { } else {
#ifndef NO_AFFINITY #ifndef NO_AFFINITY