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

Rework multithreading threshold

This commit is contained in:
Martin Kroeker 2023-03-26 00:23:57 +01:00 committed by GitHub
parent 1659750861
commit a495ffc554
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 14 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
interface/syrk.c
View File

@ -44,6 +44,7 @@
#endif #endif
#ifndef COMPLEX #ifndef COMPLEX
#define SMP_THRESHOLD_MIN 109944.
#ifdef XDOUBLE #ifdef XDOUBLE
#define ERROR_NAME "QSYRK " #define ERROR_NAME "QSYRK "
#elif defined(DOUBLE) #elif defined(DOUBLE)
@ -52,6 +53,7 @@
#define ERROR_NAME "SSYRK " #define ERROR_NAME "SSYRK "
#endif #endif
#else #else
#define SMP_THRESHOLD_MIN 14824.
#ifndef HEMM #ifndef HEMM
#ifdef XDOUBLE #ifdef XDOUBLE
#define ERROR_NAME "XSYRK " #define ERROR_NAME "XSYRK "
@ -71,6 +73,10 @@
#endif #endif
#endif #endif
#ifndef GEMM_MULTITHREAD_THRESHOLD
#define GEMM_MULTITHREAD_THRESHOLD 4
#endif
static int (*syrk[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) = { static int (*syrk[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) = {
#ifndef HEMM #ifndef HEMM
SYRK_UN, SYRK_UC, SYRK_LN, SYRK_LC, SYRK_UN, SYRK_UC, SYRK_LN, SYRK_LC,
@ -101,6 +107,7 @@ void NAME(char *UPLO, char *TRANS,
FLOAT *sa, *sb; FLOAT *sa, *sb;
#ifdef SMP #ifdef SMP
int NNK;
#ifdef USE_SIMPLE_THREADED_LEVEL3 #ifdef USE_SIMPLE_THREADED_LEVEL3
#ifndef COMPLEX #ifndef COMPLEX
#ifdef XDOUBLE #ifdef XDOUBLE
@ -225,6 +232,8 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, enum CBLAS_TRANSPOSE Tr
FLOAT *sa, *sb; FLOAT *sa, *sb;
#ifdef SMP #ifdef SMP
int NNK;
#ifdef USE_SIMPLE_THREADED_LEVEL3 #ifdef USE_SIMPLE_THREADED_LEVEL3
#ifndef COMPLEX #ifndef COMPLEX
#ifdef XDOUBLE #ifdef XDOUBLE
@ -354,18 +363,13 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, enum CBLAS_TRANSPOSE Tr
#endif #endif
args.common = NULL; args.common = NULL;
#ifndef COMPLEX
#ifdef DOUBLE NNK = (double)(args.n+1)*(double)args.n*(double)args.k;
if (args.n < 100) if (NNK <= (SMP_THRESHOLD_MIN * GEMM_MULTITHREAD_THRESHOLD)) {
#else
if (args.n < 200)
#endif
#else
if (args.n < 65)
#endif
args.nthreads = 1; args.nthreads = 1;
else } else {
args.nthreads = num_cpu_avail(3); args.nthreads = num_cpu_avail(3);
}
if (args.nthreads == 1) { if (args.nthreads == 1) {
#endif #endif
@ -373,7 +377,6 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, enum CBLAS_TRANSPOSE Tr
(syrk[(uplo << 1) | trans ])(&args, NULL, NULL, sa, sb, 0); (syrk[(uplo << 1) | trans ])(&args, NULL, NULL, sa, sb, 0);
#ifdef SMP #ifdef SMP
} else { } else {
#ifndef USE_SIMPLE_THREADED_LEVEL3 #ifndef USE_SIMPLE_THREADED_LEVEL3