Browse Source

Merge pull request #3960 from martin-frbg/symmsyrk_sp

Add multithreading threshold for SYMM and rework the one for SYRK
tags/v0.3.22^2
Martin Kroeker GitHub 2 years ago
parent
commit
2d39e715e2
No known key found for this signature in database GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 32 additions and 15 deletions
  1. +18
    -4
      interface/symm.c
  2. +14
    -11
      interface/syrk.c

+ 18
- 4
interface/symm.c View File

@@ -44,6 +44,7 @@
#endif

#ifndef COMPLEX
#define SMP_THRESHOLD_MIN 65536.
#ifdef XDOUBLE
#define ERROR_NAME "QSYMM "
#elif defined(DOUBLE)
@@ -52,6 +53,7 @@
#define ERROR_NAME "SSYMM "
#endif
#else
#define SMP_THRESHOLD_MIN 8192.
#ifndef GEMM3M
#ifndef HEMM
#ifdef XDOUBLE
@@ -91,6 +93,10 @@
#endif
#endif

#ifndef GEMM_MULTITHREAD_THRESHOLD
#define GEMM_MULTITHREAD_THRESHOLD 4
#endif


#ifdef SMP
#ifndef COMPLEX
@@ -159,7 +165,9 @@ void NAME(char *SIDE, char *UPLO,
#if defined(SMP) && !defined(NO_AFFINITY)
int nodes;
#endif

# if defined(SMP)
int MN;
#endif
blasint info;
int side;
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)
int nodes;
#endif
#if defined(SMP)
int MN;
#endif

PRINT_DEBUG_CNAME;

@@ -375,15 +386,18 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_SIDE Side, enum CBLAS_UPLO Uplo,

#ifdef SMP
args.common = NULL;
args.nthreads = num_cpu_avail(3);

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);
}
if (args.nthreads == 1) {
#endif

(symm[(side << 1) | uplo ])(&args, NULL, NULL, sa, sb, 0);

#ifdef SMP

} else {

#ifndef NO_AFFINITY


+ 14
- 11
interface/syrk.c View File

@@ -44,6 +44,7 @@
#endif

#ifndef COMPLEX
#define SMP_THRESHOLD_MIN 109944.
#ifdef XDOUBLE
#define ERROR_NAME "QSYRK "
#elif defined(DOUBLE)
@@ -52,6 +53,7 @@
#define ERROR_NAME "SSYRK "
#endif
#else
#define SMP_THRESHOLD_MIN 14824.
#ifndef HEMM
#ifdef XDOUBLE
#define ERROR_NAME "XSYRK "
@@ -71,6 +73,10 @@
#endif
#endif

#ifndef GEMM_MULTITHREAD_THRESHOLD
#define GEMM_MULTITHREAD_THRESHOLD 4
#endif

static int (*syrk[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) = {
#ifndef HEMM
SYRK_UN, SYRK_UC, SYRK_LN, SYRK_LC,
@@ -101,6 +107,7 @@ void NAME(char *UPLO, char *TRANS,
FLOAT *sa, *sb;

#ifdef SMP
int NNK;
#ifdef USE_SIMPLE_THREADED_LEVEL3
#ifndef COMPLEX
#ifdef XDOUBLE
@@ -225,6 +232,8 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, enum CBLAS_TRANSPOSE Tr
FLOAT *sa, *sb;

#ifdef SMP
int NNK;

#ifdef USE_SIMPLE_THREADED_LEVEL3
#ifndef COMPLEX
#ifdef XDOUBLE
@@ -354,18 +363,13 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, enum CBLAS_TRANSPOSE Tr
#endif

args.common = NULL;
#ifndef COMPLEX
#ifdef DOUBLE
if (args.n < 100)
#else
if (args.n < 200)
#endif
#else
if (args.n < 65)
#endif

NNK = (double)(args.n+1)*(double)args.n*(double)args.k;
if (NNK <= (SMP_THRESHOLD_MIN * GEMM_MULTITHREAD_THRESHOLD)) {
args.nthreads = 1;
else
} else {
args.nthreads = num_cpu_avail(3);
}

if (args.nthreads == 1) {
#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);

#ifdef SMP

} else {

#ifndef USE_SIMPLE_THREADED_LEVEL3


Loading…
Cancel
Save