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

Merge pull request #3843 from Mousius/switch-ratio 

Propagate SWITCH_RATIO to DYNAMIC_ARCH builds
This commit is contained in:
Martin Kroeker 2023-04-19 11:51:54 +02:00 committed by GitHub
parent c628030669 32f2fafde7
commit 437c0bf2b4
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
7 changed files with 46 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
common_param.h
View File

@ -1,5 +1,6 @@
/*********************************************************************/ /*********************************************************************/
/* Copyright 2009, 2010 The University of Texas at Austin. */ /* Copyright 2009, 2010 The University of Texas at Austin. */
/* Copyright 2023 The OpenBLAS Project. */
/* All rights reserved. */ /* All rights reserved. */
/* */ /* */
/* Redistribution and use in source and binary forms, with or */ /* Redistribution and use in source and binary forms, with or */
@ -45,6 +46,7 @@
typedef struct { typedef struct {
int dtb_entries; int dtb_entries;
int switch_ratio;
int offsetA, offsetB, align; int offsetA, offsetB, align;
#if BUILD_BFLOAT16 == 1 #if BUILD_BFLOAT16 == 1

15
driver/level3/level3_gemm3m_thread.c
View File

@ -1,5 +1,6 @@
/*********************************************************************/ /*********************************************************************/
/* Copyright 2009, 2010 The University of Texas at Austin. */ /* Copyright 2009, 2010 The University of Texas at Austin. */
/* Copyright 2023 The OpenBLAS Project. */
/* All rights reserved. */ /* All rights reserved. */
/* */ /* */
/* Redistribution and use in source and binary forms, with or */ /* Redistribution and use in source and binary forms, with or */
@ -44,10 +45,6 @@
#define DIVIDE_RATE 2 #define DIVIDE_RATE 2
#endif #endif
#ifndef SWITCH_RATIO
#define SWITCH_RATIO 2
#endif
//The array of job_t may overflow the stack. //The array of job_t may overflow the stack.
//Instead, use malloc to alloc job_t. //Instead, use malloc to alloc job_t.
#if MAX_CPU_NUMBER > BLAS3_MEM_ALLOC_THRESHOLD #if MAX_CPU_NUMBER > BLAS3_MEM_ALLOC_THRESHOLD
@ -1015,6 +1012,12 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
BLASLONG divN, divT; BLASLONG divN, divT;
int mode; int mode;
#if defined(DYNAMIC_ARCH)
int switch_ratio = gotoblas->switch_ratio;
#else
int switch_ratio = SWITCH_RATIO;
#endif
if (range_m) { if (range_m) {
BLASLONG m_from = *(((BLASLONG *)range_m) + 0); BLASLONG m_from = *(((BLASLONG *)range_m) + 0);
BLASLONG m_to = *(((BLASLONG *)range_m) + 1); BLASLONG m_to = *(((BLASLONG *)range_m) + 1);
@ -1030,7 +1033,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
} }
*/ */
if ((args -> m < nthreads * SWITCH_RATIO) || (args -> n < nthreads * SWITCH_RATIO)) { if ((args -> m < nthreads * switch_ratio) || (args -> n < nthreads * switch_ratio)) {
GEMM3M_LOCAL(args, range_m, range_n, sa, sb, 0); GEMM3M_LOCAL(args, range_m, range_n, sa, sb, 0);
return 0; return 0;
} }
@ -1038,7 +1041,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
divT = nthreads; divT = nthreads;
divN = 1; divN = 1;
while ((GEMM3M_P * divT > m * SWITCH_RATIO) && (divT > 1)) { while ((GEMM3M_P * divT > m * switch_ratio) && (divT > 1)) {
do { do {
divT --; divT --;
divN = 1; divN = 1;

13
driver/level3/level3_syrk_threaded.c
View File

@ -1,5 +1,6 @@
/*********************************************************************/ /*********************************************************************/
/* Copyright 2009, 2010 The University of Texas at Austin. */ /* Copyright 2009, 2010 The University of Texas at Austin. */
/* Copyright 2023 The OpenBLAS Project. */
/* All rights reserved. */ /* All rights reserved. */
/* */ /* */
/* Redistribution and use in source and binary forms, with or */ /* Redistribution and use in source and binary forms, with or */
@ -44,10 +45,6 @@
#define DIVIDE_RATE 2 #define DIVIDE_RATE 2
#endif #endif
#ifndef SWITCH_RATIO
#define SWITCH_RATIO 2
#endif
//The array of job_t may overflow the stack. //The array of job_t may overflow the stack.
//Instead, use malloc to alloc job_t. //Instead, use malloc to alloc job_t.
#if MAX_CPU_NUMBER > BLAS3_MEM_ALLOC_THRESHOLD #if MAX_CPU_NUMBER > BLAS3_MEM_ALLOC_THRESHOLD
@ -528,7 +525,13 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
int mode, mask; int mode, mask;
double dnum, di, dinum; double dnum, di, dinum;
if ((nthreads == 1) || (args -> n < nthreads * SWITCH_RATIO)) { #if defined(DYNAMIC_ARCH)
int switch_ratio = gotoblas->switch_ratio;
#else
int switch_ratio = SWITCH_RATIO;
#endif
if ((nthreads == 1) || (args->n < nthreads * switch_ratio)) {
SYRK_LOCAL(args, range_m, range_n, sa, sb, 0); SYRK_LOCAL(args, range_m, range_n, sa, sb, 0);
return 0; return 0;
} }

31
driver/level3/level3_thread.c
View File

@ -1,5 +1,6 @@
/*********************************************************************/ /*********************************************************************/
/* Copyright 2009, 2010 The University of Texas at Austin. */ /* Copyright 2009, 2010 The University of Texas at Austin. */
/* Copyright 2023 The OpenBLAS Project. */
/* All rights reserved. */ /* All rights reserved. */
/* */ /* */
/* Redistribution and use in source and binary forms, with or */ /* Redistribution and use in source and binary forms, with or */
@ -44,10 +45,6 @@
#define DIVIDE_RATE 2 #define DIVIDE_RATE 2
#endif #endif
#ifndef SWITCH_RATIO
#define SWITCH_RATIO 2
#endif
#ifndef GEMM_PREFERED_SIZE #ifndef GEMM_PREFERED_SIZE
#define GEMM_PREFERED_SIZE 1 #define GEMM_PREFERED_SIZE 1
#endif #endif
@ -577,6 +574,11 @@ InitializeCriticalSection((PCRITICAL_SECTION)&level3_lock);
BLASLONG width, i, j, k, js; BLASLONG width, i, j, k, js;
BLASLONG m, n, n_from, n_to; BLASLONG m, n, n_from, n_to;
int mode; int mode;
#if defined(DYNAMIC_ARCH)
int switch_ratio = gotoblas->switch_ratio;
#else
int switch_ratio = SWITCH_RATIO;
#endif
/* Get execution mode */ /* Get execution mode */
#ifndef COMPLEX #ifndef COMPLEX
@ -698,8 +700,8 @@ EnterCriticalSection((PCRITICAL_SECTION)&level3_lock);
num_parts = 0; num_parts = 0;
while (n > 0){ while (n > 0){
width = blas_quickdivide(n + nthreads - num_parts - 1, nthreads - num_parts); width = blas_quickdivide(n + nthreads - num_parts - 1, nthreads - num_parts);
if (width < SWITCH_RATIO) { if (width < switch_ratio) {
width = SWITCH_RATIO; width = switch_ratio;
} }
width = round_up(n, width, GEMM_PREFERED_SIZE); width = round_up(n, width, GEMM_PREFERED_SIZE);
@ -746,6 +748,11 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, IFLOAT *sa, IF
BLASLONG m = args -> m; BLASLONG m = args -> m;
BLASLONG n = args -> n; BLASLONG n = args -> n;
BLASLONG nthreads_m, nthreads_n; BLASLONG nthreads_m, nthreads_n;
#if defined(DYNAMIC_ARCH)
int switch_ratio = gotoblas->switch_ratio;
#else
int switch_ratio = SWITCH_RATIO;
#endif
/* Get dimensions from index ranges if available */ /* Get dimensions from index ranges if available */
if (range_m) { if (range_m) {
@ -755,21 +762,21 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, IFLOAT *sa, IF
n = range_n[1] - range_n[0]; n = range_n[1] - range_n[0];
} }
/* Partitions in m should have at least SWITCH_RATIO rows */ /* Partitions in m should have at least switch_ratio rows */
if (m < 2 * SWITCH_RATIO) { if (m < 2 * switch_ratio) {
nthreads_m = 1; nthreads_m = 1;
} else { } else {
nthreads_m = args -> nthreads; nthreads_m = args -> nthreads;
while (m < nthreads_m * SWITCH_RATIO) { while (m < nthreads_m * switch_ratio) {
nthreads_m = nthreads_m / 2; nthreads_m = nthreads_m / 2;
} }
} }
/* Partitions in n should have at most SWITCH_RATIO * nthreads_m columns */ /* Partitions in n should have at most switch_ratio * nthreads_m columns */
if (n < SWITCH_RATIO * nthreads_m) { if (n < switch_ratio * nthreads_m) {
nthreads_n = 1; nthreads_n = 1;
} else { } else {
nthreads_n = (n + SWITCH_RATIO * nthreads_m - 1) / (SWITCH_RATIO * nthreads_m); nthreads_n = (n + switch_ratio * nthreads_m - 1) / (switch_ratio * nthreads_m);
if (nthreads_m * nthreads_n > args -> nthreads) { if (nthreads_m * nthreads_n > args -> nthreads) {
nthreads_n = blas_quickdivide(args -> nthreads, nthreads_m); nthreads_n = blas_quickdivide(args -> nthreads, nthreads_m);
} }

5
kernel/setparam-ref.c
View File

@ -1,5 +1,6 @@
/*********************************************************************/ /*********************************************************************/
/* Copyright 2009, 2010 The University of Texas at Austin. */ /* Copyright 2009, 2010 The University of Texas at Austin. */
/* Copyright 2023 The OpenBLAS Project. */
/* All rights reserved. */ /* All rights reserved. */
/* */ /* */
/* Redistribution and use in source and binary forms, with or */ /* Redistribution and use in source and binary forms, with or */
@ -49,7 +50,9 @@
static void init_parameter(void); static void init_parameter(void);
gotoblas_t TABLE_NAME = { gotoblas_t TABLE_NAME = {
DTB_DEFAULT_ENTRIES , DTB_DEFAULT_ENTRIES,
SWITCH_RATIO,
GEMM_DEFAULT_OFFSET_A, GEMM_DEFAULT_OFFSET_B, GEMM_DEFAULT_ALIGN, GEMM_DEFAULT_OFFSET_A, GEMM_DEFAULT_OFFSET_B, GEMM_DEFAULT_ALIGN,

4
lapack/potrf/potrf_parallel.c
View File

@ -80,10 +80,6 @@ static FLOAT dm1 = -1.;
#define DIVIDE_RATE 2 #define DIVIDE_RATE 2
#endif #endif
#ifndef SWITCH_RATIO
#define SWITCH_RATIO 2
#endif
#ifndef LOWER #ifndef LOWER
#define TRANS #define TRANS
#endif #endif

4
param.h
View File

@ -3854,6 +3854,10 @@ Until then, just keep it different than DGEMM_DEFAULT_UNROLL_N to keep copy rout
#endif #endif
#ifndef SWITCH_RATIO
#define SWITCH_RATIO 2
#endif
#ifndef QGEMM_DEFAULT_UNROLL_M #ifndef QGEMM_DEFAULT_UNROLL_M
#define QGEMM_DEFAULT_UNROLL_M 2 #define QGEMM_DEFAULT_UNROLL_M 2
#endif #endif