Merge pull request #3843 from Mousius/switch-ratio

Propagate SWITCH_RATIO to DYNAMIC_ARCH builds

common_param.h

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

driver/level3/level3_gemm3m_thread.c

@@ -1,5 +1,6 @@
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2023 The OpenBLAS Project.                              */
 /* All rights reserved.                                              */
 /*                                                                   */
 /* Redistribution and use in source and binary forms, with or        */
@@ -44,10 +45,6 @@
 #define DIVIDE_RATE 2
 #endif
 
-#ifndef SWITCH_RATIO
-#define SWITCH_RATIO 2
-#endif
-
 //The array of job_t may overflow the stack.
 //Instead, use malloc to alloc job_t.
 #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;
   int mode;
 
+#if defined(DYNAMIC_ARCH)
+  int switch_ratio = gotoblas->switch_ratio;
+#else
+  int switch_ratio = SWITCH_RATIO;
+#endif
+
   if (range_m) {
     BLASLONG m_from = *(((BLASLONG *)range_m) + 0);
     BLASLONG m_to   = *(((BLASLONG *)range_m) + 1);
@@ -1030,15 +1033,15 @@ 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);
     return 0;
   }
 
   divT = nthreads;
   divN = 1;
 
-  while ((GEMM3M_P * divT > m * SWITCH_RATIO) && (divT > 1)) {
+  while ((GEMM3M_P * divT > m * switch_ratio) && (divT > 1)) {
     do {
       divT --;
       divN = 1;

driver/level3/level3_syrk_threaded.c

@@ -1,5 +1,6 @@
 /*********************************************************************/
 /* Copyright 2009, 2010 The University of Texas at Austin.           */
+/* Copyright 2023 The OpenBLAS Project.                              */
 /* All rights reserved.                                              */
 /*                                                                   */
 /* Redistribution and use in source and binary forms, with or        */
@@ -44,10 +45,6 @@
 #define DIVIDE_RATE 2
 #endif
 
-#ifndef SWITCH_RATIO
-#define SWITCH_RATIO 2
-#endif
-
 //The array of job_t may overflow the stack.
 //Instead, use malloc to alloc job_t.
 #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;
   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);
     return 0;
   }

driver/level3/level3_thread.c

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

kernel/setparam-ref.c

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

lapack/potrf/potrf_parallel.c

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

param.h

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