Skip to content

Commit aa972ec

Browse files
chanching the heuristic to detect the small matrices
1 parent 2ad3664 commit aa972ec

File tree

2 files changed

+2
-2
lines changed

2 files changed

+2
-2
lines changed

src/library/blas/functor/hawaii.cc

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -101,7 +101,7 @@ clblasSgemmFunctor * FunctorSelectorHawaii::select_sgemm_specific(clblasSgemmFun
101101
//TODO: the logic below is complicated; Needs cleanup;
102102
clblasSgemmFunctor * functor;
103103
bool Not_TT = ((args.transA==clblasNoTrans && args.transB==clblasTrans ) || ( args.transA==clblasNoTrans && args.transB==clblasNoTrans ) || ( args.transA==clblasTrans && args.transB==clblasNoTrans ));
104-
bool SmallMatrices = args.M/6*args.N/6<180*180 || ((args.M%64!=0 && args.N%64!=0 && args.M<1900 &&args.N<1900 ) && (args.M%96!=0 && args.N%96!=0 && args.M<1900 &&args.N<1900 ));
104+
bool SmallMatrices = args.M/6*args.N/6<100*100 || (args.M/6*args.N/6<180*180 && (args.M%32!=0||args.N%32!=0) || ((args.M%64!=0 && args.N%64!=0 && args.M<1900 &&args.N<1900 ) && (args.M%96!=0 && args.N%96!=0 && args.M<1900 &&args.N<1900 ));
105105
bool SmallMatricesMod32= (SmallMatrices && (args.M%32==0&&args.N%32==0)) ;
106106
SmallMatricesMod32 = SmallMatricesMod32&&Not_TT&&args.K % 16 == 0;
107107
//SmallMatrices= false;

src/library/blas/functor/hawaii_sgemmSplitKernel.cc

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
#if !defined CLBLAS_HAWAII_DYNAMIC_KERNEL && !defined CLBLAS_BONAIRE_DYNAMIC_KERNEL
1+
#if !defined CLBLAS_HAWAII_DYNAMIC_KERNEL || !defined CLBLAS_BONAIRE_DYNAMIC_KERNEL
22

33
#include <stdio.h>
44
#include <string.h>

0 commit comments

Comments
 (0)