generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Set minimum matrix size for AmdBlas::gemm to 20 since it works incorrect for small sizes

Browse Source
This commit is contained in:
Alexander Karsakov
2014-08-12 18:02:29 +04:00
parent b4d3b34acf
commit 713ddb89bd
2 changed files with 6 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
modules/core/src/matmul.cpp
View File

@@ -781,8 +781,9 @@ void cv::gemm( InputArray matA, InputArray matB, double alpha,
InputArray matC, double beta, OutputArray _matD, int flags )
{
#ifdef HAVE_CLAMDBLAS
CV_OCL_RUN(ocl::haveAmdBlas() && matA.dims() <= 2 && matB.dims() <= 2 && matC.dims() <= 2 && _matD.isUMat(),
ocl_gemm(matA, matB, alpha, matC, beta, _matD, flags))
CV_OCL_RUN(ocl::haveAmdBlas() && matA.dims() <= 2 && matB.dims() <= 2 && matC.dims() <= 2 && _matD.isUMat() &&
matA.cols() > 20 && matA.rows() > 20 && matB.cols() > 20, // since it works incorrect for small sizes
ocl_gemm(matA, matB, alpha, matC, beta, _matD, flags))
#endif
const int block_lin_size = 128;