ocl4dnnGEMV in case of row_size < 4

7 years ago · 65a6674c6e
parent a75840d19c
commit 65a6674c6e
1 changed files with 21 additions and 17 deletions
--- a/modules/dnn/src/ocl4dnn/src/math_functions.cpp
+++ b/modules/dnn/src/ocl4dnn/src/math_functions.cpp
@ -451,6 +451,9 @@ bool ocl4dnnGEMV<float>(const CBLAS_TRANSPOSE TransA,

        uint row_size = M;
        uint col_size = N;
+
+        if (row_size >= 4)
+        {
            size_t localsize[] = { 128 };
            size_t globalsize[] = { row_size / 4 * localsize[0] };

@ -468,6 +471,7 @@ bool ocl4dnnGEMV<float>(const CBLAS_TRANSPOSE TransA,
            k.set(argId++, NULL, localsize[0] * sizeof(cl_float4));

            ret = k.run(1, globalsize, localsize, false);
+        }

        if ((row_size % 4) != 0 && ret)
        {