mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-01-19 06:53:50 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #10754 from dkurt:dnn_ocl_gemv_min_globalsize

Browse Source
This commit is contained in:
Alexander Alekhin 2018-02-01 12:39:27 +00:00
commit 9d25bd583f
1 changed files with 19 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
modules/dnn/src/ocl4dnn/src/math_functions.cpp
View File

@ -451,23 +451,27 @@ bool ocl4dnnGEMV<float>(const CBLAS_TRANSPOSE TransA,
uint row_size = M; uint row_size = M;
uint col_size = N; uint col_size = N;
size_t localsize[] = { 128 };
size_t globalsize[] = { row_size / 4 * localsize[0] };
uint argId = 0; if (row_size >= 4)
k.set(argId++, ocl::KernelArg::PtrReadOnly(A)); {
k.set(argId++, offA); size_t localsize[] = { 128 };
k.set(argId++, cl_uint(col_size)); size_t globalsize[] = { row_size / 4 * localsize[0] };
k.set(argId++, cl_uint(col_size%4));
k.set(argId++, ocl::KernelArg::PtrReadOnly(x));
k.set(argId++, offx);
k.set(argId++, alpha);
k.set(argId++, beta);
k.set(argId++, ocl::KernelArg::PtrWriteOnly(y));
k.set(argId++, offy);
k.set(argId++, NULL, localsize[0] * sizeof(cl_float4));
ret = k.run(1, globalsize, localsize, false); uint argId = 0;
k.set(argId++, ocl::KernelArg::PtrReadOnly(A));
k.set(argId++, offA);
k.set(argId++, cl_uint(col_size));
k.set(argId++, cl_uint(col_size%4));
k.set(argId++, ocl::KernelArg::PtrReadOnly(x));
k.set(argId++, offx);
k.set(argId++, alpha);
k.set(argId++, beta);
k.set(argId++, ocl::KernelArg::PtrWriteOnly(y));
k.set(argId++, offy);
k.set(argId++, NULL, localsize[0] * sizeof(cl_float4));
ret = k.run(1, globalsize, localsize, false);
}
if ((row_size % 4) != 0 && ret) if ((row_size % 4) != 0 && ret)
{ {