ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-11 08:50:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

adding implicit GEMM v4r2

Browse Source
This commit is contained in:
Chao Liu
2019-07-13 22:10:42 -05:00
parent e87aa851ea
commit a4b524615b
1 changed files with 7 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
driver/include/device_convolution_implicit_gemm_v4r2_nchw_kcyx_nkhw.hpp
View File

@@ -55,13 +55,13 @@ void device_convolution_implicit_gemm_v4r2_nchw_kcyx_nkhw(InDesc,
#if 1
// 1x1 filter, 8x8 image
constexpr index_t N1 = 2;
constexpr index_t N0 = 1;
constexpr index_t N2 = 1;
constexpr index_t Ho1 = 8;
constexpr index_t Ho0 = 1;
constexpr index_t Ho2 = 1;
constexpr index_t Wo1 = 1;
constexpr index_t Wo0 = 2;
constexpr index_t Wo2 = 4;
constexpr index_t BlockSize = 256;
@@ -105,6 +105,10 @@ void device_convolution_implicit_gemm_v4r2_nchw_kcyx_nkhw(InDesc,
constexpr index_t WeiBlockCopyDstDataPerWrite_K = 1;
#endif
constexpr index_t N1 = N / (N0 * N2);
constexpr index_t Ho1 = Ho / (Ho0 * Ho2);
constexpr index_t Wo1 = Wo / (Wo0 * Wo2);
constexpr index_t B = N1 * Ho1 * Wo1;
constexpr index_t GridSize =