ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-17 03:19:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

add

Browse Source 
[ROCm/composable_kernel commit: 6a3f3f951d]
This commit is contained in:
Jing Zhang
2019-04-03 17:08:14 -05:00
parent 1c2ae33d1d
commit 7bea559923
4 changed files with 17 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
driver/device_implicit_gemm_convolution_2_chwn_cyxk_khwn.hpp
View File

@@ -224,7 +224,7 @@ void device_implicit_gemm_convolution_2_chwn_cyxk_khwn(InDesc,
// 1x1, 14x14, Vega 20, hack CPerBlock = 1
constexpr index_t BPerBlock = 64;
constexpr index_t KPerBlock = 128;
constexpr index_t CPerBlock = 1;
constexpr index_t CPerBlock = 8;
constexpr index_t BPerThread = 8;
constexpr index_t KPerThread = 8;
@@ -232,7 +232,7 @@ void device_implicit_gemm_convolution_2_chwn_cyxk_khwn(InDesc,
constexpr index_t GemmMPerThreadSubC = 4;
constexpr index_t GemmNPerThreadSubC = 4;
constexpr index_t GemmMLevel0Cluster = 4;
constexpr index_t GemmNLevel0Cluster = 2;
constexpr index_t GemmNLevel0Cluster = 4;
constexpr index_t GemmMLevel1Cluster = 4;
constexpr index_t GemmNLevel1Cluster = 4;
constexpr index_t GemmKPerThreadLoop = 1;
@@ -249,7 +249,7 @@ void device_implicit_gemm_convolution_2_chwn_cyxk_khwn(InDesc,
constexpr index_t InBlockCopyDataPerRead = 4;
constexpr index_t WeiBlockCopyDataPerRead = 4;
constexpr index_t BlockSize = 128;
constexpr index_t BlockSize = 256;
#endif
constexpr index_t GridSize =

2
driver/driver.hip.cpp
View File

@@ -580,7 +580,7 @@ int main(int argc, char* argv[])
constexpr index_t HPad = 0;
constexpr index_t WPad = 0;
#elif 0
#elif 1
// 1x1 filter, 14x14 image, C = 2048
constexpr index_t N = 128;
constexpr index_t C = 2048;