ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-21 21:39:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

add back some code

Browse Source 
[ROCm/composable_kernel commit: 40836ab926]
This commit is contained in:
Chao Liu
2019-08-13 12:21:38 -05:00
parent 18bc57cd93
commit 15a2fc9029
4 changed files with 54 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
driver/include/device_convolution_implicit_gemm_v1_chwn_cyxk_khwn.hpp
View File

@@ -143,7 +143,7 @@ void device_convolution_implicit_gemm_v1_chwn_cyxk_khwn(InDesc,
constexpr index_t WeiBlockCopyDataPerAccess_K = 4;
constexpr index_t OutThreadCopyDataPerAccess_N = 2;
#elif 1
#elif 0
// for 3x3, 34x34, v1r3, Pascal
// for 3x3, 28x28, v1r3, Pascal
// for 3x3, 14x14, v1r3, Pascal
@@ -266,9 +266,12 @@ void device_convolution_implicit_gemm_v1_chwn_cyxk_khwn(InDesc,
constexpr index_t GemmDataPerReadA = 4;
constexpr index_t GemmDataPerReadB = 4;
using InBlockCopySubLengths_CHWN = Sequence<1, 1, 1, 4>;
using InBlockCopyClusterLengths_CHWN = Sequence<8, 2, 4, 4>;
constexpr index_t InBlockCopyDataPerAccess_N = 4;
using WeiBlockCopySubLengths_CK = Sequence<1, 4>;
using WeiBlockCopyClusterLengths_CK = Sequence<8, 32>;
constexpr index_t WeiBlockCopyDataPerAccess_K = 4;
constexpr index_t OutThreadCopyDataPerAccess_N = 4;