ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-17 11:30:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

refactor

Browse Source 
[ROCm/composable_kernel commit: ff7a62198d]
This commit is contained in:
Chao Liu
2019-01-16 11:58:12 -06:00
parent 07817cbaae
commit 77fddd04bd
3 changed files with 12 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
driver/conv.cu
View File

@@ -336,14 +336,6 @@ void check_error(const Tensor<T>& ref, const Tensor<T>& result)
int main()
{
#if 0
constexpr unsigned N = 1;
constexpr unsigned C = 1;
constexpr unsigned HI = 4;
constexpr unsigned WI = 4;
constexpr unsigned K = 1;
constexpr unsigned S = 3;
constexpr unsigned R = 3;
#elif 0
constexpr unsigned N = 1;
constexpr unsigned C = 1;
constexpr unsigned HI = 34;
@@ -352,13 +344,13 @@ int main()
constexpr unsigned S = 3;
constexpr unsigned R = 3;
#elif 1
constexpr unsigned N = 64;
constexpr unsigned C = 256;
constexpr unsigned N = 64;
constexpr unsigned C = 256;
constexpr unsigned HI = 34;
constexpr unsigned WI = 34;
constexpr unsigned K = 64;
constexpr unsigned S = 3;
constexpr unsigned R = 3;
constexpr unsigned K = 64;
constexpr unsigned S = 3;
constexpr unsigned R = 3;
#elif 0
constexpr unsigned N = 64;
constexpr unsigned C = 64;
@@ -369,12 +361,12 @@ int main()
constexpr unsigned R = 3;
#elif 0
constexpr unsigned N = 64;
constexpr unsigned C = 64;
constexpr unsigned HI = 66;
constexpr unsigned WI = 66;
constexpr unsigned C = 256;
constexpr unsigned HI = 36;
constexpr unsigned WI = 36;
constexpr unsigned K = 64;
constexpr unsigned S = 3;
constexpr unsigned R = 3;
constexpr unsigned S = 5;
constexpr unsigned R = 5;
#endif
auto in_nchw_desc = make_ConstantTensorDescriptor(Sequence<N, C, HI, WI>{});

4
driver/device_implicit_gemm_convolution.cuh
View File

@@ -52,7 +52,7 @@ void device_implicit_gemm_convolution(
constexpr unsigned WoPerThread = 2;
constexpr unsigned BlockSize = 128;
#elif 1
#elif 0
constexpr unsigned NPerBlock = 2;
constexpr unsigned KPerBlock = 64;
constexpr unsigned CPerBlock = 4;
@@ -60,7 +60,7 @@ void device_implicit_gemm_convolution(
constexpr unsigned WoPerBlock = 32;
constexpr unsigned KPerThread = 4;
constexpr unsigned CPerThread = 1;
constexpr unsigned CPerThread = 2;
constexpr unsigned HoPerThread = 2;
constexpr unsigned WoPerThread = 2;