ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-25 07:14:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

removing old implementation of tensor descriptor

Browse Source 
[ROCm/composable_kernel commit: 39d92e7dfd]
This commit is contained in:
Chao Liu
2019-09-25 22:24:06 -05:00
parent 66f7a03ee4
commit 313eb881cd
11 changed files with 128 additions and 64 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
driver/include/device_convolution_implicit_gemm_v4r1_nchw_kcyx_nkhw_padded.hpp
View File

@@ -33,18 +33,12 @@ void device_convolution_implicit_gemm_v4r1_nchw_kcyx_nkhw_padded(InDesc,
constexpr auto I2 = Number<2>{};
constexpr auto I3 = Number<3>{};
#if 1
constexpr auto in_nchw_desc = InDesc{};
constexpr auto wei_kcyx_desc = WeiDesc{};
constexpr auto out_nkhw_desc = OutDesc{};
#else
constexpr auto in_nchw_desc =
make_native_tensor_descriptor(InDesc::GetLengths(), InDesc::GetStrides());
constexpr auto wei_kcyx_desc =
make_native_tensor_descriptor(WeiDesc::GetLengths(), WeiDesc::GetStrides());
constexpr auto out_nkhw_desc =
make_native_tensor_descriptor(OutDesc::GetLegnths(), OutDesc::GetStrides());
#endif
make_native_tensor_descriptor(OutDesc::GetLengths(), OutDesc::GetStrides());
constexpr index_t N = out_nkhw_desc.GetLength(I0);
constexpr index_t K = out_nkhw_desc.GetLength(I1);

4
driver/src/driver.cpp
View File

@@ -295,7 +295,7 @@ int main(int argc, char* argv[])
using LeftPads = Sequence<0, 0>;
using RightPads = Sequence<0, 0>;
#elif 1
#elif 0
// 3x3 filter, 2x2 stride, 35x35 input, 17x17 output
// cudnn@V100 90%, ck@V100 93%, ck@P100 83%, ck@VII 81%
constexpr index_t N = 128;
@@ -341,7 +341,7 @@ int main(int argc, char* argv[])
using LeftPads = Sequence<3, 0>;
using RightPads = Sequence<3, 0>;
#elif 0
#elif 1
// 1x7 filter, 0x3 pad, 17x17 input
constexpr index_t N = 128;
constexpr index_t C = 128;