ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-13 09:45:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

refactor

Browse Source
This commit is contained in:
Chao Liu
2019-05-31 22:28:32 -05:00
parent 8d4607403e
commit 97ba755f2f
22 changed files with 168 additions and 543 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
driver/device_convolution_implicit_gemm_v1_chwn_cyxk_khwn.hpp
View File

@@ -38,7 +38,7 @@ void device_convolution_implicit_gemm_v1_chwn_cyxk_khwn(InDesc,
constexpr index_t X = wei_kcyx_desc.GetLength(I3);
// reorder weight
auto wei_cyxk_desc = make_ConstantTensorDescriptor_default_rank_packed(Sequence<C, Y, X, K>{});
auto wei_cyxk_desc = make_ConstantTensorDescriptor_packed(Sequence<C, Y, X, K>{});
ostream_ConstantTensorDescriptor(wei_cyxk_desc, std::cout << "wei_cyxk_desc: ");
Tensor<T> wei_cyxk(make_TensorDescriptor(wei_cyxk_desc));
@@ -51,7 +51,7 @@ void device_convolution_implicit_gemm_v1_chwn_cyxk_khwn(InDesc,
std::thread::hardware_concurrency());
// reorder input
auto in_chwn_desc = make_ConstantTensorDescriptor_default_rank_packed(Sequence<C, Hi, Wi, N>{});
auto in_chwn_desc = make_ConstantTensorDescriptor_packed(Sequence<C, Hi, Wi, N>{});
ostream_ConstantTensorDescriptor(in_chwn_desc, std::cout << "in_chwn_desc: ");
Tensor<T> in_chwn(make_TensorDescriptor(in_chwn_desc));
@@ -64,8 +64,7 @@ void device_convolution_implicit_gemm_v1_chwn_cyxk_khwn(InDesc,
std::thread::hardware_concurrency());
// output
auto out_khwn_desc =
make_ConstantTensorDescriptor_default_rank_packed(Sequence<K, Ho, Wo, N>{});
auto out_khwn_desc = make_ConstantTensorDescriptor_packed(Sequence<K, Ho, Wo, N>{});
ostream_ConstantTensorDescriptor(out_khwn_desc, std::cout << "out_khwn_desc: ");
Tensor<T> out_khwn(make_TensorDescriptor(out_khwn_desc));

5
driver/device_convolution_implicit_gemm_v1_nchw_cyxk_khwn.hpp
View File

@@ -37,7 +37,7 @@ void device_convolution_implicit_gemm_v1_nchw_cyxk_khwn(InDesc,
constexpr index_t X = wei_kcyx_desc.GetLength(I3);
// reorder weight
auto wei_cyxk_desc = make_ConstantTensorDescriptor_default_rank_packed(Sequence<C, Y, X, K>{});
auto wei_cyxk_desc = make_ConstantTensorDescriptor_packed(Sequence<C, Y, X, K>{});
ostream_ConstantTensorDescriptor(wei_cyxk_desc, std::cout << "wei_cyxk_desc: ");
Tensor<T> wei_cyxk(make_TensorDescriptor(wei_cyxk_desc));
@@ -50,8 +50,7 @@ void device_convolution_implicit_gemm_v1_nchw_cyxk_khwn(InDesc,
std::thread::hardware_concurrency());
// output
auto out_khwn_desc =
make_ConstantTensorDescriptor_default_rank_packed(Sequence<K, Ho, Wo, N>{});
auto out_khwn_desc = make_ConstantTensorDescriptor_packed(Sequence<K, Ho, Wo, N>{});
ostream_ConstantTensorDescriptor(out_khwn_desc, std::cout << "out_khwn_desc: ");
Tensor<T> out_khwn(make_TensorDescriptor(out_khwn_desc));

2
driver/device_convolution_implicit_gemm_v1_nchw_cyxk_nkhw.hpp
View File

@@ -36,7 +36,7 @@ void device_convolution_implicit_gemm_v1_nchw_cyxk_nkhw(InDesc,
constexpr index_t X = wei_kcyx_desc.GetLength(I3);
// reorder weight
auto wei_cyxk_desc = make_ConstantTensorDescriptor_default_rank_packed(Sequence<C, Y, X, K>{});
auto wei_cyxk_desc = make_ConstantTensorDescriptor_packed(Sequence<C, Y, X, K>{});
ostream_ConstantTensorDescriptor(wei_cyxk_desc, std::cout << "wei_cyxk_desc: ");
Tensor<T> wei_cyxk(make_TensorDescriptor(wei_cyxk_desc));

2
driver/device_convolution_implicit_gemm_v3_nchw_cyxk_nkhw.hpp
View File

@@ -36,7 +36,7 @@ void device_convolution_implicit_gemm_v3_nchw_cyxk_nkhw(InDesc,
constexpr index_t X = wei_kcyx_desc.GetLength(I3);
// reorder weight
auto wei_cyxk_desc = make_ConstantTensorDescriptor_default_rank_packed(Sequence<C, Y, X, K>{});
auto wei_cyxk_desc = make_ConstantTensorDescriptor_packed(Sequence<C, Y, X, K>{});
ostream_ConstantTensorDescriptor(wei_cyxk_desc, std::cout << "wei_cyxk_desc: ");
Tensor<T> wei_cyxk(make_TensorDescriptor(wei_cyxk_desc));

8
driver/driver.hip.cpp
View File

@@ -443,7 +443,7 @@ int main(int argc, char* argv[])
constexpr index_t HPad = 0;
constexpr index_t WPad = 0;
#elif 0
#elif 1
// 3x3 filter, 28x28 image
constexpr index_t N = 128;
constexpr index_t C = 256;
@@ -455,7 +455,7 @@ int main(int argc, char* argv[])
constexpr index_t HPad = 0;
constexpr index_t WPad = 0;
#elif 1
#elif 0
// 1x1 filter, 28x28 image
constexpr index_t N = 128;
constexpr index_t C = 512;
@@ -568,8 +568,8 @@ int main(int argc, char* argv[])
auto lower_pads = Sequence<HPad, WPad>{};
auto upper_pads = Sequence<HPad, WPad>{};
auto in_nchw_desc = make_ConstantTensorDescriptor_default_rank_packed(Sequence<N, C, HI, WI>{});
auto wei_kcyx_desc = make_ConstantTensorDescriptor_default_rank_packed(Sequence<K, C, Y, X>{});
auto in_nchw_desc = make_ConstantTensorDescriptor_packed(Sequence<N, C, HI, WI>{});
auto wei_kcyx_desc = make_ConstantTensorDescriptor_packed(Sequence<K, C, Y, X>{});
auto out_nkhw_desc = get_convolution_with_padding_output_default_4d_tensor_descriptor(
in_nchw_desc, wei_kcyx_desc, lower_pads, upper_pads);