ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-11 09:40:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

resolve conflicts

Browse Source
This commit is contained in:
Jakub Piasecki
2025-06-16 15:33:24 +00:00
parent 60bd2a4fdf 1996c99bee
commit 116a503274
8 changed files with 106 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

0
example/ck_tile/19_grouped_convolution/CMakeLists.txt → example/ck_tile/20_grouped_convolution/CMakeLists.txt
View File

0
example/ck_tile/19_grouped_convolution/grouped_convolution_backward_weight.cpp → example/ck_tile/20_grouped_convolution/grouped_convolution_backward_weight.cpp
View File

10
example/ck_tile/19_grouped_convolution/grouped_convolution_forward.cpp → example/ck_tile/20_grouped_convolution/grouped_convolution_forward.cpp
View File

@@ -19,7 +19,10 @@ template <ck_tile::index_t NDimSpatial,
typename OutDataType,
typename InLayout,
typename WeiLayout,
typename OutLayout>
typename OutLayout,
typename DsDataType = ck_tile::tuple<>,
typename DsLayout = ck_tile::tuple<>,
typename CDEElementWise = ck_tile::element_wise::PassThrough>
float grouped_conv_fwd(const ck_tile::GroupedConvHostArgs& args, const ck_tile::stream_config& s)
{
constexpr int kBlockPerCu = 1;
@@ -49,7 +52,7 @@ float grouped_conv_fwd(const ck_tile::GroupedConvHostArgs& args, const ck_tile::
constexpr auto ConvSpec = ck_tile::ConvolutionSpecialization::Default;
using TilePartitioner = ck_tile::GemmTile1DPartitioner<CodegenShape>;
using GroupedConvTraitsType =
ck_tile::GroupedConvTraits<NDimSpatial, ConvSpec, InLayout, WeiLayout, OutLayout>;
ck_tile::GroupedConvTraits<NDimSpatial, ConvSpec, InLayout, WeiLayout, DsLayout, OutLayout>;
using CodegenPipelineProblem =
ck_tile::GemmPipelineProblem<InDataType,
WeiDataType,
@@ -68,9 +71,12 @@ float grouped_conv_fwd(const ck_tile::GroupedConvHostArgs& args, const ck_tile::
using ConvEpilogue = ck_tile::CShuffleEpilogue<
ck_tile::CShuffleEpilogueProblem<InDataType,
WeiDataType,
DsDataType,
AccDataType,
OutDataType,
typename GroupedConvTraitsType::ImplicitGemmDsLayout,
ck_tile::tensor_layout::gemm::RowMajor,
CDEElementWise,
CodegenPipelineProblem::kBlockSize,
TilePartitioner::MPerBlock,
TilePartitioner::NPerBlock,

0
example/ck_tile/19_grouped_convolution/grouped_convolution_utils.hpp → example/ck_tile/20_grouped_convolution/grouped_convolution_utils.hpp
View File

0
example/ck_tile/19_grouped_convolution/run_grouped_convolution_bwd_weight_example.inc → example/ck_tile/20_grouped_convolution/run_grouped_convolution_bwd_weight_example.inc
View File

1
example/ck_tile/19_grouped_convolution/run_grouped_convolution_fwd_example.inc → example/ck_tile/20_grouped_convolution/run_grouped_convolution_fwd_example.inc
View File

@@ -146,6 +146,7 @@ int run_grouped_conv_fwd_example_with_layouts(
ck_tile::GroupedConvHostArgs args(conv_param,
input_dev_buf.GetDeviceBuffer(),
weight_dev_buf.GetDeviceBuffer(),
{},
output_dev_buf.GetDeviceBuffer(),
kbatch);