ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-02 20:51:23 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_TILE] Multiple-D GEMM example (#2219)

Browse Source 
* Multiple d, initial commit

* Check Ds Layout

* Readme and clang format

* Update branch & conflicts

* Multiple D - fix clang-formatter

* Rename elemetwise_op

* Fix CI

* Code review part1

* Remove printf

* Remove unnecessary comment

* Add new tests with Col layout

* Review part 2

* Added support for Multiple D GEMM

* Update comment

* Remove maybe_unused

* Clang-format

* Review part 3

* Add comment to function

* Add comment to function: another

* Take number of params for a refrence function

* Remove additional d param for 0 tensor

* Change name of function

* Fix CI fails
This commit is contained in:
Mateusz Ozga
2025-06-13 19:39:11 +02:00
committed by GitHub
parent 3a0cb27966
commit bd96ac9742
34 changed files with 2267 additions and 285 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
test/ck_tile/gemm/test_gemm_pipeline_util.hpp
View File

@@ -76,12 +76,17 @@ class TestCkTileGemmPipeline : public ::testing::Test
using CDataType = std::tuple_element_t<6, Tuple>;
static constexpr auto Scheduler = std::tuple_element_t<7, Tuple>::value;
static constexpr auto PipelineType = std::tuple_element_t<8, Tuple>::value;
using DsLayout = ck_tile::tuple<>;
using DsDataType = ck_tile::tuple<>;
static constexpr bool Persistent =
ck_tile::tuple_element_or_default_t<Tuple, 9, std::false_type>::value;
// TODO: expose tile size through test t-param ?
template <bool PadM, bool PadN, bool PadK>
void invoke_gemm(const ck_tile::GemmHostArgs& args, const ck_tile::stream_config& s)
void invoke_gemm(const ck_tile::GemmHostArgs</*NumDTensor = 0*/>& args,
const ck_tile::stream_config& s)
{
// TODO: This should be parameterized in tests
constexpr ck_tile::index_t M_Tile = 256;
@@ -165,9 +170,12 @@ class TestCkTileGemmPipeline : public ::testing::Test
using GemmEpilogue = ck_tile::CShuffleEpilogue<
ck_tile::CShuffleEpilogueProblem<ADataType,
BDataType,
DsDataType,
AccDataType,
CDataType,
DsLayout,
CLayout,
ck_tile::element_wise::PassThrough,
GemmPipeline::BlockSize,
TilePartitioner::MPerBlock,
TilePartitioner::NPerBlock,
@@ -326,17 +334,17 @@ class TestCkTileGemmPipeline : public ::testing::Test
c_m_n_dev_buf.SetZero();
c_m_n_dev_result.SetZero();
ck_tile::GemmHostArgs args;
ck_tile::GemmHostArgs</*NumDTensor = 0*/> args;
args.a_ptr = a_m_k_dev_buf.GetDeviceBuffer();
args.b_ptr = b_k_n_dev_buf.GetDeviceBuffer();
args.c_ptr = c_m_n_dev_buf.GetDeviceBuffer();
args.e_ptr = c_m_n_dev_buf.GetDeviceBuffer();
args.k_batch = kbatch;
args.M = M;
args.N = N;
args.K = K;
args.stride_A = stride_A;
args.stride_B = stride_B;
args.stride_C = stride_C;
args.stride_E = stride_C;
invoke_gemm<PadM, PadN, PadK>(args, ck_tile::stream_config{nullptr, false});