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

Revert "[CK_TILE] Tile loop persistent gemm kernel (#2191)" (#2293)

Browse Source 
This reverts commit ffb52783d0.
This commit is contained in:
Illia Silin
2025-06-05 09:24:00 -07:00
committed by GitHub
parent 7ea1508b59
commit 233e274077
10 changed files with 18 additions and 232 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
test/ck_tile/gemm/test_gemm_pipeline_util.hpp
View File

@@ -76,8 +76,6 @@ 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;
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>
@@ -119,17 +117,14 @@ class TestCkTileGemmPipeline : public ::testing::Test
GemmSpatiallyLocalTilePartitioner<GemmShape, TileParitionerGroupNum, TileParitionerM01>;
using Traits = ck_tile::TileGemmTraits<kPadM, kPadN, kPadK, ALayout, BLayout, CLayout>;
static constexpr bool StructuredSparsity = false;
using GemmUniversalTraits = ck_tile::TileGemmUniversalTraits<kPadM,
using GemmUniversalTraits = ck_tile::TileGemmUniversalTraits<kPadM,
kPadN,
kPadK,
DoubleSmemBuffer,
ALayout,
BLayout,
CLayout,
TransposeC,
StructuredSparsity,
Persistent>;
TransposeC>;
using GemmPipelineProblem =
ck_tile::GemmPipelineProblem<ADataType, BDataType, AccDataType, GemmShape, Traits>;
@@ -182,15 +177,7 @@ class TestCkTileGemmPipeline : public ::testing::Test
using Kernel = ck_tile::GemmKernel<TilePartitioner, GemmPipeline, GemmEpilogue>;
auto kargs = Kernel::MakeKernelArgs(args);
dim3 grids;
if constexpr(Persistent)
{
grids = Kernel::MaxOccupancyGridSize(s);
}
else
{
grids = Kernel::GridSize(args.M, args.N, args.k_batch);
}
const dim3 grids = Kernel::GridSize(args.M, args.N, args.k_batch);
constexpr dim3 blocks = Kernel::BlockSize();
if(!Kernel::IsSupportedArgument(kargs))
@@ -359,6 +346,9 @@ class TestCkTileGemmPipeline : public ::testing::Test
"Error: Incorrect results!",
rtol_atol.at(ck_tile::number<0>{}),
rtol_atol.at(ck_tile::number<1>{}));
std::cout << "Relative error threshold: " << rtol_atol.at(ck_tile::number<0>{})
<< " Absolute error threshold: " << rtol_atol.at(ck_tile::number<1>{})
<< std::endl;
EXPECT_TRUE(pass);
}
};