Skip to content

new api changes for unit tests for supported configurations #683

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
taozha2 merged 2 commits into from
Sep 22, 2026
Merged

new api changes for unit tests for supported configurations #683

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4d66c16
new api changes for unit tests for supported configurations
rajprince-intel Jan 6, 2026
1c47644
Update gemm_universal_lincomb_per_rowbias_eltact.cpp
rajprince-intel Jan 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 13 additions & 10 deletions test/unit/gemm/device/gemm_universal_lincomb_per_rowbias_eltact.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,21 +53,24 @@ struct MainloopIntelXeXMX16_LinCombPerRowBiasEltAct_GemmConfig {
using ElementOutput = float;
using ElementBias = float;

using GmemTiledCopyA = XE_2D_U16x32x32_LD_N;
using GmemTiledCopyB = XE_2D_U16x32x32_LD_V;
// When left unspecified (void), MainloopXeL1Staged automatically selects
// appropriate 2D block copy operations
using GmemTiledCopyA = void; //XE_LOAD_2D<16, 32, 32>;
using GmemTiledCopyB = void; //XE_LOAD_2D_VNNI<16, 32, 32>;

using TileShape = Shape<_256, _256, _32>;

// Use XE_DPAS_TT atom for new API
using TiledMma =
typename TiledMMAHelper<MMA_Atom<XE_8x16x16_F32BF16BF16F32_TT>,
typename TiledMMAHelper<MMA_Atom<XE_DPAS_TT<8, float, cute::bfloat16_t>>,
Layout<TileShape>,
Layout<Shape<_8, _4, _1>, Stride<_4, _1, _0>>>::TiledMMA;

constexpr static int PipelineStages = 2;
using GEMMDispatchPolicy = cutlass::gemm::MainloopIntelXeXMX16<PipelineStages>;
using EpilogueDispatchPolicy = cutlass::epilogue::IntelXeXMX16;
using GEMMDispatchPolicy = cutlass::gemm::MainloopXeL1Staged<PipelineStages>;
using EpilogueDispatchPolicy = cutlass::epilogue::IntelXeGeneric;

using EpilogueOp = cutlass::epilogue::fusion::LinCombPerRowBiasEltAct<
using EpilogueOp = cutlass::epilogue::fusion::XeLinCombPerRowBiasEltAct<
cutlass::epilogue::thread::ReLu,
ElementOutput, ElementComputeEpilogue, ElementBias,
ElementAccumulator, ElementAccumulator,
Expand All @@ -81,15 +84,14 @@ struct MainloopIntelXeXMX16_LinCombPerRowBiasEltAct_GemmConfig {
using CollectiveEpilogue = cutlass::epilogue::collective::CollectiveEpilogue<
EpilogueDispatchPolicy,
TileShape,
void, // Epilogue tile (void = automatic)
ElementAccumulator,
cutlass::gemm::TagToStrideC_t<LayoutC>,
ElementOutput,
cutlass::gemm::TagToStrideC_t<LayoutD>,
FusionCallBacks,
XE_2D_U32x8x16_LD_N,
void, void,
XE_2D_U32x8x16_ST_N,
void, void>;
void, // Load copy atom (void = automatic)
void>; // Store copy atom (void = automatic)

using CollectiveMainloop = cutlass::gemm::collective::CollectiveMma<
GEMMDispatchPolicy,
Expand Down Expand Up @@ -177,3 +179,4 @@ TEST(MainloopIntelXeXMX16_LinCombPerRowBiasEltAct_NonParam, LargeKSmallMN) {
EXPECT_TRUE((test::gemm::device::TestXe<Gemm, cutlass::epilogue::thread::ReLu>(32, 32, 8192, 1, 1.0f, 0.0f)));
EXPECT_TRUE((test::gemm::device::TestXe<Gemm, cutlass::epilogue::thread::ReLu>(64, 64, 16384, 1, 1.0f, 0.0f)));
}

10 changes: 5 additions & 5 deletions test/unit/gemm/device/gemm_universal_mainloopintelxexmx16group.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,21 +57,21 @@ struct MainloopIntelXeXMX16Group_GemmConfig {
using LayoutC = cutlass::layout::RowMajor;
using LayoutD = cutlass::layout::RowMajor;

using GmemTiledCopyA = XE_2D_U16x32x32_LD_N;
using GmemTiledCopyB = XE_2D_U16x32x32_LD_V;
using GmemTiledCopyA = void; //XE_2D_U16x32x32_LD_N
using GmemTiledCopyB = void; //XE_2D_U16x32x32_LD_V

// Workgroup-level tile
using TileShape = Shape<_256, _256, _32>;

using TiledMma =
TiledMMA<MMA_Atom<XE_8x16x16_F32BF16BF16F32_TT>,
TiledMMA<MMA_Atom<XE_DPAS_TT<8, float, cute::bfloat16_t>>, //XE_8x16x16_F32BF16BF16F32_TT
Layout<Shape<_8, _4, _1>, Stride<_4, _1, _0>>,
Tile<Layout<Shape<_8, _8, _4>, Stride<_1, _32, _8>>,
Layout<Shape<_16, _4, _4>, Stride<_1, _64, _16>>, _32>>;

constexpr static int PipelineStages = 2;
// Dispatch to grouped gemm algorithm
using GEMMDispatchPolicy = cutlass::gemm::MainloopIntelXeXMX16Group<PipelineStages>;
using GEMMDispatchPolicy = cutlass::gemm::MainloopXeL1StagedGroup<PipelineStages>;
using EpilogueDispatchPolicy = cutlass::epilogue::IntelXeXMX16Group;

using EpilogueOp = cutlass::epilogue::fusion::LinearCombination<ElementOutput, ElementComputeEpilogue,
Expand Down Expand Up @@ -200,4 +200,4 @@ TEST(MainloopIntelXeXMX16Group_NonParam, LargeKSmallMN) {
{64, 64, 16384}
};
EXPECT_TRUE(test::gemm::device::TestXeGrouped<Gemm>(problem_sizes, 1.0f, 0.0f));
}
}
Loading