Skip to content

feat(sm120): add tileN = 8,16 for the dense F8F6F4 GEMM #3441

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

Open
sunnypatell wants to merge 1 commit into
base: main
Choose a base branch
from
+209 −1
Open

feat(sm120): add tileN = 8,16 for the dense F8F6F4 GEMM #3441

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
2 changes: 1 addition & 1 deletion include/cutlass/gemm/collective/builders/sm120_mma_builder.inl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -147,7 +147,7 @@ struct CollectiveBuilder<
"Unsupported kernel schedule by this collective mainloop dispatch policy.");

using SmemCopyAtomA = Copy_Atom<decltype(detail::sm120_rr_smem_copy_selector_A<ElementA, ElementB, UseF8f6f4>()), SmemAllocTypeA>;
using SmemCopyAtomB = Copy_Atom<decltype(detail::sm120_rr_smem_copy_selector_B<ElementA, ElementB, UseF8f6f4>()), SmemAllocTypeB>;
using SmemCopyAtomB = Copy_Atom<decltype(detail::sm120_rr_smem_copy_selector_B<ElementA, ElementB, UseF8f6f4, size<1>(TileShape_MNK{})>()), SmemAllocTypeB>;

using CollectiveOp = CollectiveMma<
DispatchPolicy,
Expand Down
208 changes: 208 additions & 0 deletions test/unit/gemm/device/sm120_tensorop_gemm/sm120_gemm_f8_f8_f32_tensor_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,4 +108,212 @@ TEST(SM120_Device_Gemm_fe4m3t_fe4m3n_f32n_tensor_op_f32, 128x64x64_1x1x1) {

///////////////////////////////////////////////////////////////////////////////

TEST(SM120_Device_Gemm_fe4m3t_fe4m3n_f32n_tensor_op_f32, 128x32x64_1x1x1) {
using ElementA = cutlass::float_e4m3_t;
using ElementB = cutlass::float_e4m3_t;
using ElementC = float;
using ElementD = float;
using ElementAccumulator = float;
using ElementCompute = float;
using LayoutA = cutlass::layout::RowMajor;
using LayoutB = cutlass::layout::ColumnMajor;
using LayoutC = cutlass::layout::ColumnMajor;
using LayoutD = cutlass::layout::ColumnMajor;

static constexpr int Alignment = 16;
static constexpr int AlignmentC = 128 / cutlass::sizeof_bits<ElementC>::value;
static constexpr int AlignmentD = 128 / cutlass::sizeof_bits<ElementD>::value;

using TileShape = Shape<_128,_32,_64>;
using ClusterShape = Shape<_1,_1,_1>;

using CollectiveEpilogue = typename cutlass::epilogue::collective::CollectiveBuilder<
cutlass::arch::Sm120, cutlass::arch::OpClassTensorOp,
TileShape, ClusterShape,
cutlass::epilogue::collective::EpilogueTileAuto,
ElementAccumulator, ElementCompute,
ElementC, LayoutC, AlignmentC,
ElementD, LayoutD, AlignmentD,
cutlass::epilogue::collective::EpilogueScheduleAuto
>::CollectiveOp;

using CollectiveMainloop = typename cutlass::gemm::collective::CollectiveBuilder<
cutlass::arch::Sm120, cutlass::arch::OpClassTensorOp,
ElementA, LayoutA, Alignment,
ElementB, LayoutB, Alignment,
ElementAccumulator,
TileShape, ClusterShape,
cutlass::gemm::collective::StageCountAutoCarveout<static_cast<int>(sizeof(typename CollectiveEpilogue::SharedStorage))>,
cutlass::gemm::collective::KernelScheduleAuto
>::CollectiveOp;

using GemmKernel = cutlass::gemm::kernel::GemmUniversal<
Shape<int,int,int,int>,
CollectiveMainloop,
CollectiveEpilogue
>;

using Gemm = cutlass::gemm::device::GemmUniversalAdapter<GemmKernel>;
bool result = test::gemm::device::TestSmall<Gemm, true>(1.0, 0.0);
EXPECT_TRUE(result);
}

///////////////////////////////////////////////////////////////////////////////

TEST(SM120_Device_Gemm_fe4m3t_fe4m3n_f32n_tensor_op_f32, 128x16x64_1x1x1) {
using ElementA = cutlass::float_e4m3_t;
using ElementB = cutlass::float_e4m3_t;
using ElementC = float;
using ElementD = float;
using ElementAccumulator = float;
using ElementCompute = float;
using LayoutA = cutlass::layout::RowMajor;
using LayoutB = cutlass::layout::ColumnMajor;
using LayoutC = cutlass::layout::ColumnMajor;
using LayoutD = cutlass::layout::ColumnMajor;

static constexpr int Alignment = 16;
static constexpr int AlignmentC = 128 / cutlass::sizeof_bits<ElementC>::value;
static constexpr int AlignmentD = 128 / cutlass::sizeof_bits<ElementD>::value;

using TileShape = Shape<_128,_16,_64>;
using ClusterShape = Shape<_1,_1,_1>;

using CollectiveEpilogue = typename cutlass::epilogue::collective::CollectiveBuilder<
cutlass::arch::Sm120, cutlass::arch::OpClassTensorOp,
TileShape, ClusterShape,
cutlass::epilogue::collective::EpilogueTileAuto,
ElementAccumulator, ElementCompute,
ElementC, LayoutC, AlignmentC,
ElementD, LayoutD, AlignmentD,
cutlass::epilogue::collective::EpilogueScheduleAuto
>::CollectiveOp;

using CollectiveMainloop = typename cutlass::gemm::collective::CollectiveBuilder<
cutlass::arch::Sm120, cutlass::arch::OpClassTensorOp,
ElementA, LayoutA, Alignment,
ElementB, LayoutB, Alignment,
ElementAccumulator,
TileShape, ClusterShape,
cutlass::gemm::collective::StageCountAutoCarveout<static_cast<int>(sizeof(typename CollectiveEpilogue::SharedStorage))>,
cutlass::gemm::collective::KernelScheduleAuto
>::CollectiveOp;

using GemmKernel = cutlass::gemm::kernel::GemmUniversal<
Shape<int,int,int,int>,
CollectiveMainloop,
CollectiveEpilogue
>;

using Gemm = cutlass::gemm::device::GemmUniversalAdapter<GemmKernel>;
bool result = test::gemm::device::TestSmall<Gemm, true>(1.0, 0.0);
EXPECT_TRUE(result);
}

///////////////////////////////////////////////////////////////////////////////

TEST(SM120_Device_Gemm_fe4m3t_fe4m3n_f32n_tensor_op_f32, 128x8x64_1x1x1) {
using ElementA = cutlass::float_e4m3_t;
using ElementB = cutlass::float_e4m3_t;
using ElementC = float;
using ElementD = float;
using ElementAccumulator = float;
using ElementCompute = float;
using LayoutA = cutlass::layout::RowMajor;
using LayoutB = cutlass::layout::ColumnMajor;
using LayoutC = cutlass::layout::ColumnMajor;
using LayoutD = cutlass::layout::ColumnMajor;

static constexpr int Alignment = 16;
static constexpr int AlignmentC = 128 / cutlass::sizeof_bits<ElementC>::value;
static constexpr int AlignmentD = 128 / cutlass::sizeof_bits<ElementD>::value;

using TileShape = Shape<_128,_8,_64>;
using ClusterShape = Shape<_1,_1,_1>;

using CollectiveEpilogue = typename cutlass::epilogue::collective::CollectiveBuilder<
cutlass::arch::Sm120, cutlass::arch::OpClassTensorOp,
TileShape, ClusterShape,
cutlass::epilogue::collective::EpilogueTileAuto,
ElementAccumulator, ElementCompute,
ElementC, LayoutC, AlignmentC,
ElementD, LayoutD, AlignmentD,
cutlass::epilogue::collective::EpilogueScheduleAuto
>::CollectiveOp;

using CollectiveMainloop = typename cutlass::gemm::collective::CollectiveBuilder<
cutlass::arch::Sm120, cutlass::arch::OpClassTensorOp,
ElementA, LayoutA, Alignment,
ElementB, LayoutB, Alignment,
ElementAccumulator,
TileShape, ClusterShape,
cutlass::gemm::collective::StageCountAutoCarveout<static_cast<int>(sizeof(typename CollectiveEpilogue::SharedStorage))>,
cutlass::gemm::collective::KernelScheduleAuto
>::CollectiveOp;

using GemmKernel = cutlass::gemm::kernel::GemmUniversal<
Shape<int,int,int,int>,
CollectiveMainloop,
CollectiveEpilogue
>;

using Gemm = cutlass::gemm::device::GemmUniversalAdapter<GemmKernel>;
bool result = test::gemm::device::TestSmall<Gemm, true>(1.0, 0.0);
EXPECT_TRUE(result);
}

///////////////////////////////////////////////////////////////////////////////

TEST(SM120_Device_Gemm_fe4m3t_fe4m3n_f32n_tensor_op_f32, 128x8x64_1x1x1_pingpong) {
using ElementA = cutlass::float_e4m3_t;
using ElementB = cutlass::float_e4m3_t;
using ElementC = float;
using ElementD = float;
using ElementAccumulator = float;
using ElementCompute = float;
using LayoutA = cutlass::layout::RowMajor;
using LayoutB = cutlass::layout::ColumnMajor;
using LayoutC = cutlass::layout::ColumnMajor;
using LayoutD = cutlass::layout::ColumnMajor;

static constexpr int Alignment = 16;
static constexpr int AlignmentC = 128 / cutlass::sizeof_bits<ElementC>::value;
static constexpr int AlignmentD = 128 / cutlass::sizeof_bits<ElementD>::value;

using TileShape = Shape<_128,_8,_64>;
using ClusterShape = Shape<_1,_1,_1>;

using CollectiveEpilogue = typename cutlass::epilogue::collective::CollectiveBuilder<
cutlass::arch::Sm120, cutlass::arch::OpClassTensorOp,
TileShape, ClusterShape,
cutlass::epilogue::collective::EpilogueTileAuto,
ElementAccumulator, ElementCompute,
ElementC, LayoutC, AlignmentC,
ElementD, LayoutD, AlignmentD,
cutlass::epilogue::collective::EpilogueScheduleAuto
>::CollectiveOp;

using CollectiveMainloop = typename cutlass::gemm::collective::CollectiveBuilder<
cutlass::arch::Sm120, cutlass::arch::OpClassTensorOp,
ElementA, LayoutA, Alignment,
ElementB, LayoutB, Alignment,
ElementAccumulator,
TileShape, ClusterShape,
cutlass::gemm::collective::StageCountAutoCarveout<static_cast<int>(sizeof(typename CollectiveEpilogue::SharedStorage))>,
cutlass::gemm::KernelTmaWarpSpecializedPingpong
>::CollectiveOp;

using GemmKernel = cutlass::gemm::kernel::GemmUniversal<
Shape<int,int,int,int>,
CollectiveMainloop,
CollectiveEpilogue
>;

using Gemm = cutlass::gemm::device::GemmUniversalAdapter<GemmKernel>;
bool result = test::gemm::device::TestSmall<Gemm, true>(1.0, 0.0);
EXPECT_TRUE(result);
}

///////////////////////////////////////////////////////////////////////////////

#endif // (defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED))