leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

v4.3 tag release update. (#2789)

Browse Source
This commit is contained in:
Junkai-Wu
2025-11-21 09:49:44 +08:00
committed by GitHub
parent 406e078b29
commit 8cd5bef43a
225 changed files with 23229 additions and 2813 deletions
Download Patch File Download Diff File Expand all files Collapse all files

47
test/unit/transform/device/CMakeLists.txt
View File

@@ -56,3 +56,50 @@ cutlass_test_unit_add_executable(
sm90_sparse_gemm_compressor_f8.cu
)
add_custom_target(
cutlass_test_unit_sm100_structured_sparse_gemm_compressor
DEPENDS
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f32
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f16
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f8
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f6
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f4_qmma
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f4_omma
)
cutlass_test_unit_add_executable(
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f32
sm100_sparse_gemm_compressor_f32.cu
)
cutlass_test_unit_add_executable(
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f16
sm100_sparse_gemm_compressor_f16.cu
)
cutlass_test_unit_add_executable(
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f8
sm100_sparse_gemm_compressor_f8.cu
)
cutlass_test_unit_add_executable(
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f6
sm100_sparse_gemm_compressor_f6.cu
)
cutlass_test_unit_add_executable(
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f4_qmma
sm100_sparse_gemm_compressor_f4_qmma.cu
)
cutlass_test_unit_add_executable(
cutlass_test_unit_sm100_structured_sparse_gemm_compressor_f4_omma
sm100_sparse_gemm_compressor_f4_omma.cu
)

92
test/unit/transform/device/sm100_sparse_gemm_compressor_f16.cu Normal file
View File

@@ -0,0 +1,92 @@
/***************************************************************************************************
* Copyright (c) 2024 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: BSD-3-Clause
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
**************************************************************************************************/
#include "cute/arch/mma_sm100_desc.hpp" // cute::UMMA::Major
#include "cutlass/gemm/collective/builders/sm100_common.inl" // tag_to_umma_major_A
#include "cutlass/gemm/collective/builders/sm1xx_sparse_config.inl" // Sm1xxGemmSparseConfig
#include "cutlass/transform/kernel/sparse_gemm_compressor.hpp" // StructuredSparseCompressor
#include "cutlass/transform/device/transform_universal_adapter.hpp" // TransformUniversalAdapter
#include "testbed_sparse_gemm_compressor.hpp" // TestbedSparseGemmCompressor
///////////////////////////////////////////////////////////////////////////////////////////////////
// * Test Plan
// ElementA : fp16
// LayoutA : row / col
// Gemm : 1x 2x 3x multiplier of alignment requirement. corner case that smaller than alignment requirement
///////////////////////////////////////////////////////////////////////////////////////////////////
#if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f16_t)
{
// Test Settings
using ElementA = cutlass::half_t;
using LayoutATag = cutlass::layout::RowMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f16_n)
{
// Test Settings
using ElementA = cutlass::bfloat16_t;
using LayoutATag = cutlass::layout::ColumnMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto() );
}
#endif // #if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)

92
test/unit/transform/device/sm100_sparse_gemm_compressor_f32.cu Normal file
View File

@@ -0,0 +1,92 @@
/***************************************************************************************************
* Copyright (c) 2024 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: BSD-3-Clause
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
**************************************************************************************************/
#include "cute/arch/mma_sm100_desc.hpp" // cute::UMMA::Major
#include "cutlass/gemm/collective/builders/sm100_common.inl" // tag_to_umma_major_A
#include "cutlass/gemm/collective/builders/sm1xx_sparse_config.inl" // Sm1xxGemmSparseConfig
#include "cutlass/transform/kernel/sparse_gemm_compressor.hpp" // StructuredSparseCompressor
#include "cutlass/transform/device/transform_universal_adapter.hpp" // TransformUniversalAdapter
#include "testbed_sparse_gemm_compressor.hpp" // TestbedSparseGemmCompressor
///////////////////////////////////////////////////////////////////////////////////////////////////
// * Test Plan
// ElementA : fp32
// LayoutA : row / col
// Gemm : 1x 2x 3x multiplier of alignment requirement. corner case that smaller than alignment requirement
///////////////////////////////////////////////////////////////////////////////////////////////////
#if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f32_t)
{
// Test Settings
using ElementA = float;
using LayoutATag = cutlass::layout::RowMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<4, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f32_n)
{
// Test Settings
using ElementA = cutlass::tfloat32_t;
using LayoutATag = cutlass::layout::ColumnMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<4, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto());
}
#endif // #if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)

92
test/unit/transform/device/sm100_sparse_gemm_compressor_f4_omma.cu Normal file
View File

@@ -0,0 +1,92 @@
/***************************************************************************************************
* Copyright (c) 2024 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: BSD-3-Clause
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
**************************************************************************************************/
#include "cute/arch/mma_sm100_desc.hpp" // cute::UMMA::Major
#include "cutlass/gemm/collective/builders/sm100_common.inl" // tag_to_umma_major_A
#include "cutlass/gemm/collective/builders/sm1xx_sparse_config.inl" // Sm1xxGemmSparseConfig
#include "cutlass/transform/kernel/sparse_gemm_compressor.hpp" // StructuredSparseCompressor
#include "cutlass/transform/device/transform_universal_adapter.hpp" // TransformUniversalAdapter
#include "testbed_sparse_gemm_compressor.hpp" // TestbedSparseGemmCompressor
///////////////////////////////////////////////////////////////////////////////////////////////////
// * Test Plan
// ElementA : fp4 (qmma), fp4 (omma)
// LayoutA : row / col
// Gemm : 1x 2x 3x multiplier of alignment requirement. corner case that smaller than alignment requirement
///////////////////////////////////////////////////////////////////////////////////////////////////
#if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, omma_f4_t)
{
// Test Settings
using ElementA = cutlass::float_e2m1_t;
using LayoutATag = cutlass::layout::RowMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<4, uint8_t>;
using ElementEMma = cute::sparse_elem<16, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, omma_f4_runtimedtype_t)
{
// Test Settings
using ElementA = cutlass::type_erased_dynamic_float4_t;
using LayoutATag = cutlass::layout::RowMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<4, uint8_t>;
using ElementEMma = cute::sparse_elem<16, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto_small());
}
#endif // #if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)

139
test/unit/transform/device/sm100_sparse_gemm_compressor_f4_qmma.cu Normal file
View File

@@ -0,0 +1,139 @@
/***************************************************************************************************
* Copyright (c) 2024 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: BSD-3-Clause
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
**************************************************************************************************/
#include "cute/arch/mma_sm100_desc.hpp" // cute::UMMA::Major
#include "cutlass/gemm/collective/builders/sm100_common.inl" // tag_to_umma_major_A
#include "cutlass/gemm/collective/builders/sm1xx_sparse_config.inl" // Sm1xxGemmSparseConfig
#include "cutlass/transform/kernel/sparse_gemm_compressor.hpp" // StructuredSparseCompressor
#include "cutlass/transform/device/transform_universal_adapter.hpp" // TransformUniversalAdapter
#include "testbed_sparse_gemm_compressor.hpp" // TestbedSparseGemmCompressor
///////////////////////////////////////////////////////////////////////////////////////////////////
// * Test Plan
// ElementA : fp4 (qmma), fp4 (omma)
// LayoutA : row / col
// Gemm : 1x 2x 3x multiplier of alignment requirement. corner case that smaller than alignment requirement
///////////////////////////////////////////////////////////////////////////////////////////////////
#if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f4_t)
{
// Test Settings
using ElementA = cutlass::float_e2m1_t;
using LayoutATag = cutlass::layout::RowMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f4_n)
{
// Test Settings
using ElementA = cutlass::float_e2m1_t;
using LayoutATag = cutlass::layout::ColumnMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f4_runtimedtype_t)
{
// Test Settings
using ElementA = cutlass::type_erased_dynamic_float4_t;
using ElementAMmaRaw = cutlass::detail::type_erased_dynamic_float4_unpacksmem_t;
using LayoutATag = cutlass::layout::RowMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementAMmaRaw>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto_small());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f4_runtimedtype_n)
{
// Test Settings
using ElementA = cutlass::type_erased_dynamic_float4_t;
using ElementAMmaRaw = cutlass::detail::type_erased_dynamic_float4_unpacksmem_t;
using LayoutATag = cutlass::layout::ColumnMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementAMmaRaw>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto_small());
}
#endif // #if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)

138
test/unit/transform/device/sm100_sparse_gemm_compressor_f6.cu Normal file
View File

@@ -0,0 +1,138 @@
/***************************************************************************************************
* Copyright (c) 2024 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: BSD-3-Clause
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
**************************************************************************************************/
#include "cute/arch/mma_sm100_desc.hpp" // cute::UMMA::Major
#include "cutlass/gemm/collective/builders/sm100_common.inl" // tag_to_umma_major_A
#include "cutlass/gemm/collective/builders/sm1xx_sparse_config.inl" // Sm1xxGemmSparseConfig
#include "cutlass/transform/kernel/sparse_gemm_compressor.hpp" // StructuredSparseCompressor
#include "cutlass/transform/device/transform_universal_adapter.hpp" // TransformUniversalAdapter
#include "testbed_sparse_gemm_compressor.hpp" // TestbedSparseGemmCompressor
///////////////////////////////////////////////////////////////////////////////////////////////////
// * Test Plan
// ElementA : fp6
// LayoutA : row / col
// Gemm : 1x 2x 3x multiplier of alignment requirement. corner case that smaller than alignment requirement
///////////////////////////////////////////////////////////////////////////////////////////////////
#if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f6_t)
{
// Test Settings
using ElementA = cutlass::float_e3m2_t;
using LayoutATag = cutlass::layout::RowMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f6_n)
{
// Test Settings
using ElementA = cutlass::float_e2m3_t;
using LayoutATag = cutlass::layout::ColumnMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f6_runtimedtype_t)
{
// Test Settings
using ElementA = cutlass::type_erased_dynamic_float6_t;
using ElementAMmaRaw = cutlass::detail::type_erased_dynamic_float6_unpacksmem_t;
using LayoutATag = cutlass::layout::RowMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementAMmaRaw>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto_small());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f6_runtimedtype_n)
{
// Test Settings
using ElementA = cutlass::type_erased_dynamic_float6_t;
using ElementAMmaRaw = cutlass::detail::type_erased_dynamic_float6_unpacksmem_t;
using LayoutATag = cutlass::layout::ColumnMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementAMmaRaw>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto_small());
}
#endif // #if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)

136
test/unit/transform/device/sm100_sparse_gemm_compressor_f8.cu Normal file
View File

@@ -0,0 +1,136 @@
/***************************************************************************************************
* Copyright (c) 2024 - 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: BSD-3-Clause
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
**************************************************************************************************/
#include "cute/arch/mma_sm100_desc.hpp" // cute::UMMA::Major
#include "cutlass/gemm/collective/builders/sm100_common.inl" // tag_to_umma_major_A
#include "cutlass/gemm/collective/builders/sm1xx_sparse_config.inl" // Sm1xxGemmSparseConfig
#include "cutlass/transform/kernel/sparse_gemm_compressor.hpp" // StructuredSparseCompressor
#include "cutlass/transform/device/transform_universal_adapter.hpp" // TransformUniversalAdapter
#include "testbed_sparse_gemm_compressor.hpp" // TestbedSparseGemmCompressor
///////////////////////////////////////////////////////////////////////////////////////////////////
// * Test Plan
// ElementA : fp8
// LayoutA : row / col
// Gemm : 1x 2x 3x multiplier of alignment requirement. corner case that smaller than alignment requirement
///////////////////////////////////////////////////////////////////////////////////////////////////
#if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f8_t)
{
// Test Settings
using ElementA = cutlass::float_e4m3_t;
using LayoutATag = cutlass::layout::RowMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f8_n)
{
// Test Settings
using ElementA = cutlass::float_e5m2_t;
using LayoutATag = cutlass::layout::ColumnMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f8_runtimedtype_t)
{
// Test Settings
using ElementA = cutlass::type_erased_dynamic_float8_t;
using LayoutATag = cutlass::layout::RowMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto_small());
}
TEST(SM100_Structured_Sparse_Gemm_Compressor_Device, f8_runtimedtype_n)
{
// Test Settings
using ElementA = cutlass::type_erased_dynamic_float8_t;
using LayoutATag = cutlass::layout::ColumnMajor;
// Deduct From Test Setting
using ElementAMma = cute::sparse_elem<2, ElementA>;
using ElementEMma = cute::sparse_elem<8, uint8_t>;
using Sm1xxSparseConfig = cutlass::Sm1xxGemmSparseConfig<ElementAMma, LayoutATag, ElementEMma>;
using CompressorKernel = cutlass::transform::kernel::
StructuredSparseCompressor<cute::Shape<int, int, int, int>, ElementA, LayoutATag, Sm1xxSparseConfig, cutlass::arch::Sm100>;
using Compressor = cutlass::transform::device::TransformUniversalAdapter<CompressorKernel>;
// Test Bed
test::transform::device::TestbedSparseGemmCompressor<Compressor> testbed;
EXPECT_TRUE(testbed.run_auto_small());
}
#endif // #if defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)

10
test/unit/transform/device/testbed_sparse_gemm_compressor.hpp
View File

@@ -107,6 +107,10 @@ initialize_tensor(cutlass::TensorView<Element, Layout> view, cutlass::Distributi
scope_max = 2;
scope_min = 0;
}
else if (bits_input <= 6) {
scope_max = 2;
scope_min = -2;
}
else if (bits_input <= 8) {
scope_max = 1;
scope_min = -1;
@@ -155,9 +159,11 @@ public:
using ElementA = typename CompressorKernel::ElementA;
using LayoutATag = typename CompressorKernel::LayoutATag;
using StrideA = typename CompressorKernel::StrideA;
static constexpr bool IsRuntimeDataTypeA = cutlass::gemm::collective::detail::is_sm10x_runtime_f8f6f4<ElementA>();
using ArrayElementA =
ElementA
;
cute::conditional_t<IsRuntimeDataTypeA,
cute::uint_bit_t<cute::sizeof_bits_v<ElementA>>,
ElementA>;
using ElementE = typename CompressorKernel::ElementEMmaRaw;
using LayoutETag = cutlass::layout::RowMajor; // We don't care about the major here, just to allocate tensor