v3.8.0 update (#2082)
* 3.8 update * fix Markus' name --------- Co-authored-by: yuzhai <yuzhai@nvidia.com>
This commit is contained in:
Yujia Zhai
@@ -276,11 +276,12 @@ endif()
|
||||
|
||||
if (CUTLASS_NVCC_MAX_ARCH GREATER_EQUAL 89)
|
||||
|
||||
# Conv - F8 input, F8 output, F32 accumulation
|
||||
# Conv - F8 input, F8 output
|
||||
cutlass_test_unit_add_executable(
|
||||
cutlass_test_unit_conv_device_tensorop_f8_sm89
|
||||
|
||||
conv2d_fprop_implicit_gemm_f8nhwc_f8nhwc_f8nhwc_tensor_op_f32_sm89.cu
|
||||
conv2d_fprop_implicit_gemm_f8nhwc_f8nhwc_f8nhwc_tensor_op_f16_sm89.cu
|
||||
)
|
||||
|
||||
endif()
|
||||
|
||||
+236
@@ -0,0 +1,236 @@
|
||||
/***************************************************************************************************
|
||||
* Copyright (c) 2025 - 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.
|
||||
*
|
||||
**************************************************************************************************/
|
||||
|
||||
/*! \file
|
||||
\brief Tests for device-wide Conv2d fprop interface with:
|
||||
A: NHWC, of type FE4M4 or FE5M2
|
||||
B: NHWC, of type FE4M3 or FE5M2
|
||||
C: NHWC, of FE4M3 or FE5M2
|
||||
Accum: F16
|
||||
*/
|
||||
|
||||
#include <iostream>
|
||||
|
||||
#include "../../common/cutlass_unit_test.h"
|
||||
#include "cutlass/cutlass.h"
|
||||
#include "cutlass/epilogue/thread/activation.h"
|
||||
#include "cutlass/epilogue/thread/linear_combination_generic_with_scaling.h"
|
||||
#include "cutlass/conv/kernel/default_conv2d_fprop_with_absmax.h"
|
||||
#include "cutlass/conv/device/implicit_gemm_convolution.h"
|
||||
#include "cutlass/util/tensor_view_io.h"
|
||||
|
||||
#include "conv2d_with_absmax_testbed.h"
|
||||
|
||||
#if defined(CUTLASS_ARCH_MMA_F16_SM89_SUPPORTED)
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
TEST(SM89_Device_Conv2d_Fprop_Analytic_ImplicitGemm_fe4m3nhwc_fe4mnhwc_fe4mnhwc_tensor_op_f16,
|
||||
identity_128x256x64_64x3_64x64x64) {
|
||||
|
||||
using ElementA = cutlass::float_e4m3_t;
|
||||
using ElementB = cutlass::float_e4m3_t;
|
||||
using ElementOutput = cutlass::float_e4m3_t;
|
||||
using ElementAuxOutput = ElementOutput;
|
||||
using ElementAccumulator = cutlass::half_t;
|
||||
static int const kStages = 3;
|
||||
|
||||
using EpilogueOutputOp = cutlass::epilogue::thread::LinearCombinationGenericWithScalingAndAbsMax<
|
||||
cutlass::epilogue::thread::Identity,
|
||||
ElementOutput,
|
||||
ElementAuxOutput,
|
||||
128 / cutlass::sizeof_bits<ElementOutput>::value,
|
||||
ElementAccumulator,
|
||||
ElementAccumulator
|
||||
>;
|
||||
|
||||
using Conv2dFpropKernel = typename cutlass::conv::kernel::DefaultConv2dFpropWithAbsMax<
|
||||
ElementA, cutlass::layout::TensorNHWC,
|
||||
ElementB, cutlass::layout::TensorNHWC,
|
||||
ElementOutput, cutlass::layout::TensorNHWC,
|
||||
ElementAccumulator,
|
||||
cutlass::arch::OpClassTensorOp,
|
||||
cutlass::arch::Sm89,
|
||||
cutlass::gemm::GemmShape<128, 256, 64>,
|
||||
cutlass::gemm::GemmShape<64, 64, 64>,
|
||||
cutlass::gemm::GemmShape<16, 8, 32>,
|
||||
EpilogueOutputOp,
|
||||
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
|
||||
kStages,
|
||||
cutlass::arch::OpMultiplyAdd,
|
||||
cutlass::conv::IteratorAlgorithm::kAnalytic
|
||||
>::Kernel;
|
||||
|
||||
using Conv2dFprop = cutlass::conv::device::ImplicitGemmConvolution<Conv2dFpropKernel>;
|
||||
|
||||
bool passed = test::conv::device::TestAllConv2dWithAbsmax<Conv2dFprop, cutlass::epilogue::thread::Identity>();
|
||||
EXPECT_TRUE(passed);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
TEST(SM89_Device_Conv2d_Fprop_Optimized_ImplicitGemm_fe4m3nhwc_fe4mnhwc_fe4mnhwc_tensor_op_f16,
|
||||
relu_128x256x64_64x3_64x64x64) {
|
||||
|
||||
using ElementA = cutlass::float_e4m3_t;
|
||||
using ElementB = cutlass::float_e4m3_t;
|
||||
using ElementOutput = cutlass::float_e4m3_t;
|
||||
using ElementAuxOutput = ElementOutput;
|
||||
using ElementAccumulator = cutlass::half_t;
|
||||
static int const kStages = 3;
|
||||
|
||||
using EpilogueOutputOp = cutlass::epilogue::thread::LinearCombinationGenericWithScalingAndAbsMax<
|
||||
cutlass::epilogue::thread::ReLu,
|
||||
ElementOutput,
|
||||
ElementAuxOutput,
|
||||
128 / cutlass::sizeof_bits<ElementOutput>::value,
|
||||
ElementAccumulator,
|
||||
ElementAccumulator
|
||||
>;
|
||||
|
||||
using Conv2dFpropKernel = typename cutlass::conv::kernel::DefaultConv2dFpropWithAbsMax<
|
||||
ElementA, cutlass::layout::TensorNHWC,
|
||||
ElementB, cutlass::layout::TensorNHWC,
|
||||
ElementOutput, cutlass::layout::TensorNHWC,
|
||||
ElementAccumulator,
|
||||
cutlass::arch::OpClassTensorOp,
|
||||
cutlass::arch::Sm89,
|
||||
cutlass::gemm::GemmShape<128, 256, 64>,
|
||||
cutlass::gemm::GemmShape<64, 64, 64>,
|
||||
cutlass::gemm::GemmShape<16, 8, 32>,
|
||||
EpilogueOutputOp,
|
||||
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
|
||||
kStages,
|
||||
cutlass::arch::OpMultiplyAdd,
|
||||
cutlass::conv::IteratorAlgorithm::kOptimized
|
||||
>::Kernel;
|
||||
|
||||
using Conv2dFprop = cutlass::conv::device::ImplicitGemmConvolution<Conv2dFpropKernel>;
|
||||
|
||||
bool passed = test::conv::device::TestAllConv2dWithAbsmax<Conv2dFprop, cutlass::epilogue::thread::ReLu>();
|
||||
EXPECT_TRUE(passed);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
TEST(SM89_Device_Conv2d_Fprop_Optimized_ImplicitGemm_fe4m3nhwc_fe4mnhwc_fe4mnhwc_tensor_op_f16,
|
||||
identity_fastacc_128x256x64_64x3_64x64x64) {
|
||||
|
||||
using ElementA = cutlass::float_e4m3_t;
|
||||
using ElementB = cutlass::float_e4m3_t;
|
||||
using ElementOutput = cutlass::float_e4m3_t;
|
||||
using ElementAuxOutput = ElementOutput;
|
||||
using ElementAccumulator = cutlass::half_t;
|
||||
static int const kStages = 3;
|
||||
|
||||
using EpilogueOutputOp = cutlass::epilogue::thread::LinearCombinationGenericWithScalingAndAbsMax<
|
||||
cutlass::epilogue::thread::Identity,
|
||||
ElementOutput,
|
||||
ElementAuxOutput,
|
||||
128 / cutlass::sizeof_bits<ElementOutput>::value,
|
||||
ElementAccumulator,
|
||||
ElementAccumulator
|
||||
>;
|
||||
|
||||
using Conv2dFpropKernel = typename cutlass::conv::kernel::DefaultConv2dFpropWithAbsMax<
|
||||
ElementA, cutlass::layout::TensorNHWC,
|
||||
ElementB, cutlass::layout::TensorNHWC,
|
||||
ElementOutput, cutlass::layout::TensorNHWC,
|
||||
ElementAccumulator,
|
||||
cutlass::arch::OpClassTensorOp,
|
||||
cutlass::arch::Sm89,
|
||||
cutlass::gemm::GemmShape<128, 256, 64>,
|
||||
cutlass::gemm::GemmShape<64, 64, 64>,
|
||||
cutlass::gemm::GemmShape<16, 8, 32>,
|
||||
EpilogueOutputOp,
|
||||
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
|
||||
kStages,
|
||||
cutlass::arch::OpMultiplyAddFastAccum,
|
||||
cutlass::conv::IteratorAlgorithm::kOptimized
|
||||
>::Kernel;
|
||||
|
||||
using Conv2dFprop = cutlass::conv::device::ImplicitGemmConvolution<Conv2dFpropKernel>;
|
||||
|
||||
bool passed = test::conv::device::TestAllConv2dWithAbsmax<Conv2dFprop, cutlass::epilogue::thread::Identity>();
|
||||
EXPECT_TRUE(passed);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
TEST(SM89_Device_Conv2d_Fprop_Optimized_ImplicitGemm_fe4m3nhwc_fe4mnhwc_fe4mnhwc_tensor_op_f16,
|
||||
identity_noScale_128x256x64_64x3_64x64x64) {
|
||||
|
||||
using ElementA = cutlass::float_e4m3_t;
|
||||
using ElementB = cutlass::float_e4m3_t;
|
||||
using ElementOutput = cutlass::float_e4m3_t;
|
||||
using ElementAuxOutput = ElementOutput;
|
||||
using ElementAccumulator = cutlass::half_t;
|
||||
static int const kStages = 3;
|
||||
|
||||
using EpilogueOutputOp = cutlass::epilogue::thread::LinearCombinationGenericWithScalingAndAbsMax<
|
||||
cutlass::epilogue::thread::Identity,
|
||||
ElementOutput,
|
||||
ElementAuxOutput,
|
||||
128 / cutlass::sizeof_bits<ElementOutput>::value,
|
||||
ElementAccumulator,
|
||||
ElementAccumulator
|
||||
>;
|
||||
|
||||
using Conv2dFpropKernel = typename cutlass::conv::kernel::DefaultConv2dFpropWithAbsMax<
|
||||
ElementA, cutlass::layout::TensorNHWC,
|
||||
ElementB, cutlass::layout::TensorNHWC,
|
||||
ElementOutput, cutlass::layout::TensorNHWC,
|
||||
ElementAccumulator,
|
||||
cutlass::arch::OpClassTensorOp,
|
||||
cutlass::arch::Sm89,
|
||||
cutlass::gemm::GemmShape<128, 256, 64>,
|
||||
cutlass::gemm::GemmShape<64, 64, 64>,
|
||||
cutlass::gemm::GemmShape<16, 8, 32>,
|
||||
EpilogueOutputOp,
|
||||
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<>,
|
||||
kStages,
|
||||
cutlass::arch::OpMultiplyAdd,
|
||||
cutlass::conv::IteratorAlgorithm::kOptimized
|
||||
>::Kernel;
|
||||
|
||||
using Conv2dFprop = cutlass::conv::device::ImplicitGemmConvolution<Conv2dFpropKernel>;
|
||||
|
||||
bool passed = test::conv::device::TestAllConv2dWithAbsmax<Conv2dFprop, cutlass::epilogue::thread::Identity>(
|
||||
/* scaleA = */false,
|
||||
/* scaleB = */false,
|
||||
/* scaleC = */false
|
||||
);
|
||||
EXPECT_TRUE(passed);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
#endif // CUTLASS_ARCH_MMA_F16_SM89_SUPPORTED
|
||||
+2
-2
@@ -49,7 +49,7 @@
|
||||
|
||||
#include "conv2d_with_absmax_testbed.h"
|
||||
|
||||
#if defined(CUTLASS_ARCH_MMA_SM89_SUPPORTED)
|
||||
#if defined(CUTLASS_ARCH_MMA_F32_SM89_SUPPORTED)
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
@@ -365,4 +365,4 @@ TEST(SM89_Device_Conv2d_Fprop_Optimized_ImplicitGemm_fe4m3nhwc_fe4mnhwc_fe4mnhwc
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
#endif // CUTLASS_ARCH_MMA_SM89_SUPPORTED
|
||||
#endif // CUTLASS_ARCH_MMA_F32_SM89_SUPPORTED
|
||||
|
||||
Reference in New Issue
Block a user