squall/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity
1d8372a8e2
cutlass/test/unit/gemm/warp/gemm_sm70.cu

290 lines
9.0 KiB
Plaintext
Raw Normal View History

CUTLASS 2.0 (#62) CUTLASS 2.0 Substantially refactored for - Better performance, particularly for native Turing Tensor Cores - Robust and durable templates spanning the design space - Encapsulated functionality embodying modern C++11 programming techniques - Optimized containers and data types for efficient, generic, portable device code Updates to: - Quick start guide - Documentation - Utilities - CUTLASS Profiler Native Turing Tensor Cores - Efficient GEMM kernels targeting Turing Tensor Cores - Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands Coverage of existing CUTLASS functionality: - GEMM kernels targeting CUDA and Tensor Cores in NVIDIA GPUs - Volta Tensor Cores through native mma.sync and through WMMA API - Optimizations such as parallel reductions, threadblock rasterization, and intra-threadblock reductions - Batched GEMM operations - Complex-valued GEMMs Note: this commit and all that follow require a host compiler supporting C++11 or greater.
2019-11-20 08:55:34 +08:00
/***************************************************************************************************
CUTLASS 2.5
2021-02-26 22:58:26 +08:00
* Copyright (c) 2017-2021, NVIDIA CORPORATION. All rights reserved.
CUTLASS 2.0 (#62) CUTLASS 2.0 Substantially refactored for - Better performance, particularly for native Turing Tensor Cores - Robust and durable templates spanning the design space - Encapsulated functionality embodying modern C++11 programming techniques - Optimized containers and data types for efficient, generic, portable device code Updates to: - Quick start guide - Documentation - Utilities - CUTLASS Profiler Native Turing Tensor Cores - Efficient GEMM kernels targeting Turing Tensor Cores - Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands Coverage of existing CUTLASS functionality: - GEMM kernels targeting CUDA and Tensor Cores in NVIDIA GPUs - Volta Tensor Cores through native mma.sync and through WMMA API - Optimizations such as parallel reductions, threadblock rasterization, and intra-threadblock reductions - Batched GEMM operations - Complex-valued GEMMs Note: this commit and all that follow require a host compiler supporting C++11 or greater.
2019-11-20 08:55:34 +08:00
*
* Redistribution and use in source and binary forms, with or without modification, are permitted
* provided that the following conditions are met:
* * Redistributions of source code must retain the above copyright notice, this list of
* conditions and the following disclaimer.
* * 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.
* * Neither the name of the NVIDIA CORPORATION 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 NVIDIA CORPORATION 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 TOR (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 Unit tests for thread-level GEMM
*/
#include "../../common/cutlass_unit_test.h"
#include "cutlass/aligned_buffer.h"
#include "cutlass/half.h"
#include "cutlass/gemm/warp/mma_tensor_op_sm70.h"
#include "cutlass/core_io.h"
#include "cutlass/util/host_tensor.h"
#include "cutlass/util/tensor_view_io.h"
#include "cutlass/util/reference/host/tensor_fill.h"
#include "cutlass/util/reference/host/tensor_compare.h"
#include "cutlass/util/reference/host/gemm.h"
#include "testbed.h"
#if defined(CUTLASS_ARCH_MMA_SM70_SUPPORTED)
/////////////////////////////////////////////////////////////////////////////////////////////////
TEST(SM70_warp_gemm_tensor_op_congruous, 128x128x16_64x64x16_16x16x4) {
using Shape = cutlass::gemm::GemmShape<64, 64, 16>;
using ElementA = cutlass::half_t;
using ElementB = cutlass::half_t;
using ElementC = cutlass::half_t;
using LayoutA = cutlass::layout::ColumnMajorVoltaTensorOpMultiplicandCongruous<cutlass::sizeof_bits<ElementA>::value>;
using LayoutB = cutlass::layout::RowMajorVoltaTensorOpMultiplicandBCongruous<cutlass::sizeof_bits<ElementB>::value>;
using Policy = cutlass::gemm::warp::MmaTensorOpPolicy<
cutlass::arch::Mma<
cutlass::gemm::GemmShape<16, 16, 4>,
32,
ElementA,
cutlass::layout::ColumnMajor,
ElementB,
cutlass::layout::RowMajor,
ElementC,
cutlass::layout::RowMajor,
cutlass::arch::OpMultiplyAdd
>,
cutlass::MatrixShape<1, 1>
>;
using MmaTensorOp = cutlass::gemm::warp::MmaVoltaTensorOp<
Shape,
ElementA,
LayoutA,
ElementB,
LayoutB,
ElementC,
cutlass::layout::RowMajor,
Policy
>;
test::gemm::warp::Testbed<MmaTensorOp, cutlass::gemm::GemmShape<128, 128, 16> >().run();
}
TEST(SM70_warp_gemm_tensor_op_congruous, 128x64x4_64x64x4_16x16x4) {
using Shape = cutlass::gemm::GemmShape<64, 64, 4>;
using ElementA = cutlass::half_t;
using ElementB = cutlass::half_t;
using ElementC = cutlass::half_t;
using LayoutA = cutlass::layout::ColumnMajorVoltaTensorOpMultiplicandCongruous<cutlass::sizeof_bits<ElementA>::value>;
using LayoutB = cutlass::layout::RowMajorVoltaTensorOpMultiplicandBCongruous<cutlass::sizeof_bits<ElementB>::value>;
using Policy = cutlass::gemm::warp::MmaTensorOpPolicy<
cutlass::arch::Mma<
cutlass::gemm::GemmShape<16, 16, 4>,
32,
ElementA,
cutlass::layout::ColumnMajor,
ElementB,
cutlass::layout::RowMajor,
ElementC,
cutlass::layout::RowMajor,
cutlass::arch::OpMultiplyAdd
>,
cutlass::MatrixShape<1, 1>
>;
using MmaTensorOp = cutlass::gemm::warp::MmaVoltaTensorOp<
Shape,
ElementA,
LayoutA,
ElementB,
LayoutB,
ElementC,
cutlass::layout::RowMajor,
Policy
>;
test::gemm::warp::Testbed<MmaTensorOp, cutlass::gemm::GemmShape<128, 64, 4> >().run();
}
TEST(SM70_warp_gemm_tensor_op_congruous, 128x128x4_32x32x4_16x16x4) {
using Shape = cutlass::gemm::GemmShape<32, 32, 4>;
using ElementA = cutlass::half_t;
using ElementB = cutlass::half_t;
using ElementC = cutlass::half_t;
using LayoutA = cutlass::layout::ColumnMajorVoltaTensorOpMultiplicandCongruous<cutlass::sizeof_bits<ElementA>::value>;
using LayoutB = cutlass::layout::RowMajorVoltaTensorOpMultiplicandBCongruous<cutlass::sizeof_bits<ElementB>::value>;
using Policy = cutlass::gemm::warp::MmaTensorOpPolicy<
cutlass::arch::Mma<
cutlass::gemm::GemmShape<16, 16, 4>,
32,
ElementA,
cutlass::layout::ColumnMajor,
ElementB,
cutlass::layout::RowMajor,
ElementC,
cutlass::layout::RowMajor,
cutlass::arch::OpMultiplyAdd
>,
cutlass::MatrixShape<1, 1>
>;
using MmaTensorOp = cutlass::gemm::warp::MmaVoltaTensorOp<
Shape,
ElementA,
LayoutA,
ElementB,
LayoutB,
ElementC,
cutlass::layout::RowMajor,
Policy
>;
test::gemm::warp::Testbed<MmaTensorOp, cutlass::gemm::GemmShape<128, 128, 4> >().run();
}
TEST(SM70_warp_gemm_tensor_op_crosswise, 64x64x32_64x64x32_16x16x4) {
using Shape = cutlass::gemm::GemmShape<64, 64, 32>;
using ElementA = cutlass::half_t;
using ElementB = cutlass::half_t;
using ElementC = cutlass::half_t;
using LayoutA = cutlass::layout::RowMajorVoltaTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<ElementA>::value, 32>;
using LayoutB = cutlass::layout::ColumnMajorVoltaTensorOpMultiplicandCrosswise<
cutlass::sizeof_bits<ElementB>::value, 32>;
using Policy = cutlass::gemm::warp::MmaTensorOpPolicy<
cutlass::arch::Mma<
cutlass::gemm::GemmShape<16, 16, 4>,
32,
ElementA,
cutlass::layout::RowMajor,
ElementB,
cutlass::layout::ColumnMajor,
ElementC,
cutlass::layout::RowMajor,
cutlass::arch::OpMultiplyAdd
>,
cutlass::MatrixShape<1, 1>
>;
using MmaTensorOp = cutlass::gemm::warp::MmaVoltaTensorOp<
Shape,
ElementA,
LayoutA,
ElementB,
LayoutB,
ElementC,
cutlass::layout::RowMajor,
Policy
>;
test::gemm::warp::Testbed<MmaTensorOp, cutlass::gemm::GemmShape<64, 64, 32> >().run();
}
CUTLASS 2.4 (Implicit GEMM convolution) (#147) CUTLASS 2.4 (Implicit GEMM Convolution) Co-authored-by: Manish Gupta <manigupta@nvidia.com>, Haicheng Wu <haichengw@nvidia.com>, Dustyn Blasig <dblasig@nvidia.com>, Andrew Kerr <akerr@nvidia.com>
2020-11-20 13:25:25 +08:00
////////////////////////////////////////////////////////////////////////////////
TEST(SM70_warp_gemm_volta_tensor_op_canonical_f32_row_col, 64x64x16_64x64x4_8x8x4) {
using Shape = cutlass::gemm::GemmShape<64, 64, 4>;
using InstructionShape = cutlass::gemm::GemmShape<8, 8, 4>;
using ElementA = cutlass::half_t;
using ElementB = cutlass::half_t;
using ElementC = float;
using LayoutA = cutlass::layout::RowMajor;
using LayoutB = cutlass::layout::ColumnMajor;
using Policy = cutlass::gemm::warp::MmaTensorOpPolicy<
cutlass::arch::Mma<
cutlass::gemm::GemmShape<16, 16, 4>,
32,
ElementA,
cutlass::layout::RowMajor,
ElementB,
cutlass::layout::ColumnMajor,
ElementC,
cutlass::layout::RowMajor,
cutlass::arch::OpMultiplyAdd
>,
cutlass::MatrixShape<1, 1>
>;
using MmaTensorOp = cutlass::gemm::warp::MmaVoltaTensorOp<
Shape,
ElementA,
LayoutA,
ElementB,
LayoutB,
ElementC,
cutlass::layout::RowMajor,
Policy
>;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<64, 64, 16> >()
.run();
}
TEST(SM70_warp_gemm_volta_tensor_op_canonical_f32_col_row, 64x64x16_64x64x4_8x8x4) {
using Shape = cutlass::gemm::GemmShape<64, 64, 4>;
using InstructionShape = cutlass::gemm::GemmShape<8, 8, 4>;
using ElementA = cutlass::half_t;
using ElementB = cutlass::half_t;
using ElementC = float;
using LayoutA = cutlass::layout::ColumnMajor;
using LayoutB = cutlass::layout::RowMajor;
using Policy = cutlass::gemm::warp::MmaTensorOpPolicy<
cutlass::arch::Mma<
cutlass::gemm::GemmShape<16, 16, 4>,
32,
ElementA,
LayoutA,
ElementB,
LayoutB,
ElementC,
cutlass::layout::RowMajor,
cutlass::arch::OpMultiplyAdd
>,
cutlass::MatrixShape<1, 1>
>;
using MmaTensorOp = cutlass::gemm::warp::MmaVoltaTensorOp<
Shape,
ElementA,
LayoutA,
ElementB,
LayoutB,
ElementC,
cutlass::layout::RowMajor,
Policy
>;
test::gemm::warp::Testbed<MmaTensorOp,
cutlass::gemm::GemmShape<64, 64, 16> >()
.run();
}
CUTLASS 2.0 (#62) CUTLASS 2.0 Substantially refactored for - Better performance, particularly for native Turing Tensor Cores - Robust and durable templates spanning the design space - Encapsulated functionality embodying modern C++11 programming techniques - Optimized containers and data types for efficient, generic, portable device code Updates to: - Quick start guide - Documentation - Utilities - CUTLASS Profiler Native Turing Tensor Cores - Efficient GEMM kernels targeting Turing Tensor Cores - Mixed-precision floating point, 8-bit integer, 4-bit integer, and binarized operands Coverage of existing CUTLASS functionality: - GEMM kernels targeting CUDA and Tensor Cores in NVIDIA GPUs - Volta Tensor Cores through native mma.sync and through WMMA API - Optimizations such as parallel reductions, threadblock rasterization, and intra-threadblock reductions - Batched GEMM operations - Complex-valued GEMMs Note: this commit and all that follow require a host compiler supporting C++11 or greater.
2019-11-20 08:55:34 +08:00
/////////////////////////////////////////////////////////////////////////////////////////////////
#endif // CUTLASS_ARCH_MMA_SM70_SUPPORTED
Reference in New Issue Copy Permalink