877bdcace6
CUTLASS 1.3 Release - Efficient GEMM kernel targeting Volta Tensor Cores via mma.sync instruction added in CUDA 10.1.
178 lines
6.8 KiB
Plaintext
178 lines
6.8 KiB
Plaintext
/***************************************************************************************************
|
|
* Copyright (c) 2017-2019, NVIDIA CORPORATION. All rights reserved.
|
|
*
|
|
* 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.
|
|
*
|
|
**************************************************************************************************/
|
|
#include "cutlass_unit_test.h"
|
|
#include "cutlass/gemm/gemm.h"
|
|
#include "cutlass/gemm/fp16_sgemm_traits.h"
|
|
#include "tools/test/unit/gemm/gemm_testbed.h"
|
|
#include "tools/test/unit/gemm/run_gemm.h"
|
|
|
|
#if !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 530
|
|
|
|
|
|
TEST(Fp16_sgemm_alphaFp32_fp32_128x128x16, fp16_sgemm_fp32_128x128x16_nn) {
|
|
/*A, B are half typed, accumulator is always float for sgemm, C, D are float typed
|
|
alpha and beta are both fp32*/
|
|
typedef cutlass::gemm::Fp16SgemmSgemmTraits<cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::Shape<16, 128, 128>,
|
|
half, /*A type*/
|
|
half, /*B type*/
|
|
float, /*C type*/
|
|
float, /*D type*/
|
|
float /*alpha, beta type*/
|
|
>
|
|
SgemmTraits;
|
|
run_gemm<SgemmTraits>(128, 128, 16);
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
TEST(Fp16_sgemm_alphaFp32_fp32_128x128x16, fp16_sgemm_fp32_128x128x16_nt) {
|
|
/*A, B are half typed, accumulator is always float for sgemm, C, D are float typed
|
|
alpha and beta are both fp32*/
|
|
typedef cutlass::gemm::Fp16SgemmSgemmTraits<cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::Shape<16, 128, 128>,
|
|
half, /*A type*/
|
|
half, /*B type*/
|
|
float, /*C type*/
|
|
float, /*D type*/
|
|
float /*alpha, beta type*/
|
|
>
|
|
SgemmTraits;
|
|
run_gemm<SgemmTraits>(128, 128, 16);
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
TEST(Fp16_sgemm_alphaFp32_fp32_128x128x16, fp16_sgemm_fp32_128x128x16_tn) {
|
|
/*A, B are half typed, accumulator is always float for sgemm, C, D are float typed
|
|
alpha and beta are both fp32*/
|
|
typedef cutlass::gemm::Fp16SgemmSgemmTraits<cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::Shape<16, 128, 128>,
|
|
half, /*A type*/
|
|
half, /*B type*/
|
|
float, /*C type*/
|
|
float, /*D type*/
|
|
float /*alpha, beta type*/
|
|
>
|
|
SgemmTraits;
|
|
run_gemm<SgemmTraits>(128, 128, 16);
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
TEST(Fp16_sgemm_alphaFp32_fp32_128x128x16, fp16_sgemm_fp32_128x128x16_tt) {
|
|
/*A, B are half typed, accumulator is always float for sgemm, C, D are float typed
|
|
alpha and beta are both fp32*/
|
|
typedef cutlass::gemm::Fp16SgemmSgemmTraits<cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::Shape<16, 128, 128>,
|
|
half, /*A type*/
|
|
half, /*B type*/
|
|
float, /*C type*/
|
|
float, /*D type*/
|
|
float /*alpha, beta type*/
|
|
>
|
|
SgemmTraits;
|
|
run_gemm<SgemmTraits>(128, 128, 16);
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
TEST(Fp16_sgemm_alphaFp32_fp32_128x128x16, fp16_sgemm_fp32_128x112x17_nn) {
|
|
/*A, B are half typed, accumulator is always float for sgemm, C, D are float typed
|
|
alpha and beta are both fp32*/
|
|
typedef cutlass::gemm::Fp16SgemmSgemmTraits<cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::Shape<16, 128, 128>,
|
|
half, /*A type*/
|
|
half, /*B type*/
|
|
float, /*C type*/
|
|
float, /*D type*/
|
|
float /*alpha, beta type*/
|
|
>
|
|
SgemmTraits;
|
|
run_gemm<SgemmTraits>(128, 112, 17);
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
TEST(Fp16_sgemm_alphaFp32_fp32_128x128x16, fp16_sgemm_fp32_128x112x17_nt) {
|
|
/*A, B are half typed, accumulator is always float for sgemm, C, D are float typed
|
|
alpha and beta are both fp32*/
|
|
typedef cutlass::gemm::Fp16SgemmSgemmTraits<cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::Shape<16, 128, 128>,
|
|
half, /*A type*/
|
|
half, /*B type*/
|
|
float, /*C type*/
|
|
float, /*D type*/
|
|
float /*alpha, beta type*/
|
|
>
|
|
SgemmTraits;
|
|
run_gemm<SgemmTraits>(128, 112, 17);
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
TEST(Fp16_sgemm_alphaFp32_fp32_128x128x16, fp16_sgemm_fp32_128x112x17_tn) {
|
|
/*A, B are half typed, accumulator is always float for sgemm, C, D are float typed
|
|
alpha and beta are both fp32*/
|
|
typedef cutlass::gemm::Fp16SgemmSgemmTraits<cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::Shape<16, 128, 128>,
|
|
half, /*A type*/
|
|
half, /*B type*/
|
|
float, /*C type*/
|
|
float, /*D type*/
|
|
float /*alpha, beta type*/
|
|
>
|
|
SgemmTraits;
|
|
run_gemm<SgemmTraits>(128, 112, 17);
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
TEST(Fp16_sgemm_alphaFp32_fp32_128x128x16, fp16_sgemm_fp32_128x112x17_tt) {
|
|
/*A, B are half typed, accumulator is always float for sgemm, C, D are float typed
|
|
alpha and beta are both fp32*/
|
|
typedef cutlass::gemm::Fp16SgemmSgemmTraits<cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::Shape<16, 128, 128>,
|
|
half, /*A type*/
|
|
half, /*B type*/
|
|
float, /*C type*/
|
|
float, /*D type*/
|
|
float /*alpha, beta type*/
|
|
>
|
|
SgemmTraits;
|
|
run_gemm<SgemmTraits>(128, 112, 17);
|
|
}
|
|
|
|
#endif
|
|
|