877bdcace6
CUTLASS 1.3 Release - Efficient GEMM kernel targeting Volta Tensor Cores via mma.sync instruction added in CUDA 10.1.
487 lines
20 KiB
Plaintext
487 lines
20 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 "tools/test/perf/cutlass_perf_test.h"
|
|
#include "tools/test/perf/gemm/gemm_perf_testbed.h"
|
|
#include "tools/test/perf/gemm/gemm_profiler.h"
|
|
|
|
#include "cutlass/wmma_matrix.h"
|
|
#ifdef CUTLASS_USE_WMMA_API
|
|
#ifdef CUTLASS_USE_INT_WMMA
|
|
#pragma warning( disable : 4503)
|
|
#include "cutlass/gemm/gemm.h"
|
|
#include "cutlass/gemm/wmma_gemm_traits.h"
|
|
#include "tools/test/perf/gemm/cutlass_dispatch.h"
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
template<typename Traits, typename ScalarA, typename ScalarB>
|
|
struct WmmaIntegerGemmDispatch {
|
|
|
|
typedef cutlass::gemm::Gemm<Traits> Gemm;
|
|
|
|
typedef typename Gemm::Params Params;
|
|
|
|
/// Indicate warp-level GEMM
|
|
static bool const kThreadMultiplyAdd = false;
|
|
|
|
static bool const kRunCuBLAS = false;
|
|
|
|
static cutlass::MatrixLayout::Kind const kLayoutA = Traits::kLayoutA;
|
|
static cutlass::MatrixLayout::Kind const kLayoutB = Traits::kLayoutB;
|
|
|
|
//
|
|
// Data members
|
|
//
|
|
|
|
/// Params argument
|
|
Params params;
|
|
|
|
//
|
|
// Methods
|
|
//
|
|
|
|
WmmaIntegerGemmDispatch() {}
|
|
|
|
/// Initializes params object
|
|
WmmaIntegerGemmDispatch(int m, int n, int k, int alpha,
|
|
ScalarA const* d_a, int lda,
|
|
ScalarB const* d_b, int ldb, int beta,
|
|
int const* d_c, int ldc, int* d_d, int ldd) {
|
|
|
|
params.initialize(m, n, k, alpha, d_a, lda, d_b, ldb, beta, d_c, ldc, d_d, ldd);
|
|
}
|
|
|
|
///
|
|
WmmaIntegerGemmDispatch(int m, int n, int k, int alpha,
|
|
ScalarA const* d_a, int lda, long long int batch_stride_a,
|
|
ScalarB const* d_b, int ldb, long long int batch_stride_b, int beta,
|
|
int const* d_c, int ldc, long long int batch_stride_c, int* d_d, int ldd, long long int batch_stride_d,
|
|
int batch_count) {
|
|
assert(0);
|
|
}
|
|
|
|
/// Initializes params object
|
|
WmmaIntegerGemmDispatch(Params const& _params) : params(_params) {}
|
|
|
|
/// Launches kernel
|
|
cudaError_t operator()() { return Gemm::launch(params); }
|
|
};
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
#ifdef CUTLASS_USE_SUBBYTE_WMMA
|
|
template<typename Traits>
|
|
struct WmmaIntegerGemmDispatch<Traits,
|
|
cutlass::Vector<cutlass::int4_t, 8>,
|
|
cutlass::Vector<cutlass::int4_t, 8> > {
|
|
|
|
typedef typename cutlass::Vector<cutlass::int4_t, 8> ScalarA;
|
|
typedef typename cutlass::Vector<cutlass::int4_t, 8> ScalarB;
|
|
|
|
typedef cutlass::gemm::Gemm<Traits> Gemm;
|
|
|
|
typedef typename Gemm::Params Params;
|
|
|
|
/// Indicate warp-level GEMM
|
|
static bool const kThreadMultiplyAdd = false;
|
|
|
|
static bool const kRunCuBLAS = false;
|
|
|
|
static cutlass::MatrixLayout::Kind const kLayoutA = Traits::kLayoutA;
|
|
static cutlass::MatrixLayout::Kind const kLayoutB = Traits::kLayoutB;
|
|
|
|
//
|
|
// Data members
|
|
//
|
|
|
|
/// Params argument
|
|
Params params;
|
|
|
|
//
|
|
// Methods
|
|
//
|
|
|
|
WmmaIntegerGemmDispatch() {}
|
|
|
|
/// Initializes params object
|
|
WmmaIntegerGemmDispatch(int m, int n, int k, int alpha,
|
|
ScalarA const* d_a, int lda,
|
|
ScalarB const* d_b, int ldb, int beta,
|
|
int const* d_c, int ldc, int* d_d, int ldd) {
|
|
|
|
params.initialize(m, n, k * 8, alpha, d_a, lda, d_b, ldb, beta, d_c, ldc, d_d, ldd);
|
|
}
|
|
|
|
///
|
|
WmmaIntegerGemmDispatch(int m, int n, int k, int alpha,
|
|
ScalarA const* d_a, int lda, long long int batch_stride_a,
|
|
ScalarB const* d_b, int ldb, long long int batch_stride_b, int beta,
|
|
int const* d_c, int ldc, long long int batch_stride_c, int* d_d, int ldd, long long int batch_stride_d,
|
|
int batch_count) {
|
|
assert(0);
|
|
}
|
|
|
|
/// Initializes params object
|
|
WmmaIntegerGemmDispatch(Params const& _params) : params(_params) {}
|
|
|
|
/// Launches kernel
|
|
cudaError_t operator()() { return Gemm::launch(params); }
|
|
};
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
template<typename Traits>
|
|
struct WmmaIntegerGemmDispatch<Traits,
|
|
cutlass::Vector<cutlass::uint4_t, 8>,
|
|
cutlass::Vector<cutlass::uint4_t, 8> > {
|
|
|
|
typedef typename cutlass::Vector<cutlass::uint4_t, 8> ScalarA;
|
|
typedef typename cutlass::Vector<cutlass::uint4_t, 8> ScalarB;
|
|
|
|
typedef cutlass::gemm::Gemm<Traits> Gemm;
|
|
|
|
typedef typename Gemm::Params Params;
|
|
|
|
/// Indicate warp-level GEMM
|
|
static bool const kThreadMultiplyAdd = false;
|
|
|
|
static bool const kRunCuBLAS = false;
|
|
|
|
static cutlass::MatrixLayout::Kind const kLayoutA = Traits::kLayoutA;
|
|
static cutlass::MatrixLayout::Kind const kLayoutB = Traits::kLayoutB;
|
|
|
|
//
|
|
// Data members
|
|
//
|
|
|
|
/// Params argument
|
|
Params params;
|
|
|
|
//
|
|
// Methods
|
|
//
|
|
|
|
WmmaIntegerGemmDispatch() {}
|
|
|
|
/// Initializes params object
|
|
WmmaIntegerGemmDispatch(int m, int n, int k, int alpha,
|
|
ScalarA const* d_a, int lda,
|
|
ScalarB const* d_b, int ldb, int beta,
|
|
int const* d_c, int ldc, int* d_d, int ldd) {
|
|
|
|
params.initialize(m, n, k * 8, alpha, d_a, lda, d_b, ldb, beta, d_c, ldc, d_d, ldd);
|
|
}
|
|
|
|
///
|
|
WmmaIntegerGemmDispatch(int m, int n, int k, int alpha,
|
|
ScalarA const* d_a, int lda, long long int batch_stride_a,
|
|
ScalarB const* d_b, int ldb, long long int batch_stride_b, int beta,
|
|
int const* d_c, int ldc, long long int batch_stride_c, int* d_d, int ldd, long long int batch_stride_d,
|
|
int batch_count) {
|
|
assert(0);
|
|
}
|
|
|
|
/// Initializes params object
|
|
WmmaIntegerGemmDispatch(Params const& _params) : params(_params) {}
|
|
|
|
/// Launches kernel
|
|
cudaError_t operator()() { return Gemm::launch(params); }
|
|
};
|
|
#endif //ifdef CUTLASS_USE_SUBBYTE_WMMA
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
namespace perf {
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
int profile_wmma_integer_gemm(TestbenchOutput<GemmProblem> &output, TestbenchOptions const &options, Config const &config) {
|
|
|
|
int results = 0;
|
|
|
|
// compute capability check
|
|
if (!options.compute_capability(7, 2)) {
|
|
return 0;
|
|
}
|
|
|
|
{
|
|
typedef cutlass::gemm::WmmaGemmTraits<cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::Shape<128, 128, 128>,
|
|
signed char,
|
|
signed char,
|
|
int,
|
|
cutlass::gemm::LinearScaling<int>,
|
|
int,
|
|
cutlass::Shape<128, 32, 32>,
|
|
cutlass::Shape<16, 16, 16>,
|
|
16,
|
|
16> WmmaGemmTraits;
|
|
|
|
typedef WmmaIntegerGemmDispatch<WmmaGemmTraits, signed char, signed char> Dispatch;
|
|
|
|
typedef perf::GemmProfiler<signed char, signed char, int, int, int> GemmProfiler;
|
|
|
|
results |= profile_gemm<Dispatch, GemmProfiler>(output, "wmma_integer_gemm_s8_16x16x16_nn", options, config);
|
|
}
|
|
|
|
{
|
|
typedef cutlass::gemm::WmmaGemmTraits<cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::Shape<128, 128, 128>,
|
|
signed char,
|
|
signed char,
|
|
int,
|
|
cutlass::gemm::LinearScaling<int>,
|
|
int,
|
|
cutlass::Shape<128, 32, 32>,
|
|
cutlass::Shape<16, 16, 16>,
|
|
16,
|
|
16> WmmaGemmTraits;
|
|
|
|
typedef WmmaIntegerGemmDispatch<WmmaGemmTraits, signed char, signed char> Dispatch;
|
|
|
|
typedef perf::GemmProfiler<signed char, signed char, int, int, int> GemmProfiler;
|
|
|
|
results |= profile_gemm<Dispatch, GemmProfiler>(output, "wmma_integer_gemm_s8_16x16x16_nt", options, config);
|
|
}
|
|
|
|
{
|
|
typedef cutlass::gemm::WmmaGemmTraits<cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::Shape<128, 128, 128>,
|
|
signed char,
|
|
signed char,
|
|
int,
|
|
cutlass::gemm::LinearScaling<int>,
|
|
int,
|
|
cutlass::Shape<128, 32, 32>,
|
|
cutlass::Shape<16, 16, 16>,
|
|
16,
|
|
16> WmmaGemmTraits;
|
|
|
|
typedef WmmaIntegerGemmDispatch<WmmaGemmTraits, signed char, signed char> Dispatch;
|
|
|
|
typedef perf::GemmProfiler<signed char, signed char, int, int, int> GemmProfiler;
|
|
|
|
results |= profile_gemm<Dispatch, GemmProfiler>(output, "wmma_integer_gemm_s8_16x16x16_tn", options, config);
|
|
}
|
|
|
|
{
|
|
typedef cutlass::gemm::WmmaGemmTraits<cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::Shape<128, 128, 128>,
|
|
signed char,
|
|
signed char,
|
|
int,
|
|
cutlass::gemm::LinearScaling<int>,
|
|
int,
|
|
cutlass::Shape<128, 32, 32>,
|
|
cutlass::Shape<16, 16, 16>,
|
|
16,
|
|
16> WmmaGemmTraits;
|
|
|
|
typedef WmmaIntegerGemmDispatch<WmmaGemmTraits, signed char, signed char> Dispatch;
|
|
|
|
typedef perf::GemmProfiler<signed char, signed char, int, int, int> GemmProfiler;
|
|
|
|
results |= profile_gemm<Dispatch, GemmProfiler>(output, "wmma_integer_gemm_s8_16x16x16_tt", options, config);
|
|
}
|
|
|
|
{
|
|
typedef cutlass::gemm::WmmaGemmTraits<cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::Shape<128, 128, 128>,
|
|
unsigned char,
|
|
unsigned char,
|
|
int,
|
|
cutlass::gemm::LinearScaling<int>,
|
|
int,
|
|
cutlass::Shape<128, 32, 32>,
|
|
cutlass::Shape<16, 16, 16>,
|
|
16,
|
|
16> WmmaGemmTraits;
|
|
|
|
typedef WmmaIntegerGemmDispatch<WmmaGemmTraits, unsigned char, unsigned char> Dispatch;
|
|
|
|
typedef perf::GemmProfiler<unsigned char, unsigned char, int, int, int> GemmProfiler;
|
|
|
|
results |= profile_gemm<Dispatch, GemmProfiler>(output, "wmma_integer_gemm_u8_16x16x16_nn", options, config);
|
|
}
|
|
|
|
{
|
|
typedef cutlass::gemm::WmmaGemmTraits<cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::Shape<128, 128, 128>,
|
|
unsigned char,
|
|
unsigned char,
|
|
int,
|
|
cutlass::gemm::LinearScaling<int>,
|
|
int,
|
|
cutlass::Shape<128, 32, 32>,
|
|
cutlass::Shape<16, 16, 16>,
|
|
16,
|
|
16> WmmaGemmTraits;
|
|
|
|
typedef WmmaIntegerGemmDispatch<WmmaGemmTraits, unsigned char, unsigned char> Dispatch;
|
|
|
|
typedef perf::GemmProfiler<unsigned char, unsigned char, int, int, int> GemmProfiler;
|
|
|
|
results |= profile_gemm<Dispatch, GemmProfiler>(output, "wmma_integer_gemm_u8_16x16x16_nt", options, config);
|
|
}
|
|
|
|
{
|
|
typedef cutlass::gemm::WmmaGemmTraits<cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::Shape<128, 128, 128>,
|
|
unsigned char,
|
|
unsigned char,
|
|
int,
|
|
cutlass::gemm::LinearScaling<int>,
|
|
int,
|
|
cutlass::Shape<128, 32, 32>,
|
|
cutlass::Shape<16, 16, 16>,
|
|
16,
|
|
16> WmmaGemmTraits;
|
|
|
|
typedef WmmaIntegerGemmDispatch<WmmaGemmTraits, unsigned char, unsigned char> Dispatch;
|
|
|
|
typedef perf::GemmProfiler<unsigned char, unsigned char, int, int, int> GemmProfiler;
|
|
|
|
results |= profile_gemm<Dispatch, GemmProfiler>(output, "wmma_integer_gemm_u8_16x16x16_tn", options, config);
|
|
}
|
|
|
|
{
|
|
typedef cutlass::gemm::WmmaGemmTraits<cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::Shape<128, 128, 128>,
|
|
unsigned char,
|
|
unsigned char,
|
|
int,
|
|
cutlass::gemm::LinearScaling<int>,
|
|
int,
|
|
cutlass::Shape<128, 32, 32>,
|
|
cutlass::Shape<16, 16, 16>,
|
|
16,
|
|
16> WmmaGemmTraits;
|
|
|
|
typedef WmmaIntegerGemmDispatch<WmmaGemmTraits, unsigned char, unsigned char> Dispatch;
|
|
|
|
typedef perf::GemmProfiler<unsigned char, unsigned char, int, int, int> GemmProfiler;
|
|
|
|
results |= profile_gemm<Dispatch, GemmProfiler>(output, "wmma_integer_gemm_u8_16x16x16_tt", options, config);
|
|
}
|
|
|
|
// compute capability check
|
|
if (!options.compute_capability_exact(7, 5)) {
|
|
return 0;
|
|
}
|
|
|
|
#ifdef CUTLASS_USE_SUBBYTE_WMMA
|
|
{
|
|
typedef cutlass::gemm::WmmaGemmTraits<cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::Shape<256, 128, 128>,
|
|
cutlass::Vector<cutlass::int4_t, 8>,
|
|
cutlass::Vector<cutlass::int4_t, 8>,
|
|
int,
|
|
cutlass::gemm::LinearScaling<int>,
|
|
int,
|
|
cutlass::Shape<256, 32, 32>,
|
|
cutlass::Shape<32, 8, 8>,
|
|
32,
|
|
32> WmmaGemmTraits;
|
|
|
|
typedef WmmaIntegerGemmDispatch<WmmaGemmTraits,
|
|
cutlass::Vector<cutlass::int4_t, 8>,
|
|
cutlass::Vector<cutlass::int4_t, 8> > Dispatch;
|
|
|
|
typedef perf::GemmProfiler<cutlass::Vector<cutlass::int4_t, 8>,
|
|
cutlass::Vector<cutlass::int4_t, 8>,
|
|
int,
|
|
int,
|
|
int> GemmProfiler;
|
|
|
|
results |= profile_gemm<Dispatch, GemmProfiler>(output, "wmma_integer_gemm_s4_tn", options, config);
|
|
}
|
|
|
|
{
|
|
typedef cutlass::gemm::WmmaGemmTraits<cutlass::MatrixLayout::kRowMajor,
|
|
cutlass::MatrixLayout::kColumnMajor,
|
|
cutlass::Shape<256, 128, 128>,
|
|
cutlass::Vector<cutlass::uint4_t, 8>,
|
|
cutlass::Vector<cutlass::uint4_t, 8>,
|
|
int,
|
|
cutlass::gemm::LinearScaling<int>,
|
|
int,
|
|
cutlass::Shape<256, 32, 32>,
|
|
cutlass::Shape<32, 8, 8>,
|
|
32,
|
|
32> WmmaGemmTraits;
|
|
|
|
typedef WmmaIntegerGemmDispatch<WmmaGemmTraits,
|
|
cutlass::Vector<cutlass::uint4_t, 8>,
|
|
cutlass::Vector<cutlass::uint4_t, 8> > Dispatch;
|
|
|
|
typedef perf::GemmProfiler<cutlass::Vector<cutlass::uint4_t, 8>,
|
|
cutlass::Vector<cutlass::uint4_t, 8>,
|
|
int,
|
|
int,
|
|
int> GemmProfiler;
|
|
|
|
results |= profile_gemm<Dispatch, GemmProfiler>(output, "wmma_integer_gemm_u4_tn", options, config);
|
|
}
|
|
#endif //ifdef CUTLASS_USE_SUBBYTE_WMMA
|
|
|
|
return results;
|
|
}
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
} // namespace perf
|
|
|
|
////////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
#else // ! CUTLASS_USE_INT_WMMA
|
|
|
|
namespace perf {
|
|
|
|
int profile_wmma_integer_gemm(TestbenchOutput<GemmProblem> &output, TestbenchOptions const &options, Config const &config) {
|
|
return 0;
|
|
}
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
struct WmmaIntegerGemmRegistrar {
|
|
WmmaIntegerGemmRegistrar() { perf::RegisterGemmProfileFunc(perf::profile_wmma_integer_gemm); }
|
|
};
|
|
|
|
volatile WmmaIntegerGemmRegistrar _WmmaIntegerGemmRegistrar;
|
|
|
|
#endif // ifdef CUTLASS_USE_WMMA_API
|