fb335f6a5f
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.
186 lines
6.2 KiB
Plaintext
186 lines
6.2 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.
|
|
*
|
|
**************************************************************************************************/
|
|
/*! \file
|
|
\brief Unit tests for conversion operators.
|
|
*/
|
|
|
|
#include "../common/cutlass_unit_test.h"
|
|
|
|
#include "cutlass/numeric_conversion.h"
|
|
|
|
#include "cutlass/layout/matrix.h"
|
|
#include "cutlass/util/host_tensor.h"
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
namespace test {
|
|
namespace core {
|
|
namespace kernel {
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
/// Conversion template
|
|
template <typename Destination, typename Source, int Count>
|
|
__global__ void convert(
|
|
cutlass::Array<Destination, Count> *destination,
|
|
cutlass::Array<Source, Count> const *source) {
|
|
|
|
cutlass::NumericArrayConverter<Destination, Source, Count> convert;
|
|
|
|
*destination = convert(*source);
|
|
}
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
} // namespace kernel
|
|
} // namespace core
|
|
} // namespace test
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
TEST(NumericConversion, f32_to_f16_rn) {
|
|
|
|
int const kN = 1;
|
|
using Source = float;
|
|
using Destination = cutlass::half_t;
|
|
|
|
dim3 grid(1, 1);
|
|
dim3 block(1, 1);
|
|
|
|
cutlass::HostTensor<cutlass::half_t, cutlass::layout::RowMajor> destination({1, kN});
|
|
cutlass::HostTensor<float, cutlass::layout::RowMajor> source({1, kN});
|
|
|
|
for (int i = 0; i < kN; ++i) {
|
|
source.host_data()[i] = float(i);
|
|
}
|
|
|
|
source.sync_device();
|
|
|
|
test::core::kernel::convert<Destination, Source, 1><<< grid, block >>>(
|
|
reinterpret_cast<cutlass::Array<Destination, 1> *>(destination.device_data()),
|
|
reinterpret_cast<cutlass::Array<Source, 1> const *>(source.device_data())
|
|
);
|
|
|
|
destination.sync_host();
|
|
|
|
for (int i = 0; i < kN; ++i) {
|
|
EXPECT_TRUE(float(destination.host_data()[i]) == source.host_data()[i]);
|
|
}
|
|
}
|
|
|
|
TEST(NumericConversion, f32x8_to_f16x8_rn) {
|
|
|
|
int const kN = 8;
|
|
using Source = float;
|
|
using Destination = cutlass::half_t;
|
|
|
|
dim3 grid(1, 1);
|
|
dim3 block(1, 1);
|
|
|
|
cutlass::HostTensor<Destination, cutlass::layout::RowMajor> destination({1, kN});
|
|
cutlass::HostTensor<Source, cutlass::layout::RowMajor> source({1, kN});
|
|
|
|
for (int i = 0; i < kN; ++i) {
|
|
source.host_data()[i] = float(i);
|
|
}
|
|
|
|
source.sync_device();
|
|
|
|
test::core::kernel::convert<Destination, Source, kN><<< grid, block >>>(
|
|
reinterpret_cast<cutlass::Array<Destination, kN> *>(destination.device_data()),
|
|
reinterpret_cast<cutlass::Array<Source, kN> const *>(source.device_data())
|
|
);
|
|
|
|
destination.sync_host();
|
|
|
|
for (int i = 0; i < kN; ++i) {
|
|
EXPECT_TRUE(float(destination.host_data()[i]) == source.host_data()[i]);
|
|
}
|
|
}
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
TEST(NumericConversion, f16_to_f32_rn) {
|
|
|
|
int const kN = 1;
|
|
using Source = cutlass::half_t;
|
|
using Destination = float;
|
|
|
|
dim3 grid(1, 1);
|
|
dim3 block(1, 1);
|
|
|
|
cutlass::HostTensor<float, cutlass::layout::RowMajor> destination({1, kN});
|
|
cutlass::HostTensor<cutlass::half_t, cutlass::layout::RowMajor> source({1, kN});
|
|
|
|
for (int i = 0; i < kN; ++i) {
|
|
source.host_data()[i] = Source(i);
|
|
}
|
|
|
|
source.sync_device();
|
|
|
|
test::core::kernel::convert<Destination, Source, kN><<< grid, block >>>(
|
|
reinterpret_cast<cutlass::Array<Destination, kN> *>(destination.device_data()),
|
|
reinterpret_cast<cutlass::Array<Source, kN> const *>(source.device_data())
|
|
);
|
|
|
|
destination.sync_host();
|
|
|
|
for (int i = 0; i < kN; ++i) {
|
|
EXPECT_TRUE(float(destination.host_data()[i]) == float(source.host_data()[i]));
|
|
}
|
|
}
|
|
|
|
TEST(NumericConversion, f16x8_to_f32x8_rn) {
|
|
|
|
int const kN = 8;
|
|
using Source = cutlass::half_t;
|
|
using Destination = float;
|
|
|
|
dim3 grid(1, 1);
|
|
dim3 block(1, 1);
|
|
|
|
cutlass::HostTensor<float, cutlass::layout::RowMajor> destination({1, kN});
|
|
cutlass::HostTensor<cutlass::half_t, cutlass::layout::RowMajor> source({1, kN});
|
|
|
|
for (int i = 0; i < kN; ++i) {
|
|
source.host_data()[i] = float(i);
|
|
}
|
|
|
|
source.sync_device();
|
|
|
|
test::core::kernel::convert<Destination, Source, kN><<< grid, block >>>(
|
|
reinterpret_cast<cutlass::Array<Destination, kN> *>(destination.device_data()),
|
|
reinterpret_cast<cutlass::Array<Source, kN> const *>(source.device_data())
|
|
);
|
|
|
|
destination.sync_host();
|
|
|
|
for (int i = 0; i < kN; ++i) {
|
|
EXPECT_TRUE(float(destination.host_data()[i]) == float(source.host_data()[i]));
|
|
}
|
|
}
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|