c53f3339bb
CUTLASS 2.3 adds GEMMs targeting Sparse Tensor Cores on the NVIDIA Ampere Architecture, fast SGEMM, and small matrix classes, bug fixes, and performance enhancements.
213 lines
7.1 KiB
Plaintext
213 lines
7.1 KiB
Plaintext
/***************************************************************************************************
|
|
* Copyright (c) 2017-2020, 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 Statically sized array of elements that accommodates all CUTLASS-supported numeric types
|
|
and is safe to use in a union.
|
|
*/
|
|
|
|
#include "../common/cutlass_unit_test.h"
|
|
|
|
#include "cutlass/array.h"
|
|
#include "cutlass/core_io.h"
|
|
#include "cutlass/numeric_types.h"
|
|
#include "cutlass/numeric_conversion.h"
|
|
#include "cutlass/layout/matrix.h"
|
|
|
|
#include "cutlass/util/device_memory.h"
|
|
#include "cutlass/util/host_tensor.h"
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
__global__ void convert_bf16_f32(cutlass::bfloat16_t *output, float const *input, int N) {
|
|
int tid = threadIdx.x + blockIdx.x * blockDim.x;
|
|
if (tid < N) {
|
|
output[tid] = static_cast<cutlass::bfloat16_t>(input[tid]);
|
|
}
|
|
}
|
|
|
|
__global__ void convert_and_pack_bf16(cutlass::bfloat16_t *output, float const *input, int N) {
|
|
int tid = threadIdx.x + blockIdx.x * blockDim.x;
|
|
if (tid * 2 < N) {
|
|
|
|
cutlass::NumericArrayConverter<cutlass::bfloat16_t, float, 2> convert;
|
|
|
|
cutlass::Array<cutlass::bfloat16_t, 2> *dst_ptr =
|
|
reinterpret_cast<cutlass::Array<cutlass::bfloat16_t, 2> *>(output + tid * 2);
|
|
|
|
cutlass::Array<float, 2> const *src_ptr =
|
|
reinterpret_cast<cutlass::Array<float, 2> const *>(input + tid * 2);
|
|
|
|
*dst_ptr = convert(*src_ptr);
|
|
}
|
|
}
|
|
|
|
TEST(bfloat16_t, device_conversion) {
|
|
using T = cutlass::bfloat16_t;
|
|
using S = float;
|
|
|
|
int const N = 256;
|
|
|
|
cutlass::HostTensor<T, cutlass::layout::RowMajor> destination({N, 1});
|
|
cutlass::HostTensor<S, cutlass::layout::RowMajor> source({N, 1});
|
|
|
|
for (int i = 0; i < N; ++i) {
|
|
source.at({i, 0}) = float(i - 128);
|
|
destination.at({i, 0}) = T(0);
|
|
}
|
|
|
|
source.sync_device();
|
|
destination.sync_device();
|
|
|
|
convert_bf16_f32<<< dim3(1,1), dim3(N, 1) >>>(destination.device_data(), source.device_data(), N);
|
|
|
|
ASSERT_EQ(cudaGetLastError(), cudaSuccess) << "Kernel launch error.";
|
|
|
|
destination.sync_host();
|
|
|
|
int errors = 0;
|
|
for (int i = 0; i < N; ++i) {
|
|
T got = destination.at({i, 0});
|
|
S expected = source.at({i, 0});
|
|
|
|
if (S(got) != expected) {
|
|
++errors;
|
|
if (errors < 10) {
|
|
std::cerr << "Basic conversion error - [" << i << "] - got " << got << ", expected " << expected << "\n";
|
|
}
|
|
}
|
|
|
|
destination.at({i, 0}) = T(0);
|
|
}
|
|
|
|
destination.sync_device();
|
|
|
|
convert_and_pack_bf16<<< dim3(1,1), dim3(N, 1) >>>(destination.device_data(), source.device_data(), N);
|
|
|
|
ASSERT_EQ(cudaGetLastError(), cudaSuccess) << "Kernel launch error.";
|
|
|
|
destination.sync_host();
|
|
|
|
for (int i = 0; i < N; ++i) {
|
|
T got = destination.at({i, 0});
|
|
S expected = source.at({i, 0});
|
|
|
|
if (S(got) != expected) {
|
|
++errors;
|
|
if (errors < 10) {
|
|
std::cerr << "Convert and pack error - [" << i << "] - got " << got << ", expected " << expected << "\n";
|
|
}
|
|
}
|
|
}
|
|
|
|
EXPECT_EQ(errors, 0);
|
|
}
|
|
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
//
|
|
// Host
|
|
//
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
TEST(bfloat16_t, host_conversion) {
|
|
for (int i = -128; i < 128; ++i) {
|
|
float f = static_cast<float>(i);
|
|
|
|
cutlass::bfloat16_t x = static_cast<cutlass::bfloat16_t>(i);
|
|
cutlass::bfloat16_t y = static_cast<cutlass::bfloat16_t>(f);
|
|
|
|
EXPECT_TRUE(static_cast<int>(x) == i);
|
|
EXPECT_TRUE(static_cast<float>(y) == f);
|
|
}
|
|
|
|
// Try out default-ctor (zero initialization of primitive proxy type)
|
|
EXPECT_TRUE(cutlass::bfloat16_t() == 0.0_bf16);
|
|
|
|
// Try out user-defined literals
|
|
EXPECT_TRUE(cutlass::bfloat16_t(7) == 7_bf16);
|
|
EXPECT_TRUE(7 == static_cast<int>(7_bf16));
|
|
}
|
|
|
|
TEST(bfloat16_t, host_arithmetic) {
|
|
|
|
for (int i = -100; i < 100; ++i) {
|
|
for (int j = -100; j < 100; ++j) {
|
|
|
|
cutlass::bfloat16_t x = static_cast<cutlass::bfloat16_t>(i);
|
|
cutlass::bfloat16_t y = static_cast<cutlass::bfloat16_t>(j);
|
|
|
|
EXPECT_TRUE(static_cast<int>(x + y) == (i + j));
|
|
}
|
|
}
|
|
}
|
|
|
|
TEST(bfloat16_t, host_round) {
|
|
|
|
struct {
|
|
uint32_t f32_bits;
|
|
uint16_t expected;
|
|
} tests[] = {
|
|
{0x40040000, 0x4004}, // M=0, R=0, S=0 => rtz
|
|
{0x40048000, 0x4004}, // M=0, R=1, S=0 => rtz
|
|
{0x40040001, 0x4004}, // M=0, R=1, S=1 => +inf
|
|
{0x4004c000, 0x4005}, // M=0, R=1, S=1 => +inf
|
|
{0x4004a000, 0x4005}, // M=0, R=1, S=1 => +inf
|
|
{0x40050000, 0x4005}, // M=1, R=0, S=0 => rtz
|
|
{0x40054000, 0x4005}, // M=1, R=0, S=1 => rtz
|
|
{0x40058000, 0x4006}, // M=1, R=1, S=0 => +inf
|
|
{0x40058001, 0x4006}, // M=1, R=1, S=1 => +inf
|
|
{0x7f800000, 0x7f80}, // +inf
|
|
{0xff800000, 0xff80}, // -inf
|
|
{0x7fffffff, 0x7fff}, // canonical NaN
|
|
{0x7ff00001, 0x7fff}, // NaN -> canonical NaN
|
|
{0xfff00010, 0x7fff}, // Nan -> canonical NaN
|
|
{0, 0}
|
|
};
|
|
|
|
bool running = true;
|
|
for (int i = 0; running; ++i) {
|
|
|
|
float f32 = reinterpret_cast<float const &>(tests[i].f32_bits);
|
|
|
|
cutlass::bfloat16_t bf16 = cutlass::bfloat16_t(f32);
|
|
|
|
bool passed = (tests[i].expected == bf16.raw());
|
|
|
|
EXPECT_TRUE(passed)
|
|
<< "Error - convert(f32: 0x" << std::hex << tests[i].f32_bits
|
|
<< ") -> 0x" << std::hex << tests[i].expected << "\ngot: 0x" << std::hex << bf16.raw();
|
|
|
|
if (!tests[i].f32_bits) {
|
|
running = false;
|
|
}
|
|
}
|
|
}
|
|
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|
|
//
|
|
// Device
|
|
//
|
|
/////////////////////////////////////////////////////////////////////////////////////////////////
|