CCCL (CUDA C++ Core Libraries) provides: - CUB: device/block/warp-level GPU primitives (reduce, scan, sort, topk) - Thrust: high-level parallel algorithms (transform_reduce, sort, scan) - libcudacxx: CUDA C++ standard library (atomics, barriers, memory) - cudax: experimental features (memory resources, allocators) - Tuning policies: per-SM hardware-specific algorithm parameters Competition optimization vectors mapped to CCCL: - Output TPS (83% weight): warp_reduce, block_reduce, device_topk - Input TPS (14% weight): device_scan, block_load, prefetch - Cache TPS (3% weight): prefix caching strategy patterns - Memory (0.9 util): pooled/cached/buddy allocators Source: https://github.com/NVIDIA/cccl (shallow clone, HEAD only) License: Apache-2.0
188 lines
6.8 KiB
Plaintext
188 lines
6.8 KiB
Plaintext
//===----------------------------------------------------------------------===//
|
|
//
|
|
// Part of CUDA Experimental in CUDA C++ Core Libraries,
|
|
// under the Apache License v2.0 with LLVM Exceptions.
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
// SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#include "common.cuh"
|
|
|
|
C2H_TEST("1d Copy", "[data_manipulation]")
|
|
{
|
|
cuda::stream _stream{cuda::device_ref{0}};
|
|
|
|
SECTION("Device resource")
|
|
{
|
|
cuda::device_memory_pool_ref device_resource = cuda::device_default_memory_pool(cuda::device_ref{0});
|
|
std::vector<int> host_vector(buffer_size);
|
|
|
|
{
|
|
cuda::__uninitialized_async_buffer<int, cuda::mr::device_accessible> buffer(device_resource, _stream, buffer_size);
|
|
cuda::fill_bytes(_stream, buffer, fill_byte);
|
|
|
|
cuda::copy_bytes(_stream, buffer, host_vector);
|
|
check_result_and_erase(_stream, host_vector);
|
|
|
|
cuda::copy_bytes(_stream, std::move(buffer), host_vector);
|
|
check_result_and_erase(_stream, host_vector);
|
|
}
|
|
{
|
|
cuda::__uninitialized_async_buffer<int, cuda::mr::device_accessible> not_yet_const_buffer(
|
|
device_resource, _stream, buffer_size);
|
|
cuda::fill_bytes(_stream, not_yet_const_buffer, fill_byte);
|
|
|
|
const auto& const_buffer = not_yet_const_buffer;
|
|
|
|
cuda::copy_bytes(_stream, const_buffer, host_vector);
|
|
check_result_and_erase(_stream, host_vector);
|
|
|
|
cuda::copy_bytes(_stream, const_buffer, cuda::std::span(host_vector));
|
|
check_result_and_erase(_stream, host_vector);
|
|
}
|
|
}
|
|
|
|
SECTION("Host and managed resource")
|
|
{
|
|
cuda::mr::legacy_managed_memory_resource managed_resource;
|
|
cuda::mr::legacy_pinned_memory_resource host_resource;
|
|
|
|
{
|
|
cudax::uninitialized_buffer<int, cuda::mr::host_accessible> host_buffer(host_resource, buffer_size);
|
|
cudax::uninitialized_buffer<int, cuda::mr::device_accessible> device_buffer(managed_resource, buffer_size);
|
|
|
|
cuda::fill_bytes(_stream, host_buffer, fill_byte);
|
|
|
|
cuda::copy_bytes(_stream, host_buffer, device_buffer);
|
|
check_result_and_erase(_stream, device_buffer);
|
|
|
|
cuda::copy_bytes(_stream, cuda::std::span(host_buffer), device_buffer);
|
|
check_result_and_erase(_stream, device_buffer);
|
|
}
|
|
|
|
{
|
|
cudax::uninitialized_buffer<int, cuda::mr::host_accessible> not_yet_const_host_buffer(host_resource, buffer_size);
|
|
cudax::uninitialized_buffer<int, cuda::mr::device_accessible> device_buffer(managed_resource, buffer_size);
|
|
cuda::fill_bytes(_stream, not_yet_const_host_buffer, fill_byte);
|
|
|
|
const auto& const_host_buffer = not_yet_const_host_buffer;
|
|
|
|
cuda::copy_bytes(_stream, const_host_buffer, device_buffer);
|
|
check_result_and_erase(_stream, device_buffer);
|
|
|
|
cuda::copy_bytes(_stream, cuda::std::span(const_host_buffer), device_buffer);
|
|
check_result_and_erase(_stream, device_buffer);
|
|
}
|
|
}
|
|
SECTION("Launch transform")
|
|
{
|
|
cuda::mr::legacy_pinned_memory_resource host_resource;
|
|
cudax::weird_buffer input(host_resource, buffer_size);
|
|
cudax::weird_buffer output(host_resource, buffer_size);
|
|
|
|
memset(input.data, fill_byte, input.size * sizeof(int));
|
|
|
|
cuda::copy_bytes(_stream, input, output);
|
|
check_result_and_erase(_stream, cuda::std::span(output.data, output.size));
|
|
}
|
|
|
|
SECTION("Asymmetric size")
|
|
{
|
|
cuda::mr::legacy_pinned_memory_resource host_resource;
|
|
cudax::uninitialized_buffer<int, cuda::mr::host_accessible> host_buffer(host_resource, 1);
|
|
cuda::fill_bytes(_stream, host_buffer, fill_byte);
|
|
|
|
::std::vector<int> vec(buffer_size, 0xbeef);
|
|
|
|
cuda::copy_bytes(_stream, host_buffer, vec);
|
|
_stream.sync();
|
|
|
|
REQUIRE(vec[0] == get_expected_value(fill_byte));
|
|
REQUIRE(vec[1] == 0xbeef);
|
|
}
|
|
}
|
|
|
|
template <typename SrcLayout = cuda::std::layout_right,
|
|
typename DstLayout = SrcLayout,
|
|
typename SrcExtents,
|
|
typename DstExtents>
|
|
void test_mdspan_copy_bytes(
|
|
cudax::stream_ref stream, SrcExtents src_extents = SrcExtents(), DstExtents dst_extents = DstExtents())
|
|
{
|
|
auto src_buffer = make_buffer_for_mdspan<SrcLayout>(src_extents, 1);
|
|
auto dst_buffer = make_buffer_for_mdspan<DstLayout>(dst_extents, 0);
|
|
|
|
cuda::std::mdspan<int, SrcExtents, SrcLayout> src(src_buffer.data(), src_extents);
|
|
cuda::std::mdspan<int, DstExtents, DstLayout> dst(dst_buffer.data(), dst_extents);
|
|
|
|
for (int i = 0; i < static_cast<int>(src.extent(1)); i++)
|
|
{
|
|
src(0, i) = i;
|
|
}
|
|
|
|
cuda::copy_bytes(stream, std::move(src), dst);
|
|
stream.sync();
|
|
|
|
for (int i = 0; i < static_cast<int>(dst.extent(1)); i++)
|
|
{
|
|
CHECK(dst(0, i) == i);
|
|
}
|
|
}
|
|
|
|
C2H_TEST("Mdspan copy", "[data_manipulation]")
|
|
{
|
|
cuda::stream stream{cuda::device_ref{0}};
|
|
|
|
SECTION("Different extents")
|
|
{
|
|
auto static_extents = cuda::std::extents<size_t, 3, 4>();
|
|
test_mdspan_copy_bytes(stream, static_extents, static_extents);
|
|
test_mdspan_copy_bytes<cuda::std::layout_left>(stream, static_extents, static_extents);
|
|
|
|
auto dynamic_extents = cuda::std::dextents<size_t, 2>(3, 4);
|
|
test_mdspan_copy_bytes(stream, dynamic_extents, dynamic_extents);
|
|
test_mdspan_copy_bytes(stream, static_extents, dynamic_extents);
|
|
test_mdspan_copy_bytes<cuda::std::layout_left>(stream, static_extents, dynamic_extents);
|
|
|
|
auto mixed_extents = cuda::std::extents<int, cuda::std::dynamic_extent, 4>(3);
|
|
test_mdspan_copy_bytes(stream, dynamic_extents, mixed_extents);
|
|
test_mdspan_copy_bytes(stream, mixed_extents, static_extents);
|
|
test_mdspan_copy_bytes<cuda::std::layout_left>(stream, mixed_extents, static_extents);
|
|
}
|
|
|
|
SECTION("Launch transform")
|
|
{
|
|
auto host_resource = cuda::mr::legacy_pinned_memory_resource{};
|
|
auto mixed_extents =
|
|
cuda::std::extents<size_t, 1024, cuda::std::dynamic_extent, 2, cuda::std::dynamic_extent>(1024, 2);
|
|
[[maybe_unused]] auto static_extents = cuda::std::extents<size_t, 1024, 1024, 2, 2>();
|
|
auto mdspan_buffer = make_buffer_for_mdspan(mixed_extents, 1);
|
|
cuda::std::mdspan<int, decltype(mixed_extents)> mdspan(mdspan_buffer.data(), mixed_extents);
|
|
cudax::weird_buffer<cuda::std::mdspan<int, decltype(static_extents)>> buffer{
|
|
host_resource, mdspan.mapping().required_span_size()};
|
|
|
|
cuda::copy_bytes(stream, mdspan, buffer);
|
|
stream.sync();
|
|
REQUIRE(!memcmp(mdspan_buffer.data(), buffer.data, mdspan_buffer.size()));
|
|
}
|
|
}
|
|
|
|
C2H_TEST("Non exhaustive mdspan copy_bytes", "[data_manipulation]")
|
|
{
|
|
cuda::stream stream{cuda::device_ref{0}};
|
|
{
|
|
auto fake_strided_mdspan = create_fake_strided_mdspan();
|
|
|
|
try
|
|
{
|
|
cuda::copy_bytes(stream, fake_strided_mdspan, fake_strided_mdspan);
|
|
}
|
|
catch (const ::std::invalid_argument& e)
|
|
{
|
|
CHECK(e.what() == ::std::string("copy_bytes supports only exhaustive mdspans"));
|
|
}
|
|
}
|
|
}
|