56fd68e7dd
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
129 lines
3.5 KiB
Plaintext
129 lines
3.5 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.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#ifndef __ALGORITHM_COMMON__
|
|
#define __ALGORITHM_COMMON__
|
|
|
|
#include <cuda/algorithm>
|
|
#include <cuda/memory_pool>
|
|
#include <cuda/memory_resource>
|
|
#include <cuda/std/mdspan>
|
|
|
|
#include <cuda/experimental/container.cuh>
|
|
#include <cuda/experimental/memory_resource.cuh>
|
|
|
|
#include <testing.cuh>
|
|
#include <utility.cuh>
|
|
|
|
inline constexpr uint8_t fill_byte = 1;
|
|
inline constexpr uint32_t buffer_size = 42;
|
|
|
|
inline int get_expected_value(uint8_t pattern_byte)
|
|
{
|
|
int result;
|
|
memset(&result, pattern_byte, sizeof(int));
|
|
return result;
|
|
}
|
|
|
|
template <typename Result>
|
|
void check_result_and_erase(cudax::stream_ref stream, Result&& result, uint8_t pattern_byte = fill_byte)
|
|
{
|
|
int expected = get_expected_value(pattern_byte);
|
|
|
|
stream.sync();
|
|
for (int& i : result)
|
|
{
|
|
REQUIRE(i == expected);
|
|
i = 0;
|
|
}
|
|
}
|
|
|
|
template <typename Layout = cuda::std::layout_right, typename Extents>
|
|
auto make_buffer_for_mdspan(Extents extents, char value = 0)
|
|
{
|
|
cuda::mr::legacy_pinned_memory_resource host_resource;
|
|
auto mapping = typename Layout::template mapping<decltype(extents)>{extents};
|
|
|
|
cudax::uninitialized_buffer<int, cuda::mr::host_accessible> buffer(host_resource, mapping.required_span_size());
|
|
|
|
memset(buffer.data(), value, buffer.size_bytes());
|
|
|
|
return buffer;
|
|
}
|
|
|
|
inline auto create_fake_strided_mdspan()
|
|
{
|
|
cuda::std::dextents<size_t, 3> dynamic_extents{1, 2, 3};
|
|
cuda::std::array<size_t, 3> strides{12, 4, 1};
|
|
#if _CCCL_CUDA_COMPILER(NVCC, <, 12, 6)
|
|
auto map = cuda::std::layout_stride::mapping{dynamic_extents, strides};
|
|
#else // ^^^ _CCCL_CUDA_COMPILER(NVCC, <, 12, 6) ^^^ / vvv _CCCL_CUDA_COMPILER(NVCC, >=, 12, 6) vvv
|
|
cuda::std::layout_stride::mapping map{dynamic_extents, strides};
|
|
#endif // ^^^ _CCCL_CUDA_COMPILER(NVCC, >=, 12, 6) ^^^
|
|
return cuda::std::mdspan<int, decltype(dynamic_extents), cuda::std::layout_stride>(nullptr, map);
|
|
};
|
|
|
|
namespace cuda::experimental
|
|
{
|
|
// Need a type that goes through all launch_transform steps, but is not a contiguous_range
|
|
template <typename RelocatableValue = cuda::std::span<int>>
|
|
struct weird_buffer
|
|
{
|
|
cuda::mr::legacy_pinned_memory_resource& resource;
|
|
int* data;
|
|
std::size_t size;
|
|
|
|
weird_buffer(::cuda::mr::legacy_pinned_memory_resource& res, std::size_t s)
|
|
: resource(res)
|
|
, data((int*) res.allocate_sync(s * sizeof(int)))
|
|
, size(s)
|
|
{
|
|
memset(data, 0, size);
|
|
}
|
|
|
|
~weird_buffer()
|
|
{
|
|
resource.deallocate_sync(data, size);
|
|
}
|
|
|
|
weird_buffer(const weird_buffer&) = delete;
|
|
weird_buffer(weird_buffer&&) = delete;
|
|
|
|
struct transform_result
|
|
{
|
|
int* data;
|
|
std::size_t size;
|
|
|
|
RelocatableValue transformed_argument()
|
|
{
|
|
return *this;
|
|
};
|
|
|
|
operator cuda::std::span<int>()
|
|
{
|
|
return {data, size};
|
|
}
|
|
|
|
template <typename Extents>
|
|
operator cuda::std::mdspan<int, Extents>()
|
|
{
|
|
return cuda::std::mdspan<int, Extents>{data};
|
|
}
|
|
};
|
|
|
|
[[nodiscard]] friend transform_result transform_launch_argument(cuda::stream_ref, const weird_buffer& self) noexcept
|
|
{
|
|
return {self.data, self.size};
|
|
}
|
|
};
|
|
} // namespace cuda::experimental
|
|
|
|
#endif // __ALGORITHM_COMMON__
|