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
270 lines
9.8 KiB
Plaintext
270 lines
9.8 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 <cuda/std/execution>
|
|
#include <cuda/std/type_traits>
|
|
|
|
#include <cuda/experimental/container.cuh>
|
|
#include <cuda/experimental/execution.cuh>
|
|
#include <cuda/experimental/memory_resource.cuh>
|
|
|
|
#include <testing.cuh>
|
|
|
|
namespace cudax = cuda::experimental;
|
|
using env_t = cudax::env_t<cuda::mr::device_accessible>;
|
|
|
|
struct test_resource
|
|
{
|
|
void* allocate_sync(size_t, size_t)
|
|
{
|
|
return nullptr;
|
|
}
|
|
void* allocate(cuda::stream_ref, size_t, size_t)
|
|
{
|
|
return nullptr;
|
|
}
|
|
void deallocate_sync(void*, size_t, size_t) noexcept {}
|
|
void deallocate(cuda::stream_ref, void*, size_t, size_t) noexcept {}
|
|
|
|
constexpr bool operator==(const test_resource&) const noexcept
|
|
{
|
|
return true;
|
|
}
|
|
|
|
constexpr bool operator!=(const test_resource&) const noexcept
|
|
{
|
|
return false;
|
|
}
|
|
|
|
friend void get_property(const test_resource&, cuda::mr::device_accessible) noexcept {}
|
|
};
|
|
|
|
C2H_TEST("env_t is queryable for all properties we want", "[execution][env]")
|
|
{
|
|
STATIC_REQUIRE(cuda::std::execution::__queryable_with<env_t, cuda::get_stream_t>);
|
|
STATIC_REQUIRE(cuda::std::execution::__queryable_with<env_t, cuda::mr::get_memory_resource_t>);
|
|
STATIC_REQUIRE(cuda::std::execution::__queryable_with<env_t, cudax::execution::get_execution_policy_t>);
|
|
}
|
|
|
|
C2H_TEST("env_t is default constructible", "[execution][env]")
|
|
{
|
|
env_t env{cuda::device_default_memory_pool(cuda::device_ref{0})};
|
|
CHECK(env.query(cuda::get_stream) == cuda::invalid_stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cudax::execution::any_execution_policy{});
|
|
CHECK(env.query(cuda::mr::get_memory_resource) == cuda::device_default_memory_pool(cuda::device_ref{0}));
|
|
}
|
|
|
|
C2H_TEST("env_t is constructible from an any_resource", "[execution][env]")
|
|
{
|
|
const cuda::mr::any_resource<cuda::mr::device_accessible> mr{test_resource{}};
|
|
|
|
SECTION("Passing an any_resource")
|
|
{
|
|
env_t env{mr};
|
|
CHECK(env.query(cuda::get_stream) == cuda::invalid_stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cudax::execution::any_execution_policy{});
|
|
CHECK(env.query(cuda::mr::get_memory_resource) == mr);
|
|
}
|
|
|
|
SECTION("Passing an any_resource and a stream")
|
|
{
|
|
cudax::stream stream{cuda::device_ref{0}};
|
|
env_t env{mr, stream};
|
|
CHECK(env.query(cuda::get_stream) == stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cudax::execution::any_execution_policy{});
|
|
CHECK(env.query(cuda::mr::get_memory_resource) == mr);
|
|
}
|
|
|
|
SECTION("Passing an any_resource, a stream and a policy")
|
|
{
|
|
cudax::stream stream{cuda::device_ref{0}};
|
|
env_t env{mr, stream, cuda::std::execution::par_unseq};
|
|
CHECK(env.query(cuda::get_stream) == stream);
|
|
CHECK((env.query(cudax::execution::get_execution_policy) == cuda::std::execution::par_unseq));
|
|
CHECK(env.query(cuda::mr::get_memory_resource) == mr);
|
|
}
|
|
}
|
|
|
|
C2H_TEST("env_t is constructible from an any_resource passed as an rvalue", "[execution][env]")
|
|
{
|
|
SECTION("Passing an any_resource")
|
|
{
|
|
env_t env{cuda::mr::any_resource<cuda::mr::device_accessible>{test_resource{}}};
|
|
CHECK(env.query(cuda::get_stream) == cuda::invalid_stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cudax::execution::any_execution_policy{});
|
|
CHECK(env.query(cuda::mr::get_memory_resource)
|
|
== cuda::mr::any_resource<cuda::mr::device_accessible>{test_resource{}});
|
|
}
|
|
|
|
SECTION("Passing an any_resource and a stream")
|
|
{
|
|
cudax::stream stream{cuda::device_ref{0}};
|
|
env_t env{cuda::mr::any_resource<cuda::mr::device_accessible>{test_resource{}}, stream};
|
|
CHECK(env.query(cuda::get_stream) == stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cudax::execution::any_execution_policy{});
|
|
CHECK(env.query(cuda::mr::get_memory_resource)
|
|
== cuda::mr::any_resource<cuda::mr::device_accessible>{test_resource{}});
|
|
}
|
|
|
|
SECTION("Passing an any_resource, a stream and a policy")
|
|
{
|
|
cudax::stream stream{cuda::device_ref{0}};
|
|
env_t env{
|
|
cuda::mr::any_resource<cuda::mr::device_accessible>{test_resource{}}, stream, cuda::std::execution::par_unseq};
|
|
CHECK(env.query(cuda::get_stream) == stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cuda::std::execution::par_unseq);
|
|
CHECK(env.query(cuda::mr::get_memory_resource)
|
|
== cuda::mr::any_resource<cuda::mr::device_accessible>{test_resource{}});
|
|
}
|
|
}
|
|
|
|
C2H_TEST("env_t is constructible from a resource", "[execution][env]")
|
|
{
|
|
test_resource mr{};
|
|
|
|
SECTION("Passing an any_resource")
|
|
{
|
|
env_t env{mr};
|
|
CHECK(env.query(cuda::get_stream) == cuda::invalid_stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cudax::execution::any_execution_policy{});
|
|
CHECK(env.query(cuda::mr::get_memory_resource) == mr);
|
|
}
|
|
|
|
SECTION("Passing an any_resource and a stream")
|
|
{
|
|
cudax::stream stream{cuda::device_ref{0}};
|
|
env_t env{mr, stream};
|
|
CHECK(env.query(cuda::get_stream) == stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cudax::execution::any_execution_policy{});
|
|
CHECK(env.query(cuda::mr::get_memory_resource) == mr);
|
|
}
|
|
|
|
SECTION("Passing an any_resource, a stream and a policy")
|
|
{
|
|
cudax::stream stream{cuda::device_ref{0}};
|
|
env_t env{mr, stream, cuda::std::execution::par_unseq};
|
|
CHECK(env.query(cuda::get_stream) == stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cuda::std::execution::par_unseq);
|
|
CHECK(env.query(cuda::mr::get_memory_resource) == mr);
|
|
}
|
|
}
|
|
|
|
C2H_TEST("env_t is constructible from a resource passed as an rvalue", "[execution][env]")
|
|
{
|
|
SECTION("Passing an any_resource")
|
|
{
|
|
env_t env{test_resource{}};
|
|
CHECK(env.query(cuda::get_stream) == cuda::invalid_stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cudax::execution::any_execution_policy{});
|
|
CHECK(env.query(cuda::mr::get_memory_resource) == test_resource{});
|
|
}
|
|
|
|
SECTION("Passing an any_resource and a stream")
|
|
{
|
|
cudax::stream stream{cuda::device_ref{0}};
|
|
env_t env{test_resource{}, stream};
|
|
CHECK(env.query(cuda::get_stream) == stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cudax::execution::any_execution_policy{});
|
|
CHECK(env.query(cuda::mr::get_memory_resource) == test_resource{});
|
|
}
|
|
|
|
SECTION("Passing an any_resource, a stream and a policy")
|
|
{
|
|
cudax::stream stream{cuda::device_ref{0}};
|
|
env_t env{test_resource{}, stream, cuda::std::execution::par_unseq};
|
|
CHECK(env.query(cuda::get_stream) == stream);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == cuda::std::execution::par_unseq);
|
|
CHECK(env.query(cuda::mr::get_memory_resource) == test_resource{});
|
|
}
|
|
}
|
|
|
|
struct some_env_t
|
|
{
|
|
test_resource res_{};
|
|
cudax::stream stream_{cuda::device_ref{0}};
|
|
cudax::execution::any_execution_policy policy_ = cuda::std::execution::par_unseq;
|
|
|
|
const test_resource& query(cuda::mr::get_memory_resource_t) const noexcept
|
|
{
|
|
return res_;
|
|
}
|
|
|
|
cudax::stream_ref query(cuda::get_stream_t) const noexcept
|
|
{
|
|
return stream_;
|
|
}
|
|
|
|
cudax::execution::any_execution_policy query(cudax::execution::get_execution_policy_t) const noexcept
|
|
{
|
|
return policy_;
|
|
}
|
|
};
|
|
C2H_TEST("env_t is constructible from a suitable env", "[execution][env]")
|
|
{
|
|
some_env_t other_env{};
|
|
env_t env{other_env};
|
|
CHECK(env.query(cuda::get_stream) == other_env.stream_);
|
|
CHECK(env.query(cudax::execution::get_execution_policy) == other_env.policy_);
|
|
CHECK(env.query(cuda::mr::get_memory_resource) == other_env.res_);
|
|
}
|
|
|
|
template <bool WithResource, bool WithStream, bool WithPolicy>
|
|
struct bad_env_t
|
|
{
|
|
test_resource res_{};
|
|
cudax::stream stream_{cuda::device_ref{0}};
|
|
cudax::execution::any_execution_policy policy_ = cuda::std::execution::par_unseq;
|
|
|
|
template <bool Enable = WithResource, cuda::std::enable_if_t<Enable, int> = 0>
|
|
const test_resource& query(cuda::mr::get_memory_resource_t) const noexcept
|
|
{
|
|
return res_;
|
|
}
|
|
|
|
template <bool Enable = WithStream, cuda::std::enable_if_t<Enable, int> = 0>
|
|
cudax::stream_ref query(cuda::get_stream_t) const noexcept
|
|
{
|
|
return stream_;
|
|
}
|
|
|
|
template <bool Enable = WithPolicy, cuda::std::enable_if_t<Enable, int> = 0>
|
|
cudax::execution::any_execution_policy query(cudax::execution::get_execution_policy_t) const noexcept
|
|
{
|
|
return policy_;
|
|
}
|
|
};
|
|
C2H_TEST("env_t is not constructible from a env missing queries", "[execution][env]")
|
|
{
|
|
STATIC_REQUIRE(cuda::std::is_constructible_v<env_t, bad_env_t<true, true, true>>);
|
|
STATIC_REQUIRE(!cuda::std::is_constructible_v<env_t, bad_env_t<false, true, true>>);
|
|
STATIC_REQUIRE(!cuda::std::is_constructible_v<env_t, bad_env_t<true, false, true>>);
|
|
STATIC_REQUIRE(!cuda::std::is_constructible_v<env_t, bad_env_t<true, true, false>>);
|
|
}
|
|
|
|
C2H_TEST("Can use query to construct various objects", "[execution][env]")
|
|
{
|
|
SECTION("Can create an any_resource")
|
|
{
|
|
env_t env{test_resource{}};
|
|
cuda::mr::any_synchronous_resource<cuda::mr::device_accessible> resource = env.query(cuda::mr::get_memory_resource);
|
|
CHECK(resource == test_resource{});
|
|
}
|
|
|
|
SECTION("Can create an __uninitialized_async_buffer")
|
|
{
|
|
cudax::stream stream_{cuda::device_ref{0}};
|
|
env_t env{test_resource{}, stream_};
|
|
cuda::__uninitialized_async_buffer<int, cuda::mr::device_accessible> buf{
|
|
env.query(cuda::mr::get_memory_resource), env.query(cuda::get_stream), 0ull};
|
|
CHECK(buf.memory_resource() == test_resource{});
|
|
CHECK(buf.stream() == stream_);
|
|
}
|
|
}
|