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
67 lines
2.3 KiB
Plaintext
67 lines
2.3 KiB
Plaintext
// SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
|
// SPDX-License-Identifier: BSD-3-Clause
|
|
|
|
#include <thrust/device_vector.h>
|
|
#include <thrust/uninitialized_copy.h>
|
|
|
|
#include <nvbench_helper.cuh>
|
|
|
|
template <typename T>
|
|
static void basic(nvbench::state& state, nvbench::type_list<T>)
|
|
{
|
|
const auto elements = static_cast<std::size_t>(state.get_int64("Elements"));
|
|
|
|
thrust::device_vector<T> input(elements, T{0xAA});
|
|
thrust::device_vector<T> output(elements, thrust::default_init);
|
|
|
|
state.add_element_count(elements);
|
|
state.add_global_memory_reads<T>(elements);
|
|
state.add_global_memory_writes<T>(elements);
|
|
|
|
caching_allocator_t alloc;
|
|
state.exec(nvbench::exec_tag::gpu | nvbench::exec_tag::no_batch | nvbench::exec_tag::sync,
|
|
[&](nvbench::launch& launch) {
|
|
thrust::uninitialized_copy(policy(alloc, launch), input.cbegin(), input.cend(), output.begin());
|
|
});
|
|
}
|
|
|
|
// Not allowed to be copied using std::memcpy or cudaMemcpy. Cannot use TMA copies.
|
|
struct no_copy
|
|
{
|
|
nvbench::uint32_t a;
|
|
|
|
no_copy() = default;
|
|
|
|
_CCCL_HOST_DEVICE no_copy(nvbench::uint32_t i)
|
|
: a(i)
|
|
{}
|
|
|
|
// the user-defined copy constructor prevents the type from being trivially copyable
|
|
// NOLINTNEXTLINE(modernize-use-equals-default)
|
|
_CCCL_HOST_DEVICE no_copy(const no_copy& nt)
|
|
: a(nt.a)
|
|
{}
|
|
};
|
|
|
|
static_assert(::cuda::std::is_trivially_default_constructible_v<no_copy>);
|
|
static_assert(!::cuda::std::is_trivially_copyable_v<no_copy>); // as required by the C++ standard
|
|
static_assert(!thrust::is_trivially_relocatable_v<no_copy>); // thrust uses this check internally
|
|
|
|
// Requires use of placement new
|
|
struct no_construct
|
|
{
|
|
nvbench::uint32_t a = 1337;
|
|
};
|
|
|
|
static_assert(!::cuda::std::is_trivially_default_constructible_v<no_construct>);
|
|
static_assert(::cuda::std::is_trivially_copyable_v<no_construct>); // as required by the C++ standard
|
|
static_assert(thrust::is_trivially_relocatable_v<no_construct>); // thrust uses this check internally
|
|
|
|
using types =
|
|
nvbench::type_list<nvbench::uint8_t, nvbench::uint16_t, nvbench::uint32_t, nvbench::uint64_t, no_copy, no_construct>;
|
|
|
|
NVBENCH_BENCH_TYPES(basic, NVBENCH_TYPE_AXES(types))
|
|
.set_name("base")
|
|
.set_type_axes_names({"T{ct}"})
|
|
.add_int64_power_of_two_axis("Elements", nvbench::range(16, 28, 4));
|