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
55 lines
2.0 KiB
Plaintext
55 lines
2.0 KiB
Plaintext
// SPDX-FileCopyrightText: Copyright (c) 2011-2023, NVIDIA CORPORATION. All rights reserved.
|
|
// SPDX-License-Identifier: BSD-3
|
|
|
|
#pragma once
|
|
|
|
#include <thrust/device_vector.h>
|
|
#include <thrust/execution_policy.h>
|
|
#include <thrust/set_operations.h>
|
|
#include <thrust/sort.h>
|
|
|
|
#include "nvbench_helper.cuh"
|
|
|
|
template <typename T, typename OpT>
|
|
static void basic(nvbench::state& state, nvbench::type_list<T>, OpT op)
|
|
{
|
|
const auto elements = static_cast<std::size_t>(state.get_int64("Elements"));
|
|
const auto size_ratio = static_cast<std::size_t>(state.get_int64("SizeRatio"));
|
|
const bit_entropy entropy = str_to_entropy(state.get_string("Entropy"));
|
|
|
|
const auto elements_in_A = static_cast<std::size_t>(static_cast<double>(size_ratio * elements) / 100.0f);
|
|
|
|
thrust::device_vector<T> input = generate(elements, entropy);
|
|
thrust::device_vector<T> output(elements);
|
|
|
|
thrust::sort(input.begin(), input.begin() + elements_in_A);
|
|
thrust::sort(input.begin() + elements_in_A, input.end());
|
|
|
|
caching_allocator_t alloc;
|
|
// not a warm-up run, we need to run once to determine the size of the output
|
|
const auto result_ends =
|
|
op(policy(alloc),
|
|
input.cbegin(),
|
|
input.cbegin() + elements_in_A,
|
|
input.cbegin() + elements_in_A,
|
|
input.cend(),
|
|
output.begin());
|
|
const std::size_t elements_in_AB = ::cuda::std::distance(output.begin(), result_ends);
|
|
|
|
state.add_element_count(elements);
|
|
state.add_global_memory_reads<T>(elements);
|
|
state.add_global_memory_writes<T>(elements_in_AB);
|
|
|
|
state.exec(nvbench::exec_tag::gpu | nvbench::exec_tag::no_batch | nvbench::exec_tag::sync,
|
|
[&](nvbench::launch& launch) {
|
|
op(policy(alloc, launch),
|
|
input.cbegin(),
|
|
input.cbegin() + elements_in_A,
|
|
input.cbegin() + elements_in_A,
|
|
input.cend(),
|
|
output.begin());
|
|
});
|
|
}
|
|
|
|
using types = nvbench::type_list<int8_t, int16_t, int32_t, int64_t>;
|