[INFRA] Import NVIDIA/CCCL upstream as optimization reference library
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
This commit is contained in:
EngineX CI
167
cccl_upstream/cub/test/catch2_test_block_histogram.cu
Normal file
167
cccl_upstream/cub/test/catch2_test_block_histogram.cu
Normal file
@@ -0,0 +1,167 @@
|
||||
// SPDX-FileCopyrightText: Copyright (c) 2011-2022, NVIDIA CORPORATION. All rights reserved.
|
||||
// SPDX-License-Identifier: BSD-3
|
||||
|
||||
/******************************************************************************
|
||||
* Test of BlockHistogram utilities
|
||||
******************************************************************************/
|
||||
|
||||
// Ensure printing of CUDA runtime errors to console
|
||||
#define CUB_STDERR
|
||||
|
||||
#include <cub/block/block_histogram.cuh>
|
||||
|
||||
#include <limits>
|
||||
#include <string>
|
||||
|
||||
#include <c2h/catch2_test_helper.h>
|
||||
|
||||
template <int BINS,
|
||||
int BLOCK_THREADS,
|
||||
int ITEMS_PER_THREAD,
|
||||
cub::BlockHistogramAlgorithm ALGORITHM,
|
||||
typename T,
|
||||
typename HistoCounter>
|
||||
__global__ void block_histogram_kernel(T* d_samples, HistoCounter* d_histogram)
|
||||
{
|
||||
// Parameterize BlockHistogram type for our thread block
|
||||
using block_histogram_t = cub::BlockHistogram<T, BLOCK_THREADS, ITEMS_PER_THREAD, BINS, ALGORITHM>;
|
||||
|
||||
// Allocate temp storage in shared memory
|
||||
__shared__ typename block_histogram_t::TempStorage temp_storage;
|
||||
|
||||
// Per-thread tile data
|
||||
T data[ITEMS_PER_THREAD];
|
||||
cub::LoadDirectStriped<BLOCK_THREADS>(threadIdx.x, d_samples, data);
|
||||
|
||||
// Test histo (writing directly to histogram buffer in global)
|
||||
block_histogram_t(temp_storage).Histogram(data, d_histogram);
|
||||
}
|
||||
|
||||
template <int ItemsPerThread, int ThreadsInBlock, int Bins, cub::BlockHistogramAlgorithm Algorithm, typename SampleT>
|
||||
void block_histogram(c2h::device_vector<SampleT>& d_samples, c2h::device_vector<int>& d_histogram)
|
||||
{
|
||||
block_histogram_kernel<Bins, ThreadsInBlock, ItemsPerThread, Algorithm>
|
||||
<<<1, ThreadsInBlock>>>(thrust::raw_pointer_cast(d_samples.data()), thrust::raw_pointer_cast(d_histogram.data()));
|
||||
|
||||
REQUIRE(cudaSuccess == cudaPeekAtLastError());
|
||||
REQUIRE(cudaSuccess == cudaDeviceSynchronize());
|
||||
}
|
||||
|
||||
// %PARAM% TEST_BINS bins 32:256:1024
|
||||
|
||||
using types = c2h::type_list<std::uint8_t, std::uint16_t>;
|
||||
using threads_in_block = c2h::enum_type_list<int, 32, 96, 128>;
|
||||
using items_per_thread = c2h::enum_type_list<int, 1, 5>;
|
||||
using bins = c2h::enum_type_list<int, TEST_BINS>;
|
||||
using algorithms = c2h::enum_type_list<cub::BlockHistogramAlgorithm, cub::BLOCK_HISTO_SORT, cub::BLOCK_HISTO_ATOMIC>;
|
||||
|
||||
template <class TestType>
|
||||
struct params_t
|
||||
{
|
||||
using sample_t = typename c2h::get<0, TestType>;
|
||||
|
||||
static constexpr int items_per_thread = c2h::get<1, TestType>::value;
|
||||
static constexpr int threads_in_block = c2h::get<2, TestType>::value;
|
||||
static constexpr int bins = c2h::get<3, TestType>::value;
|
||||
static constexpr int num_samples = threads_in_block * items_per_thread;
|
||||
static constexpr cub::BlockHistogramAlgorithm algorithm = c2h::get<4, TestType>::value;
|
||||
};
|
||||
|
||||
C2H_TEST("Block histogram can be computed with uniform input",
|
||||
"[histogram][block]",
|
||||
types,
|
||||
items_per_thread,
|
||||
threads_in_block,
|
||||
bins,
|
||||
algorithms)
|
||||
{
|
||||
using params = params_t<TestType>;
|
||||
using sample_t = typename params::sample_t;
|
||||
|
||||
const sample_t uniform_value = static_cast<sample_t>(GENERATE_COPY(take(10, random(0, params::bins - 1))));
|
||||
|
||||
c2h::host_vector<sample_t> h_samples(params::num_samples, uniform_value);
|
||||
c2h::host_vector<int> h_reference(params::bins);
|
||||
h_reference[static_cast<std::size_t>(uniform_value)] = params::num_samples;
|
||||
|
||||
// Allocate problem device arrays
|
||||
c2h::device_vector<sample_t> d_samples = h_samples;
|
||||
c2h::device_vector<int> d_histogram(params::bins);
|
||||
|
||||
// Run kernel
|
||||
block_histogram<params::items_per_thread, params::threads_in_block, params::bins, params::algorithm>(
|
||||
d_samples, d_histogram);
|
||||
|
||||
REQUIRE(h_reference == d_histogram);
|
||||
}
|
||||
|
||||
template <typename SampleT>
|
||||
c2h::host_vector<int> compute_host_reference(int bins, const c2h::host_vector<SampleT>& h_samples)
|
||||
{
|
||||
c2h::host_vector<int> h_reference(bins);
|
||||
for (const SampleT& sample : h_samples)
|
||||
{
|
||||
h_reference[sample]++;
|
||||
}
|
||||
|
||||
return h_reference;
|
||||
}
|
||||
|
||||
C2H_TEST("Block histogram can be computed with modulo input",
|
||||
"[histogram][block]",
|
||||
types,
|
||||
items_per_thread,
|
||||
threads_in_block,
|
||||
bins,
|
||||
algorithms)
|
||||
{
|
||||
using params = params_t<TestType>;
|
||||
using sample_t = typename params::sample_t;
|
||||
|
||||
// Allocate problem device arrays
|
||||
c2h::device_vector<int> d_histogram(params::bins);
|
||||
c2h::device_vector<sample_t> d_samples(params::num_samples);
|
||||
|
||||
c2h::gen(c2h::modulo_t{params::bins}, d_samples);
|
||||
|
||||
c2h::host_vector<sample_t> h_samples = d_samples;
|
||||
auto h_reference = compute_host_reference(params::bins, h_samples);
|
||||
|
||||
// Run kernel
|
||||
block_histogram<params::items_per_thread, params::threads_in_block, params::bins, params::algorithm>(
|
||||
d_samples, d_histogram);
|
||||
|
||||
REQUIRE(h_reference == d_histogram);
|
||||
}
|
||||
|
||||
C2H_TEST("Block histogram can be computed with random input",
|
||||
"[histogram][block]",
|
||||
types,
|
||||
items_per_thread,
|
||||
threads_in_block,
|
||||
bins,
|
||||
algorithms)
|
||||
{
|
||||
using params = params_t<TestType>;
|
||||
using sample_t = typename params::sample_t;
|
||||
|
||||
// Allocate problem device arrays
|
||||
c2h::device_vector<int> d_histogram(params::bins);
|
||||
c2h::device_vector<sample_t> d_samples(params::num_samples);
|
||||
|
||||
const sample_t min_bin = static_cast<sample_t>(0);
|
||||
const sample_t max_bin = static_cast<sample_t>(
|
||||
std::min(static_cast<std::int32_t>(cuda::std::numeric_limits<sample_t>::max()),
|
||||
static_cast<std::int32_t>(params::bins - 1)));
|
||||
|
||||
c2h::gen(C2H_SEED(10), d_samples, min_bin, max_bin);
|
||||
|
||||
c2h::host_vector<sample_t> h_samples = d_samples;
|
||||
auto h_reference = compute_host_reference(params::bins, h_samples);
|
||||
|
||||
// Run kernel
|
||||
block_histogram<params::items_per_thread, params::threads_in_block, params::bins, params::algorithm>(
|
||||
d_samples, d_histogram);
|
||||
|
||||
REQUIRE(h_reference == d_histogram);
|
||||
}
|
||||
Reference in New Issue
Block a user