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
65 lines
2.0 KiB
Plaintext
65 lines
2.0 KiB
Plaintext
#include <thrust/scan.h>
|
|
#include <thrust/transform.h>
|
|
|
|
#include <cuda/std/tuple>
|
|
|
|
#include <unittest/unittest.h>
|
|
|
|
#if THRUST_DEVICE_SYSTEM == THRUST_DEVICE_SYSTEM_CUDA
|
|
# include <unittest/cuda/testframework.h>
|
|
#endif
|
|
|
|
using namespace unittest;
|
|
|
|
struct SumTupleFunctor
|
|
{
|
|
template <typename Tuple>
|
|
_CCCL_HOST_DEVICE Tuple operator()(const Tuple& lhs, const Tuple& rhs)
|
|
{
|
|
using cuda::std::get;
|
|
return cuda::std::tuple(get<0>(lhs) + get<0>(rhs), get<1>(lhs) + get<1>(rhs));
|
|
}
|
|
};
|
|
|
|
struct MakeTupleFunctor
|
|
{
|
|
template <typename T1, typename T2>
|
|
_CCCL_HOST_DEVICE cuda::std::tuple<T1, T2> operator()(T1& lhs, T2& rhs)
|
|
{
|
|
return cuda::std::tuple(lhs, rhs);
|
|
}
|
|
};
|
|
|
|
template <typename T>
|
|
struct TestTupleScan
|
|
{
|
|
void operator()(const size_t n)
|
|
{
|
|
thrust::host_vector<T> h_t1 = unittest::random_integers<T>(n);
|
|
thrust::host_vector<T> h_t2 = unittest::random_integers<T>(n);
|
|
|
|
// initialize input
|
|
thrust::host_vector<cuda::std::tuple<T, T>> h_input(n);
|
|
thrust::transform(h_t1.begin(), h_t1.end(), h_t2.begin(), h_input.begin(), MakeTupleFunctor());
|
|
thrust::device_vector<cuda::std::tuple<T, T>> d_input = h_input;
|
|
|
|
// allocate output
|
|
cuda::std::tuple<T, T> zero(0, 0);
|
|
thrust::host_vector<cuda::std::tuple<T, T>> h_output(n, zero);
|
|
thrust::device_vector<cuda::std::tuple<T, T>> d_output(n, zero);
|
|
|
|
// inclusive_scan
|
|
thrust::inclusive_scan(h_input.begin(), h_input.end(), h_output.begin(), SumTupleFunctor());
|
|
thrust::inclusive_scan(d_input.begin(), d_input.end(), d_output.begin(), SumTupleFunctor());
|
|
ASSERT_EQUAL_QUIET(h_output, d_output);
|
|
|
|
// exclusive_scan
|
|
cuda::std::tuple<T, T> init(13, 17);
|
|
thrust::exclusive_scan(h_input.begin(), h_input.end(), h_output.begin(), init, SumTupleFunctor());
|
|
thrust::exclusive_scan(d_input.begin(), d_input.end(), d_output.begin(), init, SumTupleFunctor());
|
|
|
|
ASSERT_EQUAL_QUIET(h_output, d_output);
|
|
}
|
|
};
|
|
VariableUnitTest<TestTupleScan, IntegralTypes> TestTupleScanInstance;
|