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
255 lines
6.5 KiB
Plaintext
255 lines
6.5 KiB
Plaintext
#include <thrust/copy.h>
|
|
#include <thrust/iterator/counting_iterator.h>
|
|
#include <thrust/iterator/discard_iterator.h>
|
|
#include <thrust/iterator/retag.h>
|
|
#include <thrust/iterator/zip_iterator.h>
|
|
#include <thrust/sequence.h>
|
|
|
|
#include <cuda/iterator>
|
|
|
|
#include <iterator>
|
|
#include <list>
|
|
|
|
#include <unittest/unittest.h>
|
|
|
|
void TestCopyNFromConstIterator()
|
|
{
|
|
using T = int;
|
|
|
|
std::vector<T> v{0, 1, 2, 3, 4};
|
|
|
|
std::vector<int>::const_iterator begin = v.begin();
|
|
|
|
// copy to host_vector
|
|
thrust::host_vector<T> h(5, (T) 10);
|
|
thrust::host_vector<T>::iterator h_result = thrust::copy_n(begin, h.size(), h.begin());
|
|
|
|
ASSERT_EQUAL_QUIET(h_result, h.end());
|
|
|
|
// copy to device_vector
|
|
thrust::device_vector<T> d(5, (T) 10);
|
|
thrust::device_vector<T>::iterator d_result = thrust::copy_n(begin, d.size(), d.begin());
|
|
|
|
thrust::device_vector<T> dref{0, 1, 2, 3, 4};
|
|
ASSERT_EQUAL(d, dref);
|
|
ASSERT_EQUAL_QUIET(d_result, d.end());
|
|
}
|
|
DECLARE_UNITTEST(TestCopyNFromConstIterator);
|
|
|
|
void TestCopyNToDiscardIterator()
|
|
{
|
|
using T = int;
|
|
|
|
thrust::host_vector<T> h_input(5, 1);
|
|
thrust::device_vector<T> d_input = h_input;
|
|
|
|
// copy from host_vector
|
|
thrust::discard_iterator<> h_result =
|
|
thrust::copy_n(h_input.begin(), h_input.size(), thrust::make_discard_iterator());
|
|
|
|
// copy from device_vector
|
|
thrust::discard_iterator<> d_result =
|
|
thrust::copy_n(d_input.begin(), d_input.size(), thrust::make_discard_iterator());
|
|
|
|
thrust::discard_iterator<> reference(5);
|
|
|
|
ASSERT_EQUAL_QUIET(reference, h_result);
|
|
ASSERT_EQUAL_QUIET(reference, d_result);
|
|
}
|
|
DECLARE_UNITTEST(TestCopyNToDiscardIterator);
|
|
|
|
template <class Vector>
|
|
void TestCopyNMatchingTypes()
|
|
{
|
|
using T = typename Vector::value_type;
|
|
|
|
Vector v{0, 1, 2, 3, 4};
|
|
|
|
// copy to host_vector
|
|
thrust::host_vector<T> h(5, (T) 10);
|
|
typename thrust::host_vector<T>::iterator h_result = thrust::copy_n(v.begin(), v.size(), h.begin());
|
|
thrust::host_vector<T> href{0, 1, 2, 3, 4};
|
|
ASSERT_EQUAL(h, href);
|
|
ASSERT_EQUAL_QUIET(h_result, h.end());
|
|
|
|
// copy to device_vector
|
|
thrust::device_vector<T> d(5, (T) 10);
|
|
typename thrust::device_vector<T>::iterator d_result = thrust::copy_n(v.begin(), v.size(), d.begin());
|
|
thrust::device_vector<T> dref{0, 1, 2, 3, 4};
|
|
ASSERT_EQUAL(d, dref);
|
|
ASSERT_EQUAL_QUIET(d_result, d.end());
|
|
}
|
|
DECLARE_VECTOR_UNITTEST(TestCopyNMatchingTypes);
|
|
|
|
_CCCL_DIAG_PUSH
|
|
_CCCL_DIAG_SUPPRESS_MSVC(4244) // '=': conversion from 'int' to '_Ty', possible loss of data
|
|
|
|
template <class Vector>
|
|
void TestCopyNMixedTypes()
|
|
{
|
|
Vector v{0, 1, 2, 3, 4};
|
|
|
|
// copy to host_vector with different type
|
|
thrust::host_vector<float> h(5, (float) 10);
|
|
typename thrust::host_vector<float>::iterator h_result = thrust::copy_n(v.begin(), v.size(), h.begin());
|
|
|
|
thrust::host_vector<float> href{0, 1, 2, 3, 4};
|
|
ASSERT_EQUAL(h, href);
|
|
ASSERT_EQUAL_QUIET(h_result, h.end());
|
|
|
|
// copy to device_vector with different type
|
|
thrust::device_vector<float> d(5, (float) 10);
|
|
typename thrust::device_vector<float>::iterator d_result = thrust::copy_n(v.begin(), v.size(), d.begin());
|
|
thrust::device_vector<float> dref{0, 1, 2, 3, 4};
|
|
ASSERT_EQUAL(d, dref);
|
|
ASSERT_EQUAL_QUIET(d_result, d.end());
|
|
}
|
|
DECLARE_INTEGRAL_VECTOR_UNITTEST(TestCopyNMixedTypes);
|
|
|
|
_CCCL_DIAG_POP
|
|
|
|
void TestCopyNVectorBool()
|
|
{
|
|
std::vector<bool> v{true, false, true};
|
|
|
|
thrust::host_vector<bool> h(3);
|
|
thrust::device_vector<bool> d(3);
|
|
|
|
thrust::copy_n(v.begin(), v.size(), h.begin());
|
|
thrust::copy_n(v.begin(), v.size(), d.begin());
|
|
|
|
thrust::host_vector<bool> href{true, false, true};
|
|
ASSERT_EQUAL(h, href);
|
|
|
|
thrust::device_vector<bool> dref{true, false, true};
|
|
ASSERT_EQUAL(d, dref);
|
|
|
|
ASSERT_EQUAL(d, dref);
|
|
}
|
|
DECLARE_UNITTEST(TestCopyNVectorBool);
|
|
|
|
template <class Vector>
|
|
void TestCopyNListTo()
|
|
{
|
|
using T = typename Vector::value_type;
|
|
|
|
// copy from list to Vector
|
|
std::list<T> l{0, 1, 2, 3, 4};
|
|
|
|
Vector v(l.size());
|
|
|
|
typename Vector::iterator v_result = thrust::copy_n(l.begin(), l.size(), v.begin());
|
|
|
|
Vector ref{0, 1, 2, 3, 4};
|
|
ASSERT_EQUAL(v, ref);
|
|
ASSERT_EQUAL_QUIET(v_result, v.end());
|
|
|
|
l.clear();
|
|
|
|
thrust::copy_n(v.begin(), v.size(), std::back_insert_iterator<std::list<T>>(l));
|
|
|
|
ASSERT_EQUAL(l.size(), 5lu);
|
|
|
|
typename std::list<T>::const_iterator iter = l.begin();
|
|
ASSERT_EQUAL(*iter, T(0));
|
|
iter++;
|
|
ASSERT_EQUAL(*iter, T(1));
|
|
iter++;
|
|
ASSERT_EQUAL(*iter, T(2));
|
|
iter++;
|
|
ASSERT_EQUAL(*iter, T(3));
|
|
iter++;
|
|
ASSERT_EQUAL(*iter, T(4));
|
|
iter++;
|
|
}
|
|
DECLARE_VECTOR_UNITTEST(TestCopyNListTo);
|
|
|
|
template <typename Vector>
|
|
void TestCopyNCountingIterator()
|
|
{
|
|
using T = typename Vector::value_type;
|
|
|
|
thrust::counting_iterator<T> iter(1);
|
|
|
|
Vector vec(4);
|
|
|
|
thrust::copy_n(iter, 4, vec.begin());
|
|
|
|
Vector ref{1, 2, 3, 4};
|
|
ASSERT_EQUAL(vec, ref);
|
|
}
|
|
DECLARE_INTEGRAL_VECTOR_UNITTEST(TestCopyNCountingIterator);
|
|
|
|
template <typename Vector>
|
|
void TestCopyNZipIterator()
|
|
{
|
|
using T = typename Vector::value_type;
|
|
|
|
Vector v1{1, 2, 3, 4};
|
|
Vector v2{4, 5, 6, 7};
|
|
Vector v3(4, T(0));
|
|
Vector v4(4, T(0));
|
|
|
|
thrust::copy_n(
|
|
thrust::make_zip_iterator(v1.begin(), v2.begin()), 4, thrust::make_zip_iterator(v3.begin(), v4.begin()));
|
|
|
|
ASSERT_EQUAL(v1, v3);
|
|
ASSERT_EQUAL(v2, v4);
|
|
};
|
|
DECLARE_VECTOR_UNITTEST(TestCopyNZipIterator);
|
|
|
|
template <typename Vector>
|
|
void TestCopyNConstantIteratorToZipIterator()
|
|
{
|
|
using T = typename Vector::value_type;
|
|
|
|
Vector v1(4, T(0));
|
|
Vector v2(4, T(0));
|
|
|
|
thrust::copy_n(cuda::make_constant_iterator(cuda::std::tuple<T, T>(4, 7)),
|
|
v1.size(),
|
|
thrust::make_zip_iterator(v1.begin(), v2.begin()));
|
|
|
|
Vector ref1(4, 4);
|
|
Vector ref2(4, 7);
|
|
|
|
ASSERT_EQUAL(v1, ref1);
|
|
ASSERT_EQUAL(v2, ref2);
|
|
};
|
|
DECLARE_VECTOR_UNITTEST(TestCopyNConstantIteratorToZipIterator);
|
|
|
|
template <typename InputIterator, typename Size, typename OutputIterator>
|
|
OutputIterator copy_n(my_system& system, InputIterator, Size, OutputIterator result)
|
|
{
|
|
system.validate_dispatch();
|
|
return result;
|
|
}
|
|
|
|
void TestCopyNDispatchExplicit()
|
|
{
|
|
thrust::device_vector<int> vec(1);
|
|
|
|
my_system sys(0);
|
|
thrust::copy_n(sys, vec.begin(), 1, vec.begin());
|
|
|
|
ASSERT_EQUAL(true, sys.is_valid());
|
|
}
|
|
DECLARE_UNITTEST(TestCopyNDispatchExplicit);
|
|
|
|
template <typename InputIterator, typename Size, typename OutputIterator>
|
|
OutputIterator copy_n(my_tag, InputIterator, Size, OutputIterator result)
|
|
{
|
|
*result = 13;
|
|
return result;
|
|
}
|
|
|
|
void TestCopyNDispatchImplicit()
|
|
{
|
|
thrust::device_vector<int> vec(1);
|
|
|
|
thrust::copy_n(thrust::retag<my_tag>(vec.begin()), 1, thrust::retag<my_tag>(vec.begin()));
|
|
|
|
ASSERT_EQUAL(13, vec.front());
|
|
}
|
|
DECLARE_UNITTEST(TestCopyNDispatchImplicit);
|