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
110 lines
6.5 KiB
Plaintext
110 lines
6.5 KiB
Plaintext
#include <thrust/detail/allocator/allocator_system.h>
|
|
#include <thrust/functional.h>
|
|
#include <thrust/transform.h>
|
|
|
|
#include <cuda/iterator>
|
|
|
|
#include <unittest/unittest.h>
|
|
|
|
static const size_t num_samples = 10000;
|
|
|
|
template <typename Vector, typename U>
|
|
struct rebind_vector;
|
|
|
|
template <typename T, typename U, typename Allocator>
|
|
struct rebind_vector<thrust::host_vector<T, Allocator>, U>
|
|
{
|
|
using alloc_traits = typename cuda::std::allocator_traits<Allocator>;
|
|
using new_alloc = typename alloc_traits::template rebind_alloc<U>;
|
|
using type = thrust::host_vector<U, new_alloc>;
|
|
};
|
|
|
|
template <typename T, typename U, typename Allocator>
|
|
struct rebind_vector<thrust::device_vector<T, Allocator>, U>
|
|
{
|
|
using type = thrust::device_vector<U, typename Allocator::template rebind<U>::other>;
|
|
};
|
|
|
|
template <typename T, typename U, typename Allocator>
|
|
struct rebind_vector<thrust::universal_vector<T, Allocator>, U>
|
|
{
|
|
using type = thrust::universal_vector<U, typename Allocator::template rebind<U>::other>;
|
|
};
|
|
|
|
#define BINARY_FUNCTIONAL_PLACEHOLDERS_TEST(name, op, reference_functor, type_list) \
|
|
template <typename Vector> \
|
|
struct TestFunctionalPlaceholders##name \
|
|
{ \
|
|
void operator()(const size_t) \
|
|
{ \
|
|
constexpr size_t NUM_SAMPLES = 10000; \
|
|
constexpr size_t ZERO = 0; \
|
|
using T = typename Vector::value_type; \
|
|
Vector lhs = unittest::random_samples<T>(NUM_SAMPLES); \
|
|
Vector rhs = unittest::random_samples<T>(NUM_SAMPLES); \
|
|
thrust::replace(rhs.begin(), rhs.end(), T(0), T(1)); \
|
|
\
|
|
Vector reference(lhs.size()); \
|
|
Vector result(lhs.size()); \
|
|
using namespace thrust::placeholders; \
|
|
\
|
|
thrust::transform(lhs.begin(), lhs.end(), rhs.begin(), reference.begin(), reference_functor<T>()); \
|
|
thrust::transform(lhs.begin(), lhs.end(), rhs.begin(), result.begin(), _1 op _2); \
|
|
ASSERT_ALMOST_EQUAL(reference, result); \
|
|
\
|
|
thrust::transform( \
|
|
lhs.begin(), lhs.end(), cuda::make_constant_iterator<T>(1), reference.begin(), reference_functor<T>()); \
|
|
thrust::transform(lhs.begin(), lhs.end(), result.begin(), _1 op T(1)); \
|
|
ASSERT_ALMOST_EQUAL(reference, result); \
|
|
\
|
|
thrust::transform( \
|
|
cuda::make_constant_iterator<T>(1, ZERO), \
|
|
cuda::make_constant_iterator<T>(1, NUM_SAMPLES), \
|
|
rhs.begin(), \
|
|
reference.begin(), \
|
|
reference_functor<T>()); \
|
|
thrust::transform(rhs.begin(), rhs.end(), result.begin(), T(1) op _1); \
|
|
ASSERT_ALMOST_EQUAL(reference, result); \
|
|
} \
|
|
}; \
|
|
VectorUnitTest<TestFunctionalPlaceholders##name, type_list, thrust::device_vector, thrust::device_allocator> \
|
|
TestFunctionalPlaceholders##name##DeviceInstance; \
|
|
VectorUnitTest<TestFunctionalPlaceholders##name, type_list, thrust::host_vector, std::allocator> \
|
|
TestFunctionalPlaceholders##name##HostInstance;
|
|
|
|
_CCCL_DIAG_PUSH
|
|
_CCCL_DIAG_SUPPRESS_MSVC(4244) // warning C4244: '=': conversion from 'int' to '_Ty', possible loss of data
|
|
|
|
BINARY_FUNCTIONAL_PLACEHOLDERS_TEST(BitAnd, &, ::cuda::std::bit_and, SmallIntegralTypes);
|
|
BINARY_FUNCTIONAL_PLACEHOLDERS_TEST(BitOr, |, ::cuda::std::bit_or, SmallIntegralTypes);
|
|
BINARY_FUNCTIONAL_PLACEHOLDERS_TEST(BitXor, ^, ::cuda::std::bit_xor, SmallIntegralTypes);
|
|
|
|
template <typename T>
|
|
struct bit_negate_reference
|
|
{
|
|
_CCCL_HOST_DEVICE T operator()(const T& x) const
|
|
{
|
|
return ~x;
|
|
}
|
|
};
|
|
|
|
template <typename Vector>
|
|
void TestFunctionalPlaceholdersBitNegate()
|
|
{
|
|
using T = typename Vector::value_type;
|
|
using bool_vector = typename rebind_vector<Vector, bool>::type;
|
|
Vector input = unittest::random_samples<T>(num_samples);
|
|
|
|
bool_vector reference(input.size());
|
|
thrust::transform(input.begin(), input.end(), reference.begin(), bit_negate_reference<T>());
|
|
|
|
using namespace thrust::placeholders;
|
|
bool_vector result(input.size());
|
|
thrust::transform(input.begin(), input.end(), result.begin(), ~_1);
|
|
|
|
ASSERT_EQUAL(reference, result);
|
|
}
|
|
DECLARE_INTEGRAL_VECTOR_UNITTEST(TestFunctionalPlaceholdersBitNegate);
|
|
|
|
_CCCL_DIAG_POP
|