[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:
113
cccl_upstream/thrust/thrust/per_device_resource.h
Normal file
113
cccl_upstream/thrust/thrust/per_device_resource.h
Normal file
@@ -0,0 +1,113 @@
|
||||
// SPDX-FileCopyrightText: Copyright (c) 2008-2013, NVIDIA Corporation. All rights reserved.
|
||||
// SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <thrust/detail/config.h>
|
||||
|
||||
#if defined(_CCCL_IMPLICIT_SYSTEM_HEADER_GCC)
|
||||
# pragma GCC system_header
|
||||
#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_CLANG)
|
||||
# pragma clang system_header
|
||||
#elif defined(_CCCL_IMPLICIT_SYSTEM_HEADER_MSVC)
|
||||
# pragma system_header
|
||||
#endif // no system header
|
||||
#include <thrust/detail/execution_policy.h>
|
||||
#include <thrust/mr/allocator.h>
|
||||
|
||||
// Include all active backend system implementations (generic, sequential, host and device)
|
||||
#include <thrust/system/detail/generic/per_device_resource.h>
|
||||
#include <thrust/system/detail/sequential/per_device_resource.h>
|
||||
#include __THRUST_HOST_SYSTEM_ALGORITH_DETAIL_HEADER_INCLUDE(per_device_resource.h)
|
||||
#include __THRUST_DEVICE_SYSTEM_ALGORITH_DETAIL_HEADER_INCLUDE(per_device_resource.h)
|
||||
|
||||
// Some build systems need a hint to know which files we actually include
|
||||
#if 0
|
||||
# include <thrust/system/cpp/detail/per_device_resource.h>
|
||||
# include <thrust/system/cuda/detail/per_device_resource.h>
|
||||
# include <thrust/system/omp/detail/per_device_resource.h>
|
||||
# include <thrust/system/tbb/detail/per_device_resource.h>
|
||||
#endif
|
||||
|
||||
THRUST_NAMESPACE_BEGIN
|
||||
|
||||
/*! Returns a global instance of \p MR for the current device of the provided system.
|
||||
*
|
||||
* \tparam MR type of a memory resource to get an instance from. Must be \p DefaultConstructible.
|
||||
* \param system execution policy for which the resource is requested.
|
||||
* \return a pointer to a global instance of \p MR for the current device.
|
||||
*
|
||||
* \verbatim embed:rst:leading-asterisk
|
||||
* .. versionadded:: 2.2.0
|
||||
* \endverbatim
|
||||
*/
|
||||
template <typename MR, typename DerivedPolicy>
|
||||
_CCCL_HOST MR* get_per_device_resource(const thrust::detail::execution_policy_base<DerivedPolicy>& system)
|
||||
{
|
||||
using thrust::system::detail::generic::get_per_device_resource;
|
||||
|
||||
return get_per_device_resource<MR>(thrust::detail::derived_cast(thrust::detail::strip_const(system)));
|
||||
}
|
||||
|
||||
/*! A helper allocator class that uses global per device instances of a given upstream memory resource. Requires the
|
||||
* memory resource to be default constructible.
|
||||
*
|
||||
* \tparam T the type that will be allocated by this allocator.
|
||||
* \tparam MR the upstream memory resource to use for memory allocation. Must derive from
|
||||
* \p thrust::mr::memory_resource and must be \p final.
|
||||
* \tparam ExecutionPolicy the execution policy of the system to be used to retrieve the resource for the current
|
||||
* device.
|
||||
*
|
||||
* \verbatim embed:rst:leading-asterisk
|
||||
* .. versionadded:: 2.2.0
|
||||
* \endverbatim
|
||||
*/
|
||||
template <typename T, typename Upstream, typename ExecutionPolicy>
|
||||
class per_device_allocator : public thrust::mr::allocator<T, Upstream>
|
||||
{
|
||||
using base = thrust::mr::allocator<T, Upstream>;
|
||||
|
||||
public:
|
||||
/*! The \p rebind metafunction provides the type of an \p per_device_allocator instantiated with another type.
|
||||
*
|
||||
* \tparam U the other type to use for instantiation.
|
||||
*
|
||||
* \verbatim embed:rst:leading-asterisk
|
||||
* .. versionadded:: 2.2.0
|
||||
* \endverbatim
|
||||
*/
|
||||
template <typename U>
|
||||
struct rebind
|
||||
{
|
||||
/*! The alias \p other gives the type of the rebound \p per_device_allocator.
|
||||
*/
|
||||
using other = per_device_allocator<U, Upstream, ExecutionPolicy>;
|
||||
};
|
||||
|
||||
/*! Default constructor. Uses \p get_global_resource to get the global instance of \p Upstream and initializes the
|
||||
* \p allocator base subobject with that resource.
|
||||
*
|
||||
* \verbatim embed:rst:leading-asterisk
|
||||
* .. versionadded:: 2.2.0
|
||||
* \endverbatim
|
||||
*/
|
||||
_CCCL_HOST per_device_allocator()
|
||||
: base(get_per_device_resource<Upstream>(ExecutionPolicy()))
|
||||
{}
|
||||
|
||||
/*! Copy constructor. Copies the memory resource pointer. */
|
||||
_CCCL_HOST_DEVICE per_device_allocator(const per_device_allocator& other)
|
||||
: base(other)
|
||||
{}
|
||||
|
||||
/*! Conversion constructor from an allocator of a different type. Copies the memory resource pointer. */
|
||||
template <typename U>
|
||||
_CCCL_HOST_DEVICE per_device_allocator(const per_device_allocator<U, Upstream, ExecutionPolicy>& other)
|
||||
: base(other)
|
||||
{}
|
||||
|
||||
/*! Destructor. */
|
||||
_CCCL_HOST_DEVICE ~per_device_allocator() {}
|
||||
};
|
||||
|
||||
THRUST_NAMESPACE_END
|
||||
Reference in New Issue
Block a user