forked from NVIDIA/cccl
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Implement tuple interface for cuda vector types
This specializes the `std::tuple_size` and `std::tuple_element` traits so that they are usable with cuda vector types. We also provide overloads for `std::get`, which together enables structured bindings support in C++17 onwards. Fixes NVIDIA#1406
- Loading branch information
Showing
6 changed files
with
887 additions
and
0 deletions.
There are no files selected for viewing
318 changes: 318 additions & 0 deletions
318
libcudacxx/include/cuda/std/detail/libcxx/include/__tuple_dir/vector_types.h
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,318 @@ | ||
| //===----------------------------------------------------------------------===// | ||
| // | ||
| // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. | ||
| // See https://llvm.org/LICENSE.txt for license information. | ||
| // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception | ||
| // SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. | ||
| // | ||
| //===----------------------------------------------------------------------===// | ||
|
|
||
| #ifndef _LIBCUDACXX___TUPLE_VECTOR_TYPES_H | ||
| #define _LIBCUDACXX___TUPLE_VECTOR_TYPES_H | ||
|
|
||
| #ifndef __cuda_std__ | ||
| # include <__config> | ||
| #endif // __cuda_std__ | ||
|
|
||
| #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 | ||
|
|
||
| #ifdef __cuda_std__ | ||
| # if defined(_CCCL_CUDA_COMPILER) | ||
|
|
||
| _CCCL_DIAG_PUSH | ||
| _CCCL_DIAG_SUPPRESS_CLANG("-Wmismatched-tags") | ||
|
|
||
| # if !defined(_CCCL_COMPILER_NVRTC) | ||
| // Fetch utility to get primary template for ::std::tuple_size necessary for the specialization of | ||
| // ::std::tuple_size<cuda::std::tuple> to enable structured bindings. | ||
| // See https://github.com/NVIDIA/libcudacxx/issues/316 | ||
| # include <utility> | ||
| # endif | ||
|
|
||
| # include "../__fwd/get.h" | ||
| # include "../__tuple_dir/structured_bindings.h" | ||
| # include "../__tuple_dir/tuple_element.h" | ||
| # include "../__tuple_dir/tuple_size.h" | ||
| # include "../__type_traits/integral_constant.h" | ||
| # include "../__type_traits/enable_if.h" | ||
| # include "../__utility/forward.h" | ||
| # include "../__utility/move.h" | ||
|
|
||
| # define _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE(__name, __type, __size) \ | ||
| template <> \ | ||
| struct tuple_size<__name##__size> : _CUDA_VSTD::integral_constant<size_t, __size> \ | ||
| {}; \ | ||
| template <> \ | ||
| struct tuple_size<const __name##__size> : _CUDA_VSTD::integral_constant<size_t, __size> \ | ||
| {}; \ | ||
| template <> \ | ||
| struct tuple_size<volatile __name##__size> : _CUDA_VSTD::integral_constant<size_t, __size> \ | ||
| {}; \ | ||
| template <> \ | ||
| struct tuple_size<const volatile __name##__size> : _CUDA_VSTD::integral_constant<size_t, __size> \ | ||
| {}; \ | ||
| \ | ||
| template <size_t _Ip> \ | ||
| struct tuple_element<_Ip, __name##__size> \ | ||
| { \ | ||
| static_assert(_Ip < __size, "tuple_element index out of range"); \ | ||
| using type = __type; \ | ||
| }; \ | ||
| template <size_t _Ip> \ | ||
| struct tuple_element<_Ip, const __name##__size> \ | ||
| { \ | ||
| static_assert(_Ip < __size, "tuple_element index out of range"); \ | ||
| using type = const __type; \ | ||
| }; \ | ||
| template <size_t _Ip> \ | ||
| struct tuple_element<_Ip, volatile __name##__size> \ | ||
| { \ | ||
| static_assert(_Ip < __size, "tuple_element index out of range"); \ | ||
| using type = volatile __type; \ | ||
| }; \ | ||
| template <size_t _Ip> \ | ||
| struct tuple_element<_Ip, const volatile __name##__size> \ | ||
| { \ | ||
| static_assert(_Ip < __size, "tuple_element index out of range"); \ | ||
| using type = const volatile __type; \ | ||
| }; | ||
|
|
||
| # define _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(__name, __type) \ | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE(__name, __type, 1) \ | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE(__name, __type, 2) \ | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE(__name, __type, 3) \ | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE(__name, __type, 4) | ||
|
|
||
| # define _LIBCUDACXX_SPECIALIZE_GET(__name, __base_type) \ | ||
| template <size_t _Ip> \ | ||
| _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 __base_type& get(__name& __val) noexcept \ | ||
| { \ | ||
| return _CUDA_VSTD::__get_element<_Ip>::template get<__name, __base_type>(__val); \ | ||
| } \ | ||
| template <size_t _Ip> \ | ||
| _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 const __base_type& get( \ | ||
| const __name& __val) noexcept \ | ||
| { \ | ||
| return _CUDA_VSTD::__get_element<_Ip>::template get<__name, __base_type>(__val); \ | ||
| } \ | ||
| template <size_t _Ip> \ | ||
| _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 __base_type&& get(__name&& __val) noexcept \ | ||
| { \ | ||
| return _CUDA_VSTD::__get_element<_Ip>::template get<__name, __base_type>(static_cast<__name&&>(__val)); \ | ||
| } \ | ||
| template <size_t _Ip> \ | ||
| _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 const __base_type&& get( \ | ||
| const __name&& __val) noexcept \ | ||
| { \ | ||
| return _CUDA_VSTD::__get_element<_Ip>::template get<__name, __base_type>(static_cast<const __name&&>(__val)); \ | ||
| } | ||
|
|
||
| # define _LIBCUDACXX_SPECIALIZE_GET_VECTOR(__name, __base_type) \ | ||
| _LIBCUDACXX_SPECIALIZE_GET(__name##1, __base_type) \ | ||
| _LIBCUDACXX_SPECIALIZE_GET(__name##2, __base_type) \ | ||
| _LIBCUDACXX_SPECIALIZE_GET(__name##3, __base_type) \ | ||
| _LIBCUDACXX_SPECIALIZE_GET(__name##4, __base_type) | ||
|
|
||
| _LIBCUDACXX_BEGIN_NAMESPACE_STD | ||
|
|
||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(char, signed char) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(uchar, unsigned char) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(short, short) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(ushort, unsigned short) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(int, int) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(uint, unsigned int) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(long, long) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(ulong, unsigned long) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(longlong, long long) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(ulonglong, unsigned long long) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(float, float) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(double, double) | ||
|
|
||
| template <size_t _Ip> | ||
| struct __get_element; | ||
|
|
||
| template <> | ||
| struct __get_element<0> | ||
| { | ||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 _BaseType& get(_Vec& __val) noexcept | ||
| { | ||
| return __val.x; | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 const _BaseType& get(const _Vec& __val) noexcept | ||
| { | ||
| return __val.x; | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 _BaseType&& get(_Vec&& __val) noexcept | ||
| { | ||
| return static_cast<_BaseType&&>(__val.x); | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 const _BaseType&& | ||
| get(const _Vec&& __val) noexcept | ||
| { | ||
| return static_cast<const _BaseType&&>(__val.x); | ||
| } | ||
| }; | ||
|
|
||
| template <> | ||
| struct __get_element<1> | ||
| { | ||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 _BaseType& get(_Vec& __val) noexcept | ||
| { | ||
| return __val.y; | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 const _BaseType& get(const _Vec& __val) noexcept | ||
| { | ||
| return __val.y; | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 _BaseType&& get(_Vec&& __val) noexcept | ||
| { | ||
| return static_cast<_BaseType&&>(__val.y); | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 const _BaseType&& | ||
| get(const _Vec&& __val) noexcept | ||
| { | ||
| return static_cast<const _BaseType&&>(__val.y); | ||
| } | ||
| }; | ||
| template <> | ||
| struct __get_element<2> | ||
| { | ||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 _BaseType& get(_Vec& __val) noexcept | ||
| { | ||
| return __val.z; | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 const _BaseType& get(const _Vec& __val) noexcept | ||
| { | ||
| return __val.z; | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 _BaseType&& get(_Vec&& __val) noexcept | ||
| { | ||
| return static_cast<_BaseType&&>(__val.z); | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 const _BaseType&& | ||
| get(const _Vec&& __val) noexcept | ||
| { | ||
| return static_cast<const _BaseType&&>(__val.z); | ||
| } | ||
| }; | ||
|
|
||
| template <> | ||
| struct __get_element<3> | ||
| { | ||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 _BaseType& get(_Vec& __val) noexcept | ||
| { | ||
| return __val.w; | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 const _BaseType& get(const _Vec& __val) noexcept | ||
| { | ||
| return __val.w; | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 _BaseType&& get(_Vec&& __val) noexcept | ||
| { | ||
| return static_cast<_BaseType&&>(__val.w); | ||
| } | ||
|
|
||
| template <class _Vec, class _BaseType> | ||
| static _LIBCUDACXX_INLINE_VISIBILITY _LIBCUDACXX_CONSTEXPR_AFTER_CXX11 const _BaseType&& | ||
| get(const _Vec&& __val) noexcept | ||
| { | ||
| return static_cast<const _BaseType&&>(__val.w); | ||
| } | ||
| }; | ||
|
|
||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(char, signed char) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(uchar, unsigned char) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(short, short) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(ushort, unsigned short) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(int, int) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(uint, unsigned int) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(long, long) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(ulong, unsigned long) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(longlong, long long) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(ulonglong, unsigned long long) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(float, float) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(double, double) | ||
|
|
||
| _LIBCUDACXX_END_NAMESPACE_STD | ||
|
|
||
| // Those need to be defined in the global namespace because we need ADL to find them | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(char, signed char) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(uchar, unsigned char) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(short, short) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(ushort, unsigned short) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(int, int) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(uint, unsigned int) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(long, long) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(ulong, unsigned long) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(longlong, long long) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(ulonglong, unsigned long long) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(float, float) | ||
| _LIBCUDACXX_SPECIALIZE_GET_VECTOR(double, double) | ||
|
|
||
| // This is a workaround for the fact that structured bindings require that the specializations of | ||
| // `tuple_size` and `tuple_element` reside in namespace std (https://eel.is/c++draft/dcl.struct.bind#4). | ||
| // See https://github.com/NVIDIA/libcudacxx/issues/316 for a short discussion | ||
| # if _CCCL_STD_VER >= 2017 | ||
| namespace std | ||
| { | ||
|
|
||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(char, signed char) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(uchar, unsigned char) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(short, short) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(ushort, unsigned short) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(int, int) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(uint, unsigned int) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(long, long) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(ulong, unsigned long) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(longlong, long long) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(ulonglong, unsigned long long) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(float, float) | ||
| _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR(double, double) | ||
|
|
||
| } // namespace std | ||
|
|
||
| # endif // _CCCL_STD_VER >= 2017 | ||
|
|
||
| # undef _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE | ||
| # undef _LIBCUDACXX_SPECIALIZE_TUPLE_INTERFACE_VECTOR | ||
| # undef _LIBCUDACXX_SPECIALIZE_GET | ||
| # undef _LIBCUDACXX_SPECIALIZE_GET_VECTOR | ||
|
|
||
| _CCCL_DIAG_POP | ||
|
|
||
| # endif // _CCCL_CUDA_COMPILER | ||
| #endif // __cuda_std__ | ||
|
|
||
| #endif // _LIBCUDACXX___TUPLE_VECTOR_TYPES_H |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.