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
244 lines
13 KiB
Plaintext
244 lines
13 KiB
Plaintext
//===----------------------------------------------------------------------===//
|
|
//
|
|
// Part of libcu++, the C++ Standard Library for your entire system,
|
|
// 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) 2026 NVIDIA CORPORATION & AFFILIATES.
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
#ifndef _CUDA_STD_VERSION
|
|
#define _CUDA_STD_VERSION
|
|
|
|
#include <cuda/std/detail/__config> // IWYU pragma: export
|
|
|
|
#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
|
|
|
|
// We need to define our own macros to not conflict with the host stl.
|
|
// At the same time we want bring in all feature test macros from host
|
|
#if __has_include(<version>) // <version> should be the smallest include possible
|
|
# include <version>
|
|
#elif _CCCL_HOSTED()
|
|
# include <ciso646> // otherwise go for the smallest possible header
|
|
#endif // _CCCL_HOSTED()
|
|
|
|
#define __cccl_lib_bitops 201907L
|
|
#define __cccl_lib_bool_constant 201505L
|
|
#define __cccl_lib_bounded_array_traits 201902L
|
|
#define __cccl_lib_byte 201603L
|
|
#define __cccl_lib_byteswap 202110L
|
|
#define __cccl_lib_clamp 201603L
|
|
#define __cccl_lib_endian 201907L
|
|
#define __cccl_lib_forward_like 202207L
|
|
#define __cccl_lib_gcd_lcm 201606L
|
|
#define __cccl_lib_integer_comparison_functions 202002L
|
|
#define __cccl_lib_integer_sequence 202511L
|
|
#define __cccl_lib_integral_constant_callable 201304L
|
|
#define __cccl_lib_is_final 201402L
|
|
#define __cccl_lib_is_nothrow_convertible 201806L
|
|
#define __cccl_lib_is_null_pointer 201309L
|
|
#define __cccl_lib_is_scoped_enum 202011L
|
|
#define __cccl_lib_is_swappable 201603L
|
|
#if defined(_CCCL_BUILTIN_IS_VIRTUAL_BASE_OF)
|
|
# define __cccl_lib_is_virtual_base_of 202406L
|
|
#endif // _CCCL_BUILTIN_IS_VIRTUAL_BASE_OF
|
|
#define __cccl_lib_launder 201606L
|
|
#define __cccl_lib_logical_traits 201510L
|
|
#define __cccl_lib_math_constants 201907L
|
|
#if defined(_CCCL_BUILTIN_REFERENCE_CONSTRUCTS_FROM_TEMPORARY) \
|
|
&& defined(_CCCL_BUILTIN_REFERENCE_CONVERTS_FROM_TEMPORARY)
|
|
# define __cccl_lib_reference_from_temporary 202202L
|
|
#endif // _CCCL_BUILTIN_REFERENCE_CONSTRUCTS_FROM_TEMPORARY && _CCCL_BUILTIN_REFERENCE_CONVERTS_FROM_TEMPORARY
|
|
#define __cccl_lib_shift 201806L
|
|
#define __cccl_lib_to_address 201711L
|
|
#define __cccl_lib_to_array 201907L
|
|
#define __cccl_lib_to_underlying 202102L
|
|
#define __cccl_lib_transparent_operators 201210L
|
|
#define __cccl_lib_transformation_trait_aliases 201304L
|
|
#define __cccl_lib_tuple_element_t 201402L
|
|
// #define __cccl_lib_tuple_like 202311L // P2819R2 is implemented, but P2165R4 is not yet
|
|
#define __cccl_lib_type_identity 201806L
|
|
#define __cccl_lib_type_trait_variable_templates 201510L
|
|
#define __cccl_lib_unreachable 202202L
|
|
#define __cccl_lib_void_t 201411L
|
|
|
|
#define __cccl_lib_as_const 201510L
|
|
#define __cccl_lib_bit_cast 201806L
|
|
#define __cccl_lib_chrono_udls 201304L
|
|
#define __cccl_lib_complex_udls 201309L
|
|
#ifdef _CCCL_BUILTIN_IS_CONSTANT_EVALUATED
|
|
# define __cccl_lib_constexpr_complex 201711L
|
|
#endif // _CCCL_BUILTIN_IS_CONSTANT_EVALUATED
|
|
#define __cccl_lib_concepts 202002L
|
|
#define __cccl_lib_exchange_function 201304L
|
|
#define __cccl_lib_expected 202606L
|
|
// # define __cccl_lib_generic_associative_lookup 201304L
|
|
#define __cccl_lib_invoke 201411L
|
|
#define __cccl_lib_invoke_r 202106L
|
|
#define __cccl_lib_is_invocable 201703L
|
|
#define __cccl_lib_make_reverse_iterator 201402L
|
|
// # define __cccl_lib_make_unique 201304L
|
|
#define __cccl_lib_mdspan 202207L
|
|
#define __cccl_lib_null_iterators 201304L
|
|
#define __cccl_lib_optional 202506L
|
|
// # define __cccl_lib_quoted_string_io 201304L
|
|
#define __cccl_lib_result_of_sfinae 201210L
|
|
#define __cccl_lib_robust_nonmodifying_seq_ops 201304L
|
|
// # define __cccl_lib_shared_timed_mutex 201402L
|
|
#define __cccl_lib_source_location 201907L
|
|
#define __cccl_lib_span 202311L
|
|
// # define __cccl_lib_string_udls 201304L
|
|
#define __cccl_lib_submdspan 202207L
|
|
#define __cccl_lib_tuples_by_type 201304L
|
|
|
|
#ifdef _CCCL_BUILTIN_ADDRESSOF
|
|
# define __cccl_lib_addressof_constexpr 201603L
|
|
#endif // _CCCL_BUILTIN_ADDRESSOF
|
|
// # define __cccl_lib_allocator_traits_is_always_equal 201411L
|
|
// # define __cccl_lib_any 201606L
|
|
#define __cccl_lib_apply 201603L
|
|
#define __cccl_lib_array_constexpr 201603L
|
|
#define __cccl_lib_atomic_is_always_lock_free 201603L
|
|
#define __cccl_lib_bind_front 201907L
|
|
// # define __cccl_lib_boyer_moore_searcher 201603L
|
|
#define __cccl_lib_chrono 201611L
|
|
// # define __cccl_lib_enable_shared_from_this 201603L
|
|
// # define __cccl_lib_execution 201603L
|
|
// # define __cccl_lib_filesystem 201703L
|
|
// # define __cccl_lib_hardware_interference_size 201703L
|
|
#define __cccl_lib_has_unique_object_representations 201606L
|
|
#define __cccl_lib_hypot 201603L
|
|
// # define __cccl_lib_incomplete_container_elements 201505L
|
|
#ifndef _LIBCUDACXX_HAS_NO_IS_AGGREGATE
|
|
# define __cccl_lib_is_aggregate 201703L
|
|
#endif // _LIBCUDACXX_HAS_NO_IS_AGGREGATE
|
|
#if defined(_CCCL_BUILTIN_IS_LAYOUT_COMPATIBLE) && defined(_CCCL_BUILTIN_IS_CORRESPONDING_MEMBER)
|
|
# define __cccl_lib_is_layout_compatible 201907L
|
|
#endif // _CCCL_BUILTIN_IS_LAYOUT_COMPATIBLE && _CCCL_BUILTIN_IS_CORRESPONDING_MEMBER
|
|
#if defined(_CCCL_BUILTIN_IS_POINTER_INTERCONVERTIBLE_BASE_OF) \
|
|
&& defined(_CCCL_BUILTIN_IS_POINTER_INTERCONVERTIBLE_WITH_CLASS)
|
|
# define __cccl_lib_is_pointer_interconvertible 201907L
|
|
#endif // _BUILTIN_IS_POINTER_INTERCONVERTIBLE_BASE_OF && _CCCL_BUILTIN_IS_POINTER_INTERCONVERTIBLE_WITH_CLASS
|
|
#define __cccl_lib_make_from_tuple 201606L
|
|
// # define __cccl_lib_map_try_emplace 201411L
|
|
// # define __cccl_lib_math_special_functions 201603L
|
|
// # define __cccl_lib_memory_resource 201603L
|
|
// # define __cccl_lib_node_extract 201606L
|
|
// # define __cccl_lib_nonmember_container_access 201411L
|
|
#define __cccl_lib_not_fn 201603L
|
|
// # define __cccl_lib_parallel_algorithm 201603L
|
|
// # define __cccl_lib_raw_memory_algorithms 201606L
|
|
// # define __cccl_lib_sample 201603L
|
|
// # define __cccl_lib_scoped_lock 201703L
|
|
// # define __cccl_lib_shared_mutex 201505L
|
|
// # define __cccl_lib_shared_ptr_arrays 201611L
|
|
// # define __cccl_lib_shared_ptr_weak_type 201606L
|
|
#define __cccl_lib_string_view 201803L
|
|
// # define __cccl_lib_to_chars 201611L
|
|
// # define __cccl_lib_uncaught_exceptions 201411L
|
|
// # define __cccl_lib_unordered_map_try_emplace 201411L
|
|
#define __cccl_lib_variant 201606L
|
|
|
|
#define __cccl_lib_saturation_arithmetic 202603L
|
|
|
|
#if _CCCL_STD_VER >= 2020
|
|
# undef __cccl_lib_array_constexpr
|
|
# define __cccl_lib_array_constexpr 201811L
|
|
// # define __cccl_lib_assume_aligned 201811L
|
|
# define __cccl_lib_atomic_flag_test 201907L
|
|
# define __cccl_lib_atomic_float 201711L
|
|
# define __cccl_lib_atomic_lock_free_type_aliases 201907L
|
|
# define __cccl_lib_atomic_ref 202603L
|
|
// # define __cccl_lib_atomic_shared_ptr 201711L
|
|
# define __cccl_lib_atomic_value_initialization 201911L
|
|
# define __cccl_lib_atomic_wait 201907L
|
|
# define __cccl_lib_barrier 201907L
|
|
# define __cccl_lib_bit_cast 201806L
|
|
# if _CCCL_HAS_CHAR8_T()
|
|
# define __cccl_lib_char8_t 201811L
|
|
# endif // _CCCL_HAS_CHAR8_T()
|
|
// # define __cccl_lib_constexpr_algorithms 201806L
|
|
// # define __cccl_lib_constexpr_dynamic_alloc 201907L
|
|
# define __cccl_lib_constexpr_functional 201907L
|
|
// # define __cccl_lib_constexpr_iterator 201811L
|
|
// # define __cccl_lib_constexpr_memory 201811L
|
|
// # define __cccl_lib_constexpr_misc 201811L
|
|
// # define __cccl_lib_constexpr_numeric 201911L
|
|
// # define __cccl_lib_constexpr_string 201907L
|
|
// # define __cccl_lib_constexpr_swap_algorithms 201806L
|
|
// # define __cccl_lib_constexpr_tuple 201811L
|
|
// # define __cccl_lib_constexpr_utility 201811L
|
|
// # define __cccl_lib_constexpr_vector 201907L
|
|
// # define __cccl_lib_coroutine 201902L
|
|
# if __cpp_impl_destroying_delete >= 201806L && __cpp_lib_destroying_delete >= 201806L
|
|
# define __cccl_lib_destroying_delete 201806L
|
|
# endif // __cpp_impl_destroying_delete >= 201806L && __cpp_lib_destroying_delete >= 201806L
|
|
// # define __cccl_lib_erase_if 201811L
|
|
// # undef __cccl_lib_execution
|
|
// # define __cccl_lib_execution 201902L
|
|
// # define __cccl_lib_format 202106L
|
|
// # define __cccl_lib_generic_unordered_lookup 201811L
|
|
// # define __cccl_lib_int_pow2 202002L
|
|
// # define __cccl_lib_interpolate 201902L
|
|
# ifdef _CCCL_BUILTIN_IS_CONSTANT_EVALUATED
|
|
# define __cccl_lib_is_constant_evaluated 201811L
|
|
# endif // _CCCL_BUILTIN_IS_CONSTANT_EVALUATED
|
|
// # define __cccl_lib_jthread 201911L
|
|
// # define __cccl_lib_latch 201907L
|
|
// # define __cccl_lib_list_remove_return_type 201806L
|
|
// # define __cccl_lib_polymorphic_allocator 201902L
|
|
// # define __cccl_lib_ranges 201811L
|
|
// # define __cccl_lib_remove_cvref 201711L
|
|
// # define __cccl_lib_semaphore 201907L
|
|
// # undef __cccl_lib_shared_ptr_arrays
|
|
// # define __cccl_lib_shared_ptr_arrays 201707L
|
|
// # define __cccl_lib_smart_ptr_for_overwrite 202002L
|
|
// # define __cccl_lib_source_location 201907L
|
|
// # define __cccl_lib_ssize 201902L
|
|
// # define __cccl_lib_starts_ends_with 201711L
|
|
// # define __cccl_lib_syncbuf 201803L
|
|
// # define __cccl_lib_three_way_comparison 201907L
|
|
# define __cccl_lib_unwrap_ref 201811L
|
|
|
|
// todo(dabayer): Uncomment this once constant_wrapper is exposed.
|
|
// # define __cccl_lib_constant_wrapper 202603L
|
|
#endif // _CCCL_STD_VER >= 2020
|
|
|
|
#if _CCCL_STD_VER >= 2023
|
|
// # define __cccl_lib_adaptor_iterator_pair_constructor 202106L
|
|
// # define __cccl_lib_allocate_at_least 202106L
|
|
// # define __cccl_lib_associative_heterogeneous_erasure 202110L
|
|
// # define __cccl_lib_bind_back 202202L
|
|
// # define __cccl_lib_constexpr_bitset 202207L
|
|
// # define __cccl_lib_constexpr_charconv 202207L
|
|
// # define __cccl_lib_constexpr_cmath 202202L
|
|
// # undef __cccl_lib_constexpr_memory
|
|
// # define __cccl_lib_constexpr_memory 202202L
|
|
// # define __cccl_lib_constexpr_typeinfo 202106L
|
|
// # define __cccl_lib_move_only_function 202110L
|
|
// # define __cccl_lib_out_ptr 202106L
|
|
// # define __cccl_lib_ranges_chunk 202202L
|
|
// # define __cccl_lib_ranges_chunk_by 202202L
|
|
// # define __cccl_lib_ranges_iota 202202L
|
|
// # define __cccl_lib_ranges_join_with 202202L
|
|
// # define __cccl_lib_ranges_slide 202202L
|
|
// # define __cccl_lib_ranges_starts_ends_with 202106L
|
|
// # define __cccl_lib_ranges_to_container 202202L
|
|
// # define __cccl_lib_ranges_zip 202110L
|
|
// # define __cccl_lib_reference_from_temporary 202202L
|
|
// # define __cccl_lib_spanstream 202106L
|
|
// # define __cccl_lib_stacktrace 202011L
|
|
// # define __cccl_lib_stdatomic_h 202011L
|
|
// # define __cccl_lib_string_contains 202011L
|
|
// # define __cccl_lib_string_resize_and_overwrite 202110L
|
|
#endif // _CCCL_STD_VER >= 2023
|
|
|
|
#endif // _CUDA_STD_VERSION
|