[CCCL] 瘦身 + 补全: 移除 cudax/python/libcudacxx-tests 冗余文件, 新增 c2h 测试助手 + cmake 构建系统 + 8 个 CUDA thrust examples

变更摘要:
- 删除: cudax/ (783 files, 7.2M) — 实验性组件，竞赛不需要
- 删除: python/ (226 files, 2.0M) — Python 绑定，竞赛不需要
- 删除: libcudacxx/{test,benchmarks,codegen,cmake,share} (4432 files, 31M)
  保留: libcudacxx/include/ (1463 headers, cuda::std 编译依赖)
- 新增: c2h/ (27 files) — CUB Catch2 测试辅助头文件，编译 243 个测试必需
- 新增: cmake/ (29 files) — CCCL 原生 CMake 构建系统
- 新增: thrust/examples/cuda/ (7 files) + cpp_integration/ (1 file)
  async_reduce, custom_temporary_allocation, explicit_cuda_stream,
  global_device_vector, range_view, unwrap_pointer, wrap_pointer, device

结果: cccl_upstream 从 74M→35M (瘦身 53%), 核心内容 100% 保留:
  27/27 tuning headers, 78 benchmarks, 243 tests,
  60 thrust examples, 18 CUB examples, 全部编译头文件

cccl_upstream/cudax/cmake/cudaxAddSubdir.cmake

@@ -1 +0,0 @@
-cccl_add_subdir_helper(cudax)

cccl_upstream/cudax/cmake/cudaxBuildCompilerTargets.cmake

@@ -1,65 +0,0 @@
-# Including this file defines the following targets:
-#
-# cudax.compiler_interface
-# - Interface target that includes all compiler settings for cudax tests, etc.
-
-cccl_get_cub()
-cccl_get_cudax()
-cccl_get_libcudacxx()
-cccl_get_thrust()
-
-set(cuda_compile_options)
-set(cxx_compile_options)
-set(cxx_compile_definitions)
-
-if ("MSVC" STREQUAL "${CMAKE_CXX_COMPILER_ID}")
-  # C4848: support for attribute 'msvc::no_unique_address' in C++17 and earlier is a vendor extension
-  append_option_if_available("/wd4848" cxx_compile_options)
-
-  # XXX Temporary hack for STF !
-  # C4267: conversion from 'meow' to 'purr', possible loss of data
-  append_option_if_available("/wd4267" cxx_compile_options)
-
-  # C4459 : declaration of 'identifier' hides global declaration
-  # We work around std::chrono::last which hides some internal "last" variable
-  append_option_if_available("/wd4459" cxx_compile_options)
-
-  # stf used getenv which is potentially unsafe but not in our context
-  list(APPEND cxx_compile_definitions "_CRT_SECURE_NO_WARNINGS")
-endif()
-
-if ("Clang" STREQUAL "${CMAKE_CXX_COMPILER_ID}")
-  # stf heavily uses host device lambdas which break on clang due to a warning about the implicitly
-  # deleted copy constructor
-  # TODO(bgruber): remove this when NVBug 4980157 is resolved
-  append_option_if_available("-Wno-deprecated-copy" cxx_compile_options)
-endif()
-
-list(APPEND cxx_compile_definitions CCCL_ENABLE_ASSERTIONS)
-
-# Some groups related experimental code is located directly in libcu++ and is guarded by
-# _CUDAX_ENABLE_GROUP_FEATURES_IN_LIBCUDACXX macro, otherwise it would lead to a lot of code duplication. We define this
-# macro for cudax code globally, to get access to get access to the code.
-#
-# Can be removed once groups are no longer experimental.
-list(APPEND cxx_compile_definitions _CUDAX_ENABLE_GROUP_FEATURES_IN_LIBCUDACXX)
-
-cccl_build_compiler_interface(
-  cudax.compiler_flags
-  "${cuda_compile_options}"
-  "${cxx_compile_options}"
-  "${cxx_compile_definitions}"
-)
-
-add_library(cudax.compiler_interface INTERFACE)
-target_link_libraries(
-  cudax.compiler_interface
-  INTERFACE
-    # order matters here, we need the cudax options to override the cccl options.
-    cccl.compiler_interface
-    cudax.compiler_flags
-    libcudacxx::libcudacxx
-    CUB::CUB
-    Thrust::Thrust
-    cudax::cudax
-)

cccl_upstream/cudax/cmake/cudaxHeaderTesting.cmake

@@ -1,109 +0,0 @@
-# For every public header, build a translation unit containing `#include <header>`
-# to let the compiler try to figure out warnings in that header if it is not otherwise
-# included in tests, and also to verify if the headers are modular enough.
-# .inl files are not globbed for, because they are not supposed to be used as public
-# entrypoints.
-
-cccl_get_cudatoolkit()
-
-# Meta target for all configs' header builds:
-add_custom_target(cudax.all.headers)
-
-function(cudax_add_header_test label definitions)
-  ###################
-  # Non-STF headers #
-  set(headertest_target cudax.headers.${label}.no_stf)
-  cccl_generate_header_tests(
-    ${headertest_target}
-    cudax/include
-    # The cudax header template removes the check for the `small` macro.
-    HEADER_TEMPLATE "${cudax_SOURCE_DIR}/cmake/header_test.in.cu"
-    GLOBS "cuda/experimental/*.cuh"
-    EXCLUDES
-      # The following internal headers are not required to compile independently:
-      "cuda/experimental/__execution/prologue.cuh"
-      "cuda/experimental/__execution/epilogue.cuh"
-      # cuFile headers are compiled separately:
-      "cuda/experimental/cufile.cuh"
-      "cuda/experimental/__cufile/*"
-      # Places headers are compiled separately:
-      "cuda/experimental/places.cuh"
-      "cuda/experimental/__places/*"
-      # STF headers are compiled separately:
-      "cuda/experimental/stf.cuh"
-      "cuda/experimental/__stf/*"
-  )
-  target_link_libraries(${headertest_target} PUBLIC cudax.compiler_interface)
-
-  if (cudax_ENABLE_CUFILE)
-    ###############
-    # cuFile headers #
-    set(headertest_target cudax.headers.${label}.cufile)
-    cccl_generate_header_tests(
-      ${headertest_target}
-      cudax/include
-      HEADER_TEMPLATE "${cudax_SOURCE_DIR}/cmake/header_test.in.cu"
-      GLOBS #
-        "cuda/experimental/cufile.cuh"
-        "cuda/experimental/__cufile/*.cuh"
-    )
-    target_link_libraries(${headertest_target} PUBLIC cudax.compiler_interface)
-  endif()
-
-  # FIXME: Enable MSVC
-  if (cudax_ENABLE_PLACES AND NOT "MSVC" STREQUAL "${CMAKE_CXX_COMPILER_ID}")
-    ##################
-    # Places headers #
-    set(headertest_target cudax.headers.${label}.places)
-    cccl_generate_header_tests(
-      ${headertest_target}
-      cudax/include
-      GLOBS #
-        "cuda/experimental/places.cuh"
-        "cuda/experimental/__places/*.cuh"
-      HEADER_TEMPLATE "${cudax_SOURCE_DIR}/cmake/header_test.in.cu"
-    )
-    target_link_libraries(${headertest_target} PUBLIC cudax.compiler_interface)
-    target_compile_options(
-      ${headertest_target}
-      PRIVATE
-        $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--extended-lambda>
-        $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--expt-relaxed-constexpr>
-    )
-  endif()
-
-  # FIXME: Enable MSVC
-  if (cudax_ENABLE_CUDASTF AND NOT "MSVC" STREQUAL "${CMAKE_CXX_COMPILER_ID}")
-    ###############
-    # STF headers #
-    set(headertest_target cudax.headers.${label}.stf)
-    cccl_generate_header_tests(
-      ${headertest_target}
-      cudax/include
-      GLOBS #
-        "cuda/experimental/stf.cuh"
-        "cuda/experimental/__stf/*.cuh"
-      # FIXME: The cudax header template removes the check for the `small` macro.
-      # cuda/experimental/__stf/utility/memory.cuh defines functions named `small`.
-      # These should be renamed to avoid conflicts with windows system headers, and
-      # the following line removed:
-      HEADER_TEMPLATE "${cudax_SOURCE_DIR}/cmake/header_test.in.cu"
-    )
-    target_link_libraries(
-      ${headertest_target}
-      PUBLIC cudax.compiler_interface CUDA::cuda_driver
-    )
-    target_compile_options(
-      ${headertest_target}
-      PRIVATE
-        # Required by stf headers:
-        $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--extended-lambda>
-        # FIXME: We should be able to refactor away from needing this by
-        # using _CCCL_HOST_DEVICE and friends + `::cuda::std` utilities where
-        # necessary.
-        $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--expt-relaxed-constexpr>
-    )
-  endif()
-endfunction()
-
-cudax_add_header_test(basic "")

cccl_upstream/cudax/cmake/cudaxPlacesConfigureTarget.cmake

@@ -1,23 +0,0 @@
-# Configures a target for the Places framework.
-function(cudax_places_configure_target target_name)
-  target_link_libraries(
-    ${target_name}
-    PRIVATE #
-      CUDA::cudart_static
-      CUDA::cuda_driver
-  )
-
-  target_compile_options(
-    ${target_name}
-    PRIVATE
-      $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--extended-lambda>
-      $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--expt-relaxed-constexpr>
-  )
-
-  set_target_properties(
-    ${target_name}
-    PROPERTIES #
-      CUDA_RUNTIME_LIBRARY Static
-      CUDA_SEPARABLE_COMPILATION ON
-  )
-endfunction()

cccl_upstream/cudax/cmake/cudaxSTFConfigureTarget.cmake

@@ -1,58 +0,0 @@
-# Configures a target for the STF framework.
-function(cudax_stf_configure_target target_name)
-  set(options LINK_MATHLIBS)
-  set(oneValueArgs)
-  set(multiValueArgs)
-  cmake_parse_arguments(
-    CSCT
-    "${options}"
-    "${oneValueArgs}"
-    "${multiValueArgs}"
-    ${ARGN}
-  )
-
-  target_link_libraries(
-    ${target_name}
-    PRIVATE #
-      CUDA::cudart_static
-      CUDA::curand
-      CUDA::cuda_driver
-  )
-
-  if (cudax_ENABLE_CUDASTF_CODE_GENERATION)
-    target_compile_options(
-      ${target_name}
-      PRIVATE $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--extended-lambda>
-    )
-  else()
-    target_compile_definitions(
-      ${target_name}
-      PRIVATE "CUDASTF_DISABLE_CODE_GENERATION"
-    )
-  endif()
-
-  target_compile_options(
-    ${target_name}
-    PRIVATE $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:--expt-relaxed-constexpr>
-  )
-
-  set_target_properties(
-    ${target_name}
-    PROPERTIES #
-      CUDA_RUNTIME_LIBRARY Static
-      CUDA_SEPARABLE_COMPILATION ON
-  )
-
-  if (CSCT_LINK_MATHLIBS)
-    target_link_libraries(
-      ${target_name}
-      PRIVATE #
-        CUDA::cublas
-        CUDA::cusolver
-    )
-  endif()
-
-  if (cudax_ENABLE_CUDASTF_BOUNDSCHECK)
-    target_compile_definitions(${target_name} PRIVATE "CUDASTF_BOUNDSCHECK")
-  endif()
-endfunction()

cccl_upstream/cudax/cmake/header_test.in.cu

@@ -1,66 +0,0 @@
-// This source file checks that:
-// 1) Header <@header@> compiles without error.
-// 2) Common macro collisions with platform/system headers are avoided.
-// 3) half/bf16 aren't included when these are explicitly disabled.
-
-// Define CUDAX_MACRO_CHECK(macro, header), which emits a diagnostic indicating
-// a potential macro collision and halts.
-//
-// Use raw platform checks instead of the CCCL macros since we
-// don't want to #include any headers other than the one being tested.
-//
-// This is only implemented for MSVC/GCC/Clang.
-#if defined(_MSC_VER) // MSVC
-
-// Fake up an error for MSVC
-#  define CUDAX_MACRO_CHECK_IMPL(msg)             \
-    /* Print message that looks like an error: */ \
-    __pragma(message(__FILE__ ":" CUDAX_MACRO_CHECK_IMPL0(__LINE__) ": error: " #msg)) static_assert(false, #msg);
-#  define CUDAX_MACRO_CHECK_IMPL0(x) CUDAX_MACRO_CHECK_IMPL1(x)
-#  define CUDAX_MACRO_CHECK_IMPL1(x) #x
-
-#elif defined(__clang__) || defined(__GNUC__)
-
-// GCC/clang are easy:
-#  define CUDAX_MACRO_CHECK_IMPL(msg)   CUDAX_MACRO_CHECK_IMPL0(GCC error #msg)
-#  define CUDAX_MACRO_CHECK_IMPL0(expr) _Pragma(#expr)
-
-#endif
-
-// Hacky way to build a string, but it works on all tested platforms.
-#define CUDAX_MACRO_CHECK(MACRO, HEADER) \
-  CUDAX_MACRO_CHECK_IMPL(Identifier MACRO should not be used from CCCL headers due to conflicts with HEADER macros.)
-
-// complex.h conflicts
-#define I CUDAX_MACRO_CHECK('I', complex.h)
-
-// windows.h conflicts
-// @eniebler 2024-08-30: This test is disabled because it causes build
-// failures in some configurations.
-// #define small CUDAX_MACRO_CHECK('small', windows.h)
-// We can't enable these checks without breaking some builds -- some standard
-// library implementations unconditionally `#undef` these macros, which then
-// causes random failures later.
-// Leaving these commented out as a warning: Here be dragons.
-// #define min(...) CUDAX_MACRO_CHECK('min', windows.h)
-// #define max(...) CUDAX_MACRO_CHECK('max', windows.h)
-
-// termios.h conflicts (NVIDIA/thrust#1547)
-#define B0 CUDAX_MACRO_CHECK("B0", termios.h)
-
-#include <@header@>
-
-#if defined(CCCL_DISABLE_BF16_SUPPORT)
-#  if defined(__CUDA_BF16_TYPES_EXIST__)
-#    error We should not include cuda_bf16.h when BF16 support is disabled
-#  endif // __CUDA_BF16_TYPES_EXIST__
-#endif // CCCL_DISABLE_BF16_SUPPORT
-
-#if defined(CCCL_DISABLE_FP16_SUPPORT)
-#  if defined(__CUDA_FP16_TYPES_EXIST__)
-#    error We should not include cuda_fp16.h when half support is disabled
-#  endif // __CUDA_FP16_TYPES_EXIST__
-#  if defined(__CUDA_BF16_TYPES_EXIST__)
-#    error We should not include cuda_bf16.h when half support is disabled
-#  endif // __CUDA_BF16_TYPES_EXIST__
-#endif // CCCL_DISABLE_FP16_SUPPORT

cccl_upstream/cudax/cmake/places_header_unittest.in.cu

@@ -1,9 +0,0 @@
-// This file is autogenerated by configuring places_header_unittest.in.cu.
-
-// clang-format off
-#define UNITTESTED_FILE "@source@"
-
-#include <cuda/experimental/__stf/utility/unittest.cuh>
-
-#include <@source@>
-//clang-format on

cccl_upstream/cudax/cmake/stf_header_unittest.in.cu

@@ -1,9 +0,0 @@
-// This file is autogenerated by configuring stf_header_unittest.in.cu.
-
-// clang-format off
-#define UNITTESTED_FILE "@source@"
-
-#include <cuda/experimental/__stf/utility/unittest.cuh>
-
-#include <@source@>
-//clang-format on