CUDA vector_add sample project (#2160)

---------

Co-authored-by: pciolkosz <[email protected]>
Co-authored-by: Michael Schellenberger Costa <[email protected]>

cudax/CMakeLists.txt

@@ -11,7 +11,7 @@ if (cudax_TOPLEVEL_PROJECT)
   cmake_minimum_required(VERSION 3.21)
 endif()
 
-project(cudax LANGUAGES CUDA)
+project(cudax LANGUAGES CUDA CXX)
 
 option(cudax_ENABLE_INSTALL_RULES "Enable installation of CUDA Experimental." ${cudax_TOPLEVEL_PROJECT})
 if (cudax_ENABLE_INSTALL_RULES)
@@ -25,6 +25,7 @@ endif()
 
 option(cudax_ENABLE_HEADER_TESTING "Test that CUDA Experimental's public headers compile." ON)
 option(cudax_ENABLE_TESTING "Build CUDA Experimental's tests." ON)
+option(cudax_ENABLE_SAMPLES "Build CUDA Experimental's samples." ON)
 
 include(cmake/cudaxBuildCompilerTargets.cmake)
 include(cmake/cudaxBuildTargetList.cmake)
@@ -41,3 +42,13 @@ if (cudax_ENABLE_TESTING)
   enable_testing() # Must be in root directory
   add_subdirectory(test)
 endif()
+
+if (cudax_ENABLE_SAMPLES)
+  include(ExternalProject)
+  ExternalProject_Add(cudax_samples
+    PREFIX samples
+    SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/samples"
+    BUILD_ALWAYS ON
+    INSTALL_COMMAND cmake -E echo "Skipping install step.")
+  add_dependencies(cudax.all cudax_samples)
+endif()

cudax/cmake/cudaxBuildCompilerTargets.cmake

@@ -9,7 +9,7 @@
 include("${cudax_SOURCE_DIR}/cmake/AppendOptionIfAvailable.cmake")
 
 function(cudax_build_compiler_targets)
-  set(cxx_compile_definitions)
+  set(cxx_compile_definitions LIBCUDACXX_ENABLE_EXCEPTIONS)
   set(cxx_compile_options)
   set(cuda_compile_options)
 

cudax/cmake/cudaxBuildTargetList.cmake

@@ -176,6 +176,7 @@ function(cudax_build_target_list)
   file(GLOB_RECURSE all_sources
     RELATIVE "${CMAKE_CURRENT_LIST_DIR}"
     "${cudax_SOURCE_DIR}/include/cuda/experimental/*.hpp"
+    "${cudax_SOURCE_DIR}/include/cuda/experimental/*.cuh"
   )
   add_custom_target(cudax.all SOURCES ${all_sources})
 

cudax/include/cuda/experimental/__detail/utility.cuh

@@ -25,20 +25,31 @@ namespace cuda::experimental
 {
 namespace detail
 {
-struct __ignore
+// This is a helper type that can be used to ignore function arguments.
+struct [[maybe_unused]] __ignore
 {
-  template <typename... Args>
-  _CCCL_HOST_DEVICE constexpr __ignore(Args&&...) noexcept
+  __ignore() = default;
+
+  template <typename _Arg>
+  _CCCL_HOST_DEVICE constexpr __ignore(_Arg&&) noexcept
   {}
 };
+
+// Classes can inherit from this type to become immovable.
+struct __immovable
+{
+  __immovable()                         = default;
+  __immovable(__immovable&&)            = delete;
+  __immovable& operator=(__immovable&&) = delete;
+};
 } // namespace detail
 
 struct uninit_t
 {
   explicit uninit_t() = default;
 };
 
-inline constexpr uninit_t uninit{};
+_CCCL_GLOBAL_CONSTANT uninit_t uninit{};
 } // namespace cuda::experimental
 
 #endif // __CUDAX_DETAIL_UTILITY_H

cudax/include/cuda/experimental/__launch/param_kind.cuh

@@ -0,0 +1,85 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of CUDA Experimental in CUDA C++ Core Libraries,
+// 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 _CUDAX__LAUNCH_PARAM_KIND
+#define _CUDAX__LAUNCH_PARAM_KIND
+
+#include <cuda/__cccl_config>
+
+#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 <cuda/std/__type_traits/maybe_const.h>
+
+#include <cuda/experimental/__detail/utility.cuh>
+
+namespace cuda::experimental
+{
+namespace detail
+{
+enum class __param_kind : unsigned
+{
+  _in    = 1,
+  _out   = 2,
+  _inout = 3
+};
+
+_CCCL_NODISCARD _CCCL_HOST_DEVICE inline constexpr __param_kind operator&(__param_kind __a, __param_kind __b) noexcept
+{
+  return __param_kind(unsigned(__a) & unsigned(__b));
+}
+
+template <typename _Ty, __param_kind _Kind>
+struct _CCCL_NODISCARD __box
+{
+  ::cuda::std::__maybe_const<_Kind == __param_kind::_in, _Ty>& __val;
+};
+
+struct __in_t
+{
+  template <class _Ty>
+  __box<_Ty, __param_kind::_in> operator()(const _Ty& __v) const noexcept
+  {
+    return {__v};
+  }
+};
+
+struct __out_t
+{
+  template <class _Ty>
+  __box<_Ty, __param_kind::_out> operator()(_Ty& __v) const noexcept
+  {
+    return {__v};
+  }
+};
+
+struct __inout_t
+{
+  template <class _Ty>
+  __box<_Ty, __param_kind::_inout> operator()(_Ty& __v) const noexcept
+  {
+    return {__v};
+  }
+};
+
+} // namespace detail
+
+_CCCL_GLOBAL_CONSTANT detail::__in_t in{};
+_CCCL_GLOBAL_CONSTANT detail::__out_t out{};
+_CCCL_GLOBAL_CONSTANT detail::__inout_t inout{};
+
+} // namespace cuda::experimental
+
+#endif // _CUDAX__LAUNCH_PARAM_KIND

cudax/include/cuda/experimental/launch.cuh

@@ -11,6 +11,9 @@
 #ifndef __CUDAX_LAUNCH___
 #define __CUDAX_LAUNCH___
 
+#include <cuda/experimental/__launch/configuration.cuh>
 #include <cuda/experimental/__launch/launch.cuh>
+#include <cuda/experimental/__launch/launch_transform.cuh>
+#include <cuda/experimental/__launch/param_kind.cuh>
 
 #endif // __CUDAX_LAUNCH___

cudax/samples/CMakeLists.txt

@@ -0,0 +1,76 @@
+# SPDX-FileCopyrightText: Copyright (c) 2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+cmake_minimum_required(VERSION 3.14 FATAL_ERROR)
+
+project(CUDAX_SAMPLES CUDA CXX)
+
+# This example uses the CMake Package Manager (CPM) to simplify fetching CCCL from GitHub
+# For more information, see https://github.com/cpm-cmake/CPM.cmake
+include(cmake/CPM.cmake)
+
+# We define these as variables so they can be overriden in CI to pull from a PR instead of CCCL `main`
+# In your project, these variables are unncessary and you can just use the values directly
+set(CCCL_REPOSITORY "nvidia/cccl" CACHE STRING "GitHub repository to fetch CCCL from")
+set(CCCL_TAG "main" CACHE STRING "Git tag/branch to fetch from CCCL repository")
+
+# This will automatically clone CCCL from GitHub and make the exported cmake targets available
+CPMAddPackage(
+  NAME CCCL
+  GITHUB_REPOSITORY ${CCCL_REPOSITORY}
+  GIT_TAG ${CCCL_TAG}
+  GIT_SHALLOW ON
+  OPTIONS "CCCL_ENABLE_UNSTABLE ON"
+)
+
+# Default to building for the GPU on the current system
+if(NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
+  set(CMAKE_CUDA_ARCHITECTURES 86)
+endif()
+
+# Creates a cmake executable target for the main program
+add_executable(vector_add vector_add/vector_add.cu)
+
+# "Links" the CCCL::cudax CMake target to the `vector_add` executable. This
+# configures everything needed to use CCCL's headers, including setting up
+# include paths, compiler flags, etc.
+target_link_libraries(vector_add
+  PUBLIC
+    CCCL::cudax
+    CCCL::CCCL
+    CCCL::Thrust
+    CCCL::libcudacxx
+  INTERFACE cudax.compiler_interface
+)
+
+# TODO: These are temporary until the main branch catches up with the latest changes
+target_compile_definitions(vector_add PUBLIC LIBCUDACXX_ENABLE_EXCEPTIONS)
+
+if ("MSVC" STREQUAL "${CMAKE_CXX_COMPILER_ID}")
+  # mdspan on windows only works in C++20 mode
+  target_compile_features(vector_add PUBLIC cxx_std_20)
+
+  # cudax requires dim3 to be usable from a constexpr context, and the CUDART headers require
+  # __cplusplus to be defined for this to work:
+  target_compile_options(vector_add PRIVATE
+    $<$<COMPILE_LANGUAGE:CXX>:/Zc:__cplusplus /Zc:preprocessor>
+    $<$<COMPILE_LANG_AND_ID:CUDA,NVIDIA>:-Xcompiler=/Zc:__cplusplus -Xcompiler=/Zc:preprocessor>
+  )
+endif()
+
+# This is only relevant for internal testing and not needed by end users.
+include(CTest)
+enable_testing()
+add_test(NAME vector_add COMMAND vector_add)

cudax/samples/cmake/CPM.cmake

@@ -0,0 +1,33 @@
+set(CPM_DOWNLOAD_VERSION 0.38.1)
+
+if(CPM_SOURCE_CACHE)
+  set(CPM_DOWNLOAD_LOCATION "${CPM_SOURCE_CACHE}/cpm/CPM_${CPM_DOWNLOAD_VERSION}.cmake")
+elseif(DEFINED ENV{CPM_SOURCE_CACHE})
+  set(CPM_DOWNLOAD_LOCATION "$ENV{CPM_SOURCE_CACHE}/cpm/CPM_${CPM_DOWNLOAD_VERSION}.cmake")
+else()
+  set(CPM_DOWNLOAD_LOCATION "${CMAKE_BINARY_DIR}/cmake/CPM_${CPM_DOWNLOAD_VERSION}.cmake")
+endif()
+
+# Expand relative path. This is important if the provided path contains a tilde (~)
+get_filename_component(CPM_DOWNLOAD_LOCATION ${CPM_DOWNLOAD_LOCATION} ABSOLUTE)
+
+function(download_cpm)
+  message(STATUS "Downloading CPM.cmake to ${CPM_DOWNLOAD_LOCATION}")
+  file(DOWNLOAD
+       https://github.com/cpm-cmake/CPM.cmake/releases/download/v${CPM_DOWNLOAD_VERSION}/CPM.cmake
+       ${CPM_DOWNLOAD_LOCATION}
+  )
+endfunction()
+
+if(NOT (EXISTS ${CPM_DOWNLOAD_LOCATION}))
+  download_cpm()
+else()
+  # resume download if it previously failed
+  file(READ ${CPM_DOWNLOAD_LOCATION} check)
+  if("${check}" STREQUAL "")
+    download_cpm()
+  endif()
+  unset(check)
+endif()
+
+include(${CPM_DOWNLOAD_LOCATION})

cudax/samples/vector_add/param_kind.cuh

@@ -0,0 +1,85 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of CUDA Experimental in CUDA C++ Core Libraries,
+// 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 _CUDAX__LAUNCH_PARAM_KIND
+#define _CUDAX__LAUNCH_PARAM_KIND
+
+#include <cuda/__cccl_config>
+
+#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 <cuda/std/__type_traits/maybe_const.h>
+
+#include <cuda/experimental/__detail/utility.cuh>
+
+namespace cuda::experimental
+{
+namespace detail
+{
+enum class __param_kind : unsigned
+{
+  _in    = 1,
+  _out   = 2,
+  _inout = 3
+};
+
+_CCCL_NODISCARD _CCCL_HOST_DEVICE inline constexpr __param_kind operator&(__param_kind __a, __param_kind __b) noexcept
+{
+  return __param_kind(unsigned(__a) & unsigned(__b));
+}
+
+template <typename _Ty, __param_kind _Kind>
+struct _CCCL_NODISCARD __box
+{
+  ::cuda::std::__maybe_const<_Kind == __param_kind::_in, _Ty>& __val;
+};
+
+struct __in_t
+{
+  template <class _Ty>
+  __box<_Ty, __param_kind::_in> operator()(const _Ty& __v) const noexcept
+  {
+    return {__v};
+  }
+};
+
+struct __out_t
+{
+  template <class _Ty>
+  __box<_Ty, __param_kind::_out> operator()(_Ty& __v) const noexcept
+  {
+    return {__v};
+  }
+};
+
+struct __inout_t
+{
+  template <class _Ty>
+  __box<_Ty, __param_kind::_inout> operator()(_Ty& __v) const noexcept
+  {
+    return {__v};
+  }
+};
+
+} // namespace detail
+
+_CCCL_GLOBAL_CONSTANT detail::__in_t in{};
+_CCCL_GLOBAL_CONSTANT detail::__out_t out{};
+_CCCL_GLOBAL_CONSTANT detail::__inout_t inout{};
+
+} // namespace cuda::experimental
+
+#endif // _CUDAX__LAUNCH_PARAM_KIND