Initial version of mutex support and some drive-by fixes

.upstream-tests/test/support/nasty_containers.h

@@ -9,12 +9,17 @@
 #ifndef NASTY_CONTAINERS_H
 #define NASTY_CONTAINERS_H
 
-#include <cassert>
-#include <vector>
-#include <list>
+#include <cuda/std/cassert>
+#if defined(_LIBCUDACXX_HAS_VECTOR)
+#include <cuda/std/vector>
+#endif
+#if defined(_LIBCUDACXX_HAS_LIST)
+#include <cuda/std/list>
+#endif
 
 #include "test_macros.h"
 
+#if defined(_LIBCUDACXX_HAS_VECTOR)
 template <class T>
 class nasty_vector
 {
@@ -135,7 +140,9 @@ class nasty_vector
 
 template <class T>
 bool operator==(const nasty_vector<T>& x, const nasty_vector<T>& y) { return x.v_ == y.v_; }
+#endif
 
+#if defined(_LIBCUDACXX_HAS_LIST)
 template <class T>
 class nasty_list
 {
@@ -282,6 +289,7 @@ class nasty_list
 
 template <class T>
 bool operator==(const nasty_list<T>& x, const nasty_list<T>& y) { return x.l_ == y.l_; }
+#endif
 
 // Not really a mutex, but can play one in tests
 class nasty_mutex

include/cuda/mutex

@@ -0,0 +1,25 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "std/mutex"
+
+_LIBCUDACXX_BEGIN_NAMESPACE_CUDA
+
+template<thread_scope _Sco>
+using mutex = _CUDA_VSTD::__mutex_base<_Sco>;
+
+template<thread_scope _Sco>
+using timed_mutex = _CUDA_VSTD::__mutex_base<_Sco>;
+
+template<thread_scope _Sco>
+using once_flag = _CUDA_VSTD::__once_flag_base<_Sco>;
+
+using _CUDA_VSTD::call_once;
+
+_LIBCUDACXX_END_NAMESPACE_CUDA

include/cuda/std/detail/__config

@@ -79,10 +79,12 @@
 #define _LIBCUDACXX_HAS_NO_PLATFORM_WAIT
 #define _LIBCUDACXX_HAS_NO_MONOTONIC_CLOCK
 #define _LIBCUDACXX_HAS_NO_TREE_BARRIER
+#define _LIBCUDACXX_HAS_THREAD_API_EXTERNAL
+#define _LIBCUDACXX_INLINE_THREADING
+
 #ifdef __CUDACC_RTC__
     #define __ELF__
     #define _LIBCUDACXX_DISABLE_PRAGMA_GCC_SYSTEM_HEADER
-    #define _LIBCUDACXX_HAS_THREAD_API_EXTERNAL
     #define __alignof(x) alignof(x)
     #define _LIBCUDACXX_LITTLE_ENDIAN
     #define _LIBCUDACXX_DISABLE_VISIBILITY_ANNOTATIONS
@@ -104,9 +106,8 @@
 
 #include "libcxx/include/__config"
 
-#if defined(__CUDA_ARCH__)
-    #define _LIBCUDACXX_HAS_THREAD_API_CUDA
-#elif defined(_LIBCUDACXX_COMPILER_MSVC)
+#define _LIBCUDACXX_HAS_THREAD_API_CUDA
+#if defined(_LIBCUDACXX_COMPILER_MSVC)
     #define _LIBCUDACXX_HAS_THREAD_API_WIN32
 #endif
 

include/cuda/std/detail/libcxx/include/CMakeLists.txt

@@ -117,6 +117,7 @@ set(files
   __mdspan/submdspan.hpp
   __mdspan/type_list.hpp
   __memory/addressof.h
+  __memory/atomic_load.h
   __memory/pointer_traits.h
   __mutex_base
   __node_handle

include/cuda/std/detail/libcxx/include/__memory/atomic_load.h

@@ -0,0 +1,69 @@
+// -*- C++ -*-
+//===----------------------------------------------------------------------===//
+//
+// 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) 2023 NVIDIA CORPORATION & AFFILIATES.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _LIBCUDACXX___MEMORY_ATOMIMC_LOAD_H
+#define _LIBCUDACXX___MEMORY_ATOMIMC_LOAD_H
+
+#ifndef __cuda_std__
+#include <__config>
+#endif //__cuda_std__
+
+#include "../atomic"
+
+#if defined(_LIBCUDACXX_USE_PRAGMA_GCC_SYSTEM_HEADER)
+#pragma GCC system_header
+#endif
+
+_LIBCUDACXX_BEGIN_NAMESPACE_STD
+
+#ifndef __cuda_std__
+
+template <class _ValueType>
+inline _LIBCUDACXX_INLINE_VISIBILITY
+_ValueType __libcpp_relaxed_load(_ValueType const* __value) {
+#if !defined(_LIBCUDACXX_HAS_NO_THREADS) && \
+    defined(__ATOMIC_RELAXED) &&        \
+    (__has_builtin(__atomic_load_n) || defined(_LIBCUDACXX_COMPILER_GCC))
+    return __atomic_load_n(__value, __ATOMIC_RELAXED);
+#else
+    return *__value;
+#endif
+}
+
+template <class _ValueType>
+inline _LIBCUDACXX_INLINE_VISIBILITY
+_ValueType __libcpp_acquire_load(_ValueType const* __value) {
+#if !defined(_LIBCUDACXX_HAS_NO_THREADS) && \
+    defined(__ATOMIC_ACQUIRE) &&        \
+    (__has_builtin(__atomic_load_n) || defined(_LIBCUDACXX_COMPILER_GCC))
+    return __atomic_load_n(__value, __ATOMIC_ACQUIRE);
+#else
+    return *__value;
+#endif
+}
+
+#else
+
+template <class _ValueType>
+inline _LIBCUDACXX_INLINE_VISIBILITY
+_ValueType __libcpp_relaxed_load(atomic<_ValueType> const* __value) {
+    return __value->load(memory_order_relaxed);
+}
+
+template <class _ValueType>
+inline _LIBCUDACXX_INLINE_VISIBILITY
+_ValueType __libcpp_acquire_load(atomic<_ValueType> const* __value) {
+    return __value->load(memory_order_acquire);
+}
+#endif // __cuda_std__
+
+_LIBCUDACXX_END_NAMESPACE_STD
+
+#endif // _LIBCUDACXX___MEMORY_ATOMIMC_LOAD_H