cuda_stream_view.hpp

/*
 * Copyright (c) 2020-2021, NVIDIA CORPORATION.
 *
 * 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.
 */
 
#pragma once
 
#include <rmm/detail/error.hpp>
 
#include <cuda_runtime_api.h>
 
#include <atomic>
#include <cstddef>
#include <cstdint>
 
namespace rmm {
 
class cuda_stream_view {
 public:
  constexpr cuda_stream_view()                                   = default;
  constexpr cuda_stream_view(cuda_stream_view const&)            = default;
  constexpr cuda_stream_view(cuda_stream_view&&)                 = default;
  constexpr cuda_stream_view& operator=(cuda_stream_view const&) = default;
  constexpr cuda_stream_view& operator=(cuda_stream_view&&)      = default;
  ~cuda_stream_view()                                            = default;
 
  // Disable construction from literal 0
  constexpr cuda_stream_view(int)            = delete;  //< Prevent cast from 0
  constexpr cuda_stream_view(std::nullptr_t) = delete;  //< Prevent cast from nullptr
 
  constexpr cuda_stream_view(cudaStream_t stream) noexcept : stream_{stream} {}
 
  [[nodiscard]] constexpr cudaStream_t value() const noexcept { return stream_; }
 
  constexpr operator cudaStream_t() const noexcept { return value(); }
 
  [[nodiscard]] inline bool is_per_thread_default() const noexcept;
 
  [[nodiscard]] inline bool is_default() const noexcept;
 
  void synchronize() const { RMM_CUDA_TRY(cudaStreamSynchronize(stream_)); }
 
  void synchronize_no_throw() const noexcept
  {
    RMM_ASSERT_CUDA_SUCCESS(cudaStreamSynchronize(stream_));
  }
 
 private:
  cudaStream_t stream_{};
};
 
static constexpr cuda_stream_view cuda_stream_default{};
 
static const cuda_stream_view cuda_stream_legacy{
  cudaStreamLegacy  // NOLINT(cppcoreguidelines-pro-type-cstyle-cast)
};
 
static const cuda_stream_view cuda_stream_per_thread{
  cudaStreamPerThread  // NOLINT(cppcoreguidelines-pro-type-cstyle-cast)
};
 
[[nodiscard]] inline bool cuda_stream_view::is_per_thread_default() const noexcept
{
#ifdef CUDA_API_PER_THREAD_DEFAULT_STREAM
  return value() == cuda_stream_per_thread || value() == nullptr;
#else
  return value() == cuda_stream_per_thread;
#endif
}
 
[[nodiscard]] inline bool cuda_stream_view::is_default() const noexcept
{
#ifdef CUDA_API_PER_THREAD_DEFAULT_STREAM
  return value() == cuda_stream_legacy;
#else
  return value() == cuda_stream_legacy || value() == nullptr;
#endif
}
 
inline bool operator==(cuda_stream_view lhs, cuda_stream_view rhs)
{
  return lhs.value() == rhs.value();
}
 
inline bool operator!=(cuda_stream_view lhs, cuda_stream_view rhs) { return not(lhs == rhs); }
 
inline std::ostream& operator<<(std::ostream& os, cuda_stream_view stream)
{
  os << stream.value();
  return os;
}
 
}  // namespace rmm