Method gpu_stream

tensorflow/stream_executor/gpu/gpu_stream.h:72–75

Returns the GpuStreamHandle value for passing to the CUDA API. Precond: this GpuStream has been allocated (otherwise passing a nullptr into the NVIDIA library causes difficult-to-understand faults).

  // Precond: this GpuStream has been allocated (otherwise passing a nullptr
  // into the NVIDIA library causes difficult-to-understand faults).
  GpuStreamHandle gpu_stream() const {
    DCHECK(gpu_stream_ != nullptr);
    return const_cast<GpuStreamHandle>(gpu_stream_);
  }

  // TODO(timshen): Migrate away and remove this function.
  GpuStreamHandle cuda_stream() const { return gpu_stream(); }