diff options
author | Wen-Heng (Jack) Chung <whchung@gmail.com> | 2018-07-11 17:57:38 +0000 |
---|---|---|
committer | Wen-Heng (Jack) Chung <whchung@gmail.com> | 2018-07-12 17:14:09 +0000 |
commit | 25021d386cd989aedde11b72c5db36b7c1bfd2b4 (patch) | |
tree | 8c325d003489372b16747a4cbe5dec4fe3f276f1 /tensorflow/stream_executor/cuda | |
parent | 135e419e780423a888ddd45e479129493336c52b (diff) |
[ROCm] Interface changes for StreamExecutor to support both CUDA and ROCm
1) StreamInterface::CudaStreamMemberHack()
Despite the fact that StreamExecutor and GPU common runtime are largely
orthogonal, this particular routine in StreamExecutor is used in GPU common
runtime and a couple of other operators. In this commit it's renamed as
StreamInterface::GpuStreamMemberHack() and their call sites are also changed.
2) StreamExecutorInterface::CudaContextHack()
This member is renamed to StramExecutorInterface::GpuContextHack().
Changes introduced in this commit includes:
- some StreamExecutor interfaces and CUDA implementation
- GPU common runtime related to interface changes in StreamExecutor
- operators affected by interface changes in StreamExecutor
Diffstat (limited to 'tensorflow/stream_executor/cuda')
-rw-r--r-- | tensorflow/stream_executor/cuda/cuda_gpu_executor.cc | 2 | ||||
-rw-r--r-- | tensorflow/stream_executor/cuda/cuda_gpu_executor.h | 2 | ||||
-rw-r--r-- | tensorflow/stream_executor/cuda/cuda_stream.h | 4 |
3 files changed, 4 insertions, 4 deletions
diff --git a/tensorflow/stream_executor/cuda/cuda_gpu_executor.cc b/tensorflow/stream_executor/cuda/cuda_gpu_executor.cc index f11022ef1d..259c813c57 100644 --- a/tensorflow/stream_executor/cuda/cuda_gpu_executor.cc +++ b/tensorflow/stream_executor/cuda/cuda_gpu_executor.cc @@ -844,7 +844,7 @@ CUDAExecutor::GetTimerImplementation() { return std::unique_ptr<internal::TimerInterface>(new CUDATimer(this)); } -void *CUDAExecutor::CudaContextHack() { return context_; } +void *CUDAExecutor::GpuContextHack() { return context_; } CudaContext* CUDAExecutor::cuda_context() { return context_; } diff --git a/tensorflow/stream_executor/cuda/cuda_gpu_executor.h b/tensorflow/stream_executor/cuda/cuda_gpu_executor.h index 773cbfb8a1..f7c341c857 100644 --- a/tensorflow/stream_executor/cuda/cuda_gpu_executor.h +++ b/tensorflow/stream_executor/cuda/cuda_gpu_executor.h @@ -210,7 +210,7 @@ class CUDAExecutor : public internal::StreamExecutorInterface { std::unique_ptr<internal::TimerInterface> GetTimerImplementation() override; - void *CudaContextHack() override; + void *GpuContextHack() override; CudaContext* cuda_context(); diff --git a/tensorflow/stream_executor/cuda/cuda_stream.h b/tensorflow/stream_executor/cuda/cuda_stream.h index 02edff6431..bb8bda4755 100644 --- a/tensorflow/stream_executor/cuda/cuda_stream.h +++ b/tensorflow/stream_executor/cuda/cuda_stream.h @@ -40,8 +40,8 @@ class CUDAStream : public internal::StreamInterface { // Note: teardown is handled by a parent's call to DeallocateStream. ~CUDAStream() override {} - void *CudaStreamHack() override { return cuda_stream_; } - void **CudaStreamMemberHack() override { + void *GpuStreamHack() override { return cuda_stream_; } + void **GpuStreamMemberHack() override { return reinterpret_cast<void **>(&cuda_stream_); } |