Sync the 'done' operation on the GPU stream, but without all the data copying.
| 962 | // Sync the 'done' operation on the GPU stream, but without all the data |
| 963 | // copying. |
| 964 | static void StreamGPUOp(Device* gpu_device, const DeviceContext* device_context, |
| 965 | StatusCallback done) { |
| 966 | Tensor dummy1, dummy2; |
| 967 | GPUUtil::CopyGPUTensorToCPU(gpu_device, device_context, &dummy1, &dummy2, |
| 968 | done); |
| 969 | } |
| 970 | #endif // GOOGLE_CUDA |
| 971 | |
| 972 | RdmaTensorResponse* RdmaChannel::AddTensorResponse(const RdmaMessage& rm) { |
no outgoing calls
no test coverage detected