MCPcopy Create free account
hub / github.com/DeepRec-AI/DeepRec / StreamGPUOp

Function StreamGPUOp

tensorflow/contrib/verbs/rdma.cc:964–969  ·  view source on GitHub ↗

Sync the 'done' operation on the GPU stream, but without all the data copying.

Source from the content-addressed store, hash-verified

962// Sync the 'done' operation on the GPU stream, but without all the data
963// copying.
964static 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
972RdmaTensorResponse* RdmaChannel::AddTensorResponse(const RdmaMessage& rm) {

Callers 2

RecvHandlerMethod · 0.85
AllocateTensorsAsyncMethod · 0.85

Calls

no outgoing calls

Tested by

no test coverage detected