| 33 | } |
| 34 | |
| 35 | Status AsyncOutSendThunk::ExecuteOnStream(const ExecuteParams& params) { |
| 36 | auto& buffer_allocations = *params.buffer_allocations; |
| 37 | |
| 38 | auto op_profiler = |
| 39 | params.profiler->MakeScopedInstructionProfiler(hlo_instruction()); |
| 40 | se::DeviceMemoryBase addr = |
| 41 | buffer_allocations.GetDeviceAddress(input_buffer_); |
| 42 | |
| 43 | VLOG(4) << "AsyncOutSendThunk on GPU: " << hlo_instruction()->ToString() |
| 44 | << " with buf size " << addr.size() << " @" << addr.opaque() |
| 45 | << ", key " << key_ << ", hash " << key_hash_; |
| 46 | |
| 47 | tensorflow::AsyncIoRendezvous::TensorPayload payload; |
| 48 | payload.addr = addr; |
| 49 | CHECK(ShapeUtil::Equal(async_out_send_shape_, |
| 50 | hlo_instruction()->operand(0)->shape())); |
| 51 | payload.shape = hlo_instruction()->operand(0)->shape(); |
| 52 | tensorflow::GetXlaAsyncIORendezvous()->Send(key_hash_, payload); |
| 53 | |
| 54 | return Status::OK(); |
| 55 | } |
| 56 | |
| 57 | } // namespace gpu |
| 58 | } // namespace xla |
nothing calls this directly
no test coverage detected