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

Method ExecuteOnStream

tensorflow/compiler/xla/service/gpu/async_out_thunk.cc:35–55  ·  view source on GitHub ↗

Source from the content-addressed store, hash-verified

33}
34
35Status 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

Callers

nothing calls this directly

Calls 10

GetXlaAsyncIORendezvousFunction · 0.85
GetDeviceAddressMethod · 0.80
opaqueMethod · 0.80
EqualClass · 0.50
ToStringMethod · 0.45
sizeMethod · 0.45
shapeMethod · 0.45
operandMethod · 0.45
SendMethod · 0.45

Tested by

no test coverage detected