void ChannelImpl::advanceRecvOperation()

in tensorpipe/channel/cuda_gdr/channel_impl.cc [410:476]
51 lines of code
10 McCabe index (conditional complexity)

void ChannelImpl::advanceRecvOperation(
    RecvOpIter opIter,
    RecvOperation::State prevOpState) {
  TP_DCHECK(context_->inLoop());

  RecvOperation& op = *opIter;

  recvOps_.attemptTransition(
      opIter,
      /*from=*/RecvOperation::UNINITIALIZED,
      /*to=*/RecvOperation::FINISHED,
      /*cond=*/error_ || op.length == 0,
      /*actions=*/{&ChannelImpl::callRecvCallback});

  // Needs to go after previous op to ensure predictable and consistent ordering
  // of write calls on the descriptor control connection.
  recvOps_.attemptTransition(
      opIter,
      /*from=*/RecvOperation::UNINITIALIZED,
      /*to=*/RecvOperation::READING_DESCRIPTOR,
      /*cond=*/!error_ && state_ == ESTABLISHED &&
          prevOpState >= RecvOperation::READING_DESCRIPTOR,
      /*actions=*/{&ChannelImpl::readDescriptor});

  recvOps_.attemptTransition(
      opIter,
      /*from=*/RecvOperation::READING_DESCRIPTOR,
      /*to=*/RecvOperation::FINISHED,
      /*cond=*/error_ && op.doneReadingDescriptor,
      /*actions=*/{&ChannelImpl::callRecvCallback});

  // This doesn't strictly need to go after the previous op, but it doesn't make
  // sense to busy poll multiple events if only one of them is actually able to
  // then make progress.
  recvOps_.attemptTransition(
      opIter,
      /*from=*/RecvOperation::READING_DESCRIPTOR,
      /*to=*/RecvOperation::WAITING_FOR_CUDA_EVENT,
      /*cond=*/!error_ && op.doneReadingDescriptor &&
          prevOpState >= RecvOperation::RECEIVING_OVER_IB,
      /*actions=*/{&ChannelImpl::waitForRecvCudaEvent});

  recvOps_.attemptTransition(
      opIter,
      /*from=*/RecvOperation::WAITING_FOR_CUDA_EVENT,
      /*to=*/RecvOperation::FINISHED,
      /*cond=*/error_ && op.doneWaitingForCudaEvent,
      /*actions=*/{&ChannelImpl::callRecvCallback});

  // Needs to go after previous op to ensure predictable and consistent ordering
  // of recv calls on InfiniBand queue pair and write calls on the completion
  // control connection.
  recvOps_.attemptTransition(
      opIter,
      /*from=*/RecvOperation::WAITING_FOR_CUDA_EVENT,
      /*to=*/RecvOperation::RECEIVING_OVER_IB,
      /*cond=*/!error_ && op.doneWaitingForCudaEvent &&
          prevOpState >= RecvOperation::RECEIVING_OVER_IB,
      /*actions=*/{&ChannelImpl::recvOverIbAndWriteReadyToRecive});

  recvOps_.attemptTransition(
      opIter,
      /*from=*/RecvOperation::RECEIVING_OVER_IB,
      /*to=*/RecvOperation::FINISHED,
      /*cond=*/op.numChunksBeingReceived == 0,
      /*actions=*/{&ChannelImpl::callRecvCallback});
}