sunjiweiswift commented on a change in pull request #8497:
URL: https://github.com/apache/tvm/pull/8497#discussion_r681375108
##########
File path: src/runtime/graph_executor/graph_executor.cc
##########
@@ -384,9 +446,15 @@ void GraphExecutor::SetupOpExecs() {
for (size_t i = 0; i < inode.inputs.size(); i++) {
uint32_t eid = this->entry_id(inode.inputs[i]);
- // check if op input is model input
- if (input_node_eids.count(eid) > 0) {
-
input_dltensors_[eid].push_back(static_cast<DLTensor*>(op_args->arg_values[i].v_handle));
Review comment:
Inorder to modify the node input mem addr which connect to the output
node, all nodes input will push into the input_dltensors_， so I removed the
check.
Before deleting, the size of input_dltensors_ is the same as nodes_. The
non-model input node input_dltensors[X] is empty. After the modification, the
nodes that do not affect the model input, the non-input nodes will be push into
the input_dltensors_ .
The essence is that input_dltensors_ records the input memory of all nodes,
not just the memory of the model input. Does not affect the function of
set_input.
I think this will not cause bugs.What do you think
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
