Skip to content

[VitisAI] bugfix model_clone optimization #25707

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
yifei410 wants to merge 1 commit into from
Closed

[VitisAI] bugfix model_clone optimization #25707

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
35 changes: 23 additions & 12 deletions onnxruntime/core/providers/vitisai/imp/graph.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -311,19 +311,30 @@
}
auto ORT_MEM_ADDR_tag = process_ext_address(*original_tensor);
if (!ORT_MEM_ADDR_tag.empty()) {
cloned_tensor->set_data_location(ONNX_NAMESPACE::TensorProto_DataLocation_EXTERNAL);
auto external_data = cloned_tensor->mutable_external_data();
auto p = external_data->Add();
*p->mutable_key() = "location";
*p->mutable_value() = std::string("<") + graph_ptr;
} else if (size >= external_data_threshold) {
cloned_tensor->set_data_location(ONNX_NAMESPACE::TensorProto_DataLocation_EXTERNAL);
auto external_data = cloned_tensor->mutable_external_data();
auto p = external_data->Add();
*p->mutable_key() = "location";
*p->mutable_value() = std::string("<") + graph_ptr;
// ORT 1.23 intializer handling
if (size >= external_data_threshold) {
Copy link
Member

@yuslepukhin yuslepukhin Aug 11, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

if (size >= external_data_threshold) {

Use utils::HasExternalDataInMemory() to see if data is in OrtValue.

cloned_tensor->set_data_location(ONNX_NAMESPACE::TensorProto_DataLocation_EXTERNAL);
auto external_data = cloned_tensor->mutable_external_data();
auto p = external_data->Add();
*p->mutable_key() = "location";
*p->mutable_value() = std::string("<") + graph_ptr;
} else {
// Under threshold: get the data from ORT_MEM_ADDR and add it to cloned graph
Copy link
Member

@yuslepukhin yuslepukhin Aug 11, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

// Under threshold: get the data from ORT_MEM_AD

Do not rely on threshold. Some big initializers may not be converted. Use utils to test if the data is external.

We generally do not want YOU to make the decision of what is expected to be an external data in memory.

Use utilities to test and convert.

You can also get a corresponding OrtValue from the graph.

std::unique_ptr<ONNX_NAMESPACE::TensorProto> tensor_proto;

Check warning on line 323 in onnxruntime/core/providers/vitisai/imp/graph.cc

View workflow job for this annotation

GitHub Actions / Optional Lint C++ 

[cpplint] reported by reviewdog 🐶
 Add #include <memory> for unique_ptr<>  [build/include_what_you_use] [4]

Raw Output:
onnxruntime/core/providers/vitisai/imp/graph.cc:323:  Add #include <memory> for unique_ptr<>  [build/include_what_you_use] [4]
Copy link
Member

@yuslepukhin yuslepukhin Aug 11, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

tensor_proto;

If it is not in memory, then this is going to be nullptr

ORT_THROW_IF_ERROR(utils::GetTensorProtoWithDataIfInMemory(*original_tensor, tensor_proto));
*cloned_tensor = *tensor_proto;
}
} else {
*cloned_tensor = *original_tensor;
// ORT 1.22 or ealier intializer handling

Check warning on line 328 in onnxruntime/core/providers/vitisai/imp/graph.cc

View workflow job for this annotation

GitHub Actions / Optional Lint 

[misspell] reported by reviewdog 🐶
"ealier" is a misspelling of "earlier"

Raw Output:
./onnxruntime/core/providers/vitisai/imp/graph.cc:328:21: "ealier" is a misspelling of "earlier"
Copy link
Member

@yuslepukhin yuslepukhin Aug 11, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

// ORT 1.22 or ealier intializer handling

What is the perceived difference? It is inline or not.

if (size >= external_data_threshold) {
cloned_tensor->set_data_location(ONNX_NAMESPACE::TensorProto_DataLocation_EXTERNAL);
auto external_data = cloned_tensor->mutable_external_data();
auto p = external_data->Add();
*p->mutable_key() = "location";
*p->mutable_value() = std::string("<") + graph_ptr;
} else {
*cloned_tensor = *original_tensor;
}
}
}
auto ret = Model::Create(std::move(*model_proto), file_path, &local_registries, logger);
Expand Down
Loading