Skip to content

[QNN EP] Minor fix weight name missing when not valid QDQ node group #25887

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

Merged
HectorSVC merged 1 commit into from
Aug 29, 2025
Merged

[QNN EP] Minor fix weight name missing when not valid QDQ node group #25887

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
14 changes: 13 additions & 1 deletion onnxruntime/core/providers/qnn/builder/opbuilder/conv_op_builder.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -245,6 +245,12 @@ Status ConvOpBuilder::ProcessConv2D3DInputs(QnnModelWrapper& qnn_model_wrapper,
bool is_graph_input = qnn_model_wrapper.IsGraphInput(input1_name);
LOGS(logger, VERBOSE) << "Add HWCN Transpose node after input: " << input1_name;

if (!qnn_model_wrapper.IsQnnTensorWrapperExist(input1_name)) {
QnnTensorWrapper weight_tensor_wrapper;
ORT_RETURN_IF_ERROR(qnn_model_wrapper.MakeTensorWrapper(inputs[1], weight_tensor_wrapper));
ORT_RETURN_IF_NOT(qnn_model_wrapper.AddTensorWrapper(std::move(weight_tensor_wrapper)), "Failed to add weight tensor.");
}

if (conv_type == OnnxConvType::kConv) {
ORT_RETURN_IF_ERROR(qnn_model_wrapper.AddNchwToHwcnTranspose(node_unit.Index(),
input1_name,
Expand Down Expand Up @@ -425,7 +431,7 @@ Status ConvOpBuilder::ProcessConv1DInputs(QnnModelWrapper& qnn_model_wrapper,

//
// Input 1: weight
// We need to first reshape the weight inorder to handle 1D convolutions with the Conv2d operator.
// We need to first reshape the weight in order to handle 1D convolutions with the Conv2d operator.
// Next, we have to transpose the weight because ORT layout transformations do not change the weight layout.
//
{
Expand Down Expand Up @@ -511,6 +517,12 @@ Status ConvOpBuilder::ProcessConv1DInputs(QnnModelWrapper& qnn_model_wrapper,
ORT_RETURN_IF(input_info.quant_param.IsPerChannel(),
"Non-constant Conv inputs only support per-tensor quantization");

if (!qnn_model_wrapper.IsQnnTensorWrapperExist(input1_name)) {
QnnTensorWrapper weight_tensor_wrapper;
ORT_RETURN_IF_ERROR(qnn_model_wrapper.MakeTensorWrapper(inputs[1], weight_tensor_wrapper));
ORT_RETURN_IF_NOT(qnn_model_wrapper.AddTensorWrapper(std::move(weight_tensor_wrapper)), "Failed to add weight tensor.");
}

bool is_graph_input = qnn_model_wrapper.IsGraphInput(input1_name);
LOGS(logger, VERBOSE) << "Adding Reshape (to 2D) and HWCN Transpose node after input: " << input1_name;
ORT_RETURN_IF_ERROR(qnn_model_wrapper.AddReshapeNode(input1_name,
Expand Down
Loading