Skip to content

Fix quantized Pad op dimension check when QDQ optimizer is disabled #582

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
saurabhkale17 merged 1 commit into from
Feb 18, 2025
Merged

Fix quantized Pad op dimension check when QDQ optimizer is disabled #582

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
31 changes: 24 additions & 7 deletions onnxruntime/core/providers/openvino/ov_versions/data_ops.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -752,16 +752,33 @@ bool DataOps::node_is_supported(const NodeIndex node_idx, bool& has_external_wei
if (op_is_supported(optype, no_dimension_supported_)) {
return;
}
if (npu_qdq_optimizer_enabled_) {
// Pad Op with DQ inputs will be optimized out in the qdq optimization pass, so mark those no dim Pad ops
// supported here
if (optype == "Pad") {
for (Node::NodeConstIterator it_dq = node->InputNodesBegin(); it_dq != node->InputNodesEnd(); ++it_dq) {
const auto& DQ = &*it_dq;
if (DQ->OpType() == "DequantizeLinear") return;
// Special handling for the "Pad" operator
if (optype == "Pad") {
bool is_quantized = false;
// Detect a quantized model by checking for a DequantizeLinear input
for (Node::NodeConstIterator it_dq = node->InputNodesBegin(); it_dq != node->InputNodesEnd(); ++it_dq) {
const auto& DQ = &*it_dq;
if (DQ->OpType() == "DequantizeLinear") {
is_quantized = true;
break;
}
}
if (is_quantized) {
// For quantized Pad ops when the QDQ optimizer is disabled,
// bypass the unsupported dimension check to ensure 'pad_value' is constant
if (!npu_qdq_optimizer_enabled_) {
#ifndef NDEBUG
if (openvino_ep::backend_utils::IsDebugEnabled()) {
// Pad Op with DQ inputs gets optimized in the downstream,
// so mark those no dim quantized Pad ops supported here
std::cout << "QDQ optimizer disabled; quantized Pad op detected (DequantizeLinear present), so marking those no dim quantized Pad ops as supported" << std::endl;
}
#endif
}
return;
}
}
// For ops that haven't been handled above, mark as unsupported dim
has_unsupported_dimension = true;
return;
} else {
Expand Down