[Mobile]All Operators will fallback to CPU when running through QNN EP on V79 arch SOC.

[green0765]

Describe the issue

We have a V73 capable quantiazed onnx model. It works fine on Windows platform. But when trying it on Android on V79 arch, all operators will fallback to CPU.

To reproduce

1. Build ORT with QNN EP.
2. Copy ORT and all libs from QNN SDK to Android device.
3. Runnig model through onnxruntime_perf_test.
   Command: ./onnxruntime_perf_test -e qnn -i "backend_path|libQnnHtp.so htp_performance_mode|sustained_high_performance htp_graph_finalization_optimization_mode|3" -m duration -t 60 -I test_model.onnx

if enable disable_cpu_ep_fallback: ./onnxruntime_perf_test -e qnn -i "backend_path|libQnnHtp.so htp_performance_mode|sustained_high_performance htp_graph_finalization_optimization_mode|3" -m duration -t 60 -I test_model.onnx.
It will report error: This session contains graph nodes that are assigned to the default CPU EP, but fallback to CPU EP has been explicitly disabled by the user.

Urgency

This issue is related to a project that cooperator with Microsoft and Motorola Mobile. We need support ASAP.

Platform

Android

OS Version

16

ONNX Runtime Installation

Built from Source

Compiler Version (if 'Built from Source')

NDK 2.37.1.250807

Package Name (if 'Released Package')

None

ONNX Runtime Version or Commit ID

1.23.0

ONNX Runtime API

C++/C

Architecture

ARM64

Execution Provider

Other / Unknown

Execution Provider Library Version

QNN 2.37.1.250807

[skottmckay]

Set the log severity to VERBOSE and look for "Node placements" in the output to see which nodes are assigned to CPU. Based on your description it's not clear that all nodes are falling back to CPU.