qnn_executor_runner - mismatch in the skel files ?

[eliyam32]

hi,

im testing qnn_executor_runner on s25 ultra,
a Snapdragon 8 Gen 4 processor.

it seems qnn backend choses libQnnHtpV79Skel.so as the backend

but these messages seem to point to some mismatch ? it tries to call hmx_v73_convf16 ?
i.e. shouldnt it call hmx_v79_convf16 ?

V b037a:4006: CDSP0:[R]: Process "/frpc/f05c4930 qnn_executor_ru" crashed in thread "nn_3e56a57b" due to TLBMISS RW occurrence
2025-11-25 16:48:18.994 2314-2320 adsprpc cdsprpcd V b037a:4006: CDSP0:[R]: Crashed Shared Object "./libQnnHtpV79Skel.so" load address : 0x01000000
2025-11-25 16:48:18.994 2314-2320 adsprpc cdsprpcd V b037a:4006: CDSP0:[R]: [<015E5C3C>] hmx_v73_convf16_NxN_stride1+0x3C53C: (./libQnnHtpV79Skel.so)
2025-11-25 16:48:18.994 2314-2320 adsprpc cdsprpcd V b037a:4006: CDSP0:[R]: [<015E5C38>] hmx_v73_convf16_NxN_stride1+0x3C538: (./libQnnHtpV79Skel.so)
2025-11-25 16:48:18.994 2314-2320 adsprpc cdsprpcd V b037a:4006: CDSP0:[R]: [<015E5D74>] hmx_v73_convf16_NxN_stride1+0x3C674: (./libQnnHtpV79Skel.so)
2025-11-25 16:48:18.994 2314-2320 adsprpc cdsprpcd V b037a:4006: CDSP0:[R]: [<01546168>] continue_execution_bkgrnd_thread+0xA8: (./libQnnHtpV79Skel.so)
2025-11-25 16:48:18.994 2314-2320 adsprpc cdsprpcd V b037a:4006: CDSP0:[R]: [<0120EC94>] _ZN5Graph18exec_bkgrnd_workerEP12HexagonNNEnvPS_N9GraphData8ListTypeEN4hnnx3OsSE+0xD4: (./libQnnHtpV79Skel.so)
2025-11-25 16:48:18.994 2314-2320 adsprpc cdsprpcd V b037a:4006: CDSP0:[R]: [<01219EA0>] _ZNK5Graph31ubwcd_get_corresponding_surfaceEPKv+0x9E0: (./libQnnHtpV79Skel.so)

and output from adb shell
#./qnn_executor_runner --model_path ./my_model_fp16.pte --input_list_path ./raw_list.txt

[INFO] [Qnn ExecuTorch]: Deserializing processed data using QnnContextCustomProtocol
[INFO] [Qnn ExecuTorch]: create QNN Logger with log_level 1
[INFO] [Qnn ExecuTorch]: Initialize Qnn backend parameters for Qnn executorch backend type 2
[INFO] [Qnn ExecuTorch]: Caching: Caching is in RESTORE MODE.
[INFO] [Qnn ExecuTorch]: QnnContextCustomProtocol expected magic number: 0x5678abcd but get: 0x2000000
[INFO] [Qnn ExecuTorch]: Running level=1 optimization.
I 00:00:00.150474 executorch:qnn_executor_runner.cpp:313] Method loaded.
E 00:00:00.156807 executorch:method.cpp:1274] Output 0 is memory planned, or is a constant. Cannot override the existing data pointer.
I 00:00:00.156838 executorch:qnn_executor_runner.cpp:373] ignoring error from set_output_data_ptr(): 0x2
E 00:00:00.157118 executorch:method.cpp:1274] Output 1 is memory planned, or is a constant. Cannot override the existing data pointer.
I 00:00:00.157144 executorch:qnn_executor_runner.cpp:373] ignoring error from set_output_data_ptr(): 0x2
E 00:00:00.158031 executorch:method.cpp:1274] Output 2 is memory planned, or is a constant. Cannot override the existing data pointer.
I 00:00:00.158057 executorch:qnn_executor_runner.cpp:373] ignoring error from set_output_data_ptr(): 0x2
I 00:00:00.158069 executorch:qnn_executor_runner.cpp:376] Inputs prepared.
I 00:00:00.158198 executorch:qnn_executor_runner.cpp:382] Number of inputs: 1
I 00:00:00.178327 executorch:qnn_executor_runner.cpp:490] Perform 0 inference for warming up
I 00:00:00.178343 executorch:qnn_executor_runner.cpp:496] Start inference (0)
[ERROR] [Qnn ExecuTorch]: QnnDsp DspTransport call failed, error 0x00000010

[ERROR] [Qnn ExecuTorch]: QnnDsp Error from rpc transport

[ERROR] [Qnn ExecuTorch]: QnnDsp Graph forward failed in execution with err 1003

[ERROR] [Qnn ExecuTorch]: qnn_graph_execute failed. Error 1003
E 00:00:00.192908 executorch:QnnExecuTorchBackend.cpp:176] Fail to execute graph
E 00:00:00.192912 executorch:method.cpp:1426] CALL_DELEGATE execute failed at instruction 0: 0x1
I 00:00:00.192924 executorch:qnn_executor_runner.cpp:514] 1 inference took 14.576000 ms, avg 14.576000 ms
F 00:00:00.192943 executorch:qnn_executor_runner.cpp:519] In function main(), assert failed (status == Error::Ok): Execution of method forward failed with status 0x1
Aborted

cc @cccclai @winskuo-quic @shewu-quic @haowhsu-quic @DannyYuyang-quic @cbilgin

[shewu-quic]

Hi @eliyam32

Thanks to tying that. Let me check a few details.

1. Can you run a basic model on your phone?
   https://github.com/pytorch/executorch/tree/main/examples/qualcomm#simple-examples-to-verify-the-backend-is-working
2. Which qnn version is being used? Have you tried using a different qnn version?
3. If possible, could you provide the reproduce model or steps?

[eliyam32]

hi @shewu-quic

thanks for the quick reply,

1. Can you run a basic model on your phone?
   it seems to run correctly for add.pte

pa3q:/data/local/tmp/executorch_qualcomm_tutorial # ./qnn_executor_runner --model_path add.pte I 00:00:00.001286 executorch:qnn_executor_runner.cpp:232] Model file add.pte is loaded. I 00:00:00.001302 executorch:qnn_executor_runner.cpp:242] Using method forward I 00:00:00.001307 executorch:qnn_executor_runner.cpp:289] Setting up planned buffer 0, size 48. [INFO] [Qnn ExecuTorch]: Deserializing processed data using QnnContextCustomProtocol [INFO] [Qnn ExecuTorch]: create QNN Logger with log_level 1 [INFO] [Qnn ExecuTorch]: Initialize Qnn backend parameters for Qnn executorch backend type 2 [INFO] [Qnn ExecuTorch]: Caching: Caching is in RESTORE MODE. [INFO] [Qnn ExecuTorch]: QnnContextCustomProtocol expected magic number: 0x5678abcd but get: 0x2000000 [INFO] [Qnn ExecuTorch]: Running level=1 optimization. I 00:00:00.134708 executorch:qnn_executor_runner.cpp:313] Method loaded. E 00:00:00.134754 executorch:method.cpp:1274] Output 0 is memory planned, or is a constant. Cannot override the existing data pointer. I 00:00:00.134795 executorch:qnn_executor_runner.cpp:373] ignoring error from set_output_data_ptr(): 0x2 I 00:00:00.134807 executorch:qnn_executor_runner.cpp:376] Inputs prepared. I 00:00:00.134858 executorch:qnn_executor_runner.cpp:570] Input list not provided. Inputs prepared with default values set. I 00:00:00.135739 executorch:qnn_executor_runner.cpp:579] Model executed successfully. I 00:00:00.135763 executorch:qnn_executor_runner.cpp:582] Perform 0 inferences for warming up I 00:00:00.135878 executorch:qnn_executor_runner.cpp:604] 1 inferences took 0.103000 ms, avg 0.103000 ms I 00:00:00.135928 executorch:qnn_executor_runner.cpp:615] Write etdump to etdump.etdp, Size = 576 [INFO] [Qnn ExecuTorch]: Destroy Qnn backend parameters [INFO] [Qnn ExecuTorch]: Destroy Qnn context [INFO] [Qnn ExecuTorch]: Destroy Qnn device [INFO] [Qnn ExecuTorch]: Destroy Qnn backend

I tried to run the mobile_net

WARNING] the 2 th arg of node hardtanh__29 is NumberType, might need to lift
[WARNING] the 1 th arg of node hardtanh__30 is NumberType, might need to lift
[WARNING] the 2 th arg of node hardtanh__30 is NumberType, might need to lift
[WARNING] the 1 th arg of node hardtanh__31 is NumberType, might need to lift
[WARNING] the 2 th arg of node hardtanh__31 is NumberType, might need to lift
[WARNING] the 1 th arg of node hardtanh__32 is NumberType, might need to lift
[WARNING] the 2 th arg of node hardtanh__32 is NumberType, might need to lift
[WARNING] the 1 th arg of node hardtanh__33 is NumberType, might need to lift
[WARNING] the 2 th arg of node hardtanh__33 is NumberType, might need to lift
[WARNING] the 1 th arg of node hardtanh__34 is NumberType, might need to lift
[WARNING] the 2 th arg of node hardtanh__34 is NumberType, might need to lift
No quant config is implemented for op, aten.dropout.default
Quantizing(PTQ) the model...

adb: error: failed to get feature set: device 'R3CX90MTP8A' not found
adb: device <my.device.id> not found
--- there seems to be an issue with
--- but when I do adb devices its connected fine.

any advise on SimpleADB error ?
I can also manually copy and run mv2_qnn_q8.pte on device, but im not sure how

2. Which qnn version is being used? Have you tried using a different qnn version?
   qualcomm_sdk/qairt/2.37.0.250724

If possible, could you provide the reproduce model or steps?
sure , pls try to run
https://huggingface.co/qualcomm/Unet-Segmentation with s25?

thanks,
Eliya