Qualcomm AI Engine Direct - context dump utility

[haowhsu-quic]

Summary

• utility for dumping compiled binaries (QNN context_binary / QCIR)
• test cases

Test plan

python backends/qualcomm/tests/test_qnn_delegate.py -k TestQNNQuantizedUtils.test_qnn_backend_context_extraction -s $DEVICE_SERIAL -m SM8650 -b build-android/

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/7931

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

✅ No Failures

As of commit a89479d with merge base b1d76c9 ():
💚 Looks good so far! There are no failures yet. 💚

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[haowhsu-quic]

Hi @cccclai, this PR provides methods to save compiled QNN context binaries / QCIR from EdgeProgramManager or PTE file. It might be helpful for user to conduct further analysis.
Please have a look, thank you.

Hi @limintang, this one might be similar to #6745 but slightly flexible. Hopefully this could help, thanks.

[haowhsu-quic]

Hi @cccclai, sorry for pinging. We probably need this feature for performance analysis when doing XR model enablement, please help merge this if it looks good to you.

[haowhsu-quic]

@pytorchbot label "release notes: qualcomm"

[cccclai]

Sorry a bit late on this, looks good to me.

[facebook-github-bot]

@cccclai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[cccclai]

Hey there is an internal error

In file included from fbcode/executorch/backends/qualcomm/aot/python/PyQnnManagerAdaptor.cpp:8:
buck-out/v2/gen/fbcode/339a51a8c50acc11/executorch/backends/qualcomm/aot/python/__PyQnnManagerAdaptor__/buck-private-headers/executorch/backends/qualcomm/aot/python/PyQnnManagerAdaptor.h:395:32: error: unused variable 'binary' [-Werror,-Wunused-variable]
  395 |     QnnExecuTorchContextBinary binary(
      |                                ^~~~~~
1 error generated.

Can you help fixing it?

[cccclai]

Can you attach this patch?

--- a/executorch/backends/qualcomm/aot/python/PyQnnManagerAdaptor.h
+++ b/executorch/backends/qualcomm/aot/python/PyQnnManagerAdaptor.h
@@ -392,7 +392,7 @@
 
   py::array_t<char> StripProtocol(const py::bytes& preprocessed_binary) {
     py::buffer_info info(py::buffer(preprocessed_binary).request());
-    QnnExecuTorchContextBinary binary(
+    __ET_UNUSED QnnExecuTorchContextBinary binary(
         {info.ptr, static_cast<uint64_t>(info.size * info.itemsize)});
 
     void* buf_ptr = nullptr;

[haowhsu-quic]

Can you attach this patch?

--- a/executorch/backends/qualcomm/aot/python/PyQnnManagerAdaptor.h
+++ b/executorch/backends/qualcomm/aot/python/PyQnnManagerAdaptor.h
@@ -392,7 +392,7 @@
 
   py::array_t<char> StripProtocol(const py::bytes& preprocessed_binary) {
     py::buffer_info info(py::buffer(preprocessed_binary).request());
-    QnnExecuTorchContextBinary binary(
+    __ET_UNUSED QnnExecuTorchContextBinary binary(
         {info.ptr, static_cast<uint64_t>(info.size * info.itemsize)});
 
     void* buf_ptr = nullptr;

Sorry didn't catch that, I just remove it and rebase again. Thank you.

[facebook-github-bot]

@cccclai has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[cccclai]

Looks good to me. Thank you!