Repository navigation
Conversation
mouxangithub
added a commit
to mouxangithub/jetlink
that referenced
this pull request
Nov 5, 2026
… + cn features fused Conflict resolution decisions: - OrtBackend init: keep both qnnBackend (cn phone_qnn vendor backend) and the new qnn: QnnPolicy parameter - providerOptions(.htp): PR's policy-driven options (htpBackend, shared I/O allocator) + cn's backend_path override for the phone's own QNN - compileContext: keep cn's OOM precheck and fail-fast diagnostics, adopt PR's contextConfiguration (strict provider contract, no CPU fallback) - build.gradle.kts: keep cn's natives-packaging (no implementation dep); PR's QNN buildConfig fields and qnnJniDir identity kept via auto-merge (jniLibs srcDir covers the AAR's QNN libs through the overrides rule) - ServerController.kt: union (mirrors + qnn_* config keys) New opt-in QNN options (all default-off): qnn_shared_io, qnn_strict, qnn_no_activation_fusion + jetlink.qnnJniDir / jetlink.qnnSharedIO etc.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds three default-off QNN options: real HTP shared allocations, explicit CPU-fallback prevention during compilation and reload, and a matching-SDK adapter that disables advanced activation fusion. Provider discovery stays generic; changed policies get separate caches. Models, precision and ordinary builds stay unchanged.
The initial numerical problem was reproducible on the tested range-safe CTV3 graph: stock QNN fusion failed six output groups, while disabling it passed the unchanged gates on 32 identical frames. Separately, shared I/O measured 42.9 → 40.5 ms mean in sequential one-minute prototype tests. These results do not establish a universal speedup or NPU occupancy; that graph preparation is not added here.
Validation: 31 focused Swift tests, 58 Android unit tests in each default/opt-in build, both release APK builds, Android native compilation, and the adapter forwarding/failure/lifetime test passed. All 16 published files were checked against the tested source. An expanded WSL run had eight large-UDP frame-assembly issues in unchanged code; those remain outside this patch. This exact upstream port has not been device-qualified.
Options and reproduction. Related: #16, #18.