Update

[ghstack-poisoned]

Author: Gasoonjia

.ci/docker/requirements-ci.txt

@@ -9,7 +9,7 @@ torchsr==1.0.4
 transformers==5.3.0
 zstd==1.5.5.1
 pandas>=2.2.2; python_version >= '3.10'
-pytest==7.2.0
+pytest==7.3.0
 pytest-cov==4.1.0
 expecttest==0.1.6
 hypothesis==6.84.2

.ci/scripts/test-native.sh

@@ -20,14 +20,18 @@ case "${component}" in
     test_regex='^native_runtime_test$'
     ;;
   cpu)
+    # The CPU plan test covers ETDump tracing, including operator shapes.
     cmake_args+=(
       -DEXECUTORCH_BUILD_NATIVE_CPU=ON
       -DEXECUTORCH_BUILD_XNNPACK=ON
       -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON
       -DEXECUTORCH_BUILD_NATIVE_VISION_RUNNER=ON
+      -DEXECUTORCH_BUILD_DEVTOOLS=ON
+      -DEXECUTORCH_ENABLE_EVENT_TRACER=ON
+      -DEXECUTORCH_CPU_XNNPACK_OPERATOR_SHAPES=ON
     )
-    targets+=(native_cpu_engine_test run_classification_cpu)
-    test_regex='^native_cpu_engine_test$'
+    targets+=(native_cpu_engine_test cpu_plan_test run_classification_cpu)
+    test_regex='^(native_cpu_engine|cpu_plan)_test$'
     ;;
   ggml)
     cmake_args+=(
@@ -36,8 +40,9 @@ case "${component}" in
       -DEXECUTORCH_BUILD_NATIVE_VISION_RUNNER=ON
       -DGGML_NATIVE=OFF
     )
-    targets+=(run_hf_llm_ggml native_ggml_engine_test native_llm_runner_utils_test run_classification_ggml_cpu)
-    test_regex='^native_(ggml_engine|llm_runner_utils)_test$'
+    # TODO: run the other sample tests once they pass.
+    targets+=(run_hf_llm_ggml native_ggml_engine_test native_llm_runner_utils_test run_classification_ggml_cpu native_output_check_test native_ggml_sample_test)
+    test_regex='^native_(ggml_engine|llm_runner_utils|output_check)_test$|^native_ggml_sample_test_act_quant$'
     ;;
   vulkan)
     cmake_args+=(
@@ -67,7 +72,7 @@ cmake -S . -B "${build_dir}" -G Ninja \
 cmake --build "${build_dir}" \
   --parallel "${build_jobs}" \
   --target "${targets[@]}"
-ctest --test-dir "${build_dir}/backends/native/test" \
+ctest --test-dir "${build_dir}" \
   -R "${test_regex}" --output-on-failure --no-tests=error --timeout 120
 
 if [[ "${component}" == "ggml" ]]; then

.claude/faq.md

@@ -7,7 +7,7 @@ Error codes defined in `runtime/core/error.h`.
 |------|------|--------------|
 | 0x10 | InvalidArgument | Input shape mismatch - inputs don't match export shapes. Use dynamic shapes if needed. |
 | 0x14 | OperatorMissing | Selective build missing operator. Regenerate `et_operator_library` from current model. |
-| 0x20 | NotFound | Missing backend. Link with `--whole-archive`: `-Wl,--whole-archive libxnnpack_backend.a -Wl,--no-whole-archive` |
+| 0x20 | NotFound | Missing backend. Link with `--whole-archive`: `-Wl,--whole-archive libxnnpack_backend.a -Wl,--no-whole-archive libxnnpack_profiler.a` |
 
 ## Export Issues
 

.claude/skills/building/SKILL.md

@@ -202,7 +202,7 @@ Most commonly needed flags (full list: `CMakeLists.txt`):
 | Artifact | Location |
 |----------|----------|
 | Core runtime | `cmake-out/lib/libexecutorch.a` |
-| XNNPACK backend | `cmake-out/lib/libxnnpack_backend.a` |
+| XNNPACK backend | `cmake-out/lib/libxnnpack_backend.a` and `cmake-out/lib/libxnnpack_profiler.a` |
 | executor_runner | `cmake-out/executor_runner` (Ninja/Make) or `cmake-out/Release/executor_runner` (Xcode) |
 | Model runners | `cmake-out/examples/models/<model>/<runner>` |
 

.github/workflows/_llm_server.yml

@@ -42,11 +42,16 @@ jobs:
         export PYTHONPATH="$(dirname "${PWD}"):${PYTHONPATH:-}"
         export PYTHONDONTWRITEBYTECODE=1
 
-        python -m pytest -q \
+        python -m pytest -q -ra -m 'not native_worker' \
+          --junitxml="${RUNNER_TEST_RESULTS_DIR}/llm-server.xml" \
           examples/llm_server/python/tests \
           examples/models/muse-glimmer/tests/test_serve.py \
           examples/models/muse-glimmer/tests/test_serve_batching.py \
-          examples/models/muse-glimmer/tests/test_stream_parser.py
+          examples/models/muse-glimmer/tests/test_stream_parser.py \
+          examples/models/muse-glimmer/tests/test_migration_contracts.py \
+          examples/models/muse-glimmer/tests/test_atem_tool_parser.py \
+          examples/models/muse-glimmer/tests/test_*regressions.py \
+          examples/models/muse-glimmer/tests/test_client_contracts.py
 
         cmake -S . -B cmake-out \
           -DCMAKE_BUILD_TYPE=Release \
@@ -62,6 +67,10 @@ jobs:
           -j "$(nproc)"
         ctest --test-dir cmake-out/examples/llm_server/cpp --output-on-failure
 
+        export EXECUTORCH_TEST_MULTIPLEXED_WORKER="${PWD}/cmake-out/examples/llm_server/cpp/test_multiplexed_worker"
         export EXECUTORCH_BATCHING_TEST_WORKER="${PWD}/cmake-out/examples/llm_server/cpp/test_batching_worker"
+        test -x "${EXECUTORCH_TEST_MULTIPLEXED_WORKER}"
         test -x "${EXECUTORCH_BATCHING_TEST_WORKER}"
-        python -m pytest -q examples/llm_server/python/tests/test_batching_end_to_end.py
+        python -m pytest -q -ra -m native_worker \
+          --junitxml="${RUNNER_TEST_RESULTS_DIR}/llm-server-native.xml" \
+          examples/llm_server/python/tests

.github/workflows/_test_cortex_m_ops.yml

@@ -59,6 +59,6 @@ jobs:
         # op set, implementation tests check FVP numerics. Both are parametrized
         # over --cortex-m-target, so a future target-dependent lowering change is
         # caught here. (cortex-m55 runs on pull via the full-suite job.)
-        pytest --config-file=backends/arm/test/pytest.ini \
+        pytest -c backends/arm/test/pytest.ini \
           backends/cortex_m/test/ops \
           --cortex-m-target=${{ matrix.target }}

.github/workflows/cuda.yml

@@ -48,6 +48,7 @@ on:
       - exir/passes/insert_write_back_for_buffers_pass.py
       - exir/passes/propagate_device_config.py
       - exir/passes/propagate_device_pass.py
+      - exir/passes/replace_complex_with_real_pass.py
   workflow_dispatch:
     inputs:
       models:
@@ -476,9 +477,9 @@ jobs:
   unittest-cuda:
     name: unittest-cuda (${{ matrix.shard }})
     needs: [changed-files, run-decision]
-    # This job runs the pybindings device test and the device placement test,
-    # so it also has to fire on the pybindings and device pass paths the
-    # workflow filter lists.
+    # This job runs the pybindings device test, the device placement test and
+    # the complex number tests, so it also has to fire on the pybindings and
+    # exir pass paths the workflow filter lists.
     if: |
       contains(needs.changed-files.outputs.changed-files, 'backends/cuda') ||
       contains(needs.changed-files.outputs.changed-files, 'backends/aoti') ||
@@ -493,6 +494,7 @@ jobs:
       contains(needs.changed-files.outputs.changed-files, 'exir/passes/insert_write_back_for_buffers_pass.py') ||
       contains(needs.changed-files.outputs.changed-files, 'exir/passes/propagate_device_config.py') ||
       contains(needs.changed-files.outputs.changed-files, 'exir/passes/propagate_device_pass.py') ||
+      contains(needs.changed-files.outputs.changed-files, 'exir/passes/replace_complex_with_real_pass.py') ||
       needs.run-decision.outputs.is-full-run == 'true'
     # kernel: Triton kernels and their autotuning. system: lowering passes, op
     # dispatch, export/runtime, and model tests. Together they do not fit the
@@ -1219,12 +1221,8 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - variant: kquant-17gb
-            mode: solo-text
           - variant: kquant-17gb
             mode: dflash-image-batching
-          - variant: kquant-17gb
-            mode: dflash-image
           - variant: kquant-dynamic
             mode: solo-text
           - variant: kquant-dynamic