Merge branch 'main' into bundled-proper

Author: Erik-Lundell

.ci/docker/build.sh

@@ -89,10 +89,10 @@ case "${IMAGE_NAME}" in
     OS_VERSION=24.04
     GCC_VERSION=14
     ;;
-  executorch-ubuntu-26.04-gcc15)
+  executorch-ubuntu-26.04-gcc14)
     LINTRUNNER=""
     OS_VERSION=26.04
-    GCC_VERSION=15
+    GCC_VERSION=14
     ;;
   *)
     echo "Invalid image name ${IMAGE_NAME}"

.github/workflows/build-cadence-runner.yml

@@ -10,14 +10,45 @@ on:
     tags:
       - ciflow/nightly/*
   pull_request:
+  pull_request_target:
+    types: [labeled]
   workflow_dispatch:
 
 concurrency:
-  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
+  group: ${{ github.workflow }}-${{ github.event_name }}-${{ github.event.pull_request.number || github.sha }}
   cancel-in-progress: true
 
 jobs:
+  gate:
+    runs-on: ubuntu-latest
+    outputs:
+      run-cadence: ${{ steps.decide.outputs.run }}
+    steps:
+      - id: decide
+        env:
+          EVENT: ${{ github.event_name }}
+          IS_FORK: ${{ github.event.pull_request.head.repo.full_name != github.repository }}
+          HAS_CLA: ${{ contains(github.event.pull_request.labels.*.name, 'CLA Signed') }}
+          HAS_EXPORT: ${{ contains(github.event.pull_request.labels.*.name, 'meta-exported') }}
+        run: |
+          run=false
+          case "${EVENT}" in
+            push|schedule|workflow_dispatch)
+              run=true
+              ;;
+            pull_request)
+              [ "${IS_FORK}" = "false" ] && run=true
+              ;;
+            pull_request_target)
+              if [ "${IS_FORK}" = "true" ] && [ "${HAS_CLA}" = "true" ] && [ "${HAS_EXPORT}" = "true" ]; then
+                run=true
+              fi
+              ;;
+          esac
+          echo "run=${run}" >> "${GITHUB_OUTPUT}"
+
   cpu-build:
+    if: github.event_name != 'pull_request_target'
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
     permissions:
       id-token: write
@@ -44,6 +75,7 @@ jobs:
 
   cpu-test:
     needs: cpu-build
+    if: github.event_name != 'pull_request_target'
     permissions:
       id-token: write
       contents: read
@@ -56,19 +88,23 @@ jobs:
   # lives in _xtensa_build.yml. fusion_g3 is omitted until the upstream fusion_g3
   # <-> nnlib-FusionG3 API skew is fixed (its runner does not link).
   hifi-build:
+    needs: gate
+    if: needs.gate.outputs.run-cadence == 'true'
     permissions:
       id-token: write
       contents: read
     uses: ./.github/workflows/_xtensa_build.yml
     with:
       backend: hifi4
-      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      ref: ${{ (github.event_name == 'pull_request' || github.event_name == 'pull_request_target') && github.event.pull_request.head.sha || github.sha }}
 
   vision-build:
+    needs: gate
+    if: needs.gate.outputs.run-cadence == 'true'
     permissions:
       id-token: write
       contents: read
     uses: ./.github/workflows/_xtensa_build.yml
     with:
       backend: vision
-      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      ref: ${{ (github.event_name == 'pull_request' || github.event_name == 'pull_request_target') && github.event.pull_request.head.sha || github.sha }}

.github/workflows/build-cmsis-pack.yml

@@ -0,0 +1,137 @@
+# Copyright 2026 Arm Limited and/or its affiliates.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+name: Build CMSIS Pack
+
+on:
+  schedule:
+    # Nightly at 03:00 UTC, staggered after nightly.yml (02:00) so the
+    # shared runner pool isn't hit by both at the same minute.
+    - cron: 0 3 * * *
+  release:
+    # Build (and, for non-prerelease, publish) the pack when a GitHub
+    # Release is created. The tag the release points at drives the pack
+    # version via GITHUB_REF below.
+    types: [published]
+  push:
+    branches:
+      - main
+      - release/*
+    paths:
+      - .github/workflows/build-cmsis-pack.yml
+      - backends/arm/cmsis_pack/**
+      - backends/arm/cmsis_pack/scripts/**
+      - backends/arm/runtime/**
+      - backends/cortex_m/**
+      - kernels/portable/**
+      - kernels/quantized/**
+      - runtime/**
+      - schema/**
+  pull_request:
+    paths:
+      - .github/workflows/build-cmsis-pack.yml
+      - backends/arm/cmsis_pack/**
+      - backends/arm/cmsis_pack/scripts/**
+  workflow_dispatch:
+    inputs:
+      version_override:
+        description: 'Override pack version (e.g., 1.2.0). Leave empty to derive from version.txt'
+        required: false
+        type: string
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
+  cancel-in-progress: true
+
+jobs:
+  build-cmsis-pack:
+    name: build-cmsis-pack
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    permissions:
+      id-token: write
+      contents: read
+    with:
+      runner: linux.2xlarge
+      docker-image: ci-image:executorch-ubuntu-22.04-arm-sdk
+      submodules: 'recursive'
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      timeout: 60
+      upload-artifact: cmsis-pack-artifact
+      script: |
+        set -eux
+
+        echo "::group::Setup environment"
+        # The generic Linux job chooses to use base env, not the one setup by the image
+        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
+        conda activate "${CONDA_ENV}"
+
+        source .ci/scripts/utils.sh
+        install_executorch "--use-pt-pinned-commit"
+        echo "::endgroup::"
+
+        echo "::group::Install ARM toolchain"
+        .ci/scripts/setup-arm-baremetal-tools.sh
+        source examples/arm/arm-scratch/setup_path.sh
+        echo "::endgroup::"
+
+        echo "::group::Cross-compile ExecuTorch for Cortex-M"
+        # Stage 1: Build core ExecuTorch with arm-none-eabi-gcc
+        # This generates required headers (flatbuffers, schema)
+        backends/arm/scripts/build_executorch.sh
+        CMAKE_BUILD_DIR="$(pwd)/cmake-out-arm"
+        echo "::endgroup::"
+
+        echo "::group::Determine pack version"
+        # Derive version from tag, input override, schedule (nightly), or version.txt
+        BASE_VER="$(cat version.txt | sed 's/a0$//')"
+        if [[ -n "${{ inputs.version_override || '' }}" ]]; then
+          PACK_VERSION="${{ inputs.version_override }}"
+        elif [[ "${GITHUB_REF}" == refs/tags/v* ]]; then
+          # Strip leading 'v' and any -rc suffix for release tags
+          PACK_VERSION="${GITHUB_REF#refs/tags/v}"
+        elif [[ "${{ github.event_name }}" == "schedule" ]]; then
+          PACK_VERSION="${BASE_VER}-nightly-$(date -u +%Y%m%d)"
+        else
+          PACK_VERSION="${BASE_VER}-dev"
+        fi
+        echo "Pack version: ${PACK_VERSION}"
+        echo "::endgroup::"
+
+        echo "::group::Build CMSIS Pack"
+        backends/arm/cmsis_pack/scripts/build_pack.sh \
+          --executorch-root "$(pwd)" \
+          --build-dir "${CMAKE_BUILD_DIR}" \
+          --version "${PACK_VERSION}" \
+          --output-dir "$(pwd)/artifacts-to-be-uploaded"
+        echo "::endgroup::"
+
+        # Structural validation and consumer-build smoke are intentionally
+        # not run in CI yet. See:
+        #   backends/arm/cmsis_pack/test/validate_pack.py    (structural)
+        #   backends/arm/cmsis_pack/test/smoke/run.sh        (cbuild via
+        #                                                     AVH-MLOps)
+        # for the local test drivers.
+
+  # Attach the pack to the GitHub Release when a non-prerelease release is
+  # published. Prereleases still build + validate via the release trigger
+  # but are not published.
+  publish-cmsis-pack:
+    if: github.event_name == 'release' && !github.event.release.prerelease
+    needs: build-cmsis-pack
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+    steps:
+      - name: Download pack artifact
+        uses: actions/download-artifact@v4
+        with:
+          name: cmsis-pack-artifact
+          path: pack-output
+
+      - name: Upload to GitHub Release
+        uses: softprops/action-gh-release@v2
+        with:
+          files: pack-output/*.pack
+          tag_name: ${{ github.ref_name }}

.github/workflows/docker-builds.yml

@@ -43,7 +43,7 @@ jobs:
           executorch-ubuntu-22.04-mediatek-sdk,
           executorch-ubuntu-22.04-clang12-android,
           executorch-ubuntu-24.04-gcc14,
-          executorch-ubuntu-26.04-gcc15,
+          executorch-ubuntu-26.04-gcc14,
         ]
         include:
           - docker-image-name: executorch-ubuntu-22.04-gcc11-aarch64

.github/workflows/mlx.yml

@@ -79,6 +79,7 @@ jobs:
           backends/mlx/test/test_pattern_utils.py \
           backends/mlx/test/test_partitioner.py \
           backends/mlx/test/test_serialization_dedup.py \
+          backends/mlx/test/test_slot_recycling.py \
           examples/models/gemma4_31b/quant/tests/test_pack_mlx.py \
           examples/models/gemma4_31b/tests/test_mlx_pipeline.py \
           -v

.github/workflows/pull.yml

@@ -1400,7 +1400,7 @@ jobs:
       docker-image: ci-image:executorch-ubuntu-22.04-clang12
       submodules: 'recursive'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
-      timeout: 90
+      timeout: 120
       script: |
         set -eux
 

.lintrunner.toml

@@ -196,7 +196,6 @@ exclude_patterns = [
 
     # Kernel areas to onboard separately.
     'kernels/optimized/**',
-    'kernels/portable/**',
 
     # Runtime areas to onboard incrementally.
     'runtime/backend/**',
@@ -245,6 +244,23 @@ command = [
     '--extra-arg=--suppress=duplicateBranch:*kernels/test/*',
     '--extra-arg=--suppress=useStlAlgorithm:*kernels/test/*',
     '--extra-arg=--suppress=functionStatic:*kernels/test/*',
+    # Portable kernels are referenced through generated registration tables,
+    # which cppcheck CTU does not see as direct C++ calls.
+    '--extra-arg=--suppress=unusedFunction:*kernels/portable/*',
+    # Cppcheck does not parse several ExecuTorch macro idioms used heavily in
+    # portable kernels, including empty macro arguments and printf-style macros
+    # concatenated into string literals.
+    '--extra-arg=--suppress=unknownMacro:*kernels/portable/*',
+    '--extra-arg=--suppress=syntaxError:*kernels/portable/*',
+    # Keep style-only portable-kernel findings scoped to this tree rather than
+    # rewriting broad helper code for lint-only advice.
+    '--extra-arg=--suppress=useStlAlgorithm:*kernels/portable/*',
+    '--extra-arg=--suppress=variableScope:*kernels/portable/*',
+    '--extra-arg=--suppress=constParameterReference:*kernels/portable/*',
+    '--extra-arg=--suppress=constParameterPointer:*kernels/portable/*',
+    '--extra-arg=--suppress=functionStatic:*kernels/portable/*',
+    '--extra-arg=--suppress=knownConditionTrueFalse:*kernels/portable/*',
+    '--extra-arg=--suppress=invalidFunctionArg:*kernels/portable/*',
     '--',
     '@{{PATHSFILE}}'
 ]

backends/apple/coreml/runtime/delegate/ETCoreMLModel.mm

@@ -194,7 +194,15 @@ - (nullable instancetype)initWithAsset:(ETCoreMLAsset *)asset
         return nil;
     }
 
-    MLModel *mlModel = [MLModel modelWithContentsOfURL:asset.contentURL
+    NSURL *contentURL = asset.contentURL;
+    if (contentURL == nil) {
+        ETCoreMLLogErrorAndSetNSError(error,
+                                      ETCoreMLErrorCorruptedModel,
+                                      "asset.contentURL is nil");
+        return nil;
+    }
+
+    MLModel *mlModel = [MLModel modelWithContentsOfURL:contentURL
                                          configuration:configuration
                                                  error:error];
     if (!mlModel) {

backends/arm/TARGETS

@@ -100,6 +100,14 @@ runtime.python_library(
     resources = [
         "vgf/shaders/grid_sampler.glsl",
         "vgf/shaders/grid_sampler.spirv.b64",
+        "vgf/shaders/grid_sampler_sampler.glsl",
+        "vgf/shaders/grid_sampler_sampler.spirv.b64",
+        "vgf/shaders/grid_sampler_sampler_align_corners.glsl",
+        "vgf/shaders/grid_sampler_sampler_align_corners.spirv.b64",
+        "vgf/shaders/grid_sampler_sampler_int8.glsl",
+        "vgf/shaders/grid_sampler_sampler_int8.spirv.b64",
+        "vgf/shaders/grid_sampler_sampler_int8_align_corners.glsl",
+        "vgf/shaders/grid_sampler_sampler_int8_align_corners.spirv.b64",
     ],
     deps = [
         ":arm_compile_spec",

backends/arm/_passes/__init__.py

@@ -9,6 +9,7 @@
 from .accumulate_index_put_pass import AccumulateIndexPutPass  # noqa
 from .broadcast_args_pass import BroadcastArgsPass  # noqa
 from .canonicalize_gather_pass import CanonicalizeGatherPass  # noqa
+from .canonicalize_view_copy_permute_pass import CanonicalizeViewCopyPermutePass  # noqa
 from .cast_int64_pass import CastInt64BuffersToInt32Pass  # noqa
 from .cast_to_int32_pass import CastToInt32Pass  # noqa
 from .constant_folding_pass import ConstantFoldingPass  # noqa
@@ -170,6 +171,7 @@
 from .rewrite_le_lt_to_ge_gt_pass import RewriteLeLtToGeGtPass  # noqa
 from .rewrite_matmul import RewriteMatmulPass  # noqa
 from .rewrite_max_pool2d_pass import RewriteMaxPool2dPass  # noqa
+from .rewrite_mxfp_linear import RewriteMXFPLinearPass  # noqa
 from .rewrite_pad import RewritePadPass  # noqa
 from .rewrite_slice import RewriteSlicePass  # noqa
 from .rewrite_upsample import RewriteUpsamplePass  # noqa