Merge branch 'main' into export-D77479748

Author: xuhdev

.ci/scripts/setup-arm-baremetal-tools.sh

@@ -8,4 +8,4 @@
 # Setup arm example environment (including TOSA tools)
 git config --global user.email "[email protected]"
 git config --global user.name "Github Executorch"
-bash examples/arm/setup.sh --i-agree-to-the-contained-eula
+bash examples/arm/setup.sh --i-agree-to-the-contained-eula ${@:-}

.ci/scripts/test_model.sh

@@ -188,22 +188,6 @@ test_model_with_qnn() {
     EXPORT_SCRIPT=edsr
     # Additional deps for edsr
     pip install piq
-  elif [[ "${MODEL_NAME}" == "cvt" ]]; then
-    EXPORT_SCRIPT=cvt
-  elif [[ "${MODEL_NAME}" == "dit" ]]; then
-    EXPORT_SCRIPT=dit
-  elif [[ "${MODEL_NAME}" == "efficientnet" ]]; then
-    EXPORT_SCRIPT=efficientnet
-  elif [[ "${MODEL_NAME}" == "focalnet" ]]; then
-    EXPORT_SCRIPT=focalnet
-  elif [[ "${MODEL_NAME}" == "mobilevit_v1" ]]; then
-    EXPORT_SCRIPT=mobilevit_v1
-  elif [[ "${MODEL_NAME}" == "mobilevit_v2" ]]; then
-    EXPORT_SCRIPT=mobilevit_v2
-  elif [[ "${MODEL_NAME}" == "pvt" ]]; then
-    EXPORT_SCRIPT=pvt
-  elif [[ "${MODEL_NAME}" == "swin" ]]; then
-    EXPORT_SCRIPT=swin_transformer
   elif [[ "${MODEL_NAME}" == "albert" ]]; then
     EXPORT_SCRIPT=albert
   elif [[ "${MODEL_NAME}" == "bert" ]]; then
@@ -212,8 +196,6 @@ test_model_with_qnn() {
     EXPORT_SCRIPT=distilbert
   elif [[ "${MODEL_NAME}" == "eurobert" ]]; then
     EXPORT_SCRIPT=eurobert
-  elif [[ "${MODEL_NAME}" == "roberta" ]]; then
-    EXPORT_SCRIPT=roberta
   else
     echo "Unsupported model $MODEL_NAME"
     exit 1
@@ -228,13 +210,10 @@ test_model_with_qnn() {
     "dl3"|"mv3"|"mv2"|"ic4"|"ic3"|"vit"|"mb"|"w2l")
         SCRIPT_FOLDER=scripts
         ;;
-    "cvt"|"dit"|"focalnet"|"mobilevit_v2"|"pvt"|"swin")
-        SCRIPT_FOLDER=oss_scripts
-        ;;
-    "albert"|"bert"|"distilbert"|"roberta"|"efficientnet"|"mobilevit_v1")
+    "albert"|"bert"|"distilbert")
         pip install evaluate
         SCRIPT_FOLDER=oss_scripts
-        # 16bit models will encounter op validation fail on some operations,
+        # Bert models running in 16bit will encounter op validation fail on some operations,
         # which requires CHIPSET >= SM8550.
         QNN_CHIPSET=SM8550
         ;;

.github/workflows/build-presets.yml

@@ -34,6 +34,45 @@ jobs:
         ${CONDA_RUN} cmake --preset ${{ matrix.preset }}
         ${CONDA_RUN} cmake --build cmake-out -j$(( $(sysctl -n hw.ncpu) - 1 ))
 
+  zephyr:
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    strategy:
+      fail-fast: false
+      matrix:
+        preset: [zephyr]
+    with:
+      job-name: build
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      runner: linux.2xlarge
+      docker-image: executorch-ubuntu-22.04-arm-sdk
+      submodules: recursive
+      timeout: 90
+      script: |
+        set -eux
+        # The generic Linux job chooses to use base env, not the one setup by the image
+        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
+        conda activate "${CONDA_ENV}"
+
+        ./install_requirements.sh > /dev/null
+
+        # Download toolchain
+        toolchain_url="https://github.com/zephyrproject-rtos/sdk-ng/releases/download/v0.17.2/toolchain_linux-x86_64_arm-zephyr-eabi.tar.xz"
+        toolchain_dir="arm-zephyr-eabi"
+        curl --output "${toolchain_dir}.tar.xz" -L "${toolchain_url}"
+
+        # Verify download
+        echo "93128be0235cf5cf5f1ee561aa6eac5f  ${toolchain_dir}.tar.xz" > arm-zephyr-eabi.md5
+        md5sum -c --strict arm-zephyr-eabi.md5
+        
+        # Extract and install to PATH
+        tar xf "${toolchain_dir}.tar.xz"
+        rm -f "${toolchain_dir}.tar.xz"
+        toolchain_bin_path="$(cd ${toolchain_dir}/bin && pwd)"
+        export PATH=$PATH:${toolchain_bin_path}
+
+        # Build Arm Zephyr Preset
+        cmake --preset ${{ matrix.preset }}
+        cmake --build cmake-out -j$(( $(nproc) - 1 ))
   linux:
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
     strategy:

.github/workflows/trunk.yml

@@ -223,6 +223,10 @@ jobs:
     permissions:
       id-token: write
       contents: read
+    strategy:
+      matrix:
+        os: [bare_metal, zephyr-preset]
+      fail-fast: false
     with:
       runner: linux.2xlarge
       docker-image: executorch-ubuntu-22.04-arm-sdk
@@ -234,35 +238,62 @@ jobs:
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
 
+        cxx_flags="-fno-exceptions -fno-rtti -Wall -Werror -Wno-int-in-bool-context -DET_HAVE_PREAD=0"
+        setup_script_args=""
+        if [[ ${{ matrix.os}} == "bare_metal" ]]; then 
+          toolchain_prefix=arm-none-eabi-
+          threshold="103268" # ~100KiB
+          toolchain_cmake=examples/arm/ethos-u-setup/arm-none-eabi-gcc.cmake
+        elif [[ ${{ matrix.os}} == "zephyr-preset" ]]; then 
+          setup_script_args="--target-toolchain zephyr"
+          toolchain_prefix=arm-zephyr-eabi-
+          threshold="133120" # should be ~125KB, set threshold to 130KB
+          toolchain_cmake=examples/zephyr/x86_64-linux-arm-zephyr-eabi-gcc.cmake
+        else
+          echo "Fail unsupport OS selection ${{ matrix.os }}"
+          exit 1
+        fi
+
         source .ci/scripts/utils.sh
         install_executorch "--use-pt-pinned-commit"
-        .ci/scripts/setup-arm-baremetal-tools.sh
+        .ci/scripts/setup-arm-baremetal-tools.sh ${setup_script_args}
         source examples/arm/ethos-u-scratch/setup_path.sh
 
-        # User baremetal toolchain
-        arm-none-eabi-c++ --version
-        toolchain_cmake=examples/arm/ethos-u-setup/arm-none-eabi-gcc.cmake
+        # User toolchain
+        ${toolchain_prefix}c++ --version
+
+        # Setup cmake target to desired toolchain
         toolchain_cmake=$(realpath ${toolchain_cmake})
 
-        # Build and test size test
-        bash test/build_size_test.sh "-DCMAKE_TOOLCHAIN_FILE=${toolchain_cmake} -DEXECUTORCH_BUILD_ARM_BAREMETAL=ON"
+        # Build and run size test
+        if [[ ${{ matrix.os}} == "bare_metal" ]]; then
+          bash test/build_size_test.sh "-DCMAKE_TOOLCHAIN_FILE=${toolchain_cmake} -DEXECUTORCH_BUILD_ARM_BAREMETAL=ON"
+        elif [[ ${{ matrix.os}} == "zephyr-preset" ]]; then
+          CXXFLAGS=${cxx_flags} cmake --preset zephyr -DCMAKE_BUILD_TYPE=Release -DEXECUTORCH_OPTIMIZE_SIZE=ON -DCMAKE_INSTALL_PREFIX=cmake-out -Bcmake-out .
+          cmake --build cmake-out -j9 --target install --config Release
+          CXXFLAGS=${cxx_flags}  cmake -DCMAKE_TOOLCHAIN_FILE=${toolchain_cmake} -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=cmake-out -Bcmake-out/test test
+          cmake --build cmake-out/test -j9 --config Release
+        else
+          echo "Fail unsupport OS selection ${{ matrix.os }}"
+          exit 1
+        fi
+
         elf="cmake-out/test/size_test"
 
         # Dump basic info
         ls -al ${elf}
-        arm-none-eabi-size ${elf}
+        ${toolchain_prefix}size ${elf}
 
-        # Dump symbols
+        # Dump symbol
         python .github/scripts/run_nm.py -e ${elf}
-        python .github/scripts/run_nm.py -e ${elf} -f "executorch" -p "arm-none-eabi-"
-        python .github/scripts/run_nm.py -e ${elf} -f "executorch_text" -p "arm-none-eabi-"
+        python .github/scripts/run_nm.py -e ${elf} -f "executorch" -p "${toolchain_prefix}"
+        python .github/scripts/run_nm.py -e ${elf} -f "executorch_text" -p "${toolchain_prefix}"
 
         # Add basic guard - TODO: refine this!
-        arm-none-eabi-strip ${elf}
+        ${toolchain_prefix}strip ${elf}
         output=$(ls -la ${elf})
         arr=($output)
         size=${arr[4]}
-        threshold="103268" # ~100KiB
         echo "size: $size, threshold: $threshold"
         if [[ "$size" -le "$threshold" ]]; then
           echo "Success $size <= $threshold"
@@ -271,6 +302,36 @@ jobs:
           exit 1
         fi
 
+  nxp-build-test:
+    name: nxp-build-test
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    permissions:
+      id-token: write
+      contents: read
+    with:
+      runner: linux.2xlarge
+      docker-image: executorch-ubuntu-22.04-arm-sdk
+      submodules: 'recursive'
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      timeout: 90
+      script: |
+        # The generic Linux job chooses to use base env, not the one setup by the image
+        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
+        conda activate "${CONDA_ENV}"
+
+        # Build
+        cmake -DEXECUTORCH_BUILD_NXP_NEUTRON=ON -Bcmake-out .
+        cmake --build cmake-out --target executorch_delegate_neutron --config Release
+
+        # Build check for the neutron backend library
+        lib_neutron="cmake-out/backends/nxp/libexecutorch_delegate_neutron.a"
+        if [ -f $lib_neutron ]; then
+            echo "Neutron backend library built."
+        else
+            echo "Neutron backend library not found!"
+            exit 1
+        fi
+
   test-coreml-delegate:
     name: test-coreml-delegate
     uses: pytorch/test-infra/.github/workflows/macos_job.yml@main
@@ -470,7 +531,7 @@ jobs:
       docker-image: executorch-ubuntu-22.04-qnn-sdk
       submodules: 'recursive'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
-      timeout: 90
+      timeout: 900
       script: |
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
@@ -489,14 +550,14 @@ jobs:
     strategy:
       matrix:
         dtype: [fp32]
-        model: [cvt, dit, efficientnet, focalnet, mobilevit_v1, mobilevit_v2, pvt, swin, albert, bert, distilbert, roberta] # eurobert requires transfomer >= 4.48.0, skip for now
+        model: [albert, bert, distilbert] # eurobert requires transfomer >= 4.48.0, skip for now
       fail-fast: false
     with:
       runner: linux.2xlarge
       docker-image: executorch-ubuntu-22.04-qnn-sdk
       submodules: 'recursive'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
-      timeout: 90
+      timeout: 900
       script: |
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")

CMakeLists.txt

@@ -449,6 +449,12 @@ install(
   FILES_MATCHING
   PATTERN "*.h"
 )
+install(
+  DIRECTORY runtime/executor/
+  DESTINATION include/executorch/runtime/executor
+  FILES_MATCHING
+  PATTERN "*.h"
+)
 install(
   DIRECTORY runtime/kernel/
   DESTINATION include/executorch/runtime/kernel
@@ -496,6 +502,10 @@ if(EXECUTORCH_BUILD_CADENCE)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/cadence)
 endif()
 
+if(EXECUTORCH_BUILD_NXP_NEUTRON)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/nxp)
+endif()
+
 if(EXECUTORCH_BUILD_COREML)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/apple/coreml)
 endif()
@@ -542,6 +552,12 @@ endif()
 
 if(EXECUTORCH_BUILD_EXTENSION_MODULE)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/module)
+  install(
+    DIRECTORY extension/module/
+    DESTINATION include/executorch/extension/module
+    FILES_MATCHING
+    PATTERN "*.h"
+  )
 endif()
 
 if(EXECUTORCH_BUILD_EXTENSION_LLM)

CMakePresets.json

@@ -104,6 +104,17 @@
                 "Windows"
             ]
         }
+    },
+    {
+        "name": "zephyr",
+        "displayName": "Build everything buildable on Zephyr RTOS",
+        "inherits": [
+            "common"
+        ],
+        "cacheVariables": {
+            "EXECUTORCH_BUILD_PRESET_FILE": "${sourceDir}/tools/cmake/preset/zephyr.cmake",
+            "CMAKE_TOOLCHAIN_FILE": "${sourceDir}/examples/zephyr/x86_64-linux-arm-zephyr-eabi-gcc.cmake"
+        }
     }
   ]
 }

backends/arm/_passes/__init__.py

@@ -41,6 +41,7 @@
 from .decompose_ne_pass import DecomposeNotEqualPass  # noqa
 from .decompose_round_pass import DecomposeRoundPass  # noqa
 from .decompose_select import DecomposeSelectPass  # noqa
+from .decompose_sign_pass import DecomposeSignPass  # noqa
 from .decompose_silu_pass import DecomposeSiluPass  # noqa
 from .decompose_sinh_pass import DecomposeSinhPass  # noqa
 from .decompose_softmax_pass import DecomposeSoftmaxPass  # noqa

backends/arm/_passes/arm_pass_manager.py

@@ -44,6 +44,7 @@
     DecomposeNotEqualPass,
     DecomposeRoundPass,
     DecomposeSelectPass,
+    DecomposeSignPass,
     DecomposeSiluPass,
     DecomposeSinhPass,
     DecomposeSoftmaxPass,
@@ -158,6 +159,7 @@ def _tosa_080_MI_pipeline(self, exported_program: ExportedProgram) -> GraphModul
         self.add_pass(ConvertIntPowToMuls())
         self.add_pass(CastBoolToInt8Pass())
         self.add_pass(DecomposeSinhPass())
+        self.add_pass(DecomposeSignPass())
         self.add_pass(ReplaceScalarWithTensorArgPassTOSAMI())
         self.add_pass(DecomposeEmbeddingPass())
         self.add_pass(FuseQuantizedActivationPass())
@@ -242,6 +244,7 @@ def transform_for_annotation_pipeline(self, graph_module: GraphModule):
         self.add_pass(DecomposeScaledDotProductAttention())
         self.add_pass(DecomposeRoundPass())
         self.add_pass(CastBoolToInt8Pass())
+        self.add_pass(DecomposeSignPass())
         self.add_pass(ReplaceScalarWithTensorArgPassTOSABI())
         self.add_pass(ScalarsToAttributePass())
         self.add_pass(DecomposeGroupNormPass())

backends/arm/_passes/decompose_linear_pass.py

@@ -88,6 +88,7 @@ def call(self, graph_module):
                     op_target=exir_ops.edge.aten.view_copy.default,
                     args=(conv, list(output_shape)),
                     kwargs={},
+                    from_node=node,
                 )
 
             node.replace_all_uses_with(output)