Move benchmark apps to extension/benchmark dir (#5951)

Summary: Pull Request resolved: #5951

Reviewed By: huydhn

Differential Revision: D64012575

Pulled By: guangy10

fbshipit-source-id: 27bb25586af2aafb8df6c0623d502482b4fa238b
(cherry picked from commit ac2ae07)

Author: Guang Yang

.github/workflows/android-perf.yml

@@ -232,3 +232,78 @@ jobs:
       test-spec: ${{ inputs.test_spec || 'https://ossci-android.s3.amazonaws.com/executorch/android-llm-device-farm-test-spec.yml' }}
       # Uploaded to S3 from the previous job
       extra-data: https://gha-artifacts.s3.amazonaws.com/${{ github.repository }}/${{ github.run_id }}/artifacts/${{ matrix.model }}_${{ matrix.delegate }}/model.zip
+
+  upload-benchmark-results:
+    needs:
+      - benchmark-on-device
+    if: always()
+    runs-on: linux.2xlarge
+    environment: upload-benchmark-results
+    permissions:
+      id-token: write
+      contents: read
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          submodules: false
+
+      - name: Authenticate with AWS
+        uses: aws-actions/configure-aws-credentials@v4
+        with:
+          role-to-assume: arn:aws:iam::308535385114:role/gha_workflow_upload-benchmark-results
+          # The max duration enforced by the server side
+          role-duration-seconds: 18000
+          aws-region: us-east-1
+
+      - name: Setup conda
+        uses: pytorch/test-infra/.github/actions/setup-miniconda@main
+        with:
+          python-version: '3.10'
+
+      - name: Download the list of artifacts from S3
+        env:
+          ARTIFACTS_S3_DIR: s3://gha-artifacts/device_farm/${{ github.run_id }}/${{ github.run_attempt }}/artifacts/
+        shell: bash
+        run: |
+          set -eux
+          ${CONDA_RUN} python -mpip install awscli==1.32.18
+
+          mkdir -p artifacts
+          pushd artifacts
+          ${CONDA_RUN} aws s3 sync "${ARTIFACTS_S3_DIR}" .
+          popd
+
+          ls -lah artifacts
+
+      - name: Extract the benchmark results JSON
+        shell: bash
+        run: |
+          set -eux
+
+          mkdir -p benchmark-results
+
+          for ARTIFACTS_BY_JOB in artifacts/*.json; do
+            [ -f "${ARTIFACTS_BY_JOB}" ] || break
+            echo "${ARTIFACTS_BY_JOB}"
+            ${CONDA_RUN} python .github/scripts/extract_benchmark_results.py \
+              --artifacts "${ARTIFACTS_BY_JOB}" \
+              --output-dir benchmark-results \
+              --repo ${{ github.repository }} \
+              --head-branch ${{ github.head_ref || github.ref_name }} \
+              --workflow-name "${{ github.workflow }}" \
+              --workflow-run-id ${{ github.run_id }} \
+              --workflow-run-attempt ${{ github.run_attempt }}
+          done
+
+          ls -lah benchmark-results
+
+          for BENCHMARK_RESULTS in benchmark-results/*.json; do
+            cat "${BENCHMARK_RESULTS}"
+            echo
+          done
+
+      - name: Upload the benchmark results
+        uses: pytorch/test-infra/.github/actions/upload-benchmark-results@main
+        with:
+          benchmark-results-dir: 'benchmark-results'
+          dry-run: false

.github/workflows/android.yml

@@ -15,6 +15,7 @@ on:
       - install_requirements.sh
       - examples/demo-apps/android/**
       - extension/android/**
+      - extension/benchmark/android/**
       - extension/module/**
   workflow_dispatch:
 

.github/workflows/apple-perf.yml

@@ -235,17 +235,17 @@ jobs:
         PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
           build/build_apple_frameworks.sh --coreml --custom --mps --optimized --portable --quantized --xnnpack
 
-        mkdir -p extension/apple/Benchmark/Frameworks
+        mkdir -p extension/benchmark/apple/Benchmark/Frameworks
         for FRAMEWORK in "${FRAMEWORKS[@]}"; do (
-          cp -r "cmake-out/${FRAMEWORK}.xcframework" extension/apple/Benchmark/Frameworks/
+          cp -r "cmake-out/${FRAMEWORK}.xcframework" extension/benchmark/apple/Benchmark/Frameworks/
         ) done
         echo "::endgroup::"
 
         # NB: Although exported models can be copied to this directory and bundled together with the
         # app, we don't use this in CI and rely on AWS extra data parameter to make the model and the
         # tokenizer available to the benchmark. This decouples the app and the model. We just need to
         # create the directory here to pass the build
-        mkdir -p extension/apple/Benchmark/Models
+        mkdir -p extension/benchmark/apple/Benchmark/Models
         ${CONDA_RUN} --no-capture-output \
           build/build_apple_llm_demo.sh ${ARTIFACTS_DIR_NAME}
 

.github/workflows/apple.yml

@@ -17,6 +17,7 @@ on:
       - build/test_ios_ci.sh
       - examples/demo-apps/apple_ios/**
       - extension/apple/**
+      - extension/benchmark/apple/**
       - extension/module/**
   workflow_dispatch:
 
@@ -215,3 +216,70 @@ jobs:
             shasum -a 256 "${FILENAME}"
             ${AWS_CMD} "${FILENAME}" s3://ossci-ios/executorch/ --acl public-read
           done
+
+  build-benchmark-app:
+    name: build-benchmark-app
+    uses: pytorch/test-infra/.github/workflows/macos_job.yml@main
+    secrets: inherit
+    with:
+      runner: macos-latest-xlarge
+      python-version: '3.11'
+      submodules: 'true'
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      upload-artifact: ios-apps
+      secrets-env: BUILD_CERTIFICATE_BASE64 EXECUTORCH_BENCHMARK_BUILD_PROVISION_PROFILE_BASE64 KEYCHAIN_PASSWORD
+      timeout: 90
+      script: |
+        set -eux
+
+        echo "::group::Setting up CI environment"
+        .ci/scripts/setup-conda.sh
+
+        BUILD_TOOL=cmake
+        # Setup MacOS dependencies as there is no Docker support on MacOS atm
+        GITHUB_RUNNER=1 PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
+        .ci/scripts/setup-macos.sh "${BUILD_TOOL}"
+        export ARTIFACTS_DIR_NAME=artifacts-to-be-uploaded
+
+        # Setup Apple certificate for iOS development
+        BUILD_PROVISION_PROFILE_BASE64="${SECRET_EXECUTORCH_BENCHMARK_BUILD_PROVISION_PROFILE_BASE64}" \
+        BUILD_CERTIFICATE_BASE64="${SECRET_BUILD_CERTIFICATE_BASE64}" \
+        KEYCHAIN_PASSWORD="${SECRET_KEYCHAIN_PASSWORD}" \
+        .ci/scripts/setup-ios.sh
+
+        # Install CoreML Backend Requirements
+        PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
+          backends/apple/coreml/scripts/install_requirements.sh
+
+        # Install MPS Backend Requirements
+        PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
+          backends/apple/mps/install_requirements.sh
+        echo "::endgroup::"
+
+        echo "::group::Build ExecuTorch iOS frameworks"
+        FRAMEWORKS=(
+          "executorch"
+          "backend_coreml"
+          "backend_mps"
+          "backend_xnnpack"
+          "kernels_custom"
+          "kernels_optimized"
+          "kernels_portable"
+          "kernels_quantized"
+        )
+
+        # Build Release iOS Frameworks
+        PYTHON_EXECUTABLE=python ${CONDA_RUN} --no-capture-output \
+          build/build_apple_frameworks.sh --coreml --custom --mps --optimized --portable --quantized --xnnpack
+
+        mkdir -p extension/benchmark/apple/Benchmark/Frameworks
+        for FRAMEWORK in "${FRAMEWORKS[@]}"; do (
+          cp -r "cmake-out/${FRAMEWORK}.xcframework" extension/benchmark/apple/Benchmark/Frameworks/
+        ) done
+        echo "::endgroup::"
+
+        echo "::group::Build ExecuTorch benchmark app"
+        mkdir -p extension/benchmark/apple/Benchmark/Models
+        ${CONDA_RUN} --no-capture-output \
+          build/build_apple_llm_demo.sh ${ARTIFACTS_DIR_NAME}
+        echo "::endgroup::"

.github/workflows/lint.yml

@@ -66,7 +66,7 @@ jobs:
         FILES_NEEDS_FORMAT=$(/opt/google-java-format -n extension/android/src/main/java/org/pytorch/executorch/*.java \
           examples/demo-apps/android/ExecuTorchDemo/app/src/main/java/com/example/executorchdemo/*.java \
           examples/demo-apps/android/LlamaDemo/app/src/main/java/com/example/executorchllamademo/*.java \
-          extension/android/benchmark/app/src/main/java/org/pytorch/minibench/*.java)
+          extension/benchmark/android/benchmark/app/src/main/java/org/pytorch/minibench/*.java)
         if [ -n "$FILES_NEEDS_FORMAT" ]; then
           echo "Warning: The following files need formatting. Please use google-java-format."
           echo "Use a binary from https://github.com/google/google-java-format/releases/"

.github/workflows/upload-android-test-specs.yml

@@ -4,13 +4,13 @@ on:
   pull_request:
     paths:
       - .github/workflows/upload-android-test-specs.yml
-      - extension/android/benchmark/android-llm-device-farm-test-spec.yml
+      - extension/benchmark/android/benchmark/android-llm-device-farm-test-spec.yml
   push:
     branches:
       - main
     paths:
       - .github/workflows/upload-android-test-specs.yml
-      - extension/android/benchmark/android-llm-device-farm-test-spec.yml
+      - extension/benchmark/android/benchmark/android-llm-device-farm-test-spec.yml
 
 concurrency:
   # NB: This concurency group needs to be different than the one used in android-perf, otherwise
@@ -32,7 +32,7 @@ jobs:
             ${{ github.repository }}/${{ github.run_id }}/artifacts
           retention-days: 1
           if-no-files-found: error
-          path: extension/android/benchmark/android-llm-device-farm-test-spec.yml
+          path: extension/benchmark/android/benchmark/android-llm-device-farm-test-spec.yml
 
   validate-android-test-spec:
     needs: upload-android-test-spec-for-validation
@@ -77,7 +77,7 @@ jobs:
 
       - name: Upload the spec to S3 ossci-android bucket
         shell: bash
-        working-directory: extension/android/benchmark/
+        working-directory: extension/benchmark/android/benchmark/
         env:
           SPEC_FILE: android-llm-device-farm-test-spec.yml
         run: |

.github/workflows/upload-apple-test-specs.yml

@@ -4,13 +4,13 @@ on:
   pull_request:
     paths:
       - .github/workflows/upload-apple-test-specs.yml
-      - examples/demo-apps/apple_ios/default-ios-device-farm-appium-test-spec.yml
+      - extension/benchmark/apple/Benchmark/default-ios-device-farm-appium-test-spec.yml
   push:
     branches:
       - main
     paths:
       - .github/workflows/upload-apple-test-specs.yml
-      - examples/demo-apps/apple_ios/default-ios-device-farm-appium-test-spec.yml
+      - extension/benchmark/apple/Benchmark/default-ios-device-farm-appium-test-spec.yml
 
 concurrency:
   # NB: This concurency group needs to be different than the one used in apple-perf, otherwise
@@ -32,7 +32,7 @@ jobs:
             ${{ github.repository }}/${{ github.run_id }}/artifacts
           retention-days: 1
           if-no-files-found: error
-          path: examples/demo-apps/apple_ios/default-ios-device-farm-appium-test-spec.yml
+          path: extension/benchmark/apple/Benchmark/default-ios-device-farm-appium-test-spec.yml
 
   validate-apple-test-spec:
     needs: upload-apple-test-spec-for-validation
@@ -78,7 +78,7 @@ jobs:
 
       - name: Upload the spec to S3 ossci-ios bucket
         shell: bash
-        working-directory: examples/demo-apps/apple_ios
+        working-directory: extension/benchmark/apple/Benchmark/
         env:
           SPEC_FILE: default-ios-device-farm-appium-test-spec.yml
         run: |

build/build_android_llm_demo.sh

@@ -108,9 +108,9 @@ build_android_demo_apps() {
   ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew build assembleAndroidTest
   popd
 
-  mkdir -p extension/android/benchmark/app/libs
-  cp ${BUILD_AAR_DIR}/executorch.aar extension/android/benchmark/app/libs
-  pushd extension/android/benchmark
+  mkdir -p extension/benchmark/android/benchmark/app/libs
+  cp ${BUILD_AAR_DIR}/executorch.aar extension/benchmark/android/benchmark/app/libs
+  pushd extension/benchmark/android/benchmark
   ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew build assembleAndroidTest
   popd
 }
@@ -135,8 +135,8 @@ collect_artifacts_to_be_uploaded() {
   # Collect MiniBench APK
   MINIBENCH_APP_DIR="${ARTIFACTS_DIR_NAME}/minibench"
   mkdir -p "${MINIBENCH_APP_DIR}"
-  cp extension/android/benchmark/app/build/outputs/apk/debug/*.apk "${MINIBENCH_APP_DIR}"
-  cp extension/android/benchmark/app/build/outputs/apk/androidTest/debug/*.apk "${MINIBENCH_APP_DIR}"
+  cp extension/benchmark/android/benchmark/app/build/outputs/apk/debug/*.apk "${MINIBENCH_APP_DIR}"
+  cp extension/benchmark/android/benchmark/app/build/outputs/apk/androidTest/debug/*.apk "${MINIBENCH_APP_DIR}"
 }
 
 BUILD_AAR_DIR="$(mktemp -d)"

build/build_apple_llm_demo.sh

@@ -8,7 +8,7 @@
 set -euo pipefail
 
 ARTIFACTS_DIR_NAME="$1"
-APP_PATH="extension/apple/Benchmark/Benchmark"
+APP_PATH="extension/benchmark/apple/Benchmark/Benchmark"
 
 xcodebuild build-for-testing \
   -project "${APP_PATH}.xcodeproj" \