Support dynamically quantized 2D convolutions

.ci/docker/requirements-ci.txt

@@ -17,6 +17,7 @@ parameterized==0.9.0
 
 # Doc build requirements, same as https://github.com/pytorch/pytorch/blob/main/.ci/docker/requirements-docs.txt
 sphinx==5.3.0
+sphinx-reredirects==0.1.4
 sphinx-gallery==0.14.0
 breathe==4.34.0
 exhale==0.2.3

.ci/scripts/gather_benchmark_configs.py

@@ -23,6 +23,7 @@
     "samsung_galaxy_s22": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/e59f866a-30aa-4aa1-87b7-4510e5820dfa",
     "samsung_galaxy_s24": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/98f8788c-2e25-4a3c-8bb2-0d1e8897c0db",
     "google_pixel_8_pro": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/d65096ab-900b-4521-be8b-a3619b69236a",
+    "google_pixel_3_private_rooted": "arn:aws:devicefarm:us-west-2:308535385114:devicepool:02a2cf0f-6d9b-45ee-ba1a-a086587469e6/98d23ca8-ea9e-4fb7-b725-d402017b198d",
 }
 
 # Predefined benchmark configurations

.ci/scripts/test_ios_ci.sh

@@ -7,7 +7,7 @@
 
 set -e
 
-APP_PATH="examples/demo-apps/apple_ios/ExecuTorchDemo/ExecuTorchDemo"
+APP_PATH="executorch-examples/apple/ExecuTorchDemo/ExecuTorchDemo"
 MODEL_NAME="mv3"
 SIMULATOR_NAME="executorch"
 
@@ -34,6 +34,10 @@ say() {
   echo -e "\033[1m\n\t** $1 **\n\033[0m"
 }
 
+say "Cloning the Demo App"
+
+git clone --depth 1 https://github.com/pytorch-labs/executorch-examples.git
+
 say "Installing CoreML Backend Requirements"
 
 ./backends/apple/coreml/scripts/install_requirements.sh

.ci/scripts/test_llava.sh

@@ -154,7 +154,7 @@ run_and_verify() {
         EXPECTED_PREFIX="ASSISTANT: image captures a basketball game in progress, with several players on the court. One of the players is dribbling the ball, while the others are in various"
     else
         # set the expected prefix to be the same as prompt because there's a bug in sdpa_with_kv_cache that causes <unk> tokens.
-        EXPECTED_PREFIX="ASSISTANT:"
+        EXPECTED_PREFIX="ASSISTANT: image"
     fi
     if [[ "${RESULT}" == *"${EXPECTED_PREFIX}"* ]]; then
         echo "Expected result prefix: ${EXPECTED_PREFIX}"

.github/release.yml

@@ -15,57 +15,82 @@ changelog:
     - title: ARM
       labels:
         - "release notes: arm"
+        - "module: arm"
+        - "partner: arm"
     - title: NXP
-        labels:
+      labels:
         - "release notes: nxp"
+        - "module: nxp"
     - title: Exir
-        labels:
+      labels:
         - "release notes: exir"
+        - "module: exir"
     - title: Misc
-        labels:
+      labels:
         - "release notes: misc"
     - title: Apple
-        labels:
+      labels:
         - "release notes: apple"
+        - "module: coreml"
+        - "module: mps"
+    - title: Android
+      labels:
+        - "module: android"
+    - title: IOS
+      labels:
+        - "module: ios"
     - title: Build
-        labels:
+      labels:
         - "release notes: build"
     - title: Vulkan
-        labels:
+      labels:
         - "release notes: vulkan"
+        - "module: vulkan"
     - title: Cadence
-        labels:
+      labels:
         - "release notes: cadence"
+        - "module: cadence"
     - title: Runtime
-        labels:
+      labels:
         - "release notes: runtime"
+        - "module: runtime"
     - title: XNNPACK
-        labels:
+      labels:
         - "release notes: xnnpack"
+        - "module: xnnpack"
     - title: Devtools
-        labels:
+      labels:
         - "release notes: devtools"   
+        - "module: devtools"
     - title: Examples
-        labels:
+      labels:
         - "release notes: examples"
+    - title: LLM
+      labels:
+        - "module: llm"
     - title: Mediatek
-        labels:
+      labels:
         - "release notes: mediatek"
+        - "partner: mediatek"
     - title: Openvino
-        labels:
+      labels:
         - "release notes: openvino"
     - title: Qualcomm
-        labels:
+      labels:
         - "release notes: qualcomm"
+        - "partner: qualcomm"
+        - "module: qnn"
     - title: Training
-        labels:
+      labels:
         - "release notes: training"
+        - "module: training"
     - title: Quantization
-        labels:
+      labels:
         - "release notes: quantization" 
     - title: Ops & kernels
-        labels:
-        - "release notes: ops & kernels" 
+      labels:
+        - "release notes: ops & kernels"
+        - "module: kernels"
     - title: Other Changes
       labels:
         - "*"

.github/workflows/_android.yml

@@ -14,14 +14,18 @@ jobs:
     with:
       runner: linux.2xlarge
       docker-image: executorch-ubuntu-22.04-clang12-android
-      submodules: 'true'
+      submodules: 'recursive'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
       timeout: 90
       upload-artifact: android-apps
       upload-artifact-to-s3: true
       script: |
         set -eux
 
+        # Use sccache for NDK compiler as well
+        export CMAKE_CXX_COMPILER_LAUNCHER=sccache
+        export CMAKE_C_COMPILER_LAUNCHER=sccache
+
         # 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}"
@@ -36,8 +40,9 @@ jobs:
         cp ${BUILD_AAR_DIR}/executorch.aar $ARTIFACTS_DIR_NAME
 
         mkdir -p ${ARTIFACTS_DIR_NAME}/library_test_dir
-        bash .ci/scripts/build_android_instrumentation.sh
-        cp ${BUILD_AAR_DIR}/executorch_android/build/outputs/apk/androidTest/debug/executorch_android-debug-androidTest.apk "${ARTIFACTS_DIR_NAME}/library_test_dir"
+        bash extension/android/executorch_android/android_test_setup.sh
+        (cd extension/android; ANDROID_HOME="${ANDROID_SDK:-/opt/android/sdk}" ./gradlew :executorch_android:assembleAndroidTest)
+        cp extension/android/executorch_android/build/outputs/apk/androidTest/debug/executorch_android-debug-androidTest.apk "${ARTIFACTS_DIR_NAME}/library_test_dir"
 
         mkdir -p ${ARTIFACTS_DIR_NAME}/fp32-xnnpack-custom
         bash examples/models/llama/install_requirements.sh
@@ -130,7 +135,8 @@ jobs:
           # https://github.com/ReactiveCircus/android-emulator-runner. The max number
           # of cores we can set is 6, any higher number will be reduced to 6.
           cores: 6
-          ram-size: 12288M
+          ram-size: 16384M
+          heap-size: 12288M
           force-avd-creation: false
           disable-animations: true
           emulator-options: -no-snapshot-save -no-window -gpu swiftshader_indirect -noaudio -no-boot-anim -camera-back none

.github/workflows/android-perf-private-device-experiment.yml

@@ -0,0 +1,62 @@
+name: android-perf (private devices)
+
+on:
+  schedule:
+    - cron: 0 0,4,8,12,16,20 * * *
+  pull_request:
+    paths:
+      - .github/workflows/android-perf-private-device-experiment.yml
+  push:
+    branches:
+      - main
+    paths:
+      - .github/workflows/android-perf-private-device-experiment.yml
+  # Note: GitHub has an upper limit of 10 inputs
+  workflow_dispatch:
+    inputs:
+      models:
+        description: Models to be benchmarked
+        required: false
+        type: string
+        default: mv3,meta-llama/Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8,meta-llama/Llama-3.2-1B-Instruct-QLORA_INT4_EO8
+      devices:
+        description: Target devices to run benchmark
+        required: false
+        type: string
+        default: google_pixel_3_private_rooted
+      benchmark_configs:
+        description: The list of configs used the benchmark
+        required: false
+        type: string
+  workflow_call:
+    inputs:
+      models:
+        description: Models to be benchmarked
+        required: false
+        type: string
+        default: mv3,meta-llama/Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8,meta-llama/Llama-3.2-1B-Instruct-QLORA_INT4_EO8
+      devices:
+        description: Target devices to run benchmark
+        required: false
+        type: string
+        default: google_pixel_3_private_rooted
+      benchmark_configs:
+        description: The list of configs used the benchmark
+        required: false
+        type: string
+
+concurrency:
+  group: android-perf-private-devices-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}
+  cancel-in-progress: true
+
+jobs:
+  android:
+    uses: ./.github/workflows/android-perf.yml
+    secrets: inherit
+    permissions:
+      id-token: write
+      contents: read
+    with:
+      models: ${{ inputs.models || 'mv3,meta-llama/Llama-3.2-1B-Instruct-SpinQuant_INT4_EO8,meta-llama/Llama-3.2-1B-Instruct-QLORA_INT4_EO8' }}
+      devices: google_pixel_3_private_rooted
+      benchmark_configs: ${{ inputs.benchmark_configs }}

.github/workflows/android-perf.yml

@@ -345,14 +345,18 @@ jobs:
     with:
       runner: linux.2xlarge
       docker-image: executorch-ubuntu-22.04-clang12-android
-      submodules: 'true'
+      submodules: 'recursive'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
       timeout: 90
       upload-artifact: android-apps
       upload-artifact-to-s3: true
       script: |
         set -eux
 
+        # Use sccache for NDK compiler as well
+        export CMAKE_CXX_COMPILER_LAUNCHER=sccache
+        export CMAKE_C_COMPILER_LAUNCHER=sccache
+
         # 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}"
@@ -392,7 +396,7 @@ jobs:
       fail-fast: false
     with:
       # Due to scheduling a job may be pushed beyond the default 60m threshold
-      timeout: 120
+      timeout: 240
       device-type: android
       runner: linux.2xlarge
       test-infra-ref: ''