Update base for Update on "[ET-VK] Adding repeat support to add_copy_packed_dim_offset_node function."

This diff adds support for the repeat operation in `add_copy_packed_dim_offset_node` function in the Vulkan backend for Executorch. The function now takes an additional boolean parameter, "repeat", which indicates whether the copy should wrap around the tensor dimension.

`copy_packed_dim_offset` shader now has 2 functions `repeat_copy` and `no_repeat_copy` which are chosen based on specialization constant parameter.
`no_repeat_copy` function has the legacy copy code.
`repeat_copy` function reads input tensor's dim based on output pos and wraps it according to WHCB repetitions.

Push constants `src_offset` and `dst_offset` contains source and destination's WHCB dimensions (and not copy offsets) respectively, when calling repeat function.

Differential Revision: [D71477552](https://our.internmc.facebook.com/intern/diff/D71477552/)

[ghstack-poisoned]

Author: trivedivivek

.ci/scripts/gather_test_models.py

@@ -14,7 +14,7 @@
 from typing import Any
 
 from examples.models import MODEL_NAME_TO_MODEL
-from examples.xnnpack import MODEL_NAME_TO_OPTIONS
+from examples.xnnpack import MODEL_NAME_TO_OPTIONS, QuantType
 
 DEFAULT_RUNNERS = {
     "linux": "linux.2xlarge",
@@ -154,7 +154,7 @@ def export_models_for_ci() -> dict[str, dict]:
         if backend == "xnnpack":
             if name not in MODEL_NAME_TO_OPTIONS:
                 continue
-            if MODEL_NAME_TO_OPTIONS[name].quantization:
+            if MODEL_NAME_TO_OPTIONS[name].quantization != QuantType.NONE:
                 backend += "-quantization"
 
             if MODEL_NAME_TO_OPTIONS[name].delegation:

.ci/scripts/test_model.sh

@@ -224,19 +224,22 @@ test_model_with_coreml() {
 
   "${PYTHON_EXECUTABLE}" -m examples.apple.coreml.scripts.export --model_name="${MODEL_NAME}" --compute_precision "${DTYPE}"
   EXPORTED_MODEL=$(find "." -type f -name "${MODEL_NAME}*.pte" -print -quit)
-  # TODO:
+
   if [ -n "$EXPORTED_MODEL" ]; then
     EXPORTED_MODEL_WITH_DTYPE="${EXPORTED_MODEL%.pte}_${DTYPE}.pte"
     mv "$EXPORTED_MODEL" "$EXPORTED_MODEL_WITH_DTYPE"
     EXPORTED_MODEL="$EXPORTED_MODEL_WITH_DTYPE"
-    echo "Renamed file path: $EXPORTED_MODEL"
+    echo "OK exported model: $EXPORTED_MODEL"
   else
-    echo "No .pte file found"
+    echo "[error] failed to export model: no .pte file found"
     exit 1
   fi
 
   # Run the model
   if [ "${should_test}" = true ]; then
+    echo "Installing requirements needed to build coreml_executor_runner..."
+    backends/apple/coreml/scripts/install_requirements.sh
+
     echo "Testing exported model with coreml_executor_runner..."
     local out_dir=$(mktemp -d)
     COREML_EXECUTOR_RUNNER_OUT_DIR="${out_dir}" examples/apple/coreml/scripts/build_executor_runner.sh

.ci/scripts/wheel/test_macos.py

@@ -15,11 +15,9 @@
                 model=Model.Mv3,
                 backend=Backend.XnnpackQuantizationDelegation,
             ),
-            # Enable this once CoreML is suppported out-of-the-box
-            # https://github.com/pytorch/executorch/issues/9019
-            # test_base.ModelTest(
-            #     model=Model.Mv3,
-            #     backend=Backend.CoreMlTest,
-            # )
+            test_base.ModelTest(
+                model=Model.Mv3,
+                backend=Backend.CoreMlTest,
+            ),
         ]
     )

.github/workflows/build-wheels-linux.yml

@@ -6,6 +6,9 @@ on:
     paths:
       - .ci/**/*
       - .github/workflows/build-wheels-linux.yml
+      - examples/**/*
+      - pyproject.toml
+      - setup.py
   push:
     branches:
       - nightly

.github/workflows/build-wheels-macos.yml

@@ -6,6 +6,9 @@ on:
     paths:
       - .ci/**/*
       - .github/workflows/build-wheels-macos.yml
+      - examples/**/*
+      - pyproject.toml
+      - setup.py
   push:
     branches:
       - nightly
@@ -57,6 +60,8 @@ jobs:
       pre-script: ${{ matrix.pre-script }}
       post-script: ${{ matrix.post-script }}
       package-name: ${{ matrix.package-name }}
-      runner-type: macos-m1-stable
+      # Meta's macOS runners do not have Xcode, so use GitHub's runners.
+      runner-type: macos-latest-xlarge
+      setup-miniconda: true
       smoke-test-script: ${{ matrix.smoke-test-script }}
       trigger-event: ${{ github.event_name }}

.github/workflows/trunk.yml

@@ -65,22 +65,29 @@ jobs:
       matrix:
         model: [linear, add, add_mul, ic3, ic4, mv2, mv3, resnet18, resnet50, vit, w2l, mobilebert, emformer_join, emformer_transcribe]
         backend: [portable, xnnpack-quantization-delegation]
+        runner: [linux.arm64.2xlarge]
         include:
           - model: lstm
             backend: portable
+            runner: linux.arm64.2xlarge
           - model: mul
             backend: portable
+            runner: linux.arm64.2xlarge
           - model: softmax
             backend: portable
+            runner: linux.arm64.2xlarge
           - model: phi_4_mini
             backend: portable
+            runner: linux.arm64.m7g.4xlarge
           - model: qwen2_5
             backend: portable
+            runner: linux.arm64.2xlarge
           - model: llama3_2_vision_encoder
             backend: portable
+            runner: linux.arm64.2xlarge
       fail-fast: false
     with:
-      runner: linux.arm64.2xlarge
+      runner: ${{ matrix.runner }}
       docker-image: executorch-ubuntu-22.04-gcc11-aarch64
       submodules: 'true'
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
@@ -536,7 +543,7 @@ jobs:
         git clone https://github.com/huggingface/optimum-executorch
         cd optimum-executorch
         # There is no release yet, for CI stability, always test from the same commit on main
-        git checkout 1907349524b5f2d61f9c04d2e985da826d4308ba
+        git checkout 577a2b19670e4c643a5c6ecb09bf47b9a699e7c6
         pip install .[tests]
         pip list
         echo "::endgroup::"

backends/apple/coreml/compiler/coreml_preprocess.py

@@ -16,8 +16,8 @@
 
 import coremltools as ct
 import coremltools.optimize as cto
-import executorchcoreml
 
+from executorch.backends.apple.coreml import executorchcoreml
 from executorch.exir.backend.backend_details import (
     BackendDetails,
     ExportedProgram,

backends/apple/coreml/runtime/inmemoryfs/setup.py

@@ -10,6 +10,10 @@ SCRIPT_DIR_PATH="$(
     pwd -P
 )"
 
+# TODO(jathu): remove the need to fetch coremltools to build deps for coreml_executor_runner.
+# Keep this version in sync with: pyproject.toml
+COREMLTOOLS_VERSION="8.1"
+
 red=`tput setaf 1`
 green=`tput setaf 2`
 
@@ -24,7 +28,7 @@ rm -rf "$COREML_DIR_PATH/third-party"
 mkdir "$COREML_DIR_PATH/third-party"
 
 echo "${green}ExecuTorch: Cloning coremltools."
-git clone --depth 1 --branch 8.1 "https://github.com/apple/coremltools.git" $COREMLTOOLS_DIR_PATH
+git clone --depth 1 --branch "${COREMLTOOLS_VERSION}" "https://github.com/apple/coremltools.git" $COREMLTOOLS_DIR_PATH
 cd $COREMLTOOLS_DIR_PATH
 
 STATUS=$?
@@ -43,16 +47,7 @@ fi
 
 mkdir "$COREMLTOOLS_DIR_PATH/build"
 cmake -S "$COREMLTOOLS_DIR_PATH" -B "$COREMLTOOLS_DIR_PATH/build"
-cmake --build "$COREMLTOOLS_DIR_PATH/build" --parallel
-
-echo "${green}ExecuTorch: Installing coremltools."
-pip install "$COREMLTOOLS_DIR_PATH"
-
-STATUS=$?
-if [ $STATUS -ne 0 ]; then
-    echo "${red}ExecuTorch: Failed to install coremltools."
-    exit 1
-fi
+cmake --build "$COREMLTOOLS_DIR_PATH/build" --parallel --target mlmodel
 
 echo "${green}ExecuTorch: Cloning nlohmann."
 git clone https://github.com/nlohmann/json.git "$COREML_DIR_PATH/third-party/nlohmann_json"
@@ -62,8 +57,6 @@ if [ $STATUS -ne 0 ]; then
     exit 1
 fi
 
-sh "$COREML_DIR_PATH/scripts/install_inmemoryfs.sh"
-
 echo "${green}ExecuTorch: Copying protobuf files."
 mkdir -p "$COREML_DIR_PATH/runtime/sdk/format/"
 cp -rf "$PROTOBUF_FILES_DIR_PATH" "$COREML_DIR_PATH/runtime/sdk/format/"