Update on "[Executorch] make slice_copy parallel"

When doing large prefills in LLMs, slice_copy takes about 5-10% time.
Mainly coming from slicing in the rope implementation.

Differential Revision: [D85532081](https://our.internmc.facebook.com/intern/diff/D85532081/)

[ghstack-poisoned]

Author: kimishpatel

.ci/docker/common/install_arm.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+# Copyright 2025 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.
+
+set -ex
+
+install_arm_prerequiresites() {
+    apt-get update -y
+    apt-get install -y --no-install-recommends \
+            mesa-vulkan-drivers libvulkan1
+    rm -rf /var/lib/apt/lists/*
+}
+
+install_arm_prerequiresites

.ci/docker/ubuntu/Dockerfile

@@ -83,6 +83,9 @@ RUN if [ -n "${ANDROID_NDK_VERSION}" ]; then bash ./install_android.sh; fi
 RUN rm install_android.sh
 
 ARG ARM_SDK
+COPY ./common/install_arm.sh install_arm.sh
+RUN if [ -n "${ARM_SDK}" ]; then bash ./install_arm.sh; fi
+RUN rm install_arm.sh
 
 ARG ZEPHYR_SDK
 COPY ./common/install_zephyr.sh install_zephyr.sh

.ci/scripts/test_backend.sh

@@ -57,15 +57,25 @@ if [[ "$FLOW" == *vulkan* ]]; then
 fi
 
 if [[ "$FLOW" == *arm* ]]; then
+
     # Setup ARM deps.
-    .ci/scripts/setup-arm-baremetal-tools.sh
+    if [[ "$FLOW" == *vgf* ]]; then
+        .ci/scripts/setup-arm-baremetal-tools.sh --enable-mlsdk-deps --install-mlsdk-deps-with-pip
+    else
+        .ci/scripts/setup-arm-baremetal-tools.sh
+    fi
     source examples/arm/ethos-u-scratch/setup_path.sh
 
     if [[ "$FLOW" == *ethos_u* ]]; then
         # Prepare a test runner binary that can run on the Corstone-3x0 FVPs
         backends/arm/scripts/build_executorch.sh
         backends/arm/test/setup_testing.sh
     fi
+
+    if [[ "$FLOW" == *vgf* ]]; then
+        # Prepare a test runner binary for VKML runtime
+        backends/arm/test/setup_testing_vkml.sh
+    fi
 fi
 
 if [[ $IS_MACOS -eq 1 ]]; then

.ci/scripts/test_llama.sh

@@ -171,15 +171,14 @@ cmake_build_llama_runner() {
     git submodule update --init
     popd
     dir="examples/models/llama"
-    retry cmake \
-        -DEXECUTORCH_BUILD_TESTS=ON \
-        -DBUILD_TESTING=OFF \
-        -DCMAKE_INSTALL_PREFIX=cmake-out \
-        -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" \
-        -Bcmake-out/${dir} \
-        ${dir}
-    cmake --build cmake-out/${dir} -j9 --config "$CMAKE_BUILD_TYPE"
-
+    if [[ "$CMAKE_BUILD_TYPE" == "Debug" ]]; then
+        PRESET="llama-debug"
+    else
+        PRESET="llama-release"
+    fi
+    pushd "${dir}"
+    cmake --workflow --preset "${PRESET}"
+    popd
 }
 
 cleanup_files() {

.ci/scripts/test_llama_lora.sh

@@ -12,10 +12,7 @@ source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
 cmake_install_executorch_libraries() {
     echo "Installing libexecutorch.a, libextension_module.so, libportable_ops_lib.a"
     rm -rf cmake-out
-    retry cmake --preset llm \
-        -DCMAKE_INSTALL_PREFIX=cmake-out \
-        -DCMAKE_BUILD_TYPE=Release
-    cmake --build cmake-out -j9 --target install --config Release
+    cmake --workflow llm-release
 }
 
 cmake_build_llama_runner() {

.ci/scripts/test_model_e2e.sh

@@ -156,27 +156,13 @@ echo "::endgroup::"
 
 echo "::group::Build $MODEL_NAME Runner"
 
-if [ "$DEVICE" = "cuda" ]; then
-  BUILD_BACKEND="EXECUTORCH_BUILD_CUDA"
-elif [ "$DEVICE" = "metal" ]; then
-  BUILD_BACKEND="EXECUTORCH_BUILD_METAL"
-else
+if [ "$DEVICE" != "cuda" ] && [ "$DEVICE" != "metal" ]; then
   echo "Error: Unsupported device '$DEVICE'. Must be 'cuda' or 'metal'."
   exit 1
 fi
 
-cmake --preset llm \
-      -D${BUILD_BACKEND}=ON \
-      -DCMAKE_INSTALL_PREFIX=cmake-out \
-      -DCMAKE_BUILD_TYPE=Release \
-      -Bcmake-out -S.
-cmake --build cmake-out -j$(nproc) --target install --config Release
-
-cmake -D${BUILD_BACKEND}=ON \
-      -DCMAKE_BUILD_TYPE=Release \
-      -Sexamples/models/$RUNNER_PATH \
-      -Bcmake-out/examples/models/$RUNNER_PATH/
-cmake --build cmake-out/examples/models/$RUNNER_PATH --target $RUNNER_TARGET --config Release
+MAKE_TARGET="${RUNNER_PATH}-${DEVICE}"
+make "${MAKE_TARGET}"
 echo "::endgroup::"
 
 echo "::group::Run $MODEL_NAME Runner"
@@ -195,7 +181,7 @@ case "$MODEL_NAME" in
     RUNNER_ARGS="$RUNNER_ARGS --tokenizer_path ${MODEL_DIR}/$TOKENIZER_FILE --audio_path ${MODEL_DIR}/$AUDIO_FILE --processor_path ${MODEL_DIR}/$PREPROCESSOR"
     ;;
   whisper-*)
-    RUNNER_ARGS="$RUNNER_ARGS --tokenizer_path ${MODEL_DIR}/ --audio_path ${MODEL_DIR}/$AUDIO_FILE --processor_path ${MODEL_DIR}/$PREPROCESSOR --model_name ${MODEL_NAME}"
+    RUNNER_ARGS="$RUNNER_ARGS --tokenizer_path ${MODEL_DIR}/ --audio_path ${MODEL_DIR}/$AUDIO_FILE --processor_path ${MODEL_DIR}/$PREPROCESSOR"
     ;;
   gemma3)
     RUNNER_ARGS="$RUNNER_ARGS --tokenizer_path ${MODEL_DIR}/ --image_path $IMAGE_PATH"

.ci/scripts/test_phi_3_mini.sh

@@ -23,8 +23,16 @@ if hash nproc &> /dev/null; then NPROC=$(nproc); fi
 
 cmake_install_executorch_libraries() {
   rm -rf cmake-out
-  cmake --preset llm -DCMAKE_INSTALL_PREFIX=cmake-out -DCMAKE_BUILD_TYPE=${BUILD_TYPE}
-  cmake --build cmake-out -j16 --target install --config ${BUILD_TYPE}
+
+  # Select workflow preset based on BUILD_TYPE
+  if [[ "${BUILD_TYPE}" == "Debug" ]]; then
+    WORKFLOW_PRESET="llm-debug"
+  else
+    WORKFLOW_PRESET="llm-release"
+  fi
+
+  echo "Using workflow preset: ${WORKFLOW_PRESET}"
+  cmake --workflow --preset ${WORKFLOW_PRESET}
 }
 
 cmake_build_phi_3_mini() {

.ci/scripts/utils.sh

@@ -84,8 +84,8 @@ dedupe_macos_loader_path_rpaths() {
 
 install_domains() {
   echo "Install torchvision and torchaudio"
-  pip install --no-use-pep517 --user "git+https://github.com/pytorch/audio.git@${TORCHAUDIO_VERSION}"
-  pip install --no-use-pep517 --user "git+https://github.com/pytorch/vision.git@${TORCHVISION_VERSION}"
+  pip install --no-build-isolation --user "git+https://github.com/pytorch/audio.git@${TORCHAUDIO_VERSION}"
+  pip install --no-build-isolation --user "git+https://github.com/pytorch/vision.git@${TORCHVISION_VERSION}"
 }
 
 install_pytorch_and_domains() {

.github/workflows/cuda.yml

@@ -71,7 +71,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        model: [linear, add, add_mul, resnet18, conv1d]
+        model: [linear, add, add_mul, resnet18, conv1d, sdpa]
     with:
       timeout: 90
       runner: linux.g5.4xlarge.nvidia.gpu

.github/workflows/pull.yml

@@ -340,6 +340,7 @@ jobs:
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
       timeout: 90
       script: |
+        set -eux
         # The generic Linux job chooses to use base env, not the one setup by the image
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"