Merge remote-tracking branch 'origin/main' into contributing-md-merge

Author: kirklandsign

.ci/docker/build.sh

@@ -37,11 +37,15 @@ case "${IMAGE_NAME}" in
     ARM_SDK=yes
     CLANG_VERSION=12
     ;;
+  executorch-ubuntu-22.04-qnn-sdk)
+    QNN_SDK=yes
+    CLANG_VERSION=12
+    ;;
   executorch-ubuntu-22.04-clang12-android)
     LINTRUNNER=""
     CLANG_VERSION=12
     # From https://developer.android.com/ndk/downloads
-    ANDROID_NDK_VERSION=r26c
+    ANDROID_NDK_VERSION=r27b
     ;;
   *)
     echo "Invalid image name ${IMAGE_NAME}"
@@ -72,6 +76,7 @@ docker build \
   --build-arg "LINTRUNNER=${LINTRUNNER:-}" \
   --build-arg "BUILD_DOCS=${BUILD_DOCS}" \
   --build-arg "ARM_SDK=${ARM_SDK:-}" \
+  --build-arg "QNN_SDK=${QNN_SDK:-}" \
   --build-arg "ANDROID_NDK_VERSION=${ANDROID_NDK_VERSION:-}" \
   -f "${OS}"/Dockerfile \
   "$@" \

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-00e3eea170ce5db8ea9c62ce5e48f13886cd6d20
+19eff28ff3f19b50da46f5a9ff5f4d4d213806fe

.ci/docker/ci_commit_pins/torchao.txt

@@ -12,6 +12,26 @@ set -ex
 # shellcheck source=/dev/null
 source "$(dirname "${BASH_SOURCE[0]}")/utils.sh"
 
+install_ubuntu() {
+  echo "Preparing to build sccache from source"
+  apt-get update
+  # libssl-dev will not work as it is upgraded to libssl3 in Ubuntu-22.04.
+  # Instead use lib and headers from OpenSSL1.1 installed in `install_openssl.sh``
+  apt-get install -y cargo
+  echo "Checking out sccache repo"
+  git clone https://github.com/mozilla/sccache -b v0.8.2
+
+  cd sccache
+  echo "Building sccache"
+  cargo build --release
+  cp target/release/sccache /opt/cache/bin
+  echo "Cleaning up"
+  cd ..
+  rm -rf sccache
+  apt-get remove -y cargo rustc
+  apt-get autoclean && apt-get clean
+}
+
 install_binary() {
   echo "Downloading sccache binary from S3 repo"
   curl --retry 3 https://s3.amazonaws.com/ossci-linux/sccache -o /opt/cache/bin/sccache
@@ -22,15 +42,33 @@ mkdir -p /opt/cache/bin
 sed -e 's|PATH="\(.*\)"|PATH="/opt/cache/bin:\1"|g' -i /etc/environment
 export PATH="/opt/cache/bin:$PATH"
 
-# NB: Install the pre-built binary from S3 as building from source
-# https://github.com/pytorch/sccache has started failing mysteriously
-# in which sccache server couldn't start with the following error:
-#   sccache: error: Invalid argument (os error 22)
-install_binary
+install_ubuntu
 
 function write_sccache_stub() {
   BINARY=$1
-  printf "#!/bin/sh\nif [ \$(env -u LD_PRELOAD ps -p \$PPID -o comm=) != sccache ]; then\n  exec sccache %s \"\$@\"\nelse\n  exec %s \"\$@\"\nfi" "$(which "${BINARY}")" "$(which "${BINARY}")" > "/opt/cache/bin/${BINARY}"
+  if [ $1 == "gcc" ]; then
+    # Do not call sccache recursively when dumping preprocessor argument
+    # For some reason it's very important for the first cached nvcc invocation
+    cat >"/opt/cache/bin/$1" <<EOF
+#!/bin/sh
+if [ "\$1" = "-E" ] || [ "\$2" = "-E" ]; then
+  exec $(which $1) "\$@"
+elif [ \$(env -u LD_PRELOAD ps -p \$PPID -o comm=) != sccache ]; then
+  exec sccache $(which $1) "\$@"
+else
+  exec $(which $1) "\$@"
+fi
+EOF
+  else
+    cat >"/opt/cache/bin/$1" <<EOF
+#!/bin/sh
+if [ \$(env -u LD_PRELOAD ps -p \$PPID -o comm=) != sccache ]; then
+  exec sccache $(which $1) "\$@"
+else
+  exec $(which $1) "\$@"
+fi
+EOF
+  fi
   chmod a+x "/opt/cache/bin/${BINARY}"
 }
 
@@ -44,7 +82,7 @@ init_sccache() {
 
   # NB: This function is adopted from PyTorch core at
   # https://github.com/pytorch/pytorch/blob/main/.ci/pytorch/common-build.sh
-  as_ci_user sccache --stop-server > /dev/null 2>&1 || true
+  as_ci_user sccache --stop-server >/dev/null 2>&1 || true
   rm -f "${SCCACHE_ERROR_LOG}" || true
 
   # Clear sccache stats before using it

.ci/docker/common/install_cache.sh

@@ -13,7 +13,7 @@ install_ubuntu() {
   apt-get install -y --no-install-recommends clang-"$CLANG_VERSION"
   apt-get install -y --no-install-recommends llvm-"$CLANG_VERSION"
   # Also require LLD linker from llvm and libomp to build PyTorch from source
-  apt-get install -y lld "libomp-${CLANG_VERSION}-dev"
+  apt-get install -y lld "libomp-${CLANG_VERSION}-dev" "libc++-${CLANG_VERSION}-dev"
 
   # Use update-alternatives to make this version the default
   update-alternatives --install /usr/bin/clang clang /usr/bin/clang-"$CLANG_VERSION" 50

.ci/docker/common/install_clang.sh

@@ -57,6 +57,7 @@ COPY ./common/utils.sh utils.sh
 RUN bash ./install_cache.sh && rm install_cache.sh utils.sh
 ENV SCCACHE_BUCKET ossci-compiler-cache-circleci-v2
 ENV SCCACHE_S3_KEY_PREFIX executorch
+ENV SCCACHE_REGION us-east-1
 
 ARG TORCH_VERSION
 COPY ./common/install_pytorch.sh install_pytorch.sh
@@ -82,5 +83,7 @@ COPY --chown=ci-user:ci-user ./arm /opt/arm
 # Set up ARM SDK if needed
 RUN if [ -n "${ARM_SDK}" ]; then git config --global user.email "[email protected]"; git config --global user.name "OSS CI"; bash /opt/arm/setup.sh --i-agree-to-the-contained-eula /opt/arm-sdk; chown -R ci-user:ci-user /opt/arm-sdk; fi
 
+ARG QNN_SDK
+
 USER ci-user
 CMD ["bash"]

.ci/docker/ubuntu/Dockerfile

@@ -11,7 +11,7 @@ set -o xtrace
 build_qnn_backend() {
   echo "Start building qnn backend."
   export ANDROID_NDK_ROOT=/opt/ndk
-  export QNN_SDK_ROOT=/tmp/qnn/2.23.0.240531
+  export QNN_SDK_ROOT=/tmp/qnn/2.25.0.240728
   export EXECUTORCH_ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd)"
 
   bash backends/qualcomm/scripts/build.sh --skip_aarch64 --job_number 2 --release
@@ -27,7 +27,7 @@ set_up_aot() {
       -DCMAKE_INSTALL_PREFIX=$PWD \
       -DEXECUTORCH_BUILD_QNN=ON \
       -DQNN_SDK_ROOT=${QNN_SDK_ROOT} \
-      -DEXECUTORCH_BUILD_SDK=ON \
+      -DEXECUTORCH_BUILD_DEVTOOLS=ON \
       -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
       -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
       -DEXECUTORCH_ENABLE_EVENT_TRACER=ON \

.ci/scripts/build-qnn-sdk.sh

@@ -19,7 +19,6 @@ install_executorch_and_backend_lib() {
   cmake -DBUCK2="${BUCK2}" \
     -DCMAKE_TOOLCHAIN_FILE="${ANDROID_NDK}/build/cmake/android.toolchain.cmake" \
     -DANDROID_ABI="${ANDROID_ABI}" \
-    -DANDROID_PLATFORM=android-23 \
     -DCMAKE_INSTALL_PREFIX=cmake-android-out \
     -DCMAKE_BUILD_TYPE=Release \
     -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
@@ -41,16 +40,15 @@ build_llama_runner() {
     cmake -DBUCK2="${BUCK2}" \
     -DCMAKE_TOOLCHAIN_FILE="$ANDROID_NDK"/build/cmake/android.toolchain.cmake  \
     -DANDROID_ABI="${ANDROID_ABI}" \
-    -DANDROID_PLATFORM=android-23 \
     -DCMAKE_INSTALL_PREFIX=cmake-android-out \
     -DCMAKE_BUILD_TYPE=Release -DPYTHON_EXECUTABLE=python \
     -DEXECUTORCH_BUILD_XNNPACK=ON \
     -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
     -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
     -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
-    -Bcmake-android-out/examples/models/llama2 examples/models/llama2
+    -Bcmake-android-out/examples/models/llama examples/models/llama
 
-    cmake --build cmake-android-out/examples/models/llama2 -j4 --config Release
+    cmake --build cmake-android-out/examples/models/llama -j4 --config Release
 }
 install_flatc_from_source
 install_executorch_and_backend_lib

.ci/scripts/build_llama_android.sh

@@ -24,6 +24,8 @@
         "ic4": "linux.12xlarge",
         "resnet50": "linux.12xlarge",
         "llava": "linux.12xlarge",
+        "llama3_2_vision_encoder": "linux.12xlarge",
+        # "llama3_2_text_decoder": "linux.12xlarge",  # TODO: re-enable test when Huy's change is in / model gets smaller.
         # This one causes timeout on smaller runner, the root cause is unclear (T161064121)
         "dl3": "linux.12xlarge",
         "emformer_join": "linux.12xlarge",
@@ -88,8 +90,8 @@ def model_should_run_on_event(model: str, event: str) -> bool:
     if event == "pull_request":
         return model in ["mv3", "vit"]
     elif event == "push":
-        # 'emformer_predict' is running super slow. Only run it periodically
-        return model not in ["emformer_predict"]
+        # These are super slow. Only run it periodically
+        return model not in ["dl3", "edsr", "emformer_predict"]
     else:
         return True
 

.ci/scripts/gather_test_models.py

@@ -19,6 +19,8 @@ else
 fi
 
 # As Linux job is running inside a Docker container, all of its dependencies
-# have already been installed
-install_executorch
+# have already been installed, so we use PyTorch build from source here instead
+# of nightly. This allows CI to test against latest commits from PyTorch
+install_executorch "use-pt-pinned-commit"
 build_executorch_runner "${BUILD_TOOL}"
+do_not_use_nightly_on_ci