Add stories ci for qnn

.ci/docker/build.sh

@@ -37,6 +37,10 @@ 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
@@ -59,6 +63,9 @@ cp ../../requirements-lintrunner.txt ./
 # with a new image hash when the content here is updated
 cp -r ../../examples/arm/ ./arm
 
+# Copy qnn setup script from root to here
+cp -r ../../backends/qualcomm/ ./qualcomm
+
 docker build \
   --no-cache \
   --progress=plain \
@@ -72,6 +79,7 @@ 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/ubuntu/Dockerfile

@@ -82,5 +82,10 @@ 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 "ossci@example.com"; 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
+COPY --chown=ci-user:ci-user ./qualcomm /opt/qualcomm
+# Set up QNN SDK if needed
+RUN if [ -n "${QNN_SDK}" ]; then git config --global user.email "ossci@example.com"; git config --global user.name "OSS CI"; fi
+
 USER ci-user
 CMD ["bash"]

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

@@ -0,0 +1,19 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# 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
+
+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 EXECUTORCH_ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/.." && pwd)"
+
+  bash backends/qualcomm/scripts/build.sh --skip_aarch64 --job_number 2 --release
+}
+
+build_qnn_backend

.ci/scripts/setup-qnn-deps.sh

@@ -0,0 +1,29 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# 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_qnn() {
+  echo "Start installing qnn."
+  QNN_INSTALLATION_DIR=/tmp/qnn
+  mkdir -p "${QNN_INSTALLATION_DIR}"
+
+  curl -Lo /tmp/v2.23.0.24.06.24.zip "https://softwarecenter.qualcomm.com/api/download/software/qualcomm_neural_processing_sdk/v2.23.0.24.06.24.zip"
+  echo "Finishing downloading qnn sdk."
+  unzip -qo /tmp/v2.23.0.24.06.24.zip -d /tmp
+  echo "Finishing unzip qnn sdk."
+
+
+  # Print the content for manual verification
+  ls -lah "/tmp/qairt"
+  mv "/tmp/qairt"/* "${QNN_INSTALLATION_DIR}"
+  echo "Finishing installing qnn '${QNN_INSTALLATION_DIR}' ."
+
+  ls -lah "${QNN_INSTALLATION_DIR}"
+}
+
+install_qnn

.ci/scripts/test_llama.sh

@@ -72,6 +72,25 @@ fi
 
 echo "COREML option ${COREML}"
 
+if [[ "${MODE}" =~ .*qnn.* ]]; then
+  QNN=ON
+  export EXECUTORCH_ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/.." && pwd)"
+  export QNN_SDK_ROOT=/tmp/qnn/2.23.0.240531
+  export LD_LIBRARY_PATH="${QNN_SDK_ROOT}/lib/x86_64-linux-clang"
+  export PYTHONPATH=".."
+  cp schema/program.fbs exir/_serialize/program.fbs
+  cp schema/scalar_type.fbs exir/_serialize/scalar_type.fbs
+  cp -f build-x86/backends/qualcomm/PyQnnManagerAdaptor.cpython-310-x86_64-linux-gnu.so backends/qualcomm/python
+  cp -f build-x86/backends/qualcomm/PyQnnWrapperAdaptor.cpython-310-x86_64-linux-gnu.so backends/qualcomm/python
+
+else
+  QNN=OFF
+  QNN_SDK_ROOT=""
+fi
+
+echo "QNN option ${QNN}"
+echo "QNN_SDK_ROOT: ${QNN_SDK_ROOT}"
+
 if [[ -z "${BUCK:-}" ]]; then
   BUCK=buck2
 fi
@@ -96,6 +115,8 @@ cmake_install_executorch_libraries() {
         -DEXECUTORCH_BUILD_XNNPACK="$XNNPACK" \
         -DEXECUTORCH_BUILD_MPS="$MPS" \
         -DEXECUTORCH_BUILD_COREML="$COREML" \
+        -DEXECUTORCH_BUILD_QNN="$QNN" \
+        -DQNN_SDK_ROOT="$QNN_SDK_ROOT" \
         -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
         -Bcmake-out .
     cmake --build cmake-out -j9 --target install --config Debug
@@ -176,6 +197,9 @@ fi
 if [[ "${COREML}" == "ON" ]]; then
   EXPORT_ARGS="${EXPORT_ARGS} -kv -v --coreml --disable_dynamic_shape"
 fi
+if [[ "${QNN}" == "ON" ]]; then
+  EXPORT_ARGS="${EXPORT_ARGS} -kv -v --qnn --disable_dynamic_shape"
+fi
 # Add dynamically linked library location
 $PYTHON_EXECUTABLE -m examples.models.llama2.export_llama ${EXPORT_ARGS}
 

.github/workflows/docker-builds.yml

@@ -38,6 +38,7 @@ jobs:
           - docker-image-name: executorch-ubuntu-22.04-clang12
           - docker-image-name: executorch-ubuntu-22.04-linter
           - docker-image-name: executorch-ubuntu-22.04-arm-sdk
+          - docker-image-name: executorch-ubuntu-22.04-qnn-sdk
           - docker-image-name: executorch-ubuntu-22.04-clang12-android
     env:
       DOCKER_IMAGE: 308535385114.dkr.ecr.us-east-1.amazonaws.com/executorch/${{ matrix.docker-image-name }}

.github/workflows/trunk.yml

@@ -270,3 +270,38 @@ jobs:
         PYTHON_EXECUTABLE=python ${CONDA_RUN} bash examples/models/llama2/install_requirements.sh
         # Test llama2
         PYTHON_EXECUTABLE=python ${CONDA_RUN} bash .ci/scripts/test_llama.sh stories110M.pt "${BUILD_TOOL}" "${DTYPE}" "${MODE}"
+
+
+  test-llama-runner-qnn-linux:
+    name: test-llama-runner-qnn-linux
+    uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
+    strategy:
+      matrix:
+        dtype: [fp32]
+        build-tool: [cmake]
+        mode: [qnn]
+      fail-fast: false
+    with:
+      runner: linux.2xlarge
+      docker-image: executorch-ubuntu-22.04-qnn-sdk
+      submodules: 'true'
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      timeout: 900
+      script: |
+        # The generic Linux job chooses to use base env, not the one setup by the image
+        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
+        conda activate "${CONDA_ENV}"
+
+        DTYPE=${{ matrix.dtype }}
+        BUILD_TOOL=${{ matrix.build-tool }}
+        MODE=${{ matrix.mode }}
+
+        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-qnn-deps.sh
+        PYTHON_EXECUTABLE=python bash .ci/scripts/build-qnn-sdk.sh
+
+        # Setup executorch
+        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh buck2
+        # Install requirements for export_llama
+        PYTHON_EXECUTABLE=python bash examples/models/llama2/install_requirements.sh
+        # Test llama2
+        PYTHON_EXECUTABLE=python bash .ci/scripts/test_llama.sh stories110M.pt "${BUILD_TOOL}" "${DTYPE}" "${MODE}"

backends/qualcomm/aot/python/PyQnnWrapperAdaptor.h

@@ -86,7 +86,7 @@ class PyQnnOpWrapper {
         break;
       default:
         QNN_EXECUTORCH_LOG_ERROR(
-            "%s has invalid data type: %d", name, data_type);
+            "%s has invalid data type: %d", name.c_str(), data_type);
         break;
     }
   }

backends/qualcomm/scripts/build.sh

@@ -25,9 +25,9 @@ usage() {
 [ "$1" = -h ] && usage
 
 BUILD_X86_64="true"
-CMAKE_X86_64="cmake-out"
+CMAKE_X86_64="build-x86"
 BUILD_AARCH64="true"
-CMAKE_AARCH64="cmake-out-android"
+CMAKE_AARCH64="build-android"
 CLEAN="true"
 BUILD_TYPE="Debug"
 BUILD_JOB_NUMBER="16"