Qualcomm AI Engine Direct - Uplevel QNN version for ci test (#5174)

* Qualcomm AI Engine Direct - Uplevel QNN version for ci test

Summary:
- For rms norm, we need to uplevel to QNN sdk 2.25 or above
- There is no libc++.so in ${QNN_SDK_ROOT}/lib/x86_64-linux-clang for QNN
sdk 2.23 or above.
So, we need to install it manually.

* enable llama test for qnn

Author: shewu-quic

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

@@ -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

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

@@ -7,14 +7,18 @@
 
 set -ex
 
+verify_pkg_installed() {
+  echo $(dpkg-query -W --showformat='${Status}\n' $1|grep "install ok installed")
+}
+
 install_qnn() {
   echo "Start installing qnn."
   QNN_INSTALLATION_DIR=/tmp/qnn
   mkdir -p "${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"
+  curl -Lo /tmp/v2.25.0.24.07.28.zip "https://softwarecenter.qualcomm.com/api/download/software/qualcomm_neural_processing_sdk/v2.25.0.240728.zip"
   echo "Finishing downloading qnn sdk."
-  unzip -qo /tmp/v2.23.0.24.06.24.zip -d /tmp
+  unzip -qo /tmp/v2.25.0.24.07.28.zip -d /tmp
   echo "Finishing unzip qnn sdk."
 
 
@@ -26,4 +30,22 @@ install_qnn() {
   ls -lah "${QNN_INSTALLATION_DIR}"
 }
 
+setup_libc++() {
+  sudo apt-get update
+  pkgs_to_check=('libc++-dev')
+  j=0
+  while [ $j -lt ${#pkgs_to_check[*]} ]; do
+    install_status=$(verify_pkg_installed ${pkgs_to_check[$j]})
+    if [ "$install_status" == "" ]; then
+      sudo apt-get install -y ${pkgs_to_check[$j]}
+      if [[ $? -ne 0 ]]; then
+        echo "ERROR: Failed to install required packages for libc++"
+        exit 1
+      fi
+    fi
+    j=$(( $j +1));
+  done
+}
+
+setup_libc++
 install_qnn

.ci/scripts/test_llama.sh

@@ -75,7 +75,7 @@ echo "COREML option ${COREML}"
 if [[ "${MODE}" =~ .*qnn.* ]]; then
   QNN=ON
   export EXECUTORCH_ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/.." && pwd)"
-  export QNN_SDK_ROOT=/tmp/qnn/2.23.0.240531
+  export QNN_SDK_ROOT=/tmp/qnn/2.25.0.240728
   export LD_LIBRARY_PATH="${QNN_SDK_ROOT}/lib/x86_64-linux-clang"
   export PYTHONPATH=".."
   cp schema/program.fbs exir/_serialize/program.fbs

.github/workflows/pull.yml

@@ -372,3 +372,38 @@ jobs:
 
         # Run pytest with coverage
         pytest -c /dev/null -v -n auto --cov=./ --cov-report=xml backends/arm/test
+
+
+  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-clang12-android
+      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 "${BUILD_TOOL}" "${DTYPE}" "${MODE}"