Merge branch 'main' into migrate-xnnpack-vulkan

Author: metascroy

.ci/scripts/build_llama_android.sh

@@ -22,18 +22,12 @@ install_executorch_and_backend_lib() {
   ANDROID_NDK=/opt/ndk
   BUCK2=buck2
   ANDROID_ABI=arm64-v8a
-  cmake -DBUCK2="${BUCK2}" \
+  cmake --preset llm \
+    -DBUCK2="${BUCK2}" \
     -DCMAKE_TOOLCHAIN_FILE="${ANDROID_NDK}/build/cmake/android.toolchain.cmake" \
     -DANDROID_ABI="${ANDROID_ABI}" \
     -DCMAKE_INSTALL_PREFIX=cmake-android-out \
     -DCMAKE_BUILD_TYPE=Release \
-    -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
-    -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
-    -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
-    -DEXECUTORCH_BUILD_XNNPACK=ON \
-    -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-    -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
-    -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
     -DXNNPACK_ENABLE_ARM_BF16=OFF \
     -Bcmake-android-out .
 
@@ -51,11 +45,7 @@ build_llama_runner() {
     -DCMAKE_TOOLCHAIN_FILE="$ANDROID_NDK"/build/cmake/android.toolchain.cmake  \
     -DANDROID_ABI="${ANDROID_ABI}" \
     -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 \
+    -DCMAKE_BUILD_TYPE=Release \
     -Bcmake-android-out/examples/models/llama examples/models/llama
 
     cmake --build cmake-android-out/examples/models/llama -j4 --config Release

.ci/scripts/test_llama.sh

@@ -152,21 +152,11 @@ which "${PYTHON_EXECUTABLE}"
 cmake_install_executorch_libraries() {
     echo "Installing libexecutorch.a, libextension_module.so, libportable_ops_lib.a"
     rm -rf cmake-out
-    retry cmake \
+    retry cmake --preset llm \
         -DCMAKE_INSTALL_PREFIX=cmake-out \
         -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" \
-        -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
-        -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
-        -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
-        -DEXECUTORCH_BUILD_KERNELS_CUSTOM="$CUSTOM" \
-        -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-        -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
-        -DEXECUTORCH_BUILD_XNNPACK="$XNNPACK" \
-        -DEXECUTORCH_BUILD_MPS="$MPS" \
-        -DEXECUTORCH_BUILD_COREML="$COREML" \
         -DEXECUTORCH_BUILD_QNN="$QNN" \
         -DQNN_SDK_ROOT="$QNN_SDK_ROOT" \
-        -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
         -Bcmake-out .
     cmake --build cmake-out -j9 --target install --config "$CMAKE_BUILD_TYPE"
 }
@@ -181,10 +171,6 @@ cmake_build_llama_runner() {
     retry cmake \
         -DCMAKE_INSTALL_PREFIX=cmake-out \
         -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" \
-        -DEXECUTORCH_BUILD_KERNELS_CUSTOM="$CUSTOM" \
-        -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-        -DEXECUTORCH_BUILD_XNNPACK="$XNNPACK" \
-        -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
         -Bcmake-out/${dir} \
         ${dir}
     cmake --build cmake-out/${dir} -j9 --config "$CMAKE_BUILD_TYPE"

.github/workflows/build-presets.yml

@@ -20,26 +20,26 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        preset: [macos-arm64, pybind]
+        preset: [macos-arm64, pybind, llm]
     with:
       job-name: build
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
       runner: macos-latest-xlarge
       python-version: 3.12
       submodules: recursive
-      timeout: 90
+      timeout: 900
       script: |
         set -eux
         ${CONDA_RUN} ./install_requirements.sh > /dev/null
         ${CONDA_RUN} cmake --preset ${{ matrix.preset }}
-        ${CONDA_RUN} cmake --build cmake-out --parallel
+        ${CONDA_RUN} cmake --build cmake-out -j16
 
   linux:
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
     strategy:
       fail-fast: false
       matrix:
-        preset: [pybind]
+        preset: [pybind, llm]
         runner: [linux.2xlarge, linux.arm64.2xlarge]
         docker-image: [executorch-ubuntu-22.04-clang12, executorch-ubuntu-22.04-gcc11-aarch64]
         # Excluding specific runner + docker image combinations that don't make sense:

CMakePresets.json

@@ -36,6 +36,26 @@
         "string": "${hostSystemName}",
         "list": ["Darwin", "Linux", "Windows"]
       }
+    },
+    {
+        "name": "llm",
+        "displayName": "Build LLM libraries",
+        "inherits": [
+            "common"
+        ],
+        "cacheVariables": {
+            "EXECUTORCH_BUILD_PRESET_FILE": "${sourceDir}/tools/cmake/preset/llm.cmake",
+            "CMAKE_OSX_DEPLOYMENT_TARGET": "10.15"
+        },
+        "condition": {
+            "type": "inList",
+            "string": "${hostSystemName}",
+            "list": [
+                "Darwin",
+                "Linux",
+                "Windows"
+            ]
+        }
     }
   ]
 }

backends/qualcomm/CMakeLists.txt

@@ -153,12 +153,12 @@ target_link_libraries(qnn_executorch_logging PRIVATE qnn_schema)
 target_link_libraries(qnn_profiler PRIVATE qnn_executorch_logging)
 target_link_libraries(qnn_logger PRIVATE qnn_implementation ${android_log})
 target_link_libraries(qnn_backend PRIVATE qnn_implementation qnn_logger)
-target_link_libraries(qnn_custom_protocol PRIVATE qcir_utils)
+target_link_libraries(qnn_custom_protocol PRIVATE qnn_logger)
 target_link_libraries(
   qnn_device PRIVATE qnn_executorch_logging qnn_implementation qnn_logger
 )
 target_link_libraries(
-  qnn_backend_cache PRIVATE qnn_sys_implementation qcir_utils
+  qnn_backend_cache PRIVATE qnn_sys_implementation
 )
 target_link_libraries(
   qnn_context PRIVATE qnn_implementation qnn_logger qnn_backend qnn_device
@@ -184,7 +184,7 @@ target_link_libraries(
 )
 target_link_libraries(
   qnn_executorch_backend PRIVATE qnn_executorch_header qnn_schema qnn_manager
-                                 executorch_core qcir_utils extension_tensor
+                                 executorch_core extension_tensor
 )
 set_target_properties(
   qnn_executorch_backend PROPERTIES LINK_FLAGS "-Wl,-rpath='$ORIGIN'"
@@ -243,7 +243,6 @@ if(${CMAKE_SYSTEM_PROCESSOR} MATCHES "x86_64")
             qnn_manager
             qnn_executorch_header
             executorch
-            qcir_utils
             extension_tensor
   )
   target_link_libraries(

backends/qualcomm/aot/python/PyQnnManagerAdaptor.cpp

@@ -30,15 +30,14 @@ PYBIND11_MODULE(PyQnnManagerAdaptor, m) {
   py::class_<PyQnnManager, std::shared_ptr<PyQnnManager>>(m, "QnnManager")
       .def(py::init<const py::bytes&>())
       .def(py::init<const py::bytes&, const py::bytes&>())
-      .def(py::init<const py::bytes&, const py::list&>())
       .def("Init", &PyQnnManager::Init)
       .def("IsNodeSupportedByBackend", &PyQnnManager::IsNodeSupportedByBackend)
-      .def("Compile", py::overload_cast<>(&PyQnnManager::Compile))
       .def(
           "Compile",
           py::overload_cast<
-              const std::string&,
-              std::vector<std::shared_ptr<OpWrapper>>&>(&PyQnnManager::Compile))
+              const std::vector<std::string>&,
+              std::vector<std::vector<std::shared_ptr<OpWrapper>>>&>(
+              &PyQnnManager::Compile))
       .def("Destroy", &PyQnnManager::Destroy)
       .def("IsAvailable", &PyQnnManager::IsAvailable)
       .def("IsTensorDump", &PyQnnManager::IsTensorDump)