Merge branch 'main' into export-D79911643

Author: ethansfng

.Package.swift/kernels_torchao/dummy.swift

@@ -29,27 +29,22 @@ cmake -DPYTHON_EXECUTABLE=python \
     -DEXECUTORCH_ENABLE_LOGGING=1 \
     -DCMAKE_BUILD_TYPE=Release \
     -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
-    -DEXECUTORCH_BUILD_EXTENSION_LLM=ON \
-    -DEXECUTORCH_BUILD_EXTENSION_LLM_RUNNER=ON \
-    -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
     -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
+    -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
     -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
     -DEXECUTORCH_BUILD_XNNPACK=OFF \
     -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
     -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
+    -DEXECUTORCH_BUILD_KERNELS_TORCHAO=ON \
+    -DEXECUTORCH_BUILD_EXTENSION_LLM_RUNNER=ON \
+    -DEXECUTORCH_BUILD_EXTENSION_LLM=ON \
     -DEXECUTORCH_BUILD_KERNELS_LLM=ON \
     -Bcmake-out .
-cmake --build cmake-out -j16 --target install --config Release
+cmake --build cmake-out -j16 --config Release --target install
 
 # Install llama runner with torchao
 cmake -DPYTHON_EXECUTABLE=python \
-    -DBUILD_TESTING=OFF \
     -DCMAKE_BUILD_TYPE=Release \
-    -DEXECUTORCH_BUILD_KERNELS_LLM=ON \
-    -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
-    -DEXECUTORCH_BUILD_XNNPACK=OFF \
-    -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
-    -DEXECUTORCH_BUILD_TORCHAO=ON \
     -Bcmake-out/examples/models/llama \
     examples/models/llama
 cmake --build cmake-out/examples/models/llama -j16 --config Release

.Package.swift/kernels_torchao_debug/dummy.swift

@@ -485,7 +485,7 @@ jobs:
         eval "$(conda shell.bash hook)"
 
         # Install requirements
-        ${CONDA_RUN} EXECUTORCH_BUILD_TORCHAO=1 python install_executorch.py
+        ${CONDA_RUN} EXECUTORCH_BUILD_KERNELS_TORCHAO=1 python install_executorch.py
         ${CONDA_RUN} sh examples/models/llama/install_requirements.sh
 
         # Run test

.ci/scripts/test_llama_torchao_lowbit.sh

@@ -1,9 +1,6 @@
 [submodule "backends/arm/third-party/ethos-u-core-driver"]
 	path = backends/arm/third-party/ethos-u-core-driver
 	url = https://git.gitlab.arm.com/artificial-intelligence/ethos-u/ethos-u-core-driver.git
-[submodule "backends/arm/third-party/serialization_lib"]
-	path = backends/arm/third-party/serialization_lib
-	url = https://git.gitlab.arm.com/tosa/tosa-serialization.git
 [submodule "backends/vulkan/third-party/Vulkan-Headers"]
 	path = backends/vulkan/third-party/Vulkan-Headers
 	url = https://github.com/KhronosGroup/Vulkan-Headers

.github/workflows/trunk.yml

@@ -278,29 +278,6 @@ if(EXECUTORCH_BUILD_PTHREADPOOL)
   )
 endif()
 
-if(EXECUTORCH_BUILD_KERNELS_TORCHAO)
-  set(TORCHAO_BUILD_ATEN_OPS OFF)
-  set(TORCHAO_BUILD_EXECUTORCH_OPS ON)
-  set(TORCHAO_BUILD_CPU_AARCH64 ON)
-  set(TORCHAO_ENABLE_ARM_NEON_DOT ON)
-
-  list(
-    APPEND
-    TORCHAO_INCLUDE_DIRS
-    ${EXECUTORCH_ROOT}/backends/xnnpack/third-party/pthreadpool/include
-    ${EXECUTORCH_ROOT}/backends/xnnpack/third-party/cpuinfo/include
-    ${EXECUTORCH_ROOT}/third-party/ao
-  )
-
-  set(EXECUTORCH_INCLUDE_DIRS ${TORCHAO_INCLUDE_DIRS})
-
-  add_subdirectory(
-    ${CMAKE_CURRENT_SOURCE_DIR}/third-party/ao/torchao/experimental
-  )
-  executorch_target_link_options_shared_lib(torchao_ops_executorch)
-  list(APPEND _executorch_kernels torchao_ops_executorch)
-endif()
-
 if(EXECUTORCH_BUILD_TESTS)
   set(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR ON)
   include(CTest)
@@ -705,6 +682,65 @@ if(EXECUTORCH_BUILD_PTHREADPOOL AND EXECUTORCH_BUILD_CPUINFO)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/threadpool)
 endif()
 
+if(EXECUTORCH_BUILD_KERNELS_TORCHAO)
+  if(NOT TARGET cpuinfo)
+    message(
+      FATAL_ERROR
+        "EXECUTORCH_BUILD_KERNELS_TORCHAO requires EXECUTORCH_BUILD_CPUINFO be set ON"
+    )
+  endif()
+  if(NOT TARGET pthreadpool)
+    message(
+      FATAL_ERROR
+        "EXECUTORCH_BUILD_KERNELS_TORCHAO requires EXECUTORCH_BUILD_PTHREADPOOL be set ON"
+    )
+  endif()
+
+  # Configure TorchAO kernels
+  set(TORCHAO_BUILD_ATEN_OPS OFF)
+  set(TORCHAO_BUILD_EXECUTORCH_OPS ON)
+  set(TORCHAO_BUILD_CPU_AARCH64 ON)
+  set(TORCHAO_ENABLE_ARM_NEON_DOT ON)
+  set(TORCHAO_BUILD_KLEIDIAI ON)
+
+  # TorchAO kernels look for EXECUTORCH_INCLUDE_DIRS
+  if(DEFINED EXECUTORCH_INCLUDE_DIRS)
+    message(FATAL_ERROR "EXECUTORCH_INCLUDE_DIRS is already defined")
+  endif()
+  set(EXECUTORCH_INCLUDE_DIRS
+      ${EXECUTORCH_ROOT}/backends/xnnpack/third-party/pthreadpool/include
+      ${EXECUTORCH_ROOT}/backends/xnnpack/third-party/cpuinfo/include
+  )
+  add_subdirectory(
+    ${CMAKE_CURRENT_SOURCE_DIR}/third-party/ao/torchao/experimental
+  )
+  unset(EXECUTORCH_INCLUDE_DIRS)
+
+  executorch_target_link_options_shared_lib(torchao_ops_executorch)
+  list(APPEND _executorch_kernels torchao_ops_executorch)
+
+  install(
+    TARGETS torchao_ops_executorch torchao_kernels_aarch64
+    EXPORT ExecuTorchTargets
+    DESTINATION lib
+    INCLUDES
+    DESTINATION ${_common_include_directories}
+  )
+  # If using KleidiAI and XNNPACK has not installed it already, install it
+  if(TORCHAO_BUILD_KLEIDIAI AND NOT (EXECUTORCH_BUILD_XNNPACK
+                                     AND EXECUTORCH_XNNPACK_ENABLE_KLEIDI)
+  )
+    install(
+      TARGETS kleidiai
+      EXPORT ExecuTorchTargets
+      DESTINATION lib
+      INCLUDES
+      DESTINATION ${_common_include_directories}
+    )
+  endif()
+
+endif()
+
 if(EXECUTORCH_BUILD_PYBIND)
 
   # Add codegen tools subdirectory for selective_build pybind module

.gitmodules

@@ -84,6 +84,11 @@ let products = deliverables([
     ],
   ],
   "kernels_quantized": [:],
+  "kernels_torchao": [
+    "targets": [
+      "threadpool",
+    ],
+  ],
 ])
 
 let targets = deliverables([

CMakeLists.txt

@@ -120,6 +120,7 @@ runtime.python_test(
         "test/*.py",
     ]),
     deps = [
+        "fbsource//third-party/pypi/coremltools:coremltools",
         "fbsource//third-party/pypi/pytest:pytest",
         ":partitioner",
         ":quantizer",