Merge branch 'main' into fix-unsqueeze-scalar-placeholder-pass

Author: YufengShi-dudu

.ci/scripts/test_model.sh

@@ -87,10 +87,6 @@ test_model() {
     bash examples/models/llava/install_requirements.sh
     STRICT="--no-strict"
   fi
-  if [[ "$MODEL_NAME" == "llama3_2_vision_encoder" || "$MODEL_NAME" == "llama3_2_text_decoder" ]]; then
-    # Install requirements for llama vision.
-    bash examples/models/llama3_2_vision/install_requirements.sh
-  fi
   if [[ "${MODEL_NAME}" == "qwen2_5" ]]; then
       # Install requirements for export_llama
       bash examples/models/llama/install_requirements.sh

.ci/scripts/unittest-linux.sh

@@ -24,9 +24,6 @@ if [[ "$BUILD_TOOL" == "cmake" ]]; then
     CMAKE_ARGS="-DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
     .ci/scripts/setup-linux.sh "$@"
 
-    # Install llama3_2_vision dependencies.
-    PYTHON_EXECUTABLE=python ./examples/models/llama3_2_vision/install_requirements.sh
-
     .ci/scripts/unittest-linux-cmake.sh
 elif [[ "$BUILD_TOOL" == "buck2" ]]; then
     # Removing this breaks sccache in the Buck build, apparently

.ci/scripts/unittest-macos.sh

@@ -29,7 +29,6 @@ if [[ "$BUILD_TOOL" == "cmake" ]]; then
     # Install llama3_2_vision dependencies.
     PYTHON_EXECUTABLE=python \
     ${CONDA_RUN} --no-capture-output \
-    ./examples/models/llama3_2_vision/install_requirements.sh
 
     .ci/scripts/unittest-macos-cmake.sh
 elif [[ "$BUILD_TOOL" == "buck2" ]]; then

.github/workflows/build-presets.yml

@@ -11,3 +11,20 @@ on:
 concurrency:
   group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}
   cancel-in-progress: true
+
+jobs:
+  apple:
+    uses: pytorch/test-infra/.github/workflows/macos_job.yml@main
+    strategy:
+      matrix:
+        preset: [macos-arm64]
+    with:
+      job-name: build
+      runner: macos-latest-xlarge
+      python-version: 3.12
+      submodules: recursive
+      script: |
+        set -eux
+        ${CONDA_RUN} ./install_requirements.sh > /dev/null
+        ${CONDA_RUN} cmake --preset ${{ matrix.preset }}
+        ${CONDA_RUN} cmake --build cmake-out --parallel

.github/workflows/pull.yml

@@ -434,9 +434,7 @@ jobs:
         output=$(ls -la cmake-out/test/size_test)
         arr=($output)
         size=${arr[4]}
-        # threshold=48120 on devserver with gcc11.4
-        # todo(lfq): update once binary size is below 50kb.
-        threshold="47552"
+        threshold="47560"
         if [[ "$size" -le "$threshold" ]]; then
           echo "Success $size <= $threshold"
         else

.lintrunner.toml

@@ -220,6 +220,13 @@ exclude_patterns = [
     'extension/**',
     'kernels/optimized/**',
     # Justified <functional> include.
+    'kernels/portable/cpu/op_bitwise*.cpp',
+    'kernels/portable/cpu/op_eq.cpp',
+    'kernels/portable/cpu/op_ge.cpp',
+    'kernels/portable/cpu/op_gt.cpp',
+    'kernels/portable/cpu/op_le.cpp',
+    'kernels/portable/cpu/op_lt.cpp',
+    'kernels/portable/cpu/op_ne.cpp',
     'runtime/kernel/thread_parallel_interface.h',
     'scripts/**',
     'third-party/**',

CMakeLists.txt

@@ -48,12 +48,18 @@ project(executorch)
 # MARK: - Start EXECUTORCH_H12025_BUILD_MIGRATION --------------------------------------------------
 
 include(${PROJECT_SOURCE_DIR}/tools/cmake/common/preset.cmake)
+
+load_build_preset()
 include(${PROJECT_SOURCE_DIR}/tools/cmake/preset/default.cmake)
 
+# Print all the configs that were called with announce_configured_options.
+print_configured_options()
+
 # MARK: - End EXECUTORCH_H12025_BUILD_MIGRATION ----------------------------------------------------
 
 include(tools/cmake/Utils.cmake)
 include(CMakeDependentOption)
+include(ExternalProject)
 
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
@@ -175,8 +181,6 @@ option(EXECUTORCH_BUILD_ARM_BAREMETAL
        "Build the Arm Baremetal flow for Cortex-M and Ethos-U" OFF
 )
 
-option(EXECUTORCH_BUILD_COREML "Build the Core ML backend" OFF)
-
 option(EXECUTORCH_BUILD_KERNELS_CUSTOM "Build the custom kernels" OFF)
 
 option(EXECUTORCH_BUILD_KERNELS_CUSTOM_AOT "Build the custom ops lib for AOT"
@@ -239,6 +243,8 @@ option(EXECUTORCH_USE_DL "Use libdl library" ON)
 
 option(EXECUTORCH_BUILD_CADENCE "Build the Cadence DSP backend" OFF)
 
+option(EXECUTORCH_BUILD_CORTEX_M "Build the Cortex-M backend" OFF)
+
 #
 # pthreadpool: build pthreadpool library. Disable on unsupported platforms
 #
@@ -255,6 +261,8 @@ cmake_dependent_option(
   "NOT EXECUTORCH_BUILD_ARM_BAREMETAL" OFF
 )
 
+add_subdirectory(third-party)
+
 if(EXECUTORCH_BUILD_EXTENSION_TRAINING)
   set(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER ON)
   set(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR ON)
@@ -449,81 +457,6 @@ if(EXECUTORCH_USE_CPP_CODE_COVERAGE)
   endif()
 endif()
 
-#
-# flatc: Flatbuffer commandline tool to generate .h files from .fbs files
-#
-cmake_dependent_option(
-  EXECUTORCH_BUILD_FLATC "Build the flatc executable." ON
-  "NOT FLATC_EXECUTABLE" OFF
-)
-
-set(FLATBUFFERS_BUILD_FLATC OFF CACHE BOOL "")
-set(FLATBUFFERS_BUILD_FLATHASH OFF CACHE BOOL "")
-set(FLATBUFFERS_BUILD_FLATLIB OFF CACHE BOOL "")
-set(FLATBUFFERS_BUILD_TESTS OFF CACHE BOOL "")
-set(FLATBUFFERS_INSTALL OFF CACHE BOOL "")
-# exir lets users set the alignment of tensor data embedded in the flatbuffer,
-# and some users need an alignment larger than the default, which is typically
-# 32.
-set(FLATBUFFERS_MAX_ALIGNMENT 1024)
-
-if(EXECUTORCH_BUILD_FLATC)
-  if(FLATC_EXECUTABLE)
-    # We could ignore this, but it could lead to confusion about which `flatc`
-    # is actually being used.
-    message(
-      FATAL_ERROR "May not set both EXECUTORCH_BUILD_FLATC and FLATC_EXECUTABLE"
-    )
-  endif()
-
-  # Build flatc for the *host* to generate files as part of the build step.
-  include(ExternalProject)
-  ExternalProject_Add(
-    flatbuffers
-    PREFIX ${CMAKE_CURRENT_BINARY_DIR}/third-party/flatbuffers
-    BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR}/third-party/flatbuffers
-    SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/third-party/flatbuffers
-    CMAKE_ARGS -DFLATBUFFERS_BUILD_FLATC=ON
-               -DFLATBUFFERS_BUILD_FLATHASH=${FLATBUFFERS_BUILD_FLATHASH}
-               -DFLATBUFFERS_BUILD_FLATLIB=${FLATBUFFERS_BUILD_FLATLIB}
-               -DFLATBUFFERS_BUILD_TESTS=${FLATBUFFERS_BUILD_TESTS}
-               -DFLATBUFFERS_INSTALL=${FLATBUFFERS_INSTALL}
-               -DCMAKE_CXX_FLAGS="-DFLATBUFFERS_MAX_ALIGNMENT=${FLATBUFFERS_MAX_ALIGNMENT}"
-               # If building for iOS, "unset" these variables to rely on the host (macOS) defaults.
-               $<$<AND:$<BOOL:${CMAKE_TOOLCHAIN_IOS}>,$<BOOL:$<FILTER:${PLATFORM},EXCLUDE,^MAC>>>:-DCMAKE_OSX_SYSROOT=>
-    INSTALL_COMMAND ""
-    BUILD_BYPRODUCTS <BINARY_DIR>/flatc
-  )
-  ExternalProject_Get_Property(flatbuffers BINARY_DIR)
-  if(WIN32)
-    # flatbuffers does not use CMAKE_BUILD_TYPE. Internally, the build forces Release
-    # config, but from CMake's perspective the build type is always Debug.
-    set(FLATC_EXECUTABLE ${BINARY_DIR}/$<CONFIG>/flatc.exe)
-  elseif(CMAKE_GENERATOR STREQUAL "Xcode")
-    set(FLATC_EXECUTABLE ${BINARY_DIR}/$<CONFIG>/flatc)
-  else()
-    set(FLATC_EXECUTABLE ${BINARY_DIR}/flatc)
-  endif()
-  set(FLATC_EXECUTABLE_BUILT_FROM_SOURCE YES)
-endif()
-
-if(NOT FLATC_EXECUTABLE)
-  message(
-    WARNING "FLATC_EXECUTABLE not specified, looking for flatc"
-  )
-  find_program(FLATC_EXECUTABLE flatc)
-
-  if(NOT FLATC_EXECUTABLE)
-    message(FATAL_ERROR "FLATC_EXECUTABLE must be set when EXECUTORCH_BUILD_FLATC is disabled.")
-  endif()
-endif()
-
-add_executable(flatc IMPORTED GLOBAL)
-set_target_properties(flatc PROPERTIES IMPORTED_LOCATION ${FLATC_EXECUTABLE})
-
-if(FLATC_EXECUTABLE_BUILT_FROM_SOURCE)
-  add_dependencies(flatc flatbuffers)
-endif()
 
 #
 # program_schema: Generated .h files from schema/*.fbs inputs
@@ -712,6 +645,10 @@ if(EXECUTORCH_BUILD_XNNPACK)
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/xnnpack)
 endif()
 
+if(EXECUTORCH_BUILD_CORTEX_M)
+  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/cortex_m)
+endif()
+
 if(EXECUTORCH_BUILD_DEVTOOLS)
   if(NOT EXECUTORCH_BUILD_ARM_BAREMETAL)
     set(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER

CMakePresets.json

@@ -0,0 +1,33 @@
+{
+  "version": 10,
+  "cmakeMinimumRequired": {
+    "major": 3,
+    "minor": 31,
+    "patch": 0
+  },
+  "$comment": "On-device AI across mobile, embedded and edge for PyTorch.",
+  "configurePresets": [
+    {
+      "name": "common",
+      "hidden": true,
+      "binaryDir": "${sourceDir}/cmake-out",
+      "generator": "Unix Makefiles"
+    },
+    {
+      "name": "macos-arm64",
+      "inherits": ["common"],
+      "generator": "Xcode",
+      "cacheVariables": {
+        "CMAKE_TOOLCHAIN_FILE": "${sourceDir}/third-party/ios-cmake/ios.toolchain.cmake",
+        "EXECUTORCH_BUILD_PRESET_FILE": "${sourceDir}/tools/cmake/preset/macos-arm64.cmake",
+        "PLATFORM": "MAC_ARM64",
+        "DEPLOYMENT_TARGET": "10.15"
+      },
+      "condition": {
+        "lhs": "${hostSystemName}",
+        "type": "equals",
+        "rhs": "Darwin"
+      }
+    }
+  ]
+}

backends/apple/mps/CMakeLists.txt

@@ -42,7 +42,7 @@ endforeach()
 add_custom_command(
   OUTPUT ${_mps_schema__outputs}
   COMMAND
-    ${FLATC_EXECUTABLE} --cpp --cpp-std c++11 --scoped-enums -o
+    flatc --cpp --cpp-std c++11 --scoped-enums -o
     "${_mps_schema__include_dir}/executorch/backends/apple/mps"
     ${_mps_schema__srcs}
   WORKING_DIRECTORY ${EXECUTORCH_ROOT}

backends/arm/operators/op_permute.py

@@ -46,24 +46,26 @@ def permutation_matrix_to_vector(permutation_matrix: torch.Tensor) -> list[int]:
     (1,0,2)
     """
     N = len(permutation_matrix)
-    assert N == len(
-        permutation_matrix[0]
-    ), f"A permutation matrix must be square, got shape {permutation_matrix.shape}"
+    if N != len(permutation_matrix[0]):
+        raise ValueError(
+            f"A permutation matrix must be square, got shape {permutation_matrix.shape}"
+        )
 
     p = [0] * N
     for row_index, row in enumerate(permutation_matrix):
         saw_one = False
         for col_index, value in enumerate(row):
             if value == 1:
-                assert (
-                    not saw_one
-                ), f"A permutation matrix can only have one 1 per row, got row {row}."
+                if saw_one:
+                    raise ValueError(
+                        f"A permutation matrix can only have one 1 per row, got {row=}"
+                    )
                 p[row_index] = col_index
                 saw_one = True
-            else:
-                assert (
-                    value == 0
-                ), f"A permutation matrix only contains 1's and 0's, got value {value}."
+            elif value != 0:
+                raise ValueError(
+                    f"A permutation matrix only contains 1's and 0's, got {value=}"
+                )
     return p