Merge branch 'main' into addGeluSupport

Author: leafs1

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-59d5cf083b4f860dea76fe8936076177f9367f10
+01f1cc44cbbfdf6307aa01b803a4ee22f9ade946

.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"

.ci/scripts/tests/test_gather_benchmark_configs.py

@@ -53,11 +53,11 @@ def test_skip_disabled_configs(self):
                 "mv3": [
                     self.gather_benchmark_configs.DisabledConfig(
                         config_name="disabled_config1",
-                        github_issue="https://github.com/org/repo/issues/123",
+                        github_issue="https://github.com/org/repo/issues/123",  # @lint-ignore
                     ),
                     self.gather_benchmark_configs.DisabledConfig(
                         config_name="disabled_config2",
-                        github_issue="https://github.com/org/repo/issues/124",
+                        github_issue="https://github.com/org/repo/issues/124",  # @lint-ignore
                     ),
                 ]
             },
@@ -84,7 +84,9 @@ def test_skip_disabled_configs(self):
             self.assertIn("enabled_config2", result)
 
     def test_disabled_configs_have_github_links(self):
-        github_issue_regex = re.compile(r"https://github\.com/.+/.+/issues/\d+")
+        github_issue_regex = re.compile(
+            r"https://github\.com/.+/.+/issues/\d+"  # @lint-ignore
+        )
 
         for (
             model_name,

.ci/scripts/unittest-linux.sh

@@ -20,8 +20,9 @@ if [[ "$BUILD_TOOL" == "cmake" ]]; then
     # Setup swiftshader and Vulkan SDK which are required to build the Vulkan delegate
     source .ci/scripts/setup-vulkan-linux-deps.sh
 
+    # We need the runner to test the built library.
     PYTHON_EXECUTABLE=python \
-    CMAKE_ARGS="-DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
+    CMAKE_ARGS="-DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
     .ci/scripts/setup-linux.sh "$@"
 
     .ci/scripts/unittest-linux-cmake.sh

.ci/scripts/unittest-macos.sh

@@ -20,8 +20,9 @@ export PATH="${TMP_DIR}:$PATH"
 trap 'rm -rfv ${TMP_DIR}' EXIT
 
 # Setup MacOS dependencies as there is no Docker support on MacOS atm
+# We need the runner to test the built library.
 PYTHON_EXECUTABLE=python \
-CMAKE_ARGS="-DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_COREML=ON -DEXECUTORCH_BUILD_MPS=ON -DEXECUTORCH_BUILD_XNNPACK=ON -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON" \
+CMAKE_ARGS="-DEXECUTORCH_BUILD_EXECUTOR_RUNNER=ON -DEXECUTORCH_BUILD_TESTS=ON" \
 ${CONDA_RUN} --no-capture-output \
 .ci/scripts/setup-macos.sh "$@"
 

.ci/scripts/wheel/envvar_base.sh

@@ -11,7 +11,3 @@
 # Ensure that CMAKE_ARGS is defined before referencing it. Defaults to empty
 # if not defined.
 export CMAKE_ARGS="${CMAKE_ARGS:-}"
-
-# Link the XNNPACK backend into the pybindings runtime so that users can execute
-# ExecuTorch programs that delegate to it.
-CMAKE_ARGS="${CMAKE_ARGS} -DEXECUTORCH_BUILD_PYBIND=ON -DEXECUTORCH_BUILD_XNNPACK=ON"

.ci/scripts/wheel/envvar_macos.sh

@@ -9,7 +9,3 @@
 # any variables so that subprocesses will see them.
 
 source "${GITHUB_WORKSPACE}/${REPOSITORY}/.ci/scripts/wheel/envvar_base.sh"
-
-# When building for macOS, link additional backends into the pybindings runtime.
-CMAKE_ARGS="${CMAKE_ARGS} -DEXECUTORCH_BUILD_COREML=ON"
-CMAKE_ARGS="${CMAKE_ARGS} -DEXECUTORCH_BUILD_MPS=ON"

.github/scripts/label_utils.py

@@ -51,7 +51,6 @@ def get_last_page_num_from_header(header: Any) -> int:
     # rel="next", <https://api.github.com/repositories/65600975/labels?per_page=100&page=3>; rel="last"
     link_info = header["link"]
     # Docs does not specify that it should be present for projects with just few labels
-    # And https://github.com/malfet/deleteme/actions/runs/7334565243/job/19971396887 it's not the case
     if link_info is None:
         return 1
     prefix = "&page="

.github/scripts/test_extract_benchmark_results.py

@@ -29,8 +29,8 @@ def get_mock_happy_flow_content(app_type: str = "IOS_APP"):
                 "name": "Syslog",
                 "type": "DEVICE_LOG",
                 "extension": "syslog",
-                "url": "https://job_arn_1_device_log",
-                "s3_url": "https://job_arn_1/test-workflow1/1/syslog.syslog",
+                "url": "https://job_arn_1_device_log",  # @lint-ignore
+                "s3_url": "https://job_arn_1/test-workflow1/1/syslog.syslog",  # @lint-ignore
                 "app_type": app_type,
                 "job_name": "job_arn_1_name",
                 "os": "14",
@@ -55,8 +55,8 @@ def get_mock_happy_flow_content(app_type: str = "IOS_APP"):
                 "name": "Customer Artifacts",
                 "type": "CUSTOMER_ARTIFACT",
                 "extension": "zip",
-                "url": "https://job_arn_1_customer_artifact",
-                "s3_url": "https://job_arn_1_customer_artifact1",
+                "url": "https://job_arn_1_customer_artifact",  # @lint-ignore
+                "s3_url": "https://job_arn_1_customer_artifact1",  # @lint-ignore
                 "app_type": app_type,
                 "job_name": "job_arn_1_device_name",
                 "os": "14",
@@ -68,8 +68,8 @@ def get_mock_happy_flow_content(app_type: str = "IOS_APP"):
                 "name": "Syslog",
                 "type": "DEVICE_LOG",
                 "extension": "syslog",
-                "url": "https://job_arn_1_device_log",
-                "s3_url": "https://job_arn_1/test-workflow1/1/syslog.syslog",
+                "url": "https://job_arn_1_device_log",  # @lint-ignore
+                "s3_url": "https://job_arn_1/test-workflow1/1/syslog.syslog",  # @lint-ignore
                 "app_type": app_type,
                 "job_name": "job_arn_2_name",
                 "os": "14",
@@ -94,8 +94,8 @@ def get_mock_happy_flow_content(app_type: str = "IOS_APP"):
                 "name": "Customer Artifacts",
                 "type": "CUSTOMER_ARTIFACT",
                 "extension": "zip",
-                "url": "https://job_arn_1_customer_artifact",
-                "s3_url": "https://job_arn_1_customer_artifact1",
+                "url": "https://job_arn_1_customer_artifact",  # @lint-ignore
+                "s3_url": "https://job_arn_1_customer_artifact1",  # @lint-ignore
                 "app_type": app_type,
                 "job_name": "job_arn_2_name",
                 "os": "14",

.github/scripts/trymerge.py

@@ -1964,7 +1964,7 @@ def check_for_sev(org: str, project: str, skip_mandatory_checks: bool) -> None:
     response = cast(
         Dict[str, Any],
         gh_fetch_json_list(
-            "https://api.github.com/search/issues",
+            "https://api.github.com/search/issues",  # @lint-ignore
             params={"q": f'repo:{org}/{project} is:open is:issue label:"ci: sev"'},
         ),
     )

.github/workflows/build-presets.yml

@@ -20,7 +20,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        preset: [macos-arm64, pybind]
+        preset: [macos, ios, ios-simulator, pybind, llm]
     with:
       job-name: build
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
@@ -32,14 +32,14 @@ jobs:
         set -eux
         ${CONDA_RUN} ./install_requirements.sh > /dev/null
         ${CONDA_RUN} cmake --preset ${{ matrix.preset }}
-        ${CONDA_RUN} cmake --build cmake-out --parallel
+        ${CONDA_RUN} cmake --build cmake-out -j$(( $(sysctl -n hw.ncpu) - 1 ))
 
   linux:
     uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
     strategy:
       fail-fast: false
       matrix:
-        preset: [pybind]
+        preset: [linux, 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:
@@ -65,4 +65,31 @@ jobs:
 
         ./install_requirements.sh > /dev/null
         cmake --preset ${{ matrix.preset }}
-        cmake --build cmake-out --parallel
+        cmake --build cmake-out -j$(( $(nproc) - 1 ))
+
+  windows:
+    uses: pytorch/test-infra/.github/workflows/windows_job.yml@main
+    strategy:
+      fail-fast: false
+      matrix:
+        preset: [pybind]
+    with:
+      job-name: build
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      submodules: recursive
+      timeout: 90
+      script: |
+        set -eux
+        conda init powershell
+        powershell -Command "& {
+          \$ErrorActionPreference = 'Stop'
+          Set-PSDebug -Trace 1
+
+          conda create --yes --quiet -n et python=3.12
+          conda activate et
+
+          python install_requirements.py
+          cmake --preset ${{ matrix.preset }}
+          \$numCores = [System.Environment]::GetEnvironmentVariable('NUMBER_OF_PROCESSORS') - 1
+          cmake --build cmake-out -j \$numCores
+        }"

.gitignore

@@ -40,3 +40,19 @@ xcuserdata/
 .swiftpm/
 *.xcworkspace/
 *.xcframework/
+
+# misc
+/.vscode/
+*.so
+*.dylib
+/cmake_wrapper.sh
+/data/
+/devtools/bundled_program/serialize/
+/exir/_serialize/program.fbs
+/exir/_serialize/scalar_type.fbs
+/include/
+/share/
+/version.py
+
+# Android
+*.aar

.gitmodules

@@ -67,3 +67,6 @@
 [submodule "shim"]
 	path = shim
 	url = https://github.com/facebook/buck2-shims-meta
+[submodule "third-party/json"]
+	path = third-party/json
+	url = https://github.com/nlohmann/json.git

.lintrunner.toml

@@ -378,3 +378,47 @@ command = [
     '--',
     '@{{PATHSFILE}}',
 ]
+
+[[linter]]
+code = "TORCH_AO_IMPORT"
+include_patterns = ["**/*.py"]
+exclude_patterns = [
+    "third-party/**",
+    # TODO: remove exceptions as we migrate
+    # backends
+    "backends/arm/quantizer/**",
+    "backends/arm/test/ops/**",
+    "backends/vulkan/quantizer/**",
+    "backends/vulkan/test/**",
+    "backends/qualcomm/quantizer/**",
+    "examples/qualcomm/**",
+    "backends/xnnpack/quantizer/**",
+    "backends/xnnpack/test/**",
+    "exir/tests/test_passes.py",
+    "extension/llm/export/builder.py",
+    "extension/llm/export/quantizer_lib.py",
+    "exir/tests/test_memory_planning.py",
+    "backends/transforms/duplicate_dynamic_quant_chain.py",
+    "exir/backend/test/demos/test_xnnpack_qnnpack.py",
+]
+
+command = [
+  "python3",
+  "-m",
+  "lintrunner_adapters",
+  "run",
+  "grep_linter",
+  "--pattern=\\bfrom torch\\.ao\\.quantization\\.(?:quantizer|observer|quantize_pt2e|pt2e)(?:\\.[A-Za-z0-9_]+)*\\b",
+  "--linter-name=TorchAOImport",
+  "--error-name=Prohibited torch.ao.quantization import",
+  """--error-description=\
+  Imports from torch.ao.quantization are not allowed. \
+  Please import from torchao.quantization.pt2e instead.\n \
+  * torchao.quantization.pt2e (includes all the utils, including observers, fake quants etc.) \n \
+  * torchao.quantization.pt2e.quantizer (quantizer related objects and utils) \n \
+  * torchao.quantization.pt2e.quantize_pt2e (prepare_pt2e, prepare_qat_pt2e, convert_pt2e) \n\n \
+  If you need something from torch.ao.quantization, you can add your file to an exclude_patterns for TORCH_AO_IMPORT in .lintrunner.toml. \
+  """,
+  "--",
+  "@{{PATHSFILE}}",
+]

.mypy.ini

@@ -97,3 +97,6 @@ ignore_missing_imports = True
 
 [mypy-zstd]
 ignore_missing_imports = True
+
+[mypy-torchao.*]
+follow_untyped_imports = True

CMakeLists.txt

@@ -362,6 +362,9 @@ add_library(executorch_core ${_executorch_core__srcs})
 add_library(executorch_no_prim_ops ALIAS executorch_core)
 
 target_link_libraries(executorch_core PRIVATE program_schema)
+if(ANDROID)
+  target_link_libraries(executorch_core PUBLIC log)
+endif()
 if(EXECUTORCH_USE_DL)
   # Check if dl exists for this toolchain and only then link it.
   find_library(DL_LIBRARY_EXISTS NAMES dl)
@@ -653,10 +656,6 @@ if(EXECUTORCH_BUILD_PYBIND)
   install(TARGETS portable_lib
           LIBRARY DESTINATION executorch/extension/pybindings
   )
-
-  if(EXECUTORCH_BUILD_EXTENSION_TRAINING)
-
-  endif()
 endif()
 
 if(EXECUTORCH_BUILD_EXTENSION_TRAINING)