Merge branch 'pytorch:main' into main

Author: psiddh

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

@@ -32,6 +32,8 @@ set_up_aot() {
       -DQNN_SDK_ROOT=${QNN_SDK_ROOT} \
       -DEXECUTORCH_BUILD_DEVTOOLS=ON \
       -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
+      -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
+      -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
       -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
       -DEXECUTORCH_ENABLE_EVENT_TRACER=ON \
       -DPYTHON_EXECUTABLE=python3

.ci/scripts/build_llama_android.sh

@@ -42,6 +42,7 @@ build_llama_runner() {
     popd
     ANDROID_ABI=arm64-v8a
     cmake -DBUCK2="${BUCK2}" \
+    -DBUILD_TESTING=OFF \
     -DCMAKE_TOOLCHAIN_FILE="$ANDROID_NDK"/build/cmake/android.toolchain.cmake  \
     -DANDROID_ABI="${ANDROID_ABI}" \
     -DCMAKE_INSTALL_PREFIX=cmake-android-out \

.ci/scripts/test_llama.sh

@@ -169,6 +169,7 @@ cmake_build_llama_runner() {
     popd
     dir="examples/models/llama"
     retry cmake \
+        -DBUILD_TESTING=OFF \
         -DCMAKE_INSTALL_PREFIX=cmake-out \
         -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" \
         -Bcmake-out/${dir} \

.ci/scripts/test_llama_torchao_lowbit.sh

@@ -30,6 +30,7 @@ cmake -DPYTHON_EXECUTABLE=python \
     -DCMAKE_BUILD_TYPE=Release \
     -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
     -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
+    -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
     -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
     -DEXECUTORCH_BUILD_XNNPACK=OFF \
     -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
@@ -40,6 +41,7 @@ cmake --build cmake-out -j16 --target install --config Release
 
 # Install llama runner with torchao
 cmake -DPYTHON_EXECUTABLE=python \
+    -DBUILD_TESTING=OFF \
     -DCMAKE_BUILD_TYPE=Release \
     -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
     -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \

.ci/scripts/test_llava.sh

@@ -37,6 +37,7 @@ EXECUTORCH_COMMON_CMAKE_ARGS="                      \
         -DEXECUTORCH_ENABLE_LOGGING=ON              \
         -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON      \
         -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
+        -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
         -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON      \
         -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON        \
         -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON     \
@@ -64,9 +65,10 @@ cmake_install_executorch_libraries_for_android() {
 
 
 LLAVA_COMMON_CMAKE_ARGS="                        \
+        -DBUILD_TESTING=OFF                      \
         -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
         -DCMAKE_INSTALL_PREFIX=${BUILD_DIR}      \
-        -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}         \
+        -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}   \
         -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON     \
         -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON  \
         -DEXECUTORCH_BUILD_XNNPACK=ON"

.ci/scripts/test_phi_3_mini.sh

@@ -27,6 +27,7 @@ cmake_install_executorch_libraries() {
       -DEXECUTORCH_ENABLE_LOGGING=1 \
       -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
       -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
+      -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
       -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
       -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
       -DEXECUTORCH_BUILD_XNNPACK=ON \

.ci/scripts/unittest-buck2.sh

@@ -15,7 +15,7 @@ buck2 query "//backends/apple/... + //backends/example/... + \
 //kernels/optimized/... + //kernels/portable/... + //kernels/quantized/... + \
 //kernels/test/... + //runtime/... + //schema/... + //test/... + //util/..."
 
-UNBUILDABLE_OPTIMIZED_OPS_REGEX="gelu|fft_r2c|log_softmax"
+UNBUILDABLE_OPTIMIZED_OPS_REGEX="_elu|gelu|fft|log_softmax"
 BUILDABLE_OPTIMIZED_OPS=$(buck2 query //kernels/optimized/cpu/... | grep -E -v $UNBUILDABLE_OPTIMIZED_OPS_REGEX)
 
 # TODO: build prim_ops_test_cpp again once supported_features works in
@@ -24,6 +24,8 @@ BUILDABLE_KERNELS_PRIM_OPS_TARGETS=$(buck2 query //kernels/prim_ops/... | grep -
 # TODO: expand the covered scope of Buck targets.
 # //runtime/kernel/... is failing because //third-party:torchgen_files's shell script can't find python on PATH.
 # //runtime/test/... requires Python torch, which we don't have in our OSS buck setup.
-buck2 test $BUILDABLE_OPTIMIZED_OPS //kernels/portable/... \
-      $BUILDABLE_KERNELS_PRIM_OPS_TARGETS //runtime/backend/... //runtime/core/... \
-      //runtime/executor: //runtime/kernel/... //runtime/platform/...
+for op in "build" "test"; do
+    buck2 $op $BUILDABLE_OPTIMIZED_OPS //kernels/portable/... \
+          $BUILDABLE_KERNELS_PRIM_OPS_TARGETS //runtime/backend/... //runtime/core/... \
+          //runtime/executor: //runtime/kernel/... //runtime/platform/...
+done

.github/workflows/trunk.yml

@@ -262,7 +262,7 @@ jobs:
         output=$(ls -la ${elf})
         arr=($output)
         size=${arr[4]}
-        threshold="102400" # 100KiB
+        threshold="103268" # ~100KiB
         echo "size: $size, threshold: $threshold"
         if [[ "$size" -le "$threshold" ]]; then
           echo "Success $size <= $threshold"
@@ -552,6 +552,7 @@ jobs:
           -DEXECUTORCH_ENABLE_LOGGING=1 \
           -DCMAKE_BUILD_TYPE=Release \
           -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
+          -DEXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR=ON \
           -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
           -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
           -DEXECUTORCH_BUILD_XNNPACK=ON \
@@ -686,3 +687,32 @@ jobs:
       build-mode: Release
       build-tool: cmake
       docker-image: executorch-ubuntu-22.04-clang12
+
+  unittest-nxp-neutron:
+    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+    permissions:
+      id-token: write
+      contents: read
+    with:
+      runner: linux.2xlarge
+      docker-image: executorch-ubuntu-22.04-clang12
+      submodules: 'recursive'
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      timeout: 90
+      script: |
+        set -eux
+        
+        # The generic Linux job chooses to use base env, not the one setup by the image
+        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
+        conda activate "${CONDA_ENV}"
+        
+        # Build and install Executorch
+        PYTHON_EXECUTABLE=python \
+        CMAKE_ARGS="-DEXECUTORCH_BUILD_NXP_NEUTRON=ON" \
+        .ci/scripts/setup-linux.sh --build-tool "cmake"
+        
+        # Install test requirements
+        pip install -r backends/nxp/requirements-tests.txt
+        
+        # Run pytest
+        PYTHON_EXECUTABLE=python bash backends/nxp/run_unittests.sh

.lintrunner.toml

@@ -271,6 +271,10 @@ exclude_patterns = [
     'examples/**',
     'exir/verification/bindings.cpp',
     'extension/**',
+    # Uses properly-gated (ET_USE_PYTORCH_HEADERS) ATen include.
+    'kernels/portable/cpu/util/elementwise_util.h',
+    'kernels/portable/cpu/util/math_util.h',
+    'kernels/portable/cpu/util/vectorized_math.h',
     'kernels/optimized/**',
     'runtime/core/exec_aten/**',
     # Want to be able to keep c10 in sync with PyTorch core.
@@ -386,15 +390,9 @@ exclude_patterns = [
     "third-party/**",
     # TODO: remove exceptions as we migrate
     # backends
-    "backends/vulkan/quantizer/**",
-    "backends/vulkan/test/**",
-    "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",
     "exir/backend/test/demos/test_xnnpack_qnnpack.py",
+    "backends/xnnpack/test/test_xnnpack_utils.py",
 ]
 
 command = [

CMakeLists.txt

@@ -75,9 +75,13 @@ if(NOT PYTHON_EXECUTABLE)
 endif()
 announce_configured_options(PYTHON_EXECUTABLE)
 
+if(NOT BUCK2)
+  resolve_buck2()
+endif()
+announce_configured_options(BUCK2)
+
 announce_configured_options(CMAKE_CXX_COMPILER_ID)
 announce_configured_options(CMAKE_TOOLCHAIN_FILE)
-announce_configured_options(BUCK2)
 
 load_build_preset()
 include(${PROJECT_SOURCE_DIR}/tools/cmake/preset/default.cmake)
@@ -148,37 +152,11 @@ else()
 endif()
 
 if(EXECUTORCH_BUILD_TESTS)
-  set(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR ON)
   include(CTest)
 endif()
 
 add_subdirectory(third-party)
 
-if(EXECUTORCH_BUILD_EXTENSION_TRAINING)
-  set(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER ON)
-  set(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR ON)
-  set(EXECUTORCH_BUILD_EXTENSION_MODULE ON)
-  set(EXECUTORCH_BUILD_EXTENSION_TENSOR ON)
-endif()
-
-if(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR)
-  set(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER ON)
-endif()
-
-if(EXECUTORCH_BUILD_EXTENSION_MODULE)
-  set(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER ON)
-  set(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR ON)
-endif()
-
-if(EXECUTORCH_BUILD_KERNELS_CUSTOM_AOT)
-  set(EXECUTORCH_BUILD_EXTENSION_TENSOR ON)
-  set(EXECUTORCH_BUILD_KERNELS_CUSTOM ON)
-endif()
-
-if(EXECUTORCH_BUILD_KERNELS_CUSTOM)
-  set(EXECUTORCH_BUILD_KERNELS_OPTIMIZED ON)
-endif()
-
 if(NOT DEFINED FXDIV_SOURCE_DIR)
   set(ORIGINAL_CMAKE_POSITION_INDEPENDENT_CODE_FLAG
       ${CMAKE_POSITION_INDEPENDENT_CODE}
@@ -290,9 +268,6 @@ set(_common_include_directories
 #
 
 if(NOT EXECUTORCH_SRCS_FILE)
-  # Find or download buck2 binary.
-  resolve_buck2()
-
   # A file wasn't provided. Run a script to extract the source lists from the
   # buck2 build system and write them to a file we can include.
   #
@@ -335,7 +310,7 @@ if(EXECUTORCH_USE_CPP_CODE_COVERAGE)
            " -fprofile-instr-generate -fcoverage-mapping"
     )
   else()
-    message(ERROR
+    message(FATAL_ERROR
             "Code coverage for compiler ${CMAKE_CXX_COMPILER_ID} is unsupported"
     )
   endif()
@@ -455,14 +430,6 @@ endif()
 
 add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/configurations)
 
-#
-# gflags: Commandline flag host library.
-#
-
-if(EXECUTORCH_BUILD_GFLAGS)
-  add_subdirectory(third-party/gflags)
-endif()
-
 # Install `executorch` library as well as `executorch-config.cmake` under
 # ${CMAKE_INSTALL_PREFIX}/
 install(
@@ -547,17 +514,6 @@ if(EXECUTORCH_BUILD_CORTEX_M)
 endif()
 
 if(EXECUTORCH_BUILD_DEVTOOLS)
-  if(NOT EXECUTORCH_BUILD_ARM_BAREMETAL)
-    set(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER
-        ON
-        CACHE BOOL "EXECUTORCH_BUILD_EXTENSION_DATA_LOADER" FORCE
-    )
-  else()
-    set(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER
-        OFF
-        CACHE BOOL "EXECUTORCH_BUILD_EXTENSION_DATA_LOADER" FORCE
-    )
-  endif()
   add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/devtools)
 endif()
 

backends/apple/coreml/scripts/build_tests.sh

@@ -33,8 +33,7 @@ cmake "$EXECUTORCH_ROOT_PATH" -B"$CMAKE_EXECUTORCH_BUILD_DIR_PATH" \
 -DPLATFORM=MAC_UNIVERSAL \
 -DDEPLOYMENT_TARGET=13.0 \
 -DEXECUTORCH_BUILD_EXECUTOR_RUNNER=OFF \
--DEXECUTORCH_BUILD_XNNPACK=OFF \
--DEXECUTORCH_BUILD_GFLAGS=OFF
+-DEXECUTORCH_BUILD_XNNPACK=OFF
 
 cmake --build "$CMAKE_EXECUTORCH_BUILD_DIR_PATH"  -j9 -t executorch
 

backends/arm/_passes/arm_pass_manager.py

@@ -62,7 +62,10 @@
     UnsqueezeScalarPlaceholdersPass,
 )
 
-from executorch.backends.arm.tosa_specification import TosaSpecification
+from executorch.backends.arm.tosa_specification import (
+    TosaLoweringContext,
+    TosaSpecification,
+)
 from executorch.backends.transforms.decompose_sdpa import (
     DecomposeScaledDotProductAttention,
 )
@@ -80,7 +83,8 @@ def __init__(self, tosa_spec: TosaSpecification) -> None:
         super().__init__()
 
     def _transform(self, graph_module: GraphModule):
-        return self(graph_module).graph_module
+        with TosaLoweringContext(self.tosa_spec):
+            return self(graph_module).graph_module
 
     def _tosa_080_BI_pipeline(self, exported_program: ExportedProgram) -> GraphModule:
         self.add_pass(FuseQuantizedActivationPass())

backends/arm/_passes/scalars_to_attribute_pass.py

@@ -12,8 +12,8 @@
 from executorch.backends.arm._passes.arm_pass_utils import get_first_fake_tensor
 
 from executorch.exir.pass_base import ExportPass, PassResult
-from torch.ao.quantization.fx.utils import get_new_attr_name_with_prefix
 from torch.fx import GraphModule, Node
+from torchao.quantization.pt2e.utils import get_new_attr_name_with_prefix
 
 
 class ScalarsToAttributePass(ExportPass):

backends/arm/operators/op_abs.py

@@ -44,7 +44,7 @@ def define_node(
         import tosa_tools.v0_80.serializer.tosa_serializer as ts  # type: ignore
 
         validate_num_inputs(self.target, inputs, 1)
-        validate_same_dtype(self.target, [*inputs, output])
+        validate_same_dtype(self.target, [*inputs, output], ts)
 
         # Handle int8 (quantized) and int32
         if not (inputs[0].dtype in [ts.DType.INT8, ts.DType.INT32]):
@@ -106,7 +106,7 @@ def define_node(
         import tosa_tools.v0_80.serializer.tosa_serializer as ts  # type: ignore
 
         validate_num_inputs(self.target, inputs, 1)
-        validate_same_dtype(self.target, [*inputs, output])
+        validate_same_dtype(self.target, [*inputs, output], ts)
 
         if inputs[0].dtype in [ts.DType.INT8, ts.DType.INT32]:
             # Call the inherited define_node for handling integers
@@ -153,7 +153,7 @@ def define_node(
         import serializer.tosa_serializer as ts  # type: ignore
 
         validate_num_inputs(self.target, inputs, 1)
-        validate_same_dtype(self.target, [*inputs, output])
+        validate_same_dtype(self.target, [*inputs, output], ts)
 
         # Handle int8 (quantized) and int32
         if not (inputs[0].dtype in [ts.DType.INT8, ts.DType.INT32]):
@@ -216,7 +216,7 @@ def define_node(
         import serializer.tosa_serializer as ts  # type: ignore
 
         validate_num_inputs(self.target, inputs, 1)
-        validate_same_dtype(self.target, [*inputs, output])
+        validate_same_dtype(self.target, [*inputs, output], ts)
 
         if inputs[0].dtype in [ts.DType.INT8, ts.DType.INT32]:
             # Call the inherited define_node for handling integers

backends/arm/operators/op_add.py

@@ -45,7 +45,7 @@ def define_node(
         import tosa_tools.v0_80.serializer.tosa_serializer as ts  # type: ignore
 
         validate_num_inputs(self.target, inputs, 2)
-        validate_same_dtype(self.target, [*inputs, output])
+        validate_same_dtype(self.target, [*inputs, output], ts)
 
         # Handle int8 (quantized) and int32
         supported_dtypes = [ts.DType.INT8, ts.DType.INT32]
@@ -118,7 +118,7 @@ def define_node(
         import tosa_tools.v0_80.serializer.tosa_serializer as ts  # type: ignore
 
         validate_num_inputs(self.target, inputs, 2)
-        validate_same_dtype(self.target, [*inputs, output])
+        validate_same_dtype(self.target, [*inputs, output], ts)
 
         if inputs[0].dtype in [ts.DType.INT8, ts.DType.INT32]:
             # Call the inherited define_node for handling integers
@@ -163,7 +163,7 @@ def define_node(
         import serializer.tosa_serializer as ts  # type: ignore
 
         validate_num_inputs(self.target, inputs, 2)
-        validate_same_dtype(self.target, [*inputs, output])
+        validate_same_dtype(self.target, [*inputs, output], ts)
 
         # Handle int8 (quantized) and int32
         supported_dtypes = [ts.DType.INT8, ts.DType.INT32]
@@ -226,7 +226,7 @@ def define_node(
         import serializer.tosa_serializer as ts  # type: ignore
 
         validate_num_inputs(self.target, inputs, 2)
-        validate_same_dtype(self.target, [*inputs, output])
+        validate_same_dtype(self.target, [*inputs, output], ts)
 
         if inputs[0].dtype in [ts.DType.INT8, ts.DType.INT32]:
             # Call the inherited define_node for handling integers