Update

[ghstack-poisoned]

Author: swolchok

.ci/scripts/build_llama_android.sh

@@ -60,6 +60,5 @@ build_llama_runner() {
 
     cmake --build cmake-android-out/examples/models/llama -j4 --config Release
 }
-install_flatc_from_source
 install_executorch_and_backend_lib
 build_llama_runner

.ci/scripts/utils.sh

@@ -80,25 +80,6 @@ install_pytorch_and_domains() {
   sccache --show-stats || true
 }
 
-install_flatc_from_source() {
-  # NB: This function could be used to install flatbuffer from source
-  pushd third-party/flatbuffers || return
-
-  cmake -G "Unix Makefiles" -DCMAKE_BUILD_TYPE=Release
-  if [ "$(uname)" == "Darwin" ]; then
-    CMAKE_JOBS=$(( $(sysctl -n hw.ncpu) - 1 ))
-  else
-    CMAKE_JOBS=$(( $(nproc) - 1 ))
-  fi
-  cmake --build . -j "${CMAKE_JOBS}"
-
-  # Copy the flatc binary to conda path
-  EXEC_PATH=$(dirname "$(which python)")
-  cp flatc "${EXEC_PATH}"
-
-  popd || return
-}
-
 build_executorch_runner_buck2() {
   # Build executorch runtime with retry as this step is flaky on macos CI
   retry buck2 build //examples/portable/executor_runner:executor_runner
@@ -111,9 +92,14 @@ build_executorch_runner_cmake() {
   mkdir "${CMAKE_OUTPUT_DIR}"
 
   pushd "${CMAKE_OUTPUT_DIR}" || return
+  if [[ $1 == "Debug" ]]; then
+      CXXFLAGS="-fsanitize=address,undefined"
+  else
+      CXXFLAGS=""
+  fi
   # This command uses buck2 to gather source files and buck2 could crash flakily
   # on MacOS
-  retry cmake -DPYTHON_EXECUTABLE="${PYTHON_EXECUTABLE}" -DCMAKE_BUILD_TYPE="${1:-Release}" ..
+  CXXFLAGS="$CXXFLAGS" retry cmake -DPYTHON_EXECUTABLE="${PYTHON_EXECUTABLE}" -DCMAKE_BUILD_TYPE="${1:-Release}" ..
   popd || return
 
   if [ "$(uname)" == "Darwin" ]; then

CMakeLists.txt

@@ -460,7 +460,7 @@ endif()
 # tools like `flatc`, along with example executables like `executor_runner` and
 # libraries that it uses, like `gflags`. Disabling this can be helpful when
 # cross-compiling, but some required tools that would have been built need to be
-# provided directly (via, for example, FLATC_EXECUTABLE).
+# provided directly.
 cmake_dependent_option(
   EXECUTORCH_BUILD_HOST_TARGETS "Build host-only targets." ON
   "NOT CMAKE_TOOLCHAIN_IOS" OFF
@@ -471,10 +471,9 @@ cmake_dependent_option(
 #
 cmake_dependent_option(
   EXECUTORCH_BUILD_FLATC "Build the flatc executable." ON
-  "NOT FLATC_EXECUTABLE;EXECUTORCH_BUILD_HOST_TARGETS" OFF
+  "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 "")
@@ -507,6 +506,8 @@ if(EXECUTORCH_BUILD_FLATC)
                -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
   )
@@ -515,6 +516,8 @@ if(EXECUTORCH_BUILD_FLATC)
     # 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()
@@ -528,12 +531,7 @@ if(NOT FLATC_EXECUTABLE)
   find_program(FLATC_EXECUTABLE flatc)
 
   if(NOT FLATC_EXECUTABLE)
-    message(
-      FATAL_ERROR
-        "FLATC_EXECUTABLE must be set when EXECUTORCH_BUILD_FLATC is disabled. "
-        "Note that EXECUTORCH_BUILD_FLATC may be disabled implicitly when "
-        "cross-compiling or when EXECUTORCH_BUILD_HOST_TARGETS is disabled."
-    )
+    message(FATAL_ERROR "FLATC_EXECUTABLE must be set when EXECUTORCH_BUILD_FLATC is disabled.")
   endif()
 endif()
 

backends/apple/coreml/scripts/build_tests.sh

@@ -32,7 +32,6 @@ cmake "$EXECUTORCH_ROOT_PATH" -B"$CMAKE_EXECUTORCH_BUILD_DIR_PATH" \
 -DCMAKE_TOOLCHAIN_FILE="$IOS_TOOLCHAIN_PATH" \
 -DPLATFORM=MAC_UNIVERSAL \
 -DDEPLOYMENT_TARGET=13.0 \
--DFLATC_EXECUTABLE="$(which flatc)" \
 -DEXECUTORCH_BUILD_EXECUTOR_RUNNER=OFF \
 -DEXECUTORCH_BUILD_XNNPACK=OFF \
 -DEXECUTORCH_BUILD_GFLAGS=OFF

backends/apple/mps/runtime/MPSDevice.mm

@@ -22,11 +22,11 @@ static inline MTLLanguageVersion getMetalLanguageVersion(const id<MTLDevice>& de
   // MPS Advanced Indexing needs at least Metal 2.0 (support for Argument Buffers and function constants)
   // host_name attribute needs at least Metal 2.2 and ulong needs Metal 2.3 (supported on MacOS 11+)
   MTLLanguageVersion languageVersion = MTLLanguageVersion2_3;
-#if defined(__MAC_13_0)
-  if (macOS13Plus) {
-    languageVersion = MTLLanguageVersion3_0;
+  if (@available(iOS 16, macOS 13, *)) {
+    if (macOS13Plus) {
+      languageVersion = MTLLanguageVersion3_0;
+    }
   }
-#endif
 
   ET_CHECK_MSG([device supportsFamily:MTLGPUFamilyMac2], "Missing Metal support for MTLGPUFamilyMac2");
   return languageVersion;

backends/apple/mps/runtime/operations/IndexingOps.mm

@@ -206,25 +206,32 @@
 
 Error
 MPSGraphBuilder::mpsScatterOp(NodePtr nodePtr) {
-  auto graphNode = nodePtr->mpsnode_union_as_MPSScatter();
-  ET_LOG(
-    Debug, "%s %d: %d",
-    __FUNCTION__, graphNode->input1_id(), graphNode->output_id()
-  );
+  if (@available(iOS 15.4, macOS 12.3, *)) {
+    auto graphNode = nodePtr->mpsnode_union_as_MPSScatter();
+    ET_LOG(
+      Debug, "%s %d: %d",
+      __FUNCTION__, graphNode->input1_id(), graphNode->output_id()
+    );
 
-  int64_t dim = graphNode->dim();
-  MPSGraphTensor* inputTensor = getMPSGraphTensor(graphNode->input1_id());
-  MPSGraphTensor* indicesTensor = getMPSGraphTensor(graphNode->idx_id());
-  MPSGraphTensor* updatesTensor = getMPSGraphTensor(graphNode->src_id());
+    int64_t dim = graphNode->dim();
+    MPSGraphTensor* inputTensor = getMPSGraphTensor(graphNode->input1_id());
+    MPSGraphTensor* indicesTensor = getMPSGraphTensor(graphNode->idx_id());
+    MPSGraphTensor* updatesTensor = getMPSGraphTensor(graphNode->src_id());
 
-  _idToMPSGraphTensor[graphNode->output_id()] =
-    [_mpsGraph scatterAlongAxis:dim
-                 withDataTensor:inputTensor
-                  updatesTensor:updatesTensor
-                  indicesTensor:indicesTensor
-                           mode:MPSGraphScatterModeSet
-                           name:nil];
-  return Error::Ok;
+    _idToMPSGraphTensor[graphNode->output_id()] =
+      [_mpsGraph scatterAlongAxis:dim
+                   withDataTensor:inputTensor
+                    updatesTensor:updatesTensor
+                    indicesTensor:indicesTensor
+                             mode:MPSGraphScatterModeSet
+                             name:nil];
+
+    return Error::Ok;
+  } else {
+    ET_LOG(Error, "MPS: scatter op is not supported on iOS < 15.4 and macOS < 12.3");
+
+    return Error::NotSupported;
+  }
 }
 
 

backends/arm/scripts/build_executorch.sh

@@ -84,7 +84,6 @@ if [ "$build_with_etdump" = true ] ; then
         -DEXECUTORCH_ENABLE_EVENT_TRACER=ON               \
         -DEXECUTORCH_SEPARATE_FLATCC_HOST_PROJECT=ON      \
         -DFLATCC_ALLOW_WERROR=OFF                         \
-        -DFLATC_EXECUTABLE="$(which flatc)"               \
         -B"${et_build_host_dir}"                          \
         "${et_root_dir}"
 
@@ -133,7 +132,6 @@ cmake                                                 \
     -DEXECUTORCH_ENABLE_LOGGING=ON                    \
     ${build_devtools_flags}                           \
     ${build_with_etdump_flags}                        \
-    -DFLATC_EXECUTABLE="$(which flatc)"               \
     -B"${et_build_dir}"                               \
     "${et_root_dir}"
 

backends/arm/test/TARGETS

@@ -19,7 +19,6 @@ python_library(
     srcs = ["runner_utils.py"],
     deps = [
         ":conftest",
-        "//executorch/backends/xnnpack/test/tester:tester",
         "//executorch/backends/arm:arm_backend",
         "//executorch/exir:lib",
         "//executorch/exir/backend:compile_spec_schema",
@@ -31,10 +30,7 @@ python_library(
     srcs = ["common.py"],
     deps = [
         ":runner_utils",
-        "//executorch/backends/xnnpack/test/tester:tester",
-        "//executorch/backends/arm:arm_backend",
-        "//executorch/exir:lib",
-        "//executorch/exir/backend:compile_spec_schema",
+        "//executorch/backends/arm:tosa_specification",
         "fbsource//third-party/pypi/pytest:pytest",
     ]
 )
@@ -43,12 +39,11 @@ python_library(
     name = "arm_tester",
     srcs = glob(["tester/*.py"]),
     deps = [
-        ":runner_utils",
         ":common",
-        "//executorch/backends/arm:tosa_mapping",
-        "//executorch/backends/arm:tosa_specification",
-        "//executorch/backends/arm/quantizer:arm_quantizer",
+        "//executorch/backends/xnnpack/test/tester:tester",
         "//executorch/backends/arm:arm_partitioner",
+        "//executorch/backends/arm/quantizer:arm_quantizer",
+        "//executorch/backends/arm:tosa_mapping",
         "//executorch/devtools/backend_debug:delegation_info",
         "fbsource//third-party/pypi/tabulate:tabulate",
     ]

backends/arm/test/common.py

@@ -10,7 +10,7 @@
 from datetime import datetime
 
 from pathlib import Path
-from typing import Any
+from typing import Any, Optional
 
 import pytest
 from executorch.backends.arm.arm_backend import ArmCompileSpecBuilder
@@ -91,63 +91,96 @@ def get_tosa_compile_spec_unbuilt(
 
 
 def get_u55_compile_spec(
-    custom_path=None,
+    macs: int = 128,
+    system_config: str = "Ethos_U55_High_End_Embedded",
+    memory_mode: str = "Shared_Sram",
+    extra_flags: str = "--debug-force-regor --output-format=raw",
+    custom_path: Optional[str] = None,
 ) -> list[CompileSpec]:
     """
-    Default compile spec for Ethos-U55 tests.
+    Compile spec for Ethos-U55.
     """
     return get_u55_compile_spec_unbuilt(
+        macs=macs,
+        system_config=system_config,
+        memory_mode=memory_mode,
+        extra_flags=extra_flags,
         custom_path=custom_path,
     ).build()
 
 
 def get_u85_compile_spec(
+    macs: int = 128,
+    system_config="Ethos_U85_SYS_DRAM_Mid",
+    memory_mode="Shared_Sram",
+    extra_flags="--output-format=raw",
     custom_path=None,
 ) -> list[CompileSpec]:
     """
-    Default compile spec for Ethos-U85 tests.
+    Compile spec for Ethos-U85.
     """
     return get_u85_compile_spec_unbuilt(  # type: ignore[attr-defined]
+        macs=macs,
+        system_config=system_config,
+        memory_mode=memory_mode,
+        extra_flags=extra_flags,
         custom_path=custom_path,
     ).build()
 
 
 def get_u55_compile_spec_unbuilt(
-    custom_path=None,
+    macs: int,
+    system_config: str,
+    memory_mode: str,
+    extra_flags: str,
+    custom_path: Optional[str],
 ) -> ArmCompileSpecBuilder:
     """Get the ArmCompileSpecBuilder for the Ethos-U55 tests, to modify
     the compile spec before calling .build() to finalize it.
     """
     artifact_path = custom_path or tempfile.mkdtemp(prefix="arm_u55_")
     if not os.path.exists(artifact_path):
         os.makedirs(artifact_path, exist_ok=True)
+
+    # https://gitlab.arm.com/artificial-intelligence/ethos-u/ethos-u-vela/-/blob/main/OPTIONS.md
+    assert macs in [32, 64, 128, 256], "Unsupported MACs value"
+
     compile_spec = (
         ArmCompileSpecBuilder()
         .ethosu_compile_spec(
-            "ethos-u55-128",
-            system_config="Ethos_U55_High_End_Embedded",
-            memory_mode="Shared_Sram",
-            extra_flags="--debug-force-regor --output-format=raw",
+            f"ethos-u55-{macs}",
+            system_config=system_config,
+            memory_mode=memory_mode,
+            extra_flags=extra_flags,
         )
         .dump_intermediate_artifacts_to(artifact_path)
     )
     return compile_spec
 
 
 def get_u85_compile_spec_unbuilt(
-    custom_path=None,
+    macs: int,
+    system_config: str,
+    memory_mode: str,
+    extra_flags: str,
+    custom_path: Optional[str],
 ) -> list[CompileSpec]:
     """Get the ArmCompileSpecBuilder for the Ethos-U85 tests, to modify
     the compile spec before calling .build() to finalize it.
     """
     artifact_path = custom_path or tempfile.mkdtemp(prefix="arm_u85_")
+    if not os.path.exists(artifact_path):
+        os.makedirs(artifact_path, exist_ok=True)
+
+    assert macs in [128, 256, 512, 1024, 2048], "Unsupported MACs value"
+
     compile_spec = (
         ArmCompileSpecBuilder()
         .ethosu_compile_spec(
-            "ethos-u85-128",
-            system_config="Ethos_U85_SYS_DRAM_Mid",
-            memory_mode="Shared_Sram",
-            extra_flags="--output-format=raw",
+            f"ethos-u85-{macs}",
+            system_config=system_config,
+            memory_mode=memory_mode,
+            extra_flags=extra_flags,
         )
         .dump_intermediate_artifacts_to(artifact_path)
     )

backends/arm/test/passes/test_rescale_pass.py

@@ -13,7 +13,6 @@
 from executorch.backends.arm.test import common, conftest
 from executorch.backends.arm.test.tester.arm_tester import ArmTester
 from parameterized import parameterized
-from torch.testing._internal import optests
 
 
 def test_rescale_op():
@@ -64,7 +63,7 @@ def test_nonzero_zp_for_int32():
         ),
     ]
     for sample_input in sample_inputs:
-        with pytest.raises(optests.generate_tests.OpCheckError):
+        with pytest.raises(Exception, match="opcheck"):
             torch.library.opcheck(torch.ops.tosa._rescale, sample_input)
 
 
@@ -87,7 +86,7 @@ def test_zp_outside_range():
         ),
     ]
     for sample_input in sample_inputs:
-        with pytest.raises(optests.generate_tests.OpCheckError):
+        with pytest.raises(Exception, match="opcheck"):
             torch.library.opcheck(torch.ops.tosa._rescale, sample_input)