Update on "[ExecuTorch] Add broadcasting support to optimized op_div"

Summary:
Similar to broadcast support in op_mul

Test Plan:
Tests added

Reviewers:

Subscribers:

Tasks:

Tags:

cc larryliu0820 manuelcandales

Differential Revision: [D69491815](https://our.internmc.facebook.com/intern/diff/D69491815)

[ghstack-poisoned]

Author: kimishpatel

CMakeLists.txt

@@ -245,7 +245,7 @@ cmake_dependent_option(
 )
 
 if(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR)
-  set(EXECUTORCH_BUILF_EXTENSION_DATA_LOADER ON)
+  set(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER ON)
 endif()
 
 if(EXECUTORCH_BUILD_EXTENSION_TRAINING)
@@ -348,6 +348,7 @@ if(EXECUTORCH_BUILD_PTHREADPOOL)
 endif()
 
 if(EXECUTORCH_BUILD_TESTS)
+  set(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR ON)
   include(CTest)
 endif()
 
@@ -373,7 +374,7 @@ if(NOT "${_repo_dir_name}" STREQUAL "executorch")
       "fix for this restriction."
   )
 endif()
-set(_common_include_directories ${CMAKE_CURRENT_SOURCE_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/runtime/core/portable_type)
+set(_common_include_directories ${CMAKE_CURRENT_SOURCE_DIR}/.. ${CMAKE_CURRENT_SOURCE_DIR}/runtime/core/portable_type/c10)
 
 #
 # The `_<target>_srcs` lists are defined by including ${EXECUTORCH_SRCS_FILE}.

backends/apple/coreml/CMakeLists.txt

@@ -134,7 +134,7 @@ target_include_directories(
   coremldelegate PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/runtime/util
 )
 target_include_directories(coremldelegate PRIVATE ${EXECUTORCH_ROOT}/..)
-target_include_directories(coremldelegate PRIVATE ${EXECUTORCH_ROOT}/runtime/core/portable_type)
+target_include_directories(coremldelegate PRIVATE ${EXECUTORCH_ROOT}/runtime/core/portable_type/c10)
 target_compile_definitions(coremldelegate PRIVATE C10_USING_CUSTOM_GENERATED_MACROS)
 target_link_libraries(coremldelegate PRIVATE executorch_core)
 

backends/apple/coreml/runtime/workspace/executorchcoreml.xcodeproj/project.pbxproj

@@ -922,7 +922,7 @@
 					"$(SRCROOT)/../kvstore",
 					"$(SRCROOT)/../inmemoryfs",
 					"$(SRCROOT)/../include",
-					"$(SRCROOT)/../include/executorch/runtime/core/portable_type",
+					"$(SRCROOT)/../include/executorch/runtime/core/portable_type/c10",
 					"$(SRCROOT)/../sdk",
 					"$(SRCROOT)/../util",
 					"$(SRCROOT)/../../third-party/nlohmann_json/single_include",
@@ -954,7 +954,7 @@
 					"$(SRCROOT)/../kvstore",
 					"$(SRCROOT)/../inmemoryfs",
 					"$(SRCROOT)/../include",
-					"$(SRCROOT)/../include/executorch/runtime/core/portable_type",
+					"$(SRCROOT)/../include/executorch/runtime/core/portable_type/c10",
 					"$(SRCROOT)/../sdk",
 					"$(SRCROOT)/../util",
 					"$(SRCROOT)/../../third-party/nlohmann_json/single_include",

backends/arm/CMakeLists.txt

@@ -14,7 +14,7 @@ endif()
 
 include(${EXECUTORCH_ROOT}/build/Utils.cmake)
 
-set(_common_include_directories ${EXECUTORCH_ROOT}/.. ${EXECUTORCH_ROOT}/runtime/core/portable_type)
+set(_common_include_directories ${EXECUTORCH_ROOT}/.. ${EXECUTORCH_ROOT}/runtime/core/portable_type/c10)
 add_compile_definitions(C10_USING_CUSTOM_GENERATED_MACROS)
 
 # Third-party folder and Ethos-U driver inclued

backends/arm/TARGETS

@@ -4,7 +4,10 @@ load("@fbcode_macros//build_defs:python_library.bzl", "python_library")
 python_library(
     name = "arm_partitioner",
     srcs = [
-        "arm_partitioner.py",
+        "ethosu_backend.py",
+        "ethosu_partitioner.py",
+        "tosa_backend.py",
+        "tosa_partitioner.py",
     ],
     typing = True,
     deps = [

backends/arm/_passes/arm_pass_manager.py

@@ -52,7 +52,6 @@
 from executorch.backends.arm._passes.fuse_quantized_activation_pass import (  # type: ignore[import-not-found]
     FuseQuantizedActivationPass,
 )
-from executorch.backends.arm._passes.insert_rescales_pass import InsertRescalePass
 from executorch.backends.arm._passes.insert_table_ops import InsertTableOpsPass
 from executorch.backends.arm._passes.keep_dims_false_to_squeeze_pass import (
     KeepDimsFalseToSqueezePass,
@@ -76,7 +75,6 @@
     UnsqueezeScalarPlaceholdersPass,
 )
 from executorch.backends.arm.tosa_specification import TosaSpecification
-
 from executorch.backends.xnnpack._passes.remove_getitem_op import RemoveGetItemPass
 from executorch.exir import ExportedProgram
 from executorch.exir.pass_manager import PassManager
@@ -121,7 +119,7 @@ def _tosa_080_BI_pipeline(self, exported_program: ExportedProgram) -> GraphModul
         self.add_pass(ConvertSqueezesToViewPass())
 
         self.add_pass(AnnotateChannelsLastDimOrder())
-        self.add_pass(InsertRescalePass())
+
         return self._transform(exported_program.graph_module)
 
     def _tosa_080_MI_pipeline(self, exported_program: ExportedProgram) -> GraphModule:
@@ -159,7 +157,6 @@ def _tosa_080_MI_pipeline(self, exported_program: ExportedProgram) -> GraphModul
         self.add_pass(ConvertSqueezesToViewPass())
 
         self.add_pass(AnnotateChannelsLastDimOrder())
-        self.add_pass(InsertRescalePass())
 
         return self._transform(exported_program.graph_module)
 

backends/arm/_passes/fold_qdq_with_annotated_qparams_pass.py

@@ -131,9 +131,6 @@ def call(self, graph_module: GraphModule) -> PassResult:
             n = cast(Node, n)
             if n.op != "call_function":
                 continue
-            # Don't fold chains of quant-ops into each other.
-            if n.target in (q_op, dq_op):
-                continue
 
             # Make sure we haven't already set qparams meta information on the node
             assert "input_qparams" not in n.meta.keys()

backends/arm/_passes/insert_rescales_pass.py

@@ -31,7 +31,6 @@
     op_reciprocal,
     op_relu,
     op_repeat,
-    op_rescale,
     op_rshift,
     op_rsqrt,
     op_sigmoid,