Build optimized operators lib with -fexceptions

Pull Request resolved: #7546

It should be permissible for this library to use exceptions, since it is not required to supported embedded systems.

TODO: presumably we need to manage rollout of the torchgen patch?

ghstack-source-id: 260556841
@exported-using-ghexport

Differential Revision: [D67904052](https://our.internmc.facebook.com/intern/diff/D67904052/)

.ci/docker/ci_commit_pins/pytorch.txt

@@ -1 +1 @@
-2ea4b56ec872424e486c4fe2d55da061067a2ed3
+f0c50a619c282a01ef1ad33690b7ab5e615d9fef

build/Codegen.cmake

@@ -11,13 +11,14 @@
 # arguments.
 function(gen_selected_ops)
   set(arg_names LIB_NAME OPS_SCHEMA_YAML ROOT_OPS INCLUDE_ALL_OPS)
-  cmake_parse_arguments(GEN "" "" "${arg_names}" ${ARGN})
+  cmake_parse_arguments(GEN "${options}" "" "${arg_names}" ${ARGN})
 
   message(STATUS "Generating operator lib:")
   message(STATUS "  LIB_NAME: ${GEN_LIB_NAME}")
   message(STATUS "  OPS_SCHEMA_YAML: ${GEN_OPS_SCHEMA_YAML}")
   message(STATUS "  ROOT_OPS: ${GEN_ROOT_OPS}")
   message(STATUS "  INCLUDE_ALL_OPS: ${GEN_INCLUDE_ALL_OPS}")
+  message(STATUS "  ADD_EXCEPTION_BOUNDARY: ${GEN_ADD_EXCEPTION_BOUNDARTY}")
 
   set(_oplist_yaml
       ${CMAKE_CURRENT_BINARY_DIR}/${GEN_LIB_NAME}/selected_operators.yaml
@@ -59,13 +60,15 @@ endfunction()
 # Invoked as generate_bindings_for_kernels( LIB_NAME lib_name FUNCTIONS_YAML
 # functions_yaml CUSTOM_OPS_YAML custom_ops_yaml )
 function(generate_bindings_for_kernels)
+  set(options ADD_EXCEPTION_BOUNDARY)
   set(arg_names LIB_NAME FUNCTIONS_YAML CUSTOM_OPS_YAML)
-  cmake_parse_arguments(GEN "" "${arg_names}" "" ${ARGN})
+  cmake_parse_arguments(GEN "${options}" "${arg_names}" "" ${ARGN})
 
   message(STATUS "Generating kernel bindings:")
   message(STATUS "  LIB_NAME: ${GEN_LIB_NAME}")
   message(STATUS "  FUNCTIONS_YAML: ${GEN_FUNCTIONS_YAML}")
   message(STATUS "  CUSTOM_OPS_YAML: ${GEN_CUSTOM_OPS_YAML}")
+  message(STATUS "  ADD_EXCEPTION_BOUNDARY: ${GEN_ADD_EXCEPTION_BOUNDARY}")
 
   # Command to generate selected_operators.yaml from custom_ops.yaml.
   file(GLOB_RECURSE _codegen_templates "${EXECUTORCH_ROOT}/codegen/templates/*")
@@ -93,7 +96,10 @@ function(generate_bindings_for_kernels)
       --tags-path=${site-packages-out}/torchgen/packaged/ATen/native/tags.yaml
       --aten-yaml-path=${site-packages-out}/torchgen/packaged/ATen/native/native_functions.yaml
       --op-selection-yaml-path=${_oplist_yaml}
-  )
+    )
+  if(GEN_ADD_EXCEPTION_BOUNDARY)
+    set(_gen_command "${_gen_command} --add-exception-boundary")
+  endif()
 
   set(_gen_command_sources
       ${_out_dir}/RegisterCodegenUnboxedKernelsEverything.cpp

configurations/CMakeLists.txt

@@ -42,7 +42,7 @@ if(EXECUTORCH_BUILD_KERNELS_OPTIMIZED)
 
   generate_bindings_for_kernels(
     LIB_NAME "optimized_native_cpu_ops_lib" FUNCTIONS_YAML
-    ${CMAKE_CURRENT_BINARY_DIR}/merged.yaml
+    ${CMAKE_CURRENT_BINARY_DIR}/merged.yaml ADD_EXCEPTION_BOUNDARY
   )
   message("Generated files ${gen_command_sources}")
 
@@ -56,6 +56,7 @@ if(EXECUTORCH_BUILD_KERNELS_OPTIMIZED)
     DEPS
     executorch
   )
+  target_compile_options(optimized_native_cpu_ops_lib PRIVATE "-fexceptions")
 
   install(TARGETS optimized_native_cpu_ops_lib DESTINATION lib)
 endif()

kernels/optimized/CMakeLists.txt

@@ -23,6 +23,8 @@ endif()
 
 set(_common_compile_options -Wno-deprecated-declarations)
 
+add_compile_options("-fexceptions")
+
 # Note for apple platform we can rely on Accelerate framework Will come back to
 # this
 include(${CMAKE_CURRENT_LIST_DIR}/External/EigenBLAS.cmake)
@@ -55,6 +57,7 @@ gen_selected_ops(LIB_NAME "optimized_ops_lib" OPS_SCHEMA_YAML "${_yaml}")
 generate_bindings_for_kernels(
   LIB_NAME "optimized_ops_lib" FUNCTIONS_YAML
   ${CMAKE_CURRENT_SOURCE_DIR}/optimized-oss.yaml
+  ADD_EXCEPTION_BOUNDARY
 )
 message("Generated files ${gen_command_sources}")