Testing, mv2.pte, seems like yaml does not include all the needed dtypes

Author: BujSet

examples/selective_build/CMakeLists.txt

@@ -33,7 +33,7 @@ if(NOT CMAKE_CXX_STANDARD)
   # Can't set to 11 due to executor_runner.cpp make_unique
 endif()
 
-set(_common_compile_options -Wno-deprecated-declarations -fPIC)
+set(_common_compile_options -Wno-deprecated-declarations -fPIC -ffunction-sections -fdata-sections)
 
 # Let files say "include <executorch/path/to/header.h>".
 set(_common_include_directories ${EXECUTORCH_ROOT}/..)
@@ -129,7 +129,14 @@ generate_bindings_for_kernels(
 )
 
 gen_operators_lib(
-  LIB_NAME "select_build_lib" KERNEL_LIBS ${_kernel_lib} DEPS executorch_core
+  LIB_NAME 
+  "select_build_lib" 
+  KERNEL_LIBS 
+  ${_kernel_lib} 
+  DEPS 
+  executorch_core
+  DTYPE_SELECTIVE_BUILD
+  "${EXECUTORCH_DTYPE_SELECTIVE_BUILD}"
 )
 
 list(TRANSFORM _executor_runner__srcs PREPEND "${EXECUTORCH_ROOT}/")

examples/selective_build/test_selective_build.sh

@@ -191,8 +191,10 @@ test_cmake_select_ops_in_model() {
     local example_dir=examples/selective_build
     local build_dir=cmake-out/${example_dir}
     rm -rf ${build_dir}
-    retry cmake -DCMAKE_BUILD_TYPE=$CMAKE_BUILD_TYPE \
+    retry cmake -DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" \
             -DEXECUTORCH_SELECT_OPS_FROM_MODEL="./mv2.pte" \
+            -DEXECUTORCH_DTYPE_SELECTIVE_BUILD=ON \
+            -DEXECUTORCH_OPTIMIZE_SIZE=ON \
             -DCMAKE_INSTALL_PREFIX=cmake-out \
             -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
             -B${build_dir} \
@@ -204,8 +206,8 @@ test_cmake_select_ops_in_model() {
     echo 'Running selective build test'
     ${build_dir}/selective_build_test --model_path="./mv2.pte"
 
-    echo "Removing mv2.pte"
-    rm "./mv2.pte"
+#    echo "Removing mv2.pte"
+#    rm "./mv2.pte"
 }
 
 if [[ -z $BUCK ]];

kernels/portable/cpu/selective_build.h

@@ -20,34 +20,42 @@ inline constexpr bool should_include_kernel_dtype(
   const char *operator_name,
   executorch::aten::ScalarType scalar_type
 ) {
-  return ((executorch::aten::string_view(operator_name).compare("add.out") == 0)
-        && (scalar_type == executorch::aten::ScalarType::Float || scalar_type == executorch::aten::ScalarType::Int));
-// || ((executorch::aten::string_view(operator_name).compare("mul.out") == 0)
+//  return (((std::string_view(operator_name).compare("add.out") == 0))
 //        && (scalar_type == executorch::aten::ScalarType::Float));
-// || ((executorch::aten::string_view(operator_name).compare("sub.out") == 0)
-//        && (true));
-}
-
-//inline constexpr bool should_include_kernel_dtype(
-//  const char *operator_name,
-//  executorch::aten::ScalarType scalar_type
-//) {
 //  return ((executorch::aten::string_view(operator_name).compare("add.out") == 0)
 //        && (true))
 // || ((executorch::aten::string_view(operator_name).compare("mm.out") == 0)
 //        && (true));
 //}
+
+
+  return ((std::string_view(operator_name).compare("_native_batch_norm_legit_no_training.out") == 0)
+        && (scalar_type == executorch::aten::ScalarType::Float))
+ || ((std::string_view(operator_name).compare("add.out") == 0)
+        && (scalar_type == executorch::aten::ScalarType::Float))
+ || ((std::string_view(operator_name).compare("addmm.out") == 0)
+        && (scalar_type == executorch::aten::ScalarType::Float))
+ || ((std::string_view(operator_name).compare("clone.out") == 0)
+        && (scalar_type == executorch::aten::ScalarType::Float))
+ || ((std::string_view(operator_name).compare("convolution.out") == 0)
+        && (scalar_type == executorch::aten::ScalarType::Float))
+ || ((std::string_view(operator_name).compare("hardtanh.out") == 0)
+        && (scalar_type == executorch::aten::ScalarType::Float))
+ //|| ((std::string_view(operator_name).compare("hardtanh.out") == 0)
+ //       && (scalar_type == executorch::aten::ScalarType::Double))
+ || ((std::string_view(operator_name).compare("mean.out") == 0)
+        && (scalar_type == executorch::aten::ScalarType::Float))
+ || ((std::string_view(operator_name).compare("mean_dim.out") == 0)
+        && (scalar_type == executorch::aten::ScalarType::Float))
+ || ((std::string_view(operator_name).compare("permute_copy.out") == 0)
+        && (scalar_type == executorch::aten::ScalarType::Float));
+}
 /*
 inline constexpr bool should_include_kernel_dtype(
   const char* ,//operator_name,
   executorch::aten::ScalarType //scalar_type*
 ) {
   return true;
-
-//  return ((executorch::aten::string_view(operator_name).compare("my_ops::mul3.out") == 0)
-//        && (true))
-// || ((executorch::aten::string_view(operator_name).compare("my_ops::mul4.out") == 0)
-//        && (true));
 }
 */
 #endif

test/build_size_test.sh

@@ -23,7 +23,7 @@ cmake_install_executorch_lib() {
   update_tokenizers_git_submodule
   local EXTRA_BUILD_ARGS="${@}"
 
-  CXXFLAGS="$COMMON_CXXFLAGS" retry cmake -DBUCK2="$BUCK2" \
+  CXXFLAGS="$COMMON_CXXFLAGS" retry cmake \
           -DCMAKE_CXX_STANDARD_REQUIRED=ON \
           -DCMAKE_INSTALL_PREFIX=cmake-out \
           -DCMAKE_BUILD_TYPE=Release \
@@ -44,8 +44,8 @@ test_cmake_size_test() {
     echo "Build size test"
     cmake --build cmake-out/test -j9 --config Release
 
-    strip cmake-out/test/size_test
-    strip cmake-out/test/size_test_all_ops
+    #cmake-out/test/size_test
+    #cmake-out/test/size_test_all_ops
     ls -lah cmake-out/test/
 }
 

tools/cmake/Codegen.cmake

@@ -193,22 +193,35 @@ endfunction()
 
 # Generate a runtime lib for registering operators in Executorch
 function(gen_operators_lib)
-  set(multi_arg_names LIB_NAME KERNEL_LIBS DEPS)
+  set(multi_arg_names LIB_NAME KERNEL_LIBS DEPS DTYPE_SELECTIVE_BUILD)
   cmake_parse_arguments(GEN "" "" "${multi_arg_names}" ${ARGN})
 
   message("REMOVE_ME-gen_op_lib")
   message(STATUS "Generating operator lib:")
   message(STATUS "  LIB_NAME: ${GEN_LIB_NAME}")
   message(STATUS "  KERNEL_LIBS: ${GEN_KERNEL_LIBS}")
   message(STATUS "  DEPS: ${GEN_DEPS}")
+  message(STATUS "  DTYPE_SELECTIVE_BUILD: ${GEN_DTYPE_SELECTIVE_BUILD}")
 
   set(_out_dir ${CMAKE_CURRENT_BINARY_DIR}/${GEN_LIB_NAME})
+  if(DTYPE_SELECTIVE_BUILD)
+    set(_opvariant_h
+      ${CMAKE_CURRENT_BINARY_DIR}/${GEN_LIB_NAME}/selected_op_variants.h
+    )
+  endif()
 
   add_library(${GEN_LIB_NAME})
+ 
+  set(_srcs_list
+    ${_out_dir}/RegisterCodegenUnboxedKernelsEverything.cpp
+    ${_out_dir}/Functions.h ${_out_dir}/NativeFunctions.h
+  )
+  if(DTYPE_SELECTIVE_BUILD)
+    list(APPEND _srcs_list ${_opvariant_h})
+  endif()
   target_sources(
     ${GEN_LIB_NAME}
-    PRIVATE ${_out_dir}/RegisterCodegenUnboxedKernelsEverything.cpp
-            ${_out_dir}/Functions.h ${_out_dir}/NativeFunctions.h
+    PRIVATE ${_srcs_list}
   )
   target_link_libraries(${GEN_LIB_NAME} PRIVATE ${GEN_DEPS})
   if(GEN_KERNEL_LIBS)
@@ -217,6 +230,9 @@ function(gen_operators_lib)
 
   target_link_options_shared_lib(${GEN_LIB_NAME})
   set(_generated_headers ${_out_dir}/Functions.h ${_out_dir}/NativeFunctions.h)
+  if(DTYPE_SELECTIVE_BUILD)
+    list(APPEND _generated_headers ${_opvariant_h})
+  endif()
   set_target_properties(
     ${GEN_LIB_NAME} PROPERTIES PUBLIC_HEADER "${_generated_headers}"
   )