Build size_test binaries with debug info

.github/workflows/pull.yml

@@ -366,6 +366,7 @@ jobs:
         # build module for executorch.extension.pybindings.portable_lib
         bash test/build_size_test.sh
         strip cmake-out/test/size_test
+        strip --strip-debug cmake-out/test/size_test
         output=$(ls -la cmake-out/test/size_test)
         arr=($output)
         size=${arr[4]}
@@ -403,6 +404,7 @@ jobs:
         # build module for executorch.extension.pybindings.portable_lib
         bash test/build_size_test.sh
         strip cmake-out/test/size_test
+        strip --strip-debug cmake-out/test/size_test
         output=$(ls -la cmake-out/test/size_test)
         arr=($output)
         size=${arr[4]}

CMakeLists.txt

@@ -694,7 +694,7 @@ if(EXECUTORCH_BUILD_EXECUTOR_RUNNER)
   endif()
 
   add_executable(executor_runner ${_executor_runner__srcs})
-  if(CMAKE_BUILD_TYPE STREQUAL "Release")
+  if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
     if(APPLE)
       target_link_options(executor_runner PRIVATE "LINKER:-dead_strip")
     else()

backends/qualcomm/CMakeLists.txt

@@ -60,7 +60,7 @@ if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
   add_link_options("-flto=auto")
 endif()
 
-if(CMAKE_BUILD_TYPE STREQUAL "Release")
+if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
   # strip symbols
   add_link_options("-s")
 
@@ -259,7 +259,7 @@ if(${CMAKE_SYSTEM_PROCESSOR} MATCHES "x86_64")
     pybind11_strip(PyQnnWrapperAdaptor)
   endif()
 
-  if(CMAKE_BUILD_TYPE STREQUAL "Release")
+  if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
     # need to allow exceptions in pybind
     set(_pybind_compile_options -Wno-deprecated-declarations -fPIC -frtti
                                 -fexceptions

examples/models/llama/CMakeLists.txt

@@ -209,7 +209,7 @@ if(ANDROID)
 endif()
 
 add_executable(llama_main ${_srcs})
-if(CMAKE_BUILD_TYPE STREQUAL "Release")
+if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
   if(APPLE)
     target_link_options(llama_main PRIVATE "LINKER:-dead_strip")
   else()

examples/models/llava/CMakeLists.txt

@@ -197,7 +197,7 @@ list(APPEND _common_include_directories ${stb_SOURCE_DIR}
 )
 
 add_executable(llava_main ${_srcs})
-if(CMAKE_BUILD_TYPE STREQUAL "Release")
+if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
   if(APPLE)
     target_link_options(llava_main PRIVATE "LINKER:-dead_strip,-s")
   else()

examples/selective_build/CMakeLists.txt

@@ -151,7 +151,7 @@ list(TRANSFORM _executor_runner__srcs PREPEND "${EXECUTORCH_ROOT}/")
 # link to
 #
 add_executable(selective_build_test ${_executor_runner__srcs})
-if(CMAKE_BUILD_TYPE EQUAL "Release")
+if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
   target_link_options(selective_build_test PRIVATE "LINKER:--gc-sections")
 endif()
 target_link_libraries(

kernels/portable/cpu/util/normalization_ops_util.cpp

@@ -7,6 +7,7 @@
  */
 
 #include <c10/util/irange.h>
+#include <array>
 #include <cstring>
 
 #include <executorch/kernels/portable/cpu/util/normalization_ops_util.h>
@@ -92,6 +93,11 @@ bool check_layer_norm_args(
       ", ndim = %zu",
       in.dim(),
       ndim);
+  ET_CHECK_OR_RETURN_FALSE(
+      ndim <= kTensorDimensionLimit,
+      "Expected normalized shape to have at most %zu dimensions but it had %zu",
+      kTensorDimensionLimit,
+      ndim);
   size_t shift = in.dim() - ndim;
   for (const auto d : c10::irange(ndim)) {
     ET_CHECK_OR_RETURN_FALSE(
@@ -103,20 +109,20 @@ bool check_layer_norm_args(
         d,
         normalized_shape[d]);
   }
-  executorch::aten::SizesType shape[ndim];
+  std::array<executorch::aten::SizesType, kTensorDimensionLimit> shape;
   for (const auto i : c10::irange(ndim)) {
     shape[i] = static_cast<executorch::aten::SizesType>(normalized_shape[i]);
   }
 
   if (weight.has_value()) {
     ET_LOG_AND_RETURN_IF_FALSE(tensors_have_same_dtype(in, weight.value()));
     ET_LOG_AND_RETURN_IF_FALSE(
-        tensor_has_expected_size(weight.value(), {shape, ndim}));
+        tensor_has_expected_size(weight.value(), {shape.data(), ndim}));
   }
   if (bias.has_value()) {
     ET_LOG_AND_RETURN_IF_FALSE(tensors_have_same_dtype(in, bias.value()));
     ET_LOG_AND_RETURN_IF_FALSE(
-        tensor_has_expected_size(bias.value(), {shape, ndim}));
+        tensor_has_expected_size(bias.value(), {shape.data(), ndim}));
   }
   ET_LOG_AND_RETURN_IF_FALSE(tensors_have_same_dtype(in, out));
   ET_LOG_AND_RETURN_IF_FALSE(tensors_have_same_dtype(in, mean_out));

test/CMakeLists.txt

@@ -52,7 +52,7 @@ list(TRANSFORM _size_test__srcs PREPEND "${EXECUTORCH_ROOT}/")
 # when we cross compile to ios
 add_executable(size_test ${_size_test__srcs})
 target_link_libraries(size_test executorch)
-if(CMAKE_BUILD_TYPE EQUAL "Release")
+if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
   target_link_options(size_test PRIVATE "LINKER:--gc-sections")
 endif()
 
@@ -64,7 +64,7 @@ target_link_options_shared_lib(portable_ops_lib)
 target_link_libraries(
   size_test_all_ops executorch portable_ops_lib portable_kernels
 )
-if(CMAKE_BUILD_TYPE EQUAL "Release")
+if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
   target_link_options(size_test_all_ops PRIVATE "LINKER:--gc-sections")
 endif()
 
@@ -76,7 +76,7 @@ add_executable(size_test_all_optimized_ops ${_size_test__srcs})
 target_link_options_shared_lib(optimized_native_cpu_ops_lib)
 target_link_libraries(
   size_test_all_optimized_ops executorch optimized_native_cpu_ops_lib)
-if(CMAKE_BUILD_TYPE EQUAL "Release")
+if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
   target_link_options(size_test_all_optimized_ops PRIVATE "LINKER:--gc-sections")
 endif()
 endif()

test/build_optimized_size_test.sh

@@ -41,12 +41,15 @@ test_cmake_size_test() {
 
     echo 'ExecuTorch with no ops binary size, unstripped:'
     ls -al cmake-out/test/size_test
+    size cmake-out/test/size_test
 
     echo 'ExecuTorch with portable ops binary size, unstripped:'
     ls -al cmake-out/test/size_test_all_ops
+    size cmake-out/test/size_test_all_ops
 
     echo 'ExecuTorch with optimized ops binary size, unstripped:'
     ls -al cmake-out/test/size_test_all_optimized_ops
+    size cmake-out/test/size_test_all_optimized_ops
 }
 
 if [[ -z $PYTHON_EXECUTABLE ]]; then

test/build_size_test.sh

@@ -15,7 +15,10 @@ EXTRA_BUILD_ARGS="${@:-}"
 # TODO(#8357): Remove -Wno-int-in-bool-context
 # TODO: Replace -ET_HAVE_PREAD=0 with a CMake option.
 #  FileDataLoader used in the size_test breaks baremetal builds with pread when missing.
-COMMON_CXXFLAGS="-fno-exceptions -fno-rtti -Wall -Werror -Wno-int-in-bool-context -DET_HAVE_PREAD=0"
+# -Werror=maybe-uninitialized causes failures with GCC 9.5, which we
+# -currently use in CI. There were known issues with this warning and
+# -std::optional: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80635
+COMMON_CXXFLAGS="-fno-exceptions -fno-rtti -Wall -Werror -Wno-int-in-bool-context -Wno-maybe-uninitialized -DET_HAVE_PREAD=0"
 
 cmake_install_executorch_lib() {
   echo "Installing libexecutorch.a"
@@ -26,29 +29,31 @@ cmake_install_executorch_lib() {
   CXXFLAGS="$COMMON_CXXFLAGS" retry cmake -DBUCK2="$BUCK2" \
           -DCMAKE_CXX_STANDARD_REQUIRED=ON \
           -DCMAKE_INSTALL_PREFIX=cmake-out \
-          -DCMAKE_BUILD_TYPE=Release \
+          -DCMAKE_BUILD_TYPE=RelWithDebInfo \
           -DEXECUTORCH_BUILD_EXECUTOR_RUNNER=OFF \
           -DEXECUTORCH_OPTIMIZE_SIZE=ON \
           -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
           ${EXTRA_BUILD_ARGS} \
           -Bcmake-out .
-  cmake --build cmake-out -j9 --target install --config Release
+  cmake --build cmake-out -j9 --target install --config RelWithDebInfo
 }
 
 test_cmake_size_test() {
-    CXXFLAGS="$COMMON_CXXFLAGS" retry cmake -DCMAKE_BUILD_TYPE=Release \
+    CXXFLAGS="$COMMON_CXXFLAGS" retry cmake -DCMAKE_BUILD_TYPE=RelWithDebInfo \
         -DCMAKE_INSTALL_PREFIX=cmake-out \
         ${EXTRA_BUILD_ARGS} \
         -Bcmake-out/test test
 
     echo "Build size test"
-    cmake --build cmake-out/test -j9 --config Release
+    cmake --build cmake-out/test -j9 --config RelWithDebInfo
 
     echo 'ExecuTorch with no ops binary size, unstripped:'
     ls -al cmake-out/test/size_test
+    size cmake-out/test/size_test
 
     echo 'ExecuTorch with portable ops binary size, unstripped:'
     ls -al cmake-out/test/size_test_all_ops
+    size cmake-out/test/size_test_all_ops
 }
 
 if [[ -z $PYTHON_EXECUTABLE ]]; then

tools/cmake/preset/default.cmake

@@ -4,7 +4,7 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
-if(CMAKE_BUILD_TYPE STREQUAL "Release")
+if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug")
   set(_is_build_type_release ON)
   set(_is_build_type_debug OFF)
 else()