Significant Performance Regression in Apple Clang 16 for Assembly File Processing

I've observed a significant performance regression in Apple Clang 16 (Xcode 16.0/16.2) compared to Clang 15 (Xcode 15.2) when processing flutter aot compilation. Further research shows that clang -cc1as process became extremely slow. The compilation time has increased by approximately 4x.

Environment

• Machine: Apple M2 (8C8T)
• Memory: 16GB
• macOS Version: 14.7.2
• Target: Flutter AOT compilation (snapshot_assembly.o)

Performance Comparison

Reproduction Steps

The issue can be reproduced with the following command which is generated by flutter aot_assembly_release process:

time ${xcode_app_path}/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/cc \
-arch arm64 \
-miphoneos-version-min=12.0 \
-v \
-isysroot /Applications/Xcode.app/Contents/Developer/Platforms/iPhoneOS.platform/Developer/SDKs/iPhoneOS18.2.sdk \
-c ${project_path}/.dart_tool/flutter_build/f9ebf46f040933de7c8d103c84d38156/arm64/snapshot_assembly.S \
-o ${project_path}/.dart_tool/flutter_build/f9ebf46f040933de7c8d103c84d38156/arm64/snapshot_assembly.o

Additional Information

• This issue specifically affects large assembly files generated by Flutter's AOT compilation
• The performance regression appears to be consistent across different iOS SDK versions
• The same assembly file compiles significantly faster with Xcode 15.2
• Same performance regression observed on M4 Mac mini, suggesting this is not hardware-specific
• Size of object:

size -m  ${project_path}/.dart_tool/flutter_build/f9ebf46f040933de7c8d103c84d38156/arm64/snapshot_assembly.o
Segment : 64577616
	Section (__TEXT, __text): 26603344
	Section (__DATA, __bss): 48 (zerofill)
	Section (__TEXT, __const): 21292928
	Section (__DWARF, __debug_abbrev): 61
	Section (__DWARF, __debug_info): 8934534
	Section (__DWARF, __debug_line): 4464443
	Section (__LD, __compact_unwind): 3282208
	total 64577566
total 64577616

Questions

1. Is this a known issue with Apple Clang 16?
2. Are there any workarounds or compiler flags we can use to improve the performance?
3. Is this behavior expected or should it be considered a regression?

Any insights or suggestions would be greatly appreciated.