[Mlir-commits] [compiler-rt] [llvm] [libcxx] [mlir] [clang] [asan] Enable StackSafetyAnalysis by default (PR #77210)

[Fangrui Song]

MaskRay wrote:

> I guess we we should avoid ClUseStackSafety for -O0

There may be two competing goals. The traditional one that `-O0` should run very few optimizations and a practical one that instrumentations should reduce code size increase.

`[asan] Skip promotable allocas to improve performance at -O0` runs at `-O0`. (#77221 is discussing whether it should run at `-O0`). We have other optimizations e.g. within a basic block, two addresses do not need to be instrumented repeatedly unless they are separated by a call instruction.

Perhaps it isn't too bad to run StackSafetyAnalysis at -O0 for consistent with other -O levels?

https://github.com/llvm/llvm-project/pull/77210