We (Arm) intend to make -fveclib=ArmPL imply -fno-math-errno. By setting -fveclib the user shows they intend to use the vector math functions, which implies they don’t care about errno. However, currently, the vector mappings won’t be used in many cases without setting -fno-math-errno separately.
Note: We recently found some inconsistencies in how the mappings are applied (see [LAA] Don't assume libcalls with output/input pointers can be vectorized by MacDue · Pull Request #108980 · llvm/llvm-project · GitHub), and making this change would resolve them.
The question is whether this should be restricted to -fveclib=ArmPL or be implied for all values of -fveclib?