[libc-commits] [libc] [libc][math] Fix double-rounding errors with denormal inputs for hypot function. (PR #208247)

[via libc-commits]

zimmermann6 wrote:

after the change back to underflow before rounding, I now get with CORE-MATH 5a64c7b:
```
zimmerma at croissant:~/svn/core-math$ EXTRA_CFLAGS=-DCORE_MATH_UNDERFLOW_BEFORE CORE_MATH_CHECK_STD=true LIBM=$L ./check.sh hypot
Running worst cases check in --rndn mode...
Spurious invalid exception for x,y=nan,nan (z=nan)
```
and
```
Running worst cases check in --rndz mode...
Missing underflow exception for x,y=-0x0.fffffffffffffp-1022,0x0.0000000000001p-1022 (z=0x0.fffffffffffffp-1022)
```

https://github.com/llvm/llvm-project/pull/208247