ahesham-arm wrote:
> This looks like a brute force fixup (and also passes for me as-is). Can you
> more directly avoid this being a problem in the upstream computation?
@arsenm I wasn't too sure about fixing it in a function shared with `tan`
because I see that `__clc_tanpi` does the same for 1 and 2:
```
__CLC_S_GENTYPE flip = (n_1_or_2 && (reduced == __CLC_FP_LIT(0.0)))
? __CLC_GENTYPE_S_SIGNBIT
: (__CLC_S_GENTYPE)0;
__CLC_S_GENTYPE result = (__CLC_AS_S_GENTYPE(t) ^ flip) ^
(__CLC_AS_S_GENTYPE(x) & __CLC_GENTYPE_S_SIGNBIT);
```
https://github.com/llvm/llvm-project/pull/199974
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
