https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96793
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |glisse at gcc dot gnu.org
--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Uroš Bizjak from comment #7)
> Created attachment 49144 [details]
> Proposed patch
>
> Patch in testing.
OTOH we _do_ try to compensate for HONOR_SIGNED_ZEROS so I wonder whether
we really need to punt for -frounding-math? I also doubt glibc gets it
correct, does it?
I guess
/* Compensate: xa = xa - (xa > operand1 ? 1 : 0) */
tmp = ix86_expand_sse_compare_mask (UNGT, xa, res, !do_floor);
emit_insn (gen_rtx_SET (tmp, gen_rtx_AND (mode, one, tmp)));
tmp = expand_simple_binop (mode, do_floor ? MINUS : PLUS,
xa, tmp, NULL_RTX, 0, OPTAB_DIRECT);
might be in error for FE_DOWNWARD if 0 - 0 behaves differently depending
on rounding mode... or the compensation is off
if (HONOR_SIGNED_ZEROS (mode))
ix86_sse_copysign_to_positive (res, res, force_reg (mode, operand1), mask);
