On Thu, 30 Jul 2015, Richard Biener wrote:
>
> The following patch avoids folding NaN < 0 to false if -ftrapping-math
> is in effect.
>
> Bootstrap & regtest pending on x86_64-unknown-linux-gnu.
FAIL: gcc.dg/fold-abs-4.c scan-tree-dump-times gimple "ABS" 0
FAIL: gcc.dg/pr15784-1.c scan-tree-dump-times gimple "ABS_EXPR" 0
FAIL: gcc.dg/tree-ssa/pr20139.c scan-tree-dump-times optimized
"link_error" 0
all cases expect the transform to happen without -fno-trapping-math.
Shall I adjust those testcases or is my patch wrong?
Thanks,
Richard.
> Richard.
>
> 2015-07-30 Richard Biener <rguent...@suse.de>
>
> PR middle-end/67052
> * fold-const.c (fold_binary_loc): Do not fold NaN < 0 to false
> if FP math can trap.
>
> Index: gcc/fold-const.c
> ===================================================================
> --- gcc/fold-const.c (revision 226387)
> +++ gcc/fold-const.c (working copy)
> @@ -11634,7 +11455,9 @@ fold_binary_loc (location_t loc,
> /* Convert ABS_EXPR<x> < 0 to false. */
> strict_overflow_p = false;
> if (code == LT_EXPR
> - && (integer_zerop (arg1) || real_zerop (arg1))
> + && (integer_zerop (arg1)
> + || ((! HONOR_NANS (arg0) || !flag_trapping_math)
> + && real_zerop (arg1)))
> && tree_expr_nonnegative_warnv_p (arg0, &strict_overflow_p))
> {
> if (strict_overflow_p)
