https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128237
--- Comment #4 from GCC Commits <cvs-commit at gcc dot gnu.org> --- The master branch has been updated by Richard Biener <[email protected]>: https://gcc.gnu.org/g:465c1ac0dddb92d38be6b602d1732478cc16e7ae commit r17-5918-g465c1ac0dddb92d38be6b602d1732478cc16e7ae Author: Richard Biener <[email protected]> Date: Tue Nov 3 15:34:16 2026 +0100 tree-optimization/128398 - hosting of trapping FP comparisons The following fixes vn_nary_may_trap to properly handle FP comparisons and also conversions from float to integer as PR128237 notes. It adjusts gimple_could_trap_p_1 accordingly. The functions still differ in their default handling of FP expressions not explicitly handled in operation_could_trap_helper_p which I believe is a bit of an unfortunate interface erroring on the "wrong" side for both users. This should be addressed separately. This does not address the core issue of PR128237 with regard to FP to integer conversion builtins. PR tree-optimization/128398 PR tree-optimization/128237 * tree-ssa-sccvn.cc (vn_nary_may_trap): Set fp_operation correctly for comparisons and FIX_TRUNC_EXPR. Also do not trap for VEC_COND_EXPR. * gimple.cc (gimple_could_trap_p_1): Likewise for FIX_TRUNC_EXPR and VEC_COND_EXPR. * gcc.dg/torture/pr128398.c: New testcase.
