https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114365
--- Comment #2 from GCC Commits <cvs-commit at gcc dot gnu.org> --- The master branch has been updated by Jakub Jelinek <ja...@gcc.gnu.org>: https://gcc.gnu.org/g:456e10f28b36aa417e0db145556831c4f979fbd7 commit r14-9570-g456e10f28b36aa417e0db145556831c4f979fbd7 Author: Jakub Jelinek <ja...@redhat.com> Date: Wed Mar 20 10:55:07 2024 +0100 bitint: Fix handling of conditional bitfield loads [PR114365] For the m_var_msb (aka left shift) case of large/huge _BitInt bitfield loads handle_load adds a PHI node, but I forgot to actually update the temporary the code later on uses, so the PHI result was unused and the code incorrectly used something that wasn't valid SSA form. In particular, we emitted if (_29 != 2) goto <bb 4>; [80.00%] else goto <bb 5>; [20.00%] <bb 4> [local count: 1073741824]: _33 = VIEW_CONVERT_EXPR<unsigned long[3]>(s.D.2771)[_31]; <bb 5> [local count: 1073741824]: # _34 = PHI <_33(4), 0(3)> _35 = _32 >> 31; _36 = _33 << 33; _37 = _36 | _35; _38 = _37 << _19; where instead of _33 the _36 def stmt should be using _34. Fixed thusly. 2024-03-20 Jakub Jelinek <ja...@redhat.com> PR tree-optimization/114365 * gimple-lower-bitint.cc (bitint_large_huge::handle_load): When adding a PHI node, set iv2 to its result afterwards. * gcc.dg/bitint-102.c: New test.
