https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62262
--- Comment #9 from amker at gcc dot gnu.org --- (In reply to Andrew Pinski from comment #6) > (In reply to Carrot from comment #5) > > Actually the original source code is guarded by assert, and the parameter > > passed to CLZ can be guaranteed not 0, so "value <<= zeros" is well defined > > in our original source code. > > The issue is the second CLZ (not the first one). Though if I read the code > correctly value<<1 can never be zero but GCC does not optimize away that > check. How is that? Both "value >> bias" and "value << 1" are zero if value == 1 at the entry, am I misunderstanding? --- Comment #10 from carrot at gcc dot gnu.org --- Author: carrot Date: Wed Aug 27 16:48:09 2014 New Revision: 214578 URL: https://gcc.gnu.org/viewcvs?rev=214578&root=gcc&view=rev Log: PR target/62262 * config/aarch64/aarch64.md (*andim_ashift<mode>_bfiz): Check the shift amount before using it. * gcc.target/aarch64/pr62262.c: New test. Added: trunk/gcc/testsuite/gcc.target/aarch64/pr62262.c Modified: trunk/gcc/ChangeLog trunk/gcc/config/aarch64/aarch64.md trunk/gcc/testsuite/ChangeLog
