https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93346
Jakub Jelinek <jakub at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|NEW |ASSIGNED Assignee|unassigned at gcc dot gnu.org |jakub at gcc dot gnu.org --- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Created attachment 47694 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=47694&action=edit gcc10-pr93346.patch While the complicated pattern is needed to properly model the semantics of the instruction for all values of the last operand (something needed e.g. for the intrinsic), the patterns in the testcase have undefined behavior unless the shift count is valid (0 to prec-1) and there is no reason to match them as bzhi that way.