On Wed, Apr 03, 2002 at 08:22:49PM +0200, Jakub Jelinek wrote: > Is this testcase actually valid (using "r" constraints for > QImode data, not "q")? No. r~