------- Comment #2 from ubizjak at gmail dot com 2009-04-20 20:52 ------- (In reply to comment #1)
> but the widening unpacking results in absymal code generated. Where are > all the shifts coming from? Not from unpacking, but from mulv2di pattern from sse.md Can you please attach full source to create executable testcase? IIRC, execution times depend on target processor, and perhaps vect cost should be updated for this case. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39821