From: Richard Guenther <rguent...@suse.de> Date: Fri, 18 May 2012 11:48:55 +0200 (CEST)
> On Thu, 17 May 2012, David Miller wrote: > >> >> Richard, I was looking into a testsuite failure on the 4.7 branch >> on sparc and I think your fix for 52584 would fix it too. >> >> The problem eminates in gcc.c-torture/execute/vector-shift2.c with -O0 >> >> The 4 integer vector shifts get lowered to 2 integer vector shifts, >> since that is what the VIS3 shifts are capable of. >> >> However, type_for_widest_vector_mode does not take the signedness into >> consideration. So: >> >> signed_vector >> unsigned_vector >> >> gets lowered into: >> >> unsigned_vector >> unsigned_vector >> >> since type_for_widest_vector_mode unconditionally always passes >> '1' for "unsignedp", and thus the testcase aborts since we use >> a logical shift instead of an arithmetic one. >> >> I came up with the crude fix below, but then I noticed in mainline >> the change you made in order to resolve 52584. > > Backporting that change is fine with me if you can take care to > do the bootstrap & testing on the 4.7 branch. Great, I'll start doing that right now.
