http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43434
Richard Biener <rguenth at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Status|UNCONFIRMED |ASSIGNED Last reconfirmed| |2013-03-27 Blocks| |37021 AssignedTo|unassigned at gcc dot |rguenth at gcc dot gnu.org |gnu.org | Ever Confirmed|0 |1 --- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> 2013-03-27 12:45:02 UTC --- Confirmed. We for example have t.c:39: note: Build SLP failed: not grouped load _46 = *pix_1; as we fail to detect groups for _46 = *pix_1; _49 = MEM[(unsigned char *)pix_1 + 1B]; _52 = MEM[(unsigned char *)pix_1 + 2B]; ... as the step is not integral. That is, this is a strided-load SLP group. I am working on a patch for this. This is one reason why PR37021 is not vectorized in the best possible way.