[Bug target/52607] New: v4df __builtin_shuffle with {0,2,1,3} or {1,3,0,2}

[marc.glisse at normalesup dot org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52607

             Bug #: 52607
           Summary: v4df __builtin_shuffle with {0,2,1,3} or {1,3,0,2}
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: target
        AssignedTo: unassig...@gcc.gnu.org
        ReportedBy: marc.gli...@normalesup.org


Hello,

this is really just a follow-up to PR52568. The permutations {0,2,1,3} and
{1,3,0,2} can be realized with a very similar technique.

Starting from 0123:
vpermilpd+vperm2f128->3210
vblendpd(0123,3210)->0213

or:
vpermilpd->1032
vperm2f128->2301
vblendpd(1032,2301)->1302

I am not sure if there is a nice way to generalize this or if the function
expand_vec_perm_vperm2f128_vblend should be cloned a few times and slightly
modified.

(these permutations are less important to me than 1230 was)