[Bug c/20982] New: alignment attribute ignired for vector types

[Hakon dot Bugge at scali dot com]

creating a vector type aligned on a smaller unit than the size of the vector, 
produces incorrect SSE code. On Intel SSE, movups must be used instead of movaps, 
and memory operands to other SSE instructons must be avoided.

E.g. :

typedef float v4sf __attribute__ ((mode(V4SF),aligned(8)));

v4sf foo(v4sf *p, v4sf *q) {
  return *p + *q;
}

produces

   0:   0f 28 07                movaps (%rdi),%xmm0
   3:   0f 58 06                addps  (%rsi),%xmm0
   6:   c3                      retq   

it should have been something like:

   movups (%rdi),%xmm0
   movups (%rsi),%xmm1
   addps  %xmm1, %xmm0
   retq

Thanks, HÃ¥kon

-- 
           Summary: alignment attribute ignired for vector types
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: Hakon dot Bugge at scali dot com
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: 3.4.3
  GCC host triplet: 3.4.3
GCC target triplet: 3.4.3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20982