[Bug target/57601] Vector lowering could use larger modes

[glisse at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57601

--- Comment #2 from Marc Glisse <glisse at gcc dot gnu.org> ---
With this code:

#include <mmintrin.h>
__m64 f(__m64 x, __m64 y){return _mm_sub_pi32(x,y);}

I get (-m64):

    movdq2q    %xmm0, %mm0
    movdq2q    %xmm1, %mm1
    psubd    %mm1, %mm0
    movq2dq    %mm0, %xmm0

and (-m32):

    psubd    %mm1, %mm0