[Bug target/95894] New: vector shift by lane zero generates inter unit move

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95894

            Bug ID: 95894
           Summary: vector shift by lane zero generates inter unit move
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rguenth at gcc dot gnu.org
  Target Milestone: ---

typedef int v4si __attribute__((vector_size(16)));
v4si foo (v4si x)
{
  return x << x[0];
}

generates

foo:
.LFB0:
        .cfi_startproc
        movd    %xmm0, %eax
        cltq
        movq    %rax, %xmm1
        pslld   %xmm1, %xmm0
        ret

while we could use sth like

         pxor %xmm1, %xmm1
         punpckldq %xmm0, %xmm1
         pslld %xmm1, %xmm0

to zero-extend x[0] to DImode in a SSE reg.  Unfortunately even

typedef long v2di __attribute__((vector_size(16)));
v2di bar (v2di x)
{
  return x << x[0];
}

shows this behavior useless behavior:

bar:
.LFB1:
        .cfi_startproc
        movq    %xmm0, %rax
        cltq
        movq    %rax, %xmm1
        psllq   %xmm1, %xmm0
        ret

because the gimplifier casts the shift amount to unsigned int :(