[Bug target/101956] New: Miss vectorization from v4hi to v4df

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101956

            Bug ID: 101956
           Summary: Miss vectorization from v4hi to v4df
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: crazylht at gmail dot com
  Target Milestone: ---
              Host: x86_64-pc-linux-gnu
            Target: x86_64-*-* i?86-*-*

void
foo (double* p, short* q)
{
   p[0] = q[0];
   p[1] = q[1];
   p[2] = q[2];
   p[3] = q[3];
}

clang generates 

        vpmovsxwd       xmm0, qword ptr [rsi]
        vcvtdq2pd       ymm0, xmm0
        vmovups ymmword ptr [rdi], ymm0
        vzeroupper

gcc generates

foo(double*, short*):
  movswl (%rsi), %eax
  vxorps %xmm0, %xmm0, %xmm0
  vcvtsi2sdl %eax, %xmm0, %xmm1
  movswl 2(%rsi), %eax
  vcvtsi2sdl %eax, %xmm0, %xmm2
  movswl 4(%rsi), %eax
  vmovsd %xmm2, %xmm2, %xmm3
  vcvtsi2sdl %eax, %xmm0, %xmm2
  movswl 6(%rsi), %eax
  vcvtsi2sdl %eax, %xmm0, %xmm0
  vunpcklpd %xmm0, %xmm2, %xmm2
  vunpcklpd %xmm3, %xmm1, %xmm0
  vinsertf128 $0x1, %xmm2, %ymm0, %ymm0
  vmovupd %ymm0, (%rdi)
  vzeroupper
  ret