[llvm-bugs] [Bug 51746] New: Suboptimal auto vectorized code with accumulator

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=51746

            Bug ID: 51746
           Summary: Suboptimal auto vectorized code with accumulator
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: david.bolvansky at gmail.com
                CC: llvm-bugs at lists.llvm.org

typedef int v4si __attribute__ ((vector_size (16)));
int
test1 (int acc, v4si v1, v4si v2, v4si v3, v4si v4)
{
  acc &= v1[0] & v1[1] & v1[2] & v1[3];
  acc &= v2[0] & v2[1] & v2[2] & v2[3];
  acc &= v3[0] & v3[1] & v3[2] & v3[3];
  acc &= v4[0] & v4[1] & v4[2] & v4[3];
  return acc;
}


GGC -O3:
test1(int, int __vector(4), int __vector(4), int __vector(4), int __vector(4)):
        pand    xmm0, xmm1
        pand    xmm2, xmm3
        pand    xmm2, xmm0
        pshufd  xmm0, xmm2, 85
        movd    edx, xmm2
        movd    eax, xmm0
        movdqa  xmm0, xmm2
        and     eax, edx
        punpckhdq       xmm0, xmm2
        pshufd  xmm2, xmm2, 255
        and     eax, edi
        movd    edx, xmm0
        and     edx, eax
        movd    eax, xmm2
        and     eax, edx
        ret

Clang -O3:
test1(int, int __vector(4), int __vector(4), int __vector(4), int __vector(4)):
                  # @test1(int, int __vector(4), int __vector(4), int
__vector(4), int __vector(4))
        movd    eax, xmm1
        pshufd  xmm4, xmm1, 85                  # xmm4 = xmm1[1,1,1,1]
        movd    ecx, xmm4
        and     ecx, eax
        pshufd  xmm4, xmm1, 238                 # xmm4 = xmm1[2,3,2,3]
        movd    eax, xmm4
        and     eax, ecx
        pshufd  xmm1, xmm1, 255                 # xmm1 = xmm1[3,3,3,3]
        movd    ecx, xmm1
        movd    edx, xmm2
        pshufd  xmm1, xmm2, 85                  # xmm1 = xmm2[1,1,1,1]
        movd    esi, xmm1
        and     esi, ecx
        and     esi, edx
        pshufd  xmm1, xmm2, 238                 # xmm1 = xmm2[2,3,2,3]
        movd    ecx, xmm1
        and     ecx, esi
        pshufd  xmm1, xmm2, 255                 # xmm1 = xmm2[3,3,3,3]
        movd    edx, xmm1
        and     edx, ecx
        movd    ecx, xmm3
        pshufd  xmm1, xmm3, 85                  # xmm1 = xmm3[1,1,1,1]
        movd    esi, xmm1
        and     esi, ecx
        pshufd  xmm1, xmm3, 238                 # xmm1 = xmm3[2,3,2,3]
        movd    ecx, xmm1
        and     ecx, esi
        pshufd  xmm1, xmm3, 255                 # xmm1 = xmm3[3,3,3,3]
        movd    esi, xmm1
        and     esi, ecx
        pshufd  xmm1, xmm0, 238                 # xmm1 = xmm0[2,3,2,3]
        pand    xmm1, xmm0
        pshufd  xmm0, xmm1, 85                  # xmm0 = xmm1[1,1,1,1]
        pand    xmm0, xmm1
        movd    ecx, xmm0
        and     eax, edi
        and     eax, ecx
        and     eax, edx
        and     eax, esi
        ret


Codegen: https://godbolt.org/z/Y86fodv5j

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20210904/2d882c68/attachment-0001.html>