[llvm-bugs] [Bug 38917] New: Failure to hoist repeated vextractf128 from switch cases

via llvm-bugs llvm-bugs at lists.llvm.org
Wed Sep 12 11:27:25 PDT 2018


https://bugs.llvm.org/show_bug.cgi?id=38917

            Bug ID: 38917
           Summary: Failure to hoist repeated vextractf128 from switch
                    cases
           Product: new-bugs
           Version: unspecified
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: llvm-dev at redking.me.uk
                CC: llvm-bugs at lists.llvm.org

Simplified (masked store style) example:

https://gcc.godbolt.org/z/g7URz3

void store07(__m256i a, __m256i v, int *f) {
    switch(_mm256_movemask_ps(a)) {
    case 0xff:
      f[7] = _mm256_extract_epi32(v, 7);            
    case 0x7f:
      f[6] = _mm256_extract_epi32(v, 6);
    case 0x3f:
      f[5] = _mm256_extract_epi32(v, 5);
    case 0x1f:
      f[4] = _mm256_extract_epi32(v, 4);
    case 0x0f:
      f[3] = _mm256_extract_epi32(v, 3);
    case 0x07:
      f[2] = _mm256_extract_epi32(v, 2);
    case 0x03:
      f[1] = _mm256_extract_epi32(v, 1);
    case 0x01:
      f[0] = _mm256_extract_epi32(v, 0);
      break;
    }
}

_Z7store07Dv4_xS_Pi: # @_Z7store07Dv4_xS_Pi
  vmovmskps %ymm0, %eax
  cmpl $62, %eax
  jg .LBB0_3
  decl %eax
  cmpl $30, %eax
  ja .LBB0_14
  jmpq *.LJTI0_0(,%rax,8)
.LBB0_3:
  cmpl $63, %eax
  je .LBB0_8
  cmpl $255, %eax
  je .LBB0_6
  cmpl $127, %eax
  je .LBB0_7
.LBB0_14:
  retq
.LBB0_6:
  vextractf128 $1, %ymm1, %xmm0
  vextractps $3, %xmm0, 28(%rdi)
.LBB0_7:
  vextractf128 $1, %ymm1, %xmm0
  vextractps $2, %xmm0, 24(%rdi)
.LBB0_8:
  vextractf128 $1, %ymm1, %xmm0
  vextractps $1, %xmm0, 20(%rdi)
.LBB0_9:
  vextractf128 $1, %ymm1, %xmm0
  vmovss %xmm0, 16(%rdi)
.LBB0_10:
  vextractps $3, %xmm1, 12(%rdi)
.LBB0_11:
  vextractps $2, %xmm1, 8(%rdi)
.LBB0_12:
  vextractps $1, %xmm1, 4(%rdi)
.LBB0_13:
  vmovss %xmm1, (%rdi)
  retq

Shouldn't CSE be able to hoist the repeated vextractf128 instructions out from
the switch cases?

-- 
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/20180912/e61dbd56/attachment.html>


More information about the llvm-bugs mailing list