[llvm-bugs] [Bug 40083] New: [x86] psubus matching failure with undef elements

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=40083

            Bug ID: 40083
           Summary: [x86] psubus matching failure with undef elements
           Product: new-bugs
           Version: unspecified
          Hardware: PC
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: spatel+llvm at rotateright.com
                CC: htmldeveloper at gmail.com, llvm-bugs at lists.llvm.org

This example came up in the discussion for D55780:

unsigned long long test_sub_2(__m128i x) {
    __m128i c = _mm_set1_epi8(70);
    return _mm_subs_epu8(x, c)[0];
}

Currently this becomes in IR:

define i64 @test_sub_2(<2 x i64> %x) {
  %t0 = bitcast <2 x i64> %x to <16 x i8>
  %cmp = icmp ugt <16 x i8> %t0, <i8 70, i8 70, i8 70, i8 70, i8 70, i8 70, i8
70, i8 70, i8 70, i8 70, i8 70, i8 70, i8 70, i8 70, i8 70, i8 70>
  %bop = add <16 x i8> %0, <i8 -70, i8 -70, i8 -70, i8 -70, i8 -70, i8 -70, i8
-70, i8 -70, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8
undef, i8 undef>
  %sel = select <16 x i1> %cmp, <16 x i8> %bop, <16 x i8> <i8 0, i8 0, i8 0, i8
0, i8 0, i8 0, i8 0, i8 0, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8
undef, i8 undef, i8 undef>
  %bc = bitcast <16 x i8> %sel to <2 x i64>
  %ext = extractelement <2 x i64> %bc, i32 0
  ret i64 %ext
}

And that becomes this in x86 asm:
        vpmaxub LCPI0_0(%rip), %xmm0, %xmm1
        vpcmpeqb        %xmm1, %xmm0, %xmm1
        vpaddb  LCPI0_1(%rip), %xmm0, %xmm0
        vpand   %xmm0, %xmm1, %xmm0
        vmovq   %xmm0, %rax

---------------------------------------------------------------------------

We want this to become psubus again, but undefs are getting in the way.
It's possible that we also want to add an IR transform to turn the
bitcast+extract into an extracting shufflevector+bitcast.

-- 
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/20181218/f4202f42/attachment.html>