https://bugs.kde.org/show_bug.cgi?id=442168

--- Comment #8 from Julian Seward <jsew...@acm.org> ---
(In reply to Xavier Roche from comment #7)

I checked V's {V,}MOVSD implementation more, and still find no problem
I think this is unrelated to those insns.

If I had to guess, I'd say it *might* be related to the use of

+       vcmpneq_oqsd    .LCPI1_1(%rip), %xmm1, %xmm1

since the implementation is a bit shaky:

  // 0xC: this isn't really right because it returns all-1s when
  // either operand is a NaN, and it should return all-0s.
  case 0x4:  XXX(False, False, Iop_CmpEQ32Fx4, True);  break; // NEQ_UQ
  case 0xC:  XXX(False, False, Iop_CmpEQ32Fx4, True);  break; // NEQ_OQ

(from `findSSECmpOp`).  But that's just a guess; without a way to reproduce
the failure, I think the chances of resolving this is pretty low.

-- 
You are receiving this mail because:
You are watching all bug changes.

Reply via email to