[llvm-bugs] [Bug 41837] New: X86FastISel::X86FastEmitLoad - unsafe NT vector load

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=41837

            Bug ID: 41837
           Summary: X86FastISel::X86FastEmitLoad - unsafe NT vector load
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: llvm-dev at redking.me.uk
                CC: craig.topper at gmail.com, llvm-bugs at lists.llvm.org,
                    llvm-dev at redking.me.uk, spatel+llvm at rotateright.com

For V4f32/v2f64 we only attempt NT loads when SSE41 is enabled:

  case MVT::v4f32:
    if (IsNonTemporal && Alignment >= 16 && HasSSE41)
      Opc = HasVLX ? X86::VMOVNTDQAZ128rm :
            HasAVX ? X86::VMOVNTDQArm : X86::MOVNTDQArm;
    else if (Alignment >= 16)
      Opc = HasVLX ? X86::VMOVAPSZ128rm :
            HasAVX ? X86::VMOVAPSrm : X86::MOVAPSrm;
    else
      Opc = HasVLX ? X86::VMOVUPSZ128rm :
            HasAVX ? X86::VMOVUPSrm : X86::MOVUPSrm;
    RC = HasVLX ? &X86::VR128XRegClass : &X86::VR128RegClass;
    break;

But this limit appears to be missing for 128-bit integer vectors:

  case MVT::v4i32:
  case MVT::v2i64:
  case MVT::v8i16:
  case MVT::v16i8:
    if (IsNonTemporal && Alignment >= 16)
      Opc = HasVLX ? X86::VMOVNTDQAZ128rm :
            HasAVX ? X86::VMOVNTDQArm : X86::MOVNTDQArm;
    else if (Alignment >= 16)
      Opc = HasVLX ? X86::VMOVDQA64Z128rm :
            HasAVX ? X86::VMOVDQArm : X86::MOVDQArm;
    else
      Opc = HasVLX ? X86::VMOVDQU64Z128rm :
            HasAVX ? X86::VMOVDQUrm : X86::MOVDQUrm;
    RC = HasVLX ? &X86::VR128XRegClass : &X86::VR128RegClass;
    break;

-- 
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/20190510/b081b1f3/attachment.html>