[llvm-bugs] [Bug 51365] New: [ARM][Thumb2] LowerCONCAT_VECTORS_i1 assumes concatenation of only 2 vectors

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=51365

            Bug ID: 51365
           Summary: [ARM][Thumb2] LowerCONCAT_VECTORS_i1 assumes
                    concatenation of only 2 vectors
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: ARM
          Assignee: unassignedbugs at nondot.org
          Reporter: llvm-dev at redking.me.uk
                CC: david.green at arm.com, llvm-bugs at lists.llvm.org,
                    smithp352 at googlemail.com, Ties.Stuij at arm.com

I'm currently investigating adding this style of fold as an extension to
D107068:

  concat(concat(x,y,z,w),concat(a,b,c,d)) -> concat(x,y,z,w,a,b,c,d)

This has exposed an issue with LowerCONCAT_VECTORS_i1 - it assumes that it only
ever has 2 operands:

static SDValue LowerCONCAT_VECTORS_i1(SDValue Op, SelectionDAG &DAG,
                                      const ARMSubtarget *ST) {
  SDValue V1 = Op.getOperand(0);
  SDValue V2 = Op.getOperand(1);
  SDLoc dl(Op);
  EVT VT = Op.getValueType();
  EVT Op1VT = V1.getValueType();
  EVT Op2VT = V2.getValueType();
  unsigned NumElts = VT.getVectorNumElements();

and with my fold the active_lane_mask.ll v16i8 test can now generate:

    (v16i1 concat(v4i1,v4i1,v4i1,v4i1))

This needs generalizing to repeatedly concatenate pairs of subvectors together
using the existing lowering approach.

-- 
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/20210805/ee1b9f6a/attachment.html>