[AArch64][SVE] Remove redundant PTEST of MATCH/NMATCH results

[bsmith]

Summary

Co-authored-by: Paul Walker <[e-mail]>

llvm/lib/Target/AArch64/SVEInstrFormats.td

: I<(outs pprty:$Pd), (ins PPR3bAny:$Pg, zprty:$Zn, zprty:$Zm),
   let Inst{20-16} = Zm;
   let Inst{15-13} = 0b100;
   let Inst{12-10} = Pg;
   let Inst{9-5}   = Zn;
   let Inst{4}     = opc;
   let Inst{3-0}   = Pd;
 
   let Defs = [NZCV];
+  let isPTestLike = 1;
 }
 
 multiclass sve2_char_match<bit opc, string asm, SDPatternOperator op> {
   def _B : sve2_char_match<0b0, opc, asm, PPR8, ZPR8>;
   def _H : sve2_char_match<0b1, opc, asm, PPR16, ZPR16>;
 
   def : SVE_3_Op_Pat<nxv16i1, op, nxv16i1, nxv16i8, nxv16i8, !cast<Instruction>(NAME # _B)>;
   def : SVE_3_Op_Pat<nxv8i1,  op, nxv8i1,  nxv8i16, nxv8i16, !cast<Instruction>(NAME # _H)>;

llvm/test/CodeGen/AArch64/sve-ptest-removal-match.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve2 %s -o - | FileCheck %s
+
+;
+; MATCH
+;
+
+define i32 @match_nxv16i8(<vscale x 16 x i1> %pg, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
+; CHECK-LABEL: match_nxv16i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    match p0.b, p0/z, z0.b, z1.b
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.match.nxv16i8(<vscale x 16 x i1> %pg, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
+  %2 = tail call i1 @llvm.aarch64.sve.ptest.any(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
+  %conv = zext i1 %2 to i32
+  ret i32 %conv
+}
+
+;
+; NMATCH
+;
+
+define i32 @nmatch_nxv16i8(<vscale x 16 x i1> %pg, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
+; CHECK-LABEL: nmatch_nxv16i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    nmatch p0.b, p0/z, z0.b, z1.b
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+  %1 = tail call <vscale x 16 x i1> @llvm.aarch64.sve.nmatch.nxv16i8(<vscale x 16 x i1> %pg, <vscale x 16 x i8> %a, <vscale x 16 x i8> %b)
+  %2 = tail call i1 @llvm.aarch64.sve.ptest.any(<vscale x 16 x i1> %pg, <vscale x 16 x i1> %1)
+  %conv = zext i1 %2 to i32
+  ret i32 %conv
+}
+
+declare <vscale x 16 x i1> @llvm.aarch64.sve.match.nxv16i8(<vscale x 16 x i1>, <vscale x 16 x i8>, <vscale x 16 x i8>)
+declare <vscale x 16 x i1> @llvm.aarch64.sve.nmatch.nxv16i8(<vscale x 16 x i1>, <vscale x 16 x i8>, <vscale x 16 x i8>)
+declare i1 @llvm.aarch64.sve.ptest.any(<vscale x 16 x i1>, <vscale x 16 x i1>)