[AArch64][SVE2p1] Remove redundant PTESTs when predicate is a WHILEcc_x2

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -20195,13 +20195,21 @@ static bool isPredicateCCSettingOp(SDValue N) {
       (N.getOpcode() == ISD::GET_ACTIVE_LANE_MASK) ||
       (N.getOpcode() == ISD::INTRINSIC_WO_CHAIN &&
        (N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilege ||
+        N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilege_x2 ||
         N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilegt ||
+        N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilegt_x2 ||
         N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilehi ||
+        N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilehi_x2 ||
         N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilehs ||
+        N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilehs_x2 ||
         N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilele ||
+        N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilele_x2 ||
         N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilelo ||
+        N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilelo_x2 ||
         N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilels ||
-        N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilelt)))
+        N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilels_x2 ||
+        N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilelt ||
+        N.getConstantOperandVal(0) == Intrinsic::aarch64_sve_whilelt_x2)))
     return true;
 
   return false;
@@ -20227,7 +20235,7 @@ performFirstTrueTestVectorCombine(SDNode *N,
 
   // Restricted the DAG combine to only cases where we're extracting from a
   // flag-setting operation.
-  if (!isPredicateCCSettingOp(N0))
+  if (!isPredicateCCSettingOp(N0) || N0.getResNo() != 0)
     return SDValue();
 
   // Extracts of lane 0 for SVE can be expressed as PTEST(Op, FIRST) ? 1 : 0

llvm/lib/Target/AArch64/AArch64InstrInfo.cpp

@@ -1613,6 +1613,16 @@ bool AArch64InstrInfo::optimizePTestInstr(
     const MachineRegisterInfo *MRI) const {
   auto *Mask = MRI->getUniqueVRegDef(MaskReg);
   auto *Pred = MRI->getUniqueVRegDef(PredReg);
+
+  if (Pred->isCopy() && PTest->getOpcode() == AArch64::PTEST_PP_FIRST) {
+    // Instructions which return a multi-vector (e.g. WHILECC_x2) require copies
+    // before the branch to extract each subregister.
+    auto Op = Pred->getOperand(1);
+    if (Op.isReg() && Op.getReg().isVirtual() &&
+        Op.getSubReg() == AArch64::psub0)
+      Pred = MRI->getUniqueVRegDef(Op.getReg());
+  }
+
   unsigned PredOpcode = Pred->getOpcode();
   auto NewOp = canRemovePTestInstr(PTest, Mask, Pred, MRI);
   if (!NewOp)

llvm/lib/Target/AArch64/AArch64RegisterInfo.td

@@ -1164,25 +1164,22 @@ class PPRVectorListMul<int ElementWidth, int NumRegs> : PPRVectorList<ElementWid
                                                                 ", AArch64::PPRMul2RegClassID>";
 }
 
+class PPR2MulRegOp<string Suffix, int Size, ElementSizeEnum ES>
+    : RegisterOperand<PPR2Mul2, "printTypedVectorList<0,'"#Suffix#"'>"> {
+  ElementSizeEnum ElementSize;
+  let ElementSize = ES;
+  let ParserMatchClass = PPRVectorListMul<Size, 2>;
+}
+
 let EncoderMethod = "EncodeRegMul_MinMax<2, 0, 14>",
     DecoderMethod = "DecodePPR2Mul2RegisterClass" in {
-  def PP_b_mul_r : RegisterOperand<PPR2Mul2, "printTypedVectorList<0,'b'>"> {
-    let ParserMatchClass = PPRVectorListMul<8, 2>;
-  }
-
-  def PP_h_mul_r : RegisterOperand<PPR2Mul2, "printTypedVectorList<0,'h'>"> {
-    let ParserMatchClass = PPRVectorListMul<16, 2>;
-  }
 
-  def PP_s_mul_r : RegisterOperand<PPR2Mul2, "printTypedVectorList<0,'s'>"> {
-    let ParserMatchClass = PPRVectorListMul<32, 2>;
-  }
-
-  def PP_d_mul_r : RegisterOperand<PPR2Mul2, "printTypedVectorList<0,'d'>"> {
-    let ParserMatchClass = PPRVectorListMul<64, 2>;
-  }
-}  // end let EncoderMethod/DecoderMethod
+  def PP_b_mul_r : PPR2MulRegOp<"b", 8,  ElementSizeB>;
+  def PP_h_mul_r : PPR2MulRegOp<"h", 16, ElementSizeH>;
+  def PP_s_mul_r : PPR2MulRegOp<"s", 32, ElementSizeS>;
+  def PP_d_mul_r : PPR2MulRegOp<"d", 64, ElementSizeD>;
 
+} // end let EncoderMethod/DecoderMethod
 
 //===----------------------------------------------------------------------===//
 // SVE vector register classes

llvm/lib/Target/AArch64/SVEInstrFormats.td

@@ -10405,7 +10405,7 @@ multiclass sve2p1_int_while_rr_pn<string mnemonic, bits<3> opc> {
 
 // SVE integer compare scalar count and limit (predicate pair)
 class sve2p1_int_while_rr_pair<string mnemonic, bits<2> sz, bits<3> opc,
-                             RegisterOperand ppr_ty>
+                               PPR2MulRegOp ppr_ty>
     : I<(outs ppr_ty:$Pd), (ins GPR64:$Rn, GPR64:$Rm),
         mnemonic, "\t$Pd, $Rn, $Rm",
         "", []>, Sched<[]> {
@@ -10425,6 +10425,8 @@ class sve2p1_int_while_rr_pair<string mnemonic, bits<2> sz, bits<3> opc,
 
   let Defs = [NZCV];
   let hasSideEffects = 0;
+  let ElementSize = ppr_ty.ElementSize;
+  let isWhile = 1;
 }
 
 

llvm/test/CodeGen/AArch64/sve-cmp-folds.ll

@@ -1,5 +1,5 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=aarch64-linux-unknown -mattr=+sve2 -o - < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-unknown -mattr=+sve2p1 -o - < %s | FileCheck %s
 
 define <vscale x 8 x i1> @not_icmp_sle_nxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b) {
 ; CHECK-LABEL: not_icmp_sle_nxv8i16:
@@ -220,6 +220,117 @@ define i1 @lane_mask_first(i64 %next, i64 %end) {
   ret i1 %bit
 }
 
+define i1 @whilege_x2_first(i64 %next, i64 %end) {
+; CHECK-LABEL: whilege_x2_first:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    whilege { p0.s, p1.s }, x0, x1
+; CHECK-NEXT:    cset w0, mi
+; CHECK-NEXT:    ret
+  %predpair = call { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilege.x2.nxv4i1.i64(i64 %next, i64 %end)
+  %predicate = extractvalue { <vscale x 4 x i1>, <vscale x 4 x i1> } %predpair, 0
+  %bit = extractelement <vscale x 4 x i1> %predicate, i64 0
+  ret i1 %bit
+}
+
+define i1 @whilegt_x2_first(i64 %next, i64 %end) {
+; CHECK-LABEL: whilegt_x2_first:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    whilegt { p0.s, p1.s }, x0, x1
+; CHECK-NEXT:    cset w0, mi
+; CHECK-NEXT:    ret
+  %predpair = call { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilegt.x2.nxv4i1.i64(i64 %next, i64 %end)
+  %predicate = extractvalue { <vscale x 4 x i1>, <vscale x 4 x i1> } %predpair, 0
+  %bit = extractelement <vscale x 4 x i1> %predicate, i64 0
+  ret i1 %bit
+}
+
+define i1 @whilehi_x2_first(i64 %next, i64 %end) {
+; CHECK-LABEL: whilehi_x2_first:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    whilehi { p0.s, p1.s }, x0, x1
+; CHECK-NEXT:    cset w0, mi
+; CHECK-NEXT:    ret
+  %predpair = call { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilehi.x2.nxv4i1.i64(i64 %next, i64 %end)
+  %predicate = extractvalue { <vscale x 4 x i1>, <vscale x 4 x i1> } %predpair, 0
+  %bit = extractelement <vscale x 4 x i1> %predicate, i64 0
+  ret i1 %bit
+}
+
+define i1 @whilehs_x2_first(i64 %next, i64 %end) {
+; CHECK-LABEL: whilehs_x2_first:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    whilehs { p0.s, p1.s }, x0, x1
+; CHECK-NEXT:    cset w0, mi
+; CHECK-NEXT:    ret
+  %predpair = call { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilehs.x2.nxv4i1.i64(i64 %next, i64 %end)
+  %predicate = extractvalue { <vscale x 4 x i1>, <vscale x 4 x i1> } %predpair, 0
+  %bit = extractelement <vscale x 4 x i1> %predicate, i64 0
+  ret i1 %bit
+}
+
+define i1 @whilele_x2_first(i64 %next, i64 %end) {
+; CHECK-LABEL: whilele_x2_first:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    whilele { p0.s, p1.s }, x0, x1
+; CHECK-NEXT:    cset w0, mi
+; CHECK-NEXT:    ret
+  %predpair = call { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilele.x2.nxv4i1.i64(i64 %next, i64 %end)
+  %predicate = extractvalue { <vscale x 4 x i1>, <vscale x 4 x i1> } %predpair, 0
+  %bit = extractelement <vscale x 4 x i1> %predicate, i64 0
+  ret i1 %bit
+}
+
+define i1 @whilelo_x2_first(i64 %next, i64 %end) {
+; CHECK-LABEL: whilelo_x2_first:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    whilelo { p0.s, p1.s }, x0, x1
+; CHECK-NEXT:    cset w0, mi
+; CHECK-NEXT:    ret
+  %predpair = call { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilelo.x2.nxv4i1.i64(i64 %next, i64 %end)
+  %predicate = extractvalue { <vscale x 4 x i1>, <vscale x 4 x i1> } %predpair, 0
+  %bit = extractelement <vscale x 4 x i1> %predicate, i64 0
+  ret i1 %bit
+}
+
+define i1 @whilels_x2_first(i64 %next, i64 %end) {
+; CHECK-LABEL: whilels_x2_first:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    whilels { p0.s, p1.s }, x0, x1
+; CHECK-NEXT:    cset w0, mi
+; CHECK-NEXT:    ret
+  %predpair = call { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilels.x2.nxv4i1.i64(i64 %next, i64 %end)
+  %predicate = extractvalue { <vscale x 4 x i1>, <vscale x 4 x i1> } %predpair, 0
+  %bit = extractelement <vscale x 4 x i1> %predicate, i64 0
+  ret i1 %bit
+}
+
+define i1 @whilelt_x2_first(i64 %next, i64 %end) {
+; CHECK-LABEL: whilelt_x2_first:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    whilelt { p0.s, p1.s }, x0, x1
+; CHECK-NEXT:    cset w0, mi
+; CHECK-NEXT:    ret
+  %predpair = call { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilelt.x2.nxv4i1.i64(i64 %next, i64 %end)
+  %predicate = extractvalue { <vscale x 4 x i1>, <vscale x 4 x i1> } %predpair, 0
+  %bit = extractelement <vscale x 4 x i1> %predicate, i64 0
+  ret i1 %bit
+}
+
+; Do not combine to ptest when the extract is not from the first vector result
+define i1 @whilege_x2_second_result(i64 %next, i64 %end) {
+; CHECK-LABEL: whilege_x2_second_result:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    whilege { p0.s, p1.s }, x0, x1
+; CHECK-NEXT:    mov z0.s, p1/z, #1 // =0x1
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    and w0, w8, #0x1
+; CHECK-NEXT:    ret
+  %predpair = call { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilege.x2.nxv4i1.i64(i64 %next, i64 %end)
+  %predicate = extractvalue { <vscale x 4 x i1>, <vscale x 4 x i1> } %predpair, 1
+  %bit = extractelement <vscale x 4 x i1> %predicate, i64 0
+  ret i1 %bit
+}
+
 declare i64 @llvm.vscale.i64()
 declare <vscale x 4 x i1> @llvm.aarch64.sve.whilege.nxv4i1.i64(i64, i64)
 declare <vscale x 4 x i1> @llvm.aarch64.sve.whilegt.nxv4i1.i64(i64, i64)
@@ -230,3 +341,12 @@ declare <vscale x 4 x i1> @llvm.aarch64.sve.whilelo.nxv4i1.i64(i64, i64)
 declare <vscale x 4 x i1> @llvm.aarch64.sve.whilels.nxv4i1.i64(i64, i64)
 declare <vscale x 4 x i1> @llvm.aarch64.sve.whilelt.nxv4i1.i64(i64, i64)
 declare <vscale x 4 x i1> @llvm.get.active.lane.mask.nxv4i1.i64(i64, i64)
+
+declare { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilege.x2.nxv4i1(i64, i64)
+declare { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilegt.x2.nxv4i1(i64, i64)
+declare { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilehi.x2.nxv4i1(i64, i64)
+declare { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilehs.x2.nxv4i1(i64, i64)
+declare { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilele.x2.nxv4i1(i64, i64)
+declare { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilelo.x2.nxv4i1(i64, i64)
+declare { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilels.x2.nxv4i1(i64, i64)
+declare { <vscale x 4 x i1>, <vscale x 4 x i1> } @llvm.aarch64.sve.whilelt.x2.nxv4i1(i64, i64)

llvm/test/CodeGen/AArch64/sve-ptest-removal-whilelo.mir

@@ -538,3 +538,154 @@ body:             |
     RET_ReallyLR implicit $w0
 
 ...
+
+# WHILELO (predicate pair)
+---
+name:            whilelo_x2_b64_s64
+alignment:       2
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: gpr64 }
+  - { id: 1, class: gpr64 }
+  - { id: 2, class: ppr }
+  - { id: 3, class: ppr2mul2 }
+  - { id: 4, class: ppr }
+  - { id: 5, class: ppr }
+  - { id: 6, class: gpr32 }
+  - { id: 7, class: gpr32 }
+liveins:
+  - { reg: '$x0', virtual-reg: '%0' }
+  - { reg: '$x1', virtual-reg: '%1' }
+frameInfo:
+  maxCallFrameSize: 0
+body:             |
+  bb.0.entry:
+    liveins: $x0, $x1
+
+    ; CHECK-LABEL: name: whilelo_x2_b64_s64
+    ; CHECK: liveins: $x0, $x1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
+    ; CHECK-NEXT: [[PTRUE_D:%[0-9]+]]:ppr = PTRUE_D 31, implicit $vg
+    ; CHECK-NEXT: [[WHILELO_2PXX_D:%[0-9]+]]:ppr2mul2 = WHILELO_2PXX_D [[COPY]], [[COPY1]], implicit-def $nzcv
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:ppr = COPY [[WHILELO_2PXX_D]].psub0
+    ; CHECK-NEXT: [[COPY3:%[0-9]+]]:ppr = COPY [[WHILELO_2PXX_D]].psub1
+    ; CHECK-NEXT: [[COPY4:%[0-9]+]]:gpr32 = COPY $wzr
+    ; CHECK-NEXT: [[CSINCWr:%[0-9]+]]:gpr32 = CSINCWr [[COPY4]], $wzr, 0, implicit $nzcv
+    ; CHECK-NEXT: $w0 = COPY [[CSINCWr]]
+    ; CHECK-NEXT: RET_ReallyLR implicit $w0
+    %0:gpr64 = COPY $x0
+    %1:gpr64 = COPY $x1
+    %2:ppr = PTRUE_D 31, implicit $vg
+    %3:ppr2mul2 = WHILELO_2PXX_D %0, %1, implicit-def $nzcv
+    %4:ppr = COPY %3.psub0
+    %5:ppr = COPY %3.psub1
+    PTEST_PP_FIRST killed %2, killed %4, implicit-def $nzcv
+    %6:gpr32 = COPY $wzr
+    %7:gpr32 = CSINCWr %6, $wzr, 0, implicit $nzcv
+    $w0 = COPY %7
+    RET_ReallyLR implicit $w0
+...
+
+# PTEST is not redundant when it's Pg operand is a subregister copy, but not
+# from the first subregister of ppr2mul2
+---
+name:            whilelo_x2_b64_s64_psub1
+alignment:       2
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: gpr64 }
+  - { id: 1, class: gpr64 }
+  - { id: 2, class: ppr }
+  - { id: 3, class: ppr2mul2 }
+  - { id: 4, class: ppr }
+  - { id: 5, class: ppr }
+  - { id: 6, class: gpr32 }
+  - { id: 7, class: gpr32 }
+liveins:
+  - { reg: '$x0', virtual-reg: '%0' }
+  - { reg: '$x1', virtual-reg: '%1' }
+frameInfo:
+  maxCallFrameSize: 0
+body:             |
+  bb.0.entry:
+    liveins: $x0, $x1
+
+    ; CHECK-LABEL: name: whilelo_x2_b64_s64_psub1
+    ; CHECK: liveins: $x0, $x1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
+    ; CHECK-NEXT: [[PTRUE_D:%[0-9]+]]:ppr = PTRUE_D 31, implicit $vg
+    ; CHECK-NEXT: [[WHILELO_2PXX_D:%[0-9]+]]:ppr2mul2 = WHILELO_2PXX_D [[COPY]], [[COPY1]], implicit-def $nzcv
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:ppr = COPY [[WHILELO_2PXX_D]].psub0
+    ; CHECK-NEXT: [[COPY3:%[0-9]+]]:ppr = COPY [[WHILELO_2PXX_D]].psub1
+    ; CHECK-NEXT: PTEST_PP_FIRST killed [[PTRUE_D]], killed [[COPY3]], implicit-def $nzcv
+    ; CHECK-NEXT: [[COPY4:%[0-9]+]]:gpr32 = COPY $wzr
+    ; CHECK-NEXT: [[CSINCWr:%[0-9]+]]:gpr32 = CSINCWr [[COPY4]], $wzr, 0, implicit $nzcv
+    ; CHECK-NEXT: $w0 = COPY [[CSINCWr]]
+    ; CHECK-NEXT: RET_ReallyLR implicit $w0
+    %0:gpr64 = COPY $x0
+    %1:gpr64 = COPY $x1
+    %2:ppr = PTRUE_D 31, implicit $vg
+    %3:ppr2mul2 = WHILELO_2PXX_D %0, %1, implicit-def $nzcv
+    %4:ppr = COPY %3.psub0
+    %5:ppr = COPY %3.psub1
+    PTEST_PP_FIRST killed %2, killed %5, implicit-def $nzcv
+    %6:gpr32 = COPY $wzr
+    %7:gpr32 = CSINCWr %6, $wzr, 0, implicit $nzcv
+    $w0 = COPY %7
+    RET_ReallyLR implicit $w0
+...
+
+# PTEST is not redundant when it's Pg operand is a copy from subregister 0
+# if the condition is not FIRST_ACTIVE
+---
+name:            whilelo_x2_b64_s64_not_first
+alignment:       2
+tracksRegLiveness: true
+registers:
+  - { id: 0, class: gpr64 }
+  - { id: 1, class: gpr64 }
+  - { id: 2, class: ppr }
+  - { id: 3, class: ppr2mul2 }
+  - { id: 4, class: ppr }
+  - { id: 5, class: ppr }
+  - { id: 6, class: gpr32 }
+  - { id: 7, class: gpr32 }
+liveins:
+  - { reg: '$x0', virtual-reg: '%0' }
+  - { reg: '$x1', virtual-reg: '%1' }
+frameInfo:
+  maxCallFrameSize: 0
+body:             |
+  bb.0.entry:
+    liveins: $x0, $x1
+
+    ; CHECK-LABEL: name: whilelo_x2_b64_s64_not_first
+    ; CHECK: liveins: $x0, $x1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
+    ; CHECK-NEXT: [[PTRUE_D:%[0-9]+]]:ppr = PTRUE_D 31, implicit $vg
+    ; CHECK-NEXT: [[WHILELO_2PXX_D:%[0-9]+]]:ppr2mul2 = WHILELO_2PXX_D [[COPY]], [[COPY1]], implicit-def $nzcv
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:ppr = COPY [[WHILELO_2PXX_D]].psub0
+    ; CHECK-NEXT: [[COPY3:%[0-9]+]]:ppr = COPY [[WHILELO_2PXX_D]].psub1
+    ; CHECK-NEXT: PTEST_PP killed [[PTRUE_D]], killed [[COPY2]], implicit-def $nzcv
+    ; CHECK-NEXT: [[COPY4:%[0-9]+]]:gpr32 = COPY $wzr
+    ; CHECK-NEXT: [[CSINCWr:%[0-9]+]]:gpr32 = CSINCWr [[COPY4]], $wzr, 0, implicit $nzcv
+    ; CHECK-NEXT: $w0 = COPY [[CSINCWr]]
+    ; CHECK-NEXT: RET_ReallyLR implicit $w0
+    %0:gpr64 = COPY $x0
+    %1:gpr64 = COPY $x1
+    %2:ppr = PTRUE_D 31, implicit $vg
+    %3:ppr2mul2 = WHILELO_2PXX_D %0, %1, implicit-def $nzcv
+    %4:ppr = COPY %3.psub0
+    %5:ppr = COPY %3.psub1
+    PTEST_PP killed %2, killed %4, implicit-def $nzcv
+    %6:gpr32 = COPY $wzr
+    %7:gpr32 = CSINCWr %6, $wzr, 0, implicit $nzcv
+    $w0 = COPY %7
+    RET_ReallyLR implicit $w0
+...