[llvm] [WIP][SelectionDAG] Add support for the 3-way comparison intrinsics [US]CMP (PR #91871)

via llvm-commits llvm-commits at lists.llvm.org
Sat May 11 15:48:00 PDT 2024


https://github.com/Poseydon42 created https://github.com/llvm/llvm-project/pull/91871

This PR adds initial support for the `scmp`/`ucmp` 3-way comparison intrinsics in the SelectionDAG.

What works as of now:
* An invokation of the intrinsic in the IR gets properly lowered into SelectionDAG
* A node with opcodes `UCMP`/`SCMP` gets properly expanded into two comparisons and two selects
* Narrow scalar arguments and return types are properly handled (i.e. i3 or i51)

What doesn't work yet:
* Wide scalar arguments and return types are not properly expanded (the two functions that should be doing this are just `assert(false)`s for now)
* Vector arguments and return types are completely broken right now

>From 0590ce071aa2b3cd6b64b569ace8a869c16ca0d3 Mon Sep 17 00:00:00 2001
From: Poseydon42 <vvmposeydon at gmail.com>
Date: Sat, 11 May 2024 23:33:14 +0100
Subject: [PATCH] [SelectionDAG] Add support for the 3-way comparison
 intrinsics [US]CMP

---
 llvm/include/llvm/CodeGen/ISDOpcodes.h        |   5 +
 llvm/include/llvm/CodeGen/TargetLowering.h    |   4 +
 .../include/llvm/Target/TargetSelectionDAG.td |   5 +
 llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp |   6 +
 .../SelectionDAG/LegalizeIntegerTypes.cpp     |  41 +++++
 llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h |   5 +
 .../SelectionDAG/SelectionDAGBuilder.cpp      |  16 ++
 .../SelectionDAG/SelectionDAGDumper.cpp       |   2 +
 .../CodeGen/SelectionDAG/TargetLowering.cpp   |  21 +++
 llvm/lib/CodeGen/TargetLoweringBase.cpp       |   3 +
 llvm/test/CodeGen/X86/uscmp.ll                | 163 ++++++++++++++++++
 11 files changed, 271 insertions(+)
 create mode 100644 llvm/test/CodeGen/X86/uscmp.ll

diff --git a/llvm/include/llvm/CodeGen/ISDOpcodes.h b/llvm/include/llvm/CodeGen/ISDOpcodes.h
index 6429947958ee9..952e94134020a 100644
--- a/llvm/include/llvm/CodeGen/ISDOpcodes.h
+++ b/llvm/include/llvm/CodeGen/ISDOpcodes.h
@@ -677,6 +677,11 @@ enum NodeType {
   UMIN,
   UMAX,
 
+  /// [US]CMP - 3-way comparison of signed or unsigned integers. Returns -1, 0,
+  /// or 1 depending on whether Op0 <, ==, or > Op1
+  SCMP,
+  UCMP,
+
   /// Bitwise operators - logical and, logical or, logical xor.
   AND,
   OR,
diff --git a/llvm/include/llvm/CodeGen/TargetLowering.h b/llvm/include/llvm/CodeGen/TargetLowering.h
index 7ed08cfa8a202..704bd666ea5a9 100644
--- a/llvm/include/llvm/CodeGen/TargetLowering.h
+++ b/llvm/include/llvm/CodeGen/TargetLowering.h
@@ -5402,6 +5402,10 @@ class TargetLowering : public TargetLoweringBase {
   /// method accepts integers as its arguments.
   SDValue expandAddSubSat(SDNode *Node, SelectionDAG &DAG) const;
 
+  /// Method for building the DAG expansion of ISD::[US]CMP. This
+  /// method accepts integers as its arguments
+  SDValue expandCmp(SDNode *Node, SelectionDAG &DAG) const;
+
   /// Method for building the DAG expansion of ISD::[US]SHLSAT. This
   /// method accepts integers as its arguments.
   SDValue expandShlSat(SDNode *Node, SelectionDAG &DAG) const;
diff --git a/llvm/include/llvm/Target/TargetSelectionDAG.td b/llvm/include/llvm/Target/TargetSelectionDAG.td
index 1684b424e3b44..6d771521aa739 100644
--- a/llvm/include/llvm/Target/TargetSelectionDAG.td
+++ b/llvm/include/llvm/Target/TargetSelectionDAG.td
@@ -434,6 +434,11 @@ def umin       : SDNode<"ISD::UMIN"      , SDTIntBinOp,
 def umax       : SDNode<"ISD::UMAX"      , SDTIntBinOp,
                                   [SDNPCommutative, SDNPAssociative]>;
 
+def scmp       : SDNode<"ISD::SCMP"      , SDTIntBinOp,
+                                  []>;
+def ucmp       : SDNode<"ISD::UCMP"      , SDTIntBinOp,
+                                  []>;
+
 def saddsat    : SDNode<"ISD::SADDSAT"   , SDTIntBinOp, [SDNPCommutative]>;
 def uaddsat    : SDNode<"ISD::UADDSAT"   , SDTIntBinOp, [SDNPCommutative]>;
 def ssubsat    : SDNode<"ISD::SSUBSAT"   , SDTIntBinOp>;
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index bfc3e08c1632d..c10546fe58aa7 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -1148,6 +1148,8 @@ void SelectionDAGLegalize::LegalizeOp(SDNode *Node) {
   case ISD::USUBSAT:
   case ISD::SSHLSAT:
   case ISD::USHLSAT:
+  case ISD::SCMP:
+  case ISD::UCMP:
   case ISD::FP_TO_SINT_SAT:
   case ISD::FP_TO_UINT_SAT:
     Action = TLI.getOperationAction(Node->getOpcode(), Node->getValueType(0));
@@ -3864,6 +3866,10 @@ bool SelectionDAGLegalize::ExpandNode(SDNode *Node) {
   case ISD::USUBSAT:
     Results.push_back(TLI.expandAddSubSat(Node, DAG));
     break;
+  case ISD::SCMP:
+  case ISD::UCMP:
+    Results.push_back(TLI.expandCmp(Node, DAG));
+    break;
   case ISD::SSHLSAT:
   case ISD::USHLSAT:
     Results.push_back(TLI.expandShlSat(Node, DAG));
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index 0aa36deda79dc..54c71c263499a 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -232,6 +232,11 @@ void DAGTypeLegalizer::PromoteIntegerResult(SDNode *N, unsigned ResNo) {
     Res = PromoteIntRes_ADDSUBSHLSAT<VPMatchContext>(N);
     break;
 
+  case ISD::SCMP:
+  case ISD::UCMP:
+    Res = PromoteIntRes_CMP(N);
+    break;
+
   case ISD::SMULFIX:
   case ISD::SMULFIXSAT:
   case ISD::UMULFIX:
@@ -1246,6 +1251,14 @@ SDValue DAGTypeLegalizer::PromoteIntRes_SADDSUBO(SDNode *N, unsigned ResNo) {
   return Res;
 }
 
+SDValue DAGTypeLegalizer::PromoteIntRes_CMP(SDNode *N) {
+  EVT PromotedResultTy =
+      TLI.getTypeToTransformTo(*DAG.getContext(), N->getValueType(0));
+  SDValue Result = DAG.getNode(N->getOpcode(), SDLoc(N), PromotedResultTy,
+                               N->getOperand(0), N->getOperand(1));
+  return Result;
+}
+
 SDValue DAGTypeLegalizer::PromoteIntRes_Select(SDNode *N) {
   SDValue Mask = N->getOperand(0);
 
@@ -1874,6 +1887,9 @@ bool DAGTypeLegalizer::PromoteIntegerOperand(SDNode *N, unsigned OpNo) {
   case ISD::ROTL:
   case ISD::ROTR: Res = PromoteIntOp_Shift(N); break;
 
+  case ISD::SCMP:
+  case ISD::UCMP: Res = PromoteIntOp_CMP(N); break;
+
   case ISD::FSHL:
   case ISD::FSHR: Res = PromoteIntOp_FunnelShift(N); break;
 
@@ -2184,6 +2200,17 @@ SDValue DAGTypeLegalizer::PromoteIntOp_Shift(SDNode *N) {
                                 ZExtPromotedInteger(N->getOperand(1))), 0);
 }
 
+SDValue DAGTypeLegalizer::PromoteIntOp_CMP(SDNode *N) {
+  SDValue LHS = N->getOpcode() == ISD::UCMP
+                    ? ZExtPromotedInteger(N->getOperand(0))
+                    : SExtPromotedInteger(N->getOperand(0));
+  SDValue RHS = N->getOpcode() == ISD::UCMP
+                    ? ZExtPromotedInteger(N->getOperand(1))
+                    : SExtPromotedInteger(N->getOperand(1));
+
+  return SDValue(DAG.UpdateNodeOperands(N, LHS, RHS), 0);
+}
+
 SDValue DAGTypeLegalizer::PromoteIntOp_FunnelShift(SDNode *N) {
   return SDValue(DAG.UpdateNodeOperands(N, N->getOperand(0), N->getOperand(1),
                                 ZExtPromotedInteger(N->getOperand(2))), 0);
@@ -2741,6 +2768,9 @@ void DAGTypeLegalizer::ExpandIntegerResult(SDNode *N, unsigned ResNo) {
   case ISD::UMIN:
   case ISD::SMIN: ExpandIntRes_MINMAX(N, Lo, Hi); break;
 
+  case ISD::SCMP:
+  case ISD::UCMP: ExpandIntRes_CMP(N, Lo, Hi); break;
+
   case ISD::ADD:
   case ISD::SUB: ExpandIntRes_ADDSUB(N, Lo, Hi); break;
 
@@ -3233,6 +3263,10 @@ void DAGTypeLegalizer::ExpandIntRes_MINMAX(SDNode *N,
   SplitInteger(Result, Lo, Hi);
 }
 
+void DAGTypeLegalizer::ExpandIntRes_CMP(SDNode *N, SDValue &Lo, SDValue &Hi) {
+  assert(false && "fixme");
+}
+
 void DAGTypeLegalizer::ExpandIntRes_ADDSUB(SDNode *N,
                                            SDValue &Lo, SDValue &Hi) {
   SDLoc dl(N);
@@ -5137,6 +5171,9 @@ bool DAGTypeLegalizer::ExpandIntegerOperand(SDNode *N, unsigned OpNo) {
   case ISD::RETURNADDR:
   case ISD::FRAMEADDR:         Res = ExpandIntOp_RETURNADDR(N); break;
 
+  case ISD::SCMP:
+  case ISD::UCMP:              Res = ExpandIntOp_CMP(N); break;
+
   case ISD::ATOMIC_STORE:      Res = ExpandIntOp_ATOMIC_STORE(N); break;
   case ISD::STACKMAP:
     Res = ExpandIntOp_STACKMAP(N, OpNo);
@@ -5398,6 +5435,10 @@ SDValue DAGTypeLegalizer::ExpandIntOp_Shift(SDNode *N) {
   return SDValue(DAG.UpdateNodeOperands(N, N->getOperand(0), Lo), 0);
 }
 
+SDValue DAGTypeLegalizer::ExpandIntOp_CMP(SDNode *N) {
+  assert(false && "fixme");
+}
+
 SDValue DAGTypeLegalizer::ExpandIntOp_RETURNADDR(SDNode *N) {
   // The argument of RETURNADDR / FRAMEADDR builtin is 32 bit contant.  This
   // surely makes pretty nice problems on 8/16 bit targets. Just truncate this
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
index 4b06e19656ce6..035f16720c3f6 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
@@ -324,6 +324,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
   SDValue PromoteIntRes_Overflow(SDNode *N);
   SDValue PromoteIntRes_FFREXP(SDNode *N);
   SDValue PromoteIntRes_SADDSUBO(SDNode *N, unsigned ResNo);
+  SDValue PromoteIntRes_CMP(SDNode *N);
   SDValue PromoteIntRes_Select(SDNode *N);
   SDValue PromoteIntRes_SELECT_CC(SDNode *N);
   SDValue PromoteIntRes_SETCC(SDNode *N);
@@ -375,6 +376,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
   SDValue PromoteIntOp_SELECT_CC(SDNode *N, unsigned OpNo);
   SDValue PromoteIntOp_SETCC(SDNode *N, unsigned OpNo);
   SDValue PromoteIntOp_Shift(SDNode *N);
+  SDValue PromoteIntOp_CMP(SDNode *N);
   SDValue PromoteIntOp_FunnelShift(SDNode *N);
   SDValue PromoteIntOp_SIGN_EXTEND(SDNode *N);
   SDValue PromoteIntOp_VP_SIGN_EXTEND(SDNode *N);
@@ -457,6 +459,8 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
 
   void ExpandIntRes_MINMAX            (SDNode *N, SDValue &Lo, SDValue &Hi);
 
+  void ExpandIntRes_CMP               (SDNode *N, SDValue &Lo, SDValue &Hi);
+
   void ExpandIntRes_SADDSUBO          (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandIntRes_UADDSUBO          (SDNode *N, SDValue &Lo, SDValue &Hi);
   void ExpandIntRes_XMULO             (SDNode *N, SDValue &Lo, SDValue &Hi);
@@ -485,6 +489,7 @@ class LLVM_LIBRARY_VISIBILITY DAGTypeLegalizer {
   SDValue ExpandIntOp_SETCC(SDNode *N);
   SDValue ExpandIntOp_SETCCCARRY(SDNode *N);
   SDValue ExpandIntOp_Shift(SDNode *N);
+  SDValue ExpandIntOp_CMP(SDNode *N);
   SDValue ExpandIntOp_STORE(StoreSDNode *N, unsigned OpNo);
   SDValue ExpandIntOp_TRUNCATE(SDNode *N);
   SDValue ExpandIntOp_XINT_TO_FP(SDNode *N);
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index cfd82a342433f..16d8a9816b013 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -7143,6 +7143,22 @@ void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
     setValue(&I, DAG.getNode(ISD::ABS, sdl, Op1.getValueType(), Op1));
     return;
   }
+  case Intrinsic::scmp: {
+    SDValue Op1 = getValue(I.getArgOperand(0));
+    SDValue Op2 = getValue(I.getArgOperand(1));
+    EVT DestVT = DAG.getTargetLoweringInfo().getValueType(DAG.getDataLayout(),
+                                                          I.getType());
+    setValue(&I, DAG.getNode(ISD::SCMP, sdl, DestVT, Op1, Op2));
+    break;
+  }
+  case Intrinsic::ucmp: {
+    SDValue Op1 = getValue(I.getArgOperand(0));
+    SDValue Op2 = getValue(I.getArgOperand(1));
+    EVT DestVT = DAG.getTargetLoweringInfo().getValueType(DAG.getDataLayout(),
+                                                          I.getType());
+    setValue(&I, DAG.getNode(ISD::UCMP, sdl, DestVT, Op1, Op2));
+    break;
+  }
   case Intrinsic::stacksave: {
     SDValue Op = getRoot();
     EVT VT = TLI.getValueType(DAG.getDataLayout(), I.getType());
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
index 4ad4a938ca97f..d5e8256aa4085 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
@@ -289,6 +289,8 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
   case ISD::SMAX:                       return "smax";
   case ISD::UMIN:                       return "umin";
   case ISD::UMAX:                       return "umax";
+  case ISD::SCMP:                       return "scmp";
+  case ISD::UCMP:                       return "ucmp";
 
   case ISD::FLDEXP:                     return "fldexp";
   case ISD::STRICT_FLDEXP:              return "strict_fldexp";
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 336d89fbcf638..e3aa2f8f988ce 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -10274,6 +10274,27 @@ SDValue TargetLowering::expandAddSubSat(SDNode *Node, SelectionDAG &DAG) const {
   return DAG.getSelect(dl, VT, Overflow, Result, SumDiff);
 }
 
+SDValue TargetLowering::expandCmp(SDNode *Node, SelectionDAG &DAG) const {
+  unsigned Opcode = Node->getOpcode();
+  SDValue LHS = Node->getOperand(0);
+  SDValue RHS = Node->getOperand(1);
+  EVT VT = LHS.getValueType();
+  EVT ResVT = Node->getValueType(0);
+  EVT BoolVT = getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), VT);
+  SDLoc dl(Node);
+
+  auto LTPredicate = (Opcode == ISD::UCMP ? ISD::SETULT : ISD::SETLT);
+  auto GTPredicate = (Opcode == ISD::UCMP ? ISD::SETUGT : ISD::SETGT);
+
+  SDValue IsLT = DAG.getSetCC(dl, BoolVT, LHS, RHS, LTPredicate);
+  SDValue IsGT = DAG.getSetCC(dl, BoolVT, LHS, RHS, GTPredicate);
+  SDValue SelectZeroOrOne =
+      DAG.getSelect(dl, ResVT, IsGT, DAG.getConstant(1, dl, ResVT),
+                    DAG.getConstant(0, dl, ResVT));
+  return DAG.getSelect(dl, ResVT, IsLT, DAG.getConstant(-1, dl, ResVT),
+                       SelectZeroOrOne);
+}
+
 SDValue TargetLowering::expandShlSat(SDNode *Node, SelectionDAG &DAG) const {
   unsigned Opcode = Node->getOpcode();
   bool IsSigned = Opcode == ISD::SSHLSAT;
diff --git a/llvm/lib/CodeGen/TargetLoweringBase.cpp b/llvm/lib/CodeGen/TargetLoweringBase.cpp
index 6e7b67ded23c8..311ba68b96ec9 100644
--- a/llvm/lib/CodeGen/TargetLoweringBase.cpp
+++ b/llvm/lib/CodeGen/TargetLoweringBase.cpp
@@ -909,6 +909,9 @@ void TargetLoweringBase::initActions() {
     setOperationAction({ISD::ADDC, ISD::ADDE, ISD::SUBC, ISD::SUBE}, VT,
                        Expand);
 
+    // [US]CMP default to expand
+    setOperationAction({ISD::UCMP, ISD::SCMP}, VT, Expand);
+
     // Halving adds
     setOperationAction(
         {ISD::AVGFLOORS, ISD::AVGFLOORU, ISD::AVGCEILS, ISD::AVGCEILU}, VT,
diff --git a/llvm/test/CodeGen/X86/uscmp.ll b/llvm/test/CodeGen/X86/uscmp.ll
new file mode 100644
index 0000000000000..d1c504e972d46
--- /dev/null
+++ b/llvm/test/CodeGen/X86/uscmp.ll
@@ -0,0 +1,163 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -show-mc-encoding | FileCheck %s --check-prefixes=CHECK,NO-NDD
+
+define i8 @ucmp(i32 %x, i32 %y) {
+; CHECK-LABEL: ucmp:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    xorl %ecx, %ecx # encoding: [0x31,0xc9]
+; CHECK-NEXT:    cmpl %esi, %edi # encoding: [0x39,0xf7]
+; CHECK-NEXT:    seta %cl # encoding: [0x0f,0x97,0xc1]
+; CHECK-NEXT:    movl $255, %eax # encoding: [0xb8,0xff,0x00,0x00,0x00]
+; CHECK-NEXT:    cmovael %ecx, %eax # encoding: [0x0f,0x43,0xc1]
+; CHECK-NEXT:    # kill: def $al killed $al killed $eax
+; CHECK-NEXT:    retq # encoding: [0xc3]
+; NDD-LABEL: ucmp:
+; NDD:       # %bb.0:
+; NDD-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
+; NDD-NEXT:    cmpl %esi, %edi # encoding: [0x39,0xf7]
+; NDD-NEXT:    seta %al # encoding: [0x0f,0x97,0xc0]
+; NDD-NEXT:    movl $255, %ecx # encoding: [0xb9,0xff,0x00,0x00,0x00]
+; NDD-NEXT:    cmovbl %ecx, %eax # EVEX TO LEGACY Compression encoding: [0x0f,0x42,0xc1]
+; NDD-NEXT:    # kill: def $al killed $al killed $eax
+; NDD-NEXT:    retq # encoding: [0xc3]
+  %1 = call i8 @llvm.ucmp(i32 %x, i32 %y)
+  ret i8 %1
+}
+
+define i8 @scmp(i32 %x, i32 %y) {
+; CHECK-LABEL: scmp:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    xorl %ecx, %ecx # encoding: [0x31,0xc9]
+; CHECK-NEXT:    cmpl %esi, %edi # encoding: [0x39,0xf7]
+; CHECK-NEXT:    seta %cl # encoding: [0x0f,0x97,0xc1]
+; CHECK-NEXT:    movl $255, %eax # encoding: [0xb8,0xff,0x00,0x00,0x00]
+; CHECK-NEXT:    cmovael %ecx, %eax # encoding: [0x0f,0x43,0xc1]
+; CHECK-NEXT:    # kill: def $al killed $al killed $eax
+; CHECK-NEXT:    retq # encoding: [0xc3]
+; NDD-LABEL: scmp:
+; NDD:       # %bb.0:
+; NDD-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
+; NDD-NEXT:    cmpl %esi, %edi # encoding: [0x39,0xf7]
+; NDD-NEXT:    seta %al # encoding: [0x0f,0x97,0xc0]
+; NDD-NEXT:    movl $255, %ecx # encoding: [0xb9,0xff,0x00,0x00,0x00]
+; NDD-NEXT:    cmovbl %ecx, %eax # EVEX TO LEGACY Compression encoding: [0x0f,0x42,0xc1]
+; NDD-NEXT:    # kill: def $al killed $al killed $eax
+; NDD-NEXT:    retq # encoding: [0xc3]
+  %1 = call i8 @llvm.ucmp(i32 %x, i32 %y)
+  ret i8 %1
+}
+
+define i4 @ucmp_narrow_result(i32 %x, i32 %y) {
+; CHECK-LABEL: ucmp_narrow_result:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    xorl %ecx, %ecx # encoding: [0x31,0xc9]
+; CHECK-NEXT:    cmpl %esi, %edi # encoding: [0x39,0xf7]
+; CHECK-NEXT:    seta %cl # encoding: [0x0f,0x97,0xc1]
+; CHECK-NEXT:    movl $255, %eax # encoding: [0xb8,0xff,0x00,0x00,0x00]
+; CHECK-NEXT:    cmovael %ecx, %eax # encoding: [0x0f,0x43,0xc1]
+; CHECK-NEXT:    # kill: def $al killed $al killed $eax
+; CHECK-NEXT:    retq # encoding: [0xc3]
+; NDD-LABEL: ucmp_narrow_result:
+; NDD:       # %bb.0:
+; NDD-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
+; NDD-NEXT:    cmpl %esi, %edi # encoding: [0x39,0xf7]
+; NDD-NEXT:    seta %al # encoding: [0x0f,0x97,0xc0]
+; NDD-NEXT:    movl $255, %ecx # encoding: [0xb9,0xff,0x00,0x00,0x00]
+; NDD-NEXT:    cmovbl %ecx, %eax # EVEX TO LEGACY Compression encoding: [0x0f,0x42,0xc1]
+; NDD-NEXT:    # kill: def $al killed $al killed $eax
+; NDD-NEXT:    retq # encoding: [0xc3]
+  %1 = call i4 @llvm.ucmp(i32 %x, i32 %y)
+  ret i4 %1
+}
+
+define i8 @scmp_narrow_op(i5 %x, i5 %y) {
+; CHECK-LABEL: scmp_narrow_op:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    andb $31, %sil # encoding: [0x40,0x80,0xe6,0x1f]
+; CHECK-NEXT:    andb $31, %dil # encoding: [0x40,0x80,0xe7,0x1f]
+; CHECK-NEXT:    xorl %ecx, %ecx # encoding: [0x31,0xc9]
+; CHECK-NEXT:    cmpb %sil, %dil # encoding: [0x40,0x38,0xf7]
+; CHECK-NEXT:    seta %cl # encoding: [0x0f,0x97,0xc1]
+; CHECK-NEXT:    movl $255, %eax # encoding: [0xb8,0xff,0x00,0x00,0x00]
+; CHECK-NEXT:    cmovael %ecx, %eax # encoding: [0x0f,0x43,0xc1]
+; CHECK-NEXT:    # kill: def $al killed $al killed $eax
+; CHECK-NEXT:    retq # encoding: [0xc3]
+; NDD-LABEL: scmp_narrow_op:
+; NDD:       # %bb.0:
+; NDD-NEXT:    andb $31, %sil, %al # encoding: [0x62,0xf4,0x7c,0x18,0x80,0xe6,0x1f]
+; NDD-NEXT:    andb $31, %dil, %cl # encoding: [0x62,0xf4,0x74,0x18,0x80,0xe7,0x1f]
+; NDD-NEXT:    xorl %edx, %edx # encoding: [0x31,0xd2]
+; NDD-NEXT:    cmpb %al, %cl # encoding: [0x38,0xc1]
+; NDD-NEXT:    seta %dl # encoding: [0x0f,0x97,0xc2]
+; NDD-NEXT:    movl $255, %eax # encoding: [0xb8,0xff,0x00,0x00,0x00]
+; NDD-NEXT:    cmovael %edx, %eax # EVEX TO LEGACY Compression encoding: [0x0f,0x43,0xc2]
+; NDD-NEXT:    # kill: def $al killed $al killed $eax
+; NDD-NEXT:    retq # encoding: [0xc3]
+  %1 = call i8 @llvm.ucmp(i5 %x, i5 %y)
+  ret i8 %1
+}
+
+define i41 @ucmp_uncommon_types(i7 %x, i7 %y) {
+; CHECK-LABEL: ucmp_uncommon_types:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    andb $127, %sil # encoding: [0x40,0x80,0xe6,0x7f]
+; CHECK-NEXT:    andb $127, %dil # encoding: [0x40,0x80,0xe7,0x7f]
+; CHECK-NEXT:    xorl %ecx, %ecx # encoding: [0x31,0xc9]
+; CHECK-NEXT:    cmpb %sil, %dil # encoding: [0x40,0x38,0xf7]
+; CHECK-NEXT:    seta %cl # encoding: [0x0f,0x97,0xc1]
+; CHECK-NEXT:    movq $-1, %rax # encoding: [0x48,0xc7,0xc0,0xff,0xff,0xff,0xff]
+; CHECK-NEXT:    cmovaeq %rcx, %rax # encoding: [0x48,0x0f,0x43,0xc1]
+; CHECK-NEXT:    retq # encoding: [0xc3]
+; NDD-LABEL: ucmp_uncommon_types:
+; NDD:       # %bb.0:
+; NDD-NEXT:    andb $127, %sil, %al # encoding: [0x62,0xf4,0x7c,0x18,0x80,0xe6,0x7f]
+; NDD-NEXT:    andb $127, %dil, %cl # encoding: [0x62,0xf4,0x74,0x18,0x80,0xe7,0x7f]
+; NDD-NEXT:    xorl %edx, %edx # encoding: [0x31,0xd2]
+; NDD-NEXT:    cmpb %al, %cl # encoding: [0x38,0xc1]
+; NDD-NEXT:    seta %dl # encoding: [0x0f,0x97,0xc2]
+; NDD-NEXT:    movq $-1, %rax # encoding: [0x48,0xc7,0xc0,0xff,0xff,0xff,0xff]
+; NDD-NEXT:    cmovaeq %rdx, %rax # EVEX TO LEGACY Compression encoding: [0x48,0x0f,0x43,0xc2]
+; NDD-NEXT:    retq # encoding: [0xc3]
+  %1 = call i41 @llvm.ucmp(i7 %x, i7 %y)
+  ret i41 %1
+}
+
+define i13 @scmp_uncommon_types(i11 %x, i11 %y) {
+; CHECK-LABEL: scmp_uncommon_types:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    shll $5, %esi # encoding: [0xc1,0xe6,0x05]
+; CHECK-NEXT:    movswl %si, %eax # encoding: [0x0f,0xbf,0xc6]
+; CHECK-NEXT:    shrl $5, %eax # encoding: [0xc1,0xe8,0x05]
+; CHECK-NEXT:    shll $5, %edi # encoding: [0xc1,0xe7,0x05]
+; CHECK-NEXT:    movswl %di, %ecx # encoding: [0x0f,0xbf,0xcf]
+; CHECK-NEXT:    shrl $5, %ecx # encoding: [0xc1,0xe9,0x05]
+; CHECK-NEXT:    xorl %edx, %edx # encoding: [0x31,0xd2]
+; CHECK-NEXT:    cmpw %ax, %cx # encoding: [0x66,0x39,0xc1]
+; CHECK-NEXT:    setg %dl # encoding: [0x0f,0x9f,0xc2]
+; CHECK-NEXT:    movl $65535, %eax # encoding: [0xb8,0xff,0xff,0x00,0x00]
+; CHECK-NEXT:    # imm = 0xFFFF
+; CHECK-NEXT:    cmovgel %edx, %eax # encoding: [0x0f,0x4d,0xc2]
+; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
+; CHECK-NEXT:    retq # encoding: [0xc3]
+; NDD-LABEL: scmp_uncommon_types:
+; NDD:       # %bb.0:
+; NDD-NEXT:    shll $5, %esi, %eax # encoding: [0x62,0xf4,0x7c,0x18,0xc1,0xe6,0x05]
+; NDD-NEXT:    cwtl # encoding: [0x98]
+; NDD-NEXT:    shrl $5, %eax # EVEX TO LEGACY Compression encoding: [0xc1,0xe8,0x05]
+; NDD-NEXT:    shll $5, %edi, %ecx # encoding: [0x62,0xf4,0x74,0x18,0xc1,0xe7,0x05]
+; NDD-NEXT:    movswl %cx, %ecx # encoding: [0x0f,0xbf,0xc9]
+; NDD-NEXT:    shrl $5, %ecx # EVEX TO LEGACY Compression encoding: [0xc1,0xe9,0x05]
+; NDD-NEXT:    xorl %edx, %edx # encoding: [0x31,0xd2]
+; NDD-NEXT:    cmpw %ax, %cx # encoding: [0x66,0x39,0xc1]
+; NDD-NEXT:    setg %dl # encoding: [0x0f,0x9f,0xc2]
+; NDD-NEXT:    movl $65535, %eax # encoding: [0xb8,0xff,0xff,0x00,0x00]
+; NDD-NEXT:    # imm = 0xFFFF
+; NDD-NEXT:    cmovgel %edx, %eax # EVEX TO LEGACY Compression encoding: [0x0f,0x4d,0xc2]
+; NDD-NEXT:    # kill: def $ax killed $ax killed $eax
+; NDD-NEXT:    retq # encoding: [0xc3]
+  %1 = call i13 @llvm.scmp(i11 %x, i11 %y)
+  ret i13 %1
+}
+
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; NO-NDD: {{.*}}



More information about the llvm-commits mailing list