[SLP][TLI] Add vectorization support for [u|s]cmp

Author: dtcxzyw

llvm/include/llvm/CodeGen/BasicTTIImpl.h

@@ -2196,6 +2196,12 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
     case Intrinsic::bitreverse:
       ISD = ISD::BITREVERSE;
       break;
+    case Intrinsic::ucmp:
+      ISD = ISD::UCMP;
+      break;
+    case Intrinsic::scmp:
+      ISD = ISD::SCMP;
+      break;
     }
 
     auto *ST = dyn_cast<StructType>(RetTy);
@@ -2433,6 +2439,37 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
       }
       return Cost;
     }
+    case Intrinsic::ucmp:
+    case Intrinsic::scmp: {
+      InstructionCost Cost = 0;
+      bool IsSigned = IID == Intrinsic::scmp;
+      ICmpInst::Predicate GTPred =
+          IsSigned ? CmpInst::ICMP_SGT : CmpInst::ICMP_UGT;
+      ICmpInst::Predicate LTPred =
+          IsSigned ? CmpInst::ICMP_SLT : CmpInst::ICMP_ULT;
+      Type *CmpTy = Tys[0];
+
+      if (TLI->shouldExpandCmpUsingSelects()) {
+        // x < y ? -1 : (x > y ? 1 : 0)
+        Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, CmpTy, RetTy,
+                                            GTPred, CostKind);
+        Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, CmpTy, RetTy,
+                                            LTPred, CostKind);
+      } else {
+        // zext(x > y) - zext(x < y)
+        Type *CondTy = RetTy->getWithNewBitWidth(1);
+        Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, CmpTy, CondTy,
+                                            GTPred, CostKind);
+        Cost += thisT()->getCmpSelInstrCost(BinaryOperator::ICmp, CmpTy, CondTy,
+                                            LTPred, CostKind);
+        Cost +=
+            2 * thisT()->getCastInstrCost(CastInst::ZExt, RetTy, CondTy,
+                                          TTI::CastContextHint::None, CostKind);
+        Cost += thisT()->getArithmeticInstrCost(BinaryOperator::Sub, RetTy,
+                                                CostKind);
+      }
+      return Cost;
+    }
     default:
       break;
     }

llvm/lib/Analysis/VectorUtils.cpp

@@ -97,6 +97,8 @@ bool llvm::isTriviallyVectorizable(Intrinsic::ID ID) {
   case Intrinsic::fptoui_sat:
   case Intrinsic::lrint:
   case Intrinsic::llrint:
+  case Intrinsic::ucmp:
+  case Intrinsic::scmp:
     return true;
   default:
     return false;
@@ -132,6 +134,8 @@ bool llvm::isVectorIntrinsicWithOverloadTypeAtArg(Intrinsic::ID ID,
   case Intrinsic::fptoui_sat:
   case Intrinsic::lrint:
   case Intrinsic::llrint:
+  case Intrinsic::ucmp:
+  case Intrinsic::scmp:
     return OpdIdx == -1 || OpdIdx == 0;
   case Intrinsic::is_fpclass:
     return OpdIdx == 0;