[ARM] Add some basic Min/Max costs

This adds basic MVE costs for SMIN/SMAX/UMIN/UMAX, as well as MINNUM and
MAXNUM representing fmin and fmax. It tightens up the costs, not using a
ICmp+Select cost.

Differential Revision: https://reviews.llvm.org/D96603

llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp

@@ -1574,6 +1574,30 @@ int ARMTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
     }
     break;
   }
+  case Intrinsic::smin:
+  case Intrinsic::smax:
+  case Intrinsic::umin:
+  case Intrinsic::umax: {
+    if (!ST->hasMVEIntegerOps())
+      break;
+    Type *VT = ICA.getReturnType();
+
+    std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, VT);
+    if (LT.second == MVT::v4i32 || LT.second == MVT::v8i16 ||
+        LT.second == MVT::v16i8)
+      return LT.first * ST->getMVEVectorCostFactor(CostKind);
+    break;
+  }
+  case Intrinsic::minnum:
+  case Intrinsic::maxnum: {
+    if (!ST->hasMVEFloatOps())
+      break;
+    Type *VT = ICA.getReturnType();
+    std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, VT);
+    if (LT.second == MVT::v4f32 || LT.second == MVT::v8f16)
+      return LT.first * ST->getMVEVectorCostFactor(CostKind);
+    break;
+  }
   }
 
   return BaseT::getIntrinsicInstrCost(ICA, CostKind);

llvm/test/Analysis/CostModel/ARM/intrinsic-cost-kinds.ll

@@ -43,7 +43,7 @@ declare void @llvm.memcpy.p0i8.p0i8.i32(i8*, i8*, i32, i1)
 define void @smax(i32 %a, i32 %b, <16 x i32> %va, <16 x i32> %vb) {
 ; THRU-LABEL: 'smax'
 ; THRU-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s = call i32 @llvm.smax.i32(i32 %a, i32 %b)
-; THRU-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v = call <16 x i32> @llvm.smax.v16i32(<16 x i32> %va, <16 x i32> %vb)
+; THRU-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v = call <16 x i32> @llvm.smax.v16i32(<16 x i32> %va, <16 x i32> %vb)
 ; THRU-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; LATE-LABEL: 'smax'
@@ -53,12 +53,12 @@ define void @smax(i32 %a, i32 %b, <16 x i32> %va, <16 x i32> %vb) {
 ;
 ; SIZE-LABEL: 'smax'
 ; SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %s = call i32 @llvm.smax.i32(i32 %a, i32 %b)
-; SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v = call <16 x i32> @llvm.smax.v16i32(<16 x i32> %va, <16 x i32> %vb)
+; SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v = call <16 x i32> @llvm.smax.v16i32(<16 x i32> %va, <16 x i32> %vb)
 ; SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
 ; SIZE_LATE-LABEL: 'smax'
 ; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %s = call i32 @llvm.smax.i32(i32 %a, i32 %b)
-; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v = call <16 x i32> @llvm.smax.v16i32(<16 x i32> %va, <16 x i32> %vb)
+; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v = call <16 x i32> @llvm.smax.v16i32(<16 x i32> %va, <16 x i32> %vb)
 ; SIZE_LATE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
 ;
   %s = call i32 @llvm.smax.i32(i32 %a, i32 %b)