Index: llvm/include/llvm/CodeGen/BasicTTIImpl.h
===================================================================
--- llvm/include/llvm/CodeGen/BasicTTIImpl.h
+++ llvm/include/llvm/CodeGen/BasicTTIImpl.h
@@ -1921,6 +1921,32 @@
           BinaryOperator::ICmp, MulTy, OverflowTy, CmpInst::ICMP_NE, CostKind);
       return Cost;
     }
+    case Intrinsic::fptosi_sat:
+    case Intrinsic::fptoui_sat: {
+      if (Tys.empty())
+        break;
+      Type *FromTy = Tys[0];
+      bool IsSigned = IID == Intrinsic::fptosi_sat;
+
+      InstructionCost Cost = 0;
+      IntrinsicCostAttributes Attrs1(Intrinsic::minnum, FromTy,
+                                     {FromTy, FromTy});
+      Cost += thisT()->getIntrinsicInstrCost(Attrs1, CostKind);
+      IntrinsicCostAttributes Attrs2(Intrinsic::maxnum, FromTy,
+                                     {FromTy, FromTy});
+      Cost += thisT()->getIntrinsicInstrCost(Attrs2, CostKind);
+      Cost += thisT()->getCastInstrCost(
+          IsSigned ? Instruction::FPToSI : Instruction::FPToUI, RetTy, FromTy,
+          TTI::CastContextHint::None, CostKind);
+      if (IsSigned) {
+        Type *CondTy = RetTy->getWithNewBitWidth(1);
+        Cost += thisT()->getCmpSelInstrCost(
+            BinaryOperator::ICmp, FromTy, CondTy, CmpInst::FCMP_UNO, CostKind);
+        Cost += thisT()->getCmpSelInstrCost(
+            BinaryOperator::Select, RetTy, CondTy, CmpInst::FCMP_UNO, CostKind);
+      }
+      return Cost;
+    }
     case Intrinsic::ctpop:
       ISDs.push_back(ISD::CTPOP);
       // In case of legalization use TCC_Expensive. This is cheaper than a
Index: llvm/test/Analysis/CostModel/AArch64/fptoi_sat.ll
===================================================================
--- llvm/test/Analysis/CostModel/AArch64/fptoi_sat.ll
+++ llvm/test/Analysis/CostModel/AArch64/fptoi_sat.ll
@@ -4,105 +4,105 @@
 
 define void @casts() {
 ; CHECK-LABEL: 'casts'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 184 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 172 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 107 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 360 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 172 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 103 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 124 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 55 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 78 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 368 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
@@ -220,58 +220,111 @@
 }
 
 define void @fp16() {
-; CHECK-LABEL: 'fp16'
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+; CHECK-NOFP16-LABEL: 'fp16'
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 172 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 79 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 201 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 143 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 224 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 187 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 366 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 177 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 504 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 385 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 504 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 385 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 360 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 286 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 448 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 374 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 732 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 354 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-NOFP16-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; CHECK-FP16-LABEL: 'fp16'
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 172 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 109 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 74 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 400 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 78 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-FP16-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
   %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
Index: llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll
===================================================================
--- llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll
+++ llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll
@@ -4,209 +4,209 @@
 
 define void @casts() {
 ; CHECK-MVE-LABEL: 'casts'
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 124 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 148 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 106 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 378 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 306 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 268 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 268 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 268 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 376 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 334 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 218 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 536 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 498 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 536 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 498 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 536 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 496 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1348 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1304 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 906 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 794 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 794 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 794 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1336 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1296 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 690 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 458 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1652 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1578 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1652 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1576 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1648 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1568 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4888 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4800 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2986 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2762 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2828 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2762 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2828 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2760 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2824 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2752 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4848 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4768 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; CHECK-MVEFP-LABEL: 'casts'
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 284 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 278 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 306 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 268 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 268 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 268 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 250 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 396 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 376 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1148 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1104 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 906 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 794 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 794 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 794 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 828 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 792 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1336 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1296 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 290 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4488 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4400 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2986 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2762 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2828 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2762 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2828 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2760 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2824 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 2752 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4848 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4768 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
@@ -324,109 +324,109 @@
 
 define void @fp16() {
 ; CHECK-MVE-LABEL: 'fp16'
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 124 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 148 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 106 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 378 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 292 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 210 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 356 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 352 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 356 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 352 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 504 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 498 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1316 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1306 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 654 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 426 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1064 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 994 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1064 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 992 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1652 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1576 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4892 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4808 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; CHECK-MVEFP-LABEL: 'fp16'
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 284 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 278 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1112 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1102 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 246 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 124 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4484 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4400 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
Index: llvm/test/Analysis/CostModel/X86/fptoi_sat.ll
===================================================================
--- llvm/test/Analysis/CostModel/X86/fptoi_sat.ll
+++ llvm/test/Analysis/CostModel/X86/fptoi_sat.ll
@@ -1,8 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
-; RUN: opt < %s -mtriple=x86_64-apple-darwin -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx  | FileCheck %s --check-prefixes=AVX,AVX1
-; RUN: opt < %s -mtriple=x86_64-apple-darwin -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
+; RUN: opt < %s -mtriple=x86_64-apple-darwin -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx  | FileCheck %s --check-prefixes=AVX1
+; RUN: opt < %s -mtriple=x86_64-apple-darwin -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=AVX512,AVX512DQ
 ;
@@ -12,519 +12,828 @@
 
 define void @casts() {
 ; SSE2-LABEL: 'casts'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 74 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 108 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 148 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 131 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 131 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 126 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 124 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 216 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 352 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SSE42-LABEL: 'casts'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 160 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
-; AVX-LABEL: 'casts'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX1-LABEL: 'casts'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 55 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
-; AVX512-LABEL: 'casts'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX2-LABEL: 'casts'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 55 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512F-LABEL: 'casts'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 73 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'casts'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'casts'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 104 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 80 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 208 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX-LABEL: 'casts'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f32(<2 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f32u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f32(<2 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f32u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f32(<2 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f32(<2 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f32u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f32(<2 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f32(<2 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f32u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f32(<2 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f32s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f32(<2 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f32u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f32(<2 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f64(<2 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f64u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f64(<2 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f64(<2 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v2f64u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f64(<2 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f64(<2 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v2f64u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f64(<2 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v2f64s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f64(<2 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v2f64u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f64(<2 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f64s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f64(<2 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f64u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f64(<2 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f32(<4 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f32(<4 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f32s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f32(<4 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f32u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f32(<4 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f32(<4 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %v4f32u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f32(<4 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f32(<4 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v4f32u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f32(<4 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f32s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f32(<4 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v4f32u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f32(<4 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f64s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f64(<4 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f64u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f64(<4 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f64s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f64(<4 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f64u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f64(<4 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f64s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f64(<4 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v4f64u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f64(<4 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v4f64s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f64(<4 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v4f64u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f64(<4 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f64s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f64(<4 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v4f64u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f64(<4 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f32(<8 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v8f32s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f32(<8 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f32(<8 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v8f32s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f32(<8 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v8f32u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f32(<8 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v8f32s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f32(<8 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v8f32u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f32(<8 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v8f32s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f32(<8 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %v8f32u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f32(<8 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v8f64s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f64(<8 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f64u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f64(<8 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v8f64s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f64(<8 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f64u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f64(<8 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v8f64s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f64(<8 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v8f64u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f64(<8 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v8f64s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f64(<8 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v8f64u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f64(<8 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v8f64s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f64(<8 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v8f64u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f64(<8 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f32(<16 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v16f32s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f32(<16 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f32(<16 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v16f32s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f32(<16 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v16f32u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f32(<16 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v16f32s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f32(<16 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f32u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f32(<16 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v16f32s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f32(<16 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 134 for instruction: %v16f32u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f32(<16 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v16f64s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f64(<16 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f64u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f64(<16 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v16f64s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f64(<16 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f64u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f64(<16 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 55 for instruction: %v16f64s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f64(<16 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v16f64u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f64(<16 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v16f64s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f64(<16 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v16f64u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f64(<16 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
   %f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
@@ -641,269 +950,375 @@
 
 define void @fp16() {
 ; SSE2-LABEL: 'fp16'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 90 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 79 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 108 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 107 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 159 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 230 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 211 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 236 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 214 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 232 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 240 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 200 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SSE42-LABEL: 'fp16'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 79 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 91 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 78 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 159 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 163 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 184 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 166 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 156 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 200 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
-; AVX-LABEL: 'fp16'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+; AVX1-LABEL: 'fp16'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 127 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 149 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 151 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 126 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 172 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX2-LABEL: 'fp16'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 37 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 67 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 74 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 78 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 127 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 147 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 147 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 130 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 126 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 156 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512-LABEL: 'fp16'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 127 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 97 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 97 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'fp16'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 41 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 79 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 91 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 82 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 83 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 78 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 159 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 163 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 184 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 166 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 156 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 176 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 200 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX-LABEL: 'fp16'
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 77 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 68 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 86 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 127 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 149 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 151 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 126 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 172 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
   %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)