[ARM] Cost modelling for MVE vector fptoi_sat

[dmgreen]

Summary

Building on top of D125665, this adds MVE costs for fptosi.sat and fptoui.sat, providing MVE is available and the types are legal.

[SjoerdMeijer]

LGTM

llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp

case Intrinsic::fptoui_sat: {
     auto LT = TLI->getTypeLegalizationCost(DL, ICA.getArgTypes()[0]);
     EVT MTy = TLI->getValueType(DL, ICA.getReturnType());
     // Check for the legal types, with the corect subtarget features.
     if ((ST->hasVFP2Base() && LT.second == MVT::f32 && MTy == MVT::i32) ||
         (ST->hasFP64() && LT.second == MVT::f64 && MTy == MVT::i32) ||
         (ST->hasFullFP16() && LT.second == MVT::f16 && MTy == MVT::i32))
       return LT.first;
 
+    // Equally for MVE vector types
+    if (ST->hasMVEFloatOps() &&
+        (LT.second == MVT::v4f32 || LT.second == MVT::v8f16) &&
+        LT.second.getScalarSizeInBits() == MTy.getScalarSizeInBits())
+      return LT.first * ST->getMVEVectorCostFactor(CostKind);
+
     // Otherwise we use a legal convert followed by a min+max
     if (((ST->hasVFP2Base() && LT.second == MVT::f32) ||
          (ST->hasFP64() && LT.second == MVT::f64) ||
-         (ST->hasFullFP16() && LT.second == MVT::f16)) &&
+         (ST->hasFullFP16() && LT.second == MVT::f16) ||
+         (ST->hasMVEFloatOps() &&
+          (LT.second == MVT::v4f32 || LT.second == MVT::v8f16))) &&
         LT.second.getScalarSizeInBits() >= MTy.getScalarSizeInBits()) {
       Type *LegalTy = Type::getIntNTy(ICA.getReturnType()->getContext(),
                                       LT.second.getScalarSizeInBits());
-      InstructionCost Cost = 1;
+      InstructionCost Cost =
+          LT.second.isVector() ? ST->getMVEVectorCostFactor(CostKind) : 1;
       IntrinsicCostAttributes Attrs1(IsSigned ? Intrinsic::smin
                                               : Intrinsic::umin,
                                      LegalTy, {LegalTy, LegalTy});
       Cost += getIntrinsicInstrCost(Attrs1, CostKind);
       IntrinsicCostAttributes Attrs2(IsSigned ? Intrinsic::smax
                                               : Intrinsic::umax,
                                      LegalTy, {LegalTy, LegalTy});
       Cost += getIntrinsicInstrCost(Attrs2, CostKind);

llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll

 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 7 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 7 for instruction: %f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 7 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 6 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 6 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 6 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 6 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 6 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 6 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 2 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 2 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 80 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 52 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 62 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 52 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 62 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 52 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 62 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 52 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 94 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 84 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: %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: %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: %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 2 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 2 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 234 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 178 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 196 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 178 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 196 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 178 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 196 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 178 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 324 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 304 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 12 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 12 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 12 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 12 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 12 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 12 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 4 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 4 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 762 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 650 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 684 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 650 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 684 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 650 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 684 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 648 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 1192 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 1152 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 24 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 24 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 24 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 24 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 24 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 24 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 8 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 8 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 2698 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 2474 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 2540 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 2474 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 2540 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 2472 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 5 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 5 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 5 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 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 6 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 6 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 6 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 6 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 2 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 2 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: %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: %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 2 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 2 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: %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: %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 2 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 2 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 12 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 12 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 12 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 12 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 4 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 4 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)
   %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)