[llvm-branch-commits] [llvm] TTI: Check legalization cost of fptosi_sat/fptoui_sat nodes (PR #100521)

Sun, 28 Jul 2024 06:48:55 -0700 

https://github.com/arsenm updated 
https://github.com/llvm/llvm-project/pull/100521

>From 19f7331a579837b2657a5d0741c6633d6f8296da Mon Sep 17 00:00:00 2001
From: Matt Arsenault <matthew.arsena...@amd.com>
Date: Thu, 25 Jul 2024 10:33:23 +0400
Subject: [PATCH] TTI: Check legalization cost of fptosi_sat/fptoui_sat nodes

---
 llvm/include/llvm/CodeGen/BasicTTIImpl.h      |  56 +--
 llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll | 116 ++---
 llvm/test/Analysis/CostModel/X86/fptoi_sat.ll | 400 +++++++++---------
 .../AggressiveInstCombine/ARM/fptosisat.ll    |  49 ++-
 4 files changed, 324 insertions(+), 297 deletions(-)

diff --git a/llvm/include/llvm/CodeGen/BasicTTIImpl.h 
b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
index 1a089a3fa9634..ba70498bfb731 100644
--- a/llvm/include/llvm/CodeGen/BasicTTIImpl.h
+++ b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
@@ -2179,31 +2179,11 @@ class BasicTTIImplBase : public 
TargetTransformInfoImplCRTPBase<T> {
       ISD = ISD::UMULO;
       break;
     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::FCmp, FromTy, CondTy, CmpInst::FCMP_UNO, CostKind);
-        Cost += thisT()->getCmpSelInstrCost(
-            BinaryOperator::Select, RetTy, CondTy, CmpInst::FCMP_UNO, 
CostKind);
-      }
-      return Cost;
-    }
+      ISD = ISD::FP_TO_SINT_SAT;
+      break;
+    case Intrinsic::fptoui_sat:
+      ISD = ISD::FP_TO_UINT_SAT;
+      break;
     case Intrinsic::ctpop:
       ISD = ISD::CTPOP;
       // In case of legalization use TCC_Expensive. This is cheaper than a
@@ -2418,6 +2398,32 @@ class BasicTTIImplBase : public 
TargetTransformInfoImplCRTPBase<T> {
       Cost += thisT()->getArithmeticInstrCost(Instruction::Or, RetTy, 
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::FCmp, FromTy, CondTy, CmpInst::FCMP_UNO, CostKind);
+        Cost += thisT()->getCmpSelInstrCost(
+            BinaryOperator::Select, RetTy, CondTy, CmpInst::FCMP_UNO, 
CostKind);
+      }
+      return Cost;
+    }
     default:
       break;
     }
diff --git a/llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll 
b/llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll
index aff7b19a9c87a..29c86fc778a98 100644
--- a/llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll
+++ b/llvm/test/Analysis/CostModel/ARM/fptoi_sat.ll
@@ -4,26 +4,26 @@
 
 define void @casts() {
 ; CHECK-MVE-LABEL: 'casts'
-; 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 2 for instruction: 
%f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: 
%f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: 
%f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 120 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 102 for instruction: 
%v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
@@ -115,8 +115,8 @@ define void @casts() {
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 5 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 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 4 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: 
%f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: 
%f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: 
%f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
@@ -159,12 +159,12 @@ define void @casts() {
 ; 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 2 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 2 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 2 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 2 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 2 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 2 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 12 for 
instruction: %v8f32s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f32(<8 x float> 
undef)
@@ -179,12 +179,12 @@ define void @casts() {
 ; 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 2 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 2 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 2 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 2 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 4 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 4 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 24 for 
instruction: %v16f32s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f32(<16 x 
float> undef)
@@ -199,12 +199,12 @@ define void @casts() {
 ; 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 2536 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 2464 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 2 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 2 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 4 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 4 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 8 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 8 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 4560 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 4480 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
@@ -324,16 +324,16 @@ define void @casts() {
 
 define void @fp16() {
 ; CHECK-MVE-LABEL: 'fp16'
-; 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 2 for instruction: 
%f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 120 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 102 for instruction: 
%v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
@@ -385,8 +385,8 @@ define void @fp16() {
 ; 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 4 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(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)
@@ -403,8 +403,8 @@ define void @fp16() {
 ; 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 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 2 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 2 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 6 for instruction: 
%v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
@@ -413,8 +413,8 @@ define void @fp16() {
 ; 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 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 4 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 4 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 12 for 
instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x 
half> undef)
@@ -423,8 +423,8 @@ define void @fp16() {
 ; 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 8 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 8 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
diff --git a/llvm/test/Analysis/CostModel/X86/fptoi_sat.ll 
b/llvm/test/Analysis/CostModel/X86/fptoi_sat.ll
index 3f5c79f2d59c6..55b80350f595e 100644
--- a/llvm/test/Analysis/CostModel/X86/fptoi_sat.ll
+++ b/llvm/test/Analysis/CostModel/X86/fptoi_sat.ll
@@ -12,26 +12,26 @@
 
 define void @casts() {
 ; SSE2-LABEL: 'casts'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: 
%f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: 
%f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: 
%f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: 
%f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: 
%f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: 
%f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: 
%f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: 
%f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: 
%f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: 
%f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: 
%f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: 
%f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: 
%f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: 
%f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32s1 
= call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32u1 
= call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32s8 
= call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32u8 
= call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s1 
= call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u1 
= call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s8 
= call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u8 
= call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 19 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 18 for instruction: 
%v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
@@ -115,26 +115,26 @@ define void @casts() {
 ; 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 13 for instruction: 
%f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: 
%f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 17 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 13 for instruction: 
%v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
@@ -218,26 +218,26 @@ define void @casts() {
 ; SSE42-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 2 for instruction: %f32s1 
= call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32u1 
= call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32s8 
= call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32u8 
= call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s1 
= call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u1 
= call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s8 
= call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u8 
= call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
 ; AVX1-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)
 ; 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 11 for instruction: 
%v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
@@ -321,26 +321,26 @@ define void @casts() {
 ; AVX1-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 7 for instruction: %f32s1 
= call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u1 
= call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f32s8 
= call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f32u8 
= call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: 
%f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: 
%f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s1 
= call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u1 
= call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %f64s8 
= call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f64u8 
= call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: 
%f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: 
%f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: 
%f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: 
%f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32s1 
= call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32u1 
= call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32s8 
= call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32u8 
= call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s1 
= call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u1 
= call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s8 
= call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u8 
= call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 13 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 10 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 9 for instruction: 
%v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
@@ -424,26 +424,26 @@ define void @casts() {
 ; 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 2 for instruction: 
%f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 13 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 8 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 6 for instruction: 
%v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
@@ -527,26 +527,26 @@ define void @casts() {
 ; 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 2 for instruction: 
%f32s1 = call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u1 = call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s8 = call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u8 = call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s1 = call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u1 = call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s8 = call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u8 = call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 13 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 8 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 6 for instruction: 
%v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
@@ -630,26 +630,26 @@ define void @casts() {
 ; 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 13 for instruction: %f32s1 
= call i1 @llvm.fptosi.sat.i1.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32u1 
= call i1 @llvm.fptoui.sat.i1.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %f32s8 
= call i8 @llvm.fptosi.sat.i8.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f32u8 
= call i8 @llvm.fptoui.sat.i8.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f32s16 = call i16 @llvm.fptosi.sat.i16.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f32u16 = call i16 @llvm.fptoui.sat.i16.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f32s32 = call i32 @llvm.fptosi.sat.i32.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f32u32 = call i32 @llvm.fptoui.sat.i32.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f32s64 = call i64 @llvm.fptosi.sat.i64.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: 
%f32u64 = call i64 @llvm.fptoui.sat.i64.f32(float undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %f64s1 
= call i1 @llvm.fptosi.sat.i1.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64u1 
= call i1 @llvm.fptoui.sat.i1.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %f64s8 
= call i8 @llvm.fptosi.sat.i8.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f64u8 
= call i8 @llvm.fptoui.sat.i8.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f64s16 = call i16 @llvm.fptosi.sat.i16.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f64u16 = call i16 @llvm.fptoui.sat.i16.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f64s32 = call i32 @llvm.fptosi.sat.i32.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: 
%f64u32 = call i32 @llvm.fptoui.sat.i32.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: 
%f64s64 = call i64 @llvm.fptosi.sat.i64.f64(double undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: 
%f64u64 = call i64 @llvm.fptoui.sat.i64.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32s1 
= call i1 @llvm.fptosi.sat.i1.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32u1 
= call i1 @llvm.fptoui.sat.i1.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32s8 
= call i8 @llvm.fptosi.sat.i8.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32u8 
= call i8 @llvm.fptoui.sat.i8.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32s16 
= call i16 @llvm.fptosi.sat.i16.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32u16 
= call i16 @llvm.fptoui.sat.i16.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32s32 
= call i32 @llvm.fptosi.sat.i32.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32u32 
= call i32 @llvm.fptoui.sat.i32.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32s64 
= call i64 @llvm.fptosi.sat.i64.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f32u64 
= call i64 @llvm.fptoui.sat.i64.f32(float undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s1 
= call i1 @llvm.fptosi.sat.i1.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u1 
= call i1 @llvm.fptoui.sat.i1.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s8 
= call i8 @llvm.fptosi.sat.i8.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u8 
= call i8 @llvm.fptoui.sat.i8.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s16 
= call i16 @llvm.fptosi.sat.i16.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u16 
= call i16 @llvm.fptoui.sat.i16.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s32 
= call i32 @llvm.fptosi.sat.i32.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u32 
= call i32 @llvm.fptoui.sat.i32.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64s64 
= call i64 @llvm.fptosi.sat.i64.f64(double undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64u64 
= call i64 @llvm.fptoui.sat.i64.f64(double undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 19 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 15 for instruction: 
%v2f32s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f32(<2 x float> undef)
@@ -847,16 +847,16 @@ define void @casts() {
 
 define void @fp16() {
 ; SSE2-LABEL: 'fp16'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s1 
= call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u1 
= call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s8 
= call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u8 
= call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16s1 
= call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u1 
= call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16s8 
= call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u8 
= call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 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 11 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 18 for instruction: 
%v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
@@ -900,16 +900,16 @@ define void @fp16() {
 ; 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 5 for instruction: 
%f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 15 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 11 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 17 for instruction: 
%v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
@@ -953,16 +953,16 @@ define void @fp16() {
 ; SSE42-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 5 for instruction: %f16s1 
= call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u1 
= call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s8 
= call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u8 
= call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16s1 
= call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u1 
= call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16s8 
= call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u8 
= call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 15 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 11 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 16 for instruction: 
%v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
@@ -1006,16 +1006,16 @@ define void @fp16() {
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret 
void
 ;
 ; AVX512F-LABEL: 'fp16'
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: 
%v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: 
%v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: 
%v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
@@ -1059,16 +1059,16 @@ define void @fp16() {
 ; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret 
void
 ;
 ; AVX512DQ-LABEL: 'fp16'
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: 
%f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: 
%v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: 
%v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
 ; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: 
%v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
@@ -1112,16 +1112,16 @@ define void @fp16() {
 ; AVX512DQ-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 5 for instruction: %f16s1 
= call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u1 
= call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s8 
= call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u8 
= call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s16 
= call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u16 
= call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s32 
= call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u32 
= call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s64 
= call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u64 
= call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16s1 
= call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u1 
= call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16s8 
= call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u8 
= call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16s16 
= call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u16 
= call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16s32 
= call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u32 
= call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16s64 
= call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f16u64 
= call i64 @llvm.fptoui.sat.i64.f16(half undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 17 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 11 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 19 for instruction: 
%v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
diff --git a/llvm/test/Transforms/AggressiveInstCombine/ARM/fptosisat.ll 
b/llvm/test/Transforms/AggressiveInstCombine/ARM/fptosisat.ll
index f26463815c16d..da3bac2562285 100644
--- a/llvm/test/Transforms/AggressiveInstCombine/ARM/fptosisat.ll
+++ b/llvm/test/Transforms/AggressiveInstCombine/ARM/fptosisat.ll
@@ -83,10 +83,9 @@ define i64 @f64_i32(double %in) {
 ; CHECK-BASE-NEXT:    ret i64 [[MAX]]
 ;
 ; CHECK-MVEFP-LABEL: @f64_i32(
-; CHECK-MVEFP-NEXT:    [[CONV:%.*]] = fptosi double [[IN:%.*]] to i64
-; CHECK-MVEFP-NEXT:    [[MIN:%.*]] = call i64 @llvm.smin.i64(i64 [[CONV]], i64 
2147483647)
-; CHECK-MVEFP-NEXT:    [[MAX:%.*]] = call i64 @llvm.smax.i64(i64 [[MIN]], i64 
-2147483648)
-; CHECK-MVEFP-NEXT:    ret i64 [[MAX]]
+; CHECK-MVEFP-NEXT:    [[TMP1:%.*]] = call i32 @llvm.fptosi.sat.i32.f64(double 
[[IN:%.*]])
+; CHECK-MVEFP-NEXT:    [[TMP2:%.*]] = sext i32 [[TMP1]] to i64
+; CHECK-MVEFP-NEXT:    ret i64 [[TMP2]]
 ;
 ; CHECK-FP64-LABEL: @f64_i32(
 ; CHECK-FP64-NEXT:    [[TMP1:%.*]] = call i32 @llvm.fptosi.sat.i32.f64(double 
[[IN:%.*]])
@@ -100,11 +99,22 @@ define i64 @f64_i32(double %in) {
 }
 
 define i64 @f64_i31(double %in) {
-; CHECK-LABEL: @f64_i31(
-; CHECK-NEXT:    [[CONV:%.*]] = fptosi double [[IN:%.*]] to i64
-; CHECK-NEXT:    [[MIN:%.*]] = call i64 @llvm.smin.i64(i64 [[CONV]], i64 
1073741823)
-; CHECK-NEXT:    [[MAX:%.*]] = call i64 @llvm.smax.i64(i64 [[MIN]], i64 
-1073741824)
-; CHECK-NEXT:    ret i64 [[MAX]]
+; CHECK-BASE-LABEL: @f64_i31(
+; CHECK-BASE-NEXT:    [[CONV:%.*]] = fptosi double [[IN:%.*]] to i64
+; CHECK-BASE-NEXT:    [[MIN:%.*]] = call i64 @llvm.smin.i64(i64 [[CONV]], i64 
1073741823)
+; CHECK-BASE-NEXT:    [[MAX:%.*]] = call i64 @llvm.smax.i64(i64 [[MIN]], i64 
-1073741824)
+; CHECK-BASE-NEXT:    ret i64 [[MAX]]
+;
+; CHECK-MVEFP-LABEL: @f64_i31(
+; CHECK-MVEFP-NEXT:    [[TMP1:%.*]] = call i31 @llvm.fptosi.sat.i31.f64(double 
[[IN:%.*]])
+; CHECK-MVEFP-NEXT:    [[TMP2:%.*]] = sext i31 [[TMP1]] to i64
+; CHECK-MVEFP-NEXT:    ret i64 [[TMP2]]
+;
+; CHECK-FP64-LABEL: @f64_i31(
+; CHECK-FP64-NEXT:    [[CONV:%.*]] = fptosi double [[IN:%.*]] to i64
+; CHECK-FP64-NEXT:    [[MIN:%.*]] = call i64 @llvm.smin.i64(i64 [[CONV]], i64 
1073741823)
+; CHECK-FP64-NEXT:    [[MAX:%.*]] = call i64 @llvm.smax.i64(i64 [[MIN]], i64 
-1073741824)
+; CHECK-FP64-NEXT:    ret i64 [[MAX]]
 ;
   %conv = fptosi double %in to i64
   %min = call i64 @llvm.smin.i64(i64 %conv, i64 1073741823)
@@ -113,11 +123,22 @@ define i64 @f64_i31(double %in) {
 }
 
 define i32 @f64_i16(double %in) {
-; CHECK-LABEL: @f64_i16(
-; CHECK-NEXT:    [[CONV:%.*]] = fptosi double [[IN:%.*]] to i32
-; CHECK-NEXT:    [[MIN:%.*]] = call i32 @llvm.smin.i32(i32 [[CONV]], i32 32767)
-; CHECK-NEXT:    [[MAX:%.*]] = call i32 @llvm.smax.i32(i32 [[MIN]], i32 -32768)
-; CHECK-NEXT:    ret i32 [[MAX]]
+; CHECK-BASE-LABEL: @f64_i16(
+; CHECK-BASE-NEXT:    [[CONV:%.*]] = fptosi double [[IN:%.*]] to i32
+; CHECK-BASE-NEXT:    [[MIN:%.*]] = call i32 @llvm.smin.i32(i32 [[CONV]], i32 
32767)
+; CHECK-BASE-NEXT:    [[MAX:%.*]] = call i32 @llvm.smax.i32(i32 [[MIN]], i32 
-32768)
+; CHECK-BASE-NEXT:    ret i32 [[MAX]]
+;
+; CHECK-MVEFP-LABEL: @f64_i16(
+; CHECK-MVEFP-NEXT:    [[TMP1:%.*]] = call i16 @llvm.fptosi.sat.i16.f64(double 
[[IN:%.*]])
+; CHECK-MVEFP-NEXT:    [[TMP2:%.*]] = sext i16 [[TMP1]] to i32
+; CHECK-MVEFP-NEXT:    ret i32 [[TMP2]]
+;
+; CHECK-FP64-LABEL: @f64_i16(
+; CHECK-FP64-NEXT:    [[CONV:%.*]] = fptosi double [[IN:%.*]] to i32
+; CHECK-FP64-NEXT:    [[MIN:%.*]] = call i32 @llvm.smin.i32(i32 [[CONV]], i32 
32767)
+; CHECK-FP64-NEXT:    [[MAX:%.*]] = call i32 @llvm.smax.i32(i32 [[MIN]], i32 
-32768)
+; CHECK-FP64-NEXT:    ret i32 [[MAX]]
 ;
   %conv = fptosi double %in to i32
   %min = call i32 @llvm.smin.i32(i32 %conv, i32 32767)

_______________________________________________
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits

Reply via email to