[X86][FP16][CostModel][WIP] Add CostModel for more FP16 instructions

llvm/lib/Target/X86/X86TargetTransformInfo.cpp

auto AdjustCost = [&CostKind](InstructionCost Cost) -> InstructionCost {
     return Cost;
   };
 
   // The cost tables include both specific, custom (non-legal) src/dst type
   // conversions and generic, legalized types. We test for customs first, before
   // falling back to legalization.
   // FIXME: Need a better design of the cost table to handle non-simple types of
   // potential massive combinations (elem_num x src_type x dst_type).
+  static const TypeConversionCostTblEntry AVX512FP16ConversionTbl[] {
+    { ISD::UINT_TO_FP,  MVT::f16,    MVT::i64,    1 },
+    { ISD::UINT_TO_FP,  MVT::v8f16,  MVT::v2i64,  1 },
+    { ISD::UINT_TO_FP,  MVT::v16f16, MVT::v4i64,  1 },
+    { ISD::UINT_TO_FP,  MVT::v32f16, MVT::v8i64,  1 },
+    { ISD::UINT_TO_FP,  MVT::v8f16,  MVT::v4i32,  1 },
+    { ISD::UINT_TO_FP,  MVT::v16f16, MVT::v8i32,  1 },
+    { ISD::UINT_TO_FP,  MVT::v32f16, MVT::v16i32, 1 },
+    { ISD::UINT_TO_FP,  MVT::v8f16,  MVT::v8i16,  1 },
+    { ISD::UINT_TO_FP,  MVT::v16f16, MVT::v16i16, 1 },
+    { ISD::UINT_TO_FP,  MVT::v32f16, MVT::v32i16, 1 },
+    { ISD::UINT_TO_FP,  MVT::v8f16,  MVT::v16i8,  2 },
+    { ISD::UINT_TO_FP,  MVT::v16f16, MVT::v16i8,  1 },
+    { ISD::UINT_TO_FP,  MVT::v32f16, MVT::v32i8,  1 },
+    { ISD::FP_TO_UINT,  MVT::i64,    MVT::f16,    1 },
+    { ISD::FP_TO_UINT,  MVT::v2i64,  MVT::v8f16,  1 },
+    { ISD::FP_TO_UINT,  MVT::v4i64,  MVT::v16f16, 1 },
+    { ISD::FP_TO_UINT,  MVT::v8i64,  MVT::v32f16, 1 },
+    { ISD::FP_TO_UINT,  MVT::v4i32,  MVT::v8f16,  1 },
+    { ISD::FP_TO_UINT,  MVT::v8i32,  MVT::v16f16, 1 },
+    { ISD::FP_TO_UINT,  MVT::v16i32, MVT::v32f16, 1 },
+    { ISD::FP_TO_UINT,  MVT::v8i16,  MVT::v8f16,  1 },
+    { ISD::FP_TO_UINT,  MVT::v16i16, MVT::v16f16, 1 },
+    { ISD::FP_TO_UINT,  MVT::v32i16, MVT::v32f16, 1 },
+    { ISD::FP_TO_UINT,  MVT::v16i8,  MVT::v8f16,  2 },
+    { ISD::FP_TO_UINT,  MVT::v16i8,  MVT::v16f16, 1 },
+    { ISD::FP_TO_UINT,  MVT::v32i8,  MVT::v32f16, 1 },
+    { ISD::SINT_TO_FP,  MVT::f16,    MVT::i64,    1 },
+    { ISD::SINT_TO_FP,  MVT::v8f16,  MVT::v2i64,  1 },
+    { ISD::SINT_TO_FP,  MVT::v16f16, MVT::v4i64,  1 },
+    { ISD::SINT_TO_FP,  MVT::v32f16, MVT::v8i64,  1 },
+    { ISD::SINT_TO_FP,  MVT::v8f16,  MVT::v4i32,  1 },
+    { ISD::SINT_TO_FP,  MVT::v16f16, MVT::v8i32,  1 },
+    { ISD::SINT_TO_FP,  MVT::v32f16, MVT::v16i32, 1 },
+    { ISD::SINT_TO_FP,  MVT::v8f16,  MVT::v8i16,  1 },
+    { ISD::SINT_TO_FP,  MVT::v16f16, MVT::v16i16, 1 },
+    { ISD::SINT_TO_FP,  MVT::v32f16, MVT::v32i16, 1 },
+    { ISD::SINT_TO_FP,  MVT::v8f16,  MVT::v16i8,  2 },
+    { ISD::SINT_TO_FP,  MVT::v16f16, MVT::v16i8,  1 },
+    { ISD::SINT_TO_FP,  MVT::v32f16, MVT::v32i8,  1 },
+    { ISD::FP_TO_SINT,  MVT::i64,    MVT::f16,    1 },
+    { ISD::FP_TO_SINT,  MVT::v2i64,  MVT::v8f16,  1 },
+    { ISD::FP_TO_SINT,  MVT::v4i64,  MVT::v16f16, 1 },
+    { ISD::FP_TO_SINT,  MVT::v8i64,  MVT::v32f16, 1 },
+    { ISD::FP_TO_SINT,  MVT::v4i32,  MVT::v8f16,  1 },
+    { ISD::FP_TO_SINT,  MVT::v8i32,  MVT::v16f16, 1 },
+    { ISD::FP_TO_SINT,  MVT::v16i32, MVT::v32f16, 1 },
+    { ISD::FP_TO_SINT,  MVT::v8i16,  MVT::v8f16,  1 },
+    { ISD::FP_TO_SINT,  MVT::v16i16, MVT::v16f16, 1 },
+    { ISD::FP_TO_SINT,  MVT::v32i16, MVT::v32f16, 1 },
+    { ISD::FP_TO_SINT,  MVT::v16i8,  MVT::v8f16,  2 },
+    { ISD::FP_TO_SINT,  MVT::v16i8,  MVT::v16f16, 1 },
+    { ISD::FP_TO_SINT,  MVT::v32i8,  MVT::v32f16, 1 },
+  };
+
   static const TypeConversionCostTblEntry AVX512BWConversionTbl[] {
     { ISD::SIGN_EXTEND, MVT::v32i16, MVT::v32i8, 1 },
     { ISD::ZERO_EXTEND, MVT::v32i16, MVT::v32i8, 1 },
 
     // Mask sign extend has an instruction.
     { ISD::SIGN_EXTEND, MVT::v2i8,   MVT::v2i1,   1 },
     { ISD::SIGN_EXTEND, MVT::v16i8,  MVT::v2i1,   1 },
     { ISD::SIGN_EXTEND, MVT::v2i16,  MVT::v2i1,   1 },
static const TypeConversionCostTblEntry AVX512VLConversionTbl[] = {
     { ISD::ZERO_EXTEND, MVT::v4i64,  MVT::v4i32,  1 },
 
     { ISD::SINT_TO_FP,  MVT::v2f64,  MVT::v16i8,  1 },
     { ISD::SINT_TO_FP,  MVT::v8f32,  MVT::v16i8,  1 },
     { ISD::SINT_TO_FP,  MVT::v2f64,  MVT::v8i16,  1 },
     { ISD::SINT_TO_FP,  MVT::v8f32,  MVT::v8i16,  1 },
 
     { ISD::UINT_TO_FP,  MVT::f32,    MVT::i64,    1 },
+    { ISD::UINT_TO_FP,  MVT::f16,    MVT::i64,    2 },
     { ISD::UINT_TO_FP,  MVT::f64,    MVT::i64,    1 },
     { ISD::UINT_TO_FP,  MVT::v2f64,  MVT::v16i8,  1 },
     { ISD::UINT_TO_FP,  MVT::v8f32,  MVT::v16i8,  1 },
     { ISD::UINT_TO_FP,  MVT::v2f64,  MVT::v8i16,  1 },
     { ISD::UINT_TO_FP,  MVT::v8f32,  MVT::v8i16,  1 },
     { ISD::UINT_TO_FP,  MVT::v2f32,  MVT::v2i32,  1 },
     { ISD::UINT_TO_FP,  MVT::v4f32,  MVT::v4i32,  1 },
     { ISD::UINT_TO_FP,  MVT::v4f64,  MVT::v4i32,  1 },
     { ISD::UINT_TO_FP,  MVT::v8f32,  MVT::v8i32,  1 },
     { ISD::UINT_TO_FP,  MVT::v2f32,  MVT::v2i64,  5 },
     { ISD::UINT_TO_FP,  MVT::v2f64,  MVT::v2i64,  5 },
     { ISD::UINT_TO_FP,  MVT::v4f64,  MVT::v4i64,  5 },
 
     { ISD::FP_TO_SINT,  MVT::v16i8,  MVT::v8f32,  2 },
     { ISD::FP_TO_SINT,  MVT::v16i8,  MVT::v16f32, 2 },
     { ISD::FP_TO_SINT,  MVT::v32i8,  MVT::v32f32, 5 },
 
     { ISD::FP_TO_UINT,  MVT::i64,    MVT::f32,    1 },
+    { ISD::FP_TO_UINT,  MVT::i64,    MVT::f16,    2 },
     { ISD::FP_TO_UINT,  MVT::i64,    MVT::f64,    1 },
     { ISD::FP_TO_UINT,  MVT::v4i32,  MVT::v4f32,  1 },
     { ISD::FP_TO_UINT,  MVT::v4i32,  MVT::v2f64,  1 },
     { ISD::FP_TO_UINT,  MVT::v4i32,  MVT::v4f64,  1 },
     { ISD::FP_TO_UINT,  MVT::v8i32,  MVT::v8f32,  1 },
     { ISD::FP_TO_UINT,  MVT::v8i32,  MVT::v8f64,  1 },
   };
 
static const TypeConversionCostTblEntry AVX2ConversionTbl[] = {
     { ISD::FP_ROUND,    MVT::v8f32,  MVT::v8f64,  3 },
 
     { ISD::FP_TO_SINT,  MVT::v16i16, MVT::v8f32,  1 },
     { ISD::FP_TO_SINT,  MVT::v4i32,  MVT::v4f64,  1 },
     { ISD::FP_TO_SINT,  MVT::v8i32,  MVT::v8f32,  1 },
     { ISD::FP_TO_SINT,  MVT::v8i32,  MVT::v8f64,  3 },
 
     { ISD::FP_TO_UINT,  MVT::i64,    MVT::f32,    3 },
+    { ISD::FP_TO_UINT,  MVT::i64,    MVT::f16,    4 },
     { ISD::FP_TO_UINT,  MVT::i64,    MVT::f64,    3 },
     { ISD::FP_TO_UINT,  MVT::v16i16, MVT::v8f32,  1 },
     { ISD::FP_TO_UINT,  MVT::v4i32,  MVT::v4f32,  3 },
     { ISD::FP_TO_UINT,  MVT::v4i32,  MVT::v2f64,  4 },
     { ISD::FP_TO_UINT,  MVT::v4i32,  MVT::v4f64,  4 },
     { ISD::FP_TO_UINT,  MVT::v8i32,  MVT::v8f32,  3 },
     { ISD::FP_TO_UINT,  MVT::v8i32,  MVT::v4f64,  4 },
 
static const TypeConversionCostTblEntry SSE41ConversionTbl[] = {
     { ISD::TRUNCATE,    MVT::v2i1,   MVT::v2i16,  1 }, // PMOVXZWQ
     { ISD::TRUNCATE,    MVT::v4i1,   MVT::v4i8,   1 }, // PMOVXZBD
 
     { ISD::TRUNCATE,    MVT::v16i8,  MVT::v4i32,  2 },
     { ISD::TRUNCATE,    MVT::v8i16,  MVT::v4i32,  2 },
     { ISD::TRUNCATE,    MVT::v16i8,  MVT::v2i64,  2 },
 
     { ISD::SINT_TO_FP,  MVT::f32,    MVT::i32,    1 },
+    { ISD::SINT_TO_FP,  MVT::f16,    MVT::i32,    2 },
     { ISD::SINT_TO_FP,  MVT::f64,    MVT::i32,    1 },
     { ISD::SINT_TO_FP,  MVT::f32,    MVT::i64,    1 },
+    { ISD::SINT_TO_FP,  MVT::f16,    MVT::i64,    2 },
     { ISD::SINT_TO_FP,  MVT::f64,    MVT::i64,    1 },
     { ISD::SINT_TO_FP,  MVT::v4f32,  MVT::v16i8,  1 },
     { ISD::SINT_TO_FP,  MVT::v2f64,  MVT::v16i8,  1 },
     { ISD::SINT_TO_FP,  MVT::v4f32,  MVT::v8i16,  1 },
     { ISD::SINT_TO_FP,  MVT::v2f64,  MVT::v8i16,  1 },
     { ISD::SINT_TO_FP,  MVT::v4f32,  MVT::v4i32,  1 },
     { ISD::SINT_TO_FP,  MVT::v2f64,  MVT::v4i32,  1 },
     { ISD::SINT_TO_FP,  MVT::v4f64,  MVT::v4i32,  2 },
 
     { ISD::UINT_TO_FP,  MVT::f32,    MVT::i32,    1 },
+    { ISD::UINT_TO_FP,  MVT::f16,    MVT::i32,    2 },
     { ISD::UINT_TO_FP,  MVT::f64,    MVT::i32,    1 },
     { ISD::UINT_TO_FP,  MVT::f32,    MVT::i64,    4 },
+    { ISD::UINT_TO_FP,  MVT::f16,    MVT::i64,    5 },
     { ISD::UINT_TO_FP,  MVT::f64,    MVT::i64,    4 },
     { ISD::UINT_TO_FP,  MVT::v4f32,  MVT::v16i8,  1 },
     { ISD::UINT_TO_FP,  MVT::v2f64,  MVT::v16i8,  1 },
     { ISD::UINT_TO_FP,  MVT::v4f32,  MVT::v8i16,  1 },
     { ISD::UINT_TO_FP,  MVT::v2f64,  MVT::v8i16,  1 },
     { ISD::UINT_TO_FP,  MVT::v2f32,  MVT::v2i32,  3 },
     { ISD::UINT_TO_FP,  MVT::v4f32,  MVT::v4i32,  3 },
     { ISD::UINT_TO_FP,  MVT::v2f64,  MVT::v4i32,  2 },
     { ISD::UINT_TO_FP,  MVT::v4f32,  MVT::v2i64, 12 },
     { ISD::UINT_TO_FP,  MVT::v4f32,  MVT::v4i64, 22 },
     { ISD::UINT_TO_FP,  MVT::v2f64,  MVT::v2i64,  4 },
 
     { ISD::FP_TO_SINT,  MVT::i32,    MVT::f32,    1 },
     { ISD::FP_TO_SINT,  MVT::i64,    MVT::f32,    1 },
+    { ISD::FP_TO_SINT,  MVT::i32,    MVT::f16,    2 },
+    { ISD::FP_TO_SINT,  MVT::i64,    MVT::f16,    2 },
     { ISD::FP_TO_SINT,  MVT::i32,    MVT::f64,    1 },
     { ISD::FP_TO_SINT,  MVT::i64,    MVT::f64,    1 },
     { ISD::FP_TO_SINT,  MVT::v16i8,  MVT::v4f32,  2 },
     { ISD::FP_TO_SINT,  MVT::v16i8,  MVT::v2f64,  2 },
     { ISD::FP_TO_SINT,  MVT::v8i16,  MVT::v4f32,  1 },
     { ISD::FP_TO_SINT,  MVT::v8i16,  MVT::v2f64,  1 },
     { ISD::FP_TO_SINT,  MVT::v4i32,  MVT::v4f32,  1 },
     { ISD::FP_TO_SINT,  MVT::v4i32,  MVT::v2f64,  1 },
 
     { ISD::FP_TO_UINT,  MVT::i32,    MVT::f32,    1 },
     { ISD::FP_TO_UINT,  MVT::i64,    MVT::f32,    4 },
+    { ISD::FP_TO_UINT,  MVT::i32,    MVT::f16,    2 },
+    { ISD::FP_TO_UINT,  MVT::i64,    MVT::f16,    5 },
     { ISD::FP_TO_UINT,  MVT::i32,    MVT::f64,    1 },
     { ISD::FP_TO_UINT,  MVT::i64,    MVT::f64,    4 },
     { ISD::FP_TO_UINT,  MVT::v16i8,  MVT::v4f32,  2 },
     { ISD::FP_TO_UINT,  MVT::v16i8,  MVT::v2f64,  2 },
     { ISD::FP_TO_UINT,  MVT::v8i16,  MVT::v4f32,  1 },
     { ISD::FP_TO_UINT,  MVT::v8i16,  MVT::v2f64,  1 },
     { ISD::FP_TO_UINT,  MVT::v4i32,  MVT::v4f32,  4 },
     { ISD::FP_TO_UINT,  MVT::v4i32,  MVT::v2f64,  4 },
   };
 
   static const TypeConversionCostTblEntry SSE2ConversionTbl[] = {
     // These are somewhat magic numbers justified by comparing the
     // output of llvm-mca for our various supported scheduler models
     // and basing it off the worst case scenario.
     { ISD::SINT_TO_FP,  MVT::f32,    MVT::i32,    3 },
+    { ISD::SINT_TO_FP,  MVT::f16,    MVT::i32,    4 },
     { ISD::SINT_TO_FP,  MVT::f64,    MVT::i32,    3 },
     { ISD::SINT_TO_FP,  MVT::f32,    MVT::i64,    3 },
+    { ISD::SINT_TO_FP,  MVT::f16,    MVT::i64,    4 },
     { ISD::SINT_TO_FP,  MVT::f64,    MVT::i64,    3 },
     { ISD::SINT_TO_FP,  MVT::v4f32,  MVT::v16i8,  3 },
     { ISD::SINT_TO_FP,  MVT::v2f64,  MVT::v16i8,  4 },
     { ISD::SINT_TO_FP,  MVT::v4f32,  MVT::v8i16,  3 },
     { ISD::SINT_TO_FP,  MVT::v2f64,  MVT::v8i16,  4 },
     { ISD::SINT_TO_FP,  MVT::v4f32,  MVT::v4i32,  3 },
     { ISD::SINT_TO_FP,  MVT::v2f64,  MVT::v4i32,  4 },
     { ISD::SINT_TO_FP,  MVT::v4f32,  MVT::v2i64,  8 },
     { ISD::SINT_TO_FP,  MVT::v2f64,  MVT::v2i64,  8 },
 
     { ISD::UINT_TO_FP,  MVT::f32,    MVT::i32,    3 },
+    { ISD::UINT_TO_FP,  MVT::f16,    MVT::i32,    4 },
     { ISD::UINT_TO_FP,  MVT::f64,    MVT::i32,    3 },
     { ISD::UINT_TO_FP,  MVT::f32,    MVT::i64,    8 },
+    { ISD::UINT_TO_FP,  MVT::f16,    MVT::i64,    9 },
     { ISD::UINT_TO_FP,  MVT::f64,    MVT::i64,    9 },
     { ISD::UINT_TO_FP,  MVT::v2f64,  MVT::v16i8,  4 },
     { ISD::UINT_TO_FP,  MVT::v4f32,  MVT::v16i8,  4 },
     { ISD::UINT_TO_FP,  MVT::v4f32,  MVT::v8i16,  4 },
     { ISD::UINT_TO_FP,  MVT::v2f64,  MVT::v8i16,  4 },
     { ISD::UINT_TO_FP,  MVT::v2f32,  MVT::v2i32,  7 },
     { ISD::UINT_TO_FP,  MVT::v2f64,  MVT::v4i32,  7 },
     { ISD::UINT_TO_FP,  MVT::v4f32,  MVT::v4i32,  5 },
     { ISD::UINT_TO_FP,  MVT::v2f64,  MVT::v2i64, 15 },
     { ISD::UINT_TO_FP,  MVT::v4f32,  MVT::v2i64, 18 },
 
     { ISD::FP_TO_SINT,  MVT::i32,    MVT::f32,    4 },
     { ISD::FP_TO_SINT,  MVT::i64,    MVT::f32,    4 },
+    { ISD::FP_TO_SINT,  MVT::i32,    MVT::f16,    5 },
+    { ISD::FP_TO_SINT,  MVT::i64,    MVT::f16,    5 },
     { ISD::FP_TO_SINT,  MVT::i32,    MVT::f64,    4 },
     { ISD::FP_TO_SINT,  MVT::i64,    MVT::f64,    4 },
     { ISD::FP_TO_SINT,  MVT::v16i8,  MVT::v4f32,  6 },
     { ISD::FP_TO_SINT,  MVT::v16i8,  MVT::v2f64,  6 },
     { ISD::FP_TO_SINT,  MVT::v8i16,  MVT::v4f32,  5 },
     { ISD::FP_TO_SINT,  MVT::v8i16,  MVT::v2f64,  5 },
     { ISD::FP_TO_SINT,  MVT::v4i32,  MVT::v4f32,  4 },
     { ISD::FP_TO_SINT,  MVT::v4i32,  MVT::v2f64,  4 },
 
     { ISD::FP_TO_UINT,  MVT::i32,    MVT::f32,    4 },
     { ISD::FP_TO_UINT,  MVT::i64,    MVT::f32,    4 },
+    { ISD::FP_TO_UINT,  MVT::i32,    MVT::f16,    5 },
+    { ISD::FP_TO_UINT,  MVT::i64,    MVT::f16,    5 },
     { ISD::FP_TO_UINT,  MVT::i32,    MVT::f64,    4 },
     { ISD::FP_TO_UINT,  MVT::i64,    MVT::f64,   15 },
     { ISD::FP_TO_UINT,  MVT::v16i8,  MVT::v4f32,  6 },
     { ISD::FP_TO_UINT,  MVT::v16i8,  MVT::v2f64,  6 },
     { ISD::FP_TO_UINT,  MVT::v8i16,  MVT::v4f32,  5 },
     { ISD::FP_TO_UINT,  MVT::v8i16,  MVT::v2f64,  5 },
     { ISD::FP_TO_UINT,  MVT::v4i32,  MVT::v4f32,  8 },
     { ISD::FP_TO_UINT,  MVT::v4i32,  MVT::v2f64,  8 },
InstructionCost X86TTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
   EVT DstTy = TLI->getValueType(DL, Dst);
 
   // The function getSimpleVT only handles simple value types.
   if (SrcTy.isSimple() && DstTy.isSimple()) {
     MVT SimpleSrcTy = SrcTy.getSimpleVT();
     MVT SimpleDstTy = DstTy.getSimpleVT();
 
     if (ST->useAVX512Regs()) {
+      if (ST->hasFP16())
+        if (const auto *Entry = ConvertCostTableLookup(
+                AVX512FP16ConversionTbl, ISD, SimpleDstTy, SimpleSrcTy))
+          return AdjustCost(Entry->Cost);
+
       if (ST->hasBWI())
         if (const auto *Entry = ConvertCostTableLookup(
                 AVX512BWConversionTbl, ISD, SimpleDstTy, SimpleSrcTy))
           return AdjustCost(Entry->Cost);
 
       if (ST->hasDQI())
         if (const auto *Entry = ConvertCostTableLookup(
                 AVX512DQConversionTbl, ISD, SimpleDstTy, SimpleSrcTy))
X86TTIImpl::getTypeBasedIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
   // BITREVERSE: llvm\test\CodeGen\X86\vector-bitreverse.ll
   // BSWAP: llvm\test\CodeGen\X86\bswap-vector.ll
   // CTLZ: llvm\test\CodeGen\X86\vector-lzcnt-*.ll
   // CTPOP: llvm\test\CodeGen\X86\vector-popcnt-*.ll
   // CTTZ: llvm\test\CodeGen\X86\vector-tzcnt-*.ll
 
   // TODO: Overflow intrinsics (*ADDO, *SUBO, *MULO) with vector types are not
   //       specialized in these tables yet.
+  static const CostTblEntry AVX512FP16CostTbl[] = {
+    { ISD::FMAXNUM,    MVT::f16,     2 },
+    { ISD::FMAXNUM,    MVT::v8f16,   2 },
+    { ISD::FMAXNUM,    MVT::v16f16,  2 },
+    { ISD::FMAXNUM,    MVT::v32f16,  2 },
+  };
   static const CostTblEntry AVX512BITALGCostTbl[] = {
     { ISD::CTPOP,      MVT::v32i16,  1 },
     { ISD::CTPOP,      MVT::v64i8,   1 },
     { ISD::CTPOP,      MVT::v16i16,  1 },
     { ISD::CTPOP,      MVT::v32i8,   1 },
     { ISD::CTPOP,      MVT::v8i16,   1 },
     { ISD::CTPOP,      MVT::v16i8,   1 },
   };
static const CostTblEntry AVX1CostTbl[] = {
     { ISD::UMAX,       MVT::v32i8,   4 }, // 2 x 128-bit Op + extract/insert
     { ISD::UMIN,       MVT::v8i32,   4 }, // 2 x 128-bit Op + extract/insert
     { ISD::UMIN,       MVT::v16i16,  4 }, // 2 x 128-bit Op + extract/insert
     { ISD::UMIN,       MVT::v32i8,   4 }, // 2 x 128-bit Op + extract/insert
     { ISD::USUBSAT,    MVT::v16i16,  4 }, // 2 x 128-bit Op + extract/insert
     { ISD::USUBSAT,    MVT::v32i8,   4 }, // 2 x 128-bit Op + extract/insert
     { ISD::USUBSAT,    MVT::v8i32,   6 }, // 2 x 128-bit Op + extract/insert
     { ISD::FMAXNUM,    MVT::f32,     3 }, // MAXSS + CMPUNORDSS + BLENDVPS
+    { ISD::FMAXNUM,    MVT::f16,     4 }, // FPEXT + MAXSS + CMPUNORDSS + BLENDVPS
     { ISD::FMAXNUM,    MVT::v4f32,   3 }, // MAXPS + CMPUNORDPS + BLENDVPS
     { ISD::FMAXNUM,    MVT::v8f32,   5 }, // MAXPS + CMPUNORDPS + BLENDVPS + ?
     { ISD::FMAXNUM,    MVT::f64,     3 }, // MAXSD + CMPUNORDSD + BLENDVPD
     { ISD::FMAXNUM,    MVT::v2f64,   3 }, // MAXPD + CMPUNORDPD + BLENDVPD
     { ISD::FMAXNUM,    MVT::v4f64,   5 }, // MAXPD + CMPUNORDPD + BLENDVPD + ?
     { ISD::FSQRT,      MVT::f32,    14 }, // SNB from http://www.agner.org/
     { ISD::FSQRT,      MVT::v4f32,  14 }, // SNB from http://www.agner.org/
     { ISD::FSQRT,      MVT::v8f32,  28 }, // SNB from http://www.agner.org/
static const CostTblEntry SSE2CostTbl[] = {
     { ISD::UADDSAT,    MVT::v16i8,   1 },
     { ISD::UMAX,       MVT::v8i16,   2 },
     { ISD::UMAX,       MVT::v16i8,   1 },
     { ISD::UMIN,       MVT::v8i16,   2 },
     { ISD::UMIN,       MVT::v16i8,   1 },
     { ISD::USUBSAT,    MVT::v8i16,   1 },
     { ISD::USUBSAT,    MVT::v16i8,   1 },
     { ISD::FMAXNUM,    MVT::f64,     4 },
+    { ISD::FMAXNUM,    MVT::f16,     5 }, // fpext+FMAXNUM
     { ISD::FMAXNUM,    MVT::v2f64,   4 },
     { ISD::FSQRT,      MVT::f64,    32 }, // Nehalem from http://www.agner.org/
     { ISD::FSQRT,      MVT::v2f64,  32 }, // Nehalem from http://www.agner.org/
   };
   static const CostTblEntry SSE1CostTbl[] = {
     { ISD::FMAXNUM,    MVT::f32,     4 },
     { ISD::FMAXNUM,    MVT::v4f32,   4 },
     { ISD::FSQRT,      MVT::f32,    28 }, // Pentium III from http://www.agner.org/
InstructionCost X86TTIImpl::getMinMaxCost(Type *Ty, Type *CondTy,
   }
 
   static const CostTblEntry SSE1CostTbl[] = {
     {ISD::FMINNUM, MVT::v4f32, 1},
   };
 
   static const CostTblEntry SSE2CostTbl[] = {
     {ISD::FMINNUM, MVT::v2f64, 1},
+    {ISD::FMINNUM, MVT::v8f16, 2}, // fpext+FMINNUM
     {ISD::SMIN,    MVT::v8i16, 1},
     {ISD::UMIN,    MVT::v16i8, 1},
   };
 
   static const CostTblEntry SSE41CostTbl[] = {
     {ISD::SMIN,    MVT::v4i32, 1},
     {ISD::UMIN,    MVT::v4i32, 1},
     {ISD::UMIN,    MVT::v8i16, 1},
     {ISD::SMIN,    MVT::v16i8, 1},
   };
 
   static const CostTblEntry SSE42CostTbl[] = {
     {ISD::UMIN,    MVT::v2i64, 3}, // xor+pcmpgtq+blendvpd
   };
 
   static const CostTblEntry AVX1CostTbl[] = {
     {ISD::FMINNUM, MVT::v8f32,  1},
     {ISD::FMINNUM, MVT::v4f64,  1},
+    {ISD::FMINNUM, MVT::v16f16, 2}, // fpext+FMINNUM
     {ISD::SMIN,    MVT::v8i32,  3},
     {ISD::UMIN,    MVT::v8i32,  3},
     {ISD::SMIN,    MVT::v16i16, 3},
     {ISD::UMIN,    MVT::v16i16, 3},
     {ISD::SMIN,    MVT::v32i8,  3},
     {ISD::UMIN,    MVT::v32i8,  3},
   };
 
   static const CostTblEntry AVX2CostTbl[] = {
     {ISD::SMIN,    MVT::v8i32,  1},
     {ISD::UMIN,    MVT::v8i32,  1},
     {ISD::SMIN,    MVT::v16i16, 1},
     {ISD::UMIN,    MVT::v16i16, 1},
     {ISD::SMIN,    MVT::v32i8,  1},
     {ISD::UMIN,    MVT::v32i8,  1},
   };
 
   static const CostTblEntry AVX512CostTbl[] = {
     {ISD::FMINNUM, MVT::v16f32, 1},
     {ISD::FMINNUM, MVT::v8f64,  1},
+    {ISD::FMINNUM, MVT::v32f16, 2}, // fpext+FMINNUM
     {ISD::SMIN,    MVT::v2i64,  1},
     {ISD::UMIN,    MVT::v2i64,  1},
     {ISD::SMIN,    MVT::v4i64,  1},
     {ISD::UMIN,    MVT::v4i64,  1},
     {ISD::SMIN,    MVT::v8i64,  1},
     {ISD::UMIN,    MVT::v8i64,  1},
     {ISD::SMIN,    MVT::v16i32, 1},
     {ISD::UMIN,    MVT::v16i32, 1},
   };
 
   static const CostTblEntry AVX512BWCostTbl[] = {
     {ISD::SMIN,    MVT::v32i16, 1},
     {ISD::UMIN,    MVT::v32i16, 1},
     {ISD::SMIN,    MVT::v64i8,  1},
     {ISD::UMIN,    MVT::v64i8,  1},
   };
 
+  static const CostTblEntry AVX512FP16CostTbl[] = {
+    {ISD::FMINNUM, MVT::v32f16, 1},
+  };
+
   // If we have a native MIN/MAX instruction for this type, use it.
+  if (ST->hasFP16())
+    if (const auto *Entry = CostTableLookup(AVX512FP16CostTbl, ISD, MTy))
+      return LT.first * Entry->Cost;
+
   if (ST->hasBWI())
     if (const auto *Entry = CostTableLookup(AVX512BWCostTbl, ISD, MTy))
       return LT.first * Entry->Cost;
 
   if (ST->hasAVX512())
     if (const auto *Entry = CostTableLookup(AVX512CostTbl, ISD, MTy))
       return LT.first * Entry->Cost;
 

llvm/test/Analysis/CostModel/X86/fptoi_sat.ll

;
   %v16f64s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f64(<16 x double> undef)
   %v16f64u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f64(<16 x double> undef)
 
   ret void
 }
 
 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 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)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 113 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 27 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 23 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 38 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 34 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 35 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 66 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 69 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 70 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 105 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 142 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 132 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 141 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 131 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 138 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 126 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 140 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 124 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 209 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 191 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 277 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 259 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 282 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 262 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 276 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 252 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 280 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 248 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SSE42-LABEL: 'fp16'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 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 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)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 113 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 27 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 23 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 31 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 105 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 106 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 107 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 124 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 209 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 191 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 229 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 211 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 234 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 214 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 228 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 232 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 248 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX1-LABEL: 'fp16'
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; AVX1-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)
-; AVX1-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)
-; AVX1-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)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 113 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 118 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; AVX1-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)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 55 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 79 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 110 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 177 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 159 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 198 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 199 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 180 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 196 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 174 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 200 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 220 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX2-LABEL: 'fp16'
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 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 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)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 113 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; AVX2-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)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 79 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 101 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 91 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 97 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 177 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 159 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 196 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 196 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 178 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 174 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 196 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; AVX512-LABEL: 'fp16'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 63 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 127 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 97 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 97 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 94 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v2f16s1 = call <2 x i1> @llvm.fptosi.sat.v2i1.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v2f16u1 = call <2 x i1> @llvm.fptoui.sat.v2i1.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 45 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 43 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 119 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 89 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 87 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 239 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 191 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 177 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 177 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 175 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 174 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; SLM-LABEL: 'fp16'
-; SLM-NEXT:  Cost Model: Found an estimated cost of 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 15 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 17 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 61 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 46 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 60 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 113 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 99 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 122 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 92 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 120 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
-; SLM-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 13 for instruction: %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s32 = call i32 @llvm.fptosi.sat.i32.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %f16u32 = call i32 @llvm.fptoui.sat.i32.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %f16s64 = call i64 @llvm.fptosi.sat.i64.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %f16u64 = call i64 @llvm.fptoui.sat.i64.f16(half undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 27 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 23 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 31 for instruction: %v2f16s8 = call <2 x i8> @llvm.fptosi.sat.v2i8.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 27 for instruction: %v2f16u8 = call <2 x i8> @llvm.fptoui.sat.v2i8.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %v2f16s16 = call <2 x i16> @llvm.fptosi.sat.v2i16.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %v2f16u16 = call <2 x i16> @llvm.fptoui.sat.v2i16.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v2f16s32 = call <2 x i32> @llvm.fptosi.sat.v2i32.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %v2f16u32 = call <2 x i32> @llvm.fptoui.sat.v2i32.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %v2f16s64 = call <2 x i64> @llvm.fptosi.sat.v2i64.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 31 for instruction: %v2f16u64 = call <2 x i64> @llvm.fptoui.sat.v2i64.v2f16(<2 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v4f16s1 = call <4 x i1> @llvm.fptosi.sat.v4i1.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 47 for instruction: %v4f16u1 = call <4 x i1> @llvm.fptoui.sat.v4i1.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %v4f16s8 = call <4 x i8> @llvm.fptosi.sat.v4i8.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v4f16u8 = call <4 x i8> @llvm.fptoui.sat.v4i8.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 59 for instruction: %v4f16s16 = call <4 x i16> @llvm.fptosi.sat.v4i16.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 53 for instruction: %v4f16u16 = call <4 x i16> @llvm.fptoui.sat.v4i16.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 57 for instruction: %v4f16s32 = call <4 x i32> @llvm.fptosi.sat.v4i32.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 51 for instruction: %v4f16u32 = call <4 x i32> @llvm.fptoui.sat.v4i32.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %v4f16s64 = call <4 x i64> @llvm.fptosi.sat.v4i64.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 62 for instruction: %v4f16u64 = call <4 x i64> @llvm.fptoui.sat.v4i64.v4f16(<4 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 105 for instruction: %v8f16s1 = call <8 x i1> @llvm.fptosi.sat.v8i1.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 95 for instruction: %v8f16u1 = call <8 x i1> @llvm.fptoui.sat.v8i1.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v8f16s8 = call <8 x i8> @llvm.fptosi.sat.v8i8.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 106 for instruction: %v8f16u8 = call <8 x i8> @llvm.fptoui.sat.v8i8.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 117 for instruction: %v8f16s16 = call <8 x i16> @llvm.fptosi.sat.v8i16.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 107 for instruction: %v8f16u16 = call <8 x i16> @llvm.fptoui.sat.v8i16.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 114 for instruction: %v8f16s32 = call <8 x i32> @llvm.fptosi.sat.v8i32.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %v8f16u32 = call <8 x i32> @llvm.fptoui.sat.v8i32.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %v8f16s64 = call <8 x i64> @llvm.fptosi.sat.v8i64.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 124 for instruction: %v8f16u64 = call <8 x i64> @llvm.fptoui.sat.v8i64.v8f16(<8 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 209 for instruction: %v16f16s1 = call <16 x i1> @llvm.fptosi.sat.v16i1.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 191 for instruction: %v16f16u1 = call <16 x i1> @llvm.fptoui.sat.v16i1.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 229 for instruction: %v16f16s8 = call <16 x i8> @llvm.fptosi.sat.v16i8.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 211 for instruction: %v16f16u8 = call <16 x i8> @llvm.fptoui.sat.v16i8.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 234 for instruction: %v16f16s16 = call <16 x i16> @llvm.fptosi.sat.v16i16.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 214 for instruction: %v16f16u16 = call <16 x i16> @llvm.fptoui.sat.v16i16.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 228 for instruction: %v16f16s32 = call <16 x i32> @llvm.fptosi.sat.v16i32.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 204 for instruction: %v16f16u32 = call <16 x i32> @llvm.fptoui.sat.v16i32.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 232 for instruction: %v16f16s64 = call <16 x i64> @llvm.fptosi.sat.v16i64.v16f16(<16 x half> undef)
+; SLM-NEXT:  Cost Model: Found an estimated cost of 248 for instruction: %v16f16u64 = call <16 x i64> @llvm.fptoui.sat.v16i64.v16f16(<16 x half> undef)
 ; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %f16s1 = call i1 @llvm.fptosi.sat.i1.f16(half undef)
   %f16u1 = call i1 @llvm.fptoui.sat.i1.f16(half undef)
   %f16s8 = call i8 @llvm.fptosi.sat.i8.f16(half undef)
   %f16u8 = call i8 @llvm.fptoui.sat.i8.f16(half undef)
   %f16s16 = call i16 @llvm.fptosi.sat.i16.f16(half undef)
   %f16u16 = call i16 @llvm.fptoui.sat.i16.f16(half undef)