[SystemZ::TTI] Improve accuracy of costs for vector fp <-> int conversions

[jonpa]

Summary

Improve getCastInstrCost() by respecting the different types of Src and Dst for vector integer <-> fp conversions.

This means that extracting from integer becomes more expensive (by the extraction penalty), and the extraction from fp becomes cheaper (no longer has a false extraction penalty).

This only affects one file on SPEC, which gets one loop containing an uitofp no longer vectorized with VF=4, since the vector conversion now costs one more.

[uweigand]

LGTM, thanks!

[jonpa]

r346663

lib/Target/SystemZ/SystemZTargetTransformInfo.cpp

if (Opcode == Instruction::SIToFP || Opcode == Instruction::UIToFP ||
       // FP128 registers do not get inserted or extracted.
       if (DstScalarBits == 128 &&
           (Opcode == Instruction::SIToFP || Opcode == Instruction::UIToFP))
         NeedsInserts = false;
       if (SrcScalarBits == 128 &&
           (Opcode == Instruction::FPToSI || Opcode == Instruction::FPToUI))
         NeedsExtracts = false;
 
-      TotCost += getScalarizationOverhead(Dst, NeedsInserts, NeedsExtracts);
+      TotCost += getScalarizationOverhead(Src, false, NeedsExtracts);
+      TotCost += getScalarizationOverhead(Dst, NeedsInserts, false);
 
       // FIXME: VF 2 for float<->i32 is currently just as expensive as for VF 4.
       if (VF == 2 && SrcScalarBits == 32 && DstScalarBits == 32)
         TotCost *= 2;
 
       return TotCost;
     }
 

test/Analysis/CostModel/SystemZ/fp-cast.ll

 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v9 = fptosi float undef to i32
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v10 = fptosi float undef to i16
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v11 = fptosi float undef to i8
 ; CHECK: Cost Model: Found an estimated cost of 3 for instruction:   %v12 = fptosi <2 x fp128> undef to <2 x i64>
 ; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v13 = fptosi <2 x fp128> undef to <2 x i32>
 ; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v14 = fptosi <2 x fp128> undef to <2 x i16>
 ; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v15 = fptosi <2 x fp128> undef to <2 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v16 = fptosi <2 x double> undef to <2 x i64>
-; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v17 = fptosi <2 x double> undef to <2 x i32>
-; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v18 = fptosi <2 x double> undef to <2 x i16>
-; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v19 = fptosi <2 x double> undef to <2 x i8>
-; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v20 = fptosi <2 x float> undef to <2 x i64>
-; CHECK: Cost Model: Found an estimated cost of 14 for instruction:   %v21 = fptosi <2 x float> undef to <2 x i32>
-; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v22 = fptosi <2 x float> undef to <2 x i16>
-; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v23 = fptosi <2 x float> undef to <2 x i8>
+; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v17 = fptosi <2 x double> undef to <2 x i32>
+; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v18 = fptosi <2 x double> undef to <2 x i16>
+; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v19 = fptosi <2 x double> undef to <2 x i8>
+; CHECK: Cost Model: Found an estimated cost of 5 for instruction:   %v20 = fptosi <2 x float> undef to <2 x i64>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v21 = fptosi <2 x float> undef to <2 x i32>
+; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v22 = fptosi <2 x float> undef to <2 x i16>
+; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v23 = fptosi <2 x float> undef to <2 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v24 = fptosi <4 x fp128> undef to <4 x i64>
 ; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v25 = fptosi <4 x fp128> undef to <4 x i32>
 ; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v26 = fptosi <4 x fp128> undef to <4 x i16>
 ; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v27 = fptosi <4 x fp128> undef to <4 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v28 = fptosi <4 x double> undef to <4 x i64>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v29 = fptosi <4 x double> undef to <4 x i32>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v30 = fptosi <4 x double> undef to <4 x i16>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v31 = fptosi <4 x double> undef to <4 x i8>
-; CHECK: Cost Model: Found an estimated cost of 11 for instruction:   %v32 = fptosi <4 x float> undef to <4 x i64>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v33 = fptosi <4 x float> undef to <4 x i32>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v34 = fptosi <4 x float> undef to <4 x i16>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v35 = fptosi <4 x float> undef to <4 x i8>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v29 = fptosi <4 x double> undef to <4 x i32>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v30 = fptosi <4 x double> undef to <4 x i16>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v31 = fptosi <4 x double> undef to <4 x i8>
+; CHECK: Cost Model: Found an estimated cost of 10 for instruction:   %v32 = fptosi <4 x float> undef to <4 x i64>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v33 = fptosi <4 x float> undef to <4 x i32>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v34 = fptosi <4 x float> undef to <4 x i16>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v35 = fptosi <4 x float> undef to <4 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v36 = fptosi <8 x fp128> undef to <8 x i64>
 ; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v37 = fptosi <8 x fp128> undef to <8 x i32>
 ; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v38 = fptosi <8 x fp128> undef to <8 x i16>
 ; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v39 = fptosi <8 x fp128> undef to <8 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v40 = fptosi <8 x double> undef to <8 x i64>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v41 = fptosi <8 x double> undef to <8 x i32>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v42 = fptosi <8 x double> undef to <8 x i16>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v43 = fptosi <8 x double> undef to <8 x i8>
-; CHECK: Cost Model: Found an estimated cost of 21 for instruction:   %v44 = fptosi <8 x float> undef to <8 x i64>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v45 = fptosi <8 x float> undef to <8 x i32>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v46 = fptosi <8 x float> undef to <8 x i16>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v47 = fptosi <8 x float> undef to <8 x i8>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v41 = fptosi <8 x double> undef to <8 x i32>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v42 = fptosi <8 x double> undef to <8 x i16>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v43 = fptosi <8 x double> undef to <8 x i8>
+; CHECK: Cost Model: Found an estimated cost of 20 for instruction:   %v44 = fptosi <8 x float> undef to <8 x i64>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v45 = fptosi <8 x float> undef to <8 x i32>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v46 = fptosi <8 x float> undef to <8 x i16>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v47 = fptosi <8 x float> undef to <8 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v48 = fptosi <16 x double> undef to <16 x i64>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v49 = fptosi <16 x double> undef to <16 x i32>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v50 = fptosi <16 x double> undef to <16 x i16>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v51 = fptosi <16 x double> undef to <16 x i8>
-; CHECK: Cost Model: Found an estimated cost of 41 for instruction:   %v52 = fptosi <16 x float> undef to <16 x i64>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v53 = fptosi <16 x float> undef to <16 x i32>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v54 = fptosi <16 x float> undef to <16 x i16>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v55 = fptosi <16 x float> undef to <16 x i8>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v49 = fptosi <16 x double> undef to <16 x i32>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v50 = fptosi <16 x double> undef to <16 x i16>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v51 = fptosi <16 x double> undef to <16 x i8>
+; CHECK: Cost Model: Found an estimated cost of 40 for instruction:   %v52 = fptosi <16 x float> undef to <16 x i64>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v53 = fptosi <16 x float> undef to <16 x i32>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v54 = fptosi <16 x float> undef to <16 x i16>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v55 = fptosi <16 x float> undef to <16 x i8>
 
   ret void;
 }
 
 
 define void @fptoui() {
   %v0 = fptoui fp128 undef to i64
   %v1 = fptoui fp128 undef to i32

 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v9 = fptoui float undef to i32
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v10 = fptoui float undef to i16
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v11 = fptoui float undef to i8
 ; CHECK: Cost Model: Found an estimated cost of 3 for instruction:   %v12 = fptoui <2 x fp128> undef to <2 x i64>
 ; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v13 = fptoui <2 x fp128> undef to <2 x i32>
 ; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v14 = fptoui <2 x fp128> undef to <2 x i16>
 ; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v15 = fptoui <2 x fp128> undef to <2 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v16 = fptoui <2 x double> undef to <2 x i64>
-; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v17 = fptoui <2 x double> undef to <2 x i32>
-; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v18 = fptoui <2 x double> undef to <2 x i16>
-; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v19 = fptoui <2 x double> undef to <2 x i8>
-; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v20 = fptoui <2 x float> undef to <2 x i64>
-; CHECK: Cost Model: Found an estimated cost of 14 for instruction:   %v21 = fptoui <2 x float> undef to <2 x i32>
-; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v22 = fptoui <2 x float> undef to <2 x i16>
-; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v23 = fptoui <2 x float> undef to <2 x i8>
+; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v17 = fptoui <2 x double> undef to <2 x i32>
+; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v18 = fptoui <2 x double> undef to <2 x i16>
+; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v19 = fptoui <2 x double> undef to <2 x i8>
+; CHECK: Cost Model: Found an estimated cost of 5 for instruction:   %v20 = fptoui <2 x float> undef to <2 x i64>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v21 = fptoui <2 x float> undef to <2 x i32>
+; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v22 = fptoui <2 x float> undef to <2 x i16>
+; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v23 = fptoui <2 x float> undef to <2 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v24 = fptoui <4 x fp128> undef to <4 x i64>
 ; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v25 = fptoui <4 x fp128> undef to <4 x i32>
 ; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v26 = fptoui <4 x fp128> undef to <4 x i16>
 ; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v27 = fptoui <4 x fp128> undef to <4 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v28 = fptoui <4 x double> undef to <4 x i64>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v29 = fptoui <4 x double> undef to <4 x i32>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v30 = fptoui <4 x double> undef to <4 x i16>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v31 = fptoui <4 x double> undef to <4 x i8>
-; CHECK: Cost Model: Found an estimated cost of 11 for instruction:   %v32 = fptoui <4 x float> undef to <4 x i64>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v33 = fptoui <4 x float> undef to <4 x i32>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v34 = fptoui <4 x float> undef to <4 x i16>
-; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v35 = fptoui <4 x float> undef to <4 x i8>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v29 = fptoui <4 x double> undef to <4 x i32>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v30 = fptoui <4 x double> undef to <4 x i16>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v31 = fptoui <4 x double> undef to <4 x i8>
+; CHECK: Cost Model: Found an estimated cost of 10 for instruction:   %v32 = fptoui <4 x float> undef to <4 x i64>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v33 = fptoui <4 x float> undef to <4 x i32>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v34 = fptoui <4 x float> undef to <4 x i16>
+; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v35 = fptoui <4 x float> undef to <4 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v36 = fptoui <8 x fp128> undef to <8 x i64>
 ; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v37 = fptoui <8 x fp128> undef to <8 x i32>
 ; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v38 = fptoui <8 x fp128> undef to <8 x i16>
 ; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v39 = fptoui <8 x fp128> undef to <8 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v40 = fptoui <8 x double> undef to <8 x i64>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v41 = fptoui <8 x double> undef to <8 x i32>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v42 = fptoui <8 x double> undef to <8 x i16>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v43 = fptoui <8 x double> undef to <8 x i8>
-; CHECK: Cost Model: Found an estimated cost of 21 for instruction:   %v44 = fptoui <8 x float> undef to <8 x i64>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v45 = fptoui <8 x float> undef to <8 x i32>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v46 = fptoui <8 x float> undef to <8 x i16>
-; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v47 = fptoui <8 x float> undef to <8 x i8>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v41 = fptoui <8 x double> undef to <8 x i32>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v42 = fptoui <8 x double> undef to <8 x i16>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v43 = fptoui <8 x double> undef to <8 x i8>
+; CHECK: Cost Model: Found an estimated cost of 20 for instruction:   %v44 = fptoui <8 x float> undef to <8 x i64>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v45 = fptoui <8 x float> undef to <8 x i32>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v46 = fptoui <8 x float> undef to <8 x i16>
+; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v47 = fptoui <8 x float> undef to <8 x i8>
 ; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v48 = fptoui <16 x double> undef to <16 x i64>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v49 = fptoui <16 x double> undef to <16 x i32>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v50 = fptoui <16 x double> undef to <16 x i16>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v51 = fptoui <16 x double> undef to <16 x i8>
-; CHECK: Cost Model: Found an estimated cost of 41 for instruction:   %v52 = fptoui <16 x float> undef to <16 x i64>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v53 = fptoui <16 x float> undef to <16 x i32>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v54 = fptoui <16 x float> undef to <16 x i16>
-; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v55 = fptoui <16 x float> undef to <16 x i8>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v49 = fptoui <16 x double> undef to <16 x i32>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v50 = fptoui <16 x double> undef to <16 x i16>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v51 = fptoui <16 x double> undef to <16 x i8>
+; CHECK: Cost Model: Found an estimated cost of 40 for instruction:   %v52 = fptoui <16 x float> undef to <16 x i64>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v53 = fptoui <16 x float> undef to <16 x i32>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v54 = fptoui <16 x float> undef to <16 x i16>
+; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v55 = fptoui <16 x float> undef to <16 x i8>
 
   ret void;
 }
 
 define void @fptrunc() {
   %v0 = fptrunc fp128 undef to double
   %v1 = fptrunc fp128 undef to float
   %v2 = fptrunc double undef to float

 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v4 = sitofp i32 undef to double
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v5 = sitofp i32 undef to float
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v6 = sitofp i16 undef to fp128
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v7 = sitofp i16 undef to double
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v8 = sitofp i16 undef to float
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v9 = sitofp i8 undef to fp128
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v10 = sitofp i8 undef to double
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v11 = sitofp i8 undef to float
-; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v12 = sitofp <2 x i64> undef to <2 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 5 for instruction:   %v12 = sitofp <2 x i64> undef to <2 x fp128>
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v13 = sitofp <2 x i64> undef to <2 x double>
-; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v14 = sitofp <2 x i64> undef to <2 x float>
-; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v15 = sitofp <2 x i32> undef to <2 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v16 = sitofp <2 x i32> undef to <2 x double>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v17 = sitofp <2 x i32> undef to <2 x float>
-; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v18 = sitofp <2 x i16> undef to <2 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v19 = sitofp <2 x i16> undef to <2 x double>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v20 = sitofp <2 x i16> undef to <2 x float>
-; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v21 = sitofp <2 x i8> undef to <2 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v22 = sitofp <2 x i8> undef to <2 x double>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v23 = sitofp <2 x i8> undef to <2 x float>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v24 = sitofp <4 x i64> undef to <4 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v14 = sitofp <2 x i64> undef to <2 x float>
+; CHECK: Cost Model: Found an estimated cost of 5 for instruction:   %v15 = sitofp <2 x i32> undef to <2 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v16 = sitofp <2 x i32> undef to <2 x double>
+; CHECK: Cost Model: Found an estimated cost of 14 for instruction:   %v17 = sitofp <2 x i32> undef to <2 x float>
+; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v18 = sitofp <2 x i16> undef to <2 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v19 = sitofp <2 x i16> undef to <2 x double>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v20 = sitofp <2 x i16> undef to <2 x float>
+; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v21 = sitofp <2 x i8> undef to <2 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v22 = sitofp <2 x i8> undef to <2 x double>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v23 = sitofp <2 x i8> undef to <2 x float>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v24 = sitofp <4 x i64> undef to <4 x fp128>
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v25 = sitofp <4 x i64> undef to <4 x double>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v26 = sitofp <4 x i64> undef to <4 x float>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v27 = sitofp <4 x i32> undef to <4 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v28 = sitofp <4 x i32> undef to <4 x double>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v29 = sitofp <4 x i32> undef to <4 x float>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v30 = sitofp <4 x i16> undef to <4 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v31 = sitofp <4 x i16> undef to <4 x double>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v32 = sitofp <4 x i16> undef to <4 x float>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v33 = sitofp <4 x i8> undef to <4 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v34 = sitofp <4 x i8> undef to <4 x double>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v35 = sitofp <4 x i8> undef to <4 x float>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v36 = sitofp <8 x i64> undef to <8 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v26 = sitofp <4 x i64> undef to <4 x float>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v27 = sitofp <4 x i32> undef to <4 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v28 = sitofp <4 x i32> undef to <4 x double>
+; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v29 = sitofp <4 x i32> undef to <4 x float>
+; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v30 = sitofp <4 x i16> undef to <4 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v31 = sitofp <4 x i16> undef to <4 x double>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v32 = sitofp <4 x i16> undef to <4 x float>
+; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v33 = sitofp <4 x i8> undef to <4 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v34 = sitofp <4 x i8> undef to <4 x double>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v35 = sitofp <4 x i8> undef to <4 x float>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v36 = sitofp <8 x i64> undef to <8 x fp128>
 ; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v37 = sitofp <8 x i64> undef to <8 x double>
-; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v38 = sitofp <8 x i64> undef to <8 x float>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v39 = sitofp <8 x i32> undef to <8 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v40 = sitofp <8 x i32> undef to <8 x double>
-; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v41 = sitofp <8 x i32> undef to <8 x float>
-; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v42 = sitofp <8 x i16> undef to <8 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 32 for instruction:   %v43 = sitofp <8 x i16> undef to <8 x double>
-; CHECK: Cost Model: Found an estimated cost of 32 for instruction:   %v44 = sitofp <8 x i16> undef to <8 x float>
-; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v45 = sitofp <8 x i8> undef to <8 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 32 for instruction:   %v46 = sitofp <8 x i8> undef to <8 x double>
-; CHECK: Cost Model: Found an estimated cost of 32 for instruction:   %v47 = sitofp <8 x i8> undef to <8 x float>
+; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v38 = sitofp <8 x i64> undef to <8 x float>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v39 = sitofp <8 x i32> undef to <8 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v40 = sitofp <8 x i32> undef to <8 x double>
+; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v41 = sitofp <8 x i32> undef to <8 x float>
+; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v42 = sitofp <8 x i16> undef to <8 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 33 for instruction:   %v43 = sitofp <8 x i16> undef to <8 x double>
+; CHECK: Cost Model: Found an estimated cost of 33 for instruction:   %v44 = sitofp <8 x i16> undef to <8 x float>
+; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v45 = sitofp <8 x i8> undef to <8 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 33 for instruction:   %v46 = sitofp <8 x i8> undef to <8 x double>
+; CHECK: Cost Model: Found an estimated cost of 33 for instruction:   %v47 = sitofp <8 x i8> undef to <8 x float>
 ; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v48 = sitofp <16 x i64> undef to <16 x double>
-; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v49 = sitofp <16 x i64> undef to <16 x float>
-; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v50 = sitofp <16 x i32> undef to <16 x double>
-; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v51 = sitofp <16 x i32> undef to <16 x float>
-; CHECK: Cost Model: Found an estimated cost of 64 for instruction:   %v52 = sitofp <16 x i16> undef to <16 x double>
-; CHECK: Cost Model: Found an estimated cost of 64 for instruction:   %v53 = sitofp <16 x i16> undef to <16 x float>
-; CHECK: Cost Model: Found an estimated cost of 64 for instruction:   %v54 = sitofp <16 x i8> undef to <16 x double>
-; CHECK: Cost Model: Found an estimated cost of 64 for instruction:   %v55 = sitofp <16 x i8> undef to <16 x float>
+; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v49 = sitofp <16 x i64> undef to <16 x float>
+; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v50 = sitofp <16 x i32> undef to <16 x double>
+; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v51 = sitofp <16 x i32> undef to <16 x float>
+; CHECK: Cost Model: Found an estimated cost of 65 for instruction:   %v52 = sitofp <16 x i16> undef to <16 x double>
+; CHECK: Cost Model: Found an estimated cost of 65 for instruction:   %v53 = sitofp <16 x i16> undef to <16 x float>
+; CHECK: Cost Model: Found an estimated cost of 65 for instruction:   %v54 = sitofp <16 x i8> undef to <16 x double>
+; CHECK: Cost Model: Found an estimated cost of 65 for instruction:   %v55 = sitofp <16 x i8> undef to <16 x float>
 
   ret void;
 }
 
 define void @uitofp() {
   %v0 = uitofp i64 undef to fp128
   %v1 = uitofp i64 undef to double
   %v2 = uitofp i64 undef to float

 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v4 = uitofp i32 undef to double
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v5 = uitofp i32 undef to float
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v6 = uitofp i16 undef to fp128
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v7 = uitofp i16 undef to double
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v8 = uitofp i16 undef to float
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v9 = uitofp i8 undef to fp128
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v10 = uitofp i8 undef to double
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v11 = uitofp i8 undef to float
-; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v12 = uitofp <2 x i64> undef to <2 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 5 for instruction:   %v12 = uitofp <2 x i64> undef to <2 x fp128>
 ; CHECK: Cost Model: Found an estimated cost of 1 for instruction:   %v13 = uitofp <2 x i64> undef to <2 x double>
-; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v14 = uitofp <2 x i64> undef to <2 x float>
-; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v15 = uitofp <2 x i32> undef to <2 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v16 = uitofp <2 x i32> undef to <2 x double>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v17 = uitofp <2 x i32> undef to <2 x float>
-; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v18 = uitofp <2 x i16> undef to <2 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v19 = uitofp <2 x i16> undef to <2 x double>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v20 = uitofp <2 x i16> undef to <2 x float>
-; CHECK: Cost Model: Found an estimated cost of 6 for instruction:   %v21 = uitofp <2 x i8> undef to <2 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v22 = uitofp <2 x i8> undef to <2 x double>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v23 = uitofp <2 x i8> undef to <2 x float>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v24 = uitofp <4 x i64> undef to <4 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v14 = uitofp <2 x i64> undef to <2 x float>
+; CHECK: Cost Model: Found an estimated cost of 5 for instruction:   %v15 = uitofp <2 x i32> undef to <2 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v16 = uitofp <2 x i32> undef to <2 x double>
+; CHECK: Cost Model: Found an estimated cost of 14 for instruction:   %v17 = uitofp <2 x i32> undef to <2 x float>
+; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v18 = uitofp <2 x i16> undef to <2 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v19 = uitofp <2 x i16> undef to <2 x double>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v20 = uitofp <2 x i16> undef to <2 x float>
+; CHECK: Cost Model: Found an estimated cost of 7 for instruction:   %v21 = uitofp <2 x i8> undef to <2 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v22 = uitofp <2 x i8> undef to <2 x double>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v23 = uitofp <2 x i8> undef to <2 x float>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v24 = uitofp <4 x i64> undef to <4 x fp128>
 ; CHECK: Cost Model: Found an estimated cost of 2 for instruction:   %v25 = uitofp <4 x i64> undef to <4 x double>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v26 = uitofp <4 x i64> undef to <4 x float>
-; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v27 = uitofp <4 x i32> undef to <4 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v28 = uitofp <4 x i32> undef to <4 x double>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v29 = uitofp <4 x i32> undef to <4 x float>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v30 = uitofp <4 x i16> undef to <4 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v31 = uitofp <4 x i16> undef to <4 x double>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v32 = uitofp <4 x i16> undef to <4 x float>
-; CHECK: Cost Model: Found an estimated cost of 12 for instruction:   %v33 = uitofp <4 x i8> undef to <4 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v34 = uitofp <4 x i8> undef to <4 x double>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v35 = uitofp <4 x i8> undef to <4 x float>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v36 = uitofp <8 x i64> undef to <8 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v26 = uitofp <4 x i64> undef to <4 x float>
+; CHECK: Cost Model: Found an estimated cost of 9 for instruction:   %v27 = uitofp <4 x i32> undef to <4 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v28 = uitofp <4 x i32> undef to <4 x double>
+; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v29 = uitofp <4 x i32> undef to <4 x float>
+; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v30 = uitofp <4 x i16> undef to <4 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v31 = uitofp <4 x i16> undef to <4 x double>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v32 = uitofp <4 x i16> undef to <4 x float>
+; CHECK: Cost Model: Found an estimated cost of 13 for instruction:   %v33 = uitofp <4 x i8> undef to <4 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v34 = uitofp <4 x i8> undef to <4 x double>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v35 = uitofp <4 x i8> undef to <4 x float>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v36 = uitofp <8 x i64> undef to <8 x fp128>
 ; CHECK: Cost Model: Found an estimated cost of 4 for instruction:   %v37 = uitofp <8 x i64> undef to <8 x double>
-; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v38 = uitofp <8 x i64> undef to <8 x float>
-; CHECK: Cost Model: Found an estimated cost of 16 for instruction:   %v39 = uitofp <8 x i32> undef to <8 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v40 = uitofp <8 x i32> undef to <8 x double>
-; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v41 = uitofp <8 x i32> undef to <8 x float>
-; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v42 = uitofp <8 x i16> undef to <8 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 32 for instruction:   %v43 = uitofp <8 x i16> undef to <8 x double>
-; CHECK: Cost Model: Found an estimated cost of 32 for instruction:   %v44 = uitofp <8 x i16> undef to <8 x float>
-; CHECK: Cost Model: Found an estimated cost of 24 for instruction:   %v45 = uitofp <8 x i8> undef to <8 x fp128>
-; CHECK: Cost Model: Found an estimated cost of 32 for instruction:   %v46 = uitofp <8 x i8> undef to <8 x double>
-; CHECK: Cost Model: Found an estimated cost of 32 for instruction:   %v47 = uitofp <8 x i8> undef to <8 x float>
+; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v38 = uitofp <8 x i64> undef to <8 x float>
+; CHECK: Cost Model: Found an estimated cost of 17 for instruction:   %v39 = uitofp <8 x i32> undef to <8 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v40 = uitofp <8 x i32> undef to <8 x double>
+; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v41 = uitofp <8 x i32> undef to <8 x float>
+; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v42 = uitofp <8 x i16> undef to <8 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 33 for instruction:   %v43 = uitofp <8 x i16> undef to <8 x double>
+; CHECK: Cost Model: Found an estimated cost of 33 for instruction:   %v44 = uitofp <8 x i16> undef to <8 x float>
+; CHECK: Cost Model: Found an estimated cost of 25 for instruction:   %v45 = uitofp <8 x i8> undef to <8 x fp128>
+; CHECK: Cost Model: Found an estimated cost of 33 for instruction:   %v46 = uitofp <8 x i8> undef to <8 x double>
+; CHECK: Cost Model: Found an estimated cost of 33 for instruction:   %v47 = uitofp <8 x i8> undef to <8 x float>
 ; CHECK: Cost Model: Found an estimated cost of 8 for instruction:   %v48 = uitofp <16 x i64> undef to <16 x double>
-; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v49 = uitofp <16 x i64> undef to <16 x float>
-; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v50 = uitofp <16 x i32> undef to <16 x double>
-; CHECK: Cost Model: Found an estimated cost of 48 for instruction:   %v51 = uitofp <16 x i32> undef to <16 x float>
-; CHECK: Cost Model: Found an estimated cost of 64 for instruction:   %v52 = uitofp <16 x i16> undef to <16 x double>
-; CHECK: Cost Model: Found an estimated cost of 64 for instruction:   %v53 = uitofp <16 x i16> undef to <16 x float>
-; CHECK: Cost Model: Found an estimated cost of 64 for instruction:   %v54 = uitofp <16 x i8> undef to <16 x double>
-; CHECK: Cost Model: Found an estimated cost of 64 for instruction:   %v55 = uitofp <16 x i8> undef to <16 x float>
+; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v49 = uitofp <16 x i64> undef to <16 x float>
+; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v50 = uitofp <16 x i32> undef to <16 x double>
+; CHECK: Cost Model: Found an estimated cost of 49 for instruction:   %v51 = uitofp <16 x i32> undef to <16 x float>
+; CHECK: Cost Model: Found an estimated cost of 65 for instruction:   %v52 = uitofp <16 x i16> undef to <16 x double>
+; CHECK: Cost Model: Found an estimated cost of 65 for instruction:   %v53 = uitofp <16 x i16> undef to <16 x float>
+; CHECK: Cost Model: Found an estimated cost of 65 for instruction:   %v54 = uitofp <16 x i8> undef to <16 x double>
+; CHECK: Cost Model: Found an estimated cost of 65 for instruction:   %v55 = uitofp <16 x i8> undef to <16 x float>
 
   ret void;
 }
 
 define void @sitofp_extload(i16 *%src16, i8 *%src8) {
   %ld16 = load i16, i16 *%src16
   %v6 = sitofp i16 %ld16 to fp128
   %v7 = sitofp i16 %ld16 to double