ARM cost model: Fix cost of fptrunc and fpext instructions

A vector fptrunc and fpext simply gets split into scalar instructions.

radar://13192358

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177159 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 41 insertions, 2 deletions

diff --git a/lib/Target/ARM/ARMTargetTransformInfo.cpp b/lib/Target/ARM/ARMTargetTransformInfo.cpp
index 062d4d3f57..61b39e9a50 100644
--- a/lib/Target/ARM/ARMTargetTransformInfo.cpp
+++ b/lib/Target/ARM/ARMTargetTransformInfo.cpp
@@ -177,6 +177,23 @@ unsigned ARMTTI::getCastInstrCost(unsigned Opcode, Type *Dst,
   int ISD = TLI->InstructionOpcodeToISD(Opcode);
   assert(ISD && "Invalid opcode");
 
+  // Single to/from double precision conversions.
+  static const CostTblEntry<MVT> NEONFltDblTbl[] = {
+    // Vector fptrunc/fpext conversions.
+    { ISD::FP_ROUND,   MVT::v2f64, 2 },
+    { ISD::FP_EXTEND,  MVT::v2f32, 2 },
+    { ISD::FP_EXTEND,  MVT::v4f32, 4 }
+  };
+
+  if (Src->isVectorTy() && ST->hasNEON() && (ISD == ISD::FP_ROUND ||
+                                          ISD == ISD::FP_EXTEND)) {
+    std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Src);
+    int Idx = CostTableLookup<MVT>(NEONFltDblTbl, array_lengthof(NEONFltDblTbl),
+                                ISD, LT.second);
+    if (Idx != -1)
+      return LT.first * NEONFltDblTbl[Idx].Cost;
+  }
+
   EVT SrcTy = TLI->getValueType(Src);
   EVT DstTy = TLI->getValueType(Dst);
 
@@ -255,7 +272,6 @@ unsigned ARMTTI::getCastInstrCost(unsigned Opcode, Type *Dst,
         return NEONFloatConversionTbl[Idx].Cost;
   }
 
-
   // Scalar integer to float conversions.
   static const TypeConversionCostTblEntry<MVT> NEONIntegerConversionTbl[] = {
     { ISD::SINT_TO_FP,  MVT::f32, MVT::i1, 2 },
@@ -311,7 +327,6 @@ unsigned ARMTTI::getCastInstrCost(unsigned Opcode, Type *Dst,
       return ARMIntegerConversionTbl[Idx].Cost;
   }
 
-
   return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
 }
 
diff --git a/test/Analysis/CostModel/ARM/cast.ll b/test/Analysis/CostModel/ARM/cast.ll
index 779c4384e8..2b58f6ce38 100644
--- a/test/Analysis/CostModel/ARM/cast.ll
+++ b/test/Analysis/CostModel/ARM/cast.ll
@@ -166,6 +166,30 @@ define i32 @casts() {
   ; CHECK: cost of 38 {{.*}} trunc
   %r75 = trunc <16 x i32> undef to <16 x i8>
 
+  ; Floating point truncation costs.
+  ; CHECK: cost of 1 {{.*}} fptrunc double
+  %r80 = fptrunc double undef to float
+  ; CHECK: cost of 2 {{.*}} fptrunc <2 x double
+  %r81 = fptrunc <2 x double> undef to <2 x float>
+  ; CHECK: cost of 4 {{.*}} fptrunc <4 x double
+  %r82 = fptrunc <4 x double> undef to <4 x float>
+  ; CHECK: cost of 8 {{.*}} fptrunc <8 x double
+  %r83 = fptrunc <8 x double> undef to <8 x float>
+  ; CHECK: cost of 16 {{.*}} fptrunc <16 x double
+  %r84 = fptrunc <16 x double> undef to <16 x float>
+
+  ; Floating point extension costs.
+  ; CHECK: cost of 1 {{.*}} fpext float
+  %r85 = fpext float undef to double
+  ; CHECK: cost of 2 {{.*}} fpext <2 x float
+  %r86 = fpext <2 x float> undef to <2 x double>
+  ; CHECK: cost of 4 {{.*}} fpext <4 x float
+  %r87 = fpext <4 x float> undef to <4 x double>
+  ; CHECK: cost of 8 {{.*}} fpext <8 x float
+  %r88 = fpext <8 x float> undef to <8 x double>
+  ; CHECK: cost of 16 {{.*}} fpext <16 x float
+  %r89 = fpext <16 x float> undef to <16 x double>
+
   ;CHECK: cost of 0 {{.*}} ret
   ret i32 undef
 }