TTI: Estimate @llvm.fmuladd cost as fmul + fadd when FMA's aren't legal on the target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@208115 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/BasicTargetTransformInfo.cpp b/lib/CodeGen/BasicTargetTransformInfo.cpp
index 6ad1f28bb25423b6aa24fcb988578bfaf6a50abd..4b895092d3b63e80d9a5b5c98685614537a19d78 100644 (file)
--- a/lib/CodeGen/BasicTargetTransformInfo.cpp
+++ b/lib/CodeGen/BasicTargetTransformInfo.cpp
@@ -487,7 +487,7 @@ unsigned BasicTTI::getIntrinsicInstrCost(Intrinsic::ID IID, Type *RetTy,
   case Intrinsic::round:   ISD = ISD::FROUND; break;
   case Intrinsic::pow:     ISD = ISD::FPOW;   break;
   case Intrinsic::fma:     ISD = ISD::FMA;    break;
-  case Intrinsic::fmuladd: ISD = ISD::FMA;    break; // FIXME: mul + add?
+  case Intrinsic::fmuladd: ISD = ISD::FMA;    break;
   case Intrinsic::lifetime_start:
   case Intrinsic::lifetime_end:
     return 0;
@@ -512,6 +512,12 @@ unsigned BasicTTI::getIntrinsicInstrCost(Intrinsic::ID IID, Type *RetTy,
     return LT.first * 2;
   }
 
+  // If we can't lower fmuladd into an FMA estimate the cost as a floating
+  // point mul followed by an add.
+  if (IID == Intrinsic::fmuladd)
+    return TopTTI->getArithmeticInstrCost(BinaryOperator::FMul, RetTy) +
+           TopTTI->getArithmeticInstrCost(BinaryOperator::FAdd, RetTy);
+
   // Else, assume that we need to scalarize this intrinsic. For math builtins
   // this will emit a costly libcall, adding call overhead and spills. Make it
   // very expensive.

diff --git a/test/Analysis/CostModel/X86/intrinsic-cost.ll b/test/Analysis/CostModel/X86/intrinsic-cost.ll
index 8eeee8124d9ac5975f1715cca1e0c38e1b527b81..3b27b52f76549db5d577a0eb06bfedc54255158d 100644 (file)
--- a/test/Analysis/CostModel/X86/intrinsic-cost.ll
+++ b/test/Analysis/CostModel/X86/intrinsic-cost.ll
@@ -58,3 +58,31 @@ for.end:                                          ; preds = %vector.body
 }
 
 declare <4 x float> @llvm.nearbyint.v4f32(<4 x float>)  nounwind readnone
+
+define void @test3(float* nocapture %f, <4 x float> %b, <4 x float> %c) nounwind {
+vector.ph:
+  br label %vector.body
+
+vector.body:                                      ; preds = %vector.body, %vector.ph
+  %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
+  %0 = getelementptr inbounds float* %f, i64 %index
+  %1 = bitcast float* %0 to <4 x float>*
+  %wide.load = load <4 x float>* %1, align 4
+  %2 = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %wide.load, <4 x float> %b, <4 x float> %c)
+  store <4 x float> %2, <4 x float>* %1, align 4
+  %index.next = add i64 %index, 4
+  %3 = icmp eq i64 %index.next, 1024
+  br i1 %3, label %for.end, label %vector.body
+
+for.end:                                          ; preds = %vector.body
+  ret void
+
+; CORE2: Printing analysis 'Cost Model Analysis' for function 'test3':
+; CORE2: Cost Model: Found an estimated cost of 4 for instruction: %2 = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %wide.load, <4 x float> %b, <4 x float> %c)
+
+; COREI7: Printing analysis 'Cost Model Analysis' for function 'test3':
+; COREI7: Cost Model: Found an estimated cost of 4 for instruction: %2 = call <4 x float> @llvm.fmuladd.v4f32(<4 x float> %wide.load, <4 x float> %b, <4 x float> %c)
+
+}
+
+declare <4 x float> @llvm.fmuladd.v4f32(<4 x float>, <4 x float>, <4 x float>) nounwind readnone