[CostModel][X86] Fix SSE1 FADD/FSUB costs

Noticed in D56011 - handle the case that scalar fp ops are quicker on P3 than P4

Add the other costs so that we're not relying on the default "is legal/custom" cost logic.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@350403 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86TargetTransformInfo.cpp b/lib/Target/X86/X86TargetTransformInfo.cpp
index 89f45dd..bce3a06 100644 (file)
--- a/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -832,6 +832,12 @@ int X86TTIImpl::getArithmeticInstrCost(
     { ISD::FDIV, MVT::v4f32,      39 }, // Pentium IV from http://www.agner.org/
     { ISD::FDIV, MVT::f64,        38 }, // Pentium IV from http://www.agner.org/
     { ISD::FDIV, MVT::v2f64,      69 }, // Pentium IV from http://www.agner.org/
+
+    { ISD::FADD, MVT::f32,         2 }, // Pentium IV from http://www.agner.org/
+    { ISD::FADD, MVT::f64,         2 }, // Pentium IV from http://www.agner.org/
+
+    { ISD::FSUB, MVT::f32,         2 }, // Pentium IV from http://www.agner.org/
+    { ISD::FSUB, MVT::f64,         2 }, // Pentium IV from http://www.agner.org/
   };
 
   if (ST->hasSSE2())
@@ -841,6 +847,12 @@ int X86TTIImpl::getArithmeticInstrCost(
   static const CostTblEntry SSE1CostTable[] = {
     { ISD::FDIV, MVT::f32,   17 }, // Pentium III from http://www.agner.org/
     { ISD::FDIV, MVT::v4f32, 34 }, // Pentium III from http://www.agner.org/
+
+    { ISD::FADD, MVT::f32,    1 }, // Pentium III from http://www.agner.org/
+    { ISD::FADD, MVT::v4f32,  2 }, // Pentium III from http://www.agner.org/
+
+    { ISD::FSUB, MVT::f32,    1 }, // Pentium III from http://www.agner.org/
+    { ISD::FSUB, MVT::v4f32,  2 }, // Pentium III from http://www.agner.org/
   };
 
   if (ST->hasSSE1())

diff --git a/test/Analysis/CostModel/X86/arith-fp.ll b/test/Analysis/CostModel/X86/arith-fp.ll
index bd62bd9..bce7ca8 100644 (file)
--- a/test/Analysis/CostModel/X86/arith-fp.ll
+++ b/test/Analysis/CostModel/X86/arith-fp.ll
@@ -16,7 +16,7 @@ target triple = "x86_64-apple-macosx10.8.0"
 
 define i32 @fadd(i32 %arg) {
 ; SSE1-LABEL: 'fadd'
-; SSE1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F32 = fadd float undef, undef
+; SSE1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fadd float undef, undef
 ; SSE1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = fadd <4 x float> undef, undef
 ; SSE1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F32 = fadd <8 x float> undef, undef
 ; SSE1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F32 = fadd <16 x float> undef, undef
@@ -129,7 +129,7 @@ define i32 @fadd(i32 %arg) {
 
 define i32 @fsub(i32 %arg) {
 ; SSE1-LABEL: 'fsub'
-; SSE1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %F32 = fsub float undef, undef
+; SSE1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %F32 = fsub float undef, undef
 ; SSE1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = fsub <4 x float> undef, undef
 ; SSE1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F32 = fsub <8 x float> undef, undef
 ; SSE1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V16F32 = fsub <16 x float> undef, undef