Add a new target hook "predictableSelectIsExpensive".

This will be used to determine whether it's profitable to turn a select into a
branch when the branch is likely to be predicted.

Currently enabled for everything but Atom on X86 and Cortex-A9 devices on ARM.

I'm not entirely happy with the name of this flag, suggestions welcome ;)

llvm-svn: 156233

3 files changed, 7 insertions, 0 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 5132f01e866..5c9dc472370 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -605,6 +605,7 @@ TargetLowering::TargetLowering(const TargetMachine &tm,
   IntDivIsCheap = false;
   Pow2DivIsCheap = false;
   JumpIsExpensive = false;
+  predictableSelectIsExpensive = false;
   StackPointerRegisterToSaveRestore = 0;
   ExceptionPointerRegister = 0;
   ExceptionSelectorRegister = 0;
diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index 711321c6b6d..48cad6173b1 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -824,6 +824,9 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
 
   benefitFromCodePlacementOpt = true;
 
+  // Prefer likely predicted branches to selects on out-of-order cores.
+  predictableSelectIsExpensive = Subtarget->isCortexA9();
+
   setMinFunctionAlignment(Subtarget->isThumb() ? 1 : 2);
 }
 
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 7872a492f85..24d95a9977f 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1243,6 +1243,9 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
   setPrefLoopAlignment(4); // 2^4 bytes.
   benefitFromCodePlacementOpt = true;
 
+  // Predictable cmov don't hurt on atom because it's in-order.
+  predictableSelectIsExpensive = !Subtarget->isAtom();
+
   setPrefFunctionAlignment(4); // 2^4 bytes.
 }