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 ;)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@156233 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/Target/TargetLowering.h b/include/llvm/Target/TargetLowering.h
index 78ccb12d90ffc85ee3542337d63773c36e679c94..29abc191dda021f2e3e35a4da0b67127564d6c04 100644 (file)
--- a/include/llvm/Target/TargetLowering.h
+++ b/include/llvm/Target/TargetLowering.h
@@ -150,6 +150,12 @@ public:
   /// that should be avoided.
   bool isJumpExpensive() const { return JumpIsExpensive; }
 
+  /// isPredictableSelectExpensive - Return true if selects are only cheaper
+  /// than branches if the branch is unlikely to be predicted right.
+  bool isPredictableSelectExpensive() const {
+    return predictableSelectIsExpensive;
+  }
+
   /// getSetCCResultType - Return the ValueType of the result of SETCC
   /// operations.  Also used to obtain the target's preferred type for
   /// the condition operand of SELECT and BRCOND nodes.  In the case of
@@ -2028,6 +2034,10 @@ protected:
   /// optimization.
   bool benefitFromCodePlacementOpt;
 
+  /// predictableSelectIsExpensive - Tells the code generator that select is
+  /// more expensive than a branch if the branch is usually predicted right.
+  bool predictableSelectIsExpensive;
+
 private:
   /// isLegalRC - Return true if the value types that can be represented by the
   /// specified register class are all legal.

diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 5132f01e86610672feeea651d25872e5484a49da..5c9dc472370ee6a2790509b7fcb25d19109f9893 100644 (file)
--- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/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/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index 711321c6b6d37c0635a1a4bfa31f6d5ba8bd27b1..48cad6173b1fbb12eace78c5a84715c0cbb9e4e9 100644 (file)
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/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/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 7872a492f85f8005e808f4c0821944eedf3388c9..24d95a9977f3ea40aa24de615bdf84513f439a62 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/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.
 }