[PowerPC] Add Branch Hints for Highly-Biased Branches

This branch adds hints for highly biased branches on the PPC architecture. Even
in absence of profiling information, LLVM will mark code reaching unreachable
terminators and other exceptional control flow constructs as highly unlikely to
be reached.

Patch by Tom Jablin!

llvm-svn: 255398

2 files changed, 74 insertions, 2 deletions

diff --git a/llvm/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.h b/llvm/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.h
index 6075631a541..acea600fbb0 100644
--- a/llvm/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.h
+++ b/llvm/lib/Target/PowerPC/MCTargetDesc/PPCPredicates.h
@@ -56,6 +56,14 @@ namespace PPC {
     PRED_BIT_UNSET = 1025
   };
   
+  // Bit for branch taken (plus) or not-taken (minus) hint
+  enum BranchHintBit {
+    BR_NO_HINT       = 0x0,
+    BR_NONTAKEN_HINT = 0x2,
+    BR_TAKEN_HINT    = 0x3,
+    BR_HINT_MASK     = 0X3
+  };
+
   /// Invert the specified predicate.  != -> ==, < -> >=.
   Predicate InvertPredicate(Predicate Opcode);
 
diff --git a/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp b/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
index 4dfa1650c1a..d57a070bcd5 100644
--- a/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
@@ -16,6 +16,8 @@
 #include "MCTargetDesc/PPCPredicates.h"
 #include "PPCMachineFunctionInfo.h"
 #include "PPCTargetMachine.h"
+#include "llvm/Analysis/BranchProbabilityInfo.h"
+#include "llvm/CodeGen/FunctionLoweringInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
@@ -52,6 +54,11 @@ static cl::opt<bool> BPermRewriterNoMasking(
              "bit permutations"),
     cl::Hidden);
 
+static cl::opt<bool> EnableBranchHint(
+  "ppc-use-branch-hint", cl::init(true),
+    cl::desc("Enable static hinting of branches on ppc"),
+    cl::Hidden);
+
 namespace llvm {
   void initializePPCDAGToDAGISelPass(PassRegistry&);
 }
@@ -393,6 +400,57 @@ static bool isInt32Immediate(SDValue N, unsigned &Imm) {
   return isInt32Immediate(N.getNode(), Imm);
 }
 
+static unsigned getBranchHint(unsigned PCC, FunctionLoweringInfo *FuncInfo,
+                              const SDValue &DestMBB) {
+  assert(isa<BasicBlockSDNode>(DestMBB));
+
+  if (!FuncInfo->BPI) return PPC::BR_NO_HINT;
+
+  const BasicBlock *BB = FuncInfo->MBB->getBasicBlock();
+  const TerminatorInst *BBTerm = BB->getTerminator();
+
+  if (BBTerm->getNumSuccessors() != 2) return PPC::BR_NO_HINT;
+
+  const BasicBlock *TBB = BBTerm->getSuccessor(0);
+  const BasicBlock *FBB = BBTerm->getSuccessor(1);
+
+  uint32_t TWeight = FuncInfo->BPI->getEdgeWeight(BB, TBB);
+  uint32_t FWeight = FuncInfo->BPI->getEdgeWeight(BB, FBB);
+
+  // We only want to handle cases which are easy to predict at static time, e.g.
+  // C++ throw statement, that is very likely not taken, or calling never
+  // returned function, e.g. stdlib exit(). So we set Threshold to filter
+  // unwanted cases.
+  //
+  // Below is LLVM branch weight table, we only want to handle case 1, 2
+  //
+  // Case                  Taken:Nontaken  Example
+  // 1. Unreachable        1048575:1       C++ throw, stdlib exit(),
+  // 2. Invoke-terminating 1:1048575
+  // 3. Coldblock          4:64            __builtin_expect
+  // 4. Loop Branch        124:4           For loop
+  // 5. PH/ZH/FPH          20:12
+  const uint32_t Threshold = 10000;
+
+  // Minimal weight should be at least 1
+  if (std::max(TWeight, FWeight) /
+      std::max(1u, std::min(TWeight, FWeight)) < Threshold)
+    return PPC::BR_NO_HINT;
+
+  DEBUG(dbgs() << "Use branch hint for '" << FuncInfo->Fn->getName() << "::"
+               << BB->getName() << "'\n"
+               << " -> " << TBB->getName() << ": " << TWeight << "\n"
+               << " -> " << FBB->getName() << ": " << FWeight << "\n");
+
+  const BasicBlockSDNode *BBDN = cast<BasicBlockSDNode>(DestMBB);
+
+  // If Dest BasicBlock is False-BasicBlock (FBB), swap branch weight,
+  // because we want 'TWeight' stands for 'branch weight' to Dest BasicBlock
+  if (BBDN->getBasicBlock()->getBasicBlock() != TBB)
+    std::swap(TWeight, FWeight);
+
+  return (TWeight > FWeight) ? PPC::BR_TAKEN_HINT : PPC::BR_NONTAKEN_HINT;
+}
 
 // isOpcWithIntImmediate - This method tests to see if the node is a specific
 // opcode and that it has a immediate integer right operand.
@@ -2840,8 +2898,11 @@ SDNode *PPCDAGToDAGISel::Select(SDNode *N) {
     // Op #3 is the Dest MBB
     // Op #4 is the Flag.
     // Prevent PPC::PRED_* from being selected into LI.
-    SDValue Pred =
-      getI32Imm(cast<ConstantSDNode>(N->getOperand(1))->getZExtValue(), dl);
+    unsigned PCC = cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
+    if (EnableBranchHint)
+      PCC |= getBranchHint(PCC, FuncInfo, N->getOperand(3));
+
+    SDValue Pred = getI32Imm(PCC, dl);
     SDValue Ops[] = { Pred, N->getOperand(2), N->getOperand(3),
       N->getOperand(0), N->getOperand(4) };
     return CurDAG->SelectNodeTo(N, PPC::BCC, MVT::Other, Ops);
@@ -2870,6 +2931,9 @@ SDNode *PPCDAGToDAGISel::Select(SDNode *N) {
                                   BitComp, N->getOperand(4), N->getOperand(0));
     }
 
+    if (EnableBranchHint)
+      PCC |= getBranchHint(PCC, FuncInfo, N->getOperand(4));
+
     SDValue CondCode = SelectCC(N->getOperand(2), N->getOperand(3), CC, dl);
     SDValue Ops[] = { getI32Imm(PCC, dl), CondCode,
                         N->getOperand(4), N->getOperand(0) };