Introduce "expect" intrinsic instructions.

llvm-svn: 134516

include/llvm/InitializePasses.h

@@ -141,6 +141,7 @@ void initializeLoopUnrollPass(PassRegistry&);
 void initializeLoopUnswitchPass(PassRegistry&);
 void initializeLoopIdiomRecognizePass(PassRegistry&);
 void initializeLowerAtomicPass(PassRegistry&);
+void initializeLowerExpectIntrinsicPass(PassRegistry&);
 void initializeLowerIntrinsicsPass(PassRegistry&);
 void initializeLowerInvokePass(PassRegistry&);
 void initializeLowerSetJmpPass(PassRegistry&);

include/llvm/Intrinsics.td

@@ -266,6 +266,11 @@ def int_objectsize : Intrinsic<[llvm_anyint_ty], [llvm_ptr_ty, llvm_i1_ty],
                                [IntrNoMem]>,
                                GCCBuiltin<"__builtin_object_size">;
 
+//===------------------------- Expect Intrinsics --------------------------===//
+//
+def int_expect : Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>,
+                                              LLVMMatchType<0>], [IntrNoMem]>;
+
 //===-------------------- Bit Manipulation Intrinsics ---------------------===//
 //
 

include/llvm/LLVMContext.h

@@ -39,7 +39,8 @@ public:
   // compile-time performance optimization, not a correctness optimization.
   enum {
     MD_dbg = 0,  // "dbg"
-    MD_tbaa = 1  // "tbaa"
+    MD_tbaa = 1, // "tbaa"
+    MD_prof = 2  // "prof"
   };
   
   /// getMDKindID - Return a unique non-zero ID for the specified metadata kind.

include/llvm/LinkAllPasses.h

@@ -92,6 +92,7 @@ namespace {
       (void) llvm::createLoopUnswitchPass();
       (void) llvm::createLoopIdiomPass();
       (void) llvm::createLoopRotatePass();
+      (void) llvm::createLowerExpectIntrinsicPass();
       (void) llvm::createLowerInvokePass();
       (void) llvm::createLowerSetJmpPass();
       (void) llvm::createLowerSwitchPass();

include/llvm/Support/PassManagerBuilder.h

@@ -152,6 +152,7 @@ public:
     FPM.add(createCFGSimplificationPass());
     FPM.add(createScalarReplAggregatesPass());
     FPM.add(createEarlyCSEPass());
+    FPM.add(createLowerExpectIntrinsicPass());
   }
   
   /// populateModulePassManager - This sets up the primary pass manager.

include/llvm/Transforms/Scalar.h

@@ -361,6 +361,14 @@ Pass *createObjCARCOptPass();
 FunctionPass *createInstructionSimplifierPass();
 extern char &InstructionSimplifierID;
 
+
+//===----------------------------------------------------------------------===//
+//
+// LowerExpectIntriniscs - Removes llvm.expect intrinsics and creates
+// "block_weights" metadata.
+FunctionPass *createLowerExpectIntrinsicPass();
+
+
 } // End llvm namespace
 
 #endif

lib/CodeGen/IntrinsicLowering.cpp

@@ -353,6 +353,13 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) {
     report_fatal_error("Code generator does not support intrinsic function '"+
                       Callee->getName()+"'!");
 
+  case Intrinsic::expect: {
+    // Just replace __builtin_expect(exp, c) with EXP.
+    Value *V = CI->getArgOperand(0);
+    CI->replaceAllUsesWith(V);
+    break;
+  }
+
     // The setjmp/longjmp intrinsics should only exist in the code if it was
     // never optimized (ie, right out of the CFE), or if it has been hacked on
     // by the lowerinvoke pass.  In both cases, the right thing to do is to

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

@@ -4771,6 +4771,13 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) {
   case Intrinsic::flt_rounds:
     setValue(&I, DAG.getNode(ISD::FLT_ROUNDS_, dl, MVT::i32));
     return 0;
+
+  case Intrinsic::expect: {
+    // Just replace __builtin_expect(exp, c) with EXP.
+    setValue(&I, getValue(I.getArgOperand(0)));
+    return 0;
+  }
+
   case Intrinsic::trap: {
     StringRef TrapFuncName = getTrapFunctionName();
     if (TrapFuncName.empty()) {

lib/Transforms/Scalar/Scalar.cpp

@@ -48,6 +48,7 @@ void llvm::initializeScalarOpts(PassRegistry &Registry) {
   initializeLoopUnswitchPass(Registry);
   initializeLoopIdiomRecognizePass(Registry);
   initializeLowerAtomicPass(Registry);
+  initializeLowerExpectIntrinsicPass(Registry);
   initializeMemCpyOptPass(Registry);
   initializeObjCARCAliasAnalysisPass(Registry);
   initializeObjCARCExpandPass(Registry);

lib/Transforms/Utils/CMakeLists.txt

@@ -14,6 +14,7 @@ add_llvm_library(LLVMTransformUtils
   Local.cpp
   LoopSimplify.cpp
   LoopUnroll.cpp
+  LowerExpectIntrinsic.cpp
   LowerInvoke.cpp
   LowerSwitch.cpp
   Mem2Reg.cpp

lib/Transforms/Utils/LowerExpectIntrinsic.cpp

@@ -0,0 +1,163 @@
+#define DEBUG_TYPE "lower-expect-intrinsic"
+#include "llvm/Constants.h"
+#include "llvm/Function.h"
+#include "llvm/BasicBlock.h"
+#include "llvm/LLVMContext.h"
+#include "llvm/Instructions.h"
+#include "llvm/Intrinsics.h"
+#include "llvm/Metadata.h"
+#include "llvm/Pass.h"
+#include "llvm/Transforms/Scalar.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/ADT/Statistic.h"
+#include <vector>
+
+using namespace llvm;
+
+STATISTIC(IfHandled, "Number of 'expect' intrinsic intructions handled");
+
+static cl::opt<uint32_t>
+LikelyBranchWeight("likely-branch-weight", cl::Hidden, cl::init(64),
+                   cl::desc("Weight of the branch likely to be taken (default = 64)"));
+static cl::opt<uint32_t>
+UnlikelyBranchWeight("unlikely-branch-weight", cl::Hidden, cl::init(4),
+                   cl::desc("Weight of the branch unlikely to be taken (default = 4)"));
+
+namespace {
+
+  class LowerExpectIntrinsic : public FunctionPass {
+
+    bool HandleSwitchExpect(SwitchInst *SI);
+
+    bool HandleIfExpect(BranchInst *BI);
+
+  public:
+    static char ID;
+    LowerExpectIntrinsic() : FunctionPass(ID) {
+      initializeLowerExpectIntrinsicPass(*PassRegistry::getPassRegistry());
+    }
+
+    bool runOnFunction(Function &F);
+  };
+}
+
+
+bool LowerExpectIntrinsic::HandleSwitchExpect(SwitchInst *SI) {
+  CallInst *CI = dyn_cast<CallInst>(SI->getCondition());
+  if (!CI)
+    return false;
+
+  Function *Fn = CI->getCalledFunction();
+  if (!Fn || Fn->getIntrinsicID() != Intrinsic::expect)
+    return false;
+
+  Value *ArgValue = CI->getArgOperand(0);
+  ConstantInt *ExpectedValue = dyn_cast<ConstantInt>(CI->getArgOperand(1));
+  if (!ExpectedValue)
+    return false;
+
+  LLVMContext &Context = CI->getContext();
+  const Type *Int32Ty = Type::getInt32Ty(Context);
+
+  unsigned caseNo = SI->findCaseValue(ExpectedValue);
+  std::vector<Value *> Vec;
+  unsigned n = SI->getNumCases();
+  Vec.resize(n + 1); // +1 for MDString
+
+  Vec[0] = MDString::get(Context, "branch_weights");
+  for (unsigned i = 0; i < n; ++i) {
+    Vec[i + 1] = ConstantInt::get(Int32Ty, i == caseNo ? LikelyBranchWeight : UnlikelyBranchWeight);
+  }
+
+  MDNode *WeightsNode = llvm::MDNode::get(Context, Vec);
+  SI->setMetadata(LLVMContext::MD_prof, WeightsNode);
+
+  SI->setCondition(ArgValue);
+  return true;
+}
+
+
+bool LowerExpectIntrinsic::HandleIfExpect(BranchInst *BI) {
+  if (BI->isUnconditional())
+    return false;
+
+  // Handle non-optimized IR code like:
+  //   %expval = call i64 @llvm.expect.i64.i64(i64 %conv1, i64 1)
+  //   %tobool = icmp ne i64 %expval, 0
+  //   br i1 %tobool, label %if.then, label %if.end
+
+  ICmpInst *CmpI = dyn_cast<ICmpInst>(BI->getCondition());
+  if (!CmpI || CmpI->getPredicate() != CmpInst::ICMP_NE)
+    return false;
+
+  CallInst *CI = dyn_cast<CallInst>(CmpI->getOperand(0));
+  if (!CI)
+    return false;
+
+  Function *Fn = CI->getCalledFunction();
+  if (!Fn || Fn->getIntrinsicID() != Intrinsic::expect)
+    return false;
+
+  Value *ArgValue = CI->getArgOperand(0);
+  ConstantInt *ExpectedValue = dyn_cast<ConstantInt>(CI->getArgOperand(1));
+  if (!ExpectedValue)
+    return false;
+
+  LLVMContext &Context = CI->getContext();
+  const Type *Int32Ty = Type::getInt32Ty(Context);
+  bool Likely = ExpectedValue->isOne();
+
+  // If expect value is equal to 1 it means that we are more likely to take
+  // branch 0, in other case more likely is branch 1.
+  Value *Ops[] = {
+    MDString::get(Context, "branch_weights"),
+    ConstantInt::get(Int32Ty, Likely ? LikelyBranchWeight : UnlikelyBranchWeight),
+    ConstantInt::get(Int32Ty, Likely ? UnlikelyBranchWeight : LikelyBranchWeight)
+  };
+
+  MDNode *WeightsNode = MDNode::get(Context, ArrayRef<Value *>(Ops, 3));
+  BI->setMetadata(LLVMContext::MD_prof, WeightsNode);
+
+  CmpI->setOperand(0, ArgValue);
+  return true;
+}
+
+
+bool LowerExpectIntrinsic::runOnFunction(Function &F) {
+  for (Function::iterator I = F.begin(), E = F.end(); I != E;) {
+    BasicBlock *BB = I++;
+
+    // Create "block_weights" metadata.
+    if (BranchInst *BI = dyn_cast<BranchInst>(BB->getTerminator())) {
+      if (HandleIfExpect(BI))
+        IfHandled++;
+    } else if (SwitchInst *SI = dyn_cast<SwitchInst>(BB->getTerminator())) {
+      if (HandleSwitchExpect(SI))
+        IfHandled++;
+    }
+
+    // remove llvm.expect intrinsics.
+    for (BasicBlock::iterator BI = BB->begin(), BE = BB->end();
+         BI != BE; ) {
+      CallInst *CI = dyn_cast<CallInst>(BI++);
+      if (!CI)
+        continue;
+
+      Function *Fn = CI->getCalledFunction();
+      if (Fn && Fn->getIntrinsicID() == Intrinsic::expect)
+        CI->eraseFromParent();
+    }
+  }
+
+  return false;
+}
+
+
+char LowerExpectIntrinsic::ID = 0;
+INITIALIZE_PASS(LowerExpectIntrinsic, "lower-expect", "Lower 'expect' "
+                "Intrinsics", false, false)
+
+FunctionPass *llvm::createLowerExpectIntrinsicPass() {
+  return new LowerExpectIntrinsic();
+}

lib/VMCore/LLVMContext.cpp

@@ -39,6 +39,10 @@ LLVMContext::LLVMContext() : pImpl(new LLVMContextImpl(*this)) {
   // Create the 'tbaa' metadata kind.
   unsigned TBAAID = getMDKindID("tbaa");
   assert(TBAAID == MD_tbaa && "tbaa kind id drifted"); (void)TBAAID;
+
+  // Create the 'prof' metadata kind.
+  unsigned ProfID = getMDKindID("prof");
+  assert(ProfID == MD_prof && "prof kind id drifted"); (void)ProfID;
 }
 LLVMContext::~LLVMContext() { delete pImpl; }
 

test/CodeGen/Generic/builtin-expect.ll

@@ -0,0 +1,223 @@
+; RUN: llc < %s
+
+define i32 @test1(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %cmp = icmp sgt i32 %tmp, 1
+  %conv = zext i1 %cmp to i32
+  %conv1 = sext i32 %conv to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv1, i64 1)
+  %tobool = icmp ne i64 %expval, 0
+  br i1 %tobool, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+declare i64 @llvm.expect.i64(i64, i64) nounwind readnone
+
+declare i32 @f(...)
+
+define i32 @test2(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %conv = sext i32 %tmp to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv, i64 1)
+  %tobool = icmp ne i64 %expval, 0
+  br i1 %tobool, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+define i32 @test3(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %tobool = icmp ne i32 %tmp, 0
+  %lnot = xor i1 %tobool, true
+  %lnot.ext = zext i1 %lnot to i32
+  %conv = sext i32 %lnot.ext to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv, i64 1)
+  %tobool1 = icmp ne i64 %expval, 0
+  br i1 %tobool1, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+define i32 @test4(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %tobool = icmp ne i32 %tmp, 0
+  %lnot = xor i1 %tobool, true
+  %lnot1 = xor i1 %lnot, true
+  %lnot.ext = zext i1 %lnot1 to i32
+  %conv = sext i32 %lnot.ext to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv, i64 1)
+  %tobool2 = icmp ne i64 %expval, 0
+  br i1 %tobool2, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+define i32 @test5(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %cmp = icmp slt i32 %tmp, 0
+  %conv = zext i1 %cmp to i32
+  %conv1 = sext i32 %conv to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv1, i64 0)
+  %tobool = icmp ne i64 %expval, 0
+  br i1 %tobool, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+define i32 @test6(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %conv = sext i32 %tmp to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv, i64 1)
+  switch i64 %expval, label %sw.epilog [
+    i64 1, label %sw.bb
+    i64 2, label %sw.bb
+  ]
+
+sw.bb:                                            ; preds = %entry, %entry
+  store i32 0, i32* %retval
+  br label %return
+
+sw.epilog:                                        ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %sw.epilog, %sw.bb
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+define i32 @test7(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %conv = sext i32 %tmp to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv, i64 1)
+  switch i64 %expval, label %sw.epilog [
+    i64 2, label %sw.bb
+    i64 3, label %sw.bb
+  ]
+
+sw.bb:                                            ; preds = %entry, %entry
+  %tmp1 = load i32* %x.addr, align 4
+  store i32 %tmp1, i32* %retval
+  br label %return
+
+sw.epilog:                                        ; preds = %entry
+  store i32 0, i32* %retval
+  br label %return
+
+return:                                           ; preds = %sw.epilog, %sw.bb
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+define i32 @test8(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %cmp = icmp sgt i32 %tmp, 1
+  %conv = zext i1 %cmp to i32
+  %expval = call i32 @llvm.expect.i32(i32 %conv, i32 1)
+  %tobool = icmp ne i32 %expval, 0
+  br i1 %tobool, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+declare i32 @llvm.expect.i32(i32, i32) nounwind readnone
+

test/Transforms/LowerExpectIntrinsic/basic.ll

@@ -0,0 +1,251 @@
+; RUN: opt -lower-expect -strip-dead-prototypes -S -o - < %s | FileCheck %s
+
+; CHECK: @test1
+define i32 @test1(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %cmp = icmp sgt i32 %tmp, 1
+  %conv = zext i1 %cmp to i32
+  %conv1 = sext i32 %conv to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv1, i64 1)
+  %tobool = icmp ne i64 %expval, 0
+; CHECK: !prof !0
+; CHECK-NOT: @llvm.expect
+  br i1 %tobool, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+declare i64 @llvm.expect.i64(i64, i64) nounwind readnone
+
+declare i32 @f(...)
+
+; CHECK: @test2
+define i32 @test2(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %conv = sext i32 %tmp to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv, i64 1)
+  %tobool = icmp ne i64 %expval, 0
+; CHECK: !prof !0
+; CHECK-NOT: @llvm.expect
+  br i1 %tobool, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+; CHECK: @test3
+define i32 @test3(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %tobool = icmp ne i32 %tmp, 0
+  %lnot = xor i1 %tobool, true
+  %lnot.ext = zext i1 %lnot to i32
+  %conv = sext i32 %lnot.ext to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv, i64 1)
+  %tobool1 = icmp ne i64 %expval, 0
+; CHECK: !prof !0
+; CHECK-NOT: @llvm.expect
+  br i1 %tobool1, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+; CHECK: @test4
+define i32 @test4(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %tobool = icmp ne i32 %tmp, 0
+  %lnot = xor i1 %tobool, true
+  %lnot1 = xor i1 %lnot, true
+  %lnot.ext = zext i1 %lnot1 to i32
+  %conv = sext i32 %lnot.ext to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv, i64 1)
+  %tobool2 = icmp ne i64 %expval, 0
+; CHECK: !prof !0
+; CHECK-NOT: @llvm.expect
+  br i1 %tobool2, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+; CHECK: @test5
+define i32 @test5(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %cmp = icmp slt i32 %tmp, 0
+  %conv = zext i1 %cmp to i32
+  %conv1 = sext i32 %conv to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv1, i64 0)
+  %tobool = icmp ne i64 %expval, 0
+; CHECK: !prof !1
+; CHECK-NOT: @llvm.expect
+  br i1 %tobool, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+; CHECK: @test6
+define i32 @test6(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %conv = sext i32 %tmp to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv, i64 1)
+; CHECK: !prof !2
+; CHECK-NOT: @llvm.expect
+  switch i64 %expval, label %sw.epilog [
+    i64 1, label %sw.bb
+    i64 2, label %sw.bb
+  ]
+
+sw.bb:                                            ; preds = %entry, %entry
+  store i32 0, i32* %retval
+  br label %return
+
+sw.epilog:                                        ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %sw.epilog, %sw.bb
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+; CHECK: @test7
+define i32 @test7(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %conv = sext i32 %tmp to i64
+  %expval = call i64 @llvm.expect.i64(i64 %conv, i64 1)
+; CHECK: !prof !3
+; CHECK-NOT: @llvm.expect
+  switch i64 %expval, label %sw.epilog [
+    i64 2, label %sw.bb
+    i64 3, label %sw.bb
+  ]
+
+sw.bb:                                            ; preds = %entry, %entry
+  %tmp1 = load i32* %x.addr, align 4
+  store i32 %tmp1, i32* %retval
+  br label %return
+
+sw.epilog:                                        ; preds = %entry
+  store i32 0, i32* %retval
+  br label %return
+
+return:                                           ; preds = %sw.epilog, %sw.bb
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+; CHECK: @test8
+define i32 @test8(i32 %x) nounwind uwtable ssp {
+entry:
+  %retval = alloca i32, align 4
+  %x.addr = alloca i32, align 4
+  store i32 %x, i32* %x.addr, align 4
+  %tmp = load i32* %x.addr, align 4
+  %cmp = icmp sgt i32 %tmp, 1
+  %conv = zext i1 %cmp to i32
+  %expval = call i32 @llvm.expect.i32(i32 %conv, i32 1)
+  %tobool = icmp ne i32 %expval, 0
+; CHECK: !prof !0
+; CHECK-NOT: @llvm.expect
+  br i1 %tobool, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %call = call i32 (...)* @f()
+  store i32 %call, i32* %retval
+  br label %return
+
+if.end:                                           ; preds = %entry
+  store i32 1, i32* %retval
+  br label %return
+
+return:                                           ; preds = %if.end, %if.then
+  %0 = load i32* %retval
+  ret i32 %0
+}
+
+declare i32 @llvm.expect.i32(i32, i32) nounwind readnone
+
+; CHECK: !0 = metadata !{metadata !"branch_weights", i32 64, i32 4}
+; CHECK: !1 = metadata !{metadata !"branch_weights", i32 4, i32 64}
+; CHECK: !2 = metadata !{metadata !"branch_weights", i32 4, i32 64, i32 4}
+; CHECK: !3 = metadata !{metadata !"branch_weights", i32 64, i32 4, i32 4}

test/Transforms/LowerExpectIntrinsic/dg.exp

@@ -0,0 +1,3 @@
+load_lib llvm.exp
+
+RunLLVMTests [lsort [glob -nocomplain $srcdir/$subdir/*.{ll,c,cpp}]]