Generic Bypass Slow Div

- CodeGenPrepare pass for identifying div/rem ops
- Backend specifies the type mapping using addBypassSlowDivType
- Enabled only for Intel Atom with O2 32-bit -> 8-bit
- Replace IDIV with instructions which test its value and use DIVB if the value
is positive and less than 256.
- In the case when the quotient and remainder of a divide are used a DIV
and a REM instruction will be present in the IR. In the non-Atom case
they are both lowered to IDIVs and CSE removes the redundant IDIV instruction,
using the quotient and remainder from the first IDIV. However,
due to this optimization CSE is not able to eliminate redundant
IDIV instructions because they are located in different basic blocks.
This is overcome by calculating both the quotient (DIV) and remainder (REM)
in each basic block that is inserted by the optimization and reusing the result
values when a subsequent DIV or REM instruction uses the same operands.
- Test cases check for the presents of the optimization when calculating
either the quotient, remainder,  or both.

Patch by Tyler Nowicki!

llvm-svn: 163150

8 files changed, 281 insertions, 3 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 6820175c1be..10a534f0ff3 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -898,7 +898,6 @@ const char *TargetLowering::getTargetNodeName(unsigned Opcode) const {
   return NULL;
 }
 
-
 EVT TargetLowering::getSetCCResultType(EVT VT) const {
   assert(!VT.isVector() && "No default SetCC type for vectors!");
   return PointerTy.SimpleTy;
diff --git a/llvm/lib/Target/X86/X86.td b/llvm/lib/Target/X86/X86.td
index 18e6b7c3d9b..d078a7b5dfb 100644
--- a/llvm/lib/Target/X86/X86.td
+++ b/llvm/lib/Target/X86/X86.td
@@ -120,6 +120,9 @@ def FeatureBMI2    : SubtargetFeature<"bmi2", "HasBMI2", "true",
                                       "Support BMI2 instructions">;
 def FeatureLeaForSP : SubtargetFeature<"lea-sp", "UseLeaForSP", "true",
                                      "Use LEA for adjusting the stack pointer">;
+def FeatureSlowDivide : SubtargetFeature<"idiv-to-divb",
+                                     "HasSlowDivide", "true",
+                                     "Use small divide for positive values less than 256">;
 
 //===----------------------------------------------------------------------===//
 // X86 processors supported.
@@ -160,7 +163,8 @@ def : Proc<"core2",           [FeatureSSSE3, FeatureCMPXCHG16B,
 def : Proc<"penryn",          [FeatureSSE41, FeatureCMPXCHG16B,
                                FeatureSlowBTMem]>;
 def : AtomProc<"atom",        [ProcIntelAtom, FeatureSSE3, FeatureCMPXCHG16B,
-                               FeatureMOVBE, FeatureSlowBTMem, FeatureLeaForSP]>;
+                               FeatureMOVBE, FeatureSlowBTMem, FeatureLeaForSP,
+                               FeatureSlowDivide]>;
 // "Arrandale" along with corei3 and corei5
 def : Proc<"corei7",          [FeatureSSE42, FeatureCMPXCHG16B,
                                FeatureSlowBTMem, FeatureFastUAMem,
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 9f487a6b0f2..67ad99d3ce7 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -182,6 +182,10 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setSchedulingPreference(Sched::RegPressure);
   setStackPointerRegisterToSaveRestore(X86StackPtr);
 
+  // Bypass i32 with i8 on Atom when compiling with O2
+  if (Subtarget->hasSlowDivide() && TM.getOptLevel() >= CodeGenOpt::Default)
+    addBypassSlowDivType(Type::getInt32Ty(getGlobalContext()), Type::getInt8Ty(getGlobalContext()));
+
   if (Subtarget->isTargetWindows() && !Subtarget->isTargetCygMing()) {
     // Setup Windows compiler runtime calls.
     setLibcallName(RTLIB::SDIV_I64, "_alldiv");
diff --git a/llvm/lib/Target/X86/X86Subtarget.cpp b/llvm/lib/Target/X86/X86Subtarget.cpp
index 908785296d7..0d7b664e8e0 100644
--- a/llvm/lib/Target/X86/X86Subtarget.cpp
+++ b/llvm/lib/Target/X86/X86Subtarget.cpp
@@ -346,6 +346,7 @@ X86Subtarget::X86Subtarget(const std::string &TT, const std::string &CPU,
   , HasVectorUAMem(false)
   , HasCmpxchg16b(false)
   , UseLeaForSP(false)
+  , HasSlowDivide(false)
   , PostRAScheduler(false)
   , stackAlignment(4)
   // FIXME: this is a known good value for Yonah. How about others?
diff --git a/llvm/lib/Target/X86/X86Subtarget.h b/llvm/lib/Target/X86/X86Subtarget.h
index 33608bb32a6..dde7e240b9d 100644
--- a/llvm/lib/Target/X86/X86Subtarget.h
+++ b/llvm/lib/Target/X86/X86Subtarget.h
@@ -136,6 +136,10 @@ protected:
   /// the stack pointer. This is an optimization for Intel Atom processors.
   bool UseLeaForSP;
 
+  /// HasSlowDivide - True if smaller divides are significantly faster than
+  /// full divides and should be used when possible.
+  bool HasSlowDivide;
+
   /// PostRAScheduler - True if using post-register-allocation scheduler.
   bool PostRAScheduler;
 
@@ -221,6 +225,7 @@ public:
   bool hasVectorUAMem() const { return HasVectorUAMem; }
   bool hasCmpxchg16b() const { return HasCmpxchg16b; }
   bool useLeaForSP() const { return UseLeaForSP; }
+  bool hasSlowDivide() const { return HasSlowDivide; }
 
   bool isAtom() const { return X86ProcFamily == IntelAtom; }
 
diff --git a/llvm/lib/Transforms/Scalar/CodeGenPrepare.cpp b/llvm/lib/Transforms/Scalar/CodeGenPrepare.cpp
index 57a648f7b97..59121078cb1 100644
--- a/llvm/lib/Transforms/Scalar/CodeGenPrepare.cpp
+++ b/llvm/lib/Transforms/Scalar/CodeGenPrepare.cpp
@@ -43,6 +43,7 @@
 #include "llvm/Transforms/Utils/AddrModeMatcher.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/BuildLibCalls.h"
+#include "llvm/Transforms/Utils/BypassSlowDivision.h"
 #include "llvm/Transforms/Utils/Local.h"
 using namespace llvm;
 using namespace llvm::PatternMatch;
@@ -148,7 +149,19 @@ bool CodeGenPrepare::runOnFunction(Function &F) {
   PFI = getAnalysisIfAvailable<ProfileInfo>();
   OptSize = F.hasFnAttr(Attribute::OptimizeForSize);
 
-  // First pass, eliminate blocks that contain only PHI nodes and an
+  /// This optimization identifies DIV instructions that can be
+  /// profitably bypassed and carried out with a shorter, faster divide.
+  if (TLI && TLI->isSlowDivBypassed()) {
+    const DenseMap<Type *, Type *> &BypassTypeMap = TLI->getBypassSlowDivTypes();
+
+    for (Function::iterator I = F.begin(); I != F.end(); I++) {
+      EverMadeChange |= bypassSlowDivision(F,
+                                           I,
+                                           BypassTypeMap);
+    }
+  }
+
+  // Eliminate blocks that contain only PHI nodes and an
   // unconditional branch.
   EverMadeChange |= EliminateMostlyEmptyBlocks(F);
 
diff --git a/llvm/lib/Transforms/Utils/BypassSlowDivision.cpp b/llvm/lib/Transforms/Utils/BypassSlowDivision.cpp
new file mode 100644
index 00000000000..1c58bec925d
--- /dev/null
+++ b/llvm/lib/Transforms/Utils/BypassSlowDivision.cpp
@@ -0,0 +1,251 @@
+//===-- BypassSlowDivision.cpp - Bypass slow division ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains an optimization for div and rem on architectures that
+// execute short instructions significantly faster than longer instructions.
+// For example, on Intel Atom 32-bit divides are slow enough that during
+// runtime it is profitable to check the value of the operands, and if they are
+// positive and less than 256 use an unsigned 8-bit divide.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "bypass-slow-division"
+#include "llvm/Instructions.h"
+#include "llvm/Function.h"
+#include "llvm/IRBuilder.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/Transforms/Utils/BypassSlowDivision.h"
+
+using namespace llvm;
+
+namespace llvm {
+  struct DivOpInfo {
+    bool SignedOp;
+    Value *Dividend;
+    Value *Divisor;
+
+    DivOpInfo(bool InSignedOp, Value *InDividend, Value *InDivisor)
+      : SignedOp(InSignedOp), Dividend(InDividend), Divisor(InDivisor) {}
+  };
+
+  struct DivPhiNodes {
+    PHINode *Quotient;
+    PHINode *Remainder;
+
+    DivPhiNodes(PHINode *InQuotient, PHINode *InRemainder)
+      : Quotient(InQuotient), Remainder(InRemainder) {}
+  };
+
+  template<>
+  struct DenseMapInfo<DivOpInfo> {
+    static bool isEqual(const DivOpInfo &Val1, const DivOpInfo &Val2) {
+      return Val1.SignedOp == Val2.SignedOp &&
+             Val1.Dividend == Val2.Dividend &&
+             Val1.Divisor == Val2.Divisor;
+    }
+
+    static DivOpInfo getEmptyKey() {
+      return DivOpInfo(false, 0, 0);
+    }
+
+    static DivOpInfo getTombstoneKey() {
+      return DivOpInfo(true, 0, 0);
+    }
+
+    static unsigned getHashValue(const DivOpInfo &Val) {
+      return (unsigned)(reinterpret_cast<uintptr_t>(Val.Dividend) ^
+                        reinterpret_cast<uintptr_t>(Val.Divisor)) ^
+                       (unsigned)Val.SignedOp;
+    }
+  };
+
+  typedef DenseMap<DivOpInfo, DivPhiNodes> DivCacheTy;
+}
+
+// insertFastDiv - Substitutes the div/rem instruction with code that checks the
+// value of the operands and uses a shorter-faster div/rem instruction when
+// possible and the longer-slower div/rem instruction otherwise.
+static void insertFastDiv(Function &F,
+                          Function::iterator &I,
+                          BasicBlock::iterator &J,
+                          IntegerType *BypassType,
+                          bool UseDivOp,
+                          bool UseSignedOp,
+                          DivCacheTy &PerBBDivCache)
+{
+  // Get instruction operands
+  Instruction *Instr = J;
+  Value *Dividend = Instr->getOperand(0);
+  Value *Divisor = Instr->getOperand(1);
+
+  if (dyn_cast<ConstantInt>(Divisor) != 0 ||
+      (dyn_cast<ConstantInt>(Dividend) != 0 &&
+       dyn_cast<ConstantInt>(Divisor) != 0)) {
+    // Operations with immediate values should have
+    // been solved and replaced during compile time.
+    return;
+  }
+
+  // Basic Block is split before divide
+  BasicBlock *MainBB = I;
+  BasicBlock *SuccessorBB = I->splitBasicBlock(J);
+  I++; //advance iterator I to successorBB
+
+  // Add new basic block for slow divide operation
+  BasicBlock *SlowBB = BasicBlock::Create(F.getContext(), "",
+                                          MainBB->getParent(), SuccessorBB);
+  SlowBB->moveBefore(SuccessorBB);
+  IRBuilder<> SlowBuilder(SlowBB, SlowBB->begin());
+  Value *SlowQuotientV;
+  Value *SlowRemainderV;
+  if (UseSignedOp) {
+    SlowQuotientV = SlowBuilder.CreateSDiv(Dividend, Divisor);
+    SlowRemainderV = SlowBuilder.CreateSRem(Dividend, Divisor);
+  } else {
+    SlowQuotientV = SlowBuilder.CreateUDiv(Dividend, Divisor);
+    SlowRemainderV = SlowBuilder.CreateURem(Dividend, Divisor);
+  }
+  SlowBuilder.CreateBr(SuccessorBB);
+
+  // Add new basic block for fast divide operation
+  BasicBlock *FastBB = BasicBlock::Create(F.getContext(), "",
+                                          MainBB->getParent(), SuccessorBB);
+  FastBB->moveBefore(SlowBB);
+  IRBuilder<> FastBuilder(FastBB, FastBB->begin());
+  Value *ShortDivisorV = FastBuilder.CreateCast(Instruction::Trunc, Divisor, BypassType);
+  Value *ShortDividendV = FastBuilder.CreateCast(Instruction::Trunc, Dividend, BypassType);
+
+  // udiv/urem because optimization only handles positive numbers
+  Value *ShortQuotientV = FastBuilder.CreateExactUDiv(ShortDividendV,
+                                                       ShortDivisorV);
+  Value *ShortRemainderV = FastBuilder.CreateURem(ShortDividendV,
+                                                  ShortDivisorV);
+  Value *FastQuotientV = FastBuilder.CreateCast(Instruction::ZExt,
+                                                 ShortQuotientV,
+                                                 Dividend->getType());
+  Value *FastRemainderV = FastBuilder.CreateCast(Instruction::ZExt,
+                                                 ShortRemainderV,
+                                                 Dividend->getType());
+  FastBuilder.CreateBr(SuccessorBB);
+
+  // Phi nodes for result of div and rem
+  IRBuilder<> SuccessorBuilder(SuccessorBB, SuccessorBB->begin());
+  PHINode *QuoPhi = SuccessorBuilder.CreatePHI(Instr->getType(), 2);
+  QuoPhi->addIncoming(SlowQuotientV, SlowBB);
+  QuoPhi->addIncoming(FastQuotientV, FastBB);
+  PHINode *RemPhi = SuccessorBuilder.CreatePHI(Instr->getType(), 2);
+  RemPhi->addIncoming(SlowRemainderV, SlowBB);
+  RemPhi->addIncoming(FastRemainderV, FastBB);
+
+  // Replace Instr with appropriate phi node
+  if (UseDivOp) {
+    Instr->replaceAllUsesWith(QuoPhi);
+  } else {
+    Instr->replaceAllUsesWith(RemPhi);
+  }
+  Instr->eraseFromParent();
+
+  // Combine operands into a single value with OR for value testing below
+  MainBB->getInstList().back().eraseFromParent();
+  IRBuilder<> MainBuilder(MainBB, MainBB->end());
+  Value *OrV = MainBuilder.CreateOr(Dividend, Divisor);
+
+  // BitMask is inverted to check if the operands are
+  // larger than the bypass type
+  uint64_t BitMask = ~BypassType->getBitMask();
+  Value *AndV = MainBuilder.CreateAnd(OrV, BitMask);
+
+  // Compare operand values and branch
+  Value *ZeroV = MainBuilder.getInt32(0);
+  Value *CmpV = MainBuilder.CreateICmpEQ(AndV, ZeroV);
+  MainBuilder.CreateCondBr(CmpV, FastBB, SlowBB);
+
+  // point iterator J at first instruction of successorBB
+  J = I->begin();
+
+  // Cache phi nodes to be used later in place of other instances
+  // of div or rem with the same sign, dividend, and divisor
+  DivOpInfo Key(UseSignedOp, Dividend, Divisor);
+  DivPhiNodes Value(QuoPhi, RemPhi);
+  PerBBDivCache.insert(std::pair<DivOpInfo, DivPhiNodes>(Key, Value));
+}
+
+// reuseOrInsertFastDiv - Reuses previously computed dividend or remainder if
+// operands and operation are identical. Otherwise call insertFastDiv to perform
+// the optimization and cache the resulting dividend and remainder.
+static void reuseOrInsertFastDiv(Function &F,
+                                 Function::iterator &I,
+                                 BasicBlock::iterator &J,
+                                 IntegerType *BypassType,
+                                 bool UseDivOp,
+                                 bool UseSignedOp,
+                                 DivCacheTy &PerBBDivCache)
+{
+  // Get instruction operands
+  Instruction *Instr = J;
+  DivOpInfo Key(UseSignedOp, Instr->getOperand(0), Instr->getOperand(1));
+  DivCacheTy::const_iterator CacheI = PerBBDivCache.find(Key);
+
+  if (CacheI == PerBBDivCache.end()) {
+    // If previous instance does not exist, insert fast div
+    insertFastDiv(F, I, J, BypassType, UseDivOp, UseSignedOp, PerBBDivCache);
+    return;
+  }
+
+  // Replace operation value with previously generated phi node
+  DivPhiNodes Value = CacheI->second;
+  if (UseDivOp) {
+    // Replace all uses of div instruction with quotient phi node
+    J->replaceAllUsesWith(Value.Quotient);
+  } else {
+    // Replace all uses of rem instruction with remainder phi node
+    J->replaceAllUsesWith(Value.Remainder);
+  }
+
+  // Advance to next operation
+  J++;
+
+  // Remove redundant operation
+  Instr->eraseFromParent();
+}
+
+// bypassSlowDivision - This optimization identifies DIV instructions that can
+// be profitably bypassed and carried out with a shorter, faster divide.
+bool bypassSlowDivision(Function &F,
+                        Function::iterator &I,
+                        const llvm::DenseMap<Type *, Type *> &BypassTypeMap)
+{
+  DivCacheTy DivCache;
+
+  bool MadeChange = false;
+  for (BasicBlock::iterator J = I->begin(); J != I->end(); J++) {
+
+    // Get instruction details
+    unsigned Opcode = J->getOpcode();
+    bool UseDivOp = Opcode == Instruction::SDiv || Opcode == Instruction::UDiv;
+    bool UseRemOp = Opcode == Instruction::SRem || Opcode == Instruction::URem;
+    bool UseSignedOp = Opcode == Instruction::SDiv || Opcode == Instruction::SRem;
+
+    // Only optimize div or rem ops
+    if (!UseDivOp && !UseRemOp) {
+      continue;
+    }
+    // Continue if div/rem type is not bypassed
+    DenseMap<Type *, Type *>::const_iterator BT = BypassTypeMap.find(J->getType());
+    if (BT == BypassTypeMap.end()) {
+      continue;
+    }
+
+    IntegerType *BypassType = (IntegerType *)BT->second;
+    reuseOrInsertFastDiv(F, I, J, BypassType, UseDivOp, UseSignedOp, DivCache);
+    MadeChange = true;
+  }
+
+  return MadeChange;
+}
diff --git a/llvm/lib/Transforms/Utils/CMakeLists.txt b/llvm/lib/Transforms/Utils/CMakeLists.txt
index 4ff31cae62c..215a16ff3f8 100644
--- a/llvm/lib/Transforms/Utils/CMakeLists.txt
+++ b/llvm/lib/Transforms/Utils/CMakeLists.txt
@@ -3,6 +3,7 @@ add_llvm_library(LLVMTransformUtils
   BasicBlockUtils.cpp
   BreakCriticalEdges.cpp
   BuildLibCalls.cpp
+  BypassSlowDivision.cpp
   CloneFunction.cpp
   CloneModule.cpp
   CmpInstAnalysis.cpp