//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "integer-division"
-#include "llvm/Function.h"
-#include "llvm/Instructions.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/IRBuilder.h"
#include "llvm/Transforms/Utils/IntegerDivision.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/Intrinsics.h"
using namespace llvm;
-// Generate code to divide two signed integers. Returns the quotient, rounded
-// towards 0. Builder's insert point should be pointing at the sdiv
-// instruction. This will generate a udiv in the process, and Builder's insert
-// point will be pointing at the udiv (if present, i.e. not folded), ready to be
-// expanded if the user wishes.
-static Value* GenerateSignedDivisionCode(Value* Dividend, Value* Divisor,
- IRBuilder<>& Builder) {
+/// Generate code to compute the remainder of two signed integers. Returns the
+/// remainder, which will have the sign of the dividend. Builder's insert point
+/// should be pointing where the caller wants code generated, e.g. at the srem
+/// instruction. This will generate a urem in the process, and Builder's insert
+/// point will be pointing at the uren (if present, i.e. not folded), ready to
+/// be expanded if the user wishes
+static Value *generateSignedRemainderCode(Value *Dividend, Value *Divisor,
+ IRBuilder<> &Builder) {
+ ConstantInt *ThirtyOne = Builder.getInt32(31);
+
+ // ; %dividend_sgn = ashr i32 %dividend, 31
+ // ; %divisor_sgn = ashr i32 %divisor, 31
+ // ; %dvd_xor = xor i32 %dividend, %dividend_sgn
+ // ; %dvs_xor = xor i32 %divisor, %divisor_sgn
+ // ; %u_dividend = sub i32 %dvd_xor, %dividend_sgn
+ // ; %u_divisor = sub i32 %dvs_xor, %divisor_sgn
+ // ; %urem = urem i32 %dividend, %divisor
+ // ; %xored = xor i32 %urem, %dividend_sgn
+ // ; %srem = sub i32 %xored, %dividend_sgn
+ Value *DividendSign = Builder.CreateAShr(Dividend, ThirtyOne);
+ Value *DivisorSign = Builder.CreateAShr(Divisor, ThirtyOne);
+ Value *DvdXor = Builder.CreateXor(Dividend, DividendSign);
+ Value *DvsXor = Builder.CreateXor(Divisor, DivisorSign);
+ Value *UDividend = Builder.CreateSub(DvdXor, DividendSign);
+ Value *UDivisor = Builder.CreateSub(DvsXor, DivisorSign);
+ Value *URem = Builder.CreateURem(UDividend, UDivisor);
+ Value *Xored = Builder.CreateXor(URem, DividendSign);
+ Value *SRem = Builder.CreateSub(Xored, DividendSign);
+
+ if (Instruction *URemInst = dyn_cast<Instruction>(URem))
+ Builder.SetInsertPoint(URemInst);
+
+ return SRem;
+}
+
+
+/// Generate code to compute the remainder of two unsigned integers. Returns the
+/// remainder. Builder's insert point should be pointing where the caller wants
+/// code generated, e.g. at the urem instruction. This will generate a udiv in
+/// the process, and Builder's insert point will be pointing at the udiv (if
+/// present, i.e. not folded), ready to be expanded if the user wishes
+static Value *generatedUnsignedRemainderCode(Value *Dividend, Value *Divisor,
+ IRBuilder<> &Builder) {
+ // Remainder = Dividend - Quotient*Divisor
+
+ // ; %quotient = udiv i32 %dividend, %divisor
+ // ; %product = mul i32 %divisor, %quotient
+ // ; %remainder = sub i32 %dividend, %product
+ Value *Quotient = Builder.CreateUDiv(Dividend, Divisor);
+ Value *Product = Builder.CreateMul(Divisor, Quotient);
+ Value *Remainder = Builder.CreateSub(Dividend, Product);
+
+ if (Instruction *UDiv = dyn_cast<Instruction>(Quotient))
+ Builder.SetInsertPoint(UDiv);
+
+ return Remainder;
+}
+
+/// Generate code to divide two signed integers. Returns the quotient, rounded
+/// towards 0. Builder's insert point should be pointing where the caller wants
+/// code generated, e.g. at the sdiv instruction. This will generate a udiv in
+/// the process, and Builder's insert point will be pointing at the udiv (if
+/// present, i.e. not folded), ready to be expanded if the user wishes.
+static Value *generateSignedDivisionCode(Value *Dividend, Value *Divisor,
+ IRBuilder<> &Builder) {
// Implementation taken from compiler-rt's __divsi3
- ConstantInt* ThirtyOne = Builder.getInt32(31);
+ ConstantInt *ThirtyOne = Builder.getInt32(31);
// ; %tmp = ashr i32 %dividend, 31
// ; %tmp1 = ashr i32 %divisor, 31
// ; %q_mag = udiv i32 %u_dvnd, %u_dvsr
// ; %tmp4 = xor i32 %q_mag, %q_sgn
// ; %q = sub i32 %tmp4, %q_sgn
- Value* Tmp = Builder.CreateAShr(Dividend, ThirtyOne);
- Value* Tmp1 = Builder.CreateAShr(Divisor, ThirtyOne);
- Value* Tmp2 = Builder.CreateXor(Tmp, Dividend);
- Value* U_Dvnd = Builder.CreateSub(Tmp2, Tmp);
- Value* Tmp3 = Builder.CreateXor(Tmp1, Divisor);
- Value* U_Dvsr = Builder.CreateSub(Tmp3, Tmp1);
- Value* Q_Sgn = Builder.CreateXor(Tmp1, Tmp);
- Value* Q_Mag = Builder.CreateUDiv(U_Dvnd, U_Dvsr);
- Value* Tmp4 = Builder.CreateXor(Q_Mag, Q_Sgn);
- Value* Q = Builder.CreateSub(Tmp4, Q_Sgn);
-
- if (Instruction* UDiv = dyn_cast<Instruction>(Q_Mag))
+ Value *Tmp = Builder.CreateAShr(Dividend, ThirtyOne);
+ Value *Tmp1 = Builder.CreateAShr(Divisor, ThirtyOne);
+ Value *Tmp2 = Builder.CreateXor(Tmp, Dividend);
+ Value *U_Dvnd = Builder.CreateSub(Tmp2, Tmp);
+ Value *Tmp3 = Builder.CreateXor(Tmp1, Divisor);
+ Value *U_Dvsr = Builder.CreateSub(Tmp3, Tmp1);
+ Value *Q_Sgn = Builder.CreateXor(Tmp1, Tmp);
+ Value *Q_Mag = Builder.CreateUDiv(U_Dvnd, U_Dvsr);
+ Value *Tmp4 = Builder.CreateXor(Q_Mag, Q_Sgn);
+ Value *Q = Builder.CreateSub(Tmp4, Q_Sgn);
+
+ if (Instruction *UDiv = dyn_cast<Instruction>(Q_Mag))
Builder.SetInsertPoint(UDiv);
return Q;
}
-// Generates code to divide two unsigned scalar 32-bit integers. Returns the
-// quotient, rounded towards 0. Builder's insert point should be pointing at the
-// udiv instruction.
-static Value* GenerateUnsignedDivisionCode(Value* Dividend, Value* Divisor,
- IRBuilder<>& Builder) {
+/// Generates code to divide two unsigned scalar 32-bit integers. Returns the
+/// quotient, rounded towards 0. Builder's insert point should be pointing where
+/// the caller wants code generated, e.g. at the udiv instruction.
+static Value *generateUnsignedDivisionCode(Value *Dividend, Value *Divisor,
+ IRBuilder<> &Builder) {
// The basic algorithm can be found in the compiler-rt project's
// implementation of __udivsi3.c. Here, we do a lower-level IR based approach
// that's been hand-tuned to lessen the amount of control flow involved.
// Some helper values
- IntegerType* I32Ty = Builder.getInt32Ty();
+ IntegerType *I32Ty = Builder.getInt32Ty();
- ConstantInt* Zero = Builder.getInt32(0);
- ConstantInt* One = Builder.getInt32(1);
- ConstantInt* ThirtyOne = Builder.getInt32(31);
- ConstantInt* NegOne = ConstantInt::getSigned(I32Ty, -1);
- ConstantInt* True = Builder.getTrue();
+ ConstantInt *Zero = Builder.getInt32(0);
+ ConstantInt *One = Builder.getInt32(1);
+ ConstantInt *ThirtyOne = Builder.getInt32(31);
+ ConstantInt *NegOne = ConstantInt::getSigned(I32Ty, -1);
+ ConstantInt *True = Builder.getTrue();
- BasicBlock* IBB = Builder.GetInsertBlock();
- Function* F = IBB->getParent();
- Function* CTLZi32 = Intrinsic::getDeclaration(F->getParent(), Intrinsic::ctlz,
+ BasicBlock *IBB = Builder.GetInsertBlock();
+ Function *F = IBB->getParent();
+ Function *CTLZi32 = Intrinsic::getDeclaration(F->getParent(), Intrinsic::ctlz,
I32Ty);
// Our CFG is going to look like:
// | ... |
// | end |
// +-------+
- BasicBlock* SpecialCases = Builder.GetInsertBlock();
+ BasicBlock *SpecialCases = Builder.GetInsertBlock();
SpecialCases->setName(Twine(SpecialCases->getName(), "_udiv-special-cases"));
- BasicBlock* End = SpecialCases->splitBasicBlock(Builder.GetInsertPoint(),
+ BasicBlock *End = SpecialCases->splitBasicBlock(Builder.GetInsertPoint(),
"udiv-end");
- BasicBlock* LoopExit = BasicBlock::Create(Builder.getContext(),
+ BasicBlock *LoopExit = BasicBlock::Create(Builder.getContext(),
"udiv-loop-exit", F, End);
- BasicBlock* DoWhile = BasicBlock::Create(Builder.getContext(),
+ BasicBlock *DoWhile = BasicBlock::Create(Builder.getContext(),
"udiv-do-while", F, End);
- BasicBlock* Preheader = BasicBlock::Create(Builder.getContext(),
+ BasicBlock *Preheader = BasicBlock::Create(Builder.getContext(),
"udiv-preheader", F, End);
- BasicBlock* BB1 = BasicBlock::Create(Builder.getContext(),
+ BasicBlock *BB1 = BasicBlock::Create(Builder.getContext(),
"udiv-bb1", F, End);
// We'll be overwriting the terminator to insert our extra blocks
// ; %earlyRet = or i1 %ret0, %retDividend
// ; br i1 %earlyRet, label %end, label %bb1
Builder.SetInsertPoint(SpecialCases);
- Value* Ret0_1 = Builder.CreateICmpEQ(Divisor, Zero);
- Value* Ret0_2 = Builder.CreateICmpEQ(Dividend, Zero);
- Value* Ret0_3 = Builder.CreateOr(Ret0_1, Ret0_2);
- Value* Tmp0 = Builder.CreateCall2(CTLZi32, Divisor, True);
- Value* Tmp1 = Builder.CreateCall2(CTLZi32, Dividend, True);
- Value* SR = Builder.CreateSub(Tmp0, Tmp1);
- Value* Ret0_4 = Builder.CreateICmpUGT(SR, ThirtyOne);
- Value* Ret0 = Builder.CreateOr(Ret0_3, Ret0_4);
- Value* RetDividend = Builder.CreateICmpEQ(SR, ThirtyOne);
- Value* RetVal = Builder.CreateSelect(Ret0, Zero, Dividend);
- Value* EarlyRet = Builder.CreateOr(Ret0, RetDividend);
+ Value *Ret0_1 = Builder.CreateICmpEQ(Divisor, Zero);
+ Value *Ret0_2 = Builder.CreateICmpEQ(Dividend, Zero);
+ Value *Ret0_3 = Builder.CreateOr(Ret0_1, Ret0_2);
+ Value *Tmp0 = Builder.CreateCall2(CTLZi32, Divisor, True);
+ Value *Tmp1 = Builder.CreateCall2(CTLZi32, Dividend, True);
+ Value *SR = Builder.CreateSub(Tmp0, Tmp1);
+ Value *Ret0_4 = Builder.CreateICmpUGT(SR, ThirtyOne);
+ Value *Ret0 = Builder.CreateOr(Ret0_3, Ret0_4);
+ Value *RetDividend = Builder.CreateICmpEQ(SR, ThirtyOne);
+ Value *RetVal = Builder.CreateSelect(Ret0, Zero, Dividend);
+ Value *EarlyRet = Builder.CreateOr(Ret0, RetDividend);
Builder.CreateCondBr(EarlyRet, End, BB1);
// ; bb1: ; preds = %special-cases
// ; %skipLoop = icmp eq i32 %sr_1, 0
// ; br i1 %skipLoop, label %loop-exit, label %preheader
Builder.SetInsertPoint(BB1);
- Value* SR_1 = Builder.CreateAdd(SR, One);
- Value* Tmp2 = Builder.CreateSub(ThirtyOne, SR);
- Value* Q = Builder.CreateShl(Dividend, Tmp2);
- Value* SkipLoop = Builder.CreateICmpEQ(SR_1, Zero);
+ Value *SR_1 = Builder.CreateAdd(SR, One);
+ Value *Tmp2 = Builder.CreateSub(ThirtyOne, SR);
+ Value *Q = Builder.CreateShl(Dividend, Tmp2);
+ Value *SkipLoop = Builder.CreateICmpEQ(SR_1, Zero);
Builder.CreateCondBr(SkipLoop, LoopExit, Preheader);
// ; preheader: ; preds = %bb1
// ; %tmp4 = add i32 %divisor, -1
// ; br label %do-while
Builder.SetInsertPoint(Preheader);
- Value* Tmp3 = Builder.CreateLShr(Dividend, SR_1);
- Value* Tmp4 = Builder.CreateAdd(Divisor, NegOne);
+ Value *Tmp3 = Builder.CreateLShr(Dividend, SR_1);
+ Value *Tmp4 = Builder.CreateAdd(Divisor, NegOne);
Builder.CreateBr(DoWhile);
// ; do-while: ; preds = %do-while, %preheader
// ; %tmp12 = icmp eq i32 %sr_2, 0
// ; br i1 %tmp12, label %loop-exit, label %do-while
Builder.SetInsertPoint(DoWhile);
- PHINode* Carry_1 = Builder.CreatePHI(I32Ty, 2);
- PHINode* SR_3 = Builder.CreatePHI(I32Ty, 2);
- PHINode* R_1 = Builder.CreatePHI(I32Ty, 2);
- PHINode* Q_2 = Builder.CreatePHI(I32Ty, 2);
- Value* Tmp5 = Builder.CreateShl(R_1, One);
- Value* Tmp6 = Builder.CreateLShr(Q_2, ThirtyOne);
- Value* Tmp7 = Builder.CreateOr(Tmp5, Tmp6);
- Value* Tmp8 = Builder.CreateShl(Q_2, One);
- Value* Q_1 = Builder.CreateOr(Carry_1, Tmp8);
- Value* Tmp9 = Builder.CreateSub(Tmp4, Tmp7);
- Value* Tmp10 = Builder.CreateAShr(Tmp9, 31);
- Value* Carry = Builder.CreateAnd(Tmp10, One);
- Value* Tmp11 = Builder.CreateAnd(Tmp10, Divisor);
- Value* R = Builder.CreateSub(Tmp7, Tmp11);
- Value* SR_2 = Builder.CreateAdd(SR_3, NegOne);
- Value* Tmp12 = Builder.CreateICmpEQ(SR_2, Zero);
+ PHINode *Carry_1 = Builder.CreatePHI(I32Ty, 2);
+ PHINode *SR_3 = Builder.CreatePHI(I32Ty, 2);
+ PHINode *R_1 = Builder.CreatePHI(I32Ty, 2);
+ PHINode *Q_2 = Builder.CreatePHI(I32Ty, 2);
+ Value *Tmp5 = Builder.CreateShl(R_1, One);
+ Value *Tmp6 = Builder.CreateLShr(Q_2, ThirtyOne);
+ Value *Tmp7 = Builder.CreateOr(Tmp5, Tmp6);
+ Value *Tmp8 = Builder.CreateShl(Q_2, One);
+ Value *Q_1 = Builder.CreateOr(Carry_1, Tmp8);
+ Value *Tmp9 = Builder.CreateSub(Tmp4, Tmp7);
+ Value *Tmp10 = Builder.CreateAShr(Tmp9, 31);
+ Value *Carry = Builder.CreateAnd(Tmp10, One);
+ Value *Tmp11 = Builder.CreateAnd(Tmp10, Divisor);
+ Value *R = Builder.CreateSub(Tmp7, Tmp11);
+ Value *SR_2 = Builder.CreateAdd(SR_3, NegOne);
+ Value *Tmp12 = Builder.CreateICmpEQ(SR_2, Zero);
Builder.CreateCondBr(Tmp12, LoopExit, DoWhile);
// ; loop-exit: ; preds = %do-while, %bb1
// ; %q_4 = or i32 %carry_2, %tmp13
// ; br label %end
Builder.SetInsertPoint(LoopExit);
- PHINode* Carry_2 = Builder.CreatePHI(I32Ty, 2);
- PHINode* Q_3 = Builder.CreatePHI(I32Ty, 2);
- Value* Tmp13 = Builder.CreateShl(Q_3, One);
- Value* Q_4 = Builder.CreateOr(Carry_2, Tmp13);
+ PHINode *Carry_2 = Builder.CreatePHI(I32Ty, 2);
+ PHINode *Q_3 = Builder.CreatePHI(I32Ty, 2);
+ Value *Tmp13 = Builder.CreateShl(Q_3, One);
+ Value *Q_4 = Builder.CreateOr(Carry_2, Tmp13);
Builder.CreateBr(End);
// ; end: ; preds = %loop-exit, %special-cases
// ; %q_5 = phi i32 [ %q_4, %loop-exit ], [ %retVal, %special-cases ]
// ; ret i32 %q_5
Builder.SetInsertPoint(End, End->begin());
- PHINode* Q_5 = Builder.CreatePHI(I32Ty, 2);
+ PHINode *Q_5 = Builder.CreatePHI(I32Ty, 2);
// Populate the Phis, since all values have now been created. Our Phis were:
// ; %carry_1 = phi i32 [ 0, %preheader ], [ %carry, %do-while ]
return Q_5;
}
-bool llvm::expandDivision(BinaryOperator* Div) {
- assert(Div->getOpcode() == Instruction::SDiv ||
- Div->getOpcode() == Instruction::UDiv
- && "Trying to expand division from a non-division function");
+/// Generate code to calculate the remainder of two integers, replacing Rem with
+/// the generated code. This currently generates code using the udiv expansion,
+/// but future work includes generating more specialized code, e.g. when more
+/// information about the operands are known. Currently only implements 32bit
+/// scalar division (due to udiv's limitation), but future work is removing this
+/// limitation.
+///
+/// @brief Replace Rem with generated code.
+bool llvm::expandRemainder(BinaryOperator *Rem) {
+ assert((Rem->getOpcode() == Instruction::SRem ||
+ Rem->getOpcode() == Instruction::URem) &&
+ "Trying to expand remainder from a non-remainder function");
- IRBuilder<> Builder(Div);
+ IRBuilder<> Builder(Rem);
+
+ // First prepare the sign if it's a signed remainder
+ if (Rem->getOpcode() == Instruction::SRem) {
+ Value *Remainder = generateSignedRemainderCode(Rem->getOperand(0),
+ Rem->getOperand(1), Builder);
- if (Div->getType()->isVectorTy()) {
- assert(0 && "Div over vectors not supported");
- return false;
+ Rem->replaceAllUsesWith(Remainder);
+ Rem->dropAllReferences();
+ Rem->eraseFromParent();
+
+ // If we didn't actually generate a udiv instruction, we're done
+ BinaryOperator *BO = dyn_cast<BinaryOperator>(Builder.GetInsertPoint());
+ if (!BO || BO->getOpcode() != Instruction::URem)
+ return true;
+
+ Rem = BO;
+ }
+
+ Value *Remainder = generatedUnsignedRemainderCode(Rem->getOperand(0),
+ Rem->getOperand(1),
+ Builder);
+
+ Rem->replaceAllUsesWith(Remainder);
+ Rem->dropAllReferences();
+ Rem->eraseFromParent();
+
+ // Expand the udiv
+ if (BinaryOperator *UDiv = dyn_cast<BinaryOperator>(Builder.GetInsertPoint())) {
+ assert(UDiv->getOpcode() == Instruction::UDiv && "Non-udiv in expansion?");
+ expandDivision(UDiv);
}
+ return true;
+}
+
+
+/// Generate code to divide two integers, replacing Div with the generated
+/// code. This currently generates code similarly to compiler-rt's
+/// implementations, but future work includes generating more specialized code
+/// when more information about the operands are known. Currently only
+/// implements 32bit scalar division, but future work is removing this
+/// limitation.
+///
+/// @brief Replace Div with generated code.
+bool llvm::expandDivision(BinaryOperator *Div) {
+ assert((Div->getOpcode() == Instruction::SDiv ||
+ Div->getOpcode() == Instruction::UDiv) &&
+ "Trying to expand division from a non-division function");
+
+ IRBuilder<> Builder(Div);
+
+ if (Div->getType()->isVectorTy())
+ llvm_unreachable("Div over vectors not supported");
+
// First prepare the sign if it's a signed division
if (Div->getOpcode() == Instruction::SDiv) {
// Lower the code to unsigned division, and reset Div to point to the udiv.
- Value* Quotient = GenerateSignedDivisionCode(Div->getOperand(0),
- Div->getOperand(1), Builder);
+ Value *Quotient = generateSignedDivisionCode(Div->getOperand(0),
+ Div->getOperand(1), Builder);
Div->replaceAllUsesWith(Quotient);
Div->dropAllReferences();
Div->eraseFromParent();
// If we didn't actually generate a udiv instruction, we're done
- BinaryOperator* BO = dyn_cast<BinaryOperator>(Builder.GetInsertPoint());
+ BinaryOperator *BO = dyn_cast<BinaryOperator>(Builder.GetInsertPoint());
if (!BO || BO->getOpcode() != Instruction::UDiv)
return true;
}
// Insert the unsigned division code
- Value* Quotient = GenerateUnsignedDivisionCode(Div->getOperand(0),
+ Value *Quotient = generateUnsignedDivisionCode(Div->getOperand(0),
Div->getOperand(1),
Builder);
Div->replaceAllUsesWith(Quotient);
return true;
}
+
+/// Generate code to compute the remainder of two integers of bitwidth up to
+/// 32 bits. Uses the above routines and extends the inputs/truncates the
+/// outputs to operate in 32 bits; that is, these routines are good for targets
+/// that have no or very little suppport for smaller than 32 bit integer
+/// arithmetic.
+///
+/// @brief Replace Rem with emulation code.
+bool llvm::expandRemainderUpTo32Bits(BinaryOperator *Rem) {
+ assert((Rem->getOpcode() == Instruction::SRem ||
+ Rem->getOpcode() == Instruction::URem) &&
+ "Trying to expand remainder from a non-remainder function");
+
+ Type *RemTy = Rem->getType();
+ if (RemTy->isVectorTy())
+ llvm_unreachable("Div over vectors not supported");
+
+ unsigned RemTyBitWidth = RemTy->getIntegerBitWidth();
+
+ if (RemTyBitWidth > 32)
+ llvm_unreachable("Div of bitwidth greater than 32 not supported");
+
+ if (RemTyBitWidth == 32)
+ return expandRemainder(Rem);
+
+ // If bitwidth smaller than 32 extend inputs, truncate output and proceed
+ // with 32 bit division.
+ IRBuilder<> Builder(Rem);
+
+ Value *ExtDividend;
+ Value *ExtDivisor;
+ Value *ExtRem;
+ Value *Trunc;
+ Type *Int32Ty = Builder.getInt32Ty();
+
+ if (Rem->getOpcode() == Instruction::SRem) {
+ ExtDividend = Builder.CreateSExt(Rem->getOperand(0), Int32Ty);
+ ExtDivisor = Builder.CreateSExt(Rem->getOperand(1), Int32Ty);
+ ExtRem = Builder.CreateSRem(ExtDividend, ExtDivisor);
+ } else {
+ ExtDividend = Builder.CreateZExt(Rem->getOperand(0), Int32Ty);
+ ExtDivisor = Builder.CreateZExt(Rem->getOperand(1), Int32Ty);
+ ExtRem = Builder.CreateURem(ExtDividend, ExtDivisor);
+ }
+ Trunc = Builder.CreateTrunc(ExtRem, RemTy);
+
+ Rem->replaceAllUsesWith(Trunc);
+ Rem->dropAllReferences();
+ Rem->eraseFromParent();
+
+ return expandRemainder(cast<BinaryOperator>(ExtRem));
+}
+
+
+/// Generate code to divide two integers of bitwidth up to 32 bits. Uses the
+/// above routines and extends the inputs/truncates the outputs to operate
+/// in 32 bits; that is, these routines are good for targets that have no
+/// or very little support for smaller than 32 bit integer arithmetic.
+///
+/// @brief Replace Div with emulation code.
+bool llvm::expandDivisionUpTo32Bits(BinaryOperator *Div) {
+ assert((Div->getOpcode() == Instruction::SDiv ||
+ Div->getOpcode() == Instruction::UDiv) &&
+ "Trying to expand division from a non-division function");
+
+ Type *DivTy = Div->getType();
+ if (DivTy->isVectorTy())
+ llvm_unreachable("Div over vectors not supported");
+
+ unsigned DivTyBitWidth = DivTy->getIntegerBitWidth();
+
+ if (DivTyBitWidth > 32)
+ llvm_unreachable("Div of bitwidth greater than 32 not supported");
+
+ if (DivTyBitWidth == 32)
+ return expandDivision(Div);
+
+ // If bitwidth smaller than 32 extend inputs, truncate output and proceed
+ // with 32 bit division.
+ IRBuilder<> Builder(Div);
+
+ Value *ExtDividend;
+ Value *ExtDivisor;
+ Value *ExtDiv;
+ Value *Trunc;
+ Type *Int32Ty = Builder.getInt32Ty();
+
+ if (Div->getOpcode() == Instruction::SDiv) {
+ ExtDividend = Builder.CreateSExt(Div->getOperand(0), Int32Ty);
+ ExtDivisor = Builder.CreateSExt(Div->getOperand(1), Int32Ty);
+ ExtDiv = Builder.CreateSDiv(ExtDividend, ExtDivisor);
+ } else {
+ ExtDividend = Builder.CreateZExt(Div->getOperand(0), Int32Ty);
+ ExtDivisor = Builder.CreateZExt(Div->getOperand(1), Int32Ty);
+ ExtDiv = Builder.CreateUDiv(ExtDividend, ExtDivisor);
+ }
+ Trunc = Builder.CreateTrunc(ExtDiv, DivTy);
+
+ Div->replaceAllUsesWith(Trunc);
+ Div->dropAllReferences();
+ Div->eraseFromParent();
+
+ return expandDivision(cast<BinaryOperator>(ExtDiv));
+}