#include "llvm/Module.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/ADT/SmallVector.h"
+#include "llvm/Support/CallSite.h"
#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/IRBuilder.h"
#include <cstring>
using namespace llvm;
NewFn = F;
return true;
}
+ } else if (Name.compare(5, 9, "arm.neon.", 9) == 0) {
+ if (((Name.compare(14, 5, "vmovl", 5) == 0 ||
+ Name.compare(14, 5, "vaddl", 5) == 0 ||
+ Name.compare(14, 5, "vsubl", 5) == 0) &&
+ (Name.compare(19, 2, "s.", 2) == 0 ||
+ Name.compare(19, 2, "u.", 2) == 0)) ||
+
+ ((Name.compare(14, 5, "vaddw", 5) == 0 ||
+ Name.compare(14, 5, "vsubw", 5) == 0) &&
+ (Name.compare(19, 2, "s.", 2) == 0 ||
+ Name.compare(19, 2, "u.", 2) == 0)) ||
+
+ (Name.compare(14, 6, "vmovn.", 6) == 0)) {
+
+ // Calls to these are transformed into IR without intrinsics.
+ NewFn = 0;
+ return true;
+ }
+ // Old versions of NEON ld/st intrinsics are missing alignment arguments.
+ bool isVLd = (Name.compare(14, 3, "vld", 3) == 0);
+ bool isVSt = (Name.compare(14, 3, "vst", 3) == 0);
+ if (isVLd || isVSt) {
+ unsigned NumVecs = Name.at(17) - '0';
+ if (NumVecs == 0 || NumVecs > 4)
+ return false;
+ bool isLaneOp = (Name.compare(18, 5, "lane.", 5) == 0);
+ if (!isLaneOp && Name.at(18) != '.')
+ return false;
+ unsigned ExpectedArgs = 2; // for the address and alignment
+ if (isVSt || isLaneOp)
+ ExpectedArgs += NumVecs;
+ if (isLaneOp)
+ ExpectedArgs += 1; // for the lane number
+ unsigned NumP = FTy->getNumParams();
+ if (NumP != ExpectedArgs - 1)
+ return false;
+
+ // Change the name of the old (bad) intrinsic, because
+ // its type is incorrect, but we cannot overload that name.
+ F->setName("");
+
+ // One argument is missing: add the alignment argument.
+ std::vector<const Type*> NewParams;
+ for (unsigned p = 0; p < NumP; ++p)
+ NewParams.push_back(FTy->getParamType(p));
+ NewParams.push_back(Type::getInt32Ty(F->getContext()));
+ FunctionType *NewFTy = FunctionType::get(FTy->getReturnType(),
+ NewParams, false);
+ NewFn = cast<Function>(M->getOrInsertFunction(Name, NewFTy));
+ return true;
+ }
}
break;
case 'b':
}
break;
+ case 'm': {
+ // This upgrades the llvm.memcpy, llvm.memmove, and llvm.memset to the
+ // new format that allows overloading the pointer for different address
+ // space (e.g., llvm.memcpy.i16 => llvm.memcpy.p0i8.p0i8.i16)
+ const char* NewFnName = NULL;
+ if (Name.compare(5,8,"memcpy.i",8) == 0) {
+ if (Name[13] == '8')
+ NewFnName = "llvm.memcpy.p0i8.p0i8.i8";
+ else if (Name.compare(13,2,"16") == 0)
+ NewFnName = "llvm.memcpy.p0i8.p0i8.i16";
+ else if (Name.compare(13,2,"32") == 0)
+ NewFnName = "llvm.memcpy.p0i8.p0i8.i32";
+ else if (Name.compare(13,2,"64") == 0)
+ NewFnName = "llvm.memcpy.p0i8.p0i8.i64";
+ } else if (Name.compare(5,9,"memmove.i",9) == 0) {
+ if (Name[14] == '8')
+ NewFnName = "llvm.memmove.p0i8.p0i8.i8";
+ else if (Name.compare(14,2,"16") == 0)
+ NewFnName = "llvm.memmove.p0i8.p0i8.i16";
+ else if (Name.compare(14,2,"32") == 0)
+ NewFnName = "llvm.memmove.p0i8.p0i8.i32";
+ else if (Name.compare(14,2,"64") == 0)
+ NewFnName = "llvm.memmove.p0i8.p0i8.i64";
+ }
+ else if (Name.compare(5,8,"memset.i",8) == 0) {
+ if (Name[13] == '8')
+ NewFnName = "llvm.memset.p0i8.i8";
+ else if (Name.compare(13,2,"16") == 0)
+ NewFnName = "llvm.memset.p0i8.i16";
+ else if (Name.compare(13,2,"32") == 0)
+ NewFnName = "llvm.memset.p0i8.i32";
+ else if (Name.compare(13,2,"64") == 0)
+ NewFnName = "llvm.memset.p0i8.i64";
+ }
+ if (NewFnName) {
+ NewFn = cast<Function>(M->getOrInsertFunction(NewFnName,
+ FTy->getReturnType(),
+ FTy->getParamType(0),
+ FTy->getParamType(1),
+ FTy->getParamType(2),
+ FTy->getParamType(3),
+ Type::getInt1Ty(F->getContext()),
+ (Type *)0));
+ return true;
+ }
+ break;
+ }
case 'p':
// This upgrades the llvm.part.select overloaded intrinsic names to only
// use one type specifier in the name. We only care about the old format
// Calls to these intrinsics are transformed into vector multiplies.
NewFn = 0;
return true;
+ } else if (Name.compare(5, 18, "x86.ssse3.palign.r", 18) == 0 ||
+ Name.compare(5, 22, "x86.ssse3.palign.r.128", 22) == 0) {
+ // Calls to these intrinsics are transformed into vector shuffles, shifts,
+ // or 0.
+ NewFn = 0;
+ return true;
}
-
break;
}
void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
Function *F = CI->getCalledFunction();
LLVMContext &C = CI->getContext();
-
+ ImmutableCallSite CS(CI);
+
assert(F && "CallInst has no function associated with it.");
if (!NewFn) {
+ // Get the Function's name.
+ const std::string& Name = F->getName();
+
+ // Upgrade ARM NEON intrinsics.
+ if (Name.compare(5, 9, "arm.neon.", 9) == 0) {
+ Instruction *NewI;
+ if (Name.compare(14, 7, "vmovls.", 7) == 0) {
+ NewI = new SExtInst(CI->getArgOperand(0), CI->getType(),
+ "upgraded." + CI->getName(), CI);
+ } else if (Name.compare(14, 7, "vmovlu.", 7) == 0) {
+ NewI = new ZExtInst(CI->getArgOperand(0), CI->getType(),
+ "upgraded." + CI->getName(), CI);
+
+ } else if (Name.compare(14, 4, "vadd", 4) == 0 ||
+ Name.compare(14, 4, "vsub", 4) == 0) {
+ // Extend one (vaddw/vsubw) or both (vaddl/vsubl) operands.
+ Value *V0 = CI->getArgOperand(0);
+ Value *V1 = CI->getArgOperand(1);
+ if (Name.at(19) == 's') {
+ if (Name.at(18) == 'l')
+ V0 = new SExtInst(CI->getArgOperand(0), CI->getType(), "", CI);
+ V1 = new SExtInst(CI->getArgOperand(1), CI->getType(), "", CI);
+ } else {
+ assert(Name.at(19) == 'u' && "unexpected vadd/vsub intrinsic");
+ if (Name.at(18) == 'l')
+ V0 = new ZExtInst(CI->getArgOperand(0), CI->getType(), "", CI);
+ V1 = new ZExtInst(CI->getArgOperand(1), CI->getType(), "", CI);
+ }
+ if (Name.compare(14, 4, "vadd", 4) == 0)
+ NewI = BinaryOperator::CreateAdd(V0, V1,"upgraded."+CI->getName(),CI);
+ else
+ NewI = BinaryOperator::CreateSub(V0, V1,"upgraded."+CI->getName(),CI);
+
+ } else if (Name.compare(14, 6, "vmovn.", 6) == 0) {
+ NewI = new TruncInst(CI->getArgOperand(0), CI->getType(),
+ "upgraded." + CI->getName(), CI);
+ } else {
+ llvm_unreachable("Unknown arm.neon function for CallInst upgrade.");
+ }
+ // Replace any uses of the old CallInst.
+ if (!CI->use_empty())
+ CI->replaceAllUsesWith(NewI);
+ CI->eraseFromParent();
+ return;
+ }
+
bool isLoadH = false, isLoadL = false, isMovL = false;
bool isMovSD = false, isShufPD = false;
bool isUnpckhPD = false, isUnpcklPD = false;
if (isLoadH || isLoadL || isMovL || isMovSD || isShufPD ||
isUnpckhPD || isUnpcklPD || isPunpckhQPD || isPunpcklQPD) {
std::vector<Constant*> Idxs;
- Value *Op0 = CI->getOperand(1);
+ Value *Op0 = CI->getArgOperand(0);
ShuffleVectorInst *SI = NULL;
if (isLoadH || isLoadL) {
Value *Op1 = UndefValue::get(Op0->getType());
- Value *Addr = new BitCastInst(CI->getOperand(2),
+ Value *Addr = new BitCastInst(CI->getArgOperand(1),
Type::getDoublePtrTy(C),
"upgraded.", CI);
Value *Load = new LoadInst(Addr, "upgraded.", false, 8, CI);
SI = new ShuffleVectorInst(ZeroV, Op0, Mask, "upgraded.", CI);
} else if (isMovSD ||
isUnpckhPD || isUnpcklPD || isPunpckhQPD || isPunpcklQPD) {
- Value *Op1 = CI->getOperand(2);
+ Value *Op1 = CI->getArgOperand(1);
if (isMovSD) {
Idxs.push_back(ConstantInt::get(Type::getInt32Ty(C), 2));
Idxs.push_back(ConstantInt::get(Type::getInt32Ty(C), 1));
Value *Mask = ConstantVector::get(Idxs);
SI = new ShuffleVectorInst(Op0, Op1, Mask, "upgraded.", CI);
} else if (isShufPD) {
- Value *Op1 = CI->getOperand(2);
- unsigned MaskVal = cast<ConstantInt>(CI->getOperand(3))->getZExtValue();
+ Value *Op1 = CI->getArgOperand(1);
+ unsigned MaskVal =
+ cast<ConstantInt>(CI->getArgOperand(2))->getZExtValue();
Idxs.push_back(ConstantInt::get(Type::getInt32Ty(C), MaskVal & 1));
Idxs.push_back(ConstantInt::get(Type::getInt32Ty(C),
((MaskVal >> 1) & 1)+2));
CI->eraseFromParent();
} else if (F->getName() == "llvm.x86.sse41.pmulld") {
// Upgrade this set of intrinsics into vector multiplies.
- Instruction *Mul = BinaryOperator::CreateMul(CI->getOperand(1),
- CI->getOperand(2),
+ Instruction *Mul = BinaryOperator::CreateMul(CI->getArgOperand(0),
+ CI->getArgOperand(1),
CI->getName(),
CI);
// Fix up all the uses with our new multiply.
// Remove upgraded multiply.
CI->eraseFromParent();
+ } else if (F->getName() == "llvm.x86.ssse3.palign.r") {
+ Value *Op1 = CI->getArgOperand(0);
+ Value *Op2 = CI->getArgOperand(1);
+ Value *Op3 = CI->getArgOperand(2);
+ unsigned shiftVal = cast<ConstantInt>(Op3)->getZExtValue();
+ Value *Rep;
+ IRBuilder<> Builder(C);
+ Builder.SetInsertPoint(CI->getParent(), CI);
+
+ // If palignr is shifting the pair of input vectors less than 9 bytes,
+ // emit a shuffle instruction.
+ if (shiftVal <= 8) {
+ const Type *IntTy = Type::getInt32Ty(C);
+ const Type *EltTy = Type::getInt8Ty(C);
+ const Type *VecTy = VectorType::get(EltTy, 8);
+
+ Op2 = Builder.CreateBitCast(Op2, VecTy);
+ Op1 = Builder.CreateBitCast(Op1, VecTy);
+
+ llvm::SmallVector<llvm::Constant*, 8> Indices;
+ for (unsigned i = 0; i != 8; ++i)
+ Indices.push_back(ConstantInt::get(IntTy, shiftVal + i));
+
+ Value *SV = ConstantVector::get(Indices.begin(), Indices.size());
+ Rep = Builder.CreateShuffleVector(Op2, Op1, SV, "palignr");
+ Rep = Builder.CreateBitCast(Rep, F->getReturnType());
+ }
+
+ // If palignr is shifting the pair of input vectors more than 8 but less
+ // than 16 bytes, emit a logical right shift of the destination.
+ else if (shiftVal < 16) {
+ // MMX has these as 1 x i64 vectors for some odd optimization reasons.
+ const Type *EltTy = Type::getInt64Ty(C);
+ const Type *VecTy = VectorType::get(EltTy, 1);
+
+ Op1 = Builder.CreateBitCast(Op1, VecTy, "cast");
+ Op2 = ConstantInt::get(VecTy, (shiftVal-8) * 8);
+
+ // create i32 constant
+ Function *I =
+ Intrinsic::getDeclaration(F->getParent(), Intrinsic::x86_mmx_psrl_q);
+ Rep = Builder.CreateCall2(I, Op1, Op2, "palignr");
+ }
+
+ // If palignr is shifting the pair of vectors more than 32 bytes, emit zero.
+ else {
+ Rep = Constant::getNullValue(F->getReturnType());
+ }
+
+ // Replace any uses with our new instruction.
+ if (!CI->use_empty())
+ CI->replaceAllUsesWith(Rep);
+
+ // Remove upgraded instruction.
+ CI->eraseFromParent();
+
+ } else if (F->getName() == "llvm.x86.ssse3.palign.r.128") {
+ Value *Op1 = CI->getArgOperand(0);
+ Value *Op2 = CI->getArgOperand(1);
+ Value *Op3 = CI->getArgOperand(2);
+ unsigned shiftVal = cast<ConstantInt>(Op3)->getZExtValue();
+ Value *Rep;
+ IRBuilder<> Builder(C);
+ Builder.SetInsertPoint(CI->getParent(), CI);
+
+ // If palignr is shifting the pair of input vectors less than 17 bytes,
+ // emit a shuffle instruction.
+ if (shiftVal <= 16) {
+ const Type *IntTy = Type::getInt32Ty(C);
+ const Type *EltTy = Type::getInt8Ty(C);
+ const Type *VecTy = VectorType::get(EltTy, 16);
+
+ Op2 = Builder.CreateBitCast(Op2, VecTy);
+ Op1 = Builder.CreateBitCast(Op1, VecTy);
+
+ llvm::SmallVector<llvm::Constant*, 16> Indices;
+ for (unsigned i = 0; i != 16; ++i)
+ Indices.push_back(ConstantInt::get(IntTy, shiftVal + i));
+
+ Value *SV = ConstantVector::get(Indices.begin(), Indices.size());
+ Rep = Builder.CreateShuffleVector(Op2, Op1, SV, "palignr");
+ Rep = Builder.CreateBitCast(Rep, F->getReturnType());
+ }
+
+ // If palignr is shifting the pair of input vectors more than 16 but less
+ // than 32 bytes, emit a logical right shift of the destination.
+ else if (shiftVal < 32) {
+ const Type *EltTy = Type::getInt64Ty(C);
+ const Type *VecTy = VectorType::get(EltTy, 2);
+ const Type *IntTy = Type::getInt32Ty(C);
+
+ Op1 = Builder.CreateBitCast(Op1, VecTy, "cast");
+ Op2 = ConstantInt::get(IntTy, (shiftVal-16) * 8);
+
+ // create i32 constant
+ Function *I =
+ Intrinsic::getDeclaration(F->getParent(), Intrinsic::x86_sse2_psrl_dq);
+ Rep = Builder.CreateCall2(I, Op1, Op2, "palignr");
+ }
+
+ // If palignr is shifting the pair of vectors more than 32 bytes, emit zero.
+ else {
+ Rep = Constant::getNullValue(F->getReturnType());
+ }
+
+ // Replace any uses with our new instruction.
+ if (!CI->use_empty())
+ CI->replaceAllUsesWith(Rep);
+
+ // Remove upgraded instruction.
+ CI->eraseFromParent();
+
} else {
llvm_unreachable("Unknown function for CallInst upgrade.");
}
switch (NewFn->getIntrinsicID()) {
default: llvm_unreachable("Unknown function for CallInst upgrade.");
+ case Intrinsic::arm_neon_vld1:
+ case Intrinsic::arm_neon_vld2:
+ case Intrinsic::arm_neon_vld3:
+ case Intrinsic::arm_neon_vld4:
+ case Intrinsic::arm_neon_vst1:
+ case Intrinsic::arm_neon_vst2:
+ case Intrinsic::arm_neon_vst3:
+ case Intrinsic::arm_neon_vst4:
+ case Intrinsic::arm_neon_vld2lane:
+ case Intrinsic::arm_neon_vld3lane:
+ case Intrinsic::arm_neon_vld4lane:
+ case Intrinsic::arm_neon_vst2lane:
+ case Intrinsic::arm_neon_vst3lane:
+ case Intrinsic::arm_neon_vst4lane: {
+ // Add a default alignment argument of 1.
+ SmallVector<Value*, 8> Operands(CS.arg_begin(), CS.arg_end());
+ Operands.push_back(ConstantInt::get(Type::getInt32Ty(C), 1));
+ CallInst *NewCI = CallInst::Create(NewFn, Operands.begin(), Operands.end(),
+ CI->getName(), CI);
+ NewCI->setTailCall(CI->isTailCall());
+ NewCI->setCallingConv(CI->getCallingConv());
+
+ // Handle any uses of the old CallInst.
+ if (!CI->use_empty())
+ // Replace all uses of the old call with the new cast which has the
+ // correct type.
+ CI->replaceAllUsesWith(NewCI);
+
+ // Clean up the old call now that it has been completely upgraded.
+ CI->eraseFromParent();
+ break;
+ }
+
case Intrinsic::x86_mmx_psll_d:
case Intrinsic::x86_mmx_psll_q:
case Intrinsic::x86_mmx_psll_w:
case Intrinsic::x86_mmx_psrl_w: {
Value *Operands[2];
- Operands[0] = CI->getOperand(1);
+ Operands[0] = CI->getArgOperand(0);
// Cast the second parameter to the correct type.
- BitCastInst *BC = new BitCastInst(CI->getOperand(2),
+ BitCastInst *BC = new BitCastInst(CI->getArgOperand(1),
NewFn->getFunctionType()->getParamType(1),
"upgraded.", CI);
Operands[1] = BC;
case Intrinsic::ctlz:
case Intrinsic::ctpop:
case Intrinsic::cttz: {
- // Build a small vector of the 1..(N-1) operands, which are the
- // parameters.
- SmallVector<Value*, 8> Operands(CI->op_begin()+1, CI->op_end());
+ // Build a small vector of the original arguments.
+ SmallVector<Value*, 8> Operands(CS.arg_begin(), CS.arg_end());
// Construct a new CallInst
CallInst *NewCI = CallInst::Create(NewFn, Operands.begin(), Operands.end(),
case Intrinsic::eh_selector:
case Intrinsic::eh_typeid_for: {
// Only the return type changed.
- SmallVector<Value*, 8> Operands(CI->op_begin() + 1, CI->op_end());
+ SmallVector<Value*, 8> Operands(CS.arg_begin(), CS.arg_end());
CallInst *NewCI = CallInst::Create(NewFn, Operands.begin(), Operands.end(),
"upgraded." + CI->getName(), CI);
NewCI->setTailCall(CI->isTailCall());
CI->eraseFromParent();
}
break;
+ case Intrinsic::memcpy:
+ case Intrinsic::memmove:
+ case Intrinsic::memset: {
+ // Add isVolatile
+ const llvm::Type *I1Ty = llvm::Type::getInt1Ty(CI->getContext());
+ Value *Operands[5] = { CI->getArgOperand(0), CI->getArgOperand(1),
+ CI->getArgOperand(2), CI->getArgOperand(3),
+ llvm::ConstantInt::get(I1Ty, 0) };
+ CallInst *NewCI = CallInst::Create(NewFn, Operands, Operands+5,
+ CI->getName(), CI);
+ NewCI->setTailCall(CI->isTailCall());
+ NewCI->setCallingConv(CI->getCallingConv());
+ // Handle any uses of the old CallInst.
+ if (!CI->use_empty())
+ // Replace all uses of the old call with the new cast which has the
+ // correct type.
+ CI->replaceAllUsesWith(NewCI);
+
+ // Clean up the old call now that it has been completely upgraded.
+ CI->eraseFromParent();
+ break;
+ }
}
}
if (Function *Declare = M->getFunction("llvm.dbg.declare")) {
if (!Declare->use_empty()) {
DbgDeclareInst *DDI = cast<DbgDeclareInst>(Declare->use_back());
- if (!isa<MDNode>(DDI->getOperand(1)) ||!isa<MDNode>(DDI->getOperand(2))) {
+ if (!isa<MDNode>(DDI->getArgOperand(0)) ||
+ !isa<MDNode>(DDI->getArgOperand(1))) {
while (!Declare->use_empty()) {
CallInst *CI = cast<CallInst>(Declare->use_back());
CI->eraseFromParent();