llvm/lib/VMCore/AutoUpgrade.cpp
Mon P Wang e3b3a7241c Added support for overloading intrinsics (atomics) based on pointers
to different address spaces.  This alters the naming scheme for those
intrinsics, e.g., atomic.load.add.i32 => atomic.load.add.i32.p0i32


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@54195 91177308-0d34-0410-b5e6-96231b3b80d8
2008-07-30 04:36:53 +00:00

431 lines
17 KiB
C++

//===-- AutoUpgrade.cpp - Implement auto-upgrade helper functions ---------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements the auto-upgrade helper functions
//
//===----------------------------------------------------------------------===//
#include "llvm/AutoUpgrade.h"
#include "llvm/Constants.h"
#include "llvm/Function.h"
#include "llvm/Module.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
#include "llvm/ADT/SmallVector.h"
#include <cstring>
using namespace llvm;
static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) {
assert(F && "Illegal to upgrade a non-existent Function.");
// Get the Function's name.
const std::string& Name = F->getName();
// Convenience
const FunctionType *FTy = F->getFunctionType();
// Quickly eliminate it, if it's not a candidate.
if (Name.length() <= 8 || Name[0] != 'l' || Name[1] != 'l' ||
Name[2] != 'v' || Name[3] != 'm' || Name[4] != '.')
return false;
Module *M = F->getParent();
switch (Name[5]) {
default: break;
case 'a':
// This upgrades the llvm.atomic.lcs, llvm.atomic.las, llvm.atomic.lss,
// and atomics with default address spaces to their new names to their new
// function name (e.g. llvm.atomic.add.i32 => llvm.atomic.add.i32.p0i32)
if (Name.compare(5,7,"atomic.",7) == 0) {
if (Name.compare(12,3,"lcs",3) == 0) {
std::string::size_type delim = Name.find('.',12);
F->setName("llvm.atomic.cmp.swap" + Name.substr(delim) +
".p0" + Name.substr(delim+1));
NewFn = F;
return true;
}
else if (Name.compare(12,3,"las",3) == 0) {
std::string::size_type delim = Name.find('.',12);
F->setName("llvm.atomic.load.add"+Name.substr(delim)
+ ".p0" + Name.substr(delim+1));
NewFn = F;
return true;
}
else if (Name.compare(12,3,"lss",3) == 0) {
std::string::size_type delim = Name.find('.',12);
F->setName("llvm.atomic.load.sub"+Name.substr(delim)
+ ".p0" + Name.substr(delim+1));
NewFn = F;
return true;
}
else if (Name.rfind(".p") == std::string::npos) {
// We don't have an address space qualifier so this has be upgraded
// to the new name. Copy the type name at the end of the intrinsic
// and add to it
std::string::size_type delim = Name.find_last_of('.');
assert(delim != std::string::npos && "can not find type");
F->setName(Name + ".p0" + Name.substr(delim+1));
NewFn = F;
return true;
}
}
break;
case 'b':
// This upgrades the name of the llvm.bswap intrinsic function to only use
// a single type name for overloading. We only care about the old format
// 'llvm.bswap.i*.i*', so check for 'bswap.' and then for there being
// a '.' after 'bswap.'
if (Name.compare(5,6,"bswap.",6) == 0) {
std::string::size_type delim = Name.find('.',11);
if (delim != std::string::npos) {
// Construct the new name as 'llvm.bswap' + '.i*'
F->setName(Name.substr(0,10)+Name.substr(delim));
NewFn = F;
return true;
}
}
break;
case 'c':
// We only want to fix the 'llvm.ct*' intrinsics which do not have the
// correct return type, so we check for the name, and then check if the
// return type does not match the parameter type.
if ( (Name.compare(5,5,"ctpop",5) == 0 ||
Name.compare(5,4,"ctlz",4) == 0 ||
Name.compare(5,4,"cttz",4) == 0) &&
FTy->getReturnType() != FTy->getParamType(0)) {
// We first need to change the name of the old (bad) intrinsic, because
// its type is incorrect, but we cannot overload that name. We
// arbitrarily unique it here allowing us to construct a correctly named
// and typed function below.
F->setName("");
// Now construct the new intrinsic with the correct name and type. We
// leave the old function around in order to query its type, whatever it
// may be, and correctly convert up to the new type.
NewFn = cast<Function>(M->getOrInsertFunction(Name,
FTy->getParamType(0),
FTy->getParamType(0),
(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
// 'llvm.part.select.i*.i*', and solve as above with bswap.
if (Name.compare(5,12,"part.select.",12) == 0) {
std::string::size_type delim = Name.find('.',17);
if (delim != std::string::npos) {
// Construct a new name as 'llvm.part.select' + '.i*'
F->setName(Name.substr(0,16)+Name.substr(delim));
NewFn = F;
return true;
}
break;
}
// This upgrades the llvm.part.set intrinsics similarly as above, however
// we care about 'llvm.part.set.i*.i*.i*', but only the first two types
// must match. There is an additional type specifier after these two
// matching types that we must retain when upgrading. Thus, we require
// finding 2 periods, not just one, after the intrinsic name.
if (Name.compare(5,9,"part.set.",9) == 0) {
std::string::size_type delim = Name.find('.',14);
if (delim != std::string::npos &&
Name.find('.',delim+1) != std::string::npos) {
// Construct a new name as 'llvm.part.select' + '.i*.i*'
F->setName(Name.substr(0,13)+Name.substr(delim));
NewFn = F;
return true;
}
break;
}
break;
case 'x':
// This fixes all MMX shift intrinsic instructions to take a
// v1i64 instead of a v2i32 as the second parameter.
if (Name.compare(5,10,"x86.mmx.ps",10) == 0 &&
(Name.compare(13,4,"psll", 4) == 0 ||
Name.compare(13,4,"psra", 4) == 0 ||
Name.compare(13,4,"psrl", 4) == 0) && Name[17] != 'i') {
const llvm::Type *VT = VectorType::get(IntegerType::get(64), 1);
// We don't have to do anything if the parameter already has
// the correct type.
if (FTy->getParamType(1) == VT)
break;
// We first need to change the name of the old (bad) intrinsic, because
// its type is incorrect, but we cannot overload that name. We
// arbitrarily unique it here allowing us to construct a correctly named
// and typed function below.
F->setName("");
assert(FTy->getNumParams() == 2 && "MMX shift intrinsics take 2 args!");
// Now construct the new intrinsic with the correct name and type. We
// leave the old function around in order to query its type, whatever it
// may be, and correctly convert up to the new type.
NewFn = cast<Function>(M->getOrInsertFunction(Name,
FTy->getReturnType(),
FTy->getParamType(0),
VT,
(Type *)0));
return true;
} else if (Name.compare(5,17,"x86.sse2.loadh.pd",17) == 0 ||
Name.compare(5,17,"x86.sse2.loadl.pd",17) == 0 ||
Name.compare(5,16,"x86.sse2.movl.dq",16) == 0 ||
Name.compare(5,15,"x86.sse2.movs.d",15) == 0 ||
Name.compare(5,16,"x86.sse2.shuf.pd",16) == 0 ||
Name.compare(5,18,"x86.sse2.unpckh.pd",18) == 0 ||
Name.compare(5,18,"x86.sse2.unpckl.pd",18) == 0 ||
Name.compare(5,20,"x86.sse2.punpckh.qdq",20) == 0 ||
Name.compare(5,20,"x86.sse2.punpckl.qdq",20) == 0) {
// Calls to these intrinsics are transformed into ShuffleVector's.
NewFn = 0;
return true;
}
break;
}
// This may not belong here. This function is effectively being overloaded
// to both detect an intrinsic which needs upgrading, and to provide the
// upgraded form of the intrinsic. We should perhaps have two separate
// functions for this.
return false;
}
bool llvm::UpgradeIntrinsicFunction(Function *F, Function *&NewFn) {
NewFn = 0;
bool Upgraded = UpgradeIntrinsicFunction1(F, NewFn);
// Upgrade intrinsic attributes. This does not change the function.
if (NewFn)
F = NewFn;
if (unsigned id = F->getIntrinsicID(true))
F->setParamAttrs(Intrinsic::getParamAttrs((Intrinsic::ID)id));
return Upgraded;
}
// UpgradeIntrinsicCall - Upgrade a call to an old intrinsic to be a call the
// upgraded intrinsic. All argument and return casting must be provided in
// order to seamlessly integrate with existing context.
void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
Function *F = CI->getCalledFunction();
assert(F && "CallInst has no function associated with it.");
if (!NewFn) {
bool isLoadH = false, isLoadL = false, isMovL = false;
bool isMovSD = false, isShufPD = false;
bool isUnpckhPD = false, isUnpcklPD = false;
bool isPunpckhQPD = false, isPunpcklQPD = false;
if (strcmp(F->getNameStart(), "llvm.x86.sse2.loadh.pd") == 0)
isLoadH = true;
else if (strcmp(F->getNameStart(), "llvm.x86.sse2.loadl.pd") == 0)
isLoadL = true;
else if (strcmp(F->getNameStart(), "llvm.x86.sse2.movl.dq") == 0)
isMovL = true;
else if (strcmp(F->getNameStart(), "llvm.x86.sse2.movs.d") == 0)
isMovSD = true;
else if (strcmp(F->getNameStart(), "llvm.x86.sse2.shuf.pd") == 0)
isShufPD = true;
else if (strcmp(F->getNameStart(), "llvm.x86.sse2.unpckh.pd") == 0)
isUnpckhPD = true;
else if (strcmp(F->getNameStart(), "llvm.x86.sse2.unpckl.pd") == 0)
isUnpcklPD = true;
else if (strcmp(F->getNameStart(), "llvm.x86.sse2.punpckh.qdq") == 0)
isPunpckhQPD = true;
else if (strcmp(F->getNameStart(), "llvm.x86.sse2.punpckl.qdq") == 0)
isPunpcklQPD = true;
if (isLoadH || isLoadL || isMovL || isMovSD || isShufPD ||
isUnpckhPD || isUnpcklPD || isPunpckhQPD || isPunpcklQPD) {
std::vector<Constant*> Idxs;
Value *Op0 = CI->getOperand(1);
ShuffleVectorInst *SI = NULL;
if (isLoadH || isLoadL) {
Value *Op1 = UndefValue::get(Op0->getType());
Value *Addr = new BitCastInst(CI->getOperand(2),
PointerType::getUnqual(Type::DoubleTy),
"upgraded.", CI);
Value *Load = new LoadInst(Addr, "upgraded.", false, 8, CI);
Value *Idx = ConstantInt::get(Type::Int32Ty, 0);
Op1 = InsertElementInst::Create(Op1, Load, Idx, "upgraded.", CI);
if (isLoadH) {
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 0));
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 2));
} else {
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 2));
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 1));
}
Value *Mask = ConstantVector::get(Idxs);
SI = new ShuffleVectorInst(Op0, Op1, Mask, "upgraded.", CI);
} else if (isMovL) {
Constant *Zero = ConstantInt::get(Type::Int32Ty, 0);
Idxs.push_back(Zero);
Idxs.push_back(Zero);
Idxs.push_back(Zero);
Idxs.push_back(Zero);
Value *ZeroV = ConstantVector::get(Idxs);
Idxs.clear();
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 4));
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 5));
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 2));
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 3));
Value *Mask = ConstantVector::get(Idxs);
SI = new ShuffleVectorInst(ZeroV, Op0, Mask, "upgraded.", CI);
} else if (isMovSD ||
isUnpckhPD || isUnpcklPD || isPunpckhQPD || isPunpcklQPD) {
Value *Op1 = CI->getOperand(2);
if (isMovSD) {
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 2));
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 1));
} else if (isUnpckhPD || isPunpckhQPD) {
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 1));
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 3));
} else {
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 0));
Idxs.push_back(ConstantInt::get(Type::Int32Ty, 2));
}
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();
Idxs.push_back(ConstantInt::get(Type::Int32Ty, MaskVal & 1));
Idxs.push_back(ConstantInt::get(Type::Int32Ty, ((MaskVal >> 1) & 1)+2));
Value *Mask = ConstantVector::get(Idxs);
SI = new ShuffleVectorInst(Op0, Op1, Mask, "upgraded.", CI);
}
assert(SI && "Unexpected!");
// 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(SI);
// Clean up the old call now that it has been completely upgraded.
CI->eraseFromParent();
} else {
assert(0 && "Unknown function for CallInst upgrade.");
}
return;
}
switch (NewFn->getIntrinsicID()) {
default: assert(0 && "Unknown function for CallInst upgrade.");
case Intrinsic::x86_mmx_psll_d:
case Intrinsic::x86_mmx_psll_q:
case Intrinsic::x86_mmx_psll_w:
case Intrinsic::x86_mmx_psra_d:
case Intrinsic::x86_mmx_psra_w:
case Intrinsic::x86_mmx_psrl_d:
case Intrinsic::x86_mmx_psrl_q:
case Intrinsic::x86_mmx_psrl_w: {
Value *Operands[2];
Operands[0] = CI->getOperand(1);
// Cast the second parameter to the correct type.
BitCastInst *BC = new BitCastInst(CI->getOperand(2),
NewFn->getFunctionType()->getParamType(1),
"upgraded.", CI);
Operands[1] = BC;
// Construct a new CallInst
CallInst *NewCI = CallInst::Create(NewFn, Operands, Operands+2,
"upgraded."+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::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());
// Construct a new CallInst
CallInst *NewCI = CallInst::Create(NewFn, Operands.begin(), Operands.end(),
"upgraded."+CI->getName(), CI);
NewCI->setTailCall(CI->isTailCall());
NewCI->setCallingConv(CI->getCallingConv());
// Handle any uses of the old CallInst.
if (!CI->use_empty()) {
// Check for sign extend parameter attributes on the return values.
bool SrcSExt = NewFn->getParamAttrs().paramHasAttr(0, ParamAttr::SExt);
bool DestSExt = F->getParamAttrs().paramHasAttr(0, ParamAttr::SExt);
// Construct an appropriate cast from the new return type to the old.
CastInst *RetCast = CastInst::Create(
CastInst::getCastOpcode(NewCI, SrcSExt,
F->getReturnType(),
DestSExt),
NewCI, F->getReturnType(),
NewCI->getName(), CI);
NewCI->moveBefore(RetCast);
// Replace all uses of the old call with the new cast which has the
// correct type.
CI->replaceAllUsesWith(RetCast);
}
// Clean up the old call now that it has been completely upgraded.
CI->eraseFromParent();
}
break;
}
}
// This tests each Function to determine if it needs upgrading. When we find
// one we are interested in, we then upgrade all calls to reflect the new
// function.
void llvm::UpgradeCallsToIntrinsic(Function* F) {
assert(F && "Illegal attempt to upgrade a non-existent intrinsic.");
// Upgrade the function and check if it is a totaly new function.
Function* NewFn;
if (UpgradeIntrinsicFunction(F, NewFn)) {
if (NewFn != F) {
// Replace all uses to the old function with the new one if necessary.
for (Value::use_iterator UI = F->use_begin(), UE = F->use_end();
UI != UE; ) {
if (CallInst* CI = dyn_cast<CallInst>(*UI++))
UpgradeIntrinsicCall(CI, NewFn);
}
// Remove old function, no longer used, from the module.
F->eraseFromParent();
}
}
}