RPCS3/llvm
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm.git synced 2024-12-21 03:28:31 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity
3da59db637
llvm/lib/CodeGen/IntrinsicLowering.cpp
Reid Spencer 3da59db637 For PR950: 
The long awaited CAST patch. This introduces 12 new instructions into LLVM
to replace the cast instruction. Corresponding changes throughout LLVM are
provided. This passes llvm-test, llvm/test, and SPEC CPUINT2000 with the
exception of 175.vpr which fails only on a slight floating point output
difference.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31931 91177308-0d34-0410-b5e6-96231b3b80d8
2006-11-27 01:05:10 +00:00

501 lines
20 KiB
C++
Raw Blame History

//===-- IntrinsicLowering.cpp - Intrinsic Lowering default implementation -===//
//
// The LLVM Compiler Infrastructure
//
// This file was developed by the LLVM research group and is distributed under
// the University of Illinois Open Source License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements the IntrinsicLowering class.
//
//===----------------------------------------------------------------------===//
#include "llvm/CodeGen/IntrinsicLowering.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Module.h"
#include "llvm/Instructions.h"
#include "llvm/Type.h"
#include <iostream>
using namespace llvm;
template <class ArgIt>
static Function *EnsureFunctionExists(Module &M, const char *Name,
ArgIt ArgBegin, ArgIt ArgEnd,
const Type *RetTy) {
if (Function *F = M.getNamedFunction(Name)) return F;
// It doesn't already exist in the program, insert a new definition now.
std::vector<const Type *> ParamTys;
for (ArgIt I = ArgBegin; I != ArgEnd; ++I)
ParamTys.push_back(I->getType());
return M.getOrInsertFunction(Name, FunctionType::get(RetTy, ParamTys, false));
}
/// ReplaceCallWith - This function is used when we want to lower an intrinsic
/// call to a call of an external function. This handles hard cases such as
/// when there was already a prototype for the external function, and if that
/// prototype doesn't match the arguments we expect to pass in.
template <class ArgIt>
static CallInst *ReplaceCallWith(const char *NewFn, CallInst *CI,
ArgIt ArgBegin, ArgIt ArgEnd,
const unsigned *castOpcodes,
const Type *RetTy, Function *&FCache) {
if (!FCache) {
// If we haven't already looked up this function, check to see if the
// program already contains a function with this name.
Module *M = CI->getParent()->getParent()->getParent();
FCache = M->getNamedFunction(NewFn);
if (!FCache) {
// It doesn't already exist in the program, insert a new definition now.
std::vector<const Type *> ParamTys;
for (ArgIt I = ArgBegin; I != ArgEnd; ++I)
ParamTys.push_back((*I)->getType());
FCache = M->getOrInsertFunction(NewFn,
FunctionType::get(RetTy, ParamTys, false));
}
}
const FunctionType *FT = FCache->getFunctionType();
std::vector<Value*> Operands;
unsigned ArgNo = 0;
for (ArgIt I = ArgBegin; I != ArgEnd && ArgNo != FT->getNumParams();
++I, ++ArgNo) {
Value *Arg = *I;
if (Arg->getType() != FT->getParamType(ArgNo))
if (castOpcodes[ArgNo])
Arg = CastInst::create(Instruction::CastOps(castOpcodes[ArgNo]),
Arg, FT->getParamType(ArgNo), Arg->getName(), CI);
else
Arg = CastInst::createInferredCast(Arg, FT->getParamType(ArgNo),
Arg->getName(), CI);
Operands.push_back(Arg);
}
// Pass nulls into any additional arguments...
for (; ArgNo != FT->getNumParams(); ++ArgNo)
Operands.push_back(Constant::getNullValue(FT->getParamType(ArgNo)));
std::string Name = CI->getName(); CI->setName("");
if (FT->getReturnType() == Type::VoidTy) Name.clear();
CallInst *NewCI = new CallInst(FCache, Operands, Name, CI);
if (!CI->use_empty()) {
Value *V = NewCI;
if (CI->getType() != NewCI->getType())
V = CastInst::createInferredCast(NewCI, CI->getType(), Name, CI);
CI->replaceAllUsesWith(V);
}
return NewCI;
}
void IntrinsicLowering::AddPrototypes(Module &M) {
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
if (I->isExternal() && !I->use_empty())
switch (I->getIntrinsicID()) {
default: break;
case Intrinsic::setjmp:
EnsureFunctionExists(M, "setjmp", I->arg_begin(), I->arg_end(),
Type::IntTy);
break;
case Intrinsic::longjmp:
EnsureFunctionExists(M, "longjmp", I->arg_begin(), I->arg_end(),
Type::VoidTy);
break;
case Intrinsic::siglongjmp:
EnsureFunctionExists(M, "abort", I->arg_end(), I->arg_end(),
Type::VoidTy);
break;
case Intrinsic::memcpy_i32:
case Intrinsic::memcpy_i64:
EnsureFunctionExists(M, "memcpy", I->arg_begin(), --I->arg_end(),
I->arg_begin()->getType());
break;
case Intrinsic::memmove_i32:
case Intrinsic::memmove_i64:
EnsureFunctionExists(M, "memmove", I->arg_begin(), --I->arg_end(),
I->arg_begin()->getType());
break;
case Intrinsic::memset_i32:
case Intrinsic::memset_i64:
M.getOrInsertFunction("memset", PointerType::get(Type::SByteTy),
PointerType::get(Type::SByteTy),
Type::IntTy, (--(--I->arg_end()))->getType(),
(Type *)0);
break;
case Intrinsic::isunordered_f32:
case Intrinsic::isunordered_f64:
EnsureFunctionExists(M, "isunordered", I->arg_begin(), I->arg_end(),
Type::BoolTy);
break;
case Intrinsic::sqrt_f32:
case Intrinsic::sqrt_f64:
if(I->arg_begin()->getType() == Type::FloatTy)
EnsureFunctionExists(M, "sqrtf", I->arg_begin(), I->arg_end(),
Type::FloatTy);
else
EnsureFunctionExists(M, "sqrt", I->arg_begin(), I->arg_end(),
Type::DoubleTy);
break;
}
}
/// LowerBSWAP - Emit the code to lower bswap of V before the specified
/// instruction IP.
static Value *LowerBSWAP(Value *V, Instruction *IP) {
assert(V->getType()->isInteger() && "Can't bswap a non-integer type!");
unsigned BitSize = V->getType()->getPrimitiveSizeInBits();
switch(BitSize) {
default: assert(0 && "Unhandled type size of value to byteswap!");
case 16: {
Value *Tmp1 = new ShiftInst(Instruction::Shl, V,
ConstantInt::get(Type::UByteTy,8),"bswap.2",IP);
Value *Tmp2 = new ShiftInst(Instruction::LShr, V,
ConstantInt::get(Type::UByteTy,8),"bswap.1",IP);
V = BinaryOperator::createOr(Tmp1, Tmp2, "bswap.i16", IP);
break;
}
case 32: {
Value *Tmp4 = new ShiftInst(Instruction::Shl, V,
ConstantInt::get(Type::UByteTy,24),"bswap.4", IP);
Value *Tmp3 = new ShiftInst(Instruction::Shl, V,
ConstantInt::get(Type::UByteTy,8),"bswap.3",IP);
Value *Tmp2 = new ShiftInst(Instruction::LShr, V,
ConstantInt::get(Type::UByteTy,8),"bswap.2",IP);
Value *Tmp1 = new ShiftInst(Instruction::LShr, V,
ConstantInt::get(Type::UByteTy,24),"bswap.1", IP);
Tmp3 = BinaryOperator::createAnd(Tmp3,
ConstantInt::get(Type::UIntTy, 0xFF0000),
"bswap.and3", IP);
Tmp2 = BinaryOperator::createAnd(Tmp2,
ConstantInt::get(Type::UIntTy, 0xFF00),
"bswap.and2", IP);
Tmp4 = BinaryOperator::createOr(Tmp4, Tmp3, "bswap.or1", IP);
Tmp2 = BinaryOperator::createOr(Tmp2, Tmp1, "bswap.or2", IP);
V = BinaryOperator::createOr(Tmp4, Tmp3, "bswap.i32", IP);
break;
}
case 64: {
Value *Tmp8 = new ShiftInst(Instruction::Shl, V,
ConstantInt::get(Type::UByteTy,56),"bswap.8", IP);
Value *Tmp7 = new ShiftInst(Instruction::Shl, V,
ConstantInt::get(Type::UByteTy,40),"bswap.7", IP);
Value *Tmp6 = new ShiftInst(Instruction::Shl, V,
ConstantInt::get(Type::UByteTy,24),"bswap.6", IP);
Value *Tmp5 = new ShiftInst(Instruction::Shl, V,
ConstantInt::get(Type::UByteTy,8),"bswap.5", IP);
Value* Tmp4 = new ShiftInst(Instruction::LShr, V,
ConstantInt::get(Type::UByteTy,8),"bswap.4", IP);
Value* Tmp3 = new ShiftInst(Instruction::LShr, V,
ConstantInt::get(Type::UByteTy,24),"bswap.3", IP);
Value* Tmp2 = new ShiftInst(Instruction::LShr, V,
ConstantInt::get(Type::UByteTy,40),"bswap.2", IP);
Value* Tmp1 = new ShiftInst(Instruction::LShr, V,
ConstantInt::get(Type::UByteTy,56),"bswap.1", IP);
Tmp7 = BinaryOperator::createAnd(Tmp7,
ConstantInt::get(Type::ULongTy,
0xFF000000000000ULL),
"bswap.and7", IP);
Tmp6 = BinaryOperator::createAnd(Tmp6,
ConstantInt::get(Type::ULongTy, 0xFF0000000000ULL),
"bswap.and6", IP);
Tmp5 = BinaryOperator::createAnd(Tmp5,
ConstantInt::get(Type::ULongTy, 0xFF00000000ULL),
"bswap.and5", IP);
Tmp4 = BinaryOperator::createAnd(Tmp4,
ConstantInt::get(Type::ULongTy, 0xFF000000ULL),
"bswap.and4", IP);
Tmp3 = BinaryOperator::createAnd(Tmp3,
ConstantInt::get(Type::ULongTy, 0xFF0000ULL),
"bswap.and3", IP);
Tmp2 = BinaryOperator::createAnd(Tmp2,
ConstantInt::get(Type::ULongTy, 0xFF00ULL),
"bswap.and2", IP);
Tmp8 = BinaryOperator::createOr(Tmp8, Tmp7, "bswap.or1", IP);
Tmp6 = BinaryOperator::createOr(Tmp6, Tmp5, "bswap.or2", IP);
Tmp4 = BinaryOperator::createOr(Tmp4, Tmp3, "bswap.or3", IP);
Tmp2 = BinaryOperator::createOr(Tmp2, Tmp1, "bswap.or4", IP);
Tmp8 = BinaryOperator::createOr(Tmp8, Tmp6, "bswap.or5", IP);
Tmp4 = BinaryOperator::createOr(Tmp4, Tmp2, "bswap.or6", IP);
V = BinaryOperator::createOr(Tmp8, Tmp4, "bswap.i64", IP);
break;
}
}
return V;
}
/// LowerCTPOP - Emit the code to lower ctpop of V before the specified
/// instruction IP.
static Value *LowerCTPOP(Value *V, Instruction *IP) {
assert(V->getType()->isInteger() && "Can't ctpop a non-integer type!");
static const uint64_t MaskValues[6] = {
0x5555555555555555ULL, 0x3333333333333333ULL,
0x0F0F0F0F0F0F0F0FULL, 0x00FF00FF00FF00FFULL,
0x0000FFFF0000FFFFULL, 0x00000000FFFFFFFFULL
};
unsigned BitSize = V->getType()->getPrimitiveSizeInBits();
for (unsigned i = 1, ct = 0; i != BitSize; i <<= 1, ++ct) {
Value *MaskCst =
ConstantExpr::getCast(ConstantInt::get(Type::ULongTy, MaskValues[ct]),
V->getType());
Value *LHS = BinaryOperator::createAnd(V, MaskCst, "cppop.and1", IP);
Value *VShift = new ShiftInst(Instruction::LShr, V,
ConstantInt::get(Type::UByteTy, i), "ctpop.sh", IP);
Value *RHS = BinaryOperator::createAnd(VShift, MaskCst, "cppop.and2", IP);
V = BinaryOperator::createAdd(LHS, RHS, "ctpop.step", IP);
}
return V;
}
/// LowerCTLZ - Emit the code to lower ctlz of V before the specified
/// instruction IP.
static Value *LowerCTLZ(Value *V, Instruction *IP) {
unsigned BitSize = V->getType()->getPrimitiveSizeInBits();
for (unsigned i = 1; i != BitSize; i <<= 1) {
Value *ShVal = ConstantInt::get(Type::UByteTy, i);
ShVal = new ShiftInst(Instruction::LShr, V, ShVal, "ctlz.sh", IP);
V = BinaryOperator::createOr(V, ShVal, "ctlz.step", IP);
}
V = BinaryOperator::createNot(V, "", IP);
return LowerCTPOP(V, IP);
}
void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) {
Function *Callee = CI->getCalledFunction();
assert(Callee && "Cannot lower an indirect call!");
switch (Callee->getIntrinsicID()) {
case Intrinsic::not_intrinsic:
std::cerr << "Cannot lower a call to a non-intrinsic function '"
<< Callee->getName() << "'!\n";
abort();
default:
std::cerr << "Error: Code generator does not support intrinsic function '"
<< Callee->getName() << "'!\n";
abort();
// 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
// convert the call to an explicit setjmp or longjmp call.
case Intrinsic::setjmp: {
static Function *SetjmpFCache = 0;
static const unsigned castOpcodes[] = { Instruction::BitCast };
Value *V = ReplaceCallWith("setjmp", CI, CI->op_begin()+1, CI->op_end(),
castOpcodes, Type::IntTy, SetjmpFCache);
if (CI->getType() != Type::VoidTy)
CI->replaceAllUsesWith(V);
break;
}
case Intrinsic::sigsetjmp:
if (CI->getType() != Type::VoidTy)
CI->replaceAllUsesWith(Constant::getNullValue(CI->getType()));
break;
case Intrinsic::longjmp: {
static Function *LongjmpFCache = 0;
static const unsigned castOpcodes[] =
{ Instruction::BitCast, 0 };
ReplaceCallWith("longjmp", CI, CI->op_begin()+1, CI->op_end(),
castOpcodes, Type::VoidTy, LongjmpFCache);
break;
}
case Intrinsic::siglongjmp: {
// Insert the call to abort
static Function *AbortFCache = 0;
static const unsigned castOpcodes[] =
{ Instruction::BitCast, 0 };
ReplaceCallWith("abort", CI, CI->op_end(), CI->op_end(),
castOpcodes, Type::VoidTy, AbortFCache);
break;
}
case Intrinsic::ctpop_i8:
case Intrinsic::ctpop_i16:
case Intrinsic::ctpop_i32:
case Intrinsic::ctpop_i64:
CI->replaceAllUsesWith(LowerCTPOP(CI->getOperand(1), CI));
break;
case Intrinsic::bswap_i16:
case Intrinsic::bswap_i32:
case Intrinsic::bswap_i64:
CI->replaceAllUsesWith(LowerBSWAP(CI->getOperand(1), CI));
break;
case Intrinsic::ctlz_i8:
case Intrinsic::ctlz_i16:
case Intrinsic::ctlz_i32:
case Intrinsic::ctlz_i64:
CI->replaceAllUsesWith(LowerCTLZ(CI->getOperand(1), CI));
break;
case Intrinsic::cttz_i8:
case Intrinsic::cttz_i16:
case Intrinsic::cttz_i32:
case Intrinsic::cttz_i64: {
// cttz(x) -> ctpop(~X & (X-1))
Value *Src = CI->getOperand(1);
Value *NotSrc = BinaryOperator::createNot(Src, Src->getName()+".not", CI);
Value *SrcM1 = ConstantInt::get(Src->getType(), 1);
SrcM1 = BinaryOperator::createSub(Src, SrcM1, "", CI);
Src = LowerCTPOP(BinaryOperator::createAnd(NotSrc, SrcM1, "", CI), CI);
CI->replaceAllUsesWith(Src);
break;
}
case Intrinsic::stacksave:
case Intrinsic::stackrestore: {
static bool Warned = false;
if (!Warned)
std::cerr << "WARNING: this target does not support the llvm.stack"
<< (Callee->getIntrinsicID() == Intrinsic::stacksave ?
"save" : "restore") << " intrinsic.\n";
Warned = true;
if (Callee->getIntrinsicID() == Intrinsic::stacksave)
CI->replaceAllUsesWith(Constant::getNullValue(CI->getType()));
break;
}
case Intrinsic::returnaddress:
case Intrinsic::frameaddress:
std::cerr << "WARNING: this target does not support the llvm."
<< (Callee->getIntrinsicID() == Intrinsic::returnaddress ?
"return" : "frame") << "address intrinsic.\n";
CI->replaceAllUsesWith(ConstantPointerNull::get(
cast<PointerType>(CI->getType())));
break;
case Intrinsic::prefetch:
break; // Simply strip out prefetches on unsupported architectures
case Intrinsic::pcmarker:
break; // Simply strip out pcmarker on unsupported architectures
case Intrinsic::readcyclecounter: {
std::cerr << "WARNING: this target does not support the llvm.readcyclecoun"
<< "ter intrinsic. It is being lowered to a constant 0\n";
CI->replaceAllUsesWith(ConstantInt::get(Type::ULongTy, 0));
break;
}
case Intrinsic::dbg_stoppoint:
case Intrinsic::dbg_region_start:
case Intrinsic::dbg_region_end:
case Intrinsic::dbg_func_start:
case Intrinsic::dbg_declare:
break; // Simply strip out debugging intrinsics
case Intrinsic::memcpy_i32: {
// The memcpy intrinsic take an extra alignment argument that the memcpy
// libc function does not.
static unsigned opcodes[] =
{ Instruction::BitCast, Instruction::BitCast, Instruction::BitCast };
// FIXME:
// if (target_is_64_bit) opcodes[2] = Instruction::ZExt;
// else opcodes[2] = Instruction::BitCast;
static Function *MemcpyFCache = 0;
ReplaceCallWith("memcpy", CI, CI->op_begin()+1, CI->op_end()-1,
opcodes, (*(CI->op_begin()+1))->getType(), MemcpyFCache);
break;
}
case Intrinsic::memcpy_i64: {
static unsigned opcodes[] =
{ Instruction::BitCast, Instruction::BitCast, Instruction::Trunc };
// FIXME:
// if (target_is_64_bit) opcodes[2] = Instruction::BitCast;
// else opcodes[2] = Instruction::Trunc;
static Function *MemcpyFCache = 0;
ReplaceCallWith("memcpy", CI, CI->op_begin()+1, CI->op_end()-1,
opcodes, (*(CI->op_begin()+1))->getType(), MemcpyFCache);
break;
}
case Intrinsic::memmove_i32: {
// The memmove intrinsic take an extra alignment argument that the memmove
// libc function does not.
static unsigned opcodes[] =
{ Instruction::BitCast, Instruction::BitCast, Instruction::BitCast };
// FIXME:
// if (target_is_64_bit) opcodes[2] = Instruction::ZExt;
// else opcodes[2] = Instruction::BitCast;
static Function *MemmoveFCache = 0;
ReplaceCallWith("memmove", CI, CI->op_begin()+1, CI->op_end()-1,
opcodes, (*(CI->op_begin()+1))->getType(), MemmoveFCache);
break;
}
case Intrinsic::memmove_i64: {
// The memmove intrinsic take an extra alignment argument that the memmove
// libc function does not.
static const unsigned opcodes[] =
{ Instruction::BitCast, Instruction::BitCast, Instruction::Trunc };
// if (target_is_64_bit) opcodes[2] = Instruction::BitCast;
// else opcodes[2] = Instruction::Trunc;
static Function *MemmoveFCache = 0;
ReplaceCallWith("memmove", CI, CI->op_begin()+1, CI->op_end()-1,
opcodes, (*(CI->op_begin()+1))->getType(), MemmoveFCache);
break;
}
case Intrinsic::memset_i32: {
// The memset intrinsic take an extra alignment argument that the memset
// libc function does not.
static const unsigned opcodes[] =
{ Instruction::BitCast, Instruction::ZExt, Instruction::ZExt, 0 };
// if (target_is_64_bit) opcodes[2] = Instruction::BitCast;
// else opcodes[2] = Instruction::ZExt;
static Function *MemsetFCache = 0;
ReplaceCallWith("memset", CI, CI->op_begin()+1, CI->op_end()-1,
opcodes, (*(CI->op_begin()+1))->getType(), MemsetFCache);
}
case Intrinsic::memset_i64: {
// The memset intrinsic take an extra alignment argument that the memset
// libc function does not.
static const unsigned opcodes[] =
{ Instruction::BitCast, Instruction::ZExt, Instruction::Trunc, 0 };
// if (target_is_64_bit) opcodes[2] = Instruction::BitCast;
// else opcodes[2] = Instruction::Trunc;
static Function *MemsetFCache = 0;
ReplaceCallWith("memset", CI, CI->op_begin()+1, CI->op_end()-1,
opcodes, (*(CI->op_begin()+1))->getType(), MemsetFCache);
break;
}
case Intrinsic::isunordered_f32:
case Intrinsic::isunordered_f64: {
Value *L = CI->getOperand(1);
Value *R = CI->getOperand(2);
Value *LIsNan = new SetCondInst(Instruction::SetNE, L, L, "LIsNan", CI);
Value *RIsNan = new SetCondInst(Instruction::SetNE, R, R, "RIsNan", CI);
CI->replaceAllUsesWith(
BinaryOperator::create(Instruction::Or, LIsNan, RIsNan,
"isunordered", CI));
break;
}
case Intrinsic::sqrt_f32: {
static const unsigned opcodes[] = { 0 };
static Function *sqrtfFCache = 0;
ReplaceCallWith("sqrtf", CI, CI->op_begin()+1, CI->op_end(),
opcodes, Type::FloatTy, sqrtfFCache);
break;
}
case Intrinsic::sqrt_f64: {
static const unsigned opcodes[] = { 0 };
static Function *sqrtFCache = 0;
ReplaceCallWith("sqrt", CI, CI->op_begin()+1, CI->op_end(),
opcodes, Type::DoubleTy, sqrtFCache);
break;
}
}
assert(CI->use_empty() &&
"Lowering should have eliminated any uses of the intrinsic call!");
CI->eraseFromParent();
}
Reference in New Issue View Git Blame Copy Permalink