}
void NVPTXAsmPrinter::printReturnValStr(const Function *F, raw_ostream &O) {
- const DataLayout *TD = TM.getDataLayout();
+ const DataLayout &DL = getDataLayout();
const TargetLowering *TLI = nvptxSubtarget->getTargetLowering();
Type *Ty = F->getReturnType();
O << ".param .b" << size << " func_retval0";
} else if (isa<PointerType>(Ty)) {
- O << ".param .b" << TLI->getPointerTy().getSizeInBits()
+ O << ".param .b" << TLI->getPointerTy(DL).getSizeInBits()
<< " func_retval0";
} else if ((Ty->getTypeID() == Type::StructTyID) || isa<VectorType>(Ty)) {
- unsigned totalsz = TD->getTypeAllocSize(Ty);
+ unsigned totalsz = DL.getTypeAllocSize(Ty);
unsigned retAlignment = 0;
if (!llvm::getAlign(*F, 0, retAlignment))
- retAlignment = TD->getABITypeAlignment(Ty);
+ retAlignment = DL.getABITypeAlignment(Ty);
O << ".param .align " << retAlignment << " .b8 func_retval0[" << totalsz
<< "]";
} else
llvm_unreachable("Unknown return type");
} else {
SmallVector<EVT, 16> vtparts;
- ComputeValueVTs(*TLI, Ty, vtparts);
+ ComputeValueVTs(*TLI, DL, Ty, vtparts);
unsigned idx = 0;
for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
unsigned elems = 1;
// Construct a default subtarget off of the TargetMachine defaults. The
// rest of NVPTX isn't friendly to change subtargets per function and
// so the default TargetMachine will have all of the options.
- StringRef TT = TM.getTargetTriple();
+ const Triple &TT = TM.getTargetTriple();
StringRef CPU = TM.getTargetCPU();
StringRef FS = TM.getTargetFeatureString();
const NVPTXTargetMachine &NTM = static_cast<const NVPTXTargetMachine &>(TM);
raw_svector_ostream OS1(Str1);
MMI = getAnalysisIfAvailable<MachineModuleInfo>();
- MMI->AnalyzeModule(M);
// We need to call the parent's one explicitly.
//bool Result = AsmPrinter::doInitialization(M);
const_cast<TargetLoweringObjectFile &>(getObjFileLowering())
.Initialize(OutContext, TM);
- Mang = new Mangler(TM.getDataLayout());
+ Mang = new Mangler();
// Emit header before any dwarf directives are emitted below.
emitHeader(M, OS1, STI);
}
// If we're not NVCL we're CUDA, go ahead and emit filenames.
- if (Triple(TM.getTargetTriple()).getOS() != Triple::NVCL)
+ if (TM.getTargetTriple().getOS() != Triple::NVCL)
recordAndEmitFilenames(M);
GlobalsEmitted = false;
GVar->getName().startswith("nvvm."))
return;
- const DataLayout *TD = TM.getDataLayout();
+ const DataLayout &DL = getDataLayout();
// GlobalVariables are always constant pointers themselves.
const PointerType *PTy = GVar->getType();
}
if (GVar->getAlignment() == 0)
- O << " .align " << (int) TD->getPrefTypeAlignment(ETy);
+ O << " .align " << (int)DL.getPrefTypeAlignment(ETy);
else
O << " .align " << GVar->getAlignment();
case Type::StructTyID:
case Type::ArrayTyID:
case Type::VectorTyID:
- ElementSize = TD->getTypeStoreSize(ETy);
+ ElementSize = DL.getTypeStoreSize(ETy);
// Ptx allows variable initilization only for constant and
// global state spaces.
if (((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable *GVar,
raw_ostream &O) {
- const DataLayout *TD = TM.getDataLayout();
+ const DataLayout &DL = getDataLayout();
// GlobalVariables are always constant pointers themselves.
const PointerType *PTy = GVar->getType();
O << ".";
emitPTXAddressSpace(PTy->getAddressSpace(), O);
if (GVar->getAlignment() == 0)
- O << " .align " << (int) TD->getPrefTypeAlignment(ETy);
+ O << " .align " << (int)DL.getPrefTypeAlignment(ETy);
else
O << " .align " << GVar->getAlignment();
case Type::StructTyID:
case Type::ArrayTyID:
case Type::VectorTyID:
- ElementSize = TD->getTypeStoreSize(ETy);
+ ElementSize = DL.getTypeStoreSize(ETy);
O << " .b8 ";
getSymbol(GVar)->print(O, MAI);
O << "[";
return;
}
-static unsigned int getOpenCLAlignment(const DataLayout *TD, Type *Ty) {
+static unsigned int getOpenCLAlignment(const DataLayout &DL, Type *Ty) {
if (Ty->isSingleValueType())
- return TD->getPrefTypeAlignment(Ty);
+ return DL.getPrefTypeAlignment(Ty);
const ArrayType *ATy = dyn_cast<ArrayType>(Ty);
if (ATy)
- return getOpenCLAlignment(TD, ATy->getElementType());
+ return getOpenCLAlignment(DL, ATy->getElementType());
const StructType *STy = dyn_cast<StructType>(Ty);
if (STy) {
// largest alignment.
for (unsigned i = 0, e = STy->getNumElements(); i != e; i++) {
Type *ETy = STy->getElementType(i);
- unsigned int align = getOpenCLAlignment(TD, ETy);
+ unsigned int align = getOpenCLAlignment(DL, ETy);
if (align > alignStruct)
alignStruct = align;
}
const FunctionType *FTy = dyn_cast<FunctionType>(Ty);
if (FTy)
- return TD->getPointerPrefAlignment();
- return TD->getPrefTypeAlignment(Ty);
+ return DL.getPointerPrefAlignment();
+ return DL.getPrefTypeAlignment(Ty);
}
void NVPTXAsmPrinter::printParamName(Function::const_arg_iterator I,
}
void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
- const DataLayout *TD = TM.getDataLayout();
+ const DataLayout &DL = getDataLayout();
const AttributeSet &PAL = F->getAttributes();
const TargetLowering *TLI = nvptxSubtarget->getTargetLowering();
Function::const_arg_iterator I, E;
bool first = true;
bool isKernelFunc = llvm::isKernelFunction(*F);
bool isABI = (nvptxSubtarget->getSmVersion() >= 20);
- MVT thePointerTy = TLI->getPointerTy();
+ MVT thePointerTy = TLI->getPointerTy(DL);
O << "(\n";
// size = typeallocsize of element type
unsigned align = PAL.getParamAlignment(paramIndex + 1);
if (align == 0)
- align = TD->getABITypeAlignment(Ty);
+ align = DL.getABITypeAlignment(Ty);
- unsigned sz = TD->getTypeAllocSize(Ty);
+ unsigned sz = DL.getTypeAllocSize(Ty);
O << "\t.param .align " << align << " .b8 ";
printParamName(I, paramIndex, O);
O << "[" << sz << "]";
O << ".ptr .global ";
break;
}
- O << ".align " << (int) getOpenCLAlignment(TD, ETy) << " ";
+ O << ".align " << (int)getOpenCLAlignment(DL, ETy) << " ";
}
printParamName(I, paramIndex, O);
continue;
// size = typeallocsize of element type
unsigned align = PAL.getParamAlignment(paramIndex + 1);
if (align == 0)
- align = TD->getABITypeAlignment(ETy);
+ align = DL.getABITypeAlignment(ETy);
- unsigned sz = TD->getTypeAllocSize(ETy);
+ unsigned sz = DL.getTypeAllocSize(ETy);
O << "\t.param .align " << align << " .b8 ";
printParamName(I, paramIndex, O);
O << "[" << sz << "]";
// Further, if a part is vector, print the above for
// each vector element.
SmallVector<EVT, 16> vtparts;
- ComputeValueVTs(*TLI, ETy, vtparts);
+ ComputeValueVTs(*TLI, DL, ETy, vtparts);
for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
unsigned elems = 1;
EVT elemtype = vtparts[i];
llvm_unreachable("Not scalar type found in printScalarConstant()");
}
+// These utility functions assure we get the right sequence of bytes for a given
+// type even for big-endian machines
+template <typename T> static void ConvertIntToBytes(unsigned char *p, T val) {
+ int64_t vp = (int64_t)val;
+ for (unsigned i = 0; i < sizeof(T); ++i) {
+ p[i] = (unsigned char)vp;
+ vp >>= 8;
+ }
+}
+static void ConvertFloatToBytes(unsigned char *p, float val) {
+ int32_t *vp = (int32_t *)&val;
+ for (unsigned i = 0; i < sizeof(int32_t); ++i) {
+ p[i] = (unsigned char)*vp;
+ *vp >>= 8;
+ }
+}
+static void ConvertDoubleToBytes(unsigned char *p, double val) {
+ int64_t *vp = (int64_t *)&val;
+ for (unsigned i = 0; i < sizeof(int64_t); ++i) {
+ p[i] = (unsigned char)*vp;
+ *vp >>= 8;
+ }
+}
+
void NVPTXAsmPrinter::bufferLEByte(const Constant *CPV, int Bytes,
AggBuffer *aggBuffer) {
- const DataLayout *TD = TM.getDataLayout();
+ const DataLayout &DL = getDataLayout();
if (isa<UndefValue>(CPV) || CPV->isNullValue()) {
- int s = TD->getTypeAllocSize(CPV->getType());
+ int s = DL.getTypeAllocSize(CPV->getType());
if (s < Bytes)
s = Bytes;
aggBuffer->addZeros(s);
return;
}
- unsigned char *ptr;
+ unsigned char ptr[8];
switch (CPV->getType()->getTypeID()) {
case Type::IntegerTyID: {
const Type *ETy = CPV->getType();
if (ETy == Type::getInt8Ty(CPV->getContext())) {
unsigned char c = (unsigned char)cast<ConstantInt>(CPV)->getZExtValue();
- ptr = &c;
+ ConvertIntToBytes<>(ptr, c);
aggBuffer->addBytes(ptr, 1, Bytes);
} else if (ETy == Type::getInt16Ty(CPV->getContext())) {
short int16 = (short)cast<ConstantInt>(CPV)->getZExtValue();
- ptr = (unsigned char *)&int16;
+ ConvertIntToBytes<>(ptr, int16);
aggBuffer->addBytes(ptr, 2, Bytes);
} else if (ETy == Type::getInt32Ty(CPV->getContext())) {
if (const ConstantInt *constInt = dyn_cast<ConstantInt>(CPV)) {
int int32 = (int)(constInt->getZExtValue());
- ptr = (unsigned char *)&int32;
+ ConvertIntToBytes<>(ptr, int32);
aggBuffer->addBytes(ptr, 4, Bytes);
break;
} else if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
if (const ConstantInt *constInt = dyn_cast<ConstantInt>(
- ConstantFoldConstantExpression(Cexpr, *TD))) {
+ ConstantFoldConstantExpression(Cexpr, DL))) {
int int32 = (int)(constInt->getZExtValue());
- ptr = (unsigned char *)&int32;
+ ConvertIntToBytes<>(ptr, int32);
aggBuffer->addBytes(ptr, 4, Bytes);
break;
}
} else if (ETy == Type::getInt64Ty(CPV->getContext())) {
if (const ConstantInt *constInt = dyn_cast<ConstantInt>(CPV)) {
long long int64 = (long long)(constInt->getZExtValue());
- ptr = (unsigned char *)&int64;
+ ConvertIntToBytes<>(ptr, int64);
aggBuffer->addBytes(ptr, 8, Bytes);
break;
} else if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
if (const ConstantInt *constInt = dyn_cast<ConstantInt>(
- ConstantFoldConstantExpression(Cexpr, *TD))) {
+ ConstantFoldConstantExpression(Cexpr, DL))) {
long long int64 = (long long)(constInt->getZExtValue());
- ptr = (unsigned char *)&int64;
+ ConvertIntToBytes<>(ptr, int64);
aggBuffer->addBytes(ptr, 8, Bytes);
break;
}
const Type *Ty = CFP->getType();
if (Ty == Type::getFloatTy(CPV->getContext())) {
float float32 = (float) CFP->getValueAPF().convertToFloat();
- ptr = (unsigned char *)&float32;
+ ConvertFloatToBytes(ptr, float32);
aggBuffer->addBytes(ptr, 4, Bytes);
} else if (Ty == Type::getDoubleTy(CPV->getContext())) {
double float64 = CFP->getValueAPF().convertToDouble();
- ptr = (unsigned char *)&float64;
+ ConvertDoubleToBytes(ptr, float64);
aggBuffer->addBytes(ptr, 8, Bytes);
} else {
llvm_unreachable("unsupported fp const type");
const Value *v = Cexpr->stripPointerCasts();
aggBuffer->addSymbol(v, Cexpr);
}
- unsigned int s = TD->getTypeAllocSize(CPV->getType());
+ unsigned int s = DL.getTypeAllocSize(CPV->getType());
aggBuffer->addZeros(s);
break;
}
case Type::StructTyID: {
if (isa<ConstantArray>(CPV) || isa<ConstantVector>(CPV) ||
isa<ConstantStruct>(CPV) || isa<ConstantDataSequential>(CPV)) {
- int ElementSize = TD->getTypeAllocSize(CPV->getType());
+ int ElementSize = DL.getTypeAllocSize(CPV->getType());
bufferAggregateConstant(CPV, aggBuffer);
if (Bytes > ElementSize)
aggBuffer->addZeros(Bytes - ElementSize);
void NVPTXAsmPrinter::bufferAggregateConstant(const Constant *CPV,
AggBuffer *aggBuffer) {
- const DataLayout *TD = TM.getDataLayout();
+ const DataLayout &DL = getDataLayout();
int Bytes;
// Old constants
StructType *ST = cast<StructType>(CPV->getType());
for (unsigned i = 0, e = CPV->getNumOperands(); i != e; ++i) {
if (i == (e - 1))
- Bytes = TD->getStructLayout(ST)->getElementOffset(0) +
- TD->getTypeAllocSize(ST) -
- TD->getStructLayout(ST)->getElementOffset(i);
+ Bytes = DL.getStructLayout(ST)->getElementOffset(0) +
+ DL.getTypeAllocSize(ST) -
+ DL.getStructLayout(ST)->getElementOffset(i);
else
- Bytes = TD->getStructLayout(ST)->getElementOffset(i + 1) -
- TD->getStructLayout(ST)->getElementOffset(i);
+ Bytes = DL.getStructLayout(ST)->getElementOffset(i + 1) -
+ DL.getStructLayout(ST)->getElementOffset(i);
bufferLEByte(cast<Constant>(CPV->getOperand(i)), Bytes, aggBuffer);
}
}
// If the code isn't optimized, there may be outstanding folding
// opportunities. Attempt to fold the expression using DataLayout as a
// last resort before giving up.
- if (Constant *C = ConstantFoldConstantExpression(CE, *TM.getDataLayout()))
+ if (Constant *C = ConstantFoldConstantExpression(CE, getDataLayout()))
if (C != CE)
return lowerConstantForGV(C, ProcessingGeneric);
}
case Instruction::GetElementPtr: {
- const DataLayout &DL = *TM.getDataLayout();
+ const DataLayout &DL = getDataLayout();
// Generate a symbolic expression for the byte address
APInt OffsetAI(DL.getPointerTypeSizeInBits(CE->getType()), 0);
return lowerConstantForGV(CE->getOperand(0), ProcessingGeneric);
case Instruction::IntToPtr: {
- const DataLayout &DL = *TM.getDataLayout();
+ const DataLayout &DL = getDataLayout();
// Handle casts to pointers by changing them into casts to the appropriate
// integer type. This promotes constant folding and simplifies this code.
}
case Instruction::PtrToInt: {
- const DataLayout &DL = *TM.getDataLayout();
+ const DataLayout &DL = getDataLayout();
// Support only foldable casts to/from pointers that can be eliminated by
// changing the pointer to the appropriately sized integer type.