1 //===-- NVPTXAsmPrinter.cpp - NVPTX LLVM assembly writer ------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file contains a printer that converts from our internal representation
11 // of machine-dependent LLVM code to NVPTX assembly language.
13 //===----------------------------------------------------------------------===//
15 #include "NVPTXAsmPrinter.h"
16 #include "InstPrinter/NVPTXInstPrinter.h"
17 #include "MCTargetDesc/NVPTXMCAsmInfo.h"
19 #include "NVPTXInstrInfo.h"
20 #include "NVPTXMCExpr.h"
21 #include "NVPTXMachineFunctionInfo.h"
22 #include "NVPTXRegisterInfo.h"
23 #include "NVPTXTargetMachine.h"
24 #include "NVPTXUtilities.h"
25 #include "cl_common_defines.h"
26 #include "llvm/ADT/StringExtras.h"
27 #include "llvm/Analysis/ConstantFolding.h"
28 #include "llvm/CodeGen/Analysis.h"
29 #include "llvm/CodeGen/MachineFrameInfo.h"
30 #include "llvm/CodeGen/MachineModuleInfo.h"
31 #include "llvm/CodeGen/MachineRegisterInfo.h"
32 #include "llvm/IR/DebugInfo.h"
33 #include "llvm/IR/DerivedTypes.h"
34 #include "llvm/IR/Function.h"
35 #include "llvm/IR/GlobalVariable.h"
36 #include "llvm/IR/Mangler.h"
37 #include "llvm/IR/Module.h"
38 #include "llvm/IR/Operator.h"
39 #include "llvm/MC/MCStreamer.h"
40 #include "llvm/MC/MCSymbol.h"
41 #include "llvm/Support/CommandLine.h"
42 #include "llvm/Support/ErrorHandling.h"
43 #include "llvm/Support/FormattedStream.h"
44 #include "llvm/Support/Path.h"
45 #include "llvm/Support/TargetRegistry.h"
46 #include "llvm/Support/TimeValue.h"
47 #include "llvm/Target/TargetLoweringObjectFile.h"
51 #define DEPOTNAME "__local_depot"
54 EmitLineNumbers("nvptx-emit-line-numbers", cl::Hidden,
55 cl::desc("NVPTX Specific: Emit Line numbers even without -G"),
59 InterleaveSrc("nvptx-emit-src", cl::ZeroOrMore, cl::Hidden,
60 cl::desc("NVPTX Specific: Emit source line in ptx file"),
64 /// DiscoverDependentGlobals - Return a set of GlobalVariables on which \p V
66 void DiscoverDependentGlobals(const Value *V,
67 DenseSet<const GlobalVariable *> &Globals) {
68 if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
71 if (const User *U = dyn_cast<User>(V)) {
72 for (unsigned i = 0, e = U->getNumOperands(); i != e; ++i) {
73 DiscoverDependentGlobals(U->getOperand(i), Globals);
79 /// VisitGlobalVariableForEmission - Add \p GV to the list of GlobalVariable
80 /// instances to be emitted, but only after any dependents have been added
82 void VisitGlobalVariableForEmission(
83 const GlobalVariable *GV, SmallVectorImpl<const GlobalVariable *> &Order,
84 DenseSet<const GlobalVariable *> &Visited,
85 DenseSet<const GlobalVariable *> &Visiting) {
86 // Have we already visited this one?
87 if (Visited.count(GV))
90 // Do we have a circular dependency?
91 if (!Visiting.insert(GV).second)
92 report_fatal_error("Circular dependency found in global variable set");
94 // Make sure we visit all dependents first
95 DenseSet<const GlobalVariable *> Others;
96 for (unsigned i = 0, e = GV->getNumOperands(); i != e; ++i)
97 DiscoverDependentGlobals(GV->getOperand(i), Others);
99 for (DenseSet<const GlobalVariable *>::iterator I = Others.begin(),
102 VisitGlobalVariableForEmission(*I, Order, Visited, Visiting);
104 // Now we can visit ourself
111 void NVPTXAsmPrinter::emitLineNumberAsDotLoc(const MachineInstr &MI) {
112 if (!EmitLineNumbers)
117 DebugLoc curLoc = MI.getDebugLoc();
119 if (prevDebugLoc.isUnknown() && curLoc.isUnknown())
122 if (prevDebugLoc == curLoc)
125 prevDebugLoc = curLoc;
127 if (curLoc.isUnknown())
130 const MachineFunction *MF = MI.getParent()->getParent();
131 //const TargetMachine &TM = MF->getTarget();
133 const LLVMContext &ctx = MF->getFunction()->getContext();
134 DIScope Scope(curLoc.getScope(ctx));
136 assert((!Scope || Scope.isScope()) &&
137 "Scope of a DebugLoc should be null or a DIScope.");
141 StringRef fileName(Scope.getFilename());
142 StringRef dirName(Scope.getDirectory());
143 SmallString<128> FullPathName = dirName;
144 if (!dirName.empty() && !sys::path::is_absolute(fileName)) {
145 sys::path::append(FullPathName, fileName);
146 fileName = FullPathName.str();
149 if (filenameMap.find(fileName.str()) == filenameMap.end())
152 // Emit the line from the source file.
154 this->emitSrcInText(fileName.str(), curLoc.getLine());
156 std::stringstream temp;
157 temp << "\t.loc " << filenameMap[fileName.str()] << " " << curLoc.getLine()
158 << " " << curLoc.getCol();
159 OutStreamer.EmitRawText(Twine(temp.str().c_str()));
162 void NVPTXAsmPrinter::EmitInstruction(const MachineInstr *MI) {
163 SmallString<128> Str;
164 raw_svector_ostream OS(Str);
165 if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA)
166 emitLineNumberAsDotLoc(*MI);
169 lowerToMCInst(MI, Inst);
170 EmitToStreamer(OutStreamer, Inst);
173 // Handle symbol backtracking for targets that do not support image handles
174 bool NVPTXAsmPrinter::lowerImageHandleOperand(const MachineInstr *MI,
175 unsigned OpNo, MCOperand &MCOp) {
176 const MachineOperand &MO = MI->getOperand(OpNo);
177 const MCInstrDesc &MCID = MI->getDesc();
179 if (MCID.TSFlags & NVPTXII::IsTexFlag) {
180 // This is a texture fetch, so operand 4 is a texref and operand 5 is
182 if (OpNo == 4 && MO.isImm()) {
183 lowerImageHandleSymbol(MO.getImm(), MCOp);
186 if (OpNo == 5 && MO.isImm() && !(MCID.TSFlags & NVPTXII::IsTexModeUnifiedFlag)) {
187 lowerImageHandleSymbol(MO.getImm(), MCOp);
192 } else if (MCID.TSFlags & NVPTXII::IsSuldMask) {
194 1 << (((MCID.TSFlags & NVPTXII::IsSuldMask) >> NVPTXII::IsSuldShift) - 1);
196 // For a surface load of vector size N, the Nth operand will be the surfref
197 if (OpNo == VecSize && MO.isImm()) {
198 lowerImageHandleSymbol(MO.getImm(), MCOp);
203 } else if (MCID.TSFlags & NVPTXII::IsSustFlag) {
204 // This is a surface store, so operand 0 is a surfref
205 if (OpNo == 0 && MO.isImm()) {
206 lowerImageHandleSymbol(MO.getImm(), MCOp);
211 } else if (MCID.TSFlags & NVPTXII::IsSurfTexQueryFlag) {
212 // This is a query, so operand 1 is a surfref/texref
213 if (OpNo == 1 && MO.isImm()) {
214 lowerImageHandleSymbol(MO.getImm(), MCOp);
224 void NVPTXAsmPrinter::lowerImageHandleSymbol(unsigned Index, MCOperand &MCOp) {
226 TargetMachine &TM = const_cast<TargetMachine&>(MF->getTarget());
227 NVPTXTargetMachine &nvTM = static_cast<NVPTXTargetMachine&>(TM);
228 const NVPTXMachineFunctionInfo *MFI = MF->getInfo<NVPTXMachineFunctionInfo>();
229 const char *Sym = MFI->getImageHandleSymbol(Index);
230 std::string *SymNamePtr =
231 nvTM.getManagedStrPool()->getManagedString(Sym);
232 MCOp = GetSymbolRef(OutContext.GetOrCreateSymbol(
233 StringRef(SymNamePtr->c_str())));
236 void NVPTXAsmPrinter::lowerToMCInst(const MachineInstr *MI, MCInst &OutMI) {
237 OutMI.setOpcode(MI->getOpcode());
238 const NVPTXSubtarget &ST = TM.getSubtarget<NVPTXSubtarget>();
240 // Special: Do not mangle symbol operand of CALL_PROTOTYPE
241 if (MI->getOpcode() == NVPTX::CALL_PROTOTYPE) {
242 const MachineOperand &MO = MI->getOperand(0);
243 OutMI.addOperand(GetSymbolRef(
244 OutContext.GetOrCreateSymbol(Twine(MO.getSymbolName()))));
248 for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
249 const MachineOperand &MO = MI->getOperand(i);
252 if (!ST.hasImageHandles()) {
253 if (lowerImageHandleOperand(MI, i, MCOp)) {
254 OutMI.addOperand(MCOp);
259 if (lowerOperand(MO, MCOp))
260 OutMI.addOperand(MCOp);
264 bool NVPTXAsmPrinter::lowerOperand(const MachineOperand &MO,
266 switch (MO.getType()) {
267 default: llvm_unreachable("unknown operand type");
268 case MachineOperand::MO_Register:
269 MCOp = MCOperand::CreateReg(encodeVirtualRegister(MO.getReg()));
271 case MachineOperand::MO_Immediate:
272 MCOp = MCOperand::CreateImm(MO.getImm());
274 case MachineOperand::MO_MachineBasicBlock:
275 MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
276 MO.getMBB()->getSymbol(), OutContext));
278 case MachineOperand::MO_ExternalSymbol:
279 MCOp = GetSymbolRef(GetExternalSymbolSymbol(MO.getSymbolName()));
281 case MachineOperand::MO_GlobalAddress:
282 MCOp = GetSymbolRef(getSymbol(MO.getGlobal()));
284 case MachineOperand::MO_FPImmediate: {
285 const ConstantFP *Cnt = MO.getFPImm();
286 APFloat Val = Cnt->getValueAPF();
288 switch (Cnt->getType()->getTypeID()) {
289 default: report_fatal_error("Unsupported FP type"); break;
290 case Type::FloatTyID:
291 MCOp = MCOperand::CreateExpr(
292 NVPTXFloatMCExpr::CreateConstantFPSingle(Val, OutContext));
294 case Type::DoubleTyID:
295 MCOp = MCOperand::CreateExpr(
296 NVPTXFloatMCExpr::CreateConstantFPDouble(Val, OutContext));
305 unsigned NVPTXAsmPrinter::encodeVirtualRegister(unsigned Reg) {
306 if (TargetRegisterInfo::isVirtualRegister(Reg)) {
307 const TargetRegisterClass *RC = MRI->getRegClass(Reg);
309 DenseMap<unsigned, unsigned> &RegMap = VRegMapping[RC];
310 unsigned RegNum = RegMap[Reg];
312 // Encode the register class in the upper 4 bits
313 // Must be kept in sync with NVPTXInstPrinter::printRegName
315 if (RC == &NVPTX::Int1RegsRegClass) {
317 } else if (RC == &NVPTX::Int16RegsRegClass) {
319 } else if (RC == &NVPTX::Int32RegsRegClass) {
321 } else if (RC == &NVPTX::Int64RegsRegClass) {
323 } else if (RC == &NVPTX::Float32RegsRegClass) {
325 } else if (RC == &NVPTX::Float64RegsRegClass) {
328 report_fatal_error("Bad register class");
331 // Insert the vreg number
332 Ret |= (RegNum & 0x0FFFFFFF);
335 // Some special-use registers are actually physical registers.
336 // Encode this as the register class ID of 0 and the real register ID.
337 return Reg & 0x0FFFFFFF;
341 MCOperand NVPTXAsmPrinter::GetSymbolRef(const MCSymbol *Symbol) {
343 Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_None,
345 return MCOperand::CreateExpr(Expr);
348 void NVPTXAsmPrinter::printReturnValStr(const Function *F, raw_ostream &O) {
349 const DataLayout *TD = TM.getSubtargetImpl()->getDataLayout();
350 const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering();
352 Type *Ty = F->getReturnType();
354 bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
356 if (Ty->getTypeID() == Type::VoidTyID)
362 if (Ty->isFloatingPointTy() || Ty->isIntegerTy()) {
364 if (const IntegerType *ITy = dyn_cast<IntegerType>(Ty)) {
365 size = ITy->getBitWidth();
369 assert(Ty->isFloatingPointTy() && "Floating point type expected here");
370 size = Ty->getPrimitiveSizeInBits();
373 O << ".param .b" << size << " func_retval0";
374 } else if (isa<PointerType>(Ty)) {
375 O << ".param .b" << TLI->getPointerTy().getSizeInBits()
377 } else if ((Ty->getTypeID() == Type::StructTyID) || isa<VectorType>(Ty)) {
378 unsigned totalsz = TD->getTypeAllocSize(Ty);
379 unsigned retAlignment = 0;
380 if (!llvm::getAlign(*F, 0, retAlignment))
381 retAlignment = TD->getABITypeAlignment(Ty);
382 O << ".param .align " << retAlignment << " .b8 func_retval0[" << totalsz
385 llvm_unreachable("Unknown return type");
387 SmallVector<EVT, 16> vtparts;
388 ComputeValueVTs(*TLI, Ty, vtparts);
390 for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
392 EVT elemtype = vtparts[i];
393 if (vtparts[i].isVector()) {
394 elems = vtparts[i].getVectorNumElements();
395 elemtype = vtparts[i].getVectorElementType();
398 for (unsigned j = 0, je = elems; j != je; ++j) {
399 unsigned sz = elemtype.getSizeInBits();
400 if (elemtype.isInteger() && (sz < 32))
402 O << ".reg .b" << sz << " func_retval" << idx;
415 void NVPTXAsmPrinter::printReturnValStr(const MachineFunction &MF,
417 const Function *F = MF.getFunction();
418 printReturnValStr(F, O);
421 void NVPTXAsmPrinter::EmitFunctionEntryLabel() {
422 SmallString<128> Str;
423 raw_svector_ostream O(Str);
425 if (!GlobalsEmitted) {
426 emitGlobals(*MF->getFunction()->getParent());
427 GlobalsEmitted = true;
431 MRI = &MF->getRegInfo();
432 F = MF->getFunction();
433 emitLinkageDirective(F, O);
434 if (llvm::isKernelFunction(*F))
438 printReturnValStr(*MF, O);
443 emitFunctionParamList(*MF, O);
445 if (llvm::isKernelFunction(*F))
446 emitKernelFunctionDirectives(*F, O);
448 OutStreamer.EmitRawText(O.str());
450 prevDebugLoc = DebugLoc();
453 void NVPTXAsmPrinter::EmitFunctionBodyStart() {
455 OutStreamer.EmitRawText(StringRef("{\n"));
456 setAndEmitFunctionVirtualRegisters(*MF);
458 SmallString<128> Str;
459 raw_svector_ostream O(Str);
460 emitDemotedVars(MF->getFunction(), O);
461 OutStreamer.EmitRawText(O.str());
464 void NVPTXAsmPrinter::EmitFunctionBodyEnd() {
465 OutStreamer.EmitRawText(StringRef("}\n"));
469 void NVPTXAsmPrinter::emitImplicitDef(const MachineInstr *MI) const {
470 unsigned RegNo = MI->getOperand(0).getReg();
471 const TargetRegisterInfo *TRI = TM.getSubtargetImpl()->getRegisterInfo();
472 if (TRI->isVirtualRegister(RegNo)) {
473 OutStreamer.AddComment(Twine("implicit-def: ") +
474 getVirtualRegisterName(RegNo));
476 OutStreamer.AddComment(
477 Twine("implicit-def: ") +
478 TM.getSubtargetImpl()->getRegisterInfo()->getName(RegNo));
480 OutStreamer.AddBlankLine();
483 void NVPTXAsmPrinter::emitKernelFunctionDirectives(const Function &F,
484 raw_ostream &O) const {
485 // If the NVVM IR has some of reqntid* specified, then output
486 // the reqntid directive, and set the unspecified ones to 1.
487 // If none of reqntid* is specified, don't output reqntid directive.
488 unsigned reqntidx, reqntidy, reqntidz;
489 bool specified = false;
490 if (llvm::getReqNTIDx(F, reqntidx) == false)
494 if (llvm::getReqNTIDy(F, reqntidy) == false)
498 if (llvm::getReqNTIDz(F, reqntidz) == false)
504 O << ".reqntid " << reqntidx << ", " << reqntidy << ", " << reqntidz
507 // If the NVVM IR has some of maxntid* specified, then output
508 // the maxntid directive, and set the unspecified ones to 1.
509 // If none of maxntid* is specified, don't output maxntid directive.
510 unsigned maxntidx, maxntidy, maxntidz;
512 if (llvm::getMaxNTIDx(F, maxntidx) == false)
516 if (llvm::getMaxNTIDy(F, maxntidy) == false)
520 if (llvm::getMaxNTIDz(F, maxntidz) == false)
526 O << ".maxntid " << maxntidx << ", " << maxntidy << ", " << maxntidz
530 if (llvm::getMinCTASm(F, mincta))
531 O << ".minnctapersm " << mincta << "\n";
535 NVPTXAsmPrinter::getVirtualRegisterName(unsigned Reg) const {
536 const TargetRegisterClass *RC = MRI->getRegClass(Reg);
539 raw_string_ostream NameStr(Name);
541 VRegRCMap::const_iterator I = VRegMapping.find(RC);
542 assert(I != VRegMapping.end() && "Bad register class");
543 const DenseMap<unsigned, unsigned> &RegMap = I->second;
545 VRegMap::const_iterator VI = RegMap.find(Reg);
546 assert(VI != RegMap.end() && "Bad virtual register");
547 unsigned MappedVR = VI->second;
549 NameStr << getNVPTXRegClassStr(RC) << MappedVR;
555 void NVPTXAsmPrinter::emitVirtualRegister(unsigned int vr,
557 O << getVirtualRegisterName(vr);
560 void NVPTXAsmPrinter::printVecModifiedImmediate(
561 const MachineOperand &MO, const char *Modifier, raw_ostream &O) {
562 static const char vecelem[] = { '0', '1', '2', '3', '0', '1', '2', '3' };
563 int Imm = (int) MO.getImm();
564 if (0 == strcmp(Modifier, "vecelem"))
565 O << "_" << vecelem[Imm];
566 else if (0 == strcmp(Modifier, "vecv4comm1")) {
567 if ((Imm < 0) || (Imm > 3))
569 } else if (0 == strcmp(Modifier, "vecv4comm2")) {
570 if ((Imm < 4) || (Imm > 7))
572 } else if (0 == strcmp(Modifier, "vecv4pos")) {
575 O << "_" << vecelem[Imm % 4];
576 } else if (0 == strcmp(Modifier, "vecv2comm1")) {
577 if ((Imm < 0) || (Imm > 1))
579 } else if (0 == strcmp(Modifier, "vecv2comm2")) {
580 if ((Imm < 2) || (Imm > 3))
582 } else if (0 == strcmp(Modifier, "vecv2pos")) {
585 O << "_" << vecelem[Imm % 2];
587 llvm_unreachable("Unknown Modifier on immediate operand");
592 void NVPTXAsmPrinter::emitDeclaration(const Function *F, raw_ostream &O) {
594 emitLinkageDirective(F, O);
595 if (llvm::isKernelFunction(*F))
599 printReturnValStr(F, O);
600 O << *getSymbol(F) << "\n";
601 emitFunctionParamList(F, O);
605 static bool usedInGlobalVarDef(const Constant *C) {
609 if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(C)) {
610 if (GV->getName().str() == "llvm.used")
615 for (const User *U : C->users())
616 if (const Constant *C = dyn_cast<Constant>(U))
617 if (usedInGlobalVarDef(C))
623 static bool usedInOneFunc(const User *U, Function const *&oneFunc) {
624 if (const GlobalVariable *othergv = dyn_cast<GlobalVariable>(U)) {
625 if (othergv->getName().str() == "llvm.used")
629 if (const Instruction *instr = dyn_cast<Instruction>(U)) {
630 if (instr->getParent() && instr->getParent()->getParent()) {
631 const Function *curFunc = instr->getParent()->getParent();
632 if (oneFunc && (curFunc != oneFunc))
640 for (const User *UU : U->users())
641 if (usedInOneFunc(UU, oneFunc) == false)
647 /* Find out if a global variable can be demoted to local scope.
648 * Currently, this is valid for CUDA shared variables, which have local
649 * scope and global lifetime. So the conditions to check are :
650 * 1. Is the global variable in shared address space?
651 * 2. Does it have internal linkage?
652 * 3. Is the global variable referenced only in one function?
654 static bool canDemoteGlobalVar(const GlobalVariable *gv, Function const *&f) {
655 if (gv->hasInternalLinkage() == false)
657 const PointerType *Pty = gv->getType();
658 if (Pty->getAddressSpace() != llvm::ADDRESS_SPACE_SHARED)
661 const Function *oneFunc = nullptr;
663 bool flag = usedInOneFunc(gv, oneFunc);
672 static bool useFuncSeen(const Constant *C,
673 llvm::DenseMap<const Function *, bool> &seenMap) {
674 for (const User *U : C->users()) {
675 if (const Constant *cu = dyn_cast<Constant>(U)) {
676 if (useFuncSeen(cu, seenMap))
678 } else if (const Instruction *I = dyn_cast<Instruction>(U)) {
679 const BasicBlock *bb = I->getParent();
682 const Function *caller = bb->getParent();
685 if (seenMap.find(caller) != seenMap.end())
692 void NVPTXAsmPrinter::emitDeclarations(const Module &M, raw_ostream &O) {
693 llvm::DenseMap<const Function *, bool> seenMap;
694 for (Module::const_iterator FI = M.begin(), FE = M.end(); FI != FE; ++FI) {
695 const Function *F = FI;
697 if (F->isDeclaration()) {
700 if (F->getIntrinsicID())
702 emitDeclaration(F, O);
705 for (const User *U : F->users()) {
706 if (const Constant *C = dyn_cast<Constant>(U)) {
707 if (usedInGlobalVarDef(C)) {
708 // The use is in the initialization of a global variable
709 // that is a function pointer, so print a declaration
710 // for the original function
711 emitDeclaration(F, O);
714 // Emit a declaration of this function if the function that
715 // uses this constant expr has already been seen.
716 if (useFuncSeen(C, seenMap)) {
717 emitDeclaration(F, O);
722 if (!isa<Instruction>(U))
724 const Instruction *instr = cast<Instruction>(U);
725 const BasicBlock *bb = instr->getParent();
728 const Function *caller = bb->getParent();
732 // If a caller has already been seen, then the caller is
733 // appearing in the module before the callee. so print out
734 // a declaration for the callee.
735 if (seenMap.find(caller) != seenMap.end()) {
736 emitDeclaration(F, O);
744 void NVPTXAsmPrinter::recordAndEmitFilenames(Module &M) {
745 DebugInfoFinder DbgFinder;
746 DbgFinder.processModule(M);
749 for (DICompileUnit DIUnit : DbgFinder.compile_units()) {
750 StringRef Filename(DIUnit.getFilename());
751 StringRef Dirname(DIUnit.getDirectory());
752 SmallString<128> FullPathName = Dirname;
753 if (!Dirname.empty() && !sys::path::is_absolute(Filename)) {
754 sys::path::append(FullPathName, Filename);
755 Filename = FullPathName.str();
757 if (filenameMap.find(Filename.str()) != filenameMap.end())
759 filenameMap[Filename.str()] = i;
760 OutStreamer.EmitDwarfFileDirective(i, "", Filename.str());
764 for (DISubprogram SP : DbgFinder.subprograms()) {
765 StringRef Filename(SP.getFilename());
766 StringRef Dirname(SP.getDirectory());
767 SmallString<128> FullPathName = Dirname;
768 if (!Dirname.empty() && !sys::path::is_absolute(Filename)) {
769 sys::path::append(FullPathName, Filename);
770 Filename = FullPathName.str();
772 if (filenameMap.find(Filename.str()) != filenameMap.end())
774 filenameMap[Filename.str()] = i;
779 bool NVPTXAsmPrinter::doInitialization(Module &M) {
781 SmallString<128> Str1;
782 raw_svector_ostream OS1(Str1);
784 MMI = getAnalysisIfAvailable<MachineModuleInfo>();
785 MMI->AnalyzeModule(M);
787 // We need to call the parent's one explicitly.
788 //bool Result = AsmPrinter::doInitialization(M);
790 // Initialize TargetLoweringObjectFile.
791 const_cast<TargetLoweringObjectFile &>(getObjFileLowering())
792 .Initialize(OutContext, TM);
794 Mang = new Mangler(TM.getSubtargetImpl()->getDataLayout());
796 // Emit header before any dwarf directives are emitted below.
798 OutStreamer.EmitRawText(OS1.str());
800 // Already commented out
801 //bool Result = AsmPrinter::doInitialization(M);
803 // Emit module-level inline asm if it exists.
804 if (!M.getModuleInlineAsm().empty()) {
805 OutStreamer.AddComment("Start of file scope inline assembly");
806 OutStreamer.AddBlankLine();
807 OutStreamer.EmitRawText(StringRef(M.getModuleInlineAsm()));
808 OutStreamer.AddBlankLine();
809 OutStreamer.AddComment("End of file scope inline assembly");
810 OutStreamer.AddBlankLine();
813 if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA)
814 recordAndEmitFilenames(M);
816 GlobalsEmitted = false;
818 return false; // success
821 void NVPTXAsmPrinter::emitGlobals(const Module &M) {
822 SmallString<128> Str2;
823 raw_svector_ostream OS2(Str2);
825 emitDeclarations(M, OS2);
827 // As ptxas does not support forward references of globals, we need to first
828 // sort the list of module-level globals in def-use order. We visit each
829 // global variable in order, and ensure that we emit it *after* its dependent
830 // globals. We use a little extra memory maintaining both a set and a list to
831 // have fast searches while maintaining a strict ordering.
832 SmallVector<const GlobalVariable *, 8> Globals;
833 DenseSet<const GlobalVariable *> GVVisited;
834 DenseSet<const GlobalVariable *> GVVisiting;
836 // Visit each global variable, in order
837 for (Module::const_global_iterator I = M.global_begin(), E = M.global_end();
839 VisitGlobalVariableForEmission(I, Globals, GVVisited, GVVisiting);
841 assert(GVVisited.size() == M.getGlobalList().size() &&
842 "Missed a global variable");
843 assert(GVVisiting.size() == 0 && "Did not fully process a global variable");
845 // Print out module-level global variables in proper order
846 for (unsigned i = 0, e = Globals.size(); i != e; ++i)
847 printModuleLevelGV(Globals[i], OS2);
851 OutStreamer.EmitRawText(OS2.str());
854 void NVPTXAsmPrinter::emitHeader(Module &M, raw_ostream &O) {
856 O << "// Generated by LLVM NVPTX Back-End\n";
860 unsigned PTXVersion = nvptxSubtarget.getPTXVersion();
861 O << ".version " << (PTXVersion / 10) << "." << (PTXVersion % 10) << "\n";
864 O << nvptxSubtarget.getTargetName();
866 if (nvptxSubtarget.getDrvInterface() == NVPTX::NVCL)
867 O << ", texmode_independent";
868 if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA) {
869 if (!nvptxSubtarget.hasDouble())
870 O << ", map_f64_to_f32";
873 if (MAI->doesSupportDebugInformation())
878 O << ".address_size ";
879 if (nvptxSubtarget.is64Bit())
888 bool NVPTXAsmPrinter::doFinalization(Module &M) {
890 // If we did not emit any functions, then the global declarations have not
892 if (!GlobalsEmitted) {
894 GlobalsEmitted = true;
897 // XXX Temproarily remove global variables so that doFinalization() will not
898 // emit them again (global variables are emitted at beginning).
900 Module::GlobalListType &global_list = M.getGlobalList();
901 int i, n = global_list.size();
902 GlobalVariable **gv_array = new GlobalVariable *[n];
904 // first, back-up GlobalVariable in gv_array
906 for (Module::global_iterator I = global_list.begin(), E = global_list.end();
910 // second, empty global_list
911 while (!global_list.empty())
912 global_list.remove(global_list.begin());
914 // call doFinalization
915 bool ret = AsmPrinter::doFinalization(M);
917 // now we restore global variables
918 for (i = 0; i < n; i++)
919 global_list.insert(global_list.end(), gv_array[i]);
921 clearAnnotationCache(&M);
926 //bool Result = AsmPrinter::doFinalization(M);
927 // Instead of calling the parents doFinalization, we may
928 // clone parents doFinalization and customize here.
929 // Currently, we if NVISA out the EmitGlobals() in
930 // parent's doFinalization, which is too intrusive.
932 // Same for the doInitialization.
936 // This function emits appropriate linkage directives for
937 // functions and global variables.
939 // extern function declaration -> .extern
940 // extern function definition -> .visible
941 // external global variable with init -> .visible
942 // external without init -> .extern
943 // appending -> not allowed, assert.
944 // for any linkage other than
945 // internal, private, linker_private,
946 // linker_private_weak, linker_private_weak_def_auto,
949 void NVPTXAsmPrinter::emitLinkageDirective(const GlobalValue *V,
951 if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA) {
952 if (V->hasExternalLinkage()) {
953 if (isa<GlobalVariable>(V)) {
954 const GlobalVariable *GVar = cast<GlobalVariable>(V);
956 if (GVar->hasInitializer())
961 } else if (V->isDeclaration())
965 } else if (V->hasAppendingLinkage()) {
967 msg.append("Error: ");
968 msg.append("Symbol ");
970 msg.append(V->getName().str());
971 msg.append("has unsupported appending linkage type");
972 llvm_unreachable(msg.c_str());
973 } else if (!V->hasInternalLinkage() &&
974 !V->hasPrivateLinkage()) {
980 void NVPTXAsmPrinter::printModuleLevelGV(const GlobalVariable *GVar,
982 bool processDemoted) {
985 if (GVar->hasSection()) {
986 if (GVar->getSection() == StringRef("llvm.metadata"))
990 // Skip LLVM intrinsic global variables
991 if (GVar->getName().startswith("llvm.") ||
992 GVar->getName().startswith("nvvm."))
995 const DataLayout *TD = TM.getSubtargetImpl()->getDataLayout();
997 // GlobalVariables are always constant pointers themselves.
998 const PointerType *PTy = GVar->getType();
999 Type *ETy = PTy->getElementType();
1001 if (GVar->hasExternalLinkage()) {
1002 if (GVar->hasInitializer())
1006 } else if (GVar->hasLinkOnceLinkage() || GVar->hasWeakLinkage() ||
1007 GVar->hasAvailableExternallyLinkage() ||
1008 GVar->hasCommonLinkage()) {
1012 if (llvm::isTexture(*GVar)) {
1013 O << ".global .texref " << llvm::getTextureName(*GVar) << ";\n";
1017 if (llvm::isSurface(*GVar)) {
1018 O << ".global .surfref " << llvm::getSurfaceName(*GVar) << ";\n";
1022 if (GVar->isDeclaration()) {
1023 // (extern) declarations, no definition or initializer
1024 // Currently the only known declaration is for an automatic __local
1025 // (.shared) promoted to global.
1026 emitPTXGlobalVariable(GVar, O);
1031 if (llvm::isSampler(*GVar)) {
1032 O << ".global .samplerref " << llvm::getSamplerName(*GVar);
1034 const Constant *Initializer = nullptr;
1035 if (GVar->hasInitializer())
1036 Initializer = GVar->getInitializer();
1037 const ConstantInt *CI = nullptr;
1039 CI = dyn_cast<ConstantInt>(Initializer);
1041 unsigned sample = CI->getZExtValue();
1046 addr = ((sample & __CLK_ADDRESS_MASK) >> __CLK_ADDRESS_BASE);
1048 O << "addr_mode_" << i << " = ";
1054 O << "clamp_to_border";
1057 O << "clamp_to_edge";
1068 O << "filter_mode = ";
1069 switch ((sample & __CLK_FILTER_MASK) >> __CLK_FILTER_BASE) {
1077 llvm_unreachable("Anisotropic filtering is not supported");
1082 if (!((sample & __CLK_NORMALIZED_MASK) >> __CLK_NORMALIZED_BASE)) {
1083 O << ", force_unnormalized_coords = 1";
1092 if (GVar->hasPrivateLinkage()) {
1094 if (!strncmp(GVar->getName().data(), "unrollpragma", 12))
1097 // FIXME - need better way (e.g. Metadata) to avoid generating this global
1098 if (!strncmp(GVar->getName().data(), "filename", 8))
1100 if (GVar->use_empty())
1104 const Function *demotedFunc = nullptr;
1105 if (!processDemoted && canDemoteGlobalVar(GVar, demotedFunc)) {
1106 O << "// " << GVar->getName().str() << " has been demoted\n";
1107 if (localDecls.find(demotedFunc) != localDecls.end())
1108 localDecls[demotedFunc].push_back(GVar);
1110 std::vector<const GlobalVariable *> temp;
1111 temp.push_back(GVar);
1112 localDecls[demotedFunc] = temp;
1118 emitPTXAddressSpace(PTy->getAddressSpace(), O);
1120 if (isManaged(*GVar)) {
1121 O << " .attribute(.managed)";
1124 if (GVar->getAlignment() == 0)
1125 O << " .align " << (int) TD->getPrefTypeAlignment(ETy);
1127 O << " .align " << GVar->getAlignment();
1129 if (ETy->isFloatingPointTy() || ETy->isIntegerTy() || ETy->isPointerTy()) {
1131 // Special case: ABI requires that we use .u8 for predicates
1132 if (ETy->isIntegerTy(1))
1135 O << getPTXFundamentalTypeStr(ETy, false);
1137 O << *getSymbol(GVar);
1139 // Ptx allows variable initilization only for constant and global state
1141 if (GVar->hasInitializer()) {
1142 if ((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
1143 (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST)) {
1144 const Constant *Initializer = GVar->getInitializer();
1145 // 'undef' is treated as there is no value spefied.
1146 if (!Initializer->isNullValue() && !isa<UndefValue>(Initializer)) {
1148 printScalarConstant(Initializer, O);
1151 // The frontend adds zero-initializer to variables that don't have an
1152 // initial value, so skip warning for this case.
1153 if (!GVar->getInitializer()->isNullValue()) {
1154 std::string warnMsg = "initial value of '" + GVar->getName().str() +
1155 "' is not allowed in addrspace(" +
1156 llvm::utostr_32(PTy->getAddressSpace()) + ")";
1157 report_fatal_error(warnMsg.c_str());
1162 unsigned int ElementSize = 0;
1164 // Although PTX has direct support for struct type and array type and
1165 // LLVM IR is very similar to PTX, the LLVM CodeGen does not support for
1166 // targets that support these high level field accesses. Structs, arrays
1167 // and vectors are lowered into arrays of bytes.
1168 switch (ETy->getTypeID()) {
1169 case Type::StructTyID:
1170 case Type::ArrayTyID:
1171 case Type::VectorTyID:
1172 ElementSize = TD->getTypeStoreSize(ETy);
1173 // Ptx allows variable initilization only for constant and
1174 // global state spaces.
1175 if (((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
1176 (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST)) &&
1177 GVar->hasInitializer()) {
1178 const Constant *Initializer = GVar->getInitializer();
1179 if (!isa<UndefValue>(Initializer) && !Initializer->isNullValue()) {
1180 AggBuffer aggBuffer(ElementSize, O, *this);
1181 bufferAggregateConstant(Initializer, &aggBuffer);
1182 if (aggBuffer.numSymbols) {
1183 if (nvptxSubtarget.is64Bit()) {
1184 O << " .u64 " << *getSymbol(GVar) << "[";
1185 O << ElementSize / 8;
1187 O << " .u32 " << *getSymbol(GVar) << "[";
1188 O << ElementSize / 4;
1192 O << " .b8 " << *getSymbol(GVar) << "[";
1200 O << " .b8 " << *getSymbol(GVar);
1208 O << " .b8 " << *getSymbol(GVar);
1217 llvm_unreachable("type not supported yet");
1224 void NVPTXAsmPrinter::emitDemotedVars(const Function *f, raw_ostream &O) {
1225 if (localDecls.find(f) == localDecls.end())
1228 std::vector<const GlobalVariable *> &gvars = localDecls[f];
1230 for (unsigned i = 0, e = gvars.size(); i != e; ++i) {
1231 O << "\t// demoted variable\n\t";
1232 printModuleLevelGV(gvars[i], O, true);
1236 void NVPTXAsmPrinter::emitPTXAddressSpace(unsigned int AddressSpace,
1237 raw_ostream &O) const {
1238 switch (AddressSpace) {
1239 case llvm::ADDRESS_SPACE_LOCAL:
1242 case llvm::ADDRESS_SPACE_GLOBAL:
1245 case llvm::ADDRESS_SPACE_CONST:
1248 case llvm::ADDRESS_SPACE_SHARED:
1252 report_fatal_error("Bad address space found while emitting PTX");
1258 NVPTXAsmPrinter::getPTXFundamentalTypeStr(const Type *Ty, bool useB4PTR) const {
1259 switch (Ty->getTypeID()) {
1261 llvm_unreachable("unexpected type");
1263 case Type::IntegerTyID: {
1264 unsigned NumBits = cast<IntegerType>(Ty)->getBitWidth();
1267 else if (NumBits <= 64) {
1268 std::string name = "u";
1269 return name + utostr(NumBits);
1271 llvm_unreachable("Integer too large");
1276 case Type::FloatTyID:
1278 case Type::DoubleTyID:
1280 case Type::PointerTyID:
1281 if (nvptxSubtarget.is64Bit())
1291 llvm_unreachable("unexpected type");
1295 void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable *GVar,
1298 const DataLayout *TD = TM.getSubtargetImpl()->getDataLayout();
1300 // GlobalVariables are always constant pointers themselves.
1301 const PointerType *PTy = GVar->getType();
1302 Type *ETy = PTy->getElementType();
1305 emitPTXAddressSpace(PTy->getAddressSpace(), O);
1306 if (GVar->getAlignment() == 0)
1307 O << " .align " << (int) TD->getPrefTypeAlignment(ETy);
1309 O << " .align " << GVar->getAlignment();
1311 if (ETy->isFloatingPointTy() || ETy->isIntegerTy() || ETy->isPointerTy()) {
1313 O << getPTXFundamentalTypeStr(ETy);
1315 O << *getSymbol(GVar);
1319 int64_t ElementSize = 0;
1321 // Although PTX has direct support for struct type and array type and LLVM IR
1322 // is very similar to PTX, the LLVM CodeGen does not support for targets that
1323 // support these high level field accesses. Structs and arrays are lowered
1324 // into arrays of bytes.
1325 switch (ETy->getTypeID()) {
1326 case Type::StructTyID:
1327 case Type::ArrayTyID:
1328 case Type::VectorTyID:
1329 ElementSize = TD->getTypeStoreSize(ETy);
1330 O << " .b8 " << *getSymbol(GVar) << "[";
1332 O << itostr(ElementSize);
1337 llvm_unreachable("type not supported yet");
1342 static unsigned int getOpenCLAlignment(const DataLayout *TD, Type *Ty) {
1343 if (Ty->isSingleValueType())
1344 return TD->getPrefTypeAlignment(Ty);
1346 const ArrayType *ATy = dyn_cast<ArrayType>(Ty);
1348 return getOpenCLAlignment(TD, ATy->getElementType());
1350 const StructType *STy = dyn_cast<StructType>(Ty);
1352 unsigned int alignStruct = 1;
1353 // Go through each element of the struct and find the
1354 // largest alignment.
1355 for (unsigned i = 0, e = STy->getNumElements(); i != e; i++) {
1356 Type *ETy = STy->getElementType(i);
1357 unsigned int align = getOpenCLAlignment(TD, ETy);
1358 if (align > alignStruct)
1359 alignStruct = align;
1364 const FunctionType *FTy = dyn_cast<FunctionType>(Ty);
1366 return TD->getPointerPrefAlignment();
1367 return TD->getPrefTypeAlignment(Ty);
1370 void NVPTXAsmPrinter::printParamName(Function::const_arg_iterator I,
1371 int paramIndex, raw_ostream &O) {
1372 if ((nvptxSubtarget.getDrvInterface() == NVPTX::NVCL) ||
1373 (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA))
1374 O << *getSymbol(I->getParent()) << "_param_" << paramIndex;
1376 std::string argName = I->getName();
1377 const char *p = argName.c_str();
1388 void NVPTXAsmPrinter::printParamName(int paramIndex, raw_ostream &O) {
1389 Function::const_arg_iterator I, E;
1392 if ((nvptxSubtarget.getDrvInterface() == NVPTX::NVCL) ||
1393 (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA)) {
1394 O << *CurrentFnSym << "_param_" << paramIndex;
1398 for (I = F->arg_begin(), E = F->arg_end(); I != E; ++I, i++) {
1399 if (i == paramIndex) {
1400 printParamName(I, paramIndex, O);
1404 llvm_unreachable("paramIndex out of bound");
1407 void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
1408 const DataLayout *TD = TM.getSubtargetImpl()->getDataLayout();
1409 const AttributeSet &PAL = F->getAttributes();
1410 const TargetLowering *TLI = TM.getSubtargetImpl()->getTargetLowering();
1411 Function::const_arg_iterator I, E;
1412 unsigned paramIndex = 0;
1414 bool isKernelFunc = llvm::isKernelFunction(*F);
1415 bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
1416 MVT thePointerTy = TLI->getPointerTy();
1420 for (I = F->arg_begin(), E = F->arg_end(); I != E; ++I, paramIndex++) {
1421 Type *Ty = I->getType();
1428 // Handle image/sampler parameters
1429 if (isKernelFunction(*F)) {
1430 if (isSampler(*I) || isImage(*I)) {
1432 std::string sname = I->getName();
1433 if (isImageWriteOnly(*I) || isImageReadWrite(*I)) {
1434 if (nvptxSubtarget.hasImageHandles())
1435 O << "\t.param .u64 .ptr .surfref ";
1437 O << "\t.param .surfref ";
1438 O << *CurrentFnSym << "_param_" << paramIndex;
1440 else { // Default image is read_only
1441 if (nvptxSubtarget.hasImageHandles())
1442 O << "\t.param .u64 .ptr .texref ";
1444 O << "\t.param .texref ";
1445 O << *CurrentFnSym << "_param_" << paramIndex;
1448 if (nvptxSubtarget.hasImageHandles())
1449 O << "\t.param .u64 .ptr .samplerref ";
1451 O << "\t.param .samplerref ";
1452 O << *CurrentFnSym << "_param_" << paramIndex;
1458 if (PAL.hasAttribute(paramIndex + 1, Attribute::ByVal) == false) {
1459 if (Ty->isAggregateType() || Ty->isVectorTy()) {
1460 // Just print .param .align <a> .b8 .param[size];
1461 // <a> = PAL.getparamalignment
1462 // size = typeallocsize of element type
1463 unsigned align = PAL.getParamAlignment(paramIndex + 1);
1465 align = TD->getABITypeAlignment(Ty);
1467 unsigned sz = TD->getTypeAllocSize(Ty);
1468 O << "\t.param .align " << align << " .b8 ";
1469 printParamName(I, paramIndex, O);
1470 O << "[" << sz << "]";
1475 const PointerType *PTy = dyn_cast<PointerType>(Ty);
1478 // Special handling for pointer arguments to kernel
1479 O << "\t.param .u" << thePointerTy.getSizeInBits() << " ";
1481 if (nvptxSubtarget.getDrvInterface() != NVPTX::CUDA) {
1482 Type *ETy = PTy->getElementType();
1483 int addrSpace = PTy->getAddressSpace();
1484 switch (addrSpace) {
1488 case llvm::ADDRESS_SPACE_CONST:
1489 O << ".ptr .const ";
1491 case llvm::ADDRESS_SPACE_SHARED:
1492 O << ".ptr .shared ";
1494 case llvm::ADDRESS_SPACE_GLOBAL:
1495 O << ".ptr .global ";
1498 O << ".align " << (int) getOpenCLAlignment(TD, ETy) << " ";
1500 printParamName(I, paramIndex, O);
1504 // non-pointer scalar to kernel func
1506 // Special case: predicate operands become .u8 types
1507 if (Ty->isIntegerTy(1))
1510 O << getPTXFundamentalTypeStr(Ty);
1512 printParamName(I, paramIndex, O);
1515 // Non-kernel function, just print .param .b<size> for ABI
1516 // and .reg .b<size> for non-ABI
1518 if (isa<IntegerType>(Ty)) {
1519 sz = cast<IntegerType>(Ty)->getBitWidth();
1522 } else if (isa<PointerType>(Ty))
1523 sz = thePointerTy.getSizeInBits();
1525 sz = Ty->getPrimitiveSizeInBits();
1527 O << "\t.param .b" << sz << " ";
1529 O << "\t.reg .b" << sz << " ";
1530 printParamName(I, paramIndex, O);
1534 // param has byVal attribute. So should be a pointer
1535 const PointerType *PTy = dyn_cast<PointerType>(Ty);
1536 assert(PTy && "Param with byval attribute should be a pointer type");
1537 Type *ETy = PTy->getElementType();
1539 if (isABI || isKernelFunc) {
1540 // Just print .param .align <a> .b8 .param[size];
1541 // <a> = PAL.getparamalignment
1542 // size = typeallocsize of element type
1543 unsigned align = PAL.getParamAlignment(paramIndex + 1);
1545 align = TD->getABITypeAlignment(ETy);
1547 unsigned sz = TD->getTypeAllocSize(ETy);
1548 O << "\t.param .align " << align << " .b8 ";
1549 printParamName(I, paramIndex, O);
1550 O << "[" << sz << "]";
1553 // Split the ETy into constituent parts and
1554 // print .param .b<size> <name> for each part.
1555 // Further, if a part is vector, print the above for
1556 // each vector element.
1557 SmallVector<EVT, 16> vtparts;
1558 ComputeValueVTs(*TLI, ETy, vtparts);
1559 for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
1561 EVT elemtype = vtparts[i];
1562 if (vtparts[i].isVector()) {
1563 elems = vtparts[i].getVectorNumElements();
1564 elemtype = vtparts[i].getVectorElementType();
1567 for (unsigned j = 0, je = elems; j != je; ++j) {
1568 unsigned sz = elemtype.getSizeInBits();
1569 if (elemtype.isInteger() && (sz < 32))
1571 O << "\t.reg .b" << sz << " ";
1572 printParamName(I, paramIndex, O);
1588 void NVPTXAsmPrinter::emitFunctionParamList(const MachineFunction &MF,
1590 const Function *F = MF.getFunction();
1591 emitFunctionParamList(F, O);
1594 void NVPTXAsmPrinter::setAndEmitFunctionVirtualRegisters(
1595 const MachineFunction &MF) {
1596 SmallString<128> Str;
1597 raw_svector_ostream O(Str);
1599 // Map the global virtual register number to a register class specific
1600 // virtual register number starting from 1 with that class.
1601 const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
1602 //unsigned numRegClasses = TRI->getNumRegClasses();
1604 // Emit the Fake Stack Object
1605 const MachineFrameInfo *MFI = MF.getFrameInfo();
1606 int NumBytes = (int) MFI->getStackSize();
1608 O << "\t.local .align " << MFI->getMaxAlignment() << " .b8 \t" << DEPOTNAME
1609 << getFunctionNumber() << "[" << NumBytes << "];\n";
1610 if (nvptxSubtarget.is64Bit()) {
1611 O << "\t.reg .b64 \t%SP;\n";
1612 O << "\t.reg .b64 \t%SPL;\n";
1614 O << "\t.reg .b32 \t%SP;\n";
1615 O << "\t.reg .b32 \t%SPL;\n";
1619 // Go through all virtual registers to establish the mapping between the
1621 // register number and the per class virtual register number.
1622 // We use the per class virtual register number in the ptx output.
1623 unsigned int numVRs = MRI->getNumVirtRegs();
1624 for (unsigned i = 0; i < numVRs; i++) {
1625 unsigned int vr = TRI->index2VirtReg(i);
1626 const TargetRegisterClass *RC = MRI->getRegClass(vr);
1627 DenseMap<unsigned, unsigned> ®map = VRegMapping[RC];
1628 int n = regmap.size();
1629 regmap.insert(std::make_pair(vr, n + 1));
1632 // Emit register declarations
1633 // @TODO: Extract out the real register usage
1634 // O << "\t.reg .pred %p<" << NVPTXNumRegisters << ">;\n";
1635 // O << "\t.reg .s16 %rc<" << NVPTXNumRegisters << ">;\n";
1636 // O << "\t.reg .s16 %rs<" << NVPTXNumRegisters << ">;\n";
1637 // O << "\t.reg .s32 %r<" << NVPTXNumRegisters << ">;\n";
1638 // O << "\t.reg .s64 %rd<" << NVPTXNumRegisters << ">;\n";
1639 // O << "\t.reg .f32 %f<" << NVPTXNumRegisters << ">;\n";
1640 // O << "\t.reg .f64 %fd<" << NVPTXNumRegisters << ">;\n";
1642 // Emit declaration of the virtual registers or 'physical' registers for
1643 // each register class
1644 for (unsigned i=0; i< TRI->getNumRegClasses(); i++) {
1645 const TargetRegisterClass *RC = TRI->getRegClass(i);
1646 DenseMap<unsigned, unsigned> ®map = VRegMapping[RC];
1647 std::string rcname = getNVPTXRegClassName(RC);
1648 std::string rcStr = getNVPTXRegClassStr(RC);
1649 int n = regmap.size();
1651 // Only declare those registers that may be used.
1653 O << "\t.reg " << rcname << " \t" << rcStr << "<" << (n+1)
1658 OutStreamer.EmitRawText(O.str());
1661 void NVPTXAsmPrinter::printFPConstant(const ConstantFP *Fp, raw_ostream &O) {
1662 APFloat APF = APFloat(Fp->getValueAPF()); // make a copy
1664 unsigned int numHex;
1667 if (Fp->getType()->getTypeID() == Type::FloatTyID) {
1670 APF.convert(APFloat::IEEEsingle, APFloat::rmNearestTiesToEven, &ignored);
1671 } else if (Fp->getType()->getTypeID() == Type::DoubleTyID) {
1674 APF.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven, &ignored);
1676 llvm_unreachable("unsupported fp type");
1678 APInt API = APF.bitcastToAPInt();
1679 std::string hexstr(utohexstr(API.getZExtValue()));
1681 if (hexstr.length() < numHex)
1682 O << std::string(numHex - hexstr.length(), '0');
1683 O << utohexstr(API.getZExtValue());
1686 void NVPTXAsmPrinter::printScalarConstant(const Constant *CPV, raw_ostream &O) {
1687 if (const ConstantInt *CI = dyn_cast<ConstantInt>(CPV)) {
1688 O << CI->getValue();
1691 if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CPV)) {
1692 printFPConstant(CFP, O);
1695 if (isa<ConstantPointerNull>(CPV)) {
1699 if (const GlobalValue *GVar = dyn_cast<GlobalValue>(CPV)) {
1700 PointerType *PTy = dyn_cast<PointerType>(GVar->getType());
1701 bool IsNonGenericPointer = false;
1702 if (PTy && PTy->getAddressSpace() != 0) {
1703 IsNonGenericPointer = true;
1705 if (EmitGeneric && !isa<Function>(CPV) && !IsNonGenericPointer) {
1707 O << *getSymbol(GVar);
1710 O << *getSymbol(GVar);
1714 if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1715 const Value *v = Cexpr->stripPointerCasts();
1716 PointerType *PTy = dyn_cast<PointerType>(Cexpr->getType());
1717 bool IsNonGenericPointer = false;
1718 if (PTy && PTy->getAddressSpace() != 0) {
1719 IsNonGenericPointer = true;
1721 if (const GlobalValue *GVar = dyn_cast<GlobalValue>(v)) {
1722 if (EmitGeneric && !isa<Function>(v) && !IsNonGenericPointer) {
1724 O << *getSymbol(GVar);
1727 O << *getSymbol(GVar);
1731 O << *lowerConstant(CPV);
1735 llvm_unreachable("Not scalar type found in printScalarConstant()");
1738 void NVPTXAsmPrinter::bufferLEByte(const Constant *CPV, int Bytes,
1739 AggBuffer *aggBuffer) {
1741 const DataLayout *TD = TM.getSubtargetImpl()->getDataLayout();
1743 if (isa<UndefValue>(CPV) || CPV->isNullValue()) {
1744 int s = TD->getTypeAllocSize(CPV->getType());
1747 aggBuffer->addZeros(s);
1752 switch (CPV->getType()->getTypeID()) {
1754 case Type::IntegerTyID: {
1755 const Type *ETy = CPV->getType();
1756 if (ETy == Type::getInt8Ty(CPV->getContext())) {
1758 (unsigned char)(dyn_cast<ConstantInt>(CPV))->getZExtValue();
1760 aggBuffer->addBytes(ptr, 1, Bytes);
1761 } else if (ETy == Type::getInt16Ty(CPV->getContext())) {
1762 short int16 = (short)(dyn_cast<ConstantInt>(CPV))->getZExtValue();
1763 ptr = (unsigned char *)&int16;
1764 aggBuffer->addBytes(ptr, 2, Bytes);
1765 } else if (ETy == Type::getInt32Ty(CPV->getContext())) {
1766 if (const ConstantInt *constInt = dyn_cast<ConstantInt>(CPV)) {
1767 int int32 = (int)(constInt->getZExtValue());
1768 ptr = (unsigned char *)&int32;
1769 aggBuffer->addBytes(ptr, 4, Bytes);
1771 } else if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1772 if (const ConstantInt *constInt = dyn_cast<ConstantInt>(
1773 ConstantFoldConstantExpression(Cexpr, TD))) {
1774 int int32 = (int)(constInt->getZExtValue());
1775 ptr = (unsigned char *)&int32;
1776 aggBuffer->addBytes(ptr, 4, Bytes);
1779 if (Cexpr->getOpcode() == Instruction::PtrToInt) {
1780 Value *v = Cexpr->getOperand(0)->stripPointerCasts();
1781 aggBuffer->addSymbol(v);
1782 aggBuffer->addZeros(4);
1786 llvm_unreachable("unsupported integer const type");
1787 } else if (ETy == Type::getInt64Ty(CPV->getContext())) {
1788 if (const ConstantInt *constInt = dyn_cast<ConstantInt>(CPV)) {
1789 long long int64 = (long long)(constInt->getZExtValue());
1790 ptr = (unsigned char *)&int64;
1791 aggBuffer->addBytes(ptr, 8, Bytes);
1793 } else if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1794 if (const ConstantInt *constInt = dyn_cast<ConstantInt>(
1795 ConstantFoldConstantExpression(Cexpr, TD))) {
1796 long long int64 = (long long)(constInt->getZExtValue());
1797 ptr = (unsigned char *)&int64;
1798 aggBuffer->addBytes(ptr, 8, Bytes);
1801 if (Cexpr->getOpcode() == Instruction::PtrToInt) {
1802 Value *v = Cexpr->getOperand(0)->stripPointerCasts();
1803 aggBuffer->addSymbol(v);
1804 aggBuffer->addZeros(8);
1808 llvm_unreachable("unsupported integer const type");
1810 llvm_unreachable("unsupported integer const type");
1813 case Type::FloatTyID:
1814 case Type::DoubleTyID: {
1815 const ConstantFP *CFP = dyn_cast<ConstantFP>(CPV);
1816 const Type *Ty = CFP->getType();
1817 if (Ty == Type::getFloatTy(CPV->getContext())) {
1818 float float32 = (float) CFP->getValueAPF().convertToFloat();
1819 ptr = (unsigned char *)&float32;
1820 aggBuffer->addBytes(ptr, 4, Bytes);
1821 } else if (Ty == Type::getDoubleTy(CPV->getContext())) {
1822 double float64 = CFP->getValueAPF().convertToDouble();
1823 ptr = (unsigned char *)&float64;
1824 aggBuffer->addBytes(ptr, 8, Bytes);
1826 llvm_unreachable("unsupported fp const type");
1830 case Type::PointerTyID: {
1831 if (const GlobalValue *GVar = dyn_cast<GlobalValue>(CPV)) {
1832 aggBuffer->addSymbol(GVar);
1833 } else if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1834 const Value *v = Cexpr->stripPointerCasts();
1835 aggBuffer->addSymbol(v);
1837 unsigned int s = TD->getTypeAllocSize(CPV->getType());
1838 aggBuffer->addZeros(s);
1842 case Type::ArrayTyID:
1843 case Type::VectorTyID:
1844 case Type::StructTyID: {
1845 if (isa<ConstantArray>(CPV) || isa<ConstantVector>(CPV) ||
1846 isa<ConstantStruct>(CPV) || isa<ConstantDataSequential>(CPV)) {
1847 int ElementSize = TD->getTypeAllocSize(CPV->getType());
1848 bufferAggregateConstant(CPV, aggBuffer);
1849 if (Bytes > ElementSize)
1850 aggBuffer->addZeros(Bytes - ElementSize);
1851 } else if (isa<ConstantAggregateZero>(CPV))
1852 aggBuffer->addZeros(Bytes);
1854 llvm_unreachable("Unexpected Constant type");
1859 llvm_unreachable("unsupported type");
1863 void NVPTXAsmPrinter::bufferAggregateConstant(const Constant *CPV,
1864 AggBuffer *aggBuffer) {
1865 const DataLayout *TD = TM.getSubtargetImpl()->getDataLayout();
1869 if (isa<ConstantArray>(CPV) || isa<ConstantVector>(CPV)) {
1870 if (CPV->getNumOperands())
1871 for (unsigned i = 0, e = CPV->getNumOperands(); i != e; ++i)
1872 bufferLEByte(cast<Constant>(CPV->getOperand(i)), 0, aggBuffer);
1876 if (const ConstantDataSequential *CDS =
1877 dyn_cast<ConstantDataSequential>(CPV)) {
1878 if (CDS->getNumElements())
1879 for (unsigned i = 0; i < CDS->getNumElements(); ++i)
1880 bufferLEByte(cast<Constant>(CDS->getElementAsConstant(i)), 0,
1885 if (isa<ConstantStruct>(CPV)) {
1886 if (CPV->getNumOperands()) {
1887 StructType *ST = cast<StructType>(CPV->getType());
1888 for (unsigned i = 0, e = CPV->getNumOperands(); i != e; ++i) {
1890 Bytes = TD->getStructLayout(ST)->getElementOffset(0) +
1891 TD->getTypeAllocSize(ST) -
1892 TD->getStructLayout(ST)->getElementOffset(i);
1894 Bytes = TD->getStructLayout(ST)->getElementOffset(i + 1) -
1895 TD->getStructLayout(ST)->getElementOffset(i);
1896 bufferLEByte(cast<Constant>(CPV->getOperand(i)), Bytes, aggBuffer);
1901 llvm_unreachable("unsupported constant type in printAggregateConstant()");
1904 // buildTypeNameMap - Run through symbol table looking for type names.
1907 bool NVPTXAsmPrinter::isImageType(const Type *Ty) {
1909 std::map<const Type *, std::string>::iterator PI = TypeNameMap.find(Ty);
1911 if (PI != TypeNameMap.end() && (!PI->second.compare("struct._image1d_t") ||
1912 !PI->second.compare("struct._image2d_t") ||
1913 !PI->second.compare("struct._image3d_t")))
1920 bool NVPTXAsmPrinter::ignoreLoc(const MachineInstr &MI) {
1921 switch (MI.getOpcode()) {
1924 case NVPTX::CallArgBeginInst:
1925 case NVPTX::CallArgEndInst0:
1926 case NVPTX::CallArgEndInst1:
1927 case NVPTX::CallArgF32:
1928 case NVPTX::CallArgF64:
1929 case NVPTX::CallArgI16:
1930 case NVPTX::CallArgI32:
1931 case NVPTX::CallArgI32imm:
1932 case NVPTX::CallArgI64:
1933 case NVPTX::CallArgParam:
1934 case NVPTX::CallVoidInst:
1935 case NVPTX::CallVoidInstReg:
1936 case NVPTX::Callseq_End:
1937 case NVPTX::CallVoidInstReg64:
1938 case NVPTX::DeclareParamInst:
1939 case NVPTX::DeclareRetMemInst:
1940 case NVPTX::DeclareRetRegInst:
1941 case NVPTX::DeclareRetScalarInst:
1942 case NVPTX::DeclareScalarParamInst:
1943 case NVPTX::DeclareScalarRegInst:
1944 case NVPTX::StoreParamF32:
1945 case NVPTX::StoreParamF64:
1946 case NVPTX::StoreParamI16:
1947 case NVPTX::StoreParamI32:
1948 case NVPTX::StoreParamI64:
1949 case NVPTX::StoreParamI8:
1950 case NVPTX::StoreRetvalF32:
1951 case NVPTX::StoreRetvalF64:
1952 case NVPTX::StoreRetvalI16:
1953 case NVPTX::StoreRetvalI32:
1954 case NVPTX::StoreRetvalI64:
1955 case NVPTX::StoreRetvalI8:
1956 case NVPTX::LastCallArgF32:
1957 case NVPTX::LastCallArgF64:
1958 case NVPTX::LastCallArgI16:
1959 case NVPTX::LastCallArgI32:
1960 case NVPTX::LastCallArgI32imm:
1961 case NVPTX::LastCallArgI64:
1962 case NVPTX::LastCallArgParam:
1963 case NVPTX::LoadParamMemF32:
1964 case NVPTX::LoadParamMemF64:
1965 case NVPTX::LoadParamMemI16:
1966 case NVPTX::LoadParamMemI32:
1967 case NVPTX::LoadParamMemI64:
1968 case NVPTX::LoadParamMemI8:
1969 case NVPTX::PrototypeInst:
1970 case NVPTX::DBG_VALUE:
1976 /// PrintAsmOperand - Print out an operand for an inline asm expression.
1978 bool NVPTXAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
1979 unsigned AsmVariant,
1980 const char *ExtraCode, raw_ostream &O) {
1981 if (ExtraCode && ExtraCode[0]) {
1982 if (ExtraCode[1] != 0)
1983 return true; // Unknown modifier.
1985 switch (ExtraCode[0]) {
1987 // See if this is a generic print operand
1988 return AsmPrinter::PrintAsmOperand(MI, OpNo, AsmVariant, ExtraCode, O);
1994 printOperand(MI, OpNo, O);
1999 bool NVPTXAsmPrinter::PrintAsmMemoryOperand(
2000 const MachineInstr *MI, unsigned OpNo, unsigned AsmVariant,
2001 const char *ExtraCode, raw_ostream &O) {
2002 if (ExtraCode && ExtraCode[0])
2003 return true; // Unknown modifier
2006 printMemOperand(MI, OpNo, O);
2012 void NVPTXAsmPrinter::printOperand(const MachineInstr *MI, int opNum,
2013 raw_ostream &O, const char *Modifier) {
2014 const MachineOperand &MO = MI->getOperand(opNum);
2015 switch (MO.getType()) {
2016 case MachineOperand::MO_Register:
2017 if (TargetRegisterInfo::isPhysicalRegister(MO.getReg())) {
2018 if (MO.getReg() == NVPTX::VRDepot)
2019 O << DEPOTNAME << getFunctionNumber();
2021 O << NVPTXInstPrinter::getRegisterName(MO.getReg());
2023 emitVirtualRegister(MO.getReg(), O);
2027 case MachineOperand::MO_Immediate:
2030 else if (strstr(Modifier, "vec") == Modifier)
2031 printVecModifiedImmediate(MO, Modifier, O);
2034 "Don't know how to handle modifier on immediate operand");
2037 case MachineOperand::MO_FPImmediate:
2038 printFPConstant(MO.getFPImm(), O);
2041 case MachineOperand::MO_GlobalAddress:
2042 O << *getSymbol(MO.getGlobal());
2045 case MachineOperand::MO_MachineBasicBlock:
2046 O << *MO.getMBB()->getSymbol();
2050 llvm_unreachable("Operand type not supported.");
2054 void NVPTXAsmPrinter::printMemOperand(const MachineInstr *MI, int opNum,
2055 raw_ostream &O, const char *Modifier) {
2056 printOperand(MI, opNum, O);
2058 if (Modifier && !strcmp(Modifier, "add")) {
2060 printOperand(MI, opNum + 1, O);
2062 if (MI->getOperand(opNum + 1).isImm() &&
2063 MI->getOperand(opNum + 1).getImm() == 0)
2064 return; // don't print ',0' or '+0'
2066 printOperand(MI, opNum + 1, O);
2071 // Force static initialization.
2072 extern "C" void LLVMInitializeNVPTXBackendAsmPrinter() {
2073 RegisterAsmPrinter<NVPTXAsmPrinter> X(TheNVPTXTarget32);
2074 RegisterAsmPrinter<NVPTXAsmPrinter> Y(TheNVPTXTarget64);
2077 void NVPTXAsmPrinter::emitSrcInText(StringRef filename, unsigned line) {
2078 std::stringstream temp;
2079 LineReader *reader = this->getReader(filename.str());
2081 temp << filename.str();
2085 temp << reader->readLine(line);
2087 this->OutStreamer.EmitRawText(Twine(temp.str()));
2090 LineReader *NVPTXAsmPrinter::getReader(std::string filename) {
2092 reader = new LineReader(filename);
2095 if (reader->fileName() != filename) {
2097 reader = new LineReader(filename);
2103 std::string LineReader::readLine(unsigned lineNum) {
2104 if (lineNum < theCurLine) {
2106 fstr.seekg(0, std::ios::beg);
2108 while (theCurLine < lineNum) {
2109 fstr.getline(buff, 500);
2115 // Force static initialization.
2116 extern "C" void LLVMInitializeNVPTXAsmPrinter() {
2117 RegisterAsmPrinter<NVPTXAsmPrinter> X(TheNVPTXTarget32);
2118 RegisterAsmPrinter<NVPTXAsmPrinter> Y(TheNVPTXTarget64);