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 "MCTargetDesc/NVPTXMCAsmInfo.h"
18 #include "NVPTXInstrInfo.h"
19 #include "NVPTXMCExpr.h"
20 #include "NVPTXRegisterInfo.h"
21 #include "NVPTXTargetMachine.h"
22 #include "NVPTXUtilities.h"
23 #include "InstPrinter/NVPTXInstPrinter.h"
24 #include "cl_common_defines.h"
25 #include "llvm/ADT/StringExtras.h"
26 #include "llvm/Analysis/ConstantFolding.h"
27 #include "llvm/Assembly/Writer.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/DebugInfo.h"
33 #include "llvm/IR/DerivedTypes.h"
34 #include "llvm/IR/Function.h"
35 #include "llvm/IR/GlobalVariable.h"
36 #include "llvm/IR/Module.h"
37 #include "llvm/IR/Operator.h"
38 #include "llvm/MC/MCStreamer.h"
39 #include "llvm/MC/MCSymbol.h"
40 #include "llvm/Support/CommandLine.h"
41 #include "llvm/Support/ErrorHandling.h"
42 #include "llvm/Support/FormattedStream.h"
43 #include "llvm/Support/Path.h"
44 #include "llvm/Support/TargetRegistry.h"
45 #include "llvm/Support/TimeValue.h"
46 #include "llvm/Target/Mangler.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.count(GV))
92 report_fatal_error("Circular dependency found in global variable set");
94 // Start visiting this global
97 // Make sure we visit all dependents first
98 DenseSet<const GlobalVariable *> Others;
99 for (unsigned i = 0, e = GV->getNumOperands(); i != e; ++i)
100 DiscoverDependentGlobals(GV->getOperand(i), Others);
102 for (DenseSet<const GlobalVariable *>::iterator I = Others.begin(),
105 VisitGlobalVariableForEmission(*I, Order, Visited, Visiting);
107 // Now we can visit ourself
114 // @TODO: This is a copy from AsmPrinter.cpp. The function is static, so we
115 // cannot just link to the existing version.
116 /// LowerConstant - Lower the specified LLVM Constant to an MCExpr.
118 using namespace nvptx;
119 const MCExpr *nvptx::LowerConstant(const Constant *CV, AsmPrinter &AP) {
120 MCContext &Ctx = AP.OutContext;
122 if (CV->isNullValue() || isa<UndefValue>(CV))
123 return MCConstantExpr::Create(0, Ctx);
125 if (const ConstantInt *CI = dyn_cast<ConstantInt>(CV))
126 return MCConstantExpr::Create(CI->getZExtValue(), Ctx);
128 if (const GlobalValue *GV = dyn_cast<GlobalValue>(CV))
129 return MCSymbolRefExpr::Create(AP.getSymbol(GV), Ctx);
131 if (const BlockAddress *BA = dyn_cast<BlockAddress>(CV))
132 return MCSymbolRefExpr::Create(AP.GetBlockAddressSymbol(BA), Ctx);
134 const ConstantExpr *CE = dyn_cast<ConstantExpr>(CV);
136 llvm_unreachable("Unknown constant value to lower!");
138 switch (CE->getOpcode()) {
140 // If the code isn't optimized, there may be outstanding folding
141 // opportunities. Attempt to fold the expression using DataLayout as a
142 // last resort before giving up.
143 if (Constant *C = ConstantFoldConstantExpression(CE, AP.TM.getDataLayout()))
145 return LowerConstant(C, AP);
147 // Otherwise report the problem to the user.
150 raw_string_ostream OS(S);
151 OS << "Unsupported expression in static initializer: ";
152 WriteAsOperand(OS, CE, /*PrintType=*/ false,
153 !AP.MF ? 0 : AP.MF->getFunction()->getParent());
154 report_fatal_error(OS.str());
156 case Instruction::GetElementPtr: {
157 const DataLayout &TD = *AP.TM.getDataLayout();
158 // Generate a symbolic expression for the byte address
159 APInt OffsetAI(TD.getPointerSizeInBits(), 0);
160 cast<GEPOperator>(CE)->accumulateConstantOffset(TD, OffsetAI);
162 const MCExpr *Base = LowerConstant(CE->getOperand(0), AP);
166 int64_t Offset = OffsetAI.getSExtValue();
167 return MCBinaryExpr::CreateAdd(Base, MCConstantExpr::Create(Offset, Ctx),
171 case Instruction::Trunc:
172 // We emit the value and depend on the assembler to truncate the generated
173 // expression properly. This is important for differences between
174 // blockaddress labels. Since the two labels are in the same function, it
175 // is reasonable to treat their delta as a 32-bit value.
177 case Instruction::BitCast:
178 return LowerConstant(CE->getOperand(0), AP);
180 case Instruction::IntToPtr: {
181 const DataLayout &TD = *AP.TM.getDataLayout();
182 // Handle casts to pointers by changing them into casts to the appropriate
183 // integer type. This promotes constant folding and simplifies this code.
184 Constant *Op = CE->getOperand(0);
185 Op = ConstantExpr::getIntegerCast(Op, TD.getIntPtrType(CV->getContext()),
187 return LowerConstant(Op, AP);
190 case Instruction::PtrToInt: {
191 const DataLayout &TD = *AP.TM.getDataLayout();
192 // Support only foldable casts to/from pointers that can be eliminated by
193 // changing the pointer to the appropriately sized integer type.
194 Constant *Op = CE->getOperand(0);
195 Type *Ty = CE->getType();
197 const MCExpr *OpExpr = LowerConstant(Op, AP);
199 // We can emit the pointer value into this slot if the slot is an
200 // integer slot equal to the size of the pointer.
201 if (TD.getTypeAllocSize(Ty) == TD.getTypeAllocSize(Op->getType()))
204 // Otherwise the pointer is smaller than the resultant integer, mask off
205 // the high bits so we are sure to get a proper truncation if the input is
207 unsigned InBits = TD.getTypeAllocSizeInBits(Op->getType());
208 const MCExpr *MaskExpr =
209 MCConstantExpr::Create(~0ULL >> (64 - InBits), Ctx);
210 return MCBinaryExpr::CreateAnd(OpExpr, MaskExpr, Ctx);
213 // The MC library also has a right-shift operator, but it isn't consistently
214 // signed or unsigned between different targets.
215 case Instruction::Add:
216 case Instruction::Sub:
217 case Instruction::Mul:
218 case Instruction::SDiv:
219 case Instruction::SRem:
220 case Instruction::Shl:
221 case Instruction::And:
222 case Instruction::Or:
223 case Instruction::Xor: {
224 const MCExpr *LHS = LowerConstant(CE->getOperand(0), AP);
225 const MCExpr *RHS = LowerConstant(CE->getOperand(1), AP);
226 switch (CE->getOpcode()) {
228 llvm_unreachable("Unknown binary operator constant cast expr");
229 case Instruction::Add:
230 return MCBinaryExpr::CreateAdd(LHS, RHS, Ctx);
231 case Instruction::Sub:
232 return MCBinaryExpr::CreateSub(LHS, RHS, Ctx);
233 case Instruction::Mul:
234 return MCBinaryExpr::CreateMul(LHS, RHS, Ctx);
235 case Instruction::SDiv:
236 return MCBinaryExpr::CreateDiv(LHS, RHS, Ctx);
237 case Instruction::SRem:
238 return MCBinaryExpr::CreateMod(LHS, RHS, Ctx);
239 case Instruction::Shl:
240 return MCBinaryExpr::CreateShl(LHS, RHS, Ctx);
241 case Instruction::And:
242 return MCBinaryExpr::CreateAnd(LHS, RHS, Ctx);
243 case Instruction::Or:
244 return MCBinaryExpr::CreateOr(LHS, RHS, Ctx);
245 case Instruction::Xor:
246 return MCBinaryExpr::CreateXor(LHS, RHS, Ctx);
252 void NVPTXAsmPrinter::emitLineNumberAsDotLoc(const MachineInstr &MI) {
253 if (!EmitLineNumbers)
258 DebugLoc curLoc = MI.getDebugLoc();
260 if (prevDebugLoc.isUnknown() && curLoc.isUnknown())
263 if (prevDebugLoc == curLoc)
266 prevDebugLoc = curLoc;
268 if (curLoc.isUnknown())
271 const MachineFunction *MF = MI.getParent()->getParent();
272 //const TargetMachine &TM = MF->getTarget();
274 const LLVMContext &ctx = MF->getFunction()->getContext();
275 DIScope Scope(curLoc.getScope(ctx));
277 assert((!Scope || Scope.isScope()) &&
278 "Scope of a DebugLoc should be null or a DIScope.");
282 StringRef fileName(Scope.getFilename());
283 StringRef dirName(Scope.getDirectory());
284 SmallString<128> FullPathName = dirName;
285 if (!dirName.empty() && !sys::path::is_absolute(fileName)) {
286 sys::path::append(FullPathName, fileName);
287 fileName = FullPathName.str();
290 if (filenameMap.find(fileName.str()) == filenameMap.end())
293 // Emit the line from the source file.
295 this->emitSrcInText(fileName.str(), curLoc.getLine());
297 std::stringstream temp;
298 temp << "\t.loc " << filenameMap[fileName.str()] << " " << curLoc.getLine()
299 << " " << curLoc.getCol();
300 OutStreamer.EmitRawText(Twine(temp.str().c_str()));
303 void NVPTXAsmPrinter::EmitInstruction(const MachineInstr *MI) {
304 SmallString<128> Str;
305 raw_svector_ostream OS(Str);
306 if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA)
307 emitLineNumberAsDotLoc(*MI);
310 lowerToMCInst(MI, Inst);
311 OutStreamer.EmitInstruction(Inst);
314 void NVPTXAsmPrinter::lowerToMCInst(const MachineInstr *MI, MCInst &OutMI) {
315 OutMI.setOpcode(MI->getOpcode());
317 for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
318 const MachineOperand &MO = MI->getOperand(i);
321 if (lowerOperand(MO, MCOp))
322 OutMI.addOperand(MCOp);
326 bool NVPTXAsmPrinter::lowerOperand(const MachineOperand &MO,
328 switch (MO.getType()) {
329 default: llvm_unreachable("unknown operand type");
330 case MachineOperand::MO_Register:
331 MCOp = MCOperand::CreateReg(encodeVirtualRegister(MO.getReg()));
333 case MachineOperand::MO_Immediate:
334 MCOp = MCOperand::CreateImm(MO.getImm());
336 case MachineOperand::MO_MachineBasicBlock:
337 MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
338 MO.getMBB()->getSymbol(), OutContext));
340 case MachineOperand::MO_ExternalSymbol:
341 MCOp = GetSymbolRef(MO, GetExternalSymbolSymbol(MO.getSymbolName()));
343 case MachineOperand::MO_GlobalAddress:
344 MCOp = GetSymbolRef(MO, getSymbol(MO.getGlobal()));
346 case MachineOperand::MO_FPImmediate: {
347 const ConstantFP *Cnt = MO.getFPImm();
348 APFloat Val = Cnt->getValueAPF();
350 switch (Cnt->getType()->getTypeID()) {
351 default: report_fatal_error("Unsupported FP type"); break;
352 case Type::FloatTyID:
353 MCOp = MCOperand::CreateExpr(
354 NVPTXFloatMCExpr::CreateConstantFPSingle(Val, OutContext));
356 case Type::DoubleTyID:
357 MCOp = MCOperand::CreateExpr(
358 NVPTXFloatMCExpr::CreateConstantFPDouble(Val, OutContext));
367 unsigned NVPTXAsmPrinter::encodeVirtualRegister(unsigned Reg) {
368 if (TargetRegisterInfo::isVirtualRegister(Reg)) {
369 const TargetRegisterClass *RC = MRI->getRegClass(Reg);
371 DenseMap<unsigned, unsigned> &RegMap = VRegMapping[RC];
372 unsigned RegNum = RegMap[Reg];
374 // Encode the register class in the upper 4 bits
375 // Must be kept in sync with NVPTXInstPrinter::printRegName
377 if (RC == &NVPTX::Int1RegsRegClass) {
379 } else if (RC == &NVPTX::Int16RegsRegClass) {
381 } else if (RC == &NVPTX::Int32RegsRegClass) {
383 } else if (RC == &NVPTX::Int64RegsRegClass) {
385 } else if (RC == &NVPTX::Float32RegsRegClass) {
387 } else if (RC == &NVPTX::Float64RegsRegClass) {
390 report_fatal_error("Bad register class");
393 // Insert the vreg number
394 Ret |= (RegNum & 0x0FFFFFFF);
397 // Some special-use registers are actually physical registers.
398 // Encode this as the register class ID of 0 and the real register ID.
399 return Reg & 0x0FFFFFFF;
403 MCOperand NVPTXAsmPrinter::GetSymbolRef(const MachineOperand &MO,
404 const MCSymbol *Symbol) {
406 Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_None,
408 return MCOperand::CreateExpr(Expr);
411 void NVPTXAsmPrinter::printReturnValStr(const Function *F, raw_ostream &O) {
412 const DataLayout *TD = TM.getDataLayout();
413 const TargetLowering *TLI = TM.getTargetLowering();
415 Type *Ty = F->getReturnType();
417 bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
419 if (Ty->getTypeID() == Type::VoidTyID)
425 if (Ty->isPrimitiveType() || Ty->isIntegerTy()) {
427 if (const IntegerType *ITy = dyn_cast<IntegerType>(Ty)) {
428 size = ITy->getBitWidth();
432 assert(Ty->isFloatingPointTy() && "Floating point type expected here");
433 size = Ty->getPrimitiveSizeInBits();
436 O << ".param .b" << size << " func_retval0";
437 } else if (isa<PointerType>(Ty)) {
438 O << ".param .b" << TLI->getPointerTy().getSizeInBits()
441 if ((Ty->getTypeID() == Type::StructTyID) || isa<VectorType>(Ty)) {
442 SmallVector<EVT, 16> vtparts;
443 ComputeValueVTs(*TLI, Ty, vtparts);
444 unsigned totalsz = 0;
445 for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
447 EVT elemtype = vtparts[i];
448 if (vtparts[i].isVector()) {
449 elems = vtparts[i].getVectorNumElements();
450 elemtype = vtparts[i].getVectorElementType();
452 for (unsigned j = 0, je = elems; j != je; ++j) {
453 unsigned sz = elemtype.getSizeInBits();
454 if (elemtype.isInteger() && (sz < 8))
459 unsigned retAlignment = 0;
460 if (!llvm::getAlign(*F, 0, retAlignment))
461 retAlignment = TD->getABITypeAlignment(Ty);
462 O << ".param .align " << retAlignment << " .b8 func_retval0[" << totalsz
465 assert(false && "Unknown return type");
468 SmallVector<EVT, 16> vtparts;
469 ComputeValueVTs(*TLI, Ty, vtparts);
471 for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
473 EVT elemtype = vtparts[i];
474 if (vtparts[i].isVector()) {
475 elems = vtparts[i].getVectorNumElements();
476 elemtype = vtparts[i].getVectorElementType();
479 for (unsigned j = 0, je = elems; j != je; ++j) {
480 unsigned sz = elemtype.getSizeInBits();
481 if (elemtype.isInteger() && (sz < 32))
483 O << ".reg .b" << sz << " func_retval" << idx;
496 void NVPTXAsmPrinter::printReturnValStr(const MachineFunction &MF,
498 const Function *F = MF.getFunction();
499 printReturnValStr(F, O);
502 void NVPTXAsmPrinter::EmitFunctionEntryLabel() {
503 SmallString<128> Str;
504 raw_svector_ostream O(Str);
506 if (!GlobalsEmitted) {
507 emitGlobals(*MF->getFunction()->getParent());
508 GlobalsEmitted = true;
512 MRI = &MF->getRegInfo();
513 F = MF->getFunction();
514 emitLinkageDirective(F, O);
515 if (llvm::isKernelFunction(*F))
519 printReturnValStr(*MF, O);
524 emitFunctionParamList(*MF, O);
526 if (llvm::isKernelFunction(*F))
527 emitKernelFunctionDirectives(*F, O);
529 OutStreamer.EmitRawText(O.str());
531 prevDebugLoc = DebugLoc();
534 void NVPTXAsmPrinter::EmitFunctionBodyStart() {
536 OutStreamer.EmitRawText(StringRef("{\n"));
537 setAndEmitFunctionVirtualRegisters(*MF);
539 SmallString<128> Str;
540 raw_svector_ostream O(Str);
541 emitDemotedVars(MF->getFunction(), O);
542 OutStreamer.EmitRawText(O.str());
545 void NVPTXAsmPrinter::EmitFunctionBodyEnd() {
546 OutStreamer.EmitRawText(StringRef("}\n"));
550 void NVPTXAsmPrinter::emitImplicitDef(const MachineInstr *MI) const {
551 unsigned RegNo = MI->getOperand(0).getReg();
552 const TargetRegisterInfo *TRI = TM.getRegisterInfo();
553 if (TRI->isVirtualRegister(RegNo)) {
554 OutStreamer.AddComment(Twine("implicit-def: ") +
555 getVirtualRegisterName(RegNo));
557 OutStreamer.AddComment(Twine("implicit-def: ") +
558 TM.getRegisterInfo()->getName(RegNo));
560 OutStreamer.AddBlankLine();
563 void NVPTXAsmPrinter::emitKernelFunctionDirectives(const Function &F,
564 raw_ostream &O) const {
565 // If the NVVM IR has some of reqntid* specified, then output
566 // the reqntid directive, and set the unspecified ones to 1.
567 // If none of reqntid* is specified, don't output reqntid directive.
568 unsigned reqntidx, reqntidy, reqntidz;
569 bool specified = false;
570 if (llvm::getReqNTIDx(F, reqntidx) == false)
574 if (llvm::getReqNTIDy(F, reqntidy) == false)
578 if (llvm::getReqNTIDz(F, reqntidz) == false)
584 O << ".reqntid " << reqntidx << ", " << reqntidy << ", " << reqntidz
587 // If the NVVM IR has some of maxntid* specified, then output
588 // the maxntid directive, and set the unspecified ones to 1.
589 // If none of maxntid* is specified, don't output maxntid directive.
590 unsigned maxntidx, maxntidy, maxntidz;
592 if (llvm::getMaxNTIDx(F, maxntidx) == false)
596 if (llvm::getMaxNTIDy(F, maxntidy) == false)
600 if (llvm::getMaxNTIDz(F, maxntidz) == false)
606 O << ".maxntid " << maxntidx << ", " << maxntidy << ", " << maxntidz
610 if (llvm::getMinCTASm(F, mincta))
611 O << ".minnctapersm " << mincta << "\n";
615 NVPTXAsmPrinter::getVirtualRegisterName(unsigned Reg) const {
616 const TargetRegisterClass *RC = MRI->getRegClass(Reg);
619 raw_string_ostream NameStr(Name);
621 VRegRCMap::const_iterator I = VRegMapping.find(RC);
622 assert(I != VRegMapping.end() && "Bad register class");
623 const DenseMap<unsigned, unsigned> &RegMap = I->second;
625 VRegMap::const_iterator VI = RegMap.find(Reg);
626 assert(VI != RegMap.end() && "Bad virtual register");
627 unsigned MappedVR = VI->second;
629 NameStr << getNVPTXRegClassStr(RC) << MappedVR;
635 void NVPTXAsmPrinter::emitVirtualRegister(unsigned int vr,
637 O << getVirtualRegisterName(vr);
640 void NVPTXAsmPrinter::printVecModifiedImmediate(
641 const MachineOperand &MO, const char *Modifier, raw_ostream &O) {
642 static const char vecelem[] = { '0', '1', '2', '3', '0', '1', '2', '3' };
643 int Imm = (int) MO.getImm();
644 if (0 == strcmp(Modifier, "vecelem"))
645 O << "_" << vecelem[Imm];
646 else if (0 == strcmp(Modifier, "vecv4comm1")) {
647 if ((Imm < 0) || (Imm > 3))
649 } else if (0 == strcmp(Modifier, "vecv4comm2")) {
650 if ((Imm < 4) || (Imm > 7))
652 } else if (0 == strcmp(Modifier, "vecv4pos")) {
655 O << "_" << vecelem[Imm % 4];
656 } else if (0 == strcmp(Modifier, "vecv2comm1")) {
657 if ((Imm < 0) || (Imm > 1))
659 } else if (0 == strcmp(Modifier, "vecv2comm2")) {
660 if ((Imm < 2) || (Imm > 3))
662 } else if (0 == strcmp(Modifier, "vecv2pos")) {
665 O << "_" << vecelem[Imm % 2];
667 llvm_unreachable("Unknown Modifier on immediate operand");
672 void NVPTXAsmPrinter::emitDeclaration(const Function *F, raw_ostream &O) {
674 emitLinkageDirective(F, O);
675 if (llvm::isKernelFunction(*F))
679 printReturnValStr(F, O);
680 O << *getSymbol(F) << "\n";
681 emitFunctionParamList(F, O);
685 static bool usedInGlobalVarDef(const Constant *C) {
689 if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(C)) {
690 if (GV->getName().str() == "llvm.used")
695 for (Value::const_use_iterator ui = C->use_begin(), ue = C->use_end();
697 const Constant *C = dyn_cast<Constant>(*ui);
698 if (usedInGlobalVarDef(C))
704 static bool usedInOneFunc(const User *U, Function const *&oneFunc) {
705 if (const GlobalVariable *othergv = dyn_cast<GlobalVariable>(U)) {
706 if (othergv->getName().str() == "llvm.used")
710 if (const Instruction *instr = dyn_cast<Instruction>(U)) {
711 if (instr->getParent() && instr->getParent()->getParent()) {
712 const Function *curFunc = instr->getParent()->getParent();
713 if (oneFunc && (curFunc != oneFunc))
721 if (const MDNode *md = dyn_cast<MDNode>(U))
722 if (md->hasName() && ((md->getName().str() == "llvm.dbg.gv") ||
723 (md->getName().str() == "llvm.dbg.sp")))
726 for (User::const_use_iterator ui = U->use_begin(), ue = U->use_end();
728 if (usedInOneFunc(*ui, oneFunc) == false)
734 /* Find out if a global variable can be demoted to local scope.
735 * Currently, this is valid for CUDA shared variables, which have local
736 * scope and global lifetime. So the conditions to check are :
737 * 1. Is the global variable in shared address space?
738 * 2. Does it have internal linkage?
739 * 3. Is the global variable referenced only in one function?
741 static bool canDemoteGlobalVar(const GlobalVariable *gv, Function const *&f) {
742 if (gv->hasInternalLinkage() == false)
744 const PointerType *Pty = gv->getType();
745 if (Pty->getAddressSpace() != llvm::ADDRESS_SPACE_SHARED)
748 const Function *oneFunc = 0;
750 bool flag = usedInOneFunc(gv, oneFunc);
759 static bool useFuncSeen(const Constant *C,
760 llvm::DenseMap<const Function *, bool> &seenMap) {
761 for (Value::const_use_iterator ui = C->use_begin(), ue = C->use_end();
763 if (const Constant *cu = dyn_cast<Constant>(*ui)) {
764 if (useFuncSeen(cu, seenMap))
766 } else if (const Instruction *I = dyn_cast<Instruction>(*ui)) {
767 const BasicBlock *bb = I->getParent();
770 const Function *caller = bb->getParent();
773 if (seenMap.find(caller) != seenMap.end())
780 void NVPTXAsmPrinter::emitDeclarations(const Module &M, raw_ostream &O) {
781 llvm::DenseMap<const Function *, bool> seenMap;
782 for (Module::const_iterator FI = M.begin(), FE = M.end(); FI != FE; ++FI) {
783 const Function *F = FI;
785 if (F->isDeclaration()) {
788 if (F->getIntrinsicID())
790 emitDeclaration(F, O);
793 for (Value::const_use_iterator iter = F->use_begin(),
794 iterEnd = F->use_end();
795 iter != iterEnd; ++iter) {
796 if (const Constant *C = dyn_cast<Constant>(*iter)) {
797 if (usedInGlobalVarDef(C)) {
798 // The use is in the initialization of a global variable
799 // that is a function pointer, so print a declaration
800 // for the original function
801 emitDeclaration(F, O);
804 // Emit a declaration of this function if the function that
805 // uses this constant expr has already been seen.
806 if (useFuncSeen(C, seenMap)) {
807 emitDeclaration(F, O);
812 if (!isa<Instruction>(*iter))
814 const Instruction *instr = cast<Instruction>(*iter);
815 const BasicBlock *bb = instr->getParent();
818 const Function *caller = bb->getParent();
822 // If a caller has already been seen, then the caller is
823 // appearing in the module before the callee. so print out
824 // a declaration for the callee.
825 if (seenMap.find(caller) != seenMap.end()) {
826 emitDeclaration(F, O);
834 void NVPTXAsmPrinter::recordAndEmitFilenames(Module &M) {
835 DebugInfoFinder DbgFinder;
836 DbgFinder.processModule(M);
839 for (DebugInfoFinder::iterator I = DbgFinder.compile_unit_begin(),
840 E = DbgFinder.compile_unit_end();
842 DICompileUnit DIUnit(*I);
843 StringRef Filename(DIUnit.getFilename());
844 StringRef Dirname(DIUnit.getDirectory());
845 SmallString<128> FullPathName = Dirname;
846 if (!Dirname.empty() && !sys::path::is_absolute(Filename)) {
847 sys::path::append(FullPathName, Filename);
848 Filename = FullPathName.str();
850 if (filenameMap.find(Filename.str()) != filenameMap.end())
852 filenameMap[Filename.str()] = i;
853 OutStreamer.EmitDwarfFileDirective(i, "", Filename.str());
857 for (DebugInfoFinder::iterator I = DbgFinder.subprogram_begin(),
858 E = DbgFinder.subprogram_end();
861 StringRef Filename(SP.getFilename());
862 StringRef Dirname(SP.getDirectory());
863 SmallString<128> FullPathName = Dirname;
864 if (!Dirname.empty() && !sys::path::is_absolute(Filename)) {
865 sys::path::append(FullPathName, Filename);
866 Filename = FullPathName.str();
868 if (filenameMap.find(Filename.str()) != filenameMap.end())
870 filenameMap[Filename.str()] = i;
875 bool NVPTXAsmPrinter::doInitialization(Module &M) {
877 SmallString<128> Str1;
878 raw_svector_ostream OS1(Str1);
880 MMI = getAnalysisIfAvailable<MachineModuleInfo>();
881 MMI->AnalyzeModule(M);
883 // We need to call the parent's one explicitly.
884 //bool Result = AsmPrinter::doInitialization(M);
886 // Initialize TargetLoweringObjectFile.
887 const_cast<TargetLoweringObjectFile &>(getObjFileLowering())
888 .Initialize(OutContext, TM);
890 Mang = new Mangler(OutContext, &TM);
892 // Emit header before any dwarf directives are emitted below.
894 OutStreamer.EmitRawText(OS1.str());
896 // Already commented out
897 //bool Result = AsmPrinter::doInitialization(M);
899 // Emit module-level inline asm if it exists.
900 if (!M.getModuleInlineAsm().empty()) {
901 OutStreamer.AddComment("Start of file scope inline assembly");
902 OutStreamer.AddBlankLine();
903 OutStreamer.EmitRawText(StringRef(M.getModuleInlineAsm()));
904 OutStreamer.AddBlankLine();
905 OutStreamer.AddComment("End of file scope inline assembly");
906 OutStreamer.AddBlankLine();
909 if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA)
910 recordAndEmitFilenames(M);
912 GlobalsEmitted = false;
914 return false; // success
917 void NVPTXAsmPrinter::emitGlobals(const Module &M) {
918 SmallString<128> Str2;
919 raw_svector_ostream OS2(Str2);
921 emitDeclarations(M, OS2);
923 // As ptxas does not support forward references of globals, we need to first
924 // sort the list of module-level globals in def-use order. We visit each
925 // global variable in order, and ensure that we emit it *after* its dependent
926 // globals. We use a little extra memory maintaining both a set and a list to
927 // have fast searches while maintaining a strict ordering.
928 SmallVector<const GlobalVariable *, 8> Globals;
929 DenseSet<const GlobalVariable *> GVVisited;
930 DenseSet<const GlobalVariable *> GVVisiting;
932 // Visit each global variable, in order
933 for (Module::const_global_iterator I = M.global_begin(), E = M.global_end();
935 VisitGlobalVariableForEmission(I, Globals, GVVisited, GVVisiting);
937 assert(GVVisited.size() == M.getGlobalList().size() &&
938 "Missed a global variable");
939 assert(GVVisiting.size() == 0 && "Did not fully process a global variable");
941 // Print out module-level global variables in proper order
942 for (unsigned i = 0, e = Globals.size(); i != e; ++i)
943 printModuleLevelGV(Globals[i], OS2);
947 OutStreamer.EmitRawText(OS2.str());
950 void NVPTXAsmPrinter::emitHeader(Module &M, raw_ostream &O) {
952 O << "// Generated by LLVM NVPTX Back-End\n";
956 unsigned PTXVersion = nvptxSubtarget.getPTXVersion();
957 O << ".version " << (PTXVersion / 10) << "." << (PTXVersion % 10) << "\n";
960 O << nvptxSubtarget.getTargetName();
962 if (nvptxSubtarget.getDrvInterface() == NVPTX::NVCL)
963 O << ", texmode_independent";
964 if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA) {
965 if (!nvptxSubtarget.hasDouble())
966 O << ", map_f64_to_f32";
969 if (MAI->doesSupportDebugInformation())
974 O << ".address_size ";
975 if (nvptxSubtarget.is64Bit())
984 bool NVPTXAsmPrinter::doFinalization(Module &M) {
986 // If we did not emit any functions, then the global declarations have not
988 if (!GlobalsEmitted) {
990 GlobalsEmitted = true;
993 // XXX Temproarily remove global variables so that doFinalization() will not
994 // emit them again (global variables are emitted at beginning).
996 Module::GlobalListType &global_list = M.getGlobalList();
997 int i, n = global_list.size();
998 GlobalVariable **gv_array = new GlobalVariable *[n];
1000 // first, back-up GlobalVariable in gv_array
1002 for (Module::global_iterator I = global_list.begin(), E = global_list.end();
1004 gv_array[i++] = &*I;
1006 // second, empty global_list
1007 while (!global_list.empty())
1008 global_list.remove(global_list.begin());
1010 // call doFinalization
1011 bool ret = AsmPrinter::doFinalization(M);
1013 // now we restore global variables
1014 for (i = 0; i < n; i++)
1015 global_list.insert(global_list.end(), gv_array[i]);
1020 //bool Result = AsmPrinter::doFinalization(M);
1021 // Instead of calling the parents doFinalization, we may
1022 // clone parents doFinalization and customize here.
1023 // Currently, we if NVISA out the EmitGlobals() in
1024 // parent's doFinalization, which is too intrusive.
1026 // Same for the doInitialization.
1030 // This function emits appropriate linkage directives for
1031 // functions and global variables.
1033 // extern function declaration -> .extern
1034 // extern function definition -> .visible
1035 // external global variable with init -> .visible
1036 // external without init -> .extern
1037 // appending -> not allowed, assert.
1039 void NVPTXAsmPrinter::emitLinkageDirective(const GlobalValue *V,
1041 if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA) {
1042 if (V->hasExternalLinkage()) {
1043 if (isa<GlobalVariable>(V)) {
1044 const GlobalVariable *GVar = cast<GlobalVariable>(V);
1046 if (GVar->hasInitializer())
1051 } else if (V->isDeclaration())
1055 } else if (V->hasAppendingLinkage()) {
1057 msg.append("Error: ");
1058 msg.append("Symbol ");
1060 msg.append(V->getName().str());
1061 msg.append("has unsupported appending linkage type");
1062 llvm_unreachable(msg.c_str());
1067 void NVPTXAsmPrinter::printModuleLevelGV(const GlobalVariable *GVar,
1069 bool processDemoted) {
1072 if (GVar->hasSection()) {
1073 if (GVar->getSection() == "llvm.metadata")
1077 const DataLayout *TD = TM.getDataLayout();
1079 // GlobalVariables are always constant pointers themselves.
1080 const PointerType *PTy = GVar->getType();
1081 Type *ETy = PTy->getElementType();
1083 if (GVar->hasExternalLinkage()) {
1084 if (GVar->hasInitializer())
1090 if (llvm::isTexture(*GVar)) {
1091 O << ".global .texref " << llvm::getTextureName(*GVar) << ";\n";
1095 if (llvm::isSurface(*GVar)) {
1096 O << ".global .surfref " << llvm::getSurfaceName(*GVar) << ";\n";
1100 if (GVar->isDeclaration()) {
1101 // (extern) declarations, no definition or initializer
1102 // Currently the only known declaration is for an automatic __local
1103 // (.shared) promoted to global.
1104 emitPTXGlobalVariable(GVar, O);
1109 if (llvm::isSampler(*GVar)) {
1110 O << ".global .samplerref " << llvm::getSamplerName(*GVar);
1112 const Constant *Initializer = NULL;
1113 if (GVar->hasInitializer())
1114 Initializer = GVar->getInitializer();
1115 const ConstantInt *CI = NULL;
1117 CI = dyn_cast<ConstantInt>(Initializer);
1119 unsigned sample = CI->getZExtValue();
1124 addr = ((sample & __CLK_ADDRESS_MASK) >> __CLK_ADDRESS_BASE);
1126 O << "addr_mode_" << i << " = ";
1132 O << "clamp_to_border";
1135 O << "clamp_to_edge";
1146 O << "filter_mode = ";
1147 switch ((sample & __CLK_FILTER_MASK) >> __CLK_FILTER_BASE) {
1155 assert(0 && "Anisotropic filtering is not supported");
1160 if (!((sample & __CLK_NORMALIZED_MASK) >> __CLK_NORMALIZED_BASE)) {
1161 O << ", force_unnormalized_coords = 1";
1170 if (GVar->hasPrivateLinkage()) {
1172 if (!strncmp(GVar->getName().data(), "unrollpragma", 12))
1175 // FIXME - need better way (e.g. Metadata) to avoid generating this global
1176 if (!strncmp(GVar->getName().data(), "filename", 8))
1178 if (GVar->use_empty())
1182 const Function *demotedFunc = 0;
1183 if (!processDemoted && canDemoteGlobalVar(GVar, demotedFunc)) {
1184 O << "// " << GVar->getName().str() << " has been demoted\n";
1185 if (localDecls.find(demotedFunc) != localDecls.end())
1186 localDecls[demotedFunc].push_back(GVar);
1188 std::vector<const GlobalVariable *> temp;
1189 temp.push_back(GVar);
1190 localDecls[demotedFunc] = temp;
1196 emitPTXAddressSpace(PTy->getAddressSpace(), O);
1197 if (GVar->getAlignment() == 0)
1198 O << " .align " << (int) TD->getPrefTypeAlignment(ETy);
1200 O << " .align " << GVar->getAlignment();
1202 if (ETy->isPrimitiveType() || ETy->isIntegerTy() || isa<PointerType>(ETy)) {
1204 // Special case: ABI requires that we use .u8 for predicates
1205 if (ETy->isIntegerTy(1))
1208 O << getPTXFundamentalTypeStr(ETy, false);
1210 O << *getSymbol(GVar);
1212 // Ptx allows variable initilization only for constant and global state
1214 if (((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
1215 (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST)) &&
1216 GVar->hasInitializer()) {
1217 const Constant *Initializer = GVar->getInitializer();
1218 if (!Initializer->isNullValue()) {
1220 printScalarConstant(Initializer, O);
1224 unsigned int ElementSize = 0;
1226 // Although PTX has direct support for struct type and array type and
1227 // LLVM IR is very similar to PTX, the LLVM CodeGen does not support for
1228 // targets that support these high level field accesses. Structs, arrays
1229 // and vectors are lowered into arrays of bytes.
1230 switch (ETy->getTypeID()) {
1231 case Type::StructTyID:
1232 case Type::ArrayTyID:
1233 case Type::VectorTyID:
1234 ElementSize = TD->getTypeStoreSize(ETy);
1235 // Ptx allows variable initilization only for constant and
1236 // global state spaces.
1237 if (((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
1238 (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST)) &&
1239 GVar->hasInitializer()) {
1240 const Constant *Initializer = GVar->getInitializer();
1241 if (!isa<UndefValue>(Initializer) && !Initializer->isNullValue()) {
1242 AggBuffer aggBuffer(ElementSize, O, *this);
1243 bufferAggregateConstant(Initializer, &aggBuffer);
1244 if (aggBuffer.numSymbols) {
1245 if (nvptxSubtarget.is64Bit()) {
1246 O << " .u64 " << *getSymbol(GVar) << "[";
1247 O << ElementSize / 8;
1249 O << " .u32 " << *getSymbol(GVar) << "[";
1250 O << ElementSize / 4;
1254 O << " .b8 " << *getSymbol(GVar) << "[";
1262 O << " .b8 " << *getSymbol(GVar);
1270 O << " .b8 " << *getSymbol(GVar);
1279 assert(0 && "type not supported yet");
1286 void NVPTXAsmPrinter::emitDemotedVars(const Function *f, raw_ostream &O) {
1287 if (localDecls.find(f) == localDecls.end())
1290 std::vector<const GlobalVariable *> &gvars = localDecls[f];
1292 for (unsigned i = 0, e = gvars.size(); i != e; ++i) {
1293 O << "\t// demoted variable\n\t";
1294 printModuleLevelGV(gvars[i], O, true);
1298 void NVPTXAsmPrinter::emitPTXAddressSpace(unsigned int AddressSpace,
1299 raw_ostream &O) const {
1300 switch (AddressSpace) {
1301 case llvm::ADDRESS_SPACE_LOCAL:
1304 case llvm::ADDRESS_SPACE_GLOBAL:
1307 case llvm::ADDRESS_SPACE_CONST:
1310 case llvm::ADDRESS_SPACE_SHARED:
1314 report_fatal_error("Bad address space found while emitting PTX");
1320 NVPTXAsmPrinter::getPTXFundamentalTypeStr(const Type *Ty, bool useB4PTR) const {
1321 switch (Ty->getTypeID()) {
1323 llvm_unreachable("unexpected type");
1325 case Type::IntegerTyID: {
1326 unsigned NumBits = cast<IntegerType>(Ty)->getBitWidth();
1329 else if (NumBits <= 64) {
1330 std::string name = "u";
1331 return name + utostr(NumBits);
1333 llvm_unreachable("Integer too large");
1338 case Type::FloatTyID:
1340 case Type::DoubleTyID:
1342 case Type::PointerTyID:
1343 if (nvptxSubtarget.is64Bit())
1353 llvm_unreachable("unexpected type");
1357 void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable *GVar,
1360 const DataLayout *TD = TM.getDataLayout();
1362 // GlobalVariables are always constant pointers themselves.
1363 const PointerType *PTy = GVar->getType();
1364 Type *ETy = PTy->getElementType();
1367 emitPTXAddressSpace(PTy->getAddressSpace(), O);
1368 if (GVar->getAlignment() == 0)
1369 O << " .align " << (int) TD->getPrefTypeAlignment(ETy);
1371 O << " .align " << GVar->getAlignment();
1373 if (ETy->isPrimitiveType() || ETy->isIntegerTy() || isa<PointerType>(ETy)) {
1375 O << getPTXFundamentalTypeStr(ETy);
1377 O << *getSymbol(GVar);
1381 int64_t ElementSize = 0;
1383 // Although PTX has direct support for struct type and array type and LLVM IR
1384 // is very similar to PTX, the LLVM CodeGen does not support for targets that
1385 // support these high level field accesses. Structs and arrays are lowered
1386 // into arrays of bytes.
1387 switch (ETy->getTypeID()) {
1388 case Type::StructTyID:
1389 case Type::ArrayTyID:
1390 case Type::VectorTyID:
1391 ElementSize = TD->getTypeStoreSize(ETy);
1392 O << " .b8 " << *getSymbol(GVar) << "[";
1394 O << itostr(ElementSize);
1399 assert(0 && "type not supported yet");
1404 static unsigned int getOpenCLAlignment(const DataLayout *TD, Type *Ty) {
1405 if (Ty->isPrimitiveType() || Ty->isIntegerTy() || isa<PointerType>(Ty))
1406 return TD->getPrefTypeAlignment(Ty);
1408 const ArrayType *ATy = dyn_cast<ArrayType>(Ty);
1410 return getOpenCLAlignment(TD, ATy->getElementType());
1412 const VectorType *VTy = dyn_cast<VectorType>(Ty);
1414 Type *ETy = VTy->getElementType();
1415 unsigned int numE = VTy->getNumElements();
1416 unsigned int alignE = TD->getPrefTypeAlignment(ETy);
1420 return numE * alignE;
1423 const StructType *STy = dyn_cast<StructType>(Ty);
1425 unsigned int alignStruct = 1;
1426 // Go through each element of the struct and find the
1427 // largest alignment.
1428 for (unsigned i = 0, e = STy->getNumElements(); i != e; i++) {
1429 Type *ETy = STy->getElementType(i);
1430 unsigned int align = getOpenCLAlignment(TD, ETy);
1431 if (align > alignStruct)
1432 alignStruct = align;
1437 const FunctionType *FTy = dyn_cast<FunctionType>(Ty);
1439 return TD->getPointerPrefAlignment();
1440 return TD->getPrefTypeAlignment(Ty);
1443 void NVPTXAsmPrinter::printParamName(Function::const_arg_iterator I,
1444 int paramIndex, raw_ostream &O) {
1445 if ((nvptxSubtarget.getDrvInterface() == NVPTX::NVCL) ||
1446 (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA))
1447 O << *getSymbol(I->getParent()) << "_param_" << paramIndex;
1449 std::string argName = I->getName();
1450 const char *p = argName.c_str();
1461 void NVPTXAsmPrinter::printParamName(int paramIndex, raw_ostream &O) {
1462 Function::const_arg_iterator I, E;
1465 if ((nvptxSubtarget.getDrvInterface() == NVPTX::NVCL) ||
1466 (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA)) {
1467 O << *CurrentFnSym << "_param_" << paramIndex;
1471 for (I = F->arg_begin(), E = F->arg_end(); I != E; ++I, i++) {
1472 if (i == paramIndex) {
1473 printParamName(I, paramIndex, O);
1477 llvm_unreachable("paramIndex out of bound");
1480 void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
1481 const DataLayout *TD = TM.getDataLayout();
1482 const AttributeSet &PAL = F->getAttributes();
1483 const TargetLowering *TLI = TM.getTargetLowering();
1484 Function::const_arg_iterator I, E;
1485 unsigned paramIndex = 0;
1487 bool isKernelFunc = llvm::isKernelFunction(*F);
1488 bool isABI = (nvptxSubtarget.getSmVersion() >= 20);
1489 MVT thePointerTy = TLI->getPointerTy();
1493 for (I = F->arg_begin(), E = F->arg_end(); I != E; ++I, paramIndex++) {
1494 Type *Ty = I->getType();
1501 // Handle image/sampler parameters
1502 if (llvm::isSampler(*I) || llvm::isImage(*I)) {
1503 if (llvm::isImage(*I)) {
1504 std::string sname = I->getName();
1505 if (llvm::isImageWriteOnly(*I))
1506 O << "\t.param .surfref " << *getSymbol(F) << "_param_"
1508 else // Default image is read_only
1509 O << "\t.param .texref " << *getSymbol(F) << "_param_"
1511 } else // Should be llvm::isSampler(*I)
1512 O << "\t.param .samplerref " << *getSymbol(F) << "_param_"
1517 if (PAL.hasAttribute(paramIndex + 1, Attribute::ByVal) == false) {
1518 if (Ty->isVectorTy()) {
1519 // Just print .param .b8 .align <a> .param[size];
1520 // <a> = PAL.getparamalignment
1521 // size = typeallocsize of element type
1522 unsigned align = PAL.getParamAlignment(paramIndex + 1);
1524 align = TD->getABITypeAlignment(Ty);
1526 unsigned sz = TD->getTypeAllocSize(Ty);
1527 O << "\t.param .align " << align << " .b8 ";
1528 printParamName(I, paramIndex, O);
1529 O << "[" << sz << "]";
1534 const PointerType *PTy = dyn_cast<PointerType>(Ty);
1537 // Special handling for pointer arguments to kernel
1538 O << "\t.param .u" << thePointerTy.getSizeInBits() << " ";
1540 if (nvptxSubtarget.getDrvInterface() != NVPTX::CUDA) {
1541 Type *ETy = PTy->getElementType();
1542 int addrSpace = PTy->getAddressSpace();
1543 switch (addrSpace) {
1547 case llvm::ADDRESS_SPACE_CONST:
1548 O << ".ptr .const ";
1550 case llvm::ADDRESS_SPACE_SHARED:
1551 O << ".ptr .shared ";
1553 case llvm::ADDRESS_SPACE_GLOBAL:
1554 O << ".ptr .global ";
1557 O << ".align " << (int) getOpenCLAlignment(TD, ETy) << " ";
1559 printParamName(I, paramIndex, O);
1563 // non-pointer scalar to kernel func
1565 // Special case: predicate operands become .u8 types
1566 if (Ty->isIntegerTy(1))
1569 O << getPTXFundamentalTypeStr(Ty);
1571 printParamName(I, paramIndex, O);
1574 // Non-kernel function, just print .param .b<size> for ABI
1575 // and .reg .b<size> for non ABY
1577 if (isa<IntegerType>(Ty)) {
1578 sz = cast<IntegerType>(Ty)->getBitWidth();
1581 } else if (isa<PointerType>(Ty))
1582 sz = thePointerTy.getSizeInBits();
1584 sz = Ty->getPrimitiveSizeInBits();
1586 O << "\t.param .b" << sz << " ";
1588 O << "\t.reg .b" << sz << " ";
1589 printParamName(I, paramIndex, O);
1593 // param has byVal attribute. So should be a pointer
1594 const PointerType *PTy = dyn_cast<PointerType>(Ty);
1595 assert(PTy && "Param with byval attribute should be a pointer type");
1596 Type *ETy = PTy->getElementType();
1598 if (isABI || isKernelFunc) {
1599 // Just print .param .b8 .align <a> .param[size];
1600 // <a> = PAL.getparamalignment
1601 // size = typeallocsize of element type
1602 unsigned align = PAL.getParamAlignment(paramIndex + 1);
1604 align = TD->getABITypeAlignment(ETy);
1606 unsigned sz = TD->getTypeAllocSize(ETy);
1607 O << "\t.param .align " << align << " .b8 ";
1608 printParamName(I, paramIndex, O);
1609 O << "[" << sz << "]";
1612 // Split the ETy into constituent parts and
1613 // print .param .b<size> <name> for each part.
1614 // Further, if a part is vector, print the above for
1615 // each vector element.
1616 SmallVector<EVT, 16> vtparts;
1617 ComputeValueVTs(*TLI, ETy, vtparts);
1618 for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
1620 EVT elemtype = vtparts[i];
1621 if (vtparts[i].isVector()) {
1622 elems = vtparts[i].getVectorNumElements();
1623 elemtype = vtparts[i].getVectorElementType();
1626 for (unsigned j = 0, je = elems; j != je; ++j) {
1627 unsigned sz = elemtype.getSizeInBits();
1628 if (elemtype.isInteger() && (sz < 32))
1630 O << "\t.reg .b" << sz << " ";
1631 printParamName(I, paramIndex, O);
1647 void NVPTXAsmPrinter::emitFunctionParamList(const MachineFunction &MF,
1649 const Function *F = MF.getFunction();
1650 emitFunctionParamList(F, O);
1653 void NVPTXAsmPrinter::setAndEmitFunctionVirtualRegisters(
1654 const MachineFunction &MF) {
1655 SmallString<128> Str;
1656 raw_svector_ostream O(Str);
1658 // Map the global virtual register number to a register class specific
1659 // virtual register number starting from 1 with that class.
1660 const TargetRegisterInfo *TRI = MF.getTarget().getRegisterInfo();
1661 //unsigned numRegClasses = TRI->getNumRegClasses();
1663 // Emit the Fake Stack Object
1664 const MachineFrameInfo *MFI = MF.getFrameInfo();
1665 int NumBytes = (int) MFI->getStackSize();
1667 O << "\t.local .align " << MFI->getMaxAlignment() << " .b8 \t" << DEPOTNAME
1668 << getFunctionNumber() << "[" << NumBytes << "];\n";
1669 if (nvptxSubtarget.is64Bit()) {
1670 O << "\t.reg .b64 \t%SP;\n";
1671 O << "\t.reg .b64 \t%SPL;\n";
1673 O << "\t.reg .b32 \t%SP;\n";
1674 O << "\t.reg .b32 \t%SPL;\n";
1678 // Go through all virtual registers to establish the mapping between the
1680 // register number and the per class virtual register number.
1681 // We use the per class virtual register number in the ptx output.
1682 unsigned int numVRs = MRI->getNumVirtRegs();
1683 for (unsigned i = 0; i < numVRs; i++) {
1684 unsigned int vr = TRI->index2VirtReg(i);
1685 const TargetRegisterClass *RC = MRI->getRegClass(vr);
1686 DenseMap<unsigned, unsigned> ®map = VRegMapping[RC];
1687 int n = regmap.size();
1688 regmap.insert(std::make_pair(vr, n + 1));
1691 // Emit register declarations
1692 // @TODO: Extract out the real register usage
1693 // O << "\t.reg .pred %p<" << NVPTXNumRegisters << ">;\n";
1694 // O << "\t.reg .s16 %rc<" << NVPTXNumRegisters << ">;\n";
1695 // O << "\t.reg .s16 %rs<" << NVPTXNumRegisters << ">;\n";
1696 // O << "\t.reg .s32 %r<" << NVPTXNumRegisters << ">;\n";
1697 // O << "\t.reg .s64 %rl<" << NVPTXNumRegisters << ">;\n";
1698 // O << "\t.reg .f32 %f<" << NVPTXNumRegisters << ">;\n";
1699 // O << "\t.reg .f64 %fl<" << NVPTXNumRegisters << ">;\n";
1701 // Emit declaration of the virtual registers or 'physical' registers for
1702 // each register class
1703 for (unsigned i=0; i< TRI->getNumRegClasses(); i++) {
1704 const TargetRegisterClass *RC = TRI->getRegClass(i);
1705 DenseMap<unsigned, unsigned> ®map = VRegMapping[RC];
1706 std::string rcname = getNVPTXRegClassName(RC);
1707 std::string rcStr = getNVPTXRegClassStr(RC);
1708 int n = regmap.size();
1710 // Only declare those registers that may be used.
1712 O << "\t.reg " << rcname << " \t" << rcStr << "<" << (n+1)
1717 OutStreamer.EmitRawText(O.str());
1720 void NVPTXAsmPrinter::printFPConstant(const ConstantFP *Fp, raw_ostream &O) {
1721 APFloat APF = APFloat(Fp->getValueAPF()); // make a copy
1723 unsigned int numHex;
1726 if (Fp->getType()->getTypeID() == Type::FloatTyID) {
1729 APF.convert(APFloat::IEEEsingle, APFloat::rmNearestTiesToEven, &ignored);
1730 } else if (Fp->getType()->getTypeID() == Type::DoubleTyID) {
1733 APF.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven, &ignored);
1735 llvm_unreachable("unsupported fp type");
1737 APInt API = APF.bitcastToAPInt();
1738 std::string hexstr(utohexstr(API.getZExtValue()));
1740 if (hexstr.length() < numHex)
1741 O << std::string(numHex - hexstr.length(), '0');
1742 O << utohexstr(API.getZExtValue());
1745 void NVPTXAsmPrinter::printScalarConstant(const Constant *CPV, raw_ostream &O) {
1746 if (const ConstantInt *CI = dyn_cast<ConstantInt>(CPV)) {
1747 O << CI->getValue();
1750 if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CPV)) {
1751 printFPConstant(CFP, O);
1754 if (isa<ConstantPointerNull>(CPV)) {
1758 if (const GlobalValue *GVar = dyn_cast<GlobalValue>(CPV)) {
1759 O << *getSymbol(GVar);
1762 if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1763 const Value *v = Cexpr->stripPointerCasts();
1764 if (const GlobalValue *GVar = dyn_cast<GlobalValue>(v)) {
1765 O << *getSymbol(GVar);
1768 O << *LowerConstant(CPV, *this);
1772 llvm_unreachable("Not scalar type found in printScalarConstant()");
1775 void NVPTXAsmPrinter::bufferLEByte(const Constant *CPV, int Bytes,
1776 AggBuffer *aggBuffer) {
1778 const DataLayout *TD = TM.getDataLayout();
1780 if (isa<UndefValue>(CPV) || CPV->isNullValue()) {
1781 int s = TD->getTypeAllocSize(CPV->getType());
1784 aggBuffer->addZeros(s);
1789 switch (CPV->getType()->getTypeID()) {
1791 case Type::IntegerTyID: {
1792 const Type *ETy = CPV->getType();
1793 if (ETy == Type::getInt8Ty(CPV->getContext())) {
1795 (unsigned char)(dyn_cast<ConstantInt>(CPV))->getZExtValue();
1797 aggBuffer->addBytes(ptr, 1, Bytes);
1798 } else if (ETy == Type::getInt16Ty(CPV->getContext())) {
1799 short int16 = (short)(dyn_cast<ConstantInt>(CPV))->getZExtValue();
1800 ptr = (unsigned char *)&int16;
1801 aggBuffer->addBytes(ptr, 2, Bytes);
1802 } else if (ETy == Type::getInt32Ty(CPV->getContext())) {
1803 if (const ConstantInt *constInt = dyn_cast<ConstantInt>(CPV)) {
1804 int int32 = (int)(constInt->getZExtValue());
1805 ptr = (unsigned char *)&int32;
1806 aggBuffer->addBytes(ptr, 4, Bytes);
1808 } else if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1809 if (const ConstantInt *constInt = dyn_cast<ConstantInt>(
1810 ConstantFoldConstantExpression(Cexpr, TD))) {
1811 int int32 = (int)(constInt->getZExtValue());
1812 ptr = (unsigned char *)&int32;
1813 aggBuffer->addBytes(ptr, 4, Bytes);
1816 if (Cexpr->getOpcode() == Instruction::PtrToInt) {
1817 Value *v = Cexpr->getOperand(0)->stripPointerCasts();
1818 aggBuffer->addSymbol(v);
1819 aggBuffer->addZeros(4);
1823 llvm_unreachable("unsupported integer const type");
1824 } else if (ETy == Type::getInt64Ty(CPV->getContext())) {
1825 if (const ConstantInt *constInt = dyn_cast<ConstantInt>(CPV)) {
1826 long long int64 = (long long)(constInt->getZExtValue());
1827 ptr = (unsigned char *)&int64;
1828 aggBuffer->addBytes(ptr, 8, Bytes);
1830 } else if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1831 if (const ConstantInt *constInt = dyn_cast<ConstantInt>(
1832 ConstantFoldConstantExpression(Cexpr, TD))) {
1833 long long int64 = (long long)(constInt->getZExtValue());
1834 ptr = (unsigned char *)&int64;
1835 aggBuffer->addBytes(ptr, 8, Bytes);
1838 if (Cexpr->getOpcode() == Instruction::PtrToInt) {
1839 Value *v = Cexpr->getOperand(0)->stripPointerCasts();
1840 aggBuffer->addSymbol(v);
1841 aggBuffer->addZeros(8);
1845 llvm_unreachable("unsupported integer const type");
1847 llvm_unreachable("unsupported integer const type");
1850 case Type::FloatTyID:
1851 case Type::DoubleTyID: {
1852 const ConstantFP *CFP = dyn_cast<ConstantFP>(CPV);
1853 const Type *Ty = CFP->getType();
1854 if (Ty == Type::getFloatTy(CPV->getContext())) {
1855 float float32 = (float) CFP->getValueAPF().convertToFloat();
1856 ptr = (unsigned char *)&float32;
1857 aggBuffer->addBytes(ptr, 4, Bytes);
1858 } else if (Ty == Type::getDoubleTy(CPV->getContext())) {
1859 double float64 = CFP->getValueAPF().convertToDouble();
1860 ptr = (unsigned char *)&float64;
1861 aggBuffer->addBytes(ptr, 8, Bytes);
1863 llvm_unreachable("unsupported fp const type");
1867 case Type::PointerTyID: {
1868 if (const GlobalValue *GVar = dyn_cast<GlobalValue>(CPV)) {
1869 aggBuffer->addSymbol(GVar);
1870 } else if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1871 const Value *v = Cexpr->stripPointerCasts();
1872 aggBuffer->addSymbol(v);
1874 unsigned int s = TD->getTypeAllocSize(CPV->getType());
1875 aggBuffer->addZeros(s);
1879 case Type::ArrayTyID:
1880 case Type::VectorTyID:
1881 case Type::StructTyID: {
1882 if (isa<ConstantArray>(CPV) || isa<ConstantVector>(CPV) ||
1883 isa<ConstantStruct>(CPV) || isa<ConstantDataSequential>(CPV)) {
1884 int ElementSize = TD->getTypeAllocSize(CPV->getType());
1885 bufferAggregateConstant(CPV, aggBuffer);
1886 if (Bytes > ElementSize)
1887 aggBuffer->addZeros(Bytes - ElementSize);
1888 } else if (isa<ConstantAggregateZero>(CPV))
1889 aggBuffer->addZeros(Bytes);
1891 llvm_unreachable("Unexpected Constant type");
1896 llvm_unreachable("unsupported type");
1900 void NVPTXAsmPrinter::bufferAggregateConstant(const Constant *CPV,
1901 AggBuffer *aggBuffer) {
1902 const DataLayout *TD = TM.getDataLayout();
1906 if (isa<ConstantArray>(CPV) || isa<ConstantVector>(CPV)) {
1907 if (CPV->getNumOperands())
1908 for (unsigned i = 0, e = CPV->getNumOperands(); i != e; ++i)
1909 bufferLEByte(cast<Constant>(CPV->getOperand(i)), 0, aggBuffer);
1913 if (const ConstantDataSequential *CDS =
1914 dyn_cast<ConstantDataSequential>(CPV)) {
1915 if (CDS->getNumElements())
1916 for (unsigned i = 0; i < CDS->getNumElements(); ++i)
1917 bufferLEByte(cast<Constant>(CDS->getElementAsConstant(i)), 0,
1922 if (isa<ConstantStruct>(CPV)) {
1923 if (CPV->getNumOperands()) {
1924 StructType *ST = cast<StructType>(CPV->getType());
1925 for (unsigned i = 0, e = CPV->getNumOperands(); i != e; ++i) {
1927 Bytes = TD->getStructLayout(ST)->getElementOffset(0) +
1928 TD->getTypeAllocSize(ST) -
1929 TD->getStructLayout(ST)->getElementOffset(i);
1931 Bytes = TD->getStructLayout(ST)->getElementOffset(i + 1) -
1932 TD->getStructLayout(ST)->getElementOffset(i);
1933 bufferLEByte(cast<Constant>(CPV->getOperand(i)), Bytes, aggBuffer);
1938 llvm_unreachable("unsupported constant type in printAggregateConstant()");
1941 // buildTypeNameMap - Run through symbol table looking for type names.
1944 bool NVPTXAsmPrinter::isImageType(const Type *Ty) {
1946 std::map<const Type *, std::string>::iterator PI = TypeNameMap.find(Ty);
1948 if (PI != TypeNameMap.end() && (!PI->second.compare("struct._image1d_t") ||
1949 !PI->second.compare("struct._image2d_t") ||
1950 !PI->second.compare("struct._image3d_t")))
1957 bool NVPTXAsmPrinter::ignoreLoc(const MachineInstr &MI) {
1958 switch (MI.getOpcode()) {
1961 case NVPTX::CallArgBeginInst:
1962 case NVPTX::CallArgEndInst0:
1963 case NVPTX::CallArgEndInst1:
1964 case NVPTX::CallArgF32:
1965 case NVPTX::CallArgF64:
1966 case NVPTX::CallArgI16:
1967 case NVPTX::CallArgI32:
1968 case NVPTX::CallArgI32imm:
1969 case NVPTX::CallArgI64:
1970 case NVPTX::CallArgParam:
1971 case NVPTX::CallVoidInst:
1972 case NVPTX::CallVoidInstReg:
1973 case NVPTX::Callseq_End:
1974 case NVPTX::CallVoidInstReg64:
1975 case NVPTX::DeclareParamInst:
1976 case NVPTX::DeclareRetMemInst:
1977 case NVPTX::DeclareRetRegInst:
1978 case NVPTX::DeclareRetScalarInst:
1979 case NVPTX::DeclareScalarParamInst:
1980 case NVPTX::DeclareScalarRegInst:
1981 case NVPTX::StoreParamF32:
1982 case NVPTX::StoreParamF64:
1983 case NVPTX::StoreParamI16:
1984 case NVPTX::StoreParamI32:
1985 case NVPTX::StoreParamI64:
1986 case NVPTX::StoreParamI8:
1987 case NVPTX::StoreRetvalF32:
1988 case NVPTX::StoreRetvalF64:
1989 case NVPTX::StoreRetvalI16:
1990 case NVPTX::StoreRetvalI32:
1991 case NVPTX::StoreRetvalI64:
1992 case NVPTX::StoreRetvalI8:
1993 case NVPTX::LastCallArgF32:
1994 case NVPTX::LastCallArgF64:
1995 case NVPTX::LastCallArgI16:
1996 case NVPTX::LastCallArgI32:
1997 case NVPTX::LastCallArgI32imm:
1998 case NVPTX::LastCallArgI64:
1999 case NVPTX::LastCallArgParam:
2000 case NVPTX::LoadParamMemF32:
2001 case NVPTX::LoadParamMemF64:
2002 case NVPTX::LoadParamMemI16:
2003 case NVPTX::LoadParamMemI32:
2004 case NVPTX::LoadParamMemI64:
2005 case NVPTX::LoadParamMemI8:
2006 case NVPTX::PrototypeInst:
2007 case NVPTX::DBG_VALUE:
2013 /// PrintAsmOperand - Print out an operand for an inline asm expression.
2015 bool NVPTXAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
2016 unsigned AsmVariant,
2017 const char *ExtraCode, raw_ostream &O) {
2018 if (ExtraCode && ExtraCode[0]) {
2019 if (ExtraCode[1] != 0)
2020 return true; // Unknown modifier.
2022 switch (ExtraCode[0]) {
2024 // See if this is a generic print operand
2025 return AsmPrinter::PrintAsmOperand(MI, OpNo, AsmVariant, ExtraCode, O);
2031 printOperand(MI, OpNo, O);
2036 bool NVPTXAsmPrinter::PrintAsmMemoryOperand(
2037 const MachineInstr *MI, unsigned OpNo, unsigned AsmVariant,
2038 const char *ExtraCode, raw_ostream &O) {
2039 if (ExtraCode && ExtraCode[0])
2040 return true; // Unknown modifier
2043 printMemOperand(MI, OpNo, O);
2049 void NVPTXAsmPrinter::printOperand(const MachineInstr *MI, int opNum,
2050 raw_ostream &O, const char *Modifier) {
2051 const MachineOperand &MO = MI->getOperand(opNum);
2052 switch (MO.getType()) {
2053 case MachineOperand::MO_Register:
2054 if (TargetRegisterInfo::isPhysicalRegister(MO.getReg())) {
2055 if (MO.getReg() == NVPTX::VRDepot)
2056 O << DEPOTNAME << getFunctionNumber();
2058 O << NVPTXInstPrinter::getRegisterName(MO.getReg());
2060 emitVirtualRegister(MO.getReg(), O);
2064 case MachineOperand::MO_Immediate:
2067 else if (strstr(Modifier, "vec") == Modifier)
2068 printVecModifiedImmediate(MO, Modifier, O);
2071 "Don't know how to handle modifier on immediate operand");
2074 case MachineOperand::MO_FPImmediate:
2075 printFPConstant(MO.getFPImm(), O);
2078 case MachineOperand::MO_GlobalAddress:
2079 O << *getSymbol(MO.getGlobal());
2082 case MachineOperand::MO_ExternalSymbol: {
2083 const char *symbname = MO.getSymbolName();
2084 if (strstr(symbname, ".PARAM") == symbname) {
2086 sscanf(symbname + 6, "%u[];", &index);
2087 printParamName(index, O);
2088 } else if (strstr(symbname, ".HLPPARAM") == symbname) {
2090 sscanf(symbname + 9, "%u[];", &index);
2091 O << *CurrentFnSym << "_param_" << index << "_offset";
2097 case MachineOperand::MO_MachineBasicBlock:
2098 O << *MO.getMBB()->getSymbol();
2102 llvm_unreachable("Operand type not supported.");
2106 void NVPTXAsmPrinter::printMemOperand(const MachineInstr *MI, int opNum,
2107 raw_ostream &O, const char *Modifier) {
2108 printOperand(MI, opNum, O);
2110 if (Modifier && !strcmp(Modifier, "add")) {
2112 printOperand(MI, opNum + 1, O);
2114 if (MI->getOperand(opNum + 1).isImm() &&
2115 MI->getOperand(opNum + 1).getImm() == 0)
2116 return; // don't print ',0' or '+0'
2118 printOperand(MI, opNum + 1, O);
2123 // Force static initialization.
2124 extern "C" void LLVMInitializeNVPTXBackendAsmPrinter() {
2125 RegisterAsmPrinter<NVPTXAsmPrinter> X(TheNVPTXTarget32);
2126 RegisterAsmPrinter<NVPTXAsmPrinter> Y(TheNVPTXTarget64);
2129 void NVPTXAsmPrinter::emitSrcInText(StringRef filename, unsigned line) {
2130 std::stringstream temp;
2131 LineReader *reader = this->getReader(filename.str());
2133 temp << filename.str();
2137 temp << reader->readLine(line);
2139 this->OutStreamer.EmitRawText(Twine(temp.str()));
2142 LineReader *NVPTXAsmPrinter::getReader(std::string filename) {
2143 if (reader == NULL) {
2144 reader = new LineReader(filename);
2147 if (reader->fileName() != filename) {
2149 reader = new LineReader(filename);
2155 std::string LineReader::readLine(unsigned lineNum) {
2156 if (lineNum < theCurLine) {
2158 fstr.seekg(0, std::ios::beg);
2160 while (theCurLine < lineNum) {
2161 fstr.getline(buff, 500);
2167 // Force static initialization.
2168 extern "C" void LLVMInitializeNVPTXAsmPrinter() {
2169 RegisterAsmPrinter<NVPTXAsmPrinter> X(TheNVPTXTarget32);
2170 RegisterAsmPrinter<NVPTXAsmPrinter> Y(TheNVPTXTarget64);