lib/Target/NVPTX/NVPTXAsmPrinter.cpp

   1 //===-- NVPTXAsmPrinter.cpp - NVPTX LLVM assembly writer ------------------===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 // This file contains a printer that converts from our internal representation
  10 // of machine-dependent LLVM code to NVPTX assembly language.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14 #include "NVPTXAsmPrinter.h"
  15 #include "MCTargetDesc/NVPTXBaseInfo.h"
  16 #include "MCTargetDesc/NVPTXInstPrinter.h"
  17 #include "MCTargetDesc/NVPTXMCAsmInfo.h"
  18 #include "MCTargetDesc/NVPTXTargetStreamer.h"
  19 #include "NVPTX.h"
  20 #include "NVPTXMCExpr.h"
  21 #include "NVPTXMachineFunctionInfo.h"
  22 #include "NVPTXRegisterInfo.h"
  23 #include "NVPTXSubtarget.h"
  24 #include "NVPTXTargetMachine.h"
  25 #include "NVPTXUtilities.h"
  26 #include "TargetInfo/NVPTXTargetInfo.h"
  27 #include "cl_common_defines.h"
  28 #include "llvm/ADT/APFloat.h"
  29 #include "llvm/ADT/APInt.h"
  30 #include "llvm/ADT/DenseMap.h"
  31 #include "llvm/ADT/DenseSet.h"
  32 #include "llvm/ADT/SmallString.h"
  33 #include "llvm/ADT/SmallVector.h"
  34 #include "llvm/ADT/StringExtras.h"
  35 #include "llvm/ADT/StringRef.h"
  36 #include "llvm/ADT/Triple.h"
  37 #include "llvm/ADT/Twine.h"
  38 #include "llvm/Analysis/ConstantFolding.h"
  39 #include "llvm/CodeGen/Analysis.h"
  40 #include "llvm/CodeGen/MachineBasicBlock.h"
  41 #include "llvm/CodeGen/MachineFrameInfo.h"
  42 #include "llvm/CodeGen/MachineFunction.h"
  43 #include "llvm/CodeGen/MachineInstr.h"
  44 #include "llvm/CodeGen/MachineLoopInfo.h"
  45 #include "llvm/CodeGen/MachineModuleInfo.h"
  46 #include "llvm/CodeGen/MachineOperand.h"
  47 #include "llvm/CodeGen/MachineRegisterInfo.h"
  48 #include "llvm/CodeGen/TargetLowering.h"
  49 #include "llvm/CodeGen/TargetRegisterInfo.h"
  50 #include "llvm/CodeGen/ValueTypes.h"
  51 #include "llvm/IR/Attributes.h"
  52 #include "llvm/IR/BasicBlock.h"
  53 #include "llvm/IR/Constant.h"
  54 #include "llvm/IR/Constants.h"
  55 #include "llvm/IR/DataLayout.h"
  56 #include "llvm/IR/DebugInfo.h"
  57 #include "llvm/IR/DebugInfoMetadata.h"
  58 #include "llvm/IR/DebugLoc.h"
  59 #include "llvm/IR/DerivedTypes.h"
  60 #include "llvm/IR/Function.h"
  61 #include "llvm/IR/GlobalValue.h"
  62 #include "llvm/IR/GlobalVariable.h"
  63 #include "llvm/IR/Instruction.h"
  64 #include "llvm/IR/LLVMContext.h"
  65 #include "llvm/IR/Module.h"
  66 #include "llvm/IR/Operator.h"
  67 #include "llvm/IR/Type.h"
  68 #include "llvm/IR/User.h"
  69 #include "llvm/MC/MCExpr.h"
  70 #include "llvm/MC/MCInst.h"
  71 #include "llvm/MC/MCInstrDesc.h"
  72 #include "llvm/MC/MCStreamer.h"
  73 #include "llvm/MC/MCSymbol.h"
  74 #include "llvm/Support/Casting.h"
  75 #include "llvm/Support/CommandLine.h"
  76 #include "llvm/Support/ErrorHandling.h"
  77 #include "llvm/Support/MachineValueType.h"
  78 #include "llvm/Support/Path.h"
  79 #include "llvm/Support/TargetRegistry.h"
  80 #include "llvm/Support/raw_ostream.h"
  81 #include "llvm/Target/TargetLoweringObjectFile.h"
  82 #include "llvm/Target/TargetMachine.h"
  83 #include "llvm/Transforms/Utils/UnrollLoop.h"
  84 #include <cassert>
  85 #include <cstdint>
  86 #include <cstring>
  87 #include <new>
  88 #include <string>
  89 #include <utility>
  90 #include <vector>
  91
  92 using namespace llvm;
  93
  94 #define DEPOTNAME "__local_depot"
  95
  96 /// DiscoverDependentGlobals - Return a set of GlobalVariables on which \p V
  97 /// depends.
  98 static void
  99 DiscoverDependentGlobals(const Value *V,
 100                          DenseSet<const GlobalVariable *> &Globals) {
 101   if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
 102     Globals.insert(GV);
 103   else {
 104     if (const User *U = dyn_cast<User>(V)) {
 105       for (unsigned i = 0, e = U->getNumOperands(); i != e; ++i) {
 106         DiscoverDependentGlobals(U->getOperand(i), Globals);
 107       }
 108     }
 109   }
 110 }
 111
 112 /// VisitGlobalVariableForEmission - Add \p GV to the list of GlobalVariable
 113 /// instances to be emitted, but only after any dependents have been added
 114 /// first.s
 115 static void
 116 VisitGlobalVariableForEmission(const GlobalVariable *GV,
 117                                SmallVectorImpl<const GlobalVariable *> &Order,
 118                                DenseSet<const GlobalVariable *> &Visited,
 119                                DenseSet<const GlobalVariable *> &Visiting) {
 120   // Have we already visited this one?
 121   if (Visited.count(GV))
 122     return;
 123
 124   // Do we have a circular dependency?
 125   if (!Visiting.insert(GV).second)
 126     report_fatal_error("Circular dependency found in global variable set");
 127
 128   // Make sure we visit all dependents first
 129   DenseSet<const GlobalVariable *> Others;
 130   for (unsigned i = 0, e = GV->getNumOperands(); i != e; ++i)
 131     DiscoverDependentGlobals(GV->getOperand(i), Others);
 132
 133   for (DenseSet<const GlobalVariable *>::iterator I = Others.begin(),
 134                                                   E = Others.end();
 135        I != E; ++I)
 136     VisitGlobalVariableForEmission(*I, Order, Visited, Visiting);
 137
 138   // Now we can visit ourself
 139   Order.push_back(GV);
 140   Visited.insert(GV);
 141   Visiting.erase(GV);
 142 }
 143
 144 void NVPTXAsmPrinter::EmitInstruction(const MachineInstr *MI) {
 145   MCInst Inst;
 146   lowerToMCInst(MI, Inst);
 147   EmitToStreamer(*OutStreamer, Inst);
 148 }
 149
 150 // Handle symbol backtracking for targets that do not support image handles
 151 bool NVPTXAsmPrinter::lowerImageHandleOperand(const MachineInstr *MI,
 152                                            unsigned OpNo, MCOperand &MCOp) {
 153   const MachineOperand &MO = MI->getOperand(OpNo);
 154   const MCInstrDesc &MCID = MI->getDesc();
 155
 156   if (MCID.TSFlags & NVPTXII::IsTexFlag) {
 157     // This is a texture fetch, so operand 4 is a texref and operand 5 is
 158     // a samplerref
 159     if (OpNo == 4 && MO.isImm()) {
 160       lowerImageHandleSymbol(MO.getImm(), MCOp);
 161       return true;
 162     }
 163     if (OpNo == 5 && MO.isImm() && !(MCID.TSFlags & NVPTXII::IsTexModeUnifiedFlag)) {
 164       lowerImageHandleSymbol(MO.getImm(), MCOp);
 165       return true;
 166     }
 167
 168     return false;
 169   } else if (MCID.TSFlags & NVPTXII::IsSuldMask) {
 170     unsigned VecSize =
 171       1 << (((MCID.TSFlags & NVPTXII::IsSuldMask) >> NVPTXII::IsSuldShift) - 1);
 172
 173     // For a surface load of vector size N, the Nth operand will be the surfref
 174     if (OpNo == VecSize && MO.isImm()) {
 175       lowerImageHandleSymbol(MO.getImm(), MCOp);
 176       return true;
 177     }
 178
 179     return false;
 180   } else if (MCID.TSFlags & NVPTXII::IsSustFlag) {
 181     // This is a surface store, so operand 0 is a surfref
 182     if (OpNo == 0 && MO.isImm()) {
 183       lowerImageHandleSymbol(MO.getImm(), MCOp);
 184       return true;
 185     }
 186
 187     return false;
 188   } else if (MCID.TSFlags & NVPTXII::IsSurfTexQueryFlag) {
 189     // This is a query, so operand 1 is a surfref/texref
 190     if (OpNo == 1 && MO.isImm()) {
 191       lowerImageHandleSymbol(MO.getImm(), MCOp);
 192       return true;
 193     }
 194
 195     return false;
 196   }
 197
 198   return false;
 199 }
 200
 201 void NVPTXAsmPrinter::lowerImageHandleSymbol(unsigned Index, MCOperand &MCOp) {
 202   // Ewwww
 203   LLVMTargetMachine &TM = const_cast<LLVMTargetMachine&>(MF->getTarget());
 204   NVPTXTargetMachine &nvTM = static_cast<NVPTXTargetMachine&>(TM);
 205   const NVPTXMachineFunctionInfo *MFI = MF->getInfo<NVPTXMachineFunctionInfo>();
 206   const char *Sym = MFI->getImageHandleSymbol(Index);
 207   std::string *SymNamePtr =
 208     nvTM.getManagedStrPool()->getManagedString(Sym);
 209   MCOp = GetSymbolRef(OutContext.getOrCreateSymbol(StringRef(*SymNamePtr)));
 210 }
 211
 212 void NVPTXAsmPrinter::lowerToMCInst(const MachineInstr *MI, MCInst &OutMI) {
 213   OutMI.setOpcode(MI->getOpcode());
 214   // Special: Do not mangle symbol operand of CALL_PROTOTYPE
 215   if (MI->getOpcode() == NVPTX::CALL_PROTOTYPE) {
 216     const MachineOperand &MO = MI->getOperand(0);
 217     OutMI.addOperand(GetSymbolRef(
 218       OutContext.getOrCreateSymbol(Twine(MO.getSymbolName()))));
 219     return;
 220   }
 221
 222   const NVPTXSubtarget &STI = MI->getMF()->getSubtarget<NVPTXSubtarget>();
 223   for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
 224     const MachineOperand &MO = MI->getOperand(i);
 225
 226     MCOperand MCOp;
 227     if (!STI.hasImageHandles()) {
 228       if (lowerImageHandleOperand(MI, i, MCOp)) {
 229         OutMI.addOperand(MCOp);
 230         continue;
 231       }
 232     }
 233
 234     if (lowerOperand(MO, MCOp))
 235       OutMI.addOperand(MCOp);
 236   }
 237 }
 238
 239 bool NVPTXAsmPrinter::lowerOperand(const MachineOperand &MO,
 240                                    MCOperand &MCOp) {
 241   switch (MO.getType()) {
 242   default: llvm_unreachable("unknown operand type");
 243   case MachineOperand::MO_Register:
 244     MCOp = MCOperand::createReg(encodeVirtualRegister(MO.getReg()));
 245     break;
 246   case MachineOperand::MO_Immediate:
 247     MCOp = MCOperand::createImm(MO.getImm());
 248     break;
 249   case MachineOperand::MO_MachineBasicBlock:
 250     MCOp = MCOperand::createExpr(MCSymbolRefExpr::create(
 251         MO.getMBB()->getSymbol(), OutContext));
 252     break;
 253   case MachineOperand::MO_ExternalSymbol:
 254     MCOp = GetSymbolRef(GetExternalSymbolSymbol(MO.getSymbolName()));
 255     break;
 256   case MachineOperand::MO_GlobalAddress:
 257     MCOp = GetSymbolRef(getSymbol(MO.getGlobal()));
 258     break;
 259   case MachineOperand::MO_FPImmediate: {
 260     const ConstantFP *Cnt = MO.getFPImm();
 261     const APFloat &Val = Cnt->getValueAPF();
 262
 263     switch (Cnt->getType()->getTypeID()) {
 264     default: report_fatal_error("Unsupported FP type"); break;
 265     case Type::HalfTyID:
 266       MCOp = MCOperand::createExpr(
 267         NVPTXFloatMCExpr::createConstantFPHalf(Val, OutContext));
 268       break;
 269     case Type::FloatTyID:
 270       MCOp = MCOperand::createExpr(
 271         NVPTXFloatMCExpr::createConstantFPSingle(Val, OutContext));
 272       break;
 273     case Type::DoubleTyID:
 274       MCOp = MCOperand::createExpr(
 275         NVPTXFloatMCExpr::createConstantFPDouble(Val, OutContext));
 276       break;
 277     }
 278     break;
 279   }
 280   }
 281   return true;
 282 }
 283
 284 unsigned NVPTXAsmPrinter::encodeVirtualRegister(unsigned Reg) {
 285   if (TargetRegisterInfo::isVirtualRegister(Reg)) {
 286     const TargetRegisterClass *RC = MRI->getRegClass(Reg);
 287
 288     DenseMap<unsigned, unsigned> &RegMap = VRegMapping[RC];
 289     unsigned RegNum = RegMap[Reg];
 290
 291     // Encode the register class in the upper 4 bits
 292     // Must be kept in sync with NVPTXInstPrinter::printRegName
 293     unsigned Ret = 0;
 294     if (RC == &NVPTX::Int1RegsRegClass) {
 295       Ret = (1 << 28);
 296     } else if (RC == &NVPTX::Int16RegsRegClass) {
 297       Ret = (2 << 28);
 298     } else if (RC == &NVPTX::Int32RegsRegClass) {
 299       Ret = (3 << 28);
 300     } else if (RC == &NVPTX::Int64RegsRegClass) {
 301       Ret = (4 << 28);
 302     } else if (RC == &NVPTX::Float32RegsRegClass) {
 303       Ret = (5 << 28);
 304     } else if (RC == &NVPTX::Float64RegsRegClass) {
 305       Ret = (6 << 28);
 306     } else if (RC == &NVPTX::Float16RegsRegClass) {
 307       Ret = (7 << 28);
 308     } else if (RC == &NVPTX::Float16x2RegsRegClass) {
 309       Ret = (8 << 28);
 310     } else {
 311       report_fatal_error("Bad register class");
 312     }
 313
 314     // Insert the vreg number
 315     Ret |= (RegNum & 0x0FFFFFFF);
 316     return Ret;
 317   } else {
 318     // Some special-use registers are actually physical registers.
 319     // Encode this as the register class ID of 0 and the real register ID.
 320     return Reg & 0x0FFFFFFF;
 321   }
 322 }
 323
 324 MCOperand NVPTXAsmPrinter::GetSymbolRef(const MCSymbol *Symbol) {
 325   const MCExpr *Expr;
 326   Expr = MCSymbolRefExpr::create(Symbol, MCSymbolRefExpr::VK_None,
 327                                  OutContext);
 328   return MCOperand::createExpr(Expr);
 329 }
 330
 331 void NVPTXAsmPrinter::printReturnValStr(const Function *F, raw_ostream &O) {
 332   const DataLayout &DL = getDataLayout();
 333   const NVPTXSubtarget &STI = TM.getSubtarget<NVPTXSubtarget>(*F);
 334   const TargetLowering *TLI = STI.getTargetLowering();
 335
 336   Type *Ty = F->getReturnType();
 337
 338   bool isABI = (STI.getSmVersion() >= 20);
 339
 340   if (Ty->getTypeID() == Type::VoidTyID)
 341     return;
 342
 343   O << " (";
 344
 345   if (isABI) {
 346     if (Ty->isFloatingPointTy() || (Ty->isIntegerTy() && !Ty->isIntegerTy(128))) {
 347       unsigned size = 0;
 348       if (auto *ITy = dyn_cast<IntegerType>(Ty)) {
 349         size = ITy->getBitWidth();
 350       } else {
 351         assert(Ty->isFloatingPointTy() && "Floating point type expected here");
 352         size = Ty->getPrimitiveSizeInBits();
 353       }
 354       // PTX ABI requires all scalar return values to be at least 32
 355       // bits in size.  fp16 normally uses .b16 as its storage type in
 356       // PTX, so its size must be adjusted here, too.
 357       if (size < 32)
 358         size = 32;
 359
 360       O << ".param .b" << size << " func_retval0";
 361     } else if (isa<PointerType>(Ty)) {
 362       O << ".param .b" << TLI->getPointerTy(DL).getSizeInBits()
 363         << " func_retval0";
 364     } else if (Ty->isAggregateType() || Ty->isVectorTy() || Ty->isIntegerTy(128)) {
 365       unsigned totalsz = DL.getTypeAllocSize(Ty);
 366       unsigned retAlignment = 0;
 367       if (!getAlign(*F, 0, retAlignment))
 368         retAlignment = DL.getABITypeAlignment(Ty);
 369       O << ".param .align " << retAlignment << " .b8 func_retval0[" << totalsz
 370         << "]";
 371     } else
 372       llvm_unreachable("Unknown return type");
 373   } else {
 374     SmallVector<EVT, 16> vtparts;
 375     ComputeValueVTs(*TLI, DL, Ty, vtparts);
 376     unsigned idx = 0;
 377     for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
 378       unsigned elems = 1;
 379       EVT elemtype = vtparts[i];
 380       if (vtparts[i].isVector()) {
 381         elems = vtparts[i].getVectorNumElements();
 382         elemtype = vtparts[i].getVectorElementType();
 383       }
 384
 385       for (unsigned j = 0, je = elems; j != je; ++j) {
 386         unsigned sz = elemtype.getSizeInBits();
 387         if (elemtype.isInteger() && (sz < 32))
 388           sz = 32;
 389         O << ".reg .b" << sz << " func_retval" << idx;
 390         if (j < je - 1)
 391           O << ", ";
 392         ++idx;
 393       }
 394       if (i < e - 1)
 395         O << ", ";
 396     }
 397   }
 398   O << ") ";
 399 }
 400
 401 void NVPTXAsmPrinter::printReturnValStr(const MachineFunction &MF,
 402                                         raw_ostream &O) {
 403   const Function &F = MF.getFunction();
 404   printReturnValStr(&F, O);
 405 }
 406
 407 // Return true if MBB is the header of a loop marked with
 408 // llvm.loop.unroll.disable.
 409 // TODO: consider "#pragma unroll 1" which is equivalent to "#pragma nounroll".
 410 bool NVPTXAsmPrinter::isLoopHeaderOfNoUnroll(
 411     const MachineBasicBlock &MBB) const {
 412   MachineLoopInfo &LI = getAnalysis<MachineLoopInfo>();
 413   // We insert .pragma "nounroll" only to the loop header.
 414   if (!LI.isLoopHeader(&MBB))
 415     return false;
 416
 417   // llvm.loop.unroll.disable is marked on the back edges of a loop. Therefore,
 418   // we iterate through each back edge of the loop with header MBB, and check
 419   // whether its metadata contains llvm.loop.unroll.disable.
 420   for (auto I = MBB.pred_begin(); I != MBB.pred_end(); ++I) {
 421     const MachineBasicBlock *PMBB = *I;
 422     if (LI.getLoopFor(PMBB) != LI.getLoopFor(&MBB)) {
 423       // Edges from other loops to MBB are not back edges.
 424       continue;
 425     }
 426     if (const BasicBlock *PBB = PMBB->getBasicBlock()) {
 427       if (MDNode *LoopID =
 428               PBB->getTerminator()->getMetadata(LLVMContext::MD_loop)) {
 429         if (GetUnrollMetadata(LoopID, "llvm.loop.unroll.disable"))
 430           return true;
 431       }
 432     }
 433   }
 434   return false;
 435 }
 436
 437 void NVPTXAsmPrinter::EmitBasicBlockStart(const MachineBasicBlock &MBB) const {
 438   AsmPrinter::EmitBasicBlockStart(MBB);
 439   if (isLoopHeaderOfNoUnroll(MBB))
 440     OutStreamer->EmitRawText(StringRef("\t.pragma \"nounroll\";\n"));
 441 }
 442
 443 void NVPTXAsmPrinter::EmitFunctionEntryLabel() {
 444   SmallString<128> Str;
 445   raw_svector_ostream O(Str);
 446
 447   if (!GlobalsEmitted) {
 448     emitGlobals(*MF->getFunction().getParent());
 449     GlobalsEmitted = true;
 450   }
 451
 452   // Set up
 453   MRI = &MF->getRegInfo();
 454   F = &MF->getFunction();
 455   emitLinkageDirective(F, O);
 456   if (isKernelFunction(*F))
 457     O << ".entry ";
 458   else {
 459     O << ".func ";
 460     printReturnValStr(*MF, O);
 461   }
 462
 463   CurrentFnSym->print(O, MAI);
 464
 465   emitFunctionParamList(*MF, O);
 466
 467   if (isKernelFunction(*F))
 468     emitKernelFunctionDirectives(*F, O);
 469
 470   OutStreamer->EmitRawText(O.str());
 471
 472   VRegMapping.clear();
 473   // Emit open brace for function body.
 474   OutStreamer->EmitRawText(StringRef("{\n"));
 475   setAndEmitFunctionVirtualRegisters(*MF);
 476   // Emit initial .loc debug directive for correct relocation symbol data.
 477   if (MMI && MMI->hasDebugInfo())
 478     emitInitialRawDwarfLocDirective(*MF);
 479 }
 480
 481 bool NVPTXAsmPrinter::runOnMachineFunction(MachineFunction &F) {
 482   bool Result = AsmPrinter::runOnMachineFunction(F);
 483   // Emit closing brace for the body of function F.
 484   // The closing brace must be emitted here because we need to emit additional
 485   // debug labels/data after the last basic block.
 486   // We need to emit the closing brace here because we don't have function that
 487   // finished emission of the function body.
 488   OutStreamer->EmitRawText(StringRef("}\n"));
 489   return Result;
 490 }
 491
 492 void NVPTXAsmPrinter::EmitFunctionBodyStart() {
 493   SmallString<128> Str;
 494   raw_svector_ostream O(Str);
 495   emitDemotedVars(&MF->getFunction(), O);
 496   OutStreamer->EmitRawText(O.str());
 497 }
 498
 499 void NVPTXAsmPrinter::EmitFunctionBodyEnd() {
 500   VRegMapping.clear();
 501 }
 502
 503 const MCSymbol *NVPTXAsmPrinter::getFunctionFrameSymbol() const {
 504     SmallString<128> Str;
 505     raw_svector_ostream(Str) << DEPOTNAME << getFunctionNumber();
 506     return OutContext.getOrCreateSymbol(Str);
 507 }
 508
 509 void NVPTXAsmPrinter::emitImplicitDef(const MachineInstr *MI) const {
 510   unsigned RegNo = MI->getOperand(0).getReg();
 511   if (TargetRegisterInfo::isVirtualRegister(RegNo)) {
 512     OutStreamer->AddComment(Twine("implicit-def: ") +
 513                             getVirtualRegisterName(RegNo));
 514   } else {
 515     const NVPTXSubtarget &STI = MI->getMF()->getSubtarget<NVPTXSubtarget>();
 516     OutStreamer->AddComment(Twine("implicit-def: ") +
 517                             STI.getRegisterInfo()->getName(RegNo));
 518   }
 519   OutStreamer->AddBlankLine();
 520 }
 521
 522 void NVPTXAsmPrinter::emitKernelFunctionDirectives(const Function &F,
 523                                                    raw_ostream &O) const {
 524   // If the NVVM IR has some of reqntid* specified, then output
 525   // the reqntid directive, and set the unspecified ones to 1.
 526   // If none of reqntid* is specified, don't output reqntid directive.
 527   unsigned reqntidx, reqntidy, reqntidz;
 528   bool specified = false;
 529   if (!getReqNTIDx(F, reqntidx))
 530     reqntidx = 1;
 531   else
 532     specified = true;
 533   if (!getReqNTIDy(F, reqntidy))
 534     reqntidy = 1;
 535   else
 536     specified = true;
 537   if (!getReqNTIDz(F, reqntidz))
 538     reqntidz = 1;
 539   else
 540     specified = true;
 541
 542   if (specified)
 543     O << ".reqntid " << reqntidx << ", " << reqntidy << ", " << reqntidz
 544       << "\n";
 545
 546   // If the NVVM IR has some of maxntid* specified, then output
 547   // the maxntid directive, and set the unspecified ones to 1.
 548   // If none of maxntid* is specified, don't output maxntid directive.
 549   unsigned maxntidx, maxntidy, maxntidz;
 550   specified = false;
 551   if (!getMaxNTIDx(F, maxntidx))
 552     maxntidx = 1;
 553   else
 554     specified = true;
 555   if (!getMaxNTIDy(F, maxntidy))
 556     maxntidy = 1;
 557   else
 558     specified = true;
 559   if (!getMaxNTIDz(F, maxntidz))
 560     maxntidz = 1;
 561   else
 562     specified = true;
 563
 564   if (specified)
 565     O << ".maxntid " << maxntidx << ", " << maxntidy << ", " << maxntidz
 566       << "\n";
 567
 568   unsigned mincta;
 569   if (getMinCTASm(F, mincta))
 570     O << ".minnctapersm " << mincta << "\n";
 571
 572   unsigned maxnreg;
 573   if (getMaxNReg(F, maxnreg))
 574     O << ".maxnreg " << maxnreg << "\n";
 575 }
 576
 577 std::string
 578 NVPTXAsmPrinter::getVirtualRegisterName(unsigned Reg) const {
 579   const TargetRegisterClass *RC = MRI->getRegClass(Reg);
 580
 581   std::string Name;
 582   raw_string_ostream NameStr(Name);
 583
 584   VRegRCMap::const_iterator I = VRegMapping.find(RC);
 585   assert(I != VRegMapping.end() && "Bad register class");
 586   const DenseMap<unsigned, unsigned> &RegMap = I->second;
 587
 588   VRegMap::const_iterator VI = RegMap.find(Reg);
 589   assert(VI != RegMap.end() && "Bad virtual register");
 590   unsigned MappedVR = VI->second;
 591
 592   NameStr << getNVPTXRegClassStr(RC) << MappedVR;
 593
 594   NameStr.flush();
 595   return Name;
 596 }
 597
 598 void NVPTXAsmPrinter::emitVirtualRegister(unsigned int vr,
 599                                           raw_ostream &O) {
 600   O << getVirtualRegisterName(vr);
 601 }
 602
 603 void NVPTXAsmPrinter::emitDeclaration(const Function *F, raw_ostream &O) {
 604   emitLinkageDirective(F, O);
 605   if (isKernelFunction(*F))
 606     O << ".entry ";
 607   else
 608     O << ".func ";
 609   printReturnValStr(F, O);
 610   getSymbol(F)->print(O, MAI);
 611   O << "\n";
 612   emitFunctionParamList(F, O);
 613   O << ";\n";
 614 }
 615
 616 static bool usedInGlobalVarDef(const Constant *C) {
 617   if (!C)
 618     return false;
 619
 620   if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(C)) {
 621     return GV->getName() != "llvm.used";
 622   }
 623
 624   for (const User *U : C->users())
 625     if (const Constant *C = dyn_cast<Constant>(U))
 626       if (usedInGlobalVarDef(C))
 627         return true;
 628
 629   return false;
 630 }
 631
 632 static bool usedInOneFunc(const User *U, Function const *&oneFunc) {
 633   if (const GlobalVariable *othergv = dyn_cast<GlobalVariable>(U)) {
 634     if (othergv->getName() == "llvm.used")
 635       return true;
 636   }
 637
 638   if (const Instruction *instr = dyn_cast<Instruction>(U)) {
 639     if (instr->getParent() && instr->getParent()->getParent()) {
 640       const Function *curFunc = instr->getParent()->getParent();
 641       if (oneFunc && (curFunc != oneFunc))
 642         return false;
 643       oneFunc = curFunc;
 644       return true;
 645     } else
 646       return false;
 647   }
 648
 649   for (const User *UU : U->users())
 650     if (!usedInOneFunc(UU, oneFunc))
 651       return false;
 652
 653   return true;
 654 }
 655
 656 /* Find out if a global variable can be demoted to local scope.
 657  * Currently, this is valid for CUDA shared variables, which have local
 658  * scope and global lifetime. So the conditions to check are :
 659  * 1. Is the global variable in shared address space?
 660  * 2. Does it have internal linkage?
 661  * 3. Is the global variable referenced only in one function?
 662  */
 663 static bool canDemoteGlobalVar(const GlobalVariable *gv, Function const *&f) {
 664   if (!gv->hasInternalLinkage())
 665     return false;
 666   PointerType *Pty = gv->getType();
 667   if (Pty->getAddressSpace() != ADDRESS_SPACE_SHARED)
 668     return false;
 669
 670   const Function *oneFunc = nullptr;
 671
 672   bool flag = usedInOneFunc(gv, oneFunc);
 673   if (!flag)
 674     return false;
 675   if (!oneFunc)
 676     return false;
 677   f = oneFunc;
 678   return true;
 679 }
 680
 681 static bool useFuncSeen(const Constant *C,
 682                         DenseMap<const Function *, bool> &seenMap) {
 683   for (const User *U : C->users()) {
 684     if (const Constant *cu = dyn_cast<Constant>(U)) {
 685       if (useFuncSeen(cu, seenMap))
 686         return true;
 687     } else if (const Instruction *I = dyn_cast<Instruction>(U)) {
 688       const BasicBlock *bb = I->getParent();
 689       if (!bb)
 690         continue;
 691       const Function *caller = bb->getParent();
 692       if (!caller)
 693         continue;
 694       if (seenMap.find(caller) != seenMap.end())
 695         return true;
 696     }
 697   }
 698   return false;
 699 }
 700
 701 void NVPTXAsmPrinter::emitDeclarations(const Module &M, raw_ostream &O) {
 702   DenseMap<const Function *, bool> seenMap;
 703   for (Module::const_iterator FI = M.begin(), FE = M.end(); FI != FE; ++FI) {
 704     const Function *F = &*FI;
 705
 706     if (F->getAttributes().hasFnAttribute("nvptx-libcall-callee")) {
 707       emitDeclaration(F, O);
 708       continue;
 709     }
 710
 711     if (F->isDeclaration()) {
 712       if (F->use_empty())
 713         continue;
 714       if (F->getIntrinsicID())
 715         continue;
 716       emitDeclaration(F, O);
 717       continue;
 718     }
 719     for (const User *U : F->users()) {
 720       if (const Constant *C = dyn_cast<Constant>(U)) {
 721         if (usedInGlobalVarDef(C)) {
 722           // The use is in the initialization of a global variable
 723           // that is a function pointer, so print a declaration
 724           // for the original function
 725           emitDeclaration(F, O);
 726           break;
 727         }
 728         // Emit a declaration of this function if the function that
 729         // uses this constant expr has already been seen.
 730         if (useFuncSeen(C, seenMap)) {
 731           emitDeclaration(F, O);
 732           break;
 733         }
 734       }
 735
 736       if (!isa<Instruction>(U))
 737         continue;
 738       const Instruction *instr = cast<Instruction>(U);
 739       const BasicBlock *bb = instr->getParent();
 740       if (!bb)
 741         continue;
 742       const Function *caller = bb->getParent();
 743       if (!caller)
 744         continue;
 745
 746       // If a caller has already been seen, then the caller is
 747       // appearing in the module before the callee. so print out
 748       // a declaration for the callee.
 749       if (seenMap.find(caller) != seenMap.end()) {
 750         emitDeclaration(F, O);
 751         break;
 752       }
 753     }
 754     seenMap[F] = true;
 755   }
 756 }
 757
 758 static bool isEmptyXXStructor(GlobalVariable *GV) {
 759   if (!GV) return true;
 760   const ConstantArray *InitList = dyn_cast<ConstantArray>(GV->getInitializer());
 761   if (!InitList) return true;  // Not an array; we don't know how to parse.
 762   return InitList->getNumOperands() == 0;
 763 }
 764
 765 bool NVPTXAsmPrinter::doInitialization(Module &M) {
 766   // Construct a default subtarget off of the TargetMachine defaults. The
 767   // rest of NVPTX isn't friendly to change subtargets per function and
 768   // so the default TargetMachine will have all of the options.
 769   const NVPTXTargetMachine &NTM = static_cast<const NVPTXTargetMachine &>(TM);
 770   const auto* STI = static_cast<const NVPTXSubtarget*>(NTM.getSubtargetImpl());
 771
 772   if (M.alias_size()) {
 773     report_fatal_error("Module has aliases, which NVPTX does not support.");
 774     return true; // error
 775   }
 776   if (!isEmptyXXStructor(M.getNamedGlobal("llvm.global_ctors"))) {
 777     report_fatal_error(
 778         "Module has a nontrivial global ctor, which NVPTX does not support.");
 779     return true;  // error
 780   }
 781   if (!isEmptyXXStructor(M.getNamedGlobal("llvm.global_dtors"))) {
 782     report_fatal_error(
 783         "Module has a nontrivial global dtor, which NVPTX does not support.");
 784     return true;  // error
 785   }
 786
 787   SmallString<128> Str1;
 788   raw_svector_ostream OS1(Str1);
 789
 790   // We need to call the parent's one explicitly.
 791   bool Result = AsmPrinter::doInitialization(M);
 792
 793   // Emit header before any dwarf directives are emitted below.
 794   emitHeader(M, OS1, *STI);
 795   OutStreamer->EmitRawText(OS1.str());
 796
 797   // Emit module-level inline asm if it exists.
 798   if (!M.getModuleInlineAsm().empty()) {
 799     OutStreamer->AddComment("Start of file scope inline assembly");
 800     OutStreamer->AddBlankLine();
 801     OutStreamer->EmitRawText(StringRef(M.getModuleInlineAsm()));
 802     OutStreamer->AddBlankLine();
 803     OutStreamer->AddComment("End of file scope inline assembly");
 804     OutStreamer->AddBlankLine();
 805   }
 806
 807   GlobalsEmitted = false;
 808
 809   return Result;
 810 }
 811
 812 void NVPTXAsmPrinter::emitGlobals(const Module &M) {
 813   SmallString<128> Str2;
 814   raw_svector_ostream OS2(Str2);
 815
 816   emitDeclarations(M, OS2);
 817
 818   // As ptxas does not support forward references of globals, we need to first
 819   // sort the list of module-level globals in def-use order. We visit each
 820   // global variable in order, and ensure that we emit it *after* its dependent
 821   // globals. We use a little extra memory maintaining both a set and a list to
 822   // have fast searches while maintaining a strict ordering.
 823   SmallVector<const GlobalVariable *, 8> Globals;
 824   DenseSet<const GlobalVariable *> GVVisited;
 825   DenseSet<const GlobalVariable *> GVVisiting;
 826
 827   // Visit each global variable, in order
 828   for (const GlobalVariable &I : M.globals())
 829     VisitGlobalVariableForEmission(&I, Globals, GVVisited, GVVisiting);
 830
 831   assert(GVVisited.size() == M.getGlobalList().size() &&
 832          "Missed a global variable");
 833   assert(GVVisiting.size() == 0 && "Did not fully process a global variable");
 834
 835   // Print out module-level global variables in proper order
 836   for (unsigned i = 0, e = Globals.size(); i != e; ++i)
 837     printModuleLevelGV(Globals[i], OS2);
 838
 839   OS2 << '\n';
 840
 841   OutStreamer->EmitRawText(OS2.str());
 842 }
 843
 844 void NVPTXAsmPrinter::emitHeader(Module &M, raw_ostream &O,
 845                                  const NVPTXSubtarget &STI) {
 846   O << "//\n";
 847   O << "// Generated by LLVM NVPTX Back-End\n";
 848   O << "//\n";
 849   O << "\n";
 850
 851   unsigned PTXVersion = STI.getPTXVersion();
 852   O << ".version " << (PTXVersion / 10) << "." << (PTXVersion % 10) << "\n";
 853
 854   O << ".target ";
 855   O << STI.getTargetName();
 856
 857   const NVPTXTargetMachine &NTM = static_cast<const NVPTXTargetMachine &>(TM);
 858   if (NTM.getDrvInterface() == NVPTX::NVCL)
 859     O << ", texmode_independent";
 860
 861   bool HasFullDebugInfo = false;
 862   for (DICompileUnit *CU : M.debug_compile_units()) {
 863     switch(CU->getEmissionKind()) {
 864     case DICompileUnit::NoDebug:
 865     case DICompileUnit::DebugDirectivesOnly:
 866       break;
 867     case DICompileUnit::LineTablesOnly:
 868     case DICompileUnit::FullDebug:
 869       HasFullDebugInfo = true;
 870       break;
 871     }
 872     if (HasFullDebugInfo)
 873       break;
 874   }
 875   if (MMI && MMI->hasDebugInfo() && HasFullDebugInfo)
 876     O << ", debug";
 877
 878   O << "\n";
 879
 880   O << ".address_size ";
 881   if (NTM.is64Bit())
 882     O << "64";
 883   else
 884     O << "32";
 885   O << "\n";
 886
 887   O << "\n";
 888 }
 889
 890 bool NVPTXAsmPrinter::doFinalization(Module &M) {
 891   bool HasDebugInfo = MMI && MMI->hasDebugInfo();
 892
 893   // If we did not emit any functions, then the global declarations have not
 894   // yet been emitted.
 895   if (!GlobalsEmitted) {
 896     emitGlobals(M);
 897     GlobalsEmitted = true;
 898   }
 899
 900   // XXX Temproarily remove global variables so that doFinalization() will not
 901   // emit them again (global variables are emitted at beginning).
 902
 903   Module::GlobalListType &global_list = M.getGlobalList();
 904   int i, n = global_list.size();
 905   GlobalVariable **gv_array = new GlobalVariable *[n];
 906
 907   // first, back-up GlobalVariable in gv_array
 908   i = 0;
 909   for (Module::global_iterator I = global_list.begin(), E = global_list.end();
 910        I != E; ++I)
 911     gv_array[i++] = &*I;
 912
 913   // second, empty global_list
 914   while (!global_list.empty())
 915     global_list.remove(global_list.begin());
 916
 917   // call doFinalization
 918   bool ret = AsmPrinter::doFinalization(M);
 919
 920   // now we restore global variables
 921   for (i = 0; i < n; i++)
 922     global_list.insert(global_list.end(), gv_array[i]);
 923
 924   clearAnnotationCache(&M);
 925
 926   delete[] gv_array;
 927   // Close the last emitted section
 928   if (HasDebugInfo) {
 929     static_cast<NVPTXTargetStreamer *>(OutStreamer->getTargetStreamer())
 930         ->closeLastSection();
 931     // Emit empty .debug_loc section for better support of the empty files.
 932     OutStreamer->EmitRawText("\t.section\t.debug_loc\t{\t}");
 933   }
 934
 935   // Output last DWARF .file directives, if any.
 936   static_cast<NVPTXTargetStreamer *>(OutStreamer->getTargetStreamer())
 937       ->outputDwarfFileDirectives();
 938
 939   return ret;
 940
 941   //bool Result = AsmPrinter::doFinalization(M);
 942   // Instead of calling the parents doFinalization, we may
 943   // clone parents doFinalization and customize here.
 944   // Currently, we if NVISA out the EmitGlobals() in
 945   // parent's doFinalization, which is too intrusive.
 946   //
 947   // Same for the doInitialization.
 948   //return Result;
 949 }
 950
 951 // This function emits appropriate linkage directives for
 952 // functions and global variables.
 953 //
 954 // extern function declaration            -> .extern
 955 // extern function definition             -> .visible
 956 // external global variable with init     -> .visible
 957 // external without init                  -> .extern
 958 // appending                              -> not allowed, assert.
 959 // for any linkage other than
 960 // internal, private, linker_private,
 961 // linker_private_weak, linker_private_weak_def_auto,
 962 // we emit                                -> .weak.
 963
 964 void NVPTXAsmPrinter::emitLinkageDirective(const GlobalValue *V,
 965                                            raw_ostream &O) {
 966   if (static_cast<NVPTXTargetMachine &>(TM).getDrvInterface() == NVPTX::CUDA) {
 967     if (V->hasExternalLinkage()) {
 968       if (isa<GlobalVariable>(V)) {
 969         const GlobalVariable *GVar = cast<GlobalVariable>(V);
 970         if (GVar) {
 971           if (GVar->hasInitializer())
 972             O << ".visible ";
 973           else
 974             O << ".extern ";
 975         }
 976       } else if (V->isDeclaration())
 977         O << ".extern ";
 978       else
 979         O << ".visible ";
 980     } else if (V->hasAppendingLinkage()) {
 981       std::string msg;
 982       msg.append("Error: ");
 983       msg.append("Symbol ");
 984       if (V->hasName())
 985         msg.append(V->getName());
 986       msg.append("has unsupported appending linkage type");
 987       llvm_unreachable(msg.c_str());
 988     } else if (!V->hasInternalLinkage() &&
 989                !V->hasPrivateLinkage()) {
 990       O << ".weak ";
 991     }
 992   }
 993 }
 994
 995 void NVPTXAsmPrinter::printModuleLevelGV(const GlobalVariable *GVar,
 996                                          raw_ostream &O,
 997                                          bool processDemoted) {
 998   // Skip meta data
 999   if (GVar->hasSection()) {
1000     if (GVar->getSection() == "llvm.metadata")
1001       return;
1002   }
1003
1004   // Skip LLVM intrinsic global variables
1005   if (GVar->getName().startswith("llvm.") ||
1006       GVar->getName().startswith("nvvm."))
1007     return;
1008
1009   const DataLayout &DL = getDataLayout();
1010
1011   // GlobalVariables are always constant pointers themselves.
1012   PointerType *PTy = GVar->getType();
1013   Type *ETy = GVar->getValueType();
1014
1015   if (GVar->hasExternalLinkage()) {
1016     if (GVar->hasInitializer())
1017       O << ".visible ";
1018     else
1019       O << ".extern ";
1020   } else if (GVar->hasLinkOnceLinkage() || GVar->hasWeakLinkage() ||
1021              GVar->hasAvailableExternallyLinkage() ||
1022              GVar->hasCommonLinkage()) {
1023     O << ".weak ";
1024   }
1025
1026   if (isTexture(*GVar)) {
1027     O << ".global .texref " << getTextureName(*GVar) << ";\n";
1028     return;
1029   }
1030
1031   if (isSurface(*GVar)) {
1032     O << ".global .surfref " << getSurfaceName(*GVar) << ";\n";
1033     return;
1034   }
1035
1036   if (GVar->isDeclaration()) {
1037     // (extern) declarations, no definition or initializer
1038     // Currently the only known declaration is for an automatic __local
1039     // (.shared) promoted to global.
1040     emitPTXGlobalVariable(GVar, O);
1041     O << ";\n";
1042     return;
1043   }
1044
1045   if (isSampler(*GVar)) {
1046     O << ".global .samplerref " << getSamplerName(*GVar);
1047
1048     const Constant *Initializer = nullptr;
1049     if (GVar->hasInitializer())
1050       Initializer = GVar->getInitializer();
1051     const ConstantInt *CI = nullptr;
1052     if (Initializer)
1053       CI = dyn_cast<ConstantInt>(Initializer);
1054     if (CI) {
1055       unsigned sample = CI->getZExtValue();
1056
1057       O << " = { ";
1058
1059       for (int i = 0,
1060                addr = ((sample & __CLK_ADDRESS_MASK) >> __CLK_ADDRESS_BASE);
1061            i < 3; i++) {
1062         O << "addr_mode_" << i << " = ";
1063         switch (addr) {
1064         case 0:
1065           O << "wrap";
1066           break;
1067         case 1:
1068           O << "clamp_to_border";
1069           break;
1070         case 2:
1071           O << "clamp_to_edge";
1072           break;
1073         case 3:
1074           O << "wrap";
1075           break;
1076         case 4:
1077           O << "mirror";
1078           break;
1079         }
1080         O << ", ";
1081       }
1082       O << "filter_mode = ";
1083       switch ((sample & __CLK_FILTER_MASK) >> __CLK_FILTER_BASE) {
1084       case 0:
1085         O << "nearest";
1086         break;
1087       case 1:
1088         O << "linear";
1089         break;
1090       case 2:
1091         llvm_unreachable("Anisotropic filtering is not supported");
1092       default:
1093         O << "nearest";
1094         break;
1095       }
1096       if (!((sample & __CLK_NORMALIZED_MASK) >> __CLK_NORMALIZED_BASE)) {
1097         O << ", force_unnormalized_coords = 1";
1098       }
1099       O << " }";
1100     }
1101
1102     O << ";\n";
1103     return;
1104   }
1105
1106   if (GVar->hasPrivateLinkage()) {
1107     if (strncmp(GVar->getName().data(), "unrollpragma", 12) == 0)
1108       return;
1109
1110     // FIXME - need better way (e.g. Metadata) to avoid generating this global
1111     if (strncmp(GVar->getName().data(), "filename", 8) == 0)
1112       return;
1113     if (GVar->use_empty())
1114       return;
1115   }
1116
1117   const Function *demotedFunc = nullptr;
1118   if (!processDemoted && canDemoteGlobalVar(GVar, demotedFunc)) {
1119     O << "// " << GVar->getName() << " has been demoted\n";
1120     if (localDecls.find(demotedFunc) != localDecls.end())
1121       localDecls[demotedFunc].push_back(GVar);
1122     else {
1123       std::vector<const GlobalVariable *> temp;
1124       temp.push_back(GVar);
1125       localDecls[demotedFunc] = temp;
1126     }
1127     return;
1128   }
1129
1130   O << ".";
1131   emitPTXAddressSpace(PTy->getAddressSpace(), O);
1132
1133   if (isManaged(*GVar)) {
1134     O << " .attribute(.managed)";
1135   }
1136
1137   if (GVar->getAlignment() == 0)
1138     O << " .align " << (int)DL.getPrefTypeAlignment(ETy);
1139   else
1140     O << " .align " << GVar->getAlignment();
1141
1142   if (ETy->isFloatingPointTy() || ETy->isPointerTy() ||
1143       (ETy->isIntegerTy() && ETy->getScalarSizeInBits() <= 64)) {
1144     O << " .";
1145     // Special case: ABI requires that we use .u8 for predicates
1146     if (ETy->isIntegerTy(1))
1147       O << "u8";
1148     else
1149       O << getPTXFundamentalTypeStr(ETy, false);
1150     O << " ";
1151     getSymbol(GVar)->print(O, MAI);
1152
1153     // Ptx allows variable initilization only for constant and global state
1154     // spaces.
1155     if (GVar->hasInitializer()) {
1156       if ((PTy->getAddressSpace() == ADDRESS_SPACE_GLOBAL) ||
1157           (PTy->getAddressSpace() == ADDRESS_SPACE_CONST)) {
1158         const Constant *Initializer = GVar->getInitializer();
1159         // 'undef' is treated as there is no value specified.
1160         if (!Initializer->isNullValue() && !isa<UndefValue>(Initializer)) {
1161           O << " = ";
1162           printScalarConstant(Initializer, O);
1163         }
1164       } else {
1165         // The frontend adds zero-initializer to device and constant variables
1166         // that don't have an initial value, and UndefValue to shared
1167         // variables, so skip warning for this case.
1168         if (!GVar->getInitializer()->isNullValue() &&
1169             !isa<UndefValue>(GVar->getInitializer())) {
1170           report_fatal_error("initial value of '" + GVar->getName() +
1171                              "' is not allowed in addrspace(" +
1172                              Twine(PTy->getAddressSpace()) + ")");
1173         }
1174       }
1175     }
1176   } else {
1177     unsigned int ElementSize = 0;
1178
1179     // Although PTX has direct support for struct type and array type and
1180     // LLVM IR is very similar to PTX, the LLVM CodeGen does not support for
1181     // targets that support these high level field accesses. Structs, arrays
1182     // and vectors are lowered into arrays of bytes.
1183     switch (ETy->getTypeID()) {
1184     case Type::IntegerTyID: // Integers larger than 64 bits
1185     case Type::StructTyID:
1186     case Type::ArrayTyID:
1187     case Type::VectorTyID:
1188       ElementSize = DL.getTypeStoreSize(ETy);
1189       // Ptx allows variable initilization only for constant and
1190       // global state spaces.
1191       if (((PTy->getAddressSpace() == ADDRESS_SPACE_GLOBAL) ||
1192            (PTy->getAddressSpace() == ADDRESS_SPACE_CONST)) &&
1193           GVar->hasInitializer()) {
1194         const Constant *Initializer = GVar->getInitializer();
1195         if (!isa<UndefValue>(Initializer) && !Initializer->isNullValue()) {
1196           AggBuffer aggBuffer(ElementSize, O, *this);
1197           bufferAggregateConstant(Initializer, &aggBuffer);
1198           if (aggBuffer.numSymbols) {
1199             if (static_cast<const NVPTXTargetMachine &>(TM).is64Bit()) {
1200               O << " .u64 ";
1201               getSymbol(GVar)->print(O, MAI);
1202               O << "[";
1203               O << ElementSize / 8;
1204             } else {
1205               O << " .u32 ";
1206               getSymbol(GVar)->print(O, MAI);
1207               O << "[";
1208               O << ElementSize / 4;
1209             }
1210             O << "]";
1211           } else {
1212             O << " .b8 ";
1213             getSymbol(GVar)->print(O, MAI);
1214             O << "[";
1215             O << ElementSize;
1216             O << "]";
1217           }
1218           O << " = {";
1219           aggBuffer.print();
1220           O << "}";
1221         } else {
1222           O << " .b8 ";
1223           getSymbol(GVar)->print(O, MAI);
1224           if (ElementSize) {
1225             O << "[";
1226             O << ElementSize;
1227             O << "]";
1228           }
1229         }
1230       } else {
1231         O << " .b8 ";
1232         getSymbol(GVar)->print(O, MAI);
1233         if (ElementSize) {
1234           O << "[";
1235           O << ElementSize;
1236           O << "]";
1237         }
1238       }
1239       break;
1240     default:
1241       llvm_unreachable("type not supported yet");
1242     }
1243   }
1244   O << ";\n";
1245 }
1246
1247 void NVPTXAsmPrinter::emitDemotedVars(const Function *f, raw_ostream &O) {
1248   if (localDecls.find(f) == localDecls.end())
1249     return;
1250
1251   std::vector<const GlobalVariable *> &gvars = localDecls[f];
1252
1253   for (unsigned i = 0, e = gvars.size(); i != e; ++i) {
1254     O << "\t// demoted variable\n\t";
1255     printModuleLevelGV(gvars[i], O, true);
1256   }
1257 }
1258
1259 void NVPTXAsmPrinter::emitPTXAddressSpace(unsigned int AddressSpace,
1260                                           raw_ostream &O) const {
1261   switch (AddressSpace) {
1262   case ADDRESS_SPACE_LOCAL:
1263     O << "local";
1264     break;
1265   case ADDRESS_SPACE_GLOBAL:
1266     O << "global";
1267     break;
1268   case ADDRESS_SPACE_CONST:
1269     O << "const";
1270     break;
1271   case ADDRESS_SPACE_SHARED:
1272     O << "shared";
1273     break;
1274   default:
1275     report_fatal_error("Bad address space found while emitting PTX: " +
1276                        llvm::Twine(AddressSpace));
1277     break;
1278   }
1279 }
1280
1281 std::string
1282 NVPTXAsmPrinter::getPTXFundamentalTypeStr(Type *Ty, bool useB4PTR) const {
1283   switch (Ty->getTypeID()) {
1284   default:
1285     llvm_unreachable("unexpected type");
1286     break;
1287   case Type::IntegerTyID: {
1288     unsigned NumBits = cast<IntegerType>(Ty)->getBitWidth();
1289     if (NumBits == 1)
1290       return "pred";
1291     else if (NumBits <= 64) {
1292       std::string name = "u";
1293       return name + utostr(NumBits);
1294     } else {
1295       llvm_unreachable("Integer too large");
1296       break;
1297     }
1298     break;
1299   }
1300   case Type::HalfTyID:
1301     // fp16 is stored as .b16 for compatibility with pre-sm_53 PTX assembly.
1302     return "b16";
1303   case Type::FloatTyID:
1304     return "f32";
1305   case Type::DoubleTyID:
1306     return "f64";
1307   case Type::PointerTyID:
1308     if (static_cast<const NVPTXTargetMachine &>(TM).is64Bit())
1309       if (useB4PTR)
1310         return "b64";
1311       else
1312         return "u64";
1313     else if (useB4PTR)
1314       return "b32";
1315     else
1316       return "u32";
1317   }
1318   llvm_unreachable("unexpected type");
1319   return nullptr;
1320 }
1321
1322 void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable *GVar,
1323                                             raw_ostream &O) {
1324   const DataLayout &DL = getDataLayout();
1325
1326   // GlobalVariables are always constant pointers themselves.
1327   Type *ETy = GVar->getValueType();
1328
1329   O << ".";
1330   emitPTXAddressSpace(GVar->getType()->getAddressSpace(), O);
1331   if (GVar->getAlignment() == 0)
1332     O << " .align " << (int)DL.getPrefTypeAlignment(ETy);
1333   else
1334     O << " .align " << GVar->getAlignment();
1335
1336   // Special case for i128
1337   if (ETy->isIntegerTy(128)) {
1338     O << " .b8 ";
1339     getSymbol(GVar)->print(O, MAI);
1340     O << "[16]";
1341     return;
1342   }
1343
1344   if (ETy->isFloatingPointTy() || ETy->isIntOrPtrTy()) {
1345     O << " .";
1346     O << getPTXFundamentalTypeStr(ETy);
1347     O << " ";
1348     getSymbol(GVar)->print(O, MAI);
1349     return;
1350   }
1351
1352   int64_t ElementSize = 0;
1353
1354   // Although PTX has direct support for struct type and array type and LLVM IR
1355   // is very similar to PTX, the LLVM CodeGen does not support for targets that
1356   // support these high level field accesses. Structs and arrays are lowered
1357   // into arrays of bytes.
1358   switch (ETy->getTypeID()) {
1359   case Type::StructTyID:
1360   case Type::ArrayTyID:
1361   case Type::VectorTyID:
1362     ElementSize = DL.getTypeStoreSize(ETy);
1363     O << " .b8 ";
1364     getSymbol(GVar)->print(O, MAI);
1365     O << "[";
1366     if (ElementSize) {
1367       O << ElementSize;
1368     }
1369     O << "]";
1370     break;
1371   default:
1372     llvm_unreachable("type not supported yet");
1373   }
1374 }
1375
1376 static unsigned int getOpenCLAlignment(const DataLayout &DL, Type *Ty) {
1377   if (Ty->isSingleValueType())
1378     return DL.getPrefTypeAlignment(Ty);
1379
1380   auto *ATy = dyn_cast<ArrayType>(Ty);
1381   if (ATy)
1382     return getOpenCLAlignment(DL, ATy->getElementType());
1383
1384   auto *STy = dyn_cast<StructType>(Ty);
1385   if (STy) {
1386     unsigned int alignStruct = 1;
1387     // Go through each element of the struct and find the
1388     // largest alignment.
1389     for (unsigned i = 0, e = STy->getNumElements(); i != e; i++) {
1390       Type *ETy = STy->getElementType(i);
1391       unsigned int align = getOpenCLAlignment(DL, ETy);
1392       if (align > alignStruct)
1393         alignStruct = align;
1394     }
1395     return alignStruct;
1396   }
1397
1398   auto *FTy = dyn_cast<FunctionType>(Ty);
1399   if (FTy)
1400     return DL.getPointerPrefAlignment();
1401   return DL.getPrefTypeAlignment(Ty);
1402 }
1403
1404 void NVPTXAsmPrinter::printParamName(Function::const_arg_iterator I,
1405                                      int paramIndex, raw_ostream &O) {
1406   getSymbol(I->getParent())->print(O, MAI);
1407   O << "_param_" << paramIndex;
1408 }
1409
1410 void NVPTXAsmPrinter::emitFunctionParamList(const Function *F, raw_ostream &O) {
1411   const DataLayout &DL = getDataLayout();
1412   const AttributeList &PAL = F->getAttributes();
1413   const NVPTXSubtarget &STI = TM.getSubtarget<NVPTXSubtarget>(*F);
1414   const TargetLowering *TLI = STI.getTargetLowering();
1415   Function::const_arg_iterator I, E;
1416   unsigned paramIndex = 0;
1417   bool first = true;
1418   bool isKernelFunc = isKernelFunction(*F);
1419   bool isABI = (STI.getSmVersion() >= 20);
1420   bool hasImageHandles = STI.hasImageHandles();
1421   MVT thePointerTy = TLI->getPointerTy(DL);
1422
1423   if (F->arg_empty()) {
1424     O << "()\n";
1425     return;
1426   }
1427
1428   O << "(\n";
1429
1430   for (I = F->arg_begin(), E = F->arg_end(); I != E; ++I, paramIndex++) {
1431     Type *Ty = I->getType();
1432
1433     if (!first)
1434       O << ",\n";
1435
1436     first = false;
1437
1438     // Handle image/sampler parameters
1439     if (isKernelFunction(*F)) {
1440       if (isSampler(*I) || isImage(*I)) {
1441         if (isImage(*I)) {
1442           std::string sname = I->getName();
1443           if (isImageWriteOnly(*I) || isImageReadWrite(*I)) {
1444             if (hasImageHandles)
1445               O << "\t.param .u64 .ptr .surfref ";
1446             else
1447               O << "\t.param .surfref ";
1448             CurrentFnSym->print(O, MAI);
1449             O << "_param_" << paramIndex;
1450           }
1451           else { // Default image is read_only
1452             if (hasImageHandles)
1453               O << "\t.param .u64 .ptr .texref ";
1454             else
1455               O << "\t.param .texref ";
1456             CurrentFnSym->print(O, MAI);
1457             O << "_param_" << paramIndex;
1458           }
1459         } else {
1460           if (hasImageHandles)
1461             O << "\t.param .u64 .ptr .samplerref ";
1462           else
1463             O << "\t.param .samplerref ";
1464           CurrentFnSym->print(O, MAI);
1465           O << "_param_" << paramIndex;
1466         }
1467         continue;
1468       }
1469     }
1470
1471     if (!PAL.hasParamAttribute(paramIndex, Attribute::ByVal)) {
1472       if (Ty->isAggregateType() || Ty->isVectorTy() || Ty->isIntegerTy(128)) {
1473         // Just print .param .align <a> .b8 .param[size];
1474         // <a> = PAL.getparamalignment
1475         // size = typeallocsize of element type
1476         unsigned align = PAL.getParamAlignment(paramIndex);
1477         if (align == 0)
1478           align = DL.getABITypeAlignment(Ty);
1479
1480         unsigned sz = DL.getTypeAllocSize(Ty);
1481         O << "\t.param .align " << align << " .b8 ";
1482         printParamName(I, paramIndex, O);
1483         O << "[" << sz << "]";
1484
1485         continue;
1486       }
1487       // Just a scalar
1488       auto *PTy = dyn_cast<PointerType>(Ty);
1489       if (isKernelFunc) {
1490         if (PTy) {
1491           // Special handling for pointer arguments to kernel
1492           O << "\t.param .u" << thePointerTy.getSizeInBits() << " ";
1493
1494           if (static_cast<NVPTXTargetMachine &>(TM).getDrvInterface() !=
1495               NVPTX::CUDA) {
1496             Type *ETy = PTy->getElementType();
1497             int addrSpace = PTy->getAddressSpace();
1498             switch (addrSpace) {
1499             default:
1500               O << ".ptr ";
1501               break;
1502             case ADDRESS_SPACE_CONST:
1503               O << ".ptr .const ";
1504               break;
1505             case ADDRESS_SPACE_SHARED:
1506               O << ".ptr .shared ";
1507               break;
1508             case ADDRESS_SPACE_GLOBAL:
1509               O << ".ptr .global ";
1510               break;
1511             }
1512             O << ".align " << (int)getOpenCLAlignment(DL, ETy) << " ";
1513           }
1514           printParamName(I, paramIndex, O);
1515           continue;
1516         }
1517
1518         // non-pointer scalar to kernel func
1519         O << "\t.param .";
1520         // Special case: predicate operands become .u8 types
1521         if (Ty->isIntegerTy(1))
1522           O << "u8";
1523         else
1524           O << getPTXFundamentalTypeStr(Ty);
1525         O << " ";
1526         printParamName(I, paramIndex, O);
1527         continue;
1528       }
1529       // Non-kernel function, just print .param .b<size> for ABI
1530       // and .reg .b<size> for non-ABI
1531       unsigned sz = 0;
1532       if (isa<IntegerType>(Ty)) {
1533         sz = cast<IntegerType>(Ty)->getBitWidth();
1534         if (sz < 32)
1535           sz = 32;
1536       } else if (isa<PointerType>(Ty))
1537         sz = thePointerTy.getSizeInBits();
1538       else if (Ty->isHalfTy())
1539         // PTX ABI requires all scalar parameters to be at least 32
1540         // bits in size.  fp16 normally uses .b16 as its storage type
1541         // in PTX, so its size must be adjusted here, too.
1542         sz = 32;
1543       else
1544         sz = Ty->getPrimitiveSizeInBits();
1545       if (isABI)
1546         O << "\t.param .b" << sz << " ";
1547       else
1548         O << "\t.reg .b" << sz << " ";
1549       printParamName(I, paramIndex, O);
1550       continue;
1551     }
1552
1553     // param has byVal attribute. So should be a pointer
1554     auto *PTy = dyn_cast<PointerType>(Ty);
1555     assert(PTy && "Param with byval attribute should be a pointer type");
1556     Type *ETy = PTy->getElementType();
1557
1558     if (isABI || isKernelFunc) {
1559       // Just print .param .align <a> .b8 .param[size];
1560       // <a> = PAL.getparamalignment
1561       // size = typeallocsize of element type
1562       unsigned align = PAL.getParamAlignment(paramIndex);
1563       if (align == 0)
1564         align = DL.getABITypeAlignment(ETy);
1565       // Work around a bug in ptxas. When PTX code takes address of
1566       // byval parameter with alignment < 4, ptxas generates code to
1567       // spill argument into memory. Alas on sm_50+ ptxas generates
1568       // SASS code that fails with misaligned access. To work around
1569       // the problem, make sure that we align byval parameters by at
1570       // least 4. Matching change must be made in LowerCall() where we
1571       // prepare parameters for the call.
1572       //
1573       // TODO: this will need to be undone when we get to support multi-TU
1574       // device-side compilation as it breaks ABI compatibility with nvcc.
1575       // Hopefully ptxas bug is fixed by then.
1576       if (!isKernelFunc && align < 4)
1577         align = 4;
1578       unsigned sz = DL.getTypeAllocSize(ETy);
1579       O << "\t.param .align " << align << " .b8 ";
1580       printParamName(I, paramIndex, O);
1581       O << "[" << sz << "]";
1582       continue;
1583     } else {
1584       // Split the ETy into constituent parts and
1585       // print .param .b<size> <name> for each part.
1586       // Further, if a part is vector, print the above for
1587       // each vector element.
1588       SmallVector<EVT, 16> vtparts;
1589       ComputeValueVTs(*TLI, DL, ETy, vtparts);
1590       for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
1591         unsigned elems = 1;
1592         EVT elemtype = vtparts[i];
1593         if (vtparts[i].isVector()) {
1594           elems = vtparts[i].getVectorNumElements();
1595           elemtype = vtparts[i].getVectorElementType();
1596         }
1597
1598         for (unsigned j = 0, je = elems; j != je; ++j) {
1599           unsigned sz = elemtype.getSizeInBits();
1600           if (elemtype.isInteger() && (sz < 32))
1601             sz = 32;
1602           O << "\t.reg .b" << sz << " ";
1603           printParamName(I, paramIndex, O);
1604           if (j < je - 1)
1605             O << ",\n";
1606           ++paramIndex;
1607         }
1608         if (i < e - 1)
1609           O << ",\n";
1610       }
1611       --paramIndex;
1612       continue;
1613     }
1614   }
1615
1616   O << "\n)\n";
1617 }
1618
1619 void NVPTXAsmPrinter::emitFunctionParamList(const MachineFunction &MF,
1620                                             raw_ostream &O) {
1621   const Function &F = MF.getFunction();
1622   emitFunctionParamList(&F, O);
1623 }
1624
1625 void NVPTXAsmPrinter::setAndEmitFunctionVirtualRegisters(
1626     const MachineFunction &MF) {
1627   SmallString<128> Str;
1628   raw_svector_ostream O(Str);
1629
1630   // Map the global virtual register number to a register class specific
1631   // virtual register number starting from 1 with that class.
1632   const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
1633   //unsigned numRegClasses = TRI->getNumRegClasses();
1634
1635   // Emit the Fake Stack Object
1636   const MachineFrameInfo &MFI = MF.getFrameInfo();
1637   int NumBytes = (int) MFI.getStackSize();
1638   if (NumBytes) {
1639     O << "\t.local .align " << MFI.getMaxAlignment() << " .b8 \t" << DEPOTNAME
1640       << getFunctionNumber() << "[" << NumBytes << "];\n";
1641     if (static_cast<const NVPTXTargetMachine &>(MF.getTarget()).is64Bit()) {
1642       O << "\t.reg .b64 \t%SP;\n";
1643       O << "\t.reg .b64 \t%SPL;\n";
1644     } else {
1645       O << "\t.reg .b32 \t%SP;\n";
1646       O << "\t.reg .b32 \t%SPL;\n";
1647     }
1648   }
1649
1650   // Go through all virtual registers to establish the mapping between the
1651   // global virtual
1652   // register number and the per class virtual register number.
1653   // We use the per class virtual register number in the ptx output.
1654   unsigned int numVRs = MRI->getNumVirtRegs();
1655   for (unsigned i = 0; i < numVRs; i++) {
1656     unsigned int vr = TRI->index2VirtReg(i);
1657     const TargetRegisterClass *RC = MRI->getRegClass(vr);
1658     DenseMap<unsigned, unsigned> &regmap = VRegMapping[RC];
1659     int n = regmap.size();
1660     regmap.insert(std::make_pair(vr, n + 1));
1661   }
1662
1663   // Emit register declarations
1664   // @TODO: Extract out the real register usage
1665   // O << "\t.reg .pred %p<" << NVPTXNumRegisters << ">;\n";
1666   // O << "\t.reg .s16 %rc<" << NVPTXNumRegisters << ">;\n";
1667   // O << "\t.reg .s16 %rs<" << NVPTXNumRegisters << ">;\n";
1668   // O << "\t.reg .s32 %r<" << NVPTXNumRegisters << ">;\n";
1669   // O << "\t.reg .s64 %rd<" << NVPTXNumRegisters << ">;\n";
1670   // O << "\t.reg .f32 %f<" << NVPTXNumRegisters << ">;\n";
1671   // O << "\t.reg .f64 %fd<" << NVPTXNumRegisters << ">;\n";
1672
1673   // Emit declaration of the virtual registers or 'physical' registers for
1674   // each register class
1675   for (unsigned i=0; i< TRI->getNumRegClasses(); i++) {
1676     const TargetRegisterClass *RC = TRI->getRegClass(i);
1677     DenseMap<unsigned, unsigned> &regmap = VRegMapping[RC];
1678     std::string rcname = getNVPTXRegClassName(RC);
1679     std::string rcStr = getNVPTXRegClassStr(RC);
1680     int n = regmap.size();
1681
1682     // Only declare those registers that may be used.
1683     if (n) {
1684        O << "\t.reg " << rcname << " \t" << rcStr << "<" << (n+1)
1685          << ">;\n";
1686     }
1687   }
1688
1689   OutStreamer->EmitRawText(O.str());
1690 }
1691
1692 void NVPTXAsmPrinter::printFPConstant(const ConstantFP *Fp, raw_ostream &O) {
1693   APFloat APF = APFloat(Fp->getValueAPF()); // make a copy
1694   bool ignored;
1695   unsigned int numHex;
1696   const char *lead;
1697
1698   if (Fp->getType()->getTypeID() == Type::FloatTyID) {
1699     numHex = 8;
1700     lead = "0f";
1701     APF.convert(APFloat::IEEEsingle(), APFloat::rmNearestTiesToEven, &ignored);
1702   } else if (Fp->getType()->getTypeID() == Type::DoubleTyID) {
1703     numHex = 16;
1704     lead = "0d";
1705     APF.convert(APFloat::IEEEdouble(), APFloat::rmNearestTiesToEven, &ignored);
1706   } else
1707     llvm_unreachable("unsupported fp type");
1708
1709   APInt API = APF.bitcastToAPInt();
1710   O << lead << format_hex_no_prefix(API.getZExtValue(), numHex, /*Upper=*/true);
1711 }
1712
1713 void NVPTXAsmPrinter::printScalarConstant(const Constant *CPV, raw_ostream &O) {
1714   if (const ConstantInt *CI = dyn_cast<ConstantInt>(CPV)) {
1715     O << CI->getValue();
1716     return;
1717   }
1718   if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CPV)) {
1719     printFPConstant(CFP, O);
1720     return;
1721   }
1722   if (isa<ConstantPointerNull>(CPV)) {
1723     O << "0";
1724     return;
1725   }
1726   if (const GlobalValue *GVar = dyn_cast<GlobalValue>(CPV)) {
1727     bool IsNonGenericPointer = false;
1728     if (GVar->getType()->getAddressSpace() != 0) {
1729       IsNonGenericPointer = true;
1730     }
1731     if (EmitGeneric && !isa<Function>(CPV) && !IsNonGenericPointer) {
1732       O << "generic(";
1733       getSymbol(GVar)->print(O, MAI);
1734       O << ")";
1735     } else {
1736       getSymbol(GVar)->print(O, MAI);
1737     }
1738     return;
1739   }
1740   if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1741     const Value *v = Cexpr->stripPointerCasts();
1742     PointerType *PTy = dyn_cast<PointerType>(Cexpr->getType());
1743     bool IsNonGenericPointer = false;
1744     if (PTy && PTy->getAddressSpace() != 0) {
1745       IsNonGenericPointer = true;
1746     }
1747     if (const GlobalValue *GVar = dyn_cast<GlobalValue>(v)) {
1748       if (EmitGeneric && !isa<Function>(v) && !IsNonGenericPointer) {
1749         O << "generic(";
1750         getSymbol(GVar)->print(O, MAI);
1751         O << ")";
1752       } else {
1753         getSymbol(GVar)->print(O, MAI);
1754       }
1755       return;
1756     } else {
1757       lowerConstant(CPV)->print(O, MAI);
1758       return;
1759     }
1760   }
1761   llvm_unreachable("Not scalar type found in printScalarConstant()");
1762 }
1763
1764 // These utility functions assure we get the right sequence of bytes for a given
1765 // type even for big-endian machines
1766 template <typename T> static void ConvertIntToBytes(unsigned char *p, T val) {
1767   int64_t vp = (int64_t)val;
1768   for (unsigned i = 0; i < sizeof(T); ++i) {
1769     p[i] = (unsigned char)vp;
1770     vp >>= 8;
1771   }
1772 }
1773 static void ConvertFloatToBytes(unsigned char *p, float val) {
1774   int32_t *vp = (int32_t *)&val;
1775   for (unsigned i = 0; i < sizeof(int32_t); ++i) {
1776     p[i] = (unsigned char)*vp;
1777     *vp >>= 8;
1778   }
1779 }
1780 static void ConvertDoubleToBytes(unsigned char *p, double val) {
1781   int64_t *vp = (int64_t *)&val;
1782   for (unsigned i = 0; i < sizeof(int64_t); ++i) {
1783     p[i] = (unsigned char)*vp;
1784     *vp >>= 8;
1785   }
1786 }
1787
1788 void NVPTXAsmPrinter::bufferLEByte(const Constant *CPV, int Bytes,
1789                                    AggBuffer *aggBuffer) {
1790   const DataLayout &DL = getDataLayout();
1791
1792   if (isa<UndefValue>(CPV) || CPV->isNullValue()) {
1793     int s = DL.getTypeAllocSize(CPV->getType());
1794     if (s < Bytes)
1795       s = Bytes;
1796     aggBuffer->addZeros(s);
1797     return;
1798   }
1799
1800   unsigned char ptr[8];
1801   switch (CPV->getType()->getTypeID()) {
1802
1803   case Type::IntegerTyID: {
1804     Type *ETy = CPV->getType();
1805     if (ETy == Type::getInt8Ty(CPV->getContext())) {
1806       unsigned char c = (unsigned char)cast<ConstantInt>(CPV)->getZExtValue();
1807       ConvertIntToBytes<>(ptr, c);
1808       aggBuffer->addBytes(ptr, 1, Bytes);
1809     } else if (ETy == Type::getInt16Ty(CPV->getContext())) {
1810       short int16 = (short)cast<ConstantInt>(CPV)->getZExtValue();
1811       ConvertIntToBytes<>(ptr, int16);
1812       aggBuffer->addBytes(ptr, 2, Bytes);
1813     } else if (ETy == Type::getInt32Ty(CPV->getContext())) {
1814       if (const ConstantInt *constInt = dyn_cast<ConstantInt>(CPV)) {
1815         int int32 = (int)(constInt->getZExtValue());
1816         ConvertIntToBytes<>(ptr, int32);
1817         aggBuffer->addBytes(ptr, 4, Bytes);
1818         break;
1819       } else if (const auto *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1820         if (const auto *constInt = dyn_cast_or_null<ConstantInt>(
1821                 ConstantFoldConstant(Cexpr, DL))) {
1822           int int32 = (int)(constInt->getZExtValue());
1823           ConvertIntToBytes<>(ptr, int32);
1824           aggBuffer->addBytes(ptr, 4, Bytes);
1825           break;
1826         }
1827         if (Cexpr->getOpcode() == Instruction::PtrToInt) {
1828           Value *v = Cexpr->getOperand(0)->stripPointerCasts();
1829           aggBuffer->addSymbol(v, Cexpr->getOperand(0));
1830           aggBuffer->addZeros(4);
1831           break;
1832         }
1833       }
1834       llvm_unreachable("unsupported integer const type");
1835     } else if (ETy == Type::getInt64Ty(CPV->getContext())) {
1836       if (const ConstantInt *constInt = dyn_cast<ConstantInt>(CPV)) {
1837         long long int64 = (long long)(constInt->getZExtValue());
1838         ConvertIntToBytes<>(ptr, int64);
1839         aggBuffer->addBytes(ptr, 8, Bytes);
1840         break;
1841       } else if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1842         if (const auto *constInt = dyn_cast_or_null<ConstantInt>(
1843                 ConstantFoldConstant(Cexpr, DL))) {
1844           long long int64 = (long long)(constInt->getZExtValue());
1845           ConvertIntToBytes<>(ptr, int64);
1846           aggBuffer->addBytes(ptr, 8, Bytes);
1847           break;
1848         }
1849         if (Cexpr->getOpcode() == Instruction::PtrToInt) {
1850           Value *v = Cexpr->getOperand(0)->stripPointerCasts();
1851           aggBuffer->addSymbol(v, Cexpr->getOperand(0));
1852           aggBuffer->addZeros(8);
1853           break;
1854         }
1855       }
1856       llvm_unreachable("unsupported integer const type");
1857     } else
1858       llvm_unreachable("unsupported integer const type");
1859     break;
1860   }
1861   case Type::HalfTyID:
1862   case Type::FloatTyID:
1863   case Type::DoubleTyID: {
1864     const ConstantFP *CFP = dyn_cast<ConstantFP>(CPV);
1865     Type *Ty = CFP->getType();
1866     if (Ty == Type::getHalfTy(CPV->getContext())) {
1867       APInt API = CFP->getValueAPF().bitcastToAPInt();
1868       uint16_t float16 = API.getLoBits(16).getZExtValue();
1869       ConvertIntToBytes<>(ptr, float16);
1870       aggBuffer->addBytes(ptr, 2, Bytes);
1871     } else if (Ty == Type::getFloatTy(CPV->getContext())) {
1872       float float32 = (float) CFP->getValueAPF().convertToFloat();
1873       ConvertFloatToBytes(ptr, float32);
1874       aggBuffer->addBytes(ptr, 4, Bytes);
1875     } else if (Ty == Type::getDoubleTy(CPV->getContext())) {
1876       double float64 = CFP->getValueAPF().convertToDouble();
1877       ConvertDoubleToBytes(ptr, float64);
1878       aggBuffer->addBytes(ptr, 8, Bytes);
1879     } else {
1880       llvm_unreachable("unsupported fp const type");
1881     }
1882     break;
1883   }
1884   case Type::PointerTyID: {
1885     if (const GlobalValue *GVar = dyn_cast<GlobalValue>(CPV)) {
1886       aggBuffer->addSymbol(GVar, GVar);
1887     } else if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
1888       const Value *v = Cexpr->stripPointerCasts();
1889       aggBuffer->addSymbol(v, Cexpr);
1890     }
1891     unsigned int s = DL.getTypeAllocSize(CPV->getType());
1892     aggBuffer->addZeros(s);
1893     break;
1894   }
1895
1896   case Type::ArrayTyID:
1897   case Type::VectorTyID:
1898   case Type::StructTyID: {
1899     if (isa<ConstantAggregate>(CPV) || isa<ConstantDataSequential>(CPV)) {
1900       int ElementSize = DL.getTypeAllocSize(CPV->getType());
1901       bufferAggregateConstant(CPV, aggBuffer);
1902       if (Bytes > ElementSize)
1903         aggBuffer->addZeros(Bytes - ElementSize);
1904     } else if (isa<ConstantAggregateZero>(CPV))
1905       aggBuffer->addZeros(Bytes);
1906     else
1907       llvm_unreachable("Unexpected Constant type");
1908     break;
1909   }
1910
1911   default:
1912     llvm_unreachable("unsupported type");
1913   }
1914 }
1915
1916 void NVPTXAsmPrinter::bufferAggregateConstant(const Constant *CPV,
1917                                               AggBuffer *aggBuffer) {
1918   const DataLayout &DL = getDataLayout();
1919   int Bytes;
1920
1921   // Integers of arbitrary width
1922   if (const ConstantInt *CI = dyn_cast<ConstantInt>(CPV)) {
1923     APInt Val = CI->getValue();
1924     for (unsigned I = 0, E = DL.getTypeAllocSize(CPV->getType()); I < E; ++I) {
1925       uint8_t Byte = Val.getLoBits(8).getZExtValue();
1926       aggBuffer->addBytes(&Byte, 1, 1);
1927       Val.lshrInPlace(8);
1928     }
1929     return;
1930   }
1931
1932   // Old constants
1933   if (isa<ConstantArray>(CPV) || isa<ConstantVector>(CPV)) {
1934     if (CPV->getNumOperands())
1935       for (unsigned i = 0, e = CPV->getNumOperands(); i != e; ++i)
1936         bufferLEByte(cast<Constant>(CPV->getOperand(i)), 0, aggBuffer);
1937     return;
1938   }
1939
1940   if (const ConstantDataSequential *CDS =
1941           dyn_cast<ConstantDataSequential>(CPV)) {
1942     if (CDS->getNumElements())
1943       for (unsigned i = 0; i < CDS->getNumElements(); ++i)
1944         bufferLEByte(cast<Constant>(CDS->getElementAsConstant(i)), 0,
1945                      aggBuffer);
1946     return;
1947   }
1948
1949   if (isa<ConstantStruct>(CPV)) {
1950     if (CPV->getNumOperands()) {
1951       StructType *ST = cast<StructType>(CPV->getType());
1952       for (unsigned i = 0, e = CPV->getNumOperands(); i != e; ++i) {
1953         if (i == (e - 1))
1954           Bytes = DL.getStructLayout(ST)->getElementOffset(0) +
1955                   DL.getTypeAllocSize(ST) -
1956                   DL.getStructLayout(ST)->getElementOffset(i);
1957         else
1958           Bytes = DL.getStructLayout(ST)->getElementOffset(i + 1) -
1959                   DL.getStructLayout(ST)->getElementOffset(i);
1960         bufferLEByte(cast<Constant>(CPV->getOperand(i)), Bytes, aggBuffer);
1961       }
1962     }
1963     return;
1964   }
1965   llvm_unreachable("unsupported constant type in printAggregateConstant()");
1966 }
1967
1968 /// lowerConstantForGV - Return an MCExpr for the given Constant.  This is mostly
1969 /// a copy from AsmPrinter::lowerConstant, except customized to only handle
1970 /// expressions that are representable in PTX and create
1971 /// NVPTXGenericMCSymbolRefExpr nodes for addrspacecast instructions.
1972 const MCExpr *
1973 NVPTXAsmPrinter::lowerConstantForGV(const Constant *CV, bool ProcessingGeneric) {
1974   MCContext &Ctx = OutContext;
1975
1976   if (CV->isNullValue() || isa<UndefValue>(CV))
1977     return MCConstantExpr::create(0, Ctx);
1978
1979   if (const ConstantInt *CI = dyn_cast<ConstantInt>(CV))
1980     return MCConstantExpr::create(CI->getZExtValue(), Ctx);
1981
1982   if (const GlobalValue *GV = dyn_cast<GlobalValue>(CV)) {
1983     const MCSymbolRefExpr *Expr =
1984       MCSymbolRefExpr::create(getSymbol(GV), Ctx);
1985     if (ProcessingGeneric) {
1986       return NVPTXGenericMCSymbolRefExpr::create(Expr, Ctx);
1987     } else {
1988       return Expr;
1989     }
1990   }
1991
1992   const ConstantExpr *CE = dyn_cast<ConstantExpr>(CV);
1993   if (!CE) {
1994     llvm_unreachable("Unknown constant value to lower!");
1995   }
1996
1997   switch (CE->getOpcode()) {
1998   default:
1999     // If the code isn't optimized, there may be outstanding folding
2000     // opportunities. Attempt to fold the expression using DataLayout as a
2001     // last resort before giving up.
2002     if (Constant *C = ConstantFoldConstant(CE, getDataLayout()))
2003       if (C && C != CE)
2004         return lowerConstantForGV(C, ProcessingGeneric);
2005
2006     // Otherwise report the problem to the user.
2007     {
2008       std::string S;
2009       raw_string_ostream OS(S);
2010       OS << "Unsupported expression in static initializer: ";
2011       CE->printAsOperand(OS, /*PrintType=*/false,
2012                      !MF ? nullptr : MF->getFunction().getParent());
2013       report_fatal_error(OS.str());
2014     }
2015
2016   case Instruction::AddrSpaceCast: {
2017     // Strip the addrspacecast and pass along the operand
2018     PointerType *DstTy = cast<PointerType>(CE->getType());
2019     if (DstTy->getAddressSpace() == 0) {
2020       return lowerConstantForGV(cast<const Constant>(CE->getOperand(0)), true);
2021     }
2022     std::string S;
2023     raw_string_ostream OS(S);
2024     OS << "Unsupported expression in static initializer: ";
2025     CE->printAsOperand(OS, /*PrintType=*/ false,
2026                        !MF ? nullptr : MF->getFunction().getParent());
2027     report_fatal_error(OS.str());
2028   }
2029
2030   case Instruction::GetElementPtr: {
2031     const DataLayout &DL = getDataLayout();
2032
2033     // Generate a symbolic expression for the byte address
2034     APInt OffsetAI(DL.getPointerTypeSizeInBits(CE->getType()), 0);
2035     cast<GEPOperator>(CE)->accumulateConstantOffset(DL, OffsetAI);
2036
2037     const MCExpr *Base = lowerConstantForGV(CE->getOperand(0),
2038                                             ProcessingGeneric);
2039     if (!OffsetAI)
2040       return Base;
2041
2042     int64_t Offset = OffsetAI.getSExtValue();
2043     return MCBinaryExpr::createAdd(Base, MCConstantExpr::create(Offset, Ctx),
2044                                    Ctx);
2045   }
2046
2047   case Instruction::Trunc:
2048     // We emit the value and depend on the assembler to truncate the generated
2049     // expression properly.  This is important for differences between
2050     // blockaddress labels.  Since the two labels are in the same function, it
2051     // is reasonable to treat their delta as a 32-bit value.
2052     LLVM_FALLTHROUGH;
2053   case Instruction::BitCast:
2054     return lowerConstantForGV(CE->getOperand(0), ProcessingGeneric);
2055
2056   case Instruction::IntToPtr: {
2057     const DataLayout &DL = getDataLayout();
2058
2059     // Handle casts to pointers by changing them into casts to the appropriate
2060     // integer type.  This promotes constant folding and simplifies this code.
2061     Constant *Op = CE->getOperand(0);
2062     Op = ConstantExpr::getIntegerCast(Op, DL.getIntPtrType(CV->getType()),
2063                                       false/*ZExt*/);
2064     return lowerConstantForGV(Op, ProcessingGeneric);
2065   }
2066
2067   case Instruction::PtrToInt: {
2068     const DataLayout &DL = getDataLayout();
2069
2070     // Support only foldable casts to/from pointers that can be eliminated by
2071     // changing the pointer to the appropriately sized integer type.
2072     Constant *Op = CE->getOperand(0);
2073     Type *Ty = CE->getType();
2074
2075     const MCExpr *OpExpr = lowerConstantForGV(Op, ProcessingGeneric);
2076
2077     // We can emit the pointer value into this slot if the slot is an
2078     // integer slot equal to the size of the pointer.
2079     if (DL.getTypeAllocSize(Ty) == DL.getTypeAllocSize(Op->getType()))
2080       return OpExpr;
2081
2082     // Otherwise the pointer is smaller than the resultant integer, mask off
2083     // the high bits so we are sure to get a proper truncation if the input is
2084     // a constant expr.
2085     unsigned InBits = DL.getTypeAllocSizeInBits(Op->getType());
2086     const MCExpr *MaskExpr = MCConstantExpr::create(~0ULL >> (64-InBits), Ctx);
2087     return MCBinaryExpr::createAnd(OpExpr, MaskExpr, Ctx);
2088   }
2089
2090   // The MC library also has a right-shift operator, but it isn't consistently
2091   // signed or unsigned between different targets.
2092   case Instruction::Add: {
2093     const MCExpr *LHS = lowerConstantForGV(CE->getOperand(0), ProcessingGeneric);
2094     const MCExpr *RHS = lowerConstantForGV(CE->getOperand(1), ProcessingGeneric);
2095     switch (CE->getOpcode()) {
2096     default: llvm_unreachable("Unknown binary operator constant cast expr");
2097     case Instruction::Add: return MCBinaryExpr::createAdd(LHS, RHS, Ctx);
2098     }
2099   }
2100   }
2101 }
2102
2103 // Copy of MCExpr::print customized for NVPTX
2104 void NVPTXAsmPrinter::printMCExpr(const MCExpr &Expr, raw_ostream &OS) {
2105   switch (Expr.getKind()) {
2106   case MCExpr::Target:
2107     return cast<MCTargetExpr>(&Expr)->printImpl(OS, MAI);
2108   case MCExpr::Constant:
2109     OS << cast<MCConstantExpr>(Expr).getValue();
2110     return;
2111
2112   case MCExpr::SymbolRef: {
2113     const MCSymbolRefExpr &SRE = cast<MCSymbolRefExpr>(Expr);
2114     const MCSymbol &Sym = SRE.getSymbol();
2115     Sym.print(OS, MAI);
2116     return;
2117   }
2118
2119   case MCExpr::Unary: {
2120     const MCUnaryExpr &UE = cast<MCUnaryExpr>(Expr);
2121     switch (UE.getOpcode()) {
2122     case MCUnaryExpr::LNot:  OS << '!'; break;
2123     case MCUnaryExpr::Minus: OS << '-'; break;
2124     case MCUnaryExpr::Not:   OS << '~'; break;
2125     case MCUnaryExpr::Plus:  OS << '+'; break;
2126     }
2127     printMCExpr(*UE.getSubExpr(), OS);
2128     return;
2129   }
2130
2131   case MCExpr::Binary: {
2132     const MCBinaryExpr &BE = cast<MCBinaryExpr>(Expr);
2133
2134     // Only print parens around the LHS if it is non-trivial.
2135     if (isa<MCConstantExpr>(BE.getLHS()) || isa<MCSymbolRefExpr>(BE.getLHS()) ||
2136         isa<NVPTXGenericMCSymbolRefExpr>(BE.getLHS())) {
2137       printMCExpr(*BE.getLHS(), OS);
2138     } else {
2139       OS << '(';
2140       printMCExpr(*BE.getLHS(), OS);
2141       OS<< ')';
2142     }
2143
2144     switch (BE.getOpcode()) {
2145     case MCBinaryExpr::Add:
2146       // Print "X-42" instead of "X+-42".
2147       if (const MCConstantExpr *RHSC = dyn_cast<MCConstantExpr>(BE.getRHS())) {
2148         if (RHSC->getValue() < 0) {
2149           OS << RHSC->getValue();
2150           return;
2151         }
2152       }
2153
2154       OS <<  '+';
2155       break;
2156     default: llvm_unreachable("Unhandled binary operator");
2157     }
2158
2159     // Only print parens around the LHS if it is non-trivial.
2160     if (isa<MCConstantExpr>(BE.getRHS()) || isa<MCSymbolRefExpr>(BE.getRHS())) {
2161       printMCExpr(*BE.getRHS(), OS);
2162     } else {
2163       OS << '(';
2164       printMCExpr(*BE.getRHS(), OS);
2165       OS << ')';
2166     }
2167     return;
2168   }
2169   }
2170
2171   llvm_unreachable("Invalid expression kind!");
2172 }
2173
2174 /// PrintAsmOperand - Print out an operand for an inline asm expression.
2175 ///
2176 bool NVPTXAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
2177                                       const char *ExtraCode, raw_ostream &O) {
2178   if (ExtraCode && ExtraCode[0]) {
2179     if (ExtraCode[1] != 0)
2180       return true; // Unknown modifier.
2181
2182     switch (ExtraCode[0]) {
2183     default:
2184       // See if this is a generic print operand
2185       return AsmPrinter::PrintAsmOperand(MI, OpNo, ExtraCode, O);
2186     case 'r':
2187       break;
2188     }
2189   }
2190
2191   printOperand(MI, OpNo, O);
2192
2193   return false;
2194 }
2195
2196 bool NVPTXAsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI,
2197                                             unsigned OpNo,
2198                                             const char *ExtraCode,
2199                                             raw_ostream &O) {
2200   if (ExtraCode && ExtraCode[0])
2201     return true; // Unknown modifier
2202
2203   O << '[';
2204   printMemOperand(MI, OpNo, O);
2205   O << ']';
2206
2207   return false;
2208 }
2209
2210 void NVPTXAsmPrinter::printOperand(const MachineInstr *MI, int opNum,
2211                                    raw_ostream &O) {
2212   const MachineOperand &MO = MI->getOperand(opNum);
2213   switch (MO.getType()) {
2214   case MachineOperand::MO_Register:
2215     if (TargetRegisterInfo::isPhysicalRegister(MO.getReg())) {
2216       if (MO.getReg() == NVPTX::VRDepot)
2217         O << DEPOTNAME << getFunctionNumber();
2218       else
2219         O << NVPTXInstPrinter::getRegisterName(MO.getReg());
2220     } else {
2221       emitVirtualRegister(MO.getReg(), O);
2222     }
2223     break;
2224
2225   case MachineOperand::MO_Immediate:
2226     O << MO.getImm();
2227     break;
2228
2229   case MachineOperand::MO_FPImmediate:
2230     printFPConstant(MO.getFPImm(), O);
2231     break;
2232
2233   case MachineOperand::MO_GlobalAddress:
2234     PrintSymbolOperand(MO, O);
2235     break;
2236
2237   case MachineOperand::MO_MachineBasicBlock:
2238     MO.getMBB()->getSymbol()->print(O, MAI);
2239     break;
2240
2241   default:
2242     llvm_unreachable("Operand type not supported.");
2243   }
2244 }
2245
2246 void NVPTXAsmPrinter::printMemOperand(const MachineInstr *MI, int opNum,
2247                                       raw_ostream &O, const char *Modifier) {
2248   printOperand(MI, opNum, O);
2249
2250   if (Modifier && strcmp(Modifier, "add") == 0) {
2251     O << ", ";
2252     printOperand(MI, opNum + 1, O);
2253   } else {
2254     if (MI->getOperand(opNum + 1).isImm() &&
2255         MI->getOperand(opNum + 1).getImm() == 0)
2256       return; // don't print ',0' or '+0'
2257     O << "+";
2258     printOperand(MI, opNum + 1, O);
2259   }
2260 }
2261
2262 // Force static initialization.
2263 extern "C" void LLVMInitializeNVPTXAsmPrinter() {
2264   RegisterAsmPrinter<NVPTXAsmPrinter> X(getTheNVPTXTarget32());
2265   RegisterAsmPrinter<NVPTXAsmPrinter> Y(getTheNVPTXTarget64());
2266 }