Handle ARM "ret void" and function alignment with proper padding.

src/IceTargetLoweringARM32.cpp

 //===- subzero/src/IceTargetLoweringARM32.cpp - ARM32 lowering ------------===//
 //
 //                        The Subzero Code Generator
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 //
 // This file implements the TargetLoweringARM32 class, which consists almost
@@ skipping 12 matching lines @@
 #include "IceInstARM32.h"
 #include "IceLiveness.h"
 #include "IceOperand.h"
 #include "IceRegistersARM32.h"
 #include "IceTargetLoweringARM32.def"
 #include "IceTargetLoweringARM32.h"
 #include "IceUtils.h"
 
 namespace Ice {
 
+namespace {
+void UnimplementedError(const ClFlags &Flags) {
+  if (!Flags.getSkipUnimplemented()) {
+    // Use llvm_unreachable instead of report_fatal_error, which gives better
+    // stack traces.
+    llvm_unreachable("Not yet implemented");
+    abort();
+  }
+}
+} // end of anonymous namespace
+
 TargetARM32::TargetARM32(Cfg *Func)
     : TargetLowering(Func), UsesFramePointer(false) {
   // TODO: Don't initialize IntegerRegisters and friends every time.
   // Instead, initialize in some sort of static initializer for the
   // class.
   llvm::SmallBitVector IntegerRegisters(RegARM32::Reg_NUM);
   llvm::SmallBitVector FloatRegisters(RegARM32::Reg_NUM);
   llvm::SmallBitVector VectorRegisters(RegARM32::Reg_NUM);
   llvm::SmallBitVector InvalidRegisters(RegARM32::Reg_NUM);
   ScratchRegs.resize(RegARM32::Reg_NUM);
@@ skipping 155 matching lines @@
 
   // Nop insertion
   if (Ctx->getFlags().shouldDoNopInsertion()) {
     Func->doNopInsertion();
   }
 }
 
 bool TargetARM32::doBranchOpt(Inst *I, const CfgNode *NextNode) {
   (void)I;
   (void)NextNode;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
+  return false;
 }
 
 IceString TargetARM32::RegNames[] = {
 #define X(val, encode, name, scratch, preserved, stackptr, frameptr, isInt,    \
           isFP)                                                                \
   name,
     REGARM32_TABLE
 #undef X
 };
 
 IceString TargetARM32::getRegName(SizeT RegNum, Type Ty) const {
   assert(RegNum < RegARM32::Reg_NUM);
   (void)Ty;
   return RegNames[RegNum];
 }
 
 Variable *TargetARM32::getPhysicalRegister(SizeT RegNum, Type Ty) {
   if (Ty == IceType_void)
     Ty = IceType_i32;
   if (PhysicalRegisters[Ty].empty())
     PhysicalRegisters[Ty].resize(RegARM32::Reg_NUM);
   assert(RegNum < PhysicalRegisters[Ty].size());
   Variable *Reg = PhysicalRegisters[Ty][RegNum];
   if (Reg == nullptr) {
     Reg = Func->makeVariable(Ty);
     Reg->setRegNum(RegNum);
     PhysicalRegisters[Ty][RegNum] = Reg;
-    // Specially mark SP as an "argument" so that it is considered
+    // Specially mark SP and LR as an "argument" so that it is considered
     // live upon function entry.
-    if (RegNum == RegARM32::Reg_sp) {
+    if (RegNum == RegARM32::Reg_sp || RegNum == RegARM32::Reg_lr) {
       Func->addImplicitArg(Reg);
       Reg->setIgnoreLiveness();
     }
   }
   return Reg;
 }
 
 void TargetARM32::emitVariable(const Variable *Var) const {
   Ostream &Str = Ctx->getStrEmit();
-  (void)Var;
-  (void)Str;
-  llvm::report_fatal_error("emitVariable: Not yet implemented");
+  if (Var->hasReg()) {
+    Str << getRegName(Var->getRegNum(), Var->getType());
+    return;
+  }
+  if (Var->getWeight().isInf()) {
+    llvm::report_fatal_error(
+        "Infinite-weight Variable has no register assigned");
+  }
+  int32_t Offset = Var->getStackOffset();
+  if (!hasFramePointer())
+    Offset += getStackAdjustment();
+  // TODO(jvoung): Handle out of range. Perhaps we need a scratch register
+  // to materialize a larger offset.
+  const bool SignExt = false;
+  if (!OperandARM32Mem::canHoldOffset(Var->getType(), SignExt, Offset)) {
+    llvm::report_fatal_error("Illegal stack offset");
+  }
+  const Type FrameSPTy = IceType_i32;
+  Str << "[" << getRegName(getFrameOrStackReg(), FrameSPTy) << ", " << Offset
+      << "]";
 }
 
 void TargetARM32::lowerArguments() {
-  llvm::report_fatal_error("lowerArguments: Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 Type TargetARM32::stackSlotType() { return IceType_i32; }
 
 void TargetARM32::addProlog(CfgNode *Node) {
   (void)Node;
-  llvm::report_fatal_error("addProlog: Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::addEpilog(CfgNode *Node) {
   (void)Node;
-  llvm::report_fatal_error("addEpilog: Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 llvm::SmallBitVector TargetARM32::getRegisterSet(RegSetMask Include,
                                                  RegSetMask Exclude) const {
   llvm::SmallBitVector Registers(RegARM32::Reg_NUM);
 
 #define X(val, encode, name, scratch, preserved, stackptr, frameptr, isInt,    \
           isFP)                                                                \
   if (scratch && (Include & RegSet_CallerSave))                                \
     Registers[RegARM32::val] = true;                                           \
@@ skipping 21 matching lines @@
 
 void TargetARM32::lowerAlloca(const InstAlloca *Inst) {
   UsesFramePointer = true;
   // Conservatively require the stack to be aligned.  Some stack
   // adjustment operations implemented below assume that the stack is
   // aligned before the alloca.  All the alloca code ensures that the
   // stack alignment is preserved after the alloca.  The stack alignment
   // restriction can be relaxed in some cases.
   NeedsStackAlignment = true;
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::lowerArithmetic(const InstArithmetic *Inst) {
   switch (Inst->getOp()) {
   case InstArithmetic::_num:
     llvm_unreachable("Unknown arithmetic operator");
     break;
   case InstArithmetic::Add:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::And:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Or:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Xor:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Sub:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Mul:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Shl:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Lshr:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Ashr:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Udiv:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Sdiv:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Urem:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Srem:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Fadd:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Fsub:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Fmul:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Fdiv:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstArithmetic::Frem:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   }
 }
 
 void TargetARM32::lowerAssign(const InstAssign *Inst) {
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::lowerBr(const InstBr *Inst) {
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::lowerCall(const InstCall *Inst) {
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::lowerCast(const InstCast *Inst) {
   InstCast::OpKind CastKind = Inst->getCastKind();
   switch (CastKind) {
   default:
     Func->setError("Cast type not supported");
     return;
   case InstCast::Sext: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   }
   case InstCast::Zext: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   }
   case InstCast::Trunc: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   }
   case InstCast::Fptrunc:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstCast::Fpext: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   }
   case InstCast::Fptosi:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstCast::Fptoui:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstCast::Sitofp:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   case InstCast::Uitofp: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   }
   case InstCast::Bitcast: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     break;
   }
   }
 }
 
 void TargetARM32::lowerExtractElement(const InstExtractElement *Inst) {
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::lowerFcmp(const InstFcmp *Inst) {
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::lowerIcmp(const InstIcmp *Inst) {
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::lowerInsertElement(const InstInsertElement *Inst) {
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::lowerIntrinsicCall(const InstIntrinsicCall *Instr) {
   switch (Intrinsics::IntrinsicID ID = Instr->getIntrinsicInfo().ID) {
   case Intrinsics::AtomicCmpxchg: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::AtomicFence:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   case Intrinsics::AtomicFenceAll:
     // NOTE: FenceAll should prevent and load/store from being moved
     // across the fence (both atomic and non-atomic). The InstARM32Mfence
     // instruction is currently marked coarsely as "HasSideEffects".
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   case Intrinsics::AtomicIsLockFree: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::AtomicLoad: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::AtomicRMW:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   case Intrinsics::AtomicStore: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::Bswap: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::Ctpop: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::Ctlz: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::Cttz: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::Fabs: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::Longjmp: {
     InstCall *Call = makeHelperCall(H_call_longjmp, nullptr, 2);
     Call->addArg(Instr->getArg(0));
     Call->addArg(Instr->getArg(1));
     lowerCall(Call);
     return;
   }
   case Intrinsics::Memcpy: {
@@ skipping 24 matching lines @@
     lowerCast(InstCast::create(Func, InstCast::Zext, ValExt, ValOp));
     InstCall *Call = makeHelperCall(H_call_memset, nullptr, 3);
     Call->addArg(Instr->getArg(0));
     Call->addArg(ValExt);
     Call->addArg(Instr->getArg(2));
     lowerCall(Call);
     return;
   }
   case Intrinsics::NaClReadTP: {
     if (Ctx->getFlags().getUseSandboxing()) {
-      llvm::report_fatal_error("Not yet implemented");
+      UnimplementedError(Func->getContext()->getFlags());
     } else {
       InstCall *Call = makeHelperCall(H_call_read_tp, Instr->getDest(), 0);
       lowerCall(Call);
     }
     return;
   }
   case Intrinsics::Setjmp: {
     InstCall *Call = makeHelperCall(H_call_setjmp, Instr->getDest(), 1);
     Call->addArg(Instr->getArg(0));
     lowerCall(Call);
     return;
   }
   case Intrinsics::Sqrt: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::Stacksave: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::Stackrestore: {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   }
   case Intrinsics::Trap:
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
     return;
   case Intrinsics::UnknownIntrinsic:
     Func->setError("Should not be lowering UnknownIntrinsic");
     return;
   }
   return;
 }
 
 void TargetARM32::lowerLoad(const InstLoad *Inst) {
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::doAddressOptLoad() {
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::randomlyInsertNop(float Probability) {
   RandomNumberGeneratorWrapper RNG(Ctx->getRNG());
   if (RNG.getTrueWithProbability(Probability)) {
-    llvm::report_fatal_error("Not yet implemented");
+    UnimplementedError(Func->getContext()->getFlags());
   }
 }
 
 void TargetARM32::lowerPhi(const InstPhi * /*Inst*/) {
   Func->setError("Phi found in regular instruction list");
 }
 
 void TargetARM32::lowerRet(const InstRet *Inst) {
-  (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  Variable *Reg = nullptr;
+  if (Inst->hasRetValue()) {
+    UnimplementedError(Func->getContext()->getFlags());
+  }
+  // Add a ret instruction even if sandboxing is enabled, because
+  // addEpilog explicitly looks for a ret instruction as a marker for
+  // where to insert the frame removal instructions.
+  // addEpilog is responsible for restoring the "lr" register as needed
+  // prior to this ret instruction.
+  _ret(getPhysicalRegister(RegARM32::Reg_lr), Reg);
+  // Add a fake use of sp to make sure sp stays alive for the entire
+  // function.  Otherwise post-call sp adjustments get dead-code
+  // eliminated.  TODO: Are there more places where the fake use
+  // should be inserted?  E.g. "void f(int n){while(1) g(n);}" may not
+  // have a ret instruction.
+  Variable *SP = Func->getTarget()->getPhysicalRegister(RegARM32::Reg_sp);
+  Context.insert(InstFakeUse::create(Func, SP));
 }
 
 void TargetARM32::lowerSelect(const InstSelect *Inst) {
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::lowerStore(const InstStore *Inst) {
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::doAddressOptStore() {
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::lowerSwitch(const InstSwitch *Inst) {
   (void)Inst;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::lowerUnreachable(const InstUnreachable * /*Inst*/) {
   llvm_unreachable("Not yet implemented");
 }
 
 // Turn an i64 Phi instruction into a pair of i32 Phi instructions, to
 // preserve integrity of liveness analysis.  Undef values are also
 // turned into zeroes, since loOperand() and hiOperand() don't expect
 // Undef input.
 void TargetARM32::prelowerPhis() {
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 // Lower the pre-ordered list of assignments into mov instructions.
 // Also has to do some ad-hoc register allocation as necessary.
 void TargetARM32::lowerPhiAssignments(CfgNode *Node,
                                       const AssignList &Assignments) {
   (void)Node;
   (void)Assignments;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::postLower() {
   if (Ctx->getFlags().getOptLevel() == Opt_m1)
     return;
   // Find two-address non-SSA instructions where Dest==Src0, and set
   // the DestNonKillable flag to keep liveness analysis consistent.
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 void TargetARM32::makeRandomRegisterPermutation(
     llvm::SmallVectorImpl<int32_t> &Permutation,
     const llvm::SmallBitVector &ExcludeRegisters) const {
   (void)Permutation;
   (void)ExcludeRegisters;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Func->getContext()->getFlags());
 }
 
 /* TODO(jvoung): avoid duplicate symbols with multiple targets.
 void ConstantUndef::emitWithoutDollar(GlobalContext *) const {
   llvm_unreachable("Not expecting to emitWithoutDollar undef");
 }
 
 void ConstantUndef::emit(GlobalContext *) const {
   llvm_unreachable("undef value encountered by emitter.");
 }
 */
 
 TargetDataARM32::TargetDataARM32(GlobalContext *Ctx)
     : TargetDataLowering(Ctx) {}
 
 void TargetDataARM32::lowerGlobal(const VariableDeclaration &Var) const {
   (void)Var;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Ctx->getFlags());
 }
 
 void TargetDataARM32::lowerGlobals(
     std::unique_ptr<VariableDeclarationList> Vars) const {
   switch (Ctx->getFlags().getOutFileType()) {
   case FT_Elf: {
     ELFObjectWriter *Writer = Ctx->getObjectWriter();
     Writer->writeDataSection(*Vars, llvm::ELF::R_ARM_ABS32);
   } break;
   case FT_Asm:
   case FT_Iasm: {
     const IceString &TranslateOnly = Ctx->getFlags().getTranslateOnly();
     OstreamLocker L(Ctx);
     for (const VariableDeclaration *Var : *Vars) {
       if (GlobalContext::matchSymbolName(Var->getName(), TranslateOnly)) {
         lowerGlobal(*Var);
       }
     }
   } break;
   }
 }
 
 void TargetDataARM32::lowerConstants() const {
   if (Ctx->getFlags().getDisableTranslation())
     return;
-  llvm::report_fatal_error("Not yet implemented");
+  UnimplementedError(Ctx->getFlags());
 }
 
 } // end of namespace Ice