[AArch64][GlobalISel] Support sibling calls with outgoing arguments

[paquette]

Summary

This adds support for lowering sibling calls with outgoing arguments.

e.g

define void @foo(i32 %a)

Support is ported from AArch64ISelLowering's isEligibleForTailCallOptimization. The only thing that is missing is a full port of TargetLowering::parametersInCSRMatch. So, if we're using swiftself, we'll never tail call.

• Rename analyzeCallResult to analyzeArgInfo, since the function is now used for both outgoing and incoming arguments
• Teach OutgoingArgHandler about tail calls. Tail calls use frame indices for stack arguments.
• Teach lowerFormalArguments to set the bytes in the caller's stack argument area. This is used later to check if the tail call's parameters will fit on the caller's stack.
• Add areCalleeOutgoingArgsTailCallable to perform the eligibility check on the callee's outgoing arguments.

For testing:

• Update call-translator-tail-call to verify that we can now tail call with outgoing arguments, use G_FRAME_INDEX for stack arguments, and respect the size of the caller's stack
• Remove GISel-specific check lines from speculation-hardening.ll, since GISel now tail calls like the other selectors
• Add a GISel test line to tailcall-string-rvo.ll since we can tail call in that test now
• Add a GISel test line to tailcall_misched_graph.ll since we tail call there now. Add specific check lines for GISel, since the debug output from the machine-scheduler differs with GlobalISel. The dependency still holds, but the output comes out in a different order.

[aemerson]

llvm/lib/Target/AArch64/AArch64CallLowering.cpp

413 ↗

(On Diff #219812)

Not really sure what this comment means.

[paquette]

llvm/lib/Target/AArch64/AArch64CallLowering.cpp

413 ↗

(On Diff #219812)

Copied/pasted it from the SDAG code, but I can reword it.

[paquette]

Improve comment

[aemerson]

LGTM.

llvm/trunk/include/llvm/CodeGen/GlobalISel/CallLowering.h

bool handleAssignments(MachineIRBuilder &MIRBuilder,
                          SmallVectorImpl<ArgInfo> &Args,
                          ValueHandler &Handler) const;
   bool handleAssignments(CCState &CCState,
                          SmallVectorImpl<CCValAssign> &ArgLocs,
                          MachineIRBuilder &MIRBuilder,
                          SmallVectorImpl<ArgInfo> &Args,
                          ValueHandler &Handler) const;
 
-  /// Analyze the return values of a call, incorporating info about the passed
-  /// values into \p CCState.
-  bool analyzeCallResult(CCState &CCState, SmallVectorImpl<ArgInfo> &Args,
-                         CCAssignFn &Fn) const;
+  /// Analyze passed or returned values from a call, supplied in \p ArgInfo,
+  /// incorporating info about the passed values into \p CCState.
+  ///
+  /// Used to check if arguments are suitable for tail call lowering.
+  bool analyzeArgInfo(CCState &CCState, SmallVectorImpl<ArgInfo> &Args,
 
   /// \returns True if the calling convention for a callee and its caller pass
   /// results in the same way. Typically used for tail call eligibility checks.
   ///
   /// \p Info is the CallLoweringInfo for the call.
   /// \p MF is the MachineFunction for the caller.
   /// \p InArgs contains the results of the call.
   /// \p CalleeAssignFn is the CCAssignFn to be used for the callee.

llvm/trunk/lib/CodeGen/GlobalISel/CallLowering.cpp

for (unsigned i = 0, e = Args.size(), j = 0; i != e; ++i, ++j) {
     } else {
       // FIXME: Support byvals and other weirdness
       return false;
     }
   }
   return true;
 }
 
-bool CallLowering::analyzeCallResult(CCState &CCState,
+bool CallLowering::analyzeArgInfo(CCState &CCState,
                                      SmallVectorImpl<ArgInfo> &Args,
                                      CCAssignFn &Fn) const {
   for (unsigned i = 0, e = Args.size(); i < e; ++i) {
     MVT VT = MVT::getVT(Args[i].Ty);
     if (Fn(i, VT, VT, CCValAssign::Full, Args[i].Flags[0], CCState)) {
       // Bail out on anything we can't handle.
       LLVM_DEBUG(dbgs() << "Cannot analyze " << EVT(VT).getEVTString()
                         << " (arg number = " << i << "\n");
bool CallLowering::resultsCompatible(CallLoweringInfo &Info,
   CallingConv::ID CalleeCC = Info.CallConv;
   CallingConv::ID CallerCC = F.getCallingConv();
 
   if (CallerCC == CalleeCC)
     return true;
 
   SmallVector<CCValAssign, 16> ArgLocs1;
   CCState CCInfo1(CalleeCC, false, MF, ArgLocs1, F.getContext());
-  if (!analyzeCallResult(CCInfo1, InArgs, CalleeAssignFn))
+  if (!analyzeArgInfo(CCInfo1, InArgs, CalleeAssignFn))
     return false;
 
   SmallVector<CCValAssign, 16> ArgLocs2;
   CCState CCInfo2(CallerCC, false, MF, ArgLocs2, F.getContext());
-  if (!analyzeCallResult(CCInfo2, InArgs, CallerAssignFn))
+  if (!analyzeArgInfo(CCInfo2, InArgs, CallerAssignFn))
     return false;
 
   // We need the argument locations to match up exactly. If there's more in
   // one than the other, then we are done.
   if (ArgLocs1.size() != ArgLocs2.size())
     return false;
 
   // Make sure that each location is passed in exactly the same way.

llvm/trunk/lib/Target/AArch64/AArch64CallLowering.h

public:
 
   bool lowerCall(MachineIRBuilder &MIRBuilder,
                  CallLoweringInfo &Info) const override;
 
   /// Returns true if the call can be lowered as a tail call.
   bool
   isEligibleForTailCallOptimization(MachineIRBuilder &MIRBuilder,
                                     CallLoweringInfo &Info,
-                                    SmallVectorImpl<ArgInfo> &InArgs) const;
+                                    SmallVectorImpl<ArgInfo> &InArgs,
+                                    SmallVectorImpl<ArgInfo> &OutArgs) const;
 
   bool supportSwiftError() const override { return true; }
 
 private:
   using RegHandler = std::function<void(MachineIRBuilder &, Type *, unsigned,
                                         CCValAssign &)>;
 
   using MemHandler =
       std::function<void(MachineIRBuilder &, int, CCValAssign &)>;
 
   void splitToValueTypes(const ArgInfo &OrigArgInfo,
                          SmallVectorImpl<ArgInfo> &SplitArgs,
                          const DataLayout &DL, MachineRegisterInfo &MRI,
                          CallingConv::ID CallConv) const;
 
   bool
   doCallerAndCalleePassArgsTheSameWay(CallLoweringInfo &Info,
                                       MachineFunction &MF,
                                       SmallVectorImpl<ArgInfo> &InArgs) const;
+
+  bool
+  areCalleeOutgoingArgsTailCallable(CallLoweringInfo &Info, MachineFunction &MF,
+                                    SmallVectorImpl<ArgInfo> &OutArgs) const;
 };
 
 } // end namespace llvm
 
 #endif // LLVM_LIB_TARGET_AARCH64_AARCH64CALLLOWERING_H

llvm/trunk/lib/Target/AArch64/AArch64CallLowering.cpp

struct CallReturnHandler : public IncomingArgHandler {
   }
 
   MachineInstrBuilder MIB;
 };
 
 struct OutgoingArgHandler : public CallLowering::ValueHandler {
   OutgoingArgHandler(MachineIRBuilder &MIRBuilder, MachineRegisterInfo &MRI,
                      MachineInstrBuilder MIB, CCAssignFn *AssignFn,
-                     CCAssignFn *AssignFnVarArg)
+                     CCAssignFn *AssignFnVarArg, bool IsTailCall = false)
       : ValueHandler(MIRBuilder, MRI, AssignFn), MIB(MIB),
-        AssignFnVarArg(AssignFnVarArg), StackSize(0) {}
+        AssignFnVarArg(AssignFnVarArg), IsTailCall(IsTailCall), StackSize(0) {}
 
   Register getStackAddress(uint64_t Size, int64_t Offset,
                            MachinePointerInfo &MPO) override {
+    MachineFunction &MF = MIRBuilder.getMF();
     LLT p0 = LLT::pointer(0, 64);
     LLT s64 = LLT::scalar(64);
+
+    if (IsTailCall) {
+      // TODO: For -tailcallopt tail calls, Offset will need FPDiff like in
+      // ISelLowering.
+      int FI = MF.getFrameInfo().CreateFixedObject(Size, Offset, true);
+      Register FIReg = MRI.createGenericVirtualRegister(p0);
+      MIRBuilder.buildFrameIndex(FIReg, FI);
+      MPO = MachinePointerInfo::getFixedStack(MF, FI);
+      return FIReg;
+    }
+
     Register SPReg = MRI.createGenericVirtualRegister(p0);
     MIRBuilder.buildCopy(SPReg, Register(AArch64::SP));
 
     Register OffsetReg = MRI.createGenericVirtualRegister(s64);
     MIRBuilder.buildConstant(OffsetReg, Offset);
 
     Register AddrReg = MRI.createGenericVirtualRegister(p0);
     MIRBuilder.buildGEP(AddrReg, SPReg, OffsetReg);
 
-    MPO = MachinePointerInfo::getStack(MIRBuilder.getMF(), Offset);
+    MPO = MachinePointerInfo::getStack(MF, Offset);
     return AddrReg;
   }
 
   void assignValueToReg(Register ValVReg, Register PhysReg,
                         CCValAssign &VA) override {
     MIB.addUse(PhysReg, RegState::Implicit);
     Register ExtReg = extendRegister(ValVReg, VA);
     MIRBuilder.buildCopy(PhysReg, ExtReg);
else
       Res = AssignFnVarArg(ValNo, ValVT, LocVT, LocInfo, Flags, State);
 
     StackSize = State.getNextStackOffset();
     return Res;
   }
 
   MachineInstrBuilder MIB;
   CCAssignFn *AssignFnVarArg;
+  bool IsTailCall;
   uint64_t StackSize;
 };
 } // namespace
 
 void AArch64CallLowering::splitToValueTypes(
     const ArgInfo &OrigArg, SmallVectorImpl<ArgInfo> &SplitArgs,
     const DataLayout &DL, MachineRegisterInfo &MRI, CallingConv::ID CallConv) const {
   const AArch64TargetLowering &TLI = *getTLI<AArch64TargetLowering>();
bool AArch64CallLowering::lowerFormalArguments(
   const AArch64TargetLowering &TLI = *getTLI<AArch64TargetLowering>();
   CCAssignFn *AssignFn =
       TLI.CCAssignFnForCall(F.getCallingConv(), /*IsVarArg=*/false);
 
   FormalArgHandler Handler(MIRBuilder, MRI, AssignFn);
   if (!handleAssignments(MIRBuilder, SplitArgs, Handler))
     return false;
 
+  AArch64FunctionInfo *FuncInfo = MF.getInfo<AArch64FunctionInfo>();
+  uint64_t StackOffset = Handler.StackUsed;
   if (F.isVarArg()) {
     auto &Subtarget = MF.getSubtarget<AArch64Subtarget>();
     if (!Subtarget.isTargetDarwin()) {
         // FIXME: we need to reimplement saveVarArgsRegisters from
       // AArch64ISelLowering.
       return false;
     }
 
     // We currently pass all varargs at 8-byte alignment, or 4 in ILP32.
-    uint64_t StackOffset =
-        alignTo(Handler.StackUsed, Subtarget.isTargetILP32() ? 4 : 8);
+    StackOffset = alignTo(Handler.StackUsed, Subtarget.isTargetILP32() ? 4 : 8);
 
     auto &MFI = MIRBuilder.getMF().getFrameInfo();
-    AArch64FunctionInfo *FuncInfo = MF.getInfo<AArch64FunctionInfo>();
     FuncInfo->setVarArgsStackIndex(MFI.CreateFixedObject(4, StackOffset, true));
   }
 
+  // TODO: Port checks for stack to restore for -tailcallopt from ISelLowering.
+  // We need to keep track of the size of function stacks for tail call
+  // optimization. When we tail call, we need to check if the callee's arguments
+  // will fit on the caller's stack. So, whenever we lower formal arguments,
+  // we should keep track of this information, since we might lower a tail call
+  // in this function later.
+  FuncInfo->setBytesInStackArgArea(StackOffset);
+
   auto &Subtarget = MF.getSubtarget<AArch64Subtarget>();
   if (Subtarget.hasCustomCallingConv())
     Subtarget.getRegisterInfo()->UpdateCustomCalleeSavedRegs(MF);
 
   // Move back to the end of the basic block.
   MIRBuilder.setMBB(MBB);
 
   return true;
bool AArch64CallLowering::doCallerAndCalleePassArgsTheSameWay(
   if (MF.getSubtarget<AArch64Subtarget>().hasCustomCallingConv()) {
     TRI->UpdateCustomCallPreservedMask(MF, &CallerPreserved);
     TRI->UpdateCustomCallPreservedMask(MF, &CalleePreserved);
   }
 
   return TRI->regmaskSubsetEqual(CallerPreserved, CalleePreserved);
 }
 
+bool AArch64CallLowering::areCalleeOutgoingArgsTailCallable(
+    CallLoweringInfo &Info, MachineFunction &MF,
+    SmallVectorImpl<ArgInfo> &OutArgs) const {
+  // If there are no outgoing arguments, then we are done.
+  if (OutArgs.empty())
+    return true;
+
+  const Function &CallerF = MF.getFunction();
+  CallingConv::ID CalleeCC = Info.CallConv;
+  CallingConv::ID CallerCC = CallerF.getCallingConv();
+  const AArch64TargetLowering &TLI = *getTLI<AArch64TargetLowering>();
+
+  // We have outgoing arguments. Make sure that we can tail call with them.
+  SmallVector<CCValAssign, 16> OutLocs;
+  CCState OutInfo(CalleeCC, false, MF, OutLocs, CallerF.getContext());
+
+  if (!analyzeArgInfo(OutInfo, OutArgs,
+                      *TLI.CCAssignFnForCall(CalleeCC, Info.IsVarArg))) {
+    LLVM_DEBUG(dbgs() << "... Could not analyze call operands.\n");
+    return false;
+  }
+
+  // Make sure that they can fit on the caller's stack.
+  const AArch64FunctionInfo *FuncInfo = MF.getInfo<AArch64FunctionInfo>();
+  if (OutInfo.getNextStackOffset() > FuncInfo->getBytesInStackArgArea()) {
+    LLVM_DEBUG(dbgs() << "... Cannot fit call operands on caller's stack.\n");
+    return false;
+  }
+
+  // Verify that the parameters in callee-saved registers match.
+  // TODO: Port this over to CallLowering as general code once swiftself is
+  // supported.
+  auto TRI = MF.getSubtarget<AArch64Subtarget>().getRegisterInfo();
+  const uint32_t *CallerPreservedMask = TRI->getCallPreservedMask(MF, CallerCC);
+
+  for (auto &ArgLoc : OutLocs) {
+    // If it's not a register, it's fine.
+    if (!ArgLoc.isRegLoc())
+      continue;
+
+    Register Reg = ArgLoc.getLocReg();
+
+    // Only look at callee-saved registers.
+    if (MachineOperand::clobbersPhysReg(CallerPreservedMask, Reg))
+      continue;
+
+    // TODO: Port the remainder of this check from TargetLowering to support
+    // tail calling swiftself.
+    LLVM_DEBUG(
+        dbgs()
+        << "... Cannot handle callee-saved registers in outgoing args yet.\n");
+    return false;
+  }
+
+  return true;
+}
+
 bool AArch64CallLowering::isEligibleForTailCallOptimization(
     MachineIRBuilder &MIRBuilder, CallLoweringInfo &Info,
-    SmallVectorImpl<ArgInfo> &InArgs) const {
+    SmallVectorImpl<ArgInfo> &InArgs,
+    SmallVectorImpl<ArgInfo> &OutArgs) const {
   CallingConv::ID CalleeCC = Info.CallConv;
   MachineFunction &MF = MIRBuilder.getMF();
   const Function &CallerF = MF.getFunction();
 
   LLVM_DEBUG(dbgs() << "Attempting to lower call as tail call\n");
 
   if (Info.SwiftErrorVReg) {
     // TODO: We should handle this.
bool AArch64CallLowering::isEligibleForTailCallOptimization(
   // true. However, we don't have full tail call support yet. So, continue
   // checking. We want to emit a sibling call.
 
   // I want anyone implementing a new calling convention to think long and hard
   // about this assert.
   assert((!Info.IsVarArg || CalleeCC == CallingConv::C) &&
          "Unexpected variadic calling convention");
 
-  // Look at the incoming values.
+  // Verify that the incoming and outgoing arguments from the callee are
+  // safe to tail call.
   if (!doCallerAndCalleePassArgsTheSameWay(Info, MF, InArgs)) {
     LLVM_DEBUG(
         dbgs()
         << "... Caller and callee have incompatible calling conventions.\n");
     return false;
   }
 
-  // For now, only handle callees that take no arguments.
-  if (!Info.OrigArgs.empty()) {
-    LLVM_DEBUG(
-        dbgs()
-        << "... Cannot tail call callees with outgoing arguments yet.\n");
+  if (!areCalleeOutgoingArgsTailCallable(Info, MF, OutArgs))
     return false;
-  }
 
   LLVM_DEBUG(
       dbgs() << "... Call is eligible for tail call optimization.\n");
   return true;
 }
 
 static unsigned getCallOpcode(const Function &CallerF, bool IsIndirect,
                               bool IsTailCall) {
bool AArch64CallLowering::lowerCall(MachineIRBuilder &MIRBuilder,
   }
 
   if (Info.IsTailCall && MF.getTarget().Options.GuaranteedTailCallOpt) {
     // TODO: Until we lower all tail calls, we should fall back on this.
     LLVM_DEBUG(dbgs() << "Cannot handle -tailcallopt yet.\n");
     return false;
   }
 
-  SmallVector<ArgInfo, 8> SplitArgs;
+  SmallVector<ArgInfo, 8> OutArgs;
   for (auto &OrigArg : Info.OrigArgs) {
-    splitToValueTypes(OrigArg, SplitArgs, DL, MRI, Info.CallConv);
+    splitToValueTypes(OrigArg, OutArgs, DL, MRI, Info.CallConv);
     // AAPCS requires that we zero-extend i1 to 8 bits by the caller.
     if (OrigArg.Ty->isIntegerTy(1))
-      SplitArgs.back().Flags[0].setZExt();
+      OutArgs.back().Flags[0].setZExt();
   }
 
   SmallVector<ArgInfo, 8> InArgs;
   if (!Info.OrigRet.Ty->isVoidTy())
     splitToValueTypes(Info.OrigRet, InArgs, DL, MRI, F.getCallingConv());
 
-  bool IsSibCall = Info.IsTailCall &&
-                   isEligibleForTailCallOptimization(MIRBuilder, Info, InArgs);
+  bool IsSibCall = Info.IsTailCall && isEligibleForTailCallOptimization(
+                                          MIRBuilder, Info, InArgs, OutArgs);
   if (IsSibCall)
     MF.getFrameInfo().setHasTailCall();
 
   // Find out which ABI gets to decide where things go.
   CCAssignFn *AssignFnFixed =
       TLI.CCAssignFnForCall(Info.CallConv, /*IsVarArg=*/false);
   CCAssignFn *AssignFnVarArg =
       TLI.CCAssignFnForCall(Info.CallConv, /*IsVarArg=*/true);
bool AArch64CallLowering::lowerCall(MachineIRBuilder &MIRBuilder,
   MIB.addRegMask(Mask);
 
   if (TRI->isAnyArgRegReserved(MF))
     TRI->emitReservedArgRegCallError(MF);
 
   // Do the actual argument marshalling.
   SmallVector<unsigned, 8> PhysRegs;
   OutgoingArgHandler Handler(MIRBuilder, MRI, MIB, AssignFnFixed,
-                             AssignFnVarArg);
-  if (!handleAssignments(MIRBuilder, SplitArgs, Handler))
+                             AssignFnVarArg, IsSibCall);
+  if (!handleAssignments(MIRBuilder, OutArgs, Handler))
     return false;
 
   // Now we can add the actual call instruction to the correct basic block.
   MIRBuilder.insertInstr(MIB);
 
   // If Callee is a reg, since it is used by a target specific
   // instruction, it must have a register class matching the
   // constraint of that instruction.

llvm/trunk/test/CodeGen/AArch64/GlobalISel/call-translator-tail-call.ll

define void @indirect_tail_call(void()* %func) {
   ; COMMON:   liveins: $x0
   ; COMMON:   [[COPY:%[0-9]+]]:tcgpr64(p0) = COPY $x0
   ; COMMON:   TCRETURNri [[COPY]](p0), 0, csr_aarch64_aapcs, implicit $sp
   tail call void %func()
   ret void
 }
 
 declare void @outgoing_args_fn(i32)
-; Right now, callees with outgoing arguments should not be tail called.
-; TODO: Support this.
 define void @test_outgoing_args(i32 %a) {
   ; COMMON-LABEL: name: test_outgoing_args
   ; COMMON: bb.1 (%ir-block.0):
   ; COMMON:   liveins: $w0
   ; COMMON:   [[COPY:%[0-9]+]]:_(s32) = COPY $w0
-  ; COMMON:   ADJCALLSTACKDOWN 0, 0, implicit-def $sp, implicit $sp
   ; COMMON:   $w0 = COPY [[COPY]](s32)
-  ; COMMON:   BL @outgoing_args_fn, csr_aarch64_aapcs, implicit-def $lr, implicit $sp, implicit $w0
-  ; COMMON:   ADJCALLSTACKUP 0, 0, implicit-def $sp, implicit $sp
-  ; COMMON:   RET_ReallyLR
+  ; COMMON:   TCRETURNdi @outgoing_args_fn, 0, csr_aarch64_aapcs, implicit $sp, implicit $w0
   tail call void @outgoing_args_fn(i32 %a)
   ret void
 }
 
+; Verify that we create frame indices for memory arguments in tail calls.
+; We get a bunch of copies here which are unused and thus eliminated. So, let's
+; just focus on what matters, which is that we get a G_FRAME_INDEX.
+declare void @outgoing_stack_args_fn(<4 x half>)
+define void @test_outgoing_stack_args([8 x <2 x double>], <4 x half> %arg) {
+  ; COMMON-LABEL: name: test_outgoing_stack_args
+  ; COMMON:   [[FRAME_INDEX:%[0-9]+]]:_(p0) = G_FRAME_INDEX %fixed-stack.0
+  ; COMMON:   [[LOAD:%[0-9]+]]:_(<4 x s16>) = G_LOAD [[FRAME_INDEX]](p0) :: (invariant load 8 from %fixed-stack.0, align 1)
+  ; COMMON:   $d0 = COPY [[LOAD]](<4 x s16>)
+  ; COMMON:   TCRETURNdi @outgoing_stack_args_fn, 0, csr_aarch64_aapcs, implicit $sp, implicit $d0
+  tail call void @outgoing_stack_args_fn(<4 x half> %arg)
+  ret void
+}
+
+; Verify that we don't tail call when we cannot fit arguments on the caller's
+; stack.
+declare i32 @too_big_stack(i64 %x0, i64 %x1, i64 %x2, i64 %x3, i64 %x4, i64 %x5, i64 %x6, i64 %x7, i8 %c, i16 %s)
+define i32 @test_too_big_stack() {
+  ; COMMON-LABEL: name: test_too_big_stack
+  ; COMMON-NOT: TCRETURNdi
+  ; COMMON-NOT: TCRETURNri
+  ; COMMON: BL @too_big_stack
+  ; COMMON-DAG: RET_ReallyLR
+entry:
+  %call = tail call i32 @too_big_stack(i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i64 undef, i8 8, i16 9)
+  ret i32 %call
+}
+
 ; Right now, we don't want to tail call callees with nonvoid return types, since
 ; call lowering will insert COPYs after the call.
 ; TODO: Support this.
 declare i32 @nonvoid_ret()
 define i32 @test_nonvoid_ret() {
   ; COMMON-LABEL: name: test_nonvoid_ret
   ; COMMON: bb.1 (%ir-block.0):
   ; COMMON:   TCRETURNdi @nonvoid_ret, 0, csr_aarch64_aapcs, implicit $sp

llvm/trunk/test/CodeGen/AArch64/arm64-call-tailcalls.ll

 ; RUN: llc < %s -mtriple=arm64-apple-ios7.0 | FileCheck %s
+; RUN: llc -global-isel < %s -mtriple=arm64-apple-ios7.0 | FileCheck %s
 
 @t = weak global i32 ()* null
 @x = external global i32, align 4
 
 define void @t2() {
 ; CHECK-LABEL: t2:
 ; CHECK: adrp x[[GOTADDR:[0-9]+]], _t@GOTPAGE
 ; CHECK: ldr  x[[ADDR:[0-9]+]], [x[[GOTADDR]], _t@GOTPAGEOFF]

llvm/trunk/test/CodeGen/AArch64/speculation-hardening.ll

-; RUN: sed -e 's/SLHATTR/speculative_load_hardening/' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu | FileCheck %s --check-prefixes=CHECK,SLH,NOGISELSLH --dump-input-on-failure
-; RUN: sed -e 's/SLHATTR//' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu | FileCheck %s --check-prefixes=CHECK,NOSLH,NOGISELNOSLH --dump-input-on-failure
-; RUN: sed -e 's/SLHATTR/speculative_load_hardening/' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -global-isel | FileCheck %s --check-prefixes=CHECK,SLH,GISELSLH --dump-input-on-failure
-; RUN sed -e 's/SLHATTR//' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -global-isel | FileCheck %s --check-prefixes=CHECK,NOSLH,GISELNOSLH --dump-input-on-failure
-; RUN: sed -e 's/SLHATTR/speculative_load_hardening/' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -fast-isel | FileCheck %s --check-prefixes=CHECK,SLH,NOGISELSLH --dump-input-on-failure
-; RUN: sed -e 's/SLHATTR//' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -fast-isel | FileCheck %s --check-prefixes=CHECK,NOSLH,NOGISELNOSLH --dump-input-on-failure
+; RUN: sed -e 's/SLHATTR/speculative_load_hardening/' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu | FileCheck %s --check-prefixes=CHECK,SLH --dump-input-on-failure
+; RUN: sed -e 's/SLHATTR//' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu | FileCheck %s --check-prefixes=CHECK,NOSLH --dump-input-on-failure
+; RUN: sed -e 's/SLHATTR/speculative_load_hardening/' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -global-isel | FileCheck %s --check-prefixes=CHECK,SLH --dump-input-on-failure
+; RUN sed -e 's/SLHATTR//' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -global-isel | FileCheck %s --check-prefixes=CHECK,NOSLH --dump-input-on-failure
+; RUN: sed -e 's/SLHATTR/speculative_load_hardening/' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -fast-isel | FileCheck %s --check-prefixes=CHECK,SLH --dump-input-on-failure
+; RUN: sed -e 's/SLHATTR//' %s | llc -verify-machineinstrs -mtriple=aarch64-none-linux-gnu -fast-isel | FileCheck %s --check-prefixes=CHECK,NOSLH --dump-input-on-failure
 
 define i32 @f(i8* nocapture readonly %p, i32 %i, i32 %N) local_unnamed_addr SLHATTR {
 ; CHECK-LABEL: f
 entry:
 ; SLH:  cmp sp, #0
 ; SLH:  csetm x16, ne
 ; NOSLH-NOT:  cmp sp, #0
 ; NOSLH-NOT:  csetm x16, ne

 ; NOSLH-NOT:  and [[TMPREG]], [[TMPREG]], x16
 ; NOSLH-NOT:  mov sp, [[TMPREG]]
   ret i32 %retval.0
 }
 
 ; Make sure that for a tail call, taint doesn't get put into SP twice.
 define i32 @tail_caller(i32 %a) local_unnamed_addr SLHATTR {
 ; CHECK-LABEL: tail_caller:
-; NOGISELSLH:     mov [[TMPREG:x[0-9]+]], sp
-; NOGISELSLH:     and [[TMPREG]], [[TMPREG]], x16
-; NOGISELSLH:     mov sp, [[TMPREG]]
-; NOGISELNOSLH-NOT:     mov [[TMPREG:x[0-9]+]], sp
-; NOGISELNOSLH-NOT:     and [[TMPREG]], [[TMPREG]], x16
-; NOGISELNOSLH-NOT:     mov sp, [[TMPREG]]
-; GISELSLH:     mov [[TMPREG:x[0-9]+]], sp
-; GISELSLH:     and [[TMPREG]], [[TMPREG]], x16
-; GISELSLH:     mov sp, [[TMPREG]]
-; GISELNOSLH-NOT:     mov [[TMPREG:x[0-9]+]], sp
-; GISELNOSLH-NOT:     and [[TMPREG]], [[TMPREG]], x16
-; GISELNOSLH-NOT:     mov sp, [[TMPREG]]
-;  GlobalISel doesn't optimize tail calls (yet?), so only check that
-;  cross-call taint register setup code is missing if a tail call was
-;  actually produced.
-; NOGISELSLH:     b tail_callee
-; GISELSLH:       bl tail_callee
-; GISELSLH:       cmp sp, #0
+; SLH:     mov [[TMPREG:x[0-9]+]], sp
+; SLH:     and [[TMPREG]], [[TMPREG]], x16
+; SLH:     mov sp, [[TMPREG]]
+; NOSLH-NOT:     mov [[TMPREG:x[0-9]+]], sp
+; NOSLH-NOT:     and [[TMPREG]], [[TMPREG]], x16
+; NOSLH-NOT:     mov sp, [[TMPREG]]
+; SLH:     b tail_callee
 ; SLH-NOT:        cmp sp, #0
   %call = tail call i32 @tail_callee(i32 %a)
   ret i32 %call
 }
 
 declare i32 @tail_callee(i32) local_unnamed_addr
 
 ; Verify that no cb(n)z/tb(n)z instructions are produced when implementing

llvm/trunk/test/CodeGen/AArch64/tailcall-string-rvo.ll

 ; RUN: llc -relocation-model=static -verify-machineinstrs -O2 < %s | FileCheck %s
+; RUN: llc -relocation-model=static -verify-machineinstrs -global-isel -O2 < %s | FileCheck %s
 
 ; The call to function TestBar should be a tail call, when in C++ the string
 ; `ret` is RVO returned.
 ; string TestFoo() {
 ;   string ret = undef;
 ;   TestBar(&ret);  // tail call optimized
 ;   return ret;
 ; }

llvm/trunk/test/CodeGen/AArch64/tailcall_misched_graph.ll

-; RUN: llc -mcpu=cyclone -debug-only=machine-scheduler < %s 2>&1 | FileCheck %s
+; RUN: llc -mcpu=cyclone -debug-only=machine-scheduler < %s 2>&1 | FileCheck %s --check-prefixes=COMMON,SDAG
+; RUN: llc -mcpu=cyclone -global-isel -debug-only=machine-scheduler < %s 2>&1 | FileCheck %s --check-prefixes=COMMON,GISEL
 
 ; REQUIRES: asserts
 
 target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
 target triple = "arm64-apple-ios7.0.0"
 
 define void @caller2(i8* %a0, i8* %a1, i8* %a2, i8* %a3, i8* %a4, i8* %a5, i8* %a6, i8* %a7, i8* %a8, i8* %a9) {
 entry:
   tail call void @callee2(i8* %a1, i8* %a2, i8* %a3, i8* %a4, i8* %a5, i8* %a6, i8* %a7, i8* %a8, i8* %a9, i8* %a0)
   ret void
 }
 
 declare void @callee2(i8*, i8*, i8*, i8*, i8*,
                       i8*, i8*, i8*, i8*, i8*)
 
 ; Make sure there is a dependence between the load and store to the same stack
 ; location during a tail call. Tail calls clobber the incoming argument area and
 ; therefore it is not safe to assume argument locations are invariant.
 ; PR23459 has a test case that we where miscompiling because of this at the
 ; time.
 
-; CHECK: Frame Objects
-; CHECK:  fi#-4: {{.*}} fixed, at location [SP+8]
-; CHECK:  fi#-3: {{.*}} fixed, at location [SP]
-; CHECK:  fi#-2: {{.*}} fixed, at location [SP+8]
-; CHECK:  fi#-1: {{.*}} fixed, at location [SP]
-
-; CHECK:  [[VRA:%.*]]:gpr64 = LDRXui %fixed-stack.3
-; CHECK:  [[VRB:%.*]]:gpr64 = LDRXui %fixed-stack.2
-; CHECK:  STRXui %{{.*}}, %fixed-stack.0
-; CHECK:  STRXui [[VRB]]{{[^,]*}}, %fixed-stack.1
+; COMMON: Frame Objects
+; COMMON:  fi#-4: {{.*}} fixed, at location [SP+8]
+; COMMON:  fi#-3: {{.*}} fixed, at location [SP]
+; COMMON:  fi#-2: {{.*}} fixed, at location [SP+8]
+; COMMON:  fi#-1: {{.*}} fixed, at location [SP]
+
+; The order that these appear in differes in GISel than SDAG, but the
+; dependency relationship still holds.
+; COMMON:  [[VRA:%.*]]:gpr64 = LDRXui %fixed-stack.3
+; COMMON:  [[VRB:%.*]]:gpr64 = LDRXui %fixed-stack.2
+; SDAG:  STRXui %{{.*}}, %fixed-stack.0
+; SDAG:  STRXui [[VRB]]{{[^,]*}}, %fixed-stack.1
+; GISEL:  STRXui [[VRB]]{{[^,]*}}, %fixed-stack.1
+; GISEL:  STRXui %{{.*}}, %fixed-stack.0
 
 ; Make sure that there is an dependence edge between fi#-2 and fi#-4.
 ; Without this edge the scheduler would be free to move the store accross the load.
 
-; CHECK: SU({{.*}}):   [[VRB]]:gpr64 = LDRXui %fixed-stack.2
-; CHECK-NOT: SU
-; CHECK:  Successors:
-; CHECK:   SU([[DEPSTOREB:.*]]): Ord  Latency=0
-; CHECK:   SU([[DEPSTOREA:.*]]): Ord  Latency=0
+; COMMON: SU({{.*}}):   [[VRB]]:gpr64 = LDRXui %fixed-stack.2
+; COMMON-NOT: SU
+; COMMON:  Successors:
+; COMMON:   SU([[DEPSTOREB:.*]]): Ord  Latency=0
+; COMMON:   SU([[DEPSTOREA:.*]]): Ord  Latency=0
+
+; GlobalISel outputs DEPSTOREB before DEPSTOREA, but the dependency relationship
+; still holds.
+; SDAG: SU([[DEPSTOREA]]):   STRXui %{{.*}}, %fixed-stack.0
+; SDAG: SU([[DEPSTOREB]]):   STRXui %{{.*}}, %fixed-stack.1
 
-; CHECK: SU([[DEPSTOREA]]):   STRXui %{{.*}}, %fixed-stack.0
-; CHECK: SU([[DEPSTOREB]]):   STRXui %{{.*}}, %fixed-stack.1
+; GISEL: SU([[DEPSTOREB]]):   STRXui %{{.*}}, %fixed-stack.0
+; GISEL: SU([[DEPSTOREA]]):   STRXui %{{.*}}, %fixed-stack.1