MIR Statepoint refactoring. Part 4: ISEL changes.

[dantrushin]

Summary

Implement variadic-def statepoint lowering.
Assign N base/derived non-constant pointers to VRegs, pass others
on stack slots, as in old implementation.
So new statepoint looks like this:

    
reloc1,reloc2,... = STATEPOINT ..., base1, derived1<tied-def0>, base2, derived2<tied-def1>, ...

N is limited by the maximal number of tied registers machine
instruction can have (15 at the moment).
If any of gc.relocates are not in the same basic block (as the case for
invoke statepoint), use old spilling scheme too.

Full change set is available at D81603.

[reames]

Denis, I'm having a really hard time wrapping my head around this code, even knowing what it is supposed to do. We need to clean this up a bit; it's not in a state where I can approve it.

A couple of suggestions for you:

1. Since some of the complexity here seems specific to invokes, let's restrict the first patch to call instructions. If we see an invoke, we can simply use the stack lowering.
2. The handling around needing the first N GC values in a sorted order seems overly complicated. I was also unsure about the correctness of derived pointer handling when base != derived. What would you think of explicitly keeping track of which operands we've created that are tied def/use pairs? We could visit operands in the current order and then use the stack lowering when the size of the defs list is larger than our threshold.
3. Testing wise, I think it make sense to start with a new test file, build out a set of tests which exercise this, and only update all of the tests once we have something which is expected to work end to end. We're not there yet.

llvm/lib/CodeGen/SelectionDAG/StatepointLowering.cpp

71

For consistency with the above, please call this "-use-registers-for-gc-values".

[dantrushin]

In D81648#2104721, @reames wrote:

1. Since some of the complexity here seems specific to invokes, let's restrict the first patch to call instructions. If we see an invoke, we can simply use the stack lowering.

Are you talking about that cross-block usage (CopyToRegs/CopyFromRegs) stuff?
All this complexity is already there for gc.result.
For gc.relocate I do exactly the same, only I have to use separate map, because ValueMap is occupied by gc.result
Some of the complexities was required to handle undef/duplicate gc.relocate cases. Maybe now, when we have undef handling upstream, I'll be able to simplify it a bit

2. The handling around needing the first N GC values in a sorted order seems overly complicated. I was also unsure about the correctness of derived pointer handling when base != derived. What would you think of explicitly keeping track of which operands we've created that are tied def/use pairs? We could visit operands in the current order and then use the stack lowering when the size of the defs list is larger than our threshold.

I can track them all the way through lowering, yes (in FunctionLoweringInfo), but the problem is that I need to tie registers in InstrEmitter. Since STATEPOINT does not have meaningful MachineDescription, I need to get this information from somewhere else.
I cannot store that mapping in SDNode, and InstrEmitter does not have access to SelectionDAG or FunctionLoweringInfo, so I have to figure it out from the statepoint SDNode itself.
"First N" approach is the only one I can think of.
I can map gc values to registers in order until I encounter something 'non-register' and move that sorting stuff to say, CodeGenPrepare. Does it sounds better?
Could you clarify why you're worrying about base!=derived case? Base pointers are not interesting at all - they all can be assigned to registers (no need to tie them). It is base==derived case which is interesting/complicated. I've been worrying about LLVM's
ability to handle to identical uses only one of which is tied to def, but apparently it works.

3. Testing wise, I think it make sense to start with a new test file, build out a set of tests which exercise this, and only update all of the tests once we have something which is expected to work end to end. We're not there yet.

Could you explain? This change is last one in series and completes the whole sequence. What's missing here?

[dantrushin]

Version without GC pointers sorting.

[dantrushin]

Do not handle statepoints with cross-block relocates, spill their args as before.

[dantrushin]

Correctly handle vectors of pointers.

[reames]

Initial comments, continue to look for other suggestions.

llvm/lib/CodeGen/SelectionDAG/StatepointLowering.cpp

230

Rebase on commit b172cd781.

587

Please replace your modifications to this function with the following:

DenseSet<SDValue> LowerAsVReg;
if (UseRegistersForGCPointers && isa<CallInst>(SI.StatepointInstr)) {

for (unsigned i = 0; i < SI.Bases.size(); ++i) {
  if (willDirectlyLower(getValue(SI.Bases[i])) || <is vector type>
    continue;
  LowerAsVReg.insert(getValue(SI.Bases[i]));
  if (LowerAsVReg.size() == N) break;

}
}

auto requireSpillSlot = [&](const Value *V) {

if (isGCValue(V))
   return LowerAsVReg.contains(getValue(V)));
return !(LiveInDeopt || UseRegistersForDeoptValues);

};

[reames]

Required Change - Please introduce a runtime flag which controls how many values are handled via vregs. Default this value to zero. This will remove all existing test diffs; if it doesn't you have a bug. Then introduce a new test file, optional copied from an existing one, called statepoint-gc-regs.ll which enumerates sufficient coverage for the new feature.

This is not a suggestion, it is a requirement. I specifically need to see a zero diff in the existing tests to have confidence in the changes.

[reames]

I've spent several hours today wrapping my head around this patch. I think I've found a material simplification which should greatly simplify the code.

You track the def offset for each pointer of interest. You have plumbed through the merge value code with the purpose of being able to get to the SDNode corresponding to the actual statepoint. In the gc.relocate code, you combine these two pieces into enough information to extract the relevant definition from the statepoint. Under the assumption that we're only talking about relocates within a single basic block, this is correct.

My suggested simplification is the following. Rather than maintaining a map to offsets in the STATEPOINT node, simply maintain a map from pointer to SDValue representing the particular output of the node. Doing this means that you do not need any of the merge value code, all of the result propagation remains the same, and the gc.relocate changes are isolated. You would need to export these values in the cross block case, but this is fairly easily handled by copying each output to a vreg during lowering of the statepoint and having the vreg be the value tracked in the FuncInfo.

If my written description is not clear, or you disagree with my conclusions, let's find a time to talk offline.

In addition to this larger change, I have added a couple of smaller FIXMEs throughout the day. I've also been landing NFCs where doing so made sense in the process of understanding your changes. You should expect a non-trivial, but not particularly difficult either, rebase.

You should consider this comment a blocking comment. I do not intend to spend further time on this review until either a) the changes noted today have been made, or b) we've talked offline and we've agreed to an alternate approach.

llvm/include/llvm/CodeGen/FunctionLoweringInfo.h

106 ↗

(On Diff #275439)

Marker (see overall comment)

[dantrushin]

Changed DerivedPtrMap and moved it to StatepointLowering.h as it only
needed for local gc.relocate processing;
Rebased on tip, rewrote lowerStatepointMetaArgs as been told;
Deleted test mods, will add separate test later;

[reames]

On first skim, looks much much better. I'm going to do a detailed pass through, but thank you for making the major design change requested.

[reames]

Detailed comments on the new implementation. These are on the whole minor. Remember to add your new test file.

I need to look more closely at the code outside StatepointLowering, will do that in a separate comment shortly.

llvm/lib/CodeGen/SelectionDAG/StatepointLowering.cpp

562

I don't think the isNonIntegralPointer check is needed here. Can you either remove or explain?

602

This map population isn't related to stack reservation. Please either separate it into it's own loop, or merge it with the population of the LowerAsVReg set above.

(If I understood this, it's simply the index into a flattened list of the values to be spilled in registers.)

603

I'd suggest simply not adding anything for the map for the spill case. Using contains checks are more idiomatic of the purpose.

637–641

This block of code is functionally broken when base != derived. You have only added the vreg information for the derived, and would need to spill the base so that the GC can find it. The fix is trivial, pass "true" for the base case when Base != Derived.

(Also, this is the probably the profitable lowering, so don't be tempted to add the base to a reg.)

681–683

See comment above about count checks. (Contains)

731

Isn't this simply Ptr2ResNo.size()? (after the change suggest above)

847

Again, contains check, not default value.

1116

Add a comment describing what this does.

[reames]

(Comments on code outside StatepointLowering. Minor +1 question)

Overall, if you address the style comments from the this and the last comment, I think we're close to an LGTM.

llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp

225

This line looks unnecessary provided the defs list is marked variadic. If you want to leave it for the moment, that's fine, but I think we can clean this up in a follow up change.

(To be specific, OutOperandList in Target.td for STATEPOINT can be variable_ops, and this line disappears.)

llvm/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp

128

I don't understand the implication of this change, can you explain? On the surface, this looks likely okay, but not understanding the reason for the change is bothering me.

llvm/lib/CodeGen/TargetLoweringBase.cpp

1055

Please add a comment which explains you're relying on all the definitions coming before any FI and thus the indices not changing.

[dantrushin]

llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp

225

II here is MCInstrDesc.
Statepoint has 0 defs in it MCIntrDesc. (It is MachineInstr::getNumDefs which work correctly for variadic outs).
So I really need it here

llvm/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp

128

Again, MCInstr.getNumDefs() for statepoint always returns 0.
This code assumes that all 'dynamic' defs are implicit, so it accesses II.ImplicitDefs without any checks.
But statepoint has no implicit defs, so for it II.ImplicitDefs is NULL.
That functions does the the same thing, but with proper checks.

llvm/lib/CodeGen/SelectionDAG/StatepointLowering.cpp

637–641

vreg information is only needed for derived pointers to relocate them.
(base pointers are not relocated). So I could just assign all bases to VRegs (if at all possible).
I just did not want to have same value both in stack slot and vreg.

So IMHO the better approach is to check if base is found in LowerAsVReg (or that is what you mean?)

As for profitability, our goal is to shift as much spilling as possible to register allocator , doesn't it?

[dantrushin]

Addressed comments.

Now it appears that Ptr2ResNo and LowerAsVReg somewhat duplicate
each other. Perhaps I can rename Ptr2ResNo to LowerAsVReg and get
rid of old LowerAsVReg? Or it will hurt readability?

[reames]

I took some time today to apply this patch with the attention of addressing some of the unaddressed comments and landing it to unblock progress here. However, I hit two problems.

First, a trivial test case crashes with this flag enabled.

define void @in_register(i8 addrspace(1)* %a) #1 gc "statepoint-example" {

%sp = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* undef, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i8 addrspace(1)* %a)]
%a1 = call i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %sp, i32 0, i32 0)
call void @use(i8 addrspace(1)* %a1)
ret void

}

Second, there appears to be a semantic problem around the handling of base vs derived slots unless we *always* spill the base. We can't tie both uses to a single def. This may warrant some offline discussion.

Denis, please address all of the previous applied comments, including writing targetting tests.

llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp

225

Please read the line above your change with the explicit handling for variadic defs.

(Not a blocking item. I'll fix after submit if needed)

[dantrushin]

Another attempt to submit update.

Additionally in this version:

• Added tests (used multiple checks per function to better demonstrate how it works in practice - after ISel, after RA and resulting asm).

• Changed boolean flag to counter to simplify possible landing ahead of D81646.

• Use single map LowerAsVreg instead of pair of set LowerAsVReg and map Ptr2ResNo.

NOTES:

• This can be landed only after D81645 (Part 1: Basic MI level changes).
• If landed before D81646 (Part 2: Operand Folding), vreg limit cannot be raised too high or regalloc will run out of registers.
• If landed after D81647 (Part 3; Spill GC Ptr regs), last run line in test must be updated to allow passing pointers in CSRs.

[dantrushin]

In D81648#2146051, @reames wrote:

Second, there appears to be a semantic problem around the handling of base vs derived slots unless we *always* spill the base. We can't tie both uses to a single def. This may warrant some offline discussion.

Yes, we can't. But apparently, we do not need to.
Basically, that tied-ness information is only needed to glue two live ranges into one during SSA flattening. And it does not matter how many uses end live range at the same instruction.
Luckily, LLVM appears to handle it properly.

But if included tests (and checks) do not make you believe it works, we should throw all this stuff away and try some different approach.

llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp

225

First, please notice !MF->getTarget().usesPhysRegsForValues() there.

/// True if the target uses physical regs (as nearly all targets do). False
/// for stack machines such as WebAssembly and other virtual-register
/// machines. If true, all vregs must be allocated before PEI. If false, then
/// callee-save register spilling and scavenging are not needed or used. If
/// false, implicitly defined registers will still be assumed to be physical
/// registers, except that variadic defs will be allocated vregs.
virtual bool usesPhysRegsForValues() const { return true; }

This function returns true for all targets except WebAssembly, so HasVRegVariadicDefs is
always false on e.g., X86

Second, looks like variadicOpsAreDefs as it was implemented for ARM does not means exactly what I wanted.
Looking at MCInstrDesc::hasDefOfPhysReg I see that variadic defs are supposed to go after static uses.
Even though this would work for current definition of STATEPOINT - because it have no static uses
(and this is somewhat questionable from my POW) - I want STATEPOINT defs to come first, just like in all
ordinary instructions.
WebAssembly abused this attribute for its own purposes, but I did not want to.
In fact, for our purposes we can compute isVariadicDefs property same way as isVariadic is computed,
without any need for additional attributes. I'm not sure if isVariadicDefs would work for WebAssembly/ARM
so I decided not to dig into this until we have first variadic-defs instruction in tree.

[dantrushin]

Clear DerivedPtrMap at new statepoint.

[reames]

LGTM

I'm approving this patch not because it's perfect, but because it doesn't break anything in the existing code and iterating here is no longer productive. Once this lands, we can continue the discussion about base/derived handling with specific concrete examples.

Since Denis is out on vacation, I plan to land this patch and then iterate on top with a couple of minor improvements.

[reames]

In D81648#2150053, @dantrushin wrote:

In D81648#2146051, @reames wrote:

Second, there appears to be a semantic problem around the handling of base vs derived slots unless we *always* spill the base. We can't tie both uses to a single def. This may warrant some offline discussion.

Yes, we can't. But apparently, we do not need to.
Basically, that tied-ness information is only needed to glue two live ranges into one during SSA flattening. And it does not matter how many uses end live range at the same instruction.
Luckily, LLVM appears to handle it properly.

But if included tests (and checks) do not make you believe it works, we should throw all this stuff away and try some different approach.

I filed a bug with a detailed description of the problem and one suggestion as to how to approach. See https://bugs.llvm.org/show_bug.cgi?id=46917, let's take discussion there (or offline).

llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp

if (RegisterSDNode *RN = dyn_cast<RegisterSDNode>(Node->getOperand(I - 1)))
         continue;
     NumImpUses = N - I;
     break;
   }
 
   return N;
 }
 
+/// Return starting index of GC operand list.
+// FIXME: need a better place for this. Put it in StackMaps?
+static unsigned getStatepointGCArgStartIdx(MachineInstr *MI) {
+  assert(MI->getOpcode() == TargetOpcode::STATEPOINT &&
+         "STATEPOINT node expected");
+  unsigned OperIdx = StatepointOpers(MI).getNumDeoptArgsIdx();
+  unsigned NumDeopts = MI->getOperand(OperIdx).getImm();
+  // At this point stack references has not been lowered yet, so they
+  // take single operand.
+  ++OperIdx;
+  while (NumDeopts--) {
+    MachineOperand &MO = MI->getOperand(OperIdx);
+    if (MO.isImm() && MO.getImm() == StackMaps::ConstantOp) {
+      ++OperIdx;
+      assert(MI->getOperand(OperIdx).isImm() &&
+             "Unexpected statepoint operand");
+    }
+    ++OperIdx;
+  }
+  return OperIdx;
+}
+
 /// EmitCopyFromReg - Generate machine code for an CopyFromReg node or an
 /// implicit physical register output.
 void InstrEmitter::
 EmitCopyFromReg(SDNode *Node, unsigned ResNo, bool IsClone, bool IsCloned,
                 Register SrcReg, DenseMap<SDValue, Register> &VRBaseMap) {
   Register VRBase;
   if (SrcReg.isVirtual()) {
     // Just use the input register directly!
void InstrEmitter::CreateVirtualRegisters(SDNode *Node,
                                        DenseMap<SDValue, Register> &VRBaseMap) {
   assert(Node->getMachineOpcode() != TargetOpcode::IMPLICIT_DEF &&
          "IMPLICIT_DEF should have been handled as a special case elsewhere!");
 
   unsigned NumResults = CountResults(Node);
   bool HasVRegVariadicDefs = !MF->getTarget().usesPhysRegsForValues() &&
                              II.isVariadic() && II.variadicOpsAreDefs();
   unsigned NumVRegs = HasVRegVariadicDefs ? NumResults : II.getNumDefs();
+  if (Node->getMachineOpcode() == TargetOpcode::STATEPOINT)
+    NumVRegs = NumResults;
   for (unsigned i = 0; i < NumVRegs; ++i) {
     // If the specific node value is only used by a CopyToReg and the dest reg
     // is a vreg in the same register class, use the CopyToReg'd destination
     // register instead of creating a new vreg.
     Register VRBase;
     const TargetRegisterClass *RC =
       TRI->getAllocatableClass(TII->getRegClass(II, i, TRI, *MF));
     // Always let the value type influence the used register class. The
if (Opc == TargetOpcode::STACKMAP || Opc == TargetOpcode::PATCHPOINT) {
     // convention. However, to simplify runtime support, they clobber the same
     // scratch registers as AnyRegCC.
     unsigned CC = CallingConv::AnyReg;
     if (Opc == TargetOpcode::PATCHPOINT) {
       CC = Node->getConstantOperandVal(PatchPointOpers::CCPos);
       NumDefs = NumResults;
     }
     ScratchRegs = TLI->getScratchRegisters((CallingConv::ID) CC);
+  } else if (Opc == TargetOpcode::STATEPOINT) {
+    NumDefs = NumResults;
   }
 
   unsigned NumImpUses = 0;
   unsigned NodeOperands =
     countOperands(Node, II.getNumOperands() - NumDefs, NumImpUses);
   bool HasVRegVariadicDefs = !MF->getTarget().usesPhysRegsForValues() &&
                              II.isVariadic() && II.variadicOpsAreDefs();
   bool HasPhysRegOuts = NumResults > NumDefs &&
for (SDNode *F = Node->getGluedUser(); F; F = F->getGluedUser()) {
         }
     }
   }
 
   // Finally mark unused registers as dead.
   if (!UsedRegs.empty() || II.getImplicitDefs() || II.hasOptionalDef())
     MIB->setPhysRegsDeadExcept(UsedRegs, *TRI);
 
+  // STATEPOINT is too 'dynamic' to have meaningful machine description.
+  // We have to manually tie operands.
+  if (Opc == TargetOpcode::STATEPOINT && NumDefs > 0) {
+    assert(!HasPhysRegOuts && "STATEPOINT mishandled");
+    MachineInstr *MI = MIB;
+    unsigned Def = 0;
+    unsigned Use = getStatepointGCArgStartIdx(MI) + 1;
+    while (Def < NumDefs) {
+      if (MI->getOperand(Use).isReg())
+        MI->tieOperands(Def++, Use);
+      Use += 2;
+    }
+  }
+
   // Run post-isel target hook to adjust this instruction if needed.
   if (II.hasPostISelHook())
     TLI->AdjustInstrPostInstrSelection(*MIB, Node);
 }
 
 /// EmitSpecialNode - Generate machine code for a target-independent node and
 /// needed dependencies.
 void InstrEmitter::

llvm/lib/CodeGen/SelectionDAG/ScheduleDAGSDNodes.cpp

if (Register::isVirtualRegister(Reg))
     return;
 
   unsigned ResNo = User->getOperand(2).getResNo();
   if (Def->getOpcode() == ISD::CopyFromReg &&
       cast<RegisterSDNode>(Def->getOperand(1))->getReg() == Reg) {
     PhysReg = Reg;
   } else if (Def->isMachineOpcode()) {
     const MCInstrDesc &II = TII->get(Def->getMachineOpcode());
-    if (ResNo >= II.getNumDefs() &&
-        II.ImplicitDefs[ResNo - II.getNumDefs()] == Reg)
+    if (ResNo >= II.getNumDefs() && II.hasImplicitDefOfPhysReg(Reg))
       PhysReg = Reg;
   }
 
   if (PhysReg != 0) {
     const TargetRegisterClass *RC =
         TRI->getMinimalPhysRegClass(Reg, Def->getSimpleValueType(ResNo));
     Cost = RC->getCopyCost();
   }

llvm/lib/CodeGen/SelectionDAG/StatepointLowering.h

public:
   }
 
   bool isStackSlotAllocated(int Offset) {
     assert(Offset >= 0 && Offset < (int)AllocatedStackSlots.size() &&
            "out of bounds");
     return AllocatedStackSlots.test(Offset);
   }
 
+  /// For each statepoint keep mapping from original derived pointer to
+  /// the statepoint node result defining its new value.
+  DenseMap<const Value *, SDValue> DerivedPtrMap;
+
 private:
   /// Maps pre-relocation value (gc pointer directly incoming into statepoint)
   /// into it's location (currently only stack slots)
   DenseMap<SDValue, SDValue> Locations;
 
   /// A boolean indicator for each slot listed in the FunctionInfo as to
   /// whether it has been used in the current statepoint.  Since we try to
   /// preserve stack slots across safepoints, there can be gaps in which

llvm/lib/CodeGen/SelectionDAG/StatepointLowering.cpp

 STATISTIC(NumOfStatepoints, "Number of statepoint nodes encountered");
 STATISTIC(StatepointMaxSlotsRequired,
           "Maximum number of stack slots required for a singe statepoint");
 
 cl::opt<bool> UseRegistersForDeoptValues(
     "use-registers-for-deopt-values", cl::Hidden, cl::init(false),
     cl::desc("Allow using registers for non pointer deopt args"));
 
+cl::opt<unsigned> MaxRegistersForGCPointers(
+    "max-registers-for-gc-values", cl::Hidden, cl::init(0),
+    cl::desc("Max number of VRegs allowed to pass GC pointer meta args in"));
+
 static void pushStackMapConstant(SmallVectorImpl<SDValue>& Ops,
                                  SelectionDAGBuilder &Builder, uint64_t Value) {
   SDLoc L = Builder.getCurSDLoc();
   Ops.push_back(Builder.DAG.getTargetConstant(StackMaps::ConstantOp, L,
                                               MVT::i64));
   Ops.push_back(Builder.DAG.getTargetConstant(Value, L, MVT::i64));
 }
 
 void StatepointLoweringState::startNewStatepoint(SelectionDAGBuilder &Builder) {
   // Consistency check
   assert(PendingGCRelocateCalls.empty() &&
          "Trying to visit statepoint before finished processing previous one");
   Locations.clear();
   NextSlotToAllocate = 0;
   // Need to resize this on each safepoint - we need the two to stay in sync and
   // the clear patterns of a SelectionDAGBuilder have no relation to
   // FunctionLoweringInfo.  Also need to ensure used bits get cleared.
   AllocatedStackSlots.clear();
   AllocatedStackSlots.resize(Builder.FuncInfo.StatepointStackSlots.size());
+  DerivedPtrMap.clear();
 }
 
 void StatepointLoweringState::clear() {
   Locations.clear();
   AllocatedStackSlots.clear();
+  DerivedPtrMap.clear();
   assert(PendingGCRelocateCalls.empty() &&
          "cleared before statepoint sequence completed");
 }
 
 SDValue
 StatepointLoweringState::allocateStackSlot(EVT ValueType,
                                            SelectionDAGBuilder &Builder) {
   NumSlotsAllocatedForStatepoints++;
static Optional<int> findPreviousSpillSlot(const Value *Val,
   // put handling of simple modifications in this function like it's done
   // for bitcasts we might end up reserving i's slot for 'i+1' because order in
   // which we visit values is unspecified.
 
   // Don't know any information about this instruction
   return None;
 }
 
-
 /// Return true if-and-only-if the given SDValue can be lowered as either a
 /// constant argument or a stack reference.  The key point is that the value
 /// doesn't need to be spilled or tracked as a vreg use.
 static bool willLowerDirectly(SDValue Incoming) {
   // We are making an unchecked assumption that the frame size <= 2^16 as that
   // is the largest offset which can be encoded in the stackmap format.
   if (isa<FrameIndexSDNode>(Incoming))
     return true;
 
   // The largest constant describeable in the StackMap format is 64 bits.
   // Potential Optimization:  Constants values are sign extended by consumer,
   // and thus there are many constants of static type > 64 bits whose value
   // happens to be sext(Con64) and could thus be lowered directly.
   if (Incoming.getValueType().getSizeInBits() > 64)
     return false;
 
   return (isa<ConstantSDNode>(Incoming) || isa<ConstantFPSDNode>(Incoming) ||
           Incoming.isUndef());
 }
 
-
 /// Try to find existing copies of the incoming values in stack slots used for
 /// statepoint spilling.  If we can find a spill slot for the incoming value,
 /// mark that slot as allocated, and reuse the same slot for this safepoint.
 /// This helps to avoid series of loads and stores that only serve to reshuffle
 /// values on the stack between calls.
 static void reservePreviousStackSlotForValue(const Value *IncomingValue,
                                              SelectionDAGBuilder &Builder) {
   SDValue Incoming = Builder.getValue(IncomingValue);
if (!Loc.getNode()) {
     auto PtrInfo = MachinePointerInfo::getFixedStack(MF, Index);
     auto *StoreMMO = MF.getMachineMemOperand(
         PtrInfo, MachineMemOperand::MOStore, MFI.getObjectSize(Index),
         MFI.getObjectAlign(Index));
     Chain = Builder.DAG.getStore(Chain, Builder.getCurSDLoc(), Incoming, Loc,
                                  StoreMMO);
 
     MMO = getMachineMemOperand(MF, *cast<FrameIndexSDNode>(Loc));
-    
     Builder.StatepointLowering.setLocation(Incoming, Loc);
   }
 
   assert(Loc.getNode());
   return std::make_tuple(Loc, Chain, MMO);
 }
 
 /// Lower a single value incoming to a statepoint node.  This value can be

 /// lowering is described in lowerIncomingStatepointValue.  This function is
 /// responsible for lowering everything in the right position and playing some
 /// tricks to avoid redundant stack manipulation where possible.  On
 /// completion, 'Ops' will contain ready to use operands for machine code
 /// statepoint. The chain nodes will have already been created and the DAG root
 /// will be set to the last value spilled (if any were).
 static void
 lowerStatepointMetaArgs(SmallVectorImpl<SDValue> &Ops,
-                        SmallVectorImpl<MachineMemOperand*> &MemRefs,                                    SelectionDAGBuilder::StatepointLoweringInfo &SI,
+                        SmallVectorImpl<MachineMemOperand *> &MemRefs,
+                        DenseMap<SDValue, int> &LowerAsVReg,
+                        SelectionDAGBuilder::StatepointLoweringInfo &SI,
                         SelectionDAGBuilder &Builder) {
   // Lower the deopt and gc arguments for this statepoint.  Layout will be:
   // deopt argument length, deopt arguments.., gc arguments...
 #ifndef NDEBUG
   if (auto *GFI = Builder.GFI) {
     // Check that each of the gc pointer and bases we've gotten out of the
     // safepoint is something the strategy thinks might be a pointer (or vector
     // of pointers) into the GC heap.  This is basically just here to help catch
#endif
   // particular, if the callee spills callee preserved registers we may not
   // be able to find a value placed in that register during the call.  This is
   // fine for live-out, but not for live-through.  If we were willing to make
   // assumptions about the code generator producing the callee, we could
   // potentially allow live-through values in callee saved registers.
   const bool LiveInDeopt =
     SI.StatepointFlags & (uint64_t)StatepointFlags::DeoptLiveIn;
 
+  // Decide which deriver pointers will go on VRegs
+  const unsigned MaxTiedRegs = 15; // Max  number of tied regs MI can have.
+  unsigned MaxVRegPtrs =
+      std::min(MaxTiedRegs, MaxRegistersForGCPointers.getValue());
+  // Use old spill scheme for cross-block relocates.
+  if (SI.StatepointInstr) {
+    const BasicBlock *BB = SI.StatepointInstr->getParent();
+    bool NonLocalReloc =
+        llvm::any_of(SI.GCRelocates, [BB](const GCRelocateInst *R) {
+          return R->getParent() != BB;
+        });
+    if (NonLocalReloc)
+      MaxVRegPtrs = 0;
+  }
+
+  LLVM_DEBUG(dbgs() << "Desiding how to lower GC Pointers:\n");
+  unsigned CurNumVRegs = 0;
+  for (const Value *P : SI.Ptrs) {
+    if (LowerAsVReg.size() == MaxVRegPtrs)
+      break;
+    SDValue PtrSD = Builder.getValue(P);
+    if (willLowerDirectly(PtrSD) || P->getType()->isVectorTy()) {
+      LLVM_DEBUG(dbgs() << "direct/spill "; PtrSD.dump(&Builder.DAG));
+      continue;
+    }
+    LLVM_DEBUG(dbgs() << "vreg "; PtrSD.dump(&Builder.DAG));
+    LowerAsVReg[PtrSD] = CurNumVRegs++;
+  }
+  LLVM_DEBUG(dbgs() << LowerAsVReg.size()
+                    << " derived pointers will go in vregs\n");
+
   auto isGCValue = [&](const Value *V) {
     auto *Ty = V->getType();
     if (!Ty->isPtrOrPtrVectorTy())
       return false;
     if (auto *GFI = Builder.GFI)
       if (auto IsManaged = GFI->getStrategy().isGCManagedPointer(Ty))
         return *IsManaged;
     return true; // conservative
   };
 
   auto requireSpillSlot = [&](const Value *V) {
-    return !(LiveInDeopt || UseRegistersForDeoptValues) || isGCValue(V);
+    if (isGCValue(V))
+      return !LowerAsVReg.count(Builder.getValue(V));
+    return !(LiveInDeopt || UseRegistersForDeoptValues);
   };
 
   // Before we actually start lowering (and allocating spill slots for values),
   // reserve any stack slots which we judge to be profitable to reuse for a
   // particular value.  This is purely an optimization over the code below and
   // doesn't change semantics at all.  It is important for performance that we
   // reserve slots for both deopt and gc values before lowering either.
   for (const Value *V : SI.DeoptState) {
     if (requireSpillSlot(V))
       reservePreviousStackSlotForValue(V, Builder);
   }
+
   for (unsigned i = 0; i < SI.Bases.size(); ++i) {
+    SDValue SDV = Builder.getValue(SI.Bases[i]);
+    if (!LowerAsVReg.count(SDV))
       reservePreviousStackSlotForValue(SI.Bases[i], Builder);
+    SDV = Builder.getValue(SI.Ptrs[i]);
+    if (!LowerAsVReg.count(SDV))
       reservePreviousStackSlotForValue(SI.Ptrs[i], Builder);
   }
 
   // First, prefix the list with the number of unique values to be
   // lowered.  Note that this is the number of *Values* not the
   // number of SDValues required to lower them.
   const int NumVMSArgs = SI.DeoptState.size();
   pushStackMapConstant(Ops, Builder, NumVMSArgs);
 
   // The vm state arguments are lowered in an opaque manner.  We do not know
   // what type of values are contained within.
+  LLVM_DEBUG(dbgs() << "Lowering deopt state\n");
   for (const Value *V : SI.DeoptState) {
     SDValue Incoming;
     // If this is a function argument at a static frame index, generate it as
     // the frame index.
     if (const Argument *Arg = dyn_cast<Argument>(V)) {
       int FI = Builder.FuncInfo.getArgumentFrameIndex(Arg);
       if (FI != INT_MAX)
         Incoming = Builder.DAG.getFrameIndex(FI, Builder.getFrameIndexTy());
     }
     if (!Incoming.getNode())
       Incoming = Builder.getValue(V);
+    LLVM_DEBUG(dbgs() << "Value " << *V
+                      << " requireSpillSlot = " << requireSpillSlot(V) << "\n");
     lowerIncomingStatepointValue(Incoming, requireSpillSlot(V), Ops, MemRefs,
                                  Builder);
   }
 
   // Finally, go ahead and lower all the gc arguments.  There's no prefixed
   // length for this one.  After lowering, we'll have the base and pointer
   // arrays interwoven with each (lowered) base pointer immediately followed by
   // it's (lowered) derived pointer.  i.e
   // (base[0], ptr[0], base[1], ptr[1], ...)
   for (unsigned i = 0; i < SI.Bases.size(); ++i) {
-    const Value *Base = SI.Bases[i];
-    lowerIncomingStatepointValue(Builder.getValue(Base),
-                                 /*RequireSpillSlot*/ true, Ops, MemRefs,
+    bool RequireSpillSlot;
+    SDValue Base = Builder.getValue(SI.Bases[i]);
+    RequireSpillSlot = !LowerAsVReg.count(Base);
+    lowerIncomingStatepointValue(Base, RequireSpillSlot, Ops, MemRefs,
                                  Builder);
 
-    const Value *Ptr = SI.Ptrs[i];
-    lowerIncomingStatepointValue(Builder.getValue(Ptr),
-                                 /*RequireSpillSlot*/ true, Ops, MemRefs,
+    SDValue Derived = Builder.getValue(SI.Ptrs[i]);
+    RequireSpillSlot = !LowerAsVReg.count(Derived);
+    lowerIncomingStatepointValue(Derived, RequireSpillSlot, Ops, MemRefs,
                                  Builder);
   }
 
   // If there are any explicit spill slots passed to the statepoint, record
   // them, but otherwise do not do anything special.  These are user provided
   // allocas and give control over placement to the consumer.  In this case,
   // it is the contents of the slot which may get updated, not the pointer to
   // the alloca
#endif
   const Instruction *StatepointInstr = SI.StatepointInstr;
   auto &SpillMap = Builder.FuncInfo.StatepointSpillMaps[StatepointInstr];
 
   for (const GCRelocateInst *Relocate : SI.GCRelocates) {
     const Value *V = Relocate->getDerivedPtr();
     SDValue SDV = Builder.getValue(V);
     SDValue Loc = Builder.StatepointLowering.getLocation(SDV);
 
-    if (Loc.getNode()) {
+    if (LowerAsVReg.count(SDV)) {
+      SpillMap[V] = None;
+    } else if (Loc.getNode()) {
       SpillMap[V] = cast<FrameIndexSDNode>(Loc)->getIndex();
     } else {
       // Record value as visited, but not spilled. This is case for allocas
       // and constants. For this values we can avoid emitting spill load while
       // visiting corresponding gc_relocate.
       // Actually we do not need to record them in this map at all.
       // We do this only to check that we are not relocating any unvisited
       // value.
SDValue SelectionDAGBuilder::LowerAsSTATEPOINT(
   // lower it, then reverse engineer the calling sequence.
 
   NumOfStatepoints++;
   // Clear state
   StatepointLowering.startNewStatepoint(*this);
   assert(SI.Bases.size() == SI.Ptrs.size() &&
          SI.Ptrs.size() <= SI.GCRelocates.size());
 
+  LLVM_DEBUG(dbgs() << "Lowering statepoint " << *SI.StatepointInstr << "\n");
 #ifndef NDEBUG
   for (auto *Reloc : SI.GCRelocates)
     if (Reloc->getParent() == SI.StatepointInstr->getParent())
       StatepointLowering.scheduleRelocCall(*Reloc);
 #endif
 
   // Lower statepoint vmstate and gcstate arguments
   SmallVector<SDValue, 10> LoweredMetaArgs;
   SmallVector<MachineMemOperand*, 16> MemRefs;
-  lowerStatepointMetaArgs(LoweredMetaArgs, MemRefs, SI, *this);
+  // Maps derived pointer SDValue to statepoint result of relocated pointer.
+  DenseMap<SDValue, int> LowerAsVReg;
+  lowerStatepointMetaArgs(LoweredMetaArgs, MemRefs, LowerAsVReg, SI, *this);
 
   // Now that we've emitted the spills, we need to update the root so that the
   // call sequence is ordered correctly.
   SI.CLI.setChain(getRoot());
 
   // Get call node, we will replace it later with statepoint
   SDValue ReturnVal;
   SDNode *CallNode;
   std::tie(ReturnVal, CallNode) =
#endif
   Ops.push_back(Chain);
 
   // Same for the glue, but we add it only if original call had it
   if (Glue.getNode())
     Ops.push_back(Glue);
 
   // Compute return values.  Provide a glue output since we consume one as
   // input.  This allows someone else to chain off us as needed.
-  SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
+  SmallVector<EVT, 8> NodeTys;
+  for (auto &Ptr : SI.Ptrs) {
+    SDValue SD = getValue(Ptr);
+    if (LowerAsVReg.count(SD)) {
+      NodeTys.push_back(SD.getValueType());
+    }
+  }
+  LLVM_DEBUG(dbgs() << "Statepoint has " << NodeTys.size() << " results\n");
+  assert(NodeTys.size() == LowerAsVReg.size() && "Inconsistent GC Ptr lowering");
+  NodeTys.push_back(MVT::Other);
+  NodeTys.push_back(MVT::Glue);
 
+  unsigned NumResults = NodeTys.size();
   MachineSDNode *StatepointMCNode =
     DAG.getMachineNode(TargetOpcode::STATEPOINT, getCurSDLoc(), NodeTys, Ops);
   DAG.setNodeMemRefs(StatepointMCNode, MemRefs);
 
   SDNode *SinkNode = StatepointMCNode;
 
+  // Fill mapping from derived pointer to statepoint result denoting its
+  // relocated value.
+  auto &DPtrMap = StatepointLowering.DerivedPtrMap;
+  for (const auto *Relocate : SI.GCRelocates) {
+    Value *Derived = Relocate->getDerivedPtr();
+    SDValue SD = getValue(Derived);
+    if (LowerAsVReg.count(SD))
+      DPtrMap[Derived] = SDValue(StatepointMCNode, LowerAsVReg[SD]);
+  }
+
   // Build the GC_TRANSITION_END node if necessary.
   //
   // See the comment above regarding GC_TRANSITION_START for the layout of
   // the operands to the GC_TRANSITION_END node.
   if (IsGCTransition) {
     SmallVector<SDValue, 8> TEOps;
 
     // Add chain
-    TEOps.push_back(SDValue(StatepointMCNode, 0));
+    TEOps.push_back(SDValue(StatepointMCNode, NumResults - 2));
 
     // Add GC transition arguments
     for (const Value *V : SI.GCTransitionArgs) {
       TEOps.push_back(getValue(V));
       if (V->getType()->isPointerTy())
         TEOps.push_back(DAG.getSrcValue(V));
     }
 
     // Add glue
-    TEOps.push_back(SDValue(StatepointMCNode, 1));
+    TEOps.push_back(SDValue(StatepointMCNode, NumResults - 1));
 
     SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
 
     SDValue GCTransitionStart =
         DAG.getNode(ISD::GC_TRANSITION_END, getCurSDLoc(), NodeTys, TEOps);
 
     SinkNode = GCTransitionStart.getNode();
   }
 
   // Replace original call
-  DAG.ReplaceAllUsesWith(CallNode, SinkNode); // This may update Root
+  // Call: ch,glue = CALL ...
+  // Statepoint: [gc relocates],ch,glue = STATEPOINT ...
+  unsigned NumSinkValues = SinkNode->getNumValues();
+  SDValue StatepointValues[2] = {SDValue(SinkNode, NumSinkValues - 2),
+                                 SDValue(SinkNode, NumSinkValues - 1)};
+  DAG.ReplaceAllUsesWith(CallNode, StatepointValues);
   // Remove original call node
   DAG.DeleteNode(CallNode);
 
   // DON'T set the root - under the assumption that it's already set past the
   // inserted node we created.
 
   // TODO: A better future implementation would be to emit a single variable
   // argument, variable return value STATEPOINT node here and then hookup the
#endif
 
   if (GCResult->getParent() == I.getParent()) {
     // Result value will be used in a same basic block. Don't export it or
     // perform any explicit register copies. The gc_result will simply grab
     // this value. 
     setValue(&I, ReturnValue);
     return;
   }
-  
   // Result value will be used in a different basic block so we need to export
   // it now.  Default exporting mechanism will not work here because statepoint
   // call has a different type than the actual call. It means that by default
   // llvm will create export register of the wrong type (always i32 in our
   // case). So instead we need to create export register with correct type
   // manually.
   // TODO: To eliminate this problem we can remove gc.result intrinsics
   //       completely and make statepoint call to return a tuple.
void SelectionDAGBuilder::visitGCResult(const GCResultInst &CI) {
   Type *RetTy = SI->getActualReturnType();
   SDValue CopyFromReg = getCopyFromRegs(SI, RetTy);
   
   assert(CopyFromReg.getNode());
   setValue(&CI, CopyFromReg);
 }
 
 void SelectionDAGBuilder::visitGCRelocate(const GCRelocateInst &Relocate) {
+  const BasicBlock *StatepointBB = Relocate.getStatepoint()->getParent();
 #ifndef NDEBUG
   // Consistency check
   // We skip this check for relocates not in the same basic block as their
   // statepoint. It would be too expensive to preserve validation info through
   // different basic blocks.
-  if (Relocate.getStatepoint()->getParent() == Relocate.getParent())
+  if (StatepointBB == Relocate.getParent())
     StatepointLowering.relocCallVisited(Relocate);
 
   auto *Ty = Relocate.getType()->getScalarType();
   if (auto IsManaged = GFI->getStrategy().isGCManagedPointer(Ty))
     assert(*IsManaged && "Non gc managed pointer relocated!");
 #endif
 
   const Value *DerivedPtr = Relocate.getDerivedPtr();
   SDValue SD = getValue(DerivedPtr);
 
   if (SD.isUndef() && SD.getValueType().getSizeInBits() <= 64) {
     // Lowering relocate(undef) as arbitrary constant. Current constant value
     // is chosen such that it's unlikely to be a valid pointer.
     setValue(&Relocate, DAG.getTargetConstant(0xFEFEFEFE, SDLoc(SD), MVT::i64));
     return;
   }
 
+  // Relocate is local to statepoint block and its pointer was assigned
+  // to VReg. Use corresponding statepoint result.
+  auto &DPtrMap = StatepointLowering.DerivedPtrMap;
+  auto It = DPtrMap.find(DerivedPtr);
+  if (It != DPtrMap.end()) {
+    setValue(&Relocate, It->second);
+    assert(Relocate.getParent() == StatepointBB && "unexpected DPtrMap entry");
+    return;
+  }
+
   auto &SpillMap = FuncInfo.StatepointSpillMaps[Relocate.getStatepoint()];
   auto SlotIt = SpillMap.find(DerivedPtr);
   assert(SlotIt != SpillMap.end() && "Relocating not lowered gc value");
   Optional<int> DerivedPtrLocation = SlotIt->second;
 
   // We didn't need to spill these special cases (constants and allocas).
   // See the handling in spillIncomingValueForStatepoint for detail.
   if (!DerivedPtrLocation) {

llvm/lib/CodeGen/TargetLoweringBase.cpp

if (!llvm::any_of(MI->operands(),
                     [](MachineOperand &Operand) { return Operand.isFI(); }))
     return MBB;
 
   MachineInstrBuilder MIB = BuildMI(MF, MI->getDebugLoc(), MI->getDesc());
 
   // Inherit previous memory operands.
   MIB.cloneMemRefs(*MI);
 
-  for (auto &MO : MI->operands()) {
+  for (unsigned i = 0; i < MI->getNumOperands(); ++i) {
+    MachineOperand &MO = MI->getOperand(i);
     if (!MO.isFI()) {
+      // Index of Def operand this Use it tied to.
+      // Since Defs are coming before Uses, if Use is tied, then
+      // index of Def must be smaller that index of that Use.
+      // Also, Defs preserve their position in new MI.
+      unsigned TiedTo = i;
+      if (MO.isReg() && MO.isTied())
+        TiedTo = MI->findTiedOperandIdx(i);
       MIB.add(MO);
+      if (TiedTo < i)
+        MIB->tieOperands(TiedTo, MIB->getNumOperands() - 1);
       continue;
     }
 
     // foldMemoryOperand builds a new MI after replacing a single FI operand
     // with the canonical set of five x86 addressing-mode operands.
     int FI = MO.getIndex();
 
     // Add frame index operands recognized by stackmaps.cpp

llvm/test/CodeGen/X86/statepoint-vreg.ll

+; This run is to demonstrate what MIR SSA looks like.
+; RUN: llc -max-registers-for-gc-values=4 -stop-after finalize-isel < %s | FileCheck --check-prefix=CHECK-VREG %s
+; This run is to demonstrate register allocator work.
+; RUN: llc -max-registers-for-gc-values=4 -stop-after virtregrewriter < %s | FileCheck --check-prefix=CHECK-PREG %s
+; This run is to demonstrate resulting assembly/stackmaps.
+; NOTE: When D81647 is landed this run line will need to be adjusted!
+; RUN: llc -max-registers-for-gc-values=4 < %s | FileCheck --check-prefix=CHECK-ASM %s
+
+target datalayout = "e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-pc-linux-gnu"
+
+declare i1 @return_i1()
+declare void @func()
+declare void @consume(i32 addrspace(1)*)
+declare void @consume2(i32 addrspace(1)*, i32 addrspace(1)*)
+declare void @consume5(i32 addrspace(1)*, i32 addrspace(1)*, i32 addrspace(1)*, i32 addrspace(1)*, i32 addrspace(1)*)
+declare void @use1(i32 addrspace(1)*, i8 addrspace(1)*)
+
+; test most simple relocate
+define i1 @test_relocate(i32 addrspace(1)* %a) gc "statepoint-example" {
+; CHECK-VREG-LABEL: name:            test_relocate
+; CHECK-VREG:    %0:gr64 = COPY $rdi
+; CHECK-VREG:    %1:gr64 = STATEPOINT 0, 0, 0, @return_i1, 2, 0, 2, 0, 2, 0, %0, %0(tied-def 0), csr_64, implicit-def $rsp, implicit-def $ssp, implicit-def $al
+; CHECK-VREG:    %2:gr8 = COPY $al
+; CHECK-VREG:    $rdi = COPY %1
+; CHECK-VREG:    CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-PREG-LABEL: name:            test_relocate
+; CHECK-PREG:    renamable $rbx = COPY $rdi
+; CHECK-PREG:    renamable $rbx = STATEPOINT 0, 0, 0, @return_i1, 2, 0, 2, 0, 2, 0, killed renamable $rbx, renamable $rbx(tied-def 0), csr_64, implicit-def $rsp, implicit-def $ssp, implicit-def $al
+; CHECK-PREG:    renamable $bpl = COPY killed $al
+; CHECK-PREG:    $rdi = COPY killed renamable $rbx
+; CHECK-PREG:    CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-ASM-LABEL: test_relocate:
+; CHECK-ASM:       # %bb.0:
+; CHECK-ASM-NEXT: pushq %rbp
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: pushq %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 24
+; CHECK-ASM-NEXT: pushq %rax
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 32
+; CHECK-ASM-NEXT: .cfi_offset %rbx, -24
+; CHECK-ASM-NEXT: .cfi_offset %rbp, -16
+; CHECK-ASM-NEXT: movq  %rdi, %rbx
+; CHECK-ASM-NEXT: callq return_i1
+; CHECK-ASM-NEXT:  .Ltmp0:
+; CHECK-ASM-NEXT: movl  %eax, %ebp
+; CHECK-ASM-NEXT: movq  %rbx, %rdi
+; CHECK-ASM-NEXT: callq consume
+; CHECK-ASM-NEXT: movl  %ebp, %eax
+; CHECK-ASM-NEXT: addq  $8, %rsp
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 24
+; CHECK-ASM-NEXT: popq  %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: popq  %rbp
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 8
+; CHECK-ASM-NEXT: retq
+entry:
+  %safepoint_token = tail call token (i64, i32, i1 ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_i1f(i64 0, i32 0, i1 ()* @return_i1, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %a)]
+  %rel1 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 0, i32 0)
+  %res1 = call zeroext i1 @llvm.experimental.gc.result.i1(token %safepoint_token)
+  call void @consume(i32 addrspace(1)* %rel1)
+  ret i1 %res1
+}
+; test pointer variables intermixed with pointer constants
+define void @test_mixed(i32 addrspace(1)* %a, i32 addrspace(1)* %b, i32 addrspace(1)* %c) gc "statepoint-example" {
+; CHECK-VREG-LABEL: name:            test_mixed
+; CHECK-VREG:    %2:gr64 = COPY $rdx
+; CHECK-VREG:    %1:gr64 = COPY $rsi
+; CHECK-VREG:    %0:gr64 = COPY $rdi
+; CHECK-VREG:    %3:gr64, %4:gr64, %5:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, %2, %2(tied-def 0), 2, 0, 2, 0, %1, %1(tied-def 1), %0, %0(tied-def 2), csr_64
+; CHECK-VREG:    %6:gr32 = MOV32r0 implicit-def dead $eflags
+; CHECK-VREG:    %7:gr64 = SUBREG_TO_REG 0, killed %6, %subreg.sub_32bit
+; CHECK-VREG:    $rdi = COPY %5
+; CHECK-VREG:    $rsi = COPY %7
+; CHECK-VREG:    $rdx = COPY %4
+; CHECK-VREG:    $rcx = COPY %7
+; CHECK-VREG:    $r8 = COPY %3
+; CHECK-VREG:    CALL64pcrel32 @consume5, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit $rsi, implicit $rdx, implicit $rcx, implicit $r8, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-PREG-LABEL: name:            test_mixed
+; CHECK-PREG:    renamable $r14 = COPY $rdx
+; CHECK-PREG:    renamable $r15 = COPY $rsi
+; CHECK-PREG:    renamable $rbx = COPY $rdi
+; CHECK-PREG:    renamable $r14, renamable $r15, renamable $rbx = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, killed renamable $r14, renamable $r14(tied-def 0), 2, 0, 2, 0, killed renamable $r15, renamable $r15(tied-def 1), killed renamable $rbx, renamable $rbx(tied-def 2), csr_64, implicit-def $rsp, implicit-def $ssp
+; CHECK-PREG:    $rdi = COPY killed renamable $rbx
+; CHECK-PREG:    dead $esi = MOV32r0 implicit-def dead $eflags, implicit-def $rsi
+; CHECK-PREG:    $rdx = COPY killed renamable $r15
+; CHECK-PREG:    dead $ecx = MOV32r0 implicit-def dead $eflags, implicit-def $rcx
+; CHECK-PREG:    $r8 = COPY killed renamable $r14
+; CHECK-PREG:    CALL64pcrel32 @consume5, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit $rsi, implicit $rdx, implicit killed $rcx, implicit killed $r8, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-ASM-LABEL: test_mixed:
+; CHECK-ASM:        # %bb.0:                                # %entry
+; CHECK-ASM-NEXT: pushq %r15
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: pushq %r14
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 24
+; CHECK-ASM-NEXT: pushq %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 32
+; CHECK-ASM-NEXT: .cfi_offset %rbx, -32
+; CHECK-ASM-NEXT: .cfi_offset %r14, -24
+; CHECK-ASM-NEXT: .cfi_offset %r15, -16
+; CHECK-ASM-NEXT: movq  %rdx, %r14
+; CHECK-ASM-NEXT: movq  %rsi, %r15
+; CHECK-ASM-NEXT: movq  %rdi, %rbx
+; CHECK-ASM-NEXT: callq func
+; CHECK-ASM-NEXT:.Ltmp1:
+; CHECK-ASM-NEXT: movq  %rbx, %rdi
+; CHECK-ASM-NEXT: xorl  %esi, %esi
+; CHECK-ASM-NEXT: movq  %r15, %rdx
+; CHECK-ASM-NEXT: xorl  %ecx, %ecx
+; CHECK-ASM-NEXT: movq  %r14, %r8
+; CHECK-ASM-NEXT: callq consume5
+; CHECK-ASM-NEXT: popq  %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 24
+; CHECK-ASM-NEXT: popq  %r14
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: popq  %r15
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 8
+; CHECK-ASM-NEXT: retq
+entry:
+  %safepoint_token = tail call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %a, i32 addrspace(1)* null, i32 addrspace(1)* %b, i32 addrspace(1)* null, i32 addrspace(1)* %c)]
+  %rel1 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 0, i32 0)
+  %rel2 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 1, i32 1)
+  %rel3 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 2, i32 2)
+  %rel4 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 3, i32 3)
+  %rel5 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 4, i32 4)
+  call void @consume5(i32 addrspace(1)* %rel1, i32 addrspace(1)* %rel2, i32 addrspace(1)* %rel3, i32 addrspace(1)* %rel4, i32 addrspace(1)* %rel5)
+  ret void
+}
+
+; same as above, but for alloca
+define i32 addrspace(1)* @test_alloca(i32 addrspace(1)* %ptr) gc "statepoint-example" {
+; CHECK-VREG-LABEL: name:            test_alloca
+; CHECK-VREG:    %0:gr64 = COPY $rdi
+; CHECK-VREG:    MOV64mr %stack.0.alloca, 1, $noreg, 0, $noreg, %0 :: (store 8 into %ir.alloca)
+; CHECK-VREG:    %1:gr64 = STATEPOINT 0, 0, 0, @return_i1, 2, 0, 2, 0, 2, 0, %0, %0(tied-def 0), 0, %stack.0.alloca, 0, csr_64, implicit-def $rsp, implicit-def $ssp, implicit-def $al :: (volatile load store 8 on %stack.0.alloca)
+; CHECK-VREG:    %2:gr8 = COPY $al
+; CHECK-VREG:    %3:gr64 = MOV64rm %stack.0.alloca, 1, $noreg, 0, $noreg :: (dereferenceable load 8 from %ir.alloca)
+; CHECK-VREG:    $rdi = COPY %1
+; CHECK-VREG:    CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-PREG-LABEL: name:            test_alloca
+; CHECK-PREG:    renamable $rbx = COPY $rdi
+; CHECK-PREG:    MOV64mr %stack.0.alloca, 1, $noreg, 0, $noreg, renamable $rbx :: (store 8 into %ir.alloca)
+; CHECK-PREG:    renamable $rbx = STATEPOINT 0, 0, 0, @return_i1, 2, 0, 2, 0, 2, 0, killed renamable $rbx, renamable $rbx(tied-def 0), 0, %stack.0.alloca, 0, csr_64, implicit-def $rsp, implicit-def $ssp, implicit-def dead $al :: (volatile load store 8 on %stack.0.alloca)
+; CHECK-PREG:    renamable $r14 = MOV64rm %stack.0.alloca, 1, $noreg, 0, $noreg :: (dereferenceable load 8 from %ir.alloca)
+; CHECK-PREG:    $rdi = COPY killed renamable $rbx
+; CHECK-PREG:    CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-ASM-LABEL: test_alloca:
+; CHECK-ASM:       # %bb.0: # %entry
+; CHECK-ASM-NEXT: pushq %r14
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: pushq %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 24
+; CHECK-ASM-NEXT: pushq %rax
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 32
+; CHECK-ASM-NEXT: .cfi_offset %rbx, -24
+; CHECK-ASM-NEXT: .cfi_offset %r14, -16
+; CHECK-ASM-NEXT: movq  %rdi, %rbx
+; CHECK-ASM-NEXT: movq  %rdi, (%rsp)
+; CHECK-ASM-NEXT: callq return_i1
+; CHECK-ASM-NEXT:  .Ltmp2:
+; CHECK-ASM-NEXT: movq  (%rsp), %r14
+; CHECK-ASM-NEXT: movq  %rbx, %rdi
+; CHECK-ASM-NEXT: callq consume
+; CHECK-ASM-NEXT: movq  %r14, %rax
+; CHECK-ASM-NEXT: addq  $8, %rsp
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 24
+; CHECK-ASM-NEXT: popq  %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: popq  %r14
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 8
+; CHECK-ASM-NEXT: retq
+entry:
+  %alloca = alloca i32 addrspace(1)*, align 8
+  store i32 addrspace(1)* %ptr, i32 addrspace(1)** %alloca
+  %safepoint_token = call token (i64, i32, i1 ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_i1f(i64 0, i32 0, i1 ()* @return_i1, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)** %alloca, i32 addrspace(1)* %ptr)]
+  %rel1 = load i32 addrspace(1)*, i32 addrspace(1)** %alloca
+  %rel2 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 1, i32 1)
+  call void @consume(i32 addrspace(1)* %rel2)
+  ret i32 addrspace(1)* %rel1
+}
+
+; test base != derived
+define void @test_base_derived(i32 addrspace(1)* %base, i32 addrspace(1)* %derived) gc "statepoint-example" {
+; CHECK-VREG-LABEL: name:            test_base_derived
+; CHECK-VREG:    %1:gr64 = COPY $rsi
+; CHECK-VREG:    %0:gr64 = COPY $rdi
+; CHECK-VREG:    MOV64mr %stack.0, 1, $noreg, 0, $noreg, %0 :: (store 8 into %stack.0)
+; CHECK-VREG:    %2:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 1, 8, %stack.0, 0, %1(tied-def 0), csr_64, implicit-def $rsp, implicit-def $ssp :: (volatile load store 8 on %stack.0)
+; CHECK-VREG:    $rdi = COPY %2
+; CHECK-VREG:    CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-PREG-LABEL: name:            test_base_derived
+; CHECK-PREG:    renamable $rbx = COPY $rsi
+; CHECK-PREG:    MOV64mr %stack.0, 1, $noreg, 0, $noreg, killed renamable $rdi :: (store 8 into %stack.0)
+; CHECK-PREG:    renamable $rbx = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 1, 8, %stack.0, 0, killed renamable $rbx(tied-def 0), csr_64, implicit-def $rsp, implicit-def $ssp :: (volatile load store 8 on %stack.0)
+; CHECK-PREG:    $rdi = COPY killed renamable $rbx
+; CHECK-PREG:    CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-ASM-LABEL: test_base_derived:
+; CHECK-ASM:       # %bb.0:
+; CHECK-ASM-NEXT: pushq %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: subq  $16, %rsp
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 32
+; CHECK-ASM-NEXT: .cfi_offset %rbx, -16
+; CHECK-ASM-NEXT: movq  %rsi, %rbx
+; CHECK-ASM-NEXT: movq  %rdi, 8(%rsp)
+; CHECK-ASM-NEXT: callq func
+; CHECK-ASM-NEXT:  .Ltmp3:
+; CHECK-ASM-NEXT: movq  %rbx, %rdi
+; CHECK-ASM-NEXT: callq consume
+; CHECK-ASM-NEXT: addq  $16, %rsp
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: popq  %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 8
+; CHECK-ASM-NEXT: retq
+  %safepoint_token = tail call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %base, i32 addrspace(1)* %derived)]
+  %reloc = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 0, i32 1)
+  call void @consume(i32 addrspace(1)* %reloc)
+  ret void
+}
+
+; deopt GC pointer not present in GC args must be spilled
+define void @test_deopt_gcpointer(i32 addrspace(1)* %a, i32 addrspace(1)* %b) gc "statepoint-example" {
+; CHECK-VREG-LABEL: name:            test_deopt_gcpointer
+; CHECK-VREG:    %1:gr64 = COPY $rsi
+; CHECK-VREG:    %0:gr64 = COPY $rdi
+; CHECK-VREG:    MOV64mr %stack.0, 1, $noreg, 0, $noreg, %0 :: (store 8 into %stack.0)
+; CHECK-VREG:    %2:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 1, 1, 8, %stack.0, 0, %1, %1(tied-def 0), csr_64, implicit-def $rsp, implicit-def $ssp :: (volatile load store 8 on %stack.0)
+; CHECK-VREG:    $rdi = COPY %2
+; CHECK-VREG:    CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+; CHECK-VREG:    RET 0
+
+; CHECK-PREG-LABEL: name:            test_deopt_gcpointer
+; CHECK-PREG:    renamable $rbx = COPY $rsi
+; CHECK-PREG:    MOV64mr %stack.0, 1, $noreg, 0, $noreg, killed renamable $rdi :: (store 8 into %stack.0)
+; CHECK-PREG:    renamable $rbx = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 1, 1, 8, %stack.0, 0, killed renamable $rbx, renamable $rbx(tied-def 0), csr_64, implicit-def $rsp, implicit-def $ssp :: (volatile load store 8 on %stack.0)
+; CHECK-PREG:    $rdi = COPY killed renamable $rbx
+; CHECK-PREG:    CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-ASM-LABEL: test_deopt_gcpointer:
+; CHECK-ASM:       # %bb.0:
+; CHECK-ASM-NEXT: pushq %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: subq  $16, %rsp
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 32
+; CHECK-ASM-NEXT: .cfi_offset %rbx, -16
+; CHECK-ASM-NEXT: movq  %rsi, %rbx
+; CHECK-ASM-NEXT: movq  %rdi, 8(%rsp)
+; CHECK-ASM-NEXT: callq func
+; CHECK-ASM-NEXT:  .Ltmp4:
+; CHECK-ASM-NEXT: movq  %rbx, %rdi
+; CHECK-ASM-NEXT: callq consume
+; CHECK-ASM-NEXT: addq  $16, %rsp
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: popq  %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 8
+; CHECK-ASM-NEXT: retq
+  %safepoint_token = tail call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["deopt" (i32 addrspace(1)* %a), "gc-live" (i32 addrspace(1)* %b)]
+  %rel = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 0, i32 0)
+  call void @consume(i32 addrspace(1)* %rel)
+  ret void
+}
+
+;; Two gc.relocates of the same input, should require only a single spill/fill
+define void @test_gcrelocate_uniqueing(i32 addrspace(1)* %ptr) gc "statepoint-example" {
+; CHECK-VREG-LABEL: name:            test_gcrelocate_uniqueing
+; CHECK-VREG:    %0:gr64 = COPY $rdi
+; CHECK-VREG:    %1:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 2, %0, 2, 4278124286, %0, %0(tied-def 0), csr_64, implicit-def $rsp, implicit-def $ssp
+; CHECK-VREG:    $rdi = COPY %1
+; CHECK-VREG:    $rsi = COPY %1
+; CHECK-VREG:    CALL64pcrel32 @consume2, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit $rsi, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-PREG-LABEL: name:            test_gcrelocate_uniqueing
+; CHECK-PREG:    renamable $rbx = COPY $rdi
+; CHECK-PREG:    renamable $rbx = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 2, killed renamable $rbx, 2, 4278124286, renamable $rbx, renamable $rbx(tied-def 0), csr_64, implicit-def $rsp, implicit-def $ssp
+; CHECK-PREG:    $rdi = COPY renamable $rbx
+; CHECK-PREG:    $rsi = COPY killed renamable $rbx
+; CHECK-PREG:    CALL64pcrel32 @consume2, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit killed $rsi, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-ASM-LABEL: test_gcrelocate_uniqueing:
+; CHECK-ASM:       # %bb.0:
+; CHECK-ASM-NEXT: pushq %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: .cfi_offset %rbx, -16
+; CHECK-ASM-NEXT: movq  %rdi, %rbx
+; CHECK-ASM-NEXT: callq func
+; CHECK-ASM-NEXT: .Ltmp5:
+; CHECK-ASM-NEXT: movq  %rbx, %rdi
+; CHECK-ASM-NEXT: movq  %rbx, %rsi
+; CHECK-ASM-NEXT: callq consume2
+; CHECK-ASM-NEXT: popq  %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 8
+; CHECK-ASM-NEXT: retq
+  %tok = tail call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["deopt" (i32 addrspace(1)* %ptr, i32 undef), "gc-live" (i32 addrspace(1)* %ptr, i32 addrspace(1)* %ptr)]
+  %a = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %tok, i32 0, i32 0)
+  %b = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %tok, i32 1, i32 1)
+  call void @consume2(i32 addrspace(1)* %a, i32 addrspace(1)* %b)
+  ret void
+}
+
+; Two gc.relocates of a bitcasted pointer should only require a single spill/fill
+define void @test_gcptr_uniqueing(i32 addrspace(1)* %ptr) gc "statepoint-example" {
+; CHECK-VREG-LABEL: name:            test_gcptr_uniqueing
+; CHECK-VREG:    %0:gr64 = COPY $rdi
+; CHECK-VREG:    ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+; CHECK-VREG:    %1:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 2, %0, 2, 4278124286, %0, %0(tied-def 0), csr_64, implicit-def $rsp, implicit-def $ssp
+; CHECK-VREG:    ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+; CHECK-VREG:    ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+; CHECK-VREG:    $rdi = COPY %1
+; CHECK-VREG:    $rsi = COPY %1
+; CHECK-VREG:    CALL64pcrel32 @use1, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit $rsi, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-PREG-LABEL: name:            test_gcptr_uniqueing
+; CHECK-PREG:    renamable $rbx = COPY $rdi
+; CHECK-PREG:    renamable $rbx = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 2, killed renamable $rbx, 2, 4278124286, renamable $rbx, renamable $rbx(tied-def 0), csr_64, implicit-def $rsp, implicit-def $ssp
+; CHECK-PREG:    $rdi = COPY renamable $rbx
+; CHECK-PREG:    $rsi = COPY killed renamable $rbx
+; CHECK-PREG:    CALL64pcrel32 @use1, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit killed $rsi, implicit-def $rsp, implicit-def $ssp
+
+; CHECK-ASM-LABEL: test_gcptr_uniqueing:
+; CHECK-ASM:       # %bb.0:
+; CHECK-ASM-NEXT: pushq %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT: .cfi_offset %rbx, -16
+; CHECK-ASM-NEXT: movq  %rdi, %rbx
+; CHECK-ASM-NEXT: callq func
+; CHECK-ASM-NEXT:  .Ltmp6:
+; CHECK-ASM-NEXT: movq  %rbx, %rdi
+; CHECK-ASM-NEXT: movq  %rbx, %rsi
+; CHECK-ASM-NEXT: callq use1
+; CHECK-ASM-NEXT: popq  %rbx
+; CHECK-ASM-NEXT: .cfi_def_cfa_offset 8
+; CHECK-ASM-NEXT: retq
+  %ptr2 = bitcast i32 addrspace(1)* %ptr to i8 addrspace(1)*
+  %tok = tail call token (i64, i32, void ()*, i32, i32, ...)
+      @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["deopt" (i32 addrspace(1)* %ptr, i32 undef), "gc-live" (i32 addrspace(1)* %ptr, i8 addrspace(1)* %ptr2)]
+  %a = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %tok, i32 0, i32 0)
+  %b = call i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token %tok, i32 1, i32 1)
+  call void @use1(i32 addrspace(1)* %a, i8 addrspace(1)* %b)
+  ret void
+}
+
+;
+; Cross-basicblock relocates are handled with spilling for now.
+; No need to check post-RA output
+define i1 @test_cross_bb(i32 addrspace(1)* %a, i1 %external_cond) gc "statepoint-example" {
+; CHECK-VREG-LABEL: name:            test_cross_bb
+; CHECK-VREG:  bb.0.entry:
+; CHECK-VREG:         %1:gr32 = COPY $esi
+; CHECK-VREG-NEXT:    %0:gr64 = COPY $rdi
+; CHECK-VREG-NEXT:    %3:gr8 = COPY %1.sub_8bit
+; CHECK-VREG-NEXT:    MOV64mr %stack.0, 1, $noreg, 0, $noreg, %0 :: (store 8 into %stack.0)
+; CHECK-VREG-NEXT:    ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+; CHECK-VREG-NEXT:    STATEPOINT 0, 0, 0, @return_i1, 2, 0, 2, 0, 2, 0, 1, 8, %stack.0, 0, 1, 8, %stack.0, 0, csr_64, implicit-def $rsp, implicit-def $ssp, implicit-def $al :: (volatile load store 8 on %stack.0)
+; CHECK-VREG-NEXT:    ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+; CHECK-VREG-NEXT:    %4:gr8 = COPY $al
+; CHECK-VREG-NEXT:    %2:gr8 = COPY %4
+; CHECK-VREG-NEXT:    TEST8ri killed %3, 1, implicit-def $eflags
+; CHECK-VREG-NEXT:    JCC_1 %bb.2, 4, implicit $eflags
+; CHECK-VREG-NEXT:    JMP_1 %bb.1
+; CHECK-VREG:       bb.1.left:
+; CHECK-VREG-NEXT:    %6:gr64 = MOV64rm %stack.0, 1, $noreg, 0, $noreg :: (load 8 from %stack.0)
+; CHECK-VREG-NEXT:    ADJCALLSTACKDOWN64 0, 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+; CHECK-VREG-NEXT:    $rdi = COPY %6
+; CHECK-VREG-NEXT:    CALL64pcrel32 @consume, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit-def $rsp, implicit-def $ssp
+; CHECK-VREG-NEXT:    ADJCALLSTACKUP64 0, 0, implicit-def dead $rsp, implicit-def dead $eflags, implicit-def dead $ssp, implicit $rsp, implicit $ssp
+; CHECK-VREG-NEXT:    $al = COPY %2
+; CHECK-VREG-NEXT:    RET 0, $al
+; CHECK-VREG:       bb.2.right:
+; CHECK-VREG-NEXT:    %5:gr8 = MOV8ri 1
+; CHECK-VREG-NEXT:    $al = COPY %5
+; CHECK-VREG-NEXT:    RET 0, $al
+
+entry:
+  %safepoint_token = tail call token (i64, i32, i1 ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_i1f(i64 0, i32 0, i1 ()* @return_i1, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %a)]
+  br i1 %external_cond, label %left, label %right
+
+left:
+  %call1 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 0, i32 0)
+  %call2 = call zeroext i1 @llvm.experimental.gc.result.i1(token %safepoint_token)
+  call void @consume(i32 addrspace(1)* %call1)
+  ret i1 %call2
+
+right:
+  ret i1 true
+}
+
+; No need to check post-regalloc output as it is the same
+define i1 @duplicate_reloc() gc "statepoint-example" {
+; CHECK-VREG-LABEL: name:            duplicate_reloc
+; CHECK-VREG:  bb.0.entry:
+; CHECK-VREG:    STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, csr_64, implicit-def $rsp, implicit-def $ssp
+; CHECK-VREG:    STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 2, 0, 2, 0, csr_64, implicit-def $rsp, implicit-def $ssp
+; CHECK-VREG:    %0:gr8 = MOV8ri 1
+; CHECK-VREG:    $al = COPY %0
+; CHECK-VREG:    RET 0, $al
+
+; CHECK-ASM-LABEL: duplicate_reloc:
+; CHECK-ASM:       # %bb.0: # %entry
+; CHECK-ASM-NEXT:          pushq  %rax
+; CHECK-ASM-NEXT:          .cfi_def_cfa_offset 16
+; CHECK-ASM-NEXT:          callq  func
+; CHECK-ASM-NEXT:  .Ltmp8:
+; CHECK-ASM-NEXT:          callq  func
+; CHECK-ASM-NEXT:  .Ltmp9:
+; CHECK-ASM-NEXT:          movb $1, %al
+; CHECK-ASM-NEXT:          popq %rcx
+; CHECK-ASM-NEXT:          .cfi_def_cfa_offset 8
+; CHECK-ASM-NEXT:          retq
+entry:
+  %safepoint_token = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* null, i32 addrspace(1)* null)]
+  %base = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 0, i32 0)
+  %derived = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 0, i32 1)
+  %safepoint_token2 = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %base, i32 addrspace(1)* %derived)]
+  %base_reloc = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token2,  i32 0, i32 0)
+  %derived_reloc = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token2,  i32 0, i32 1)
+  %cmp1 = icmp eq i32 addrspace(1)* %base_reloc, null
+  %cmp2 = icmp eq i32 addrspace(1)* %derived_reloc, null
+  %cmp = and i1 %cmp1, %cmp2
+  ret i1 %cmp
+}
+
+; Vectors cannot go in VRegs
+; No need to check post-regalloc output as it is lowered using old scheme
+define <2 x i8 addrspace(1)*> @test_vector(<2 x i8 addrspace(1)*> %obj) gc "statepoint-example" {
+; CHECK-VREG-LABEL: name:            test_vector
+; CHECK-VREG:    %0:vr128 = COPY $xmm0
+; CHECK-VREG:    MOVAPSmr %stack.0, 1, $noreg, 0, $noreg, %0 :: (store 16 into %stack.0)
+; CHECK-VREG:    STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, 1, 16, %stack.0, 0, 1, 16, %stack.0, 0, csr_64, implicit-def $rsp, implicit-def $ssp :: (volatile load store 16 on %stack.0)
+; CHECK-VREG:    %1:vr128 = MOVAPSrm %stack.0, 1, $noreg, 0, $noreg :: (load 16 from %stack.0)
+; CHECK-VREG:    $xmm0 = COPY %1
+; CHECK-VREG:    RET 0, $xmm0
+
+; CHECK-ASM-LABEL: test_vector:
+; CHECK-ASM:       # %bb.0: # %entry
+; CHECK-ASM-NEXT:          subq $24, %rsp
+; CHECK-ASM-NEXT:          .cfi_def_cfa_offset 32
+; CHECK-ASM-NEXT:          movaps %xmm0, (%rsp)
+; CHECK-ASM-NEXT:          callq  func
+; CHECK-ASM-NEXT:  .Ltmp10:
+; CHECK-ASM-NEXT:          movaps (%rsp), %xmm0
+; CHECK-ASM-NEXT:          addq $24, %rsp
+; CHECK-ASM-NEXT:          .cfi_def_cfa_offset 8
+; CHECK-ASM-NEXT:          retq
+entry:
+  %safepoint_token = call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (<2 x i8 addrspace(1)*> %obj)]
+  %obj.relocated = call coldcc <2 x i8 addrspace(1)*> @llvm.experimental.gc.relocate.v2p1i8(token %safepoint_token, i32 0, i32 0) ; (%obj, %obj)
+  ret <2 x i8 addrspace(1)*> %obj.relocated
+}
+
+
+; test limit on amount of vregs
+define void @test_limit(i32 addrspace(1)* %a, i32 addrspace(1)* %b, i32 addrspace(1)* %c, i32 addrspace(1)* %d, i32 addrspace(1)*  %e) gc "statepoint-example" {
+; CHECK-VREG-LABEL: name:            test_limit
+; CHECK-VREG:    %4:gr64 = COPY $r8
+; CHECK-VREG:    %3:gr64 = COPY $rcx
+; CHECK-VREG:    %2:gr64 = COPY $rdx
+; CHECK-VREG:    %1:gr64 = COPY $rsi
+; CHECK-VREG:    %0:gr64 = COPY $rdi
+; CHECK-VREG:    MOV64mr %stack.0, 1, $noreg, 0, $noreg, %0 :: (store 8 into %stack.0)
+; CHECK-VREG:    %5:gr64, %6:gr64, %7:gr64, %8:gr64 = STATEPOINT 0, 0, 0, @func, 2, 0, 2, 0, 2, 0, %4, %4(tied-def 0), %3, %3(tied-def 1), %2, %2(tied-def 2), %1, %1(tied-def 3), 1, 8, %stack.0, 0, 1, 8, %stack.0, 0, csr_64, implicit-def $rsp, implicit-def $ssp :: (volatile load store 8 on %stack.0)
+; CHECK-VREG:    %9:gr64 = MOV64rm %stack.0, 1, $noreg, 0, $noreg :: (load 8 from %stack.0)
+; CHECK-VREG:    $rdi = COPY %9
+; CHECK-VREG:    $rsi = COPY %8
+; CHECK-VREG:    $rdx = COPY %7
+; CHECK-VREG:    $rcx = COPY %6
+; CHECK-VREG:    $r8 = COPY %5
+; CHECK-VREG:    CALL64pcrel32 @consume5, csr_64, implicit $rsp, implicit $ssp, implicit $rdi, implicit $rsi, implicit $rdx, implicit $rcx, implicit $r8, implicit-def $rsp, implicit-def $ssp
+; CHECK-VREG:    RET 0
+entry:
+  %safepoint_token = tail call token (i64, i32, void ()*, i32, i32, ...) @llvm.experimental.gc.statepoint.p0f_isVoidf(i64 0, i32 0, void ()* @func, i32 0, i32 0, i32 0, i32 0) ["gc-live" (i32 addrspace(1)* %a, i32 addrspace(1)* %b, i32 addrspace(1)* %c, i32 addrspace(1)* %d, i32 addrspace(1)* %e)]
+  %rel1 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 0, i32 0)
+  %rel2 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 1, i32 1)
+  %rel3 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 2, i32 2)
+  %rel4 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 3, i32 3)
+  %rel5 = call i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token %safepoint_token,  i32 4, i32 4)
+  call void @consume5(i32 addrspace(1)* %rel1, i32 addrspace(1)* %rel2, i32 addrspace(1)* %rel3, i32 addrspace(1)* %rel4, i32 addrspace(1)* %rel5)
+  ret void
+}
+
+declare token @llvm.experimental.gc.statepoint.p0f_i1f(i64, i32, i1 ()*, i32, i32, ...)
+declare token @llvm.experimental.gc.statepoint.p0f_isVoidf(i64, i32, void ()*, i32, i32, ...)
+declare i32 addrspace(1)* @llvm.experimental.gc.relocate.p1i32(token, i32, i32)
+declare i8 addrspace(1)* @llvm.experimental.gc.relocate.p1i8(token, i32, i32)
+declare <2 x i8 addrspace(1)*> @llvm.experimental.gc.relocate.v2p1i8(token, i32, i32)
+declare i1 @llvm.experimental.gc.result.i1(token)
+
+; CHECK-ASM-LABEL: .section .llvm_stackmaps
+; CHECK-ASM-NEXT:  __LLVM_StackMaps:
+; Entry for test_relocate
+; CHECK-ASM:          .quad 0
+; CHECK-ASM-NEXT:       .long .Ltmp0-test_relocate
+; CHECK-ASM-NEXT: .short  0
+; Num locations
+; CHECK-ASM-NEXT: .short  5
+; Location 1 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 2 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 3 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 4 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 5 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+;  Entry for test_mixed
+; CHECK-ASM:      .quad 0
+; CHECK-ASM-NEXT: .long .Ltmp1-test_mixed
+; CHECK-ASM-NEXT: .short  0
+; Num locations
+; CHECK-ASM-NEXT: .short  11
+; Location 1 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 2 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 3 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 4 Register $r14
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  14
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 5 Register $r14
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  14
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 6 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 7 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 8 Register $r15
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  15
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 9 Register $r15
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  15
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 10 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 11 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Entry for test_alloca
+; CHECK-ASM:      .quad 0
+; CHECK-ASM-NEXT: .long .Ltmp2-test_alloca
+; CHECK-ASM-NEXT: .short  0
+; Num locations
+; CHECK-ASM-NEXT: .short  6
+; Location 1 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 2 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 3 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 4 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 5 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 6 Direct $rsp + 0
+; CHECK-ASM-NEXT: .byte 2
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  7
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Entry for test_base_derive
+; CHECK-ASM:      .quad 0
+; CHECK-ASM-NEXT: .long .Ltmp3-test_base_derived
+; CHECK-ASM-NEXT: .short  0
+; Num locations
+; CHECK-ASM-NEXT: .short  5
+; Location 1 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 2 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 3 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 4 Indirect $rsp + 8
+; CHECK-ASM-NEXT: .byte 3
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  7
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 8
+; Location 5 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Entry for test_deopt_gcpointer
+; CHECK-ASM:      .quad 0
+; CHECK-ASM-NEXT: .long .Ltmp4-test_deopt_gcpointer
+; CHECK-ASM-NEXT: .short  0
+; Num locations
+; CHECK-ASM-NEXT: .short  6
+; Location 1 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 2 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 3 Constant 1
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 1
+; Location 4Indirect $rsp + 8
+; CHECK-ASM-NEXT: .byte 3
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  7
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 8
+; Location 5 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 6
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Entry for test_gcrelocate_uniqueing
+; CHECK-ASM:      .quad 0
+; CHECK-ASM-NEXT: .long .Ltmp5-test_gcrelocate_uniqueing
+; CHECK-ASM-NEXT: .short  0
+; Num locations
+; CHECK-ASM-NEXT: .short  7
+; Location 1 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 2 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 3 Constant 2
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 2
+; Location 4 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 5 Constant Index 0
+; CHECK-ASM-NEXT: .byte 5
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 6 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 7 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Entry for test_gcptr_uniqueing
+; CHECK-ASM:      .long .Ltmp6-test_gcptr_uniqueing
+; CHECK-ASM-NEXT: .short  0
+; Num locations
+; CHECK-ASM-NEXT: .short  7
+; Location 1 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 2 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 3 Constant 2
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 2
+; Location 4 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 5 Constant Index 0
+; CHECK-ASM-NEXT: .byte 5
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 6 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 7 Register $rbx
+; CHECK-ASM-NEXT: .byte 1
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  3
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Entry for test_cross_bb
+; CHECK-ASM:      .quad 0
+; CHECK-ASM-NEXT: .long .Ltmp7-test_cross_bb
+; CHECK-ASM-NEXT: .short  0
+; Num locations
+; CHECK-ASM-NEXT: .short  5
+; Location 1 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 2 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 3 Constant 0
+; CHECK-ASM-NEXT: .byte 4
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 4 Indirect $rsp + 0
+; CHECK-ASM-NEXT: .byte 3
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  7
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0
+; Location 5 Indirect $rsp + 0
+; CHECK-ASM-NEXT: .byte 3
+; CHECK-ASM-NEXT: .byte 0
+; CHECK-ASM-NEXT: .short  8
+; CHECK-ASM-NEXT: .short  7
+; CHECK-ASM-NEXT: .short  0
+; CHECK-ASM-NEXT: .long 0