Generalize tail continues (#1298)

* Generalize tail continues

* Fix DecodeBasicBlock

`Next` and `Branch` would be null, which is not the state expected by
the branch instructions. They end up branching or falling into a block
which is never populated by the `Translator`. This causes an assert to
be fired when building the CFG.

* Clean up Decode overloads

* Do not synchronize when branching into exit block

If we're branching into an exit block, that exit block will tail
continue into another translation which already has a synchronization.

* Remove A32 predicate tail continue

If `block` is not an exit block then the `block.Next` must exist (as
per the last instruction of `block`).

* Throw if decoded 0 blocks

Address gdkchan's feedback

* Rebuild block list instead of setting to null

Address gdkchan's feedback

5 files changed, 55 insertions, 134 deletions

diff --git a/ARMeilleure/Instructions/InstEmitException.cs b/ARMeilleure/Instructions/InstEmitException.cs
index cdf6cf34..9cfd066f 100644
--- a/ARMeilleure/Instructions/InstEmitException.cs
+++ b/ARMeilleure/Instructions/InstEmitException.cs
@@ -27,11 +27,6 @@ namespace ARMeilleure.Instructions
             context.Call(typeof(NativeInterface).GetMethod(name), Const(op.Address), Const(op.Id));
 
             context.LoadFromContext();
-
-            if (context.CurrBlock.Next == null)
-            {
-                EmitTailContinue(context, Const(op.Address + 4));
-            }
         }
 
         public static void Und(ArmEmitterContext context)
@@ -45,11 +40,6 @@ namespace ARMeilleure.Instructions
             context.Call(typeof(NativeInterface).GetMethod(name), Const(op.Address), Const(op.RawOpCode));
 
             context.LoadFromContext();
-
-            if (context.CurrBlock.Next == null)
-            {
-                EmitTailContinue(context, Const(op.Address + 4));
-            }
         }
     }
 }
\ No newline at end of file
diff --git a/ARMeilleure/Instructions/InstEmitException32.cs b/ARMeilleure/Instructions/InstEmitException32.cs
index fd64e7bf..fd67ed52 100644
--- a/ARMeilleure/Instructions/InstEmitException32.cs
+++ b/ARMeilleure/Instructions/InstEmitException32.cs
@@ -27,11 +27,6 @@ namespace ARMeilleure.Instructions
             context.Call(typeof(NativeInterface).GetMethod(name), Const(op.Address), Const(op.Id));
 
             context.LoadFromContext();
-
-            if (context.CurrBlock.Next == null)
-            {
-                EmitTailContinue(context, Const(op.Address + 4));
-            }
         }
     }
 }
diff --git a/ARMeilleure/Instructions/InstEmitFlow.cs b/ARMeilleure/Instructions/InstEmitFlow.cs
index 6f2a346e..2fcf50db 100644
--- a/ARMeilleure/Instructions/InstEmitFlow.cs
+++ b/ARMeilleure/Instructions/InstEmitFlow.cs
@@ -15,14 +15,7 @@ namespace ARMeilleure.Instructions
         {
             OpCodeBImmAl op = (OpCodeBImmAl)context.CurrOp;
 
-            if (context.CurrBlock.Branch != null)
-            {
-                context.Branch(context.GetLabel((ulong)op.Immediate));
-            }
-            else
-            {
-                EmitTailContinue(context, Const(op.Immediate), context.CurrBlock.TailCall);
-            }
+            context.Branch(context.GetLabel((ulong)op.Immediate));
         }
 
         public static void B_Cond(ArmEmitterContext context)
@@ -92,69 +85,22 @@ namespace ARMeilleure.Instructions
         {
             OpCodeBImm op = (OpCodeBImm)context.CurrOp;
 
-            if (context.CurrBlock.Branch != null)
-            {
-                EmitCondBranch(context, context.GetLabel((ulong)op.Immediate), cond);
-
-                if (context.CurrBlock.Next == null)
-                {
-                    EmitTailContinue(context, Const(op.Address + 4));
-                }
-            }
-            else
-            {
-                Operand lblTaken = Label();
-
-                EmitCondBranch(context, lblTaken, cond);
-
-                EmitTailContinue(context, Const(op.Address + 4));
-
-                context.MarkLabel(lblTaken);
-
-                EmitTailContinue(context, Const(op.Immediate));
-            }
+            EmitCondBranch(context, context.GetLabel((ulong)op.Immediate), cond);
         }
 
         private static void EmitBranch(ArmEmitterContext context, Operand value, bool onNotZero)
         {
             OpCodeBImm op = (OpCodeBImm)context.CurrOp;
 
-            if (context.CurrBlock.Branch != null)
+            Operand lblTarget = context.GetLabel((ulong)op.Immediate);
+
+            if (onNotZero)
             {
-                Operand lblTarget = context.GetLabel((ulong)op.Immediate);
-
-                if (onNotZero)
-                {
-                    context.BranchIfTrue(lblTarget, value);
-                }
-                else
-                {
-                    context.BranchIfFalse(lblTarget, value);
-                }
-
-                if (context.CurrBlock.Next == null)
-                {
-                    EmitTailContinue(context, Const(op.Address + 4));
-                }
+                context.BranchIfTrue(lblTarget, value);
             }
             else
             {
-                Operand lblTaken = Label();
-
-                if (onNotZero)
-                {
-                    context.BranchIfTrue(lblTaken, value);
-                }
-                else
-                {
-                    context.BranchIfFalse(lblTaken, value);
-                }
-
-                EmitTailContinue(context, Const(op.Address + 4));
-
-                context.MarkLabel(lblTaken);
-
-                EmitTailContinue(context, Const(op.Immediate));
+                context.BranchIfFalse(lblTarget, value);
             }
         }
     }
diff --git a/ARMeilleure/Instructions/InstEmitFlow32.cs b/ARMeilleure/Instructions/InstEmitFlow32.cs
index 47233eb9..d7ebc945 100644
--- a/ARMeilleure/Instructions/InstEmitFlow32.cs
+++ b/ARMeilleure/Instructions/InstEmitFlow32.cs
@@ -15,14 +15,7 @@ namespace ARMeilleure.Instructions
         {
             IOpCode32BImm op = (IOpCode32BImm)context.CurrOp;
 
-            if (context.CurrBlock.Branch != null)
-            {
-                context.Branch(context.GetLabel((ulong)op.Immediate));
-            }
-            else
-            {
-                EmitTailContinue(context, Const(op.Immediate));
-            }
+            context.Branch(context.GetLabel((ulong)op.Immediate));
         }
 
         public static void Bl(ArmEmitterContext context)
diff --git a/ARMeilleure/Instructions/InstEmitFlowHelper.cs b/ARMeilleure/Instructions/InstEmitFlowHelper.cs
index e99afa75..6906f782 100644
--- a/ARMeilleure/Instructions/InstEmitFlowHelper.cs
+++ b/ARMeilleure/Instructions/InstEmitFlowHelper.cs
@@ -150,17 +150,32 @@ namespace ARMeilleure.Instructions
         private static void EmitNativeCall(ArmEmitterContext context, Operand nativeContextPtr, Operand funcAddr, bool isJump = false)
         {
             context.StoreToContext();
-            Operand returnAddress;
+
             if (isJump)
             {
                 context.Tailcall(funcAddr, nativeContextPtr);
             }
             else
             {
-                returnAddress = context.Call(funcAddr, OperandType.I64, nativeContextPtr);
+                OpCode op = context.CurrOp;
+
+                Operand returnAddress = context.Call(funcAddr, OperandType.I64, nativeContextPtr);
+
                 context.LoadFromContext();
 
-                EmitContinueOrReturnCheck(context, returnAddress);
+                // Note: The return value of a translated function is always an Int64 with the
+                // address execution has returned to. We expect this address to be immediately after the
+                // current instruction, if it isn't we keep returning until we reach the dispatcher.
+                Operand nextAddr = Const((long)op.Address + op.OpCodeSizeInBytes);
+
+                // Try to continue within this block.
+                // If the return address isn't to our next instruction, we need to return so the JIT can figure out what to do.
+                Operand lblContinue = context.GetLabel(nextAddr.Value);
+
+                // We need to clear out the call flag for the return address before comparing it.
+                context.BranchIfTrue(lblContinue, context.ICompareEqual(context.BitwiseAnd(returnAddress, Const(~CallFlag)), nextAddr));
+
+                context.Return(returnAddress);
             }
         }
 
@@ -191,46 +206,18 @@ namespace ARMeilleure.Instructions
             }
         }
 
-        private static void EmitContinueOrReturnCheck(ArmEmitterContext context, Operand returnAddress)
-        {
-            // Note: The return value of a translated function is always an Int64 with the
-            // address execution has returned to. We expect this address to be immediately after the
-            // current instruction, if it isn't we keep returning until we reach the dispatcher.
-            Operand nextAddr = Const(GetNextOpAddress(context.CurrOp));
-
-            // Try to continue within this block.
-            // If the return address isn't to our next instruction, we need to return so the JIT can figure out what to do.
-            Operand lblContinue = Label();
-
-            // We need to clear out the call flag for the return address before comparing it.
-            context.BranchIfTrue(lblContinue, context.ICompareEqual(context.BitwiseAnd(returnAddress, Const(~CallFlag)), nextAddr));
-
-            context.Return(returnAddress);
-
-            context.MarkLabel(lblContinue);
-
-            if (context.CurrBlock.Next == null)
-            {
-                // No code following this instruction, try and find the next block and jump to it.
-                EmitTailContinue(context, nextAddr);
-            }
-        }
-
-        private static ulong GetNextOpAddress(OpCode op)
-        {
-            return op.Address + (ulong)op.OpCodeSizeInBytes;
-        }
-
         public static void EmitTailContinue(ArmEmitterContext context, Operand address, bool allowRejit = false)
         {
-            bool useTailContinue = true; // Left option here as it may be useful if we need to return to managed rather than tail call in future. (eg. for debug)
+            // Left option here as it may be useful if we need to return to managed rather than tail call in future. 
+            // (eg. for debug)
+            bool useTailContinue = true;
 
             if (useTailContinue)
             {
                 if (context.HighCq)
                 {
-                    // If we're doing a tail continue in HighCq, reserve a space in the jump table to avoid calling back to the translator.
-                    // This will always try to get a HighCq version of our continue target as well.
+                    // If we're doing a tail continue in HighCq, reserve a space in the jump table to avoid calling back
+                    // to the translator. This will always try to get a HighCq version of our continue target as well.
                     EmitJumpTableBranch(context, address, true);
                 }
                 else
@@ -263,6 +250,7 @@ namespace ARMeilleure.Instructions
         {
             address = context.BitwiseOr(address, Const(address.Type, (long)CallFlag)); // Set call flag.
             Operand fallbackAddr = context.Call(typeof(NativeInterface).GetMethod(nameof(NativeInterface.GetFunctionAddress)), address);
+
             EmitNativeCall(context, fallbackAddr, isJump);
         }
 
@@ -273,39 +261,48 @@ namespace ARMeilleure.Instructions
             Operand endLabel = Label();
             Operand fallbackLabel = Label();
 
-            Action<Operand> emitTableEntry = (Operand entrySkipLabel) =>
+            void EmitTableEntry(Operand entrySkipLabel)
             {
                 // Try to take this entry in the table if its guest address equals 0.
                 Operand gotResult = context.CompareAndSwap(tableAddress, Const(0L), address);
 
                 // Is the address ours? (either taken via CompareAndSwap (0), or what was already here)
-                context.BranchIfFalse(entrySkipLabel, context.BitwiseOr(context.ICompareEqual(gotResult, address), context.ICompareEqual(gotResult, Const(0L))));
+                context.BranchIfFalse(entrySkipLabel, 
+                    context.BitwiseOr(
+                        context.ICompareEqual(gotResult, address), 
+                        context.ICompareEqual(gotResult, Const(0L)))
+                );
 
                 // It's ours, so what function is it pointing to?
                 Operand targetFunctionPtr = context.Add(tableAddress, Const(8L));
                 Operand targetFunction = context.Load(OperandType.I64, targetFunctionPtr);
 
                 // Call the function.
-                // We pass in the entry address as the guest address, as the entry may need to be updated by the indirect call stub.
+                // We pass in the entry address as the guest address, as the entry may need to be updated by the 
+                // indirect call stub.
                 EmitNativeCallWithGuestAddress(context, targetFunction, tableAddress, isJump);
+
                 context.Branch(endLabel);
-            };
+            }
 
             // Currently this uses a size of 1, as higher values inflate code size for no real benefit.
             for (int i = 0; i < JumpTable.DynamicTableElems; i++)
             {
                 if (i == JumpTable.DynamicTableElems - 1)
                 {
-                    emitTableEntry(fallbackLabel); // If this is the last entry, avoid emitting the additional label and add.
-                }
+                    // If this is the last entry, avoid emitting the additional label and add.
+                    EmitTableEntry(fallbackLabel);
+                } 
                 else
                 {
                     Operand nextLabel = Label();
 
-                    emitTableEntry(nextLabel);
+                    EmitTableEntry(nextLabel);
 
                     context.MarkLabel(nextLabel);
-                    tableAddress = context.Add(tableAddress, Const((long)JumpTable.JumpTableStride)); // Move to the next table entry.
+
+                    // Move to the next table entry.
+                    tableAddress = context.Add(tableAddress, Const((long)JumpTable.JumpTableStride));
                 }
             }
 
@@ -323,16 +320,15 @@ namespace ARMeilleure.Instructions
                 address = context.ZeroExtend32(OperandType.I64, address);
             }
 
-            // TODO: Constant folding. Indirect calls are slower in the best case and emit more code so we want to avoid them when possible.
+            // TODO: Constant folding. Indirect calls are slower in the best case and emit more code so we want to 
+            // avoid them when possible.
             bool isConst = address.Kind == OperandKind.Constant;
             long constAddr = (long)address.Value;
 
             if (!context.HighCq)
             {
-                // Don't emit indirect calls or jumps if we're compiling in lowCq mode.
-                // This avoids wasting space on the jump and indirect tables.
-                // Just ask the translator for the function address.
-
+                // Don't emit indirect calls or jumps if we're compiling in lowCq mode. This avoids wasting space on the
+                // jump and indirect tables. Just ask the translator for the function address.
                 EmitBranchFallback(context, address, isJump);
             }
             else if (!isConst)
@@ -376,7 +372,8 @@ namespace ARMeilleure.Instructions
 
                 Operand funcAddr = context.Load(OperandType.I64, tableEntryPtr);
 
-                EmitNativeCallWithGuestAddress(context, funcAddr, address, isJump); // Call the function directly. If it's not present yet, this will call the direct call stub.
+                // Call the function directly. If it's not present yet, this will call the direct call stub.
+                EmitNativeCallWithGuestAddress(context, funcAddr, address, isJump);
             }
         }
     }