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

1 files changed, 32 insertions, 18 deletions

diff --git a/ARMeilleure/Decoders/Optimizations/TailCallRemover.cs b/ARMeilleure/Decoders/Optimizations/TailCallRemover.cs
index 2d6439ba..e64f9a54 100644
--- a/ARMeilleure/Decoders/Optimizations/TailCallRemover.cs
+++ b/ARMeilleure/Decoders/Optimizations/TailCallRemover.cs
@@ -1,16 +1,15 @@
-using ARMeilleure.Decoders;
-using System;
+using System;
 using System.Collections.Generic;
 
 namespace ARMeilleure.Decoders.Optimizations
 {
     static class TailCallRemover
     {
-        public static void RunPass(ulong entryAddress, List<Block> blocks)
+        public static Block[] RunPass(ulong entryAddress, List<Block> blocks)
         {
             // Detect tail calls:
             // - Assume this function spans the space covered by contiguous code blocks surrounding the entry address.
-            // - Unconditional jump to an area outside this contiguous region will be treated as a tail call.
+            // - A jump to an area outside this contiguous region will be treated as an exit block.
             // - Include a small allowance for jumps outside the contiguous range.
 
             if (!Decoder.BinarySearch(blocks, entryAddress, out int entryBlockId))
@@ -19,57 +18,72 @@ namespace ARMeilleure.Decoders.Optimizations
             }
 
             const ulong allowance = 4;
+
             Block entryBlock = blocks[entryBlockId];
-            int startBlockIndex = entryBlockId;
+
             Block startBlock = entryBlock;
-            int endBlockIndex = entryBlockId;
-            Block endBlock = entryBlock;
+            Block endBlock   = entryBlock;
+
+            int startBlockIndex = entryBlockId;
+            int endBlockIndex   = entryBlockId;
 
             for (int i = entryBlockId + 1; i < blocks.Count; i++) // Search forwards.
             {
                 Block block = blocks[i];
+
                 if (endBlock.EndAddress < block.Address - allowance)
                 {
                     break; // End of contiguous function.
                 }
 
-                endBlock = block;
+                endBlock      = block;
                 endBlockIndex = i;
             }
 
             for (int i = entryBlockId - 1; i >= 0; i--) // Search backwards.
             {
                 Block block = blocks[i];
+
                 if (startBlock.Address > block.EndAddress + allowance)
                 {
                     break; // End of contiguous function.
                 }
 
-                startBlock = block;
+                startBlock      = block;
                 startBlockIndex = i;
             }
 
             if (startBlockIndex == 0 && endBlockIndex == blocks.Count - 1)
             {
-                return; // Nothing to do here.
+                return blocks.ToArray(); // Nothing to do here.
             }
-
-            // Replace all branches to blocks outside the range with null, and force a tail call.
-
+            
+            // Mark branches outside of contiguous region as exit blocks.
             for (int i = startBlockIndex; i <= endBlockIndex; i++)
             {
                 Block block = blocks[i];
+
                 if (block.Branch != null && (block.Branch.Address > endBlock.EndAddress || block.Branch.EndAddress < startBlock.Address))
                 {
-                    block.Branch = null;
-                    block.TailCall = true;
+                    block.Branch.Exit     = true;
+                    block.Branch.TailCall = true;
                 }
             }
 
-            // Finally, delete all blocks outside the contiguous range.
+           var newBlocks = new List<Block>(blocks.Count);
+
+            // Finally, rebuild decoded block list, ignoring blocks outside the contiguous range.
+            for (int i = 0; i < blocks.Count; i++)
+            {
+                Block block = blocks[i];
+
+                if (block.Exit || (i >= startBlockIndex && i <= endBlockIndex))
+                {
+                    newBlocks.Add(block);
+                }
+            }
 
-            blocks.RemoveRange(endBlockIndex + 1, (blocks.Count - endBlockIndex) - 1);
-            blocks.RemoveRange(0, startBlockIndex);
+            return newBlocks.ToArray();
         }
     }
 }