2 files changed, 71 insertions, 1 deletions
diff --git a/ARMeilleure/CodeGen/Optimizations/BlockPlacement.cs b/ARMeilleure/CodeGen/Optimizations/BlockPlacement.cs
new file mode 100644
index 00000000..a200f54e
--- /dev/null
+++ b/ARMeilleure/CodeGen/Optimizations/BlockPlacement.cs
@@ -0,0 +1,66 @@
+using ARMeilleure.IntermediateRepresentation;
+using ARMeilleure.Translation;
+using System.Diagnostics;
+
+using static ARMeilleure.IntermediateRepresentation.OperandHelper;
+
+namespace ARMeilleure.CodeGen.Optimizations
+{
+    static class BlockPlacement
+    {
+        public static void RunPass(ControlFlowGraph cfg)
+        {
+            bool update = false;
+
+            BasicBlock block;
+            BasicBlock nextBlock;
+
+            BasicBlock lastBlock = cfg.Blocks.Last;
+
+            // Move cold blocks at the end of the list, so that they are emitted away from hot code.
+            for (block = cfg.Blocks.First; block != lastBlock; block = nextBlock)
+            {
+                nextBlock = block.ListNext;
+
+                if (block.Frequency == BasicBlockFrequency.Cold)
+                {
+                    cfg.Blocks.Remove(block);
+                    cfg.Blocks.AddLast(block);
+                }
+            }
+
+            for (block = cfg.Blocks.First; block != null; block = nextBlock)
+            {
+                nextBlock = block.ListNext;
+
+                if (block.SuccessorCount == 2 && block.Operations.Last is Operation branchOp)
+                {
+                    Debug.Assert(branchOp.Instruction == Instruction.BranchIf);
+
+                    BasicBlock falseSucc = block.GetSuccessor(0);
+                    BasicBlock trueSucc = block.GetSuccessor(1);
+
+                    // If true successor is next block in list, invert the condition. We avoid extra branching by
+                    // making the true side the fallthrough (i.e, convert it to the false side).
+                    if (trueSucc == block.ListNext)
+                    {
+                        Comparison comp = (Comparison)branchOp.GetSource(2).AsInt32();
+                        Comparison compInv = comp.Invert();
+
+                        branchOp.SetSource(2, Const((int)compInv));
+
+                        block.SetSuccessor(0, trueSucc);
+                        block.SetSuccessor(1, falseSucc);
+
+                        update = true;
+                    }
+                }
+            }
+
+            if (update)
+            {
+                cfg.Update(removeUnreachableBlocks: false);
+            }
+        }
+    }
+}
diff --git a/ARMeilleure/CodeGen/X86/CodeGenerator.cs b/ARMeilleure/CodeGen/X86/CodeGenerator.cs
index a51f4a13..c9acd945 100644
--- a/ARMeilleure/CodeGen/X86/CodeGenerator.cs
+++ b/ARMeilleure/CodeGen/X86/CodeGenerator.cs
@@ -106,6 +106,8 @@ namespace ARMeilleure.CodeGen.X86
 
             X86Optimizer.RunPass(cfg);
 
+            BlockPlacement.RunPass(cfg);
+
             Logger.EndPass(PassName.Optimization, cfg);
 
             Logger.StartPass(PassName.PreAllocation);
@@ -186,9 +188,11 @@ namespace ARMeilleure.CodeGen.X86
                     }
                 }
 
+                byte[] code = context.GetCode();
+
                 Logger.EndPass(PassName.CodeGeneration);
 
-                return new CompiledFunction(context.GetCode(), unwindInfo);
+                return new CompiledFunction(code, unwindInfo);
             }
         }