Fix arguments count with struct return values, other fixes

ARMeilleure/CodeGen/RegisterAllocators/AllocationResult.cs

@@ -5,18 +5,15 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
         public int IntUsedRegisters { get; }
         public int VecUsedRegisters { get; }
         public int SpillRegionSize  { get; }
-        public int MaxCallArgs      { get; }
 
         public AllocationResult(
             int intUsedRegisters,
             int vecUsedRegisters,
-            int spillRegionSize,
+            int spillRegionSize)
-            int maxCallArgs)
         {
             IntUsedRegisters = intUsedRegisters;
             VecUsedRegisters = vecUsedRegisters;
             SpillRegionSize  = spillRegionSize;
-            MaxCallArgs      = maxCallArgs;
         }
     }
 }

ARMeilleure/CodeGen/RegisterAllocators/LinearScan.cs

@@ -77,7 +77,7 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
 
             AllocationContext context = new AllocationContext(regMasks, _intervals.Count);
 
-            BuildIntervals(cfg, context, out int maxCallArgs);
+            BuildIntervals(cfg, context);
 
             for (int index = 0; index < _intervals.Count; index++)
             {
@@ -115,8 +115,7 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             return new AllocationResult(
                 context.IntUsedRegisters,
                 context.VecUsedRegisters,
-                context.StackAlloc.TotalSize,
+                context.StackAlloc.TotalSize);
-                maxCallArgs);
         }
 
         private void AllocateInterval(AllocationContext context, LiveInterval current, int cIndex)
@@ -774,10 +773,8 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             _parentIntervals = _intervals.ToArray();
         }
 
-        private void BuildIntervals(ControlFlowGraph cfg, AllocationContext context, out int maxCallArgs)
+        private void BuildIntervals(ControlFlowGraph cfg, AllocationContext context)
         {
-            maxCallArgs = 0;
-
             _blockRanges = new LiveRange[cfg.Blocks.Count];
 
             int mapSize = _intervals.Count;
@@ -912,11 +909,6 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
                         {
                             AddIntervalCallerSavedReg(context.Masks.IntCallerSavedRegisters, operationPos, RegisterType.Integer);
                             AddIntervalCallerSavedReg(context.Masks.VecCallerSavedRegisters, operationPos, RegisterType.Vector);
-
-                            if (maxCallArgs < operation.SourcesCount - 1)
-                            {
-                                maxCallArgs = operation.SourcesCount - 1;
-                            }
                         }
                         else if (operation.Inst == Instruction.StackAlloc)
                         {

ARMeilleure/CodeGen/X86/CodeGenContext.cs

@@ -71,7 +71,7 @@ namespace ARMeilleure.CodeGen.X86
         private long _jNearPosition;
         private int  _jNearLength;
 
-        public CodeGenContext(Stream stream, AllocationResult allocResult, int blocksCount)
+        public CodeGenContext(Stream stream, AllocationResult allocResult, int maxCallArgs, int blocksCount)
         {
             _stream = stream;
 
@@ -79,7 +79,7 @@ namespace ARMeilleure.CodeGen.X86
 
             Assembler = new Assembler(stream);
 
-            CallArgsRegionSize = GetCallArgsRegionSize(allocResult, out int vecCalleeSaveSize);
+            CallArgsRegionSize = GetCallArgsRegionSize(allocResult, maxCallArgs, out int vecCalleeSaveSize);
 
             VecCalleeSaveSize = vecCalleeSaveSize;
 
@@ -88,7 +88,7 @@ namespace ARMeilleure.CodeGen.X86
             _jumps = new List<Jump>();
         }
 
-        private int GetCallArgsRegionSize(AllocationResult allocResult, out int vecCalleeSaveSize)
+        private int GetCallArgsRegionSize(AllocationResult allocResult, int maxCallArgs, out int vecCalleeSaveSize)
         {
             //We need to add 8 bytes to the total size, as the call to this
             //function already pushed 8 bytes (the return address).
@@ -101,7 +101,7 @@ namespace ARMeilleure.CodeGen.X86
 
             int calleeSaveRegionSize = BitUtils.CountBits(intMask) * 8 + vecCalleeSaveSize + 8;
 
-            int argsCount = allocResult.MaxCallArgs;
+            int argsCount = maxCallArgs;
 
             //The ABI mandates that the space for at least 4 arguments
             //is reserved on the stack (this is called shadow space).

ARMeilleure/CodeGen/X86/CodeGenerator.cs

@@ -231,7 +231,7 @@ namespace ARMeilleure.CodeGen.X86
 
             Logger.StartPass(PassName.PreAllocation);
 
-            PreAllocator.RunPass(cfg, memory);
+            PreAllocator.RunPass(cfg, memory, out int maxCallArgs);
 
             Logger.EndPass(PassName.PreAllocation, cfg);
 
@@ -253,7 +253,7 @@ namespace ARMeilleure.CodeGen.X86
 
             using (MemoryStream stream = new MemoryStream())
             {
-                CodeGenContext context = new CodeGenContext(stream, allocResult, cfg.Blocks.Count);
+                CodeGenContext context = new CodeGenContext(stream, allocResult, maxCallArgs, cfg.Blocks.Count);
 
                 WritePrologue(context);
 

ARMeilleure/CodeGen/X86/PreAllocator.cs

@@ -10,8 +10,10 @@ namespace ARMeilleure.CodeGen.X86
 {
     static class PreAllocator
     {
-        public static void RunPass(ControlFlowGraph cfg, MemoryManager memory)
+        public static void RunPass(ControlFlowGraph cfg, MemoryManager memory, out int maxCallArgs)
         {
+            maxCallArgs = 0;
+
             foreach (BasicBlock block in cfg.Blocks)
             {
                 LinkedListNode<Node> nextNode;
@@ -55,6 +57,24 @@ namespace ARMeilleure.CodeGen.X86
                         ReplaceNegateWithXor(node, operation);
                     }
 
+                    //Get the maximum number of arguments used on a call. On windows,
+                    //when a struct is returned from the call, we also need to pass
+                    //the pointer where the struct should be written on the first argument.
+                    if (inst == Instruction.Call)
+                    {
+                        int argsCount = operation.SourcesCount - 1;
+
+                        if (operation.Dest != null && operation.Dest.Type == OperandType.V128)
+                        {
+                            argsCount++;
+                        }
+
+                        if (maxCallArgs < argsCount)
+                        {
+                            maxCallArgs = argsCount;
+                        }
+                    }
+
                     HandleFixedRegisterCopy(node, operation);
 
                     HandleSameDestSrc1Copy(node, operation);
@@ -597,6 +617,8 @@ namespace ARMeilleure.CodeGen.X86
             {
                 case Instruction.Copy:
                 case Instruction.LoadFromContext:
+                case Instruction.Spill:
+                case Instruction.SpillArg:
                 case Instruction.StoreToContext:
                     return true;
             }

ARMeilleure/Instructions/SoftFallback.cs

@@ -882,7 +882,7 @@ namespace ARMeilleure.Instructions
             byte[] res   = new byte[16];
             byte[] table = new byte[tb.Length * 16];
 
-            for (byte index  = 0; index  < tb.Length; index++)
+            for (byte index = 0; index < tb.Length; index++)
             {
                 Buffer.BlockCopy(tb[index].ToArray(), 0, table, index * 16, 16);
             }
@@ -891,11 +891,11 @@ namespace ARMeilleure.Instructions
 
             for (byte index = 0; index < bytes; index++)
             {
-                byte tblIdx = v[index];
+                byte tblIndex = v[index];
 
-                if (tblIdx < table.Length)
+                if (tblIndex < table.Length)
                 {
-                    res[index] = table[tblIdx];
+                    res[index] = table[tblIndex];
                 }
             }