Change placement of XMM callee save and restore code to match other compilers

ARMeilleure/CodeGen/X86/CodeGenContext.cs

@@ -22,7 +22,7 @@ namespace ARMeilleure.CodeGen.X86
         public BasicBlock CurrBlock { get; private set; }
 
         public int CallArgsRegionSize { get; }
-        public int VecCalleeSaveSize  { get; }
+        public int XmmSaveRegionSize  { get; }
 
         private long[] _blockOffsets;
 
@@ -80,24 +80,24 @@ namespace ARMeilleure.CodeGen.X86
 
             Assembler = new Assembler(stream);
 
-            CallArgsRegionSize = GetCallArgsRegionSize(allocResult, maxCallArgs, out int vecCalleeSaveSize);
+            CallArgsRegionSize = GetCallArgsRegionSize(allocResult, maxCallArgs, out int xmmSaveRegionSize);
-            VecCalleeSaveSize  = vecCalleeSaveSize;
+            XmmSaveRegionSize  = xmmSaveRegionSize;
 
             _blockOffsets = new long[blocksCount];
 
             _jumps = new List<Jump>();
         }
 
-        private int GetCallArgsRegionSize(AllocationResult allocResult, int maxCallArgs, out int vecCalleeSaveSize)
+        private int GetCallArgsRegionSize(AllocationResult allocResult, int maxCallArgs, out int xmmSaveRegionSize)
         {
             //We need to add 8 bytes to the total size, as the call to this
             //function already pushed 8 bytes (the return address).
             int intMask = CallingConvention.GetIntCalleeSavedRegisters() & allocResult.IntUsedRegisters;
             int vecMask = CallingConvention.GetVecCalleeSavedRegisters() & allocResult.VecUsedRegisters;
 
-            vecCalleeSaveSize = BitUtils.CountBits(vecMask) * 16;
+            xmmSaveRegionSize = BitUtils.CountBits(vecMask) * 16;
 
-            int calleeSaveRegionSize = BitUtils.CountBits(intMask) * 8 + vecCalleeSaveSize + 8;
+            int calleeSaveRegionSize = BitUtils.CountBits(intMask) * 8 + xmmSaveRegionSize + 8;
 
             int argsCount = maxCallArgs;
 

ARMeilleure/CodeGen/X86/CodeGenerator.cs

@@ -1523,9 +1523,23 @@ namespace ARMeilleure.CodeGen.X86
                 mask &= ~(1 << bit);
             }
 
-            mask = CallingConvention.GetVecCalleeSavedRegisters() & context.AllocResult.VecUsedRegisters;
+            int reservedStackSize = context.CallArgsRegionSize + context.AllocResult.SpillRegionSize;
 
-            int offset = 0;
+            reservedStackSize += context.XmmSaveRegionSize;
+
+            if (reservedStackSize >= StackGuardSize)
+            {
+                GenerateInlineStackProbe(context, reservedStackSize);
+            }
+
+            if (reservedStackSize != 0)
+            {
+                context.Assembler.Sub(rsp, new Operand(reservedStackSize), OperandType.I64);
+            }
+
+            int offset = reservedStackSize;
+
+            mask = CallingConvention.GetVecCalleeSavedRegisters() & context.AllocResult.VecUsedRegisters;
 
             while (mask != 0)
             {
@@ -1542,20 +1556,6 @@ namespace ARMeilleure.CodeGen.X86
                 mask &= ~(1 << bit);
             }
 
-            int reservedStackSize = context.CallArgsRegionSize + context.AllocResult.SpillRegionSize;
-
-            reservedStackSize += context.VecCalleeSaveSize;
-
-            if (reservedStackSize >= StackGuardSize)
-            {
-                GenerateInlineStackProbe(context, reservedStackSize);
-            }
-
-            if (reservedStackSize != 0)
-            {
-                context.Assembler.Sub(rsp, new Operand(reservedStackSize), OperandType.I64);
-            }
-
             return new UnwindInfo(pushEntries.ToArray(), context.StreamOffset, reservedStackSize);
         }
 
@@ -1565,17 +1565,12 @@ namespace ARMeilleure.CodeGen.X86
 
             int reservedStackSize = context.CallArgsRegionSize + context.AllocResult.SpillRegionSize;
 
-            reservedStackSize += context.VecCalleeSaveSize;
+            reservedStackSize += context.XmmSaveRegionSize;
 
-            if (reservedStackSize != 0)
+            int offset = reservedStackSize;
-            {
-                context.Assembler.Add(rsp, new Operand(reservedStackSize), OperandType.I64);
-            }
 
             int mask = CallingConvention.GetVecCalleeSavedRegisters() & context.AllocResult.VecUsedRegisters;
 
-            int offset = 0;
-
             while (mask != 0)
             {
                 int bit = BitUtils.LowestBitSet(mask);
@@ -1589,6 +1584,11 @@ namespace ARMeilleure.CodeGen.X86
                 mask &= ~(1 << bit);
             }
 
+            if (reservedStackSize != 0)
+            {
+                context.Assembler.Add(rsp, new Operand(reservedStackSize), OperandType.I64);
+            }
+
             mask = CallingConvention.GetIntCalleeSavedRegisters() & context.AllocResult.IntUsedRegisters;
 
             while (mask != 0)

ARMeilleure/Translation/JitCache.cs

@@ -58,7 +58,7 @@ namespace ARMeilleure.Translation
         {
             // Map pages that are already full as RX.
             // Map pages that are not full yet as RWX.
-            // On unix, the address and size must be page aligned.
+            // On unix, the address must be page aligned.
             int endOffs = offset + size;
 
             int pageStart = offset  & ~PageMask;

ARMeilleure/Translation/JitUnwindWindows.cs

@@ -99,31 +99,48 @@ namespace ARMeilleure.Translation
 
             var unwindInfo = funcEntry.UnwindInfo;
 
-            _unwindInfo->UnwindCodes[0] = PackUwop(UnwindOperation.AllocLarge, unwindInfo.PrologueSize, 1);
+            int codeIndex = 0;
-            _unwindInfo->UnwindCodes[1] = (ushort)(unwindInfo.FixedAllocSize >> 0);
-            _unwindInfo->UnwindCodes[2] = (ushort)(unwindInfo.FixedAllocSize >> 16);
 
-            int codeIndex = 3;
+            int spOffset = unwindInfo.FixedAllocSize;
 
-            int spOffset = 0;
+            foreach (var entry in unwindInfo.PushEntries)
+            {
+                if (entry.Type == RegisterType.Vector)
+                {
+                    spOffset -= 16;
+                }
+            }
 
             for (int index = unwindInfo.PushEntries.Length - 1; index >= 0; index--)
             {
                 var entry = unwindInfo.PushEntries[index];
 
-                bool isInteger = entry.Type == RegisterType.Integer;
+                if (entry.Type == RegisterType.Vector)
-
-                UnwindOperation uwop = isInteger
-                    ? UnwindOperation.PushNonvol
-                    : UnwindOperation.SaveXmm128;
-
-                _unwindInfo->UnwindCodes[codeIndex++] = PackUwop(uwop, entry.StreamEndOffset, entry.Index);
-
-                if (!isInteger)
                 {
+                    ushort uwop = PackUwop(UnwindOperation.SaveXmm128, entry.StreamEndOffset, entry.Index);
+
+                    _unwindInfo->UnwindCodes[codeIndex++] = uwop;
                     _unwindInfo->UnwindCodes[codeIndex++] = (ushort)spOffset;
 
-                    spOffset -= 16;
+                    spOffset += 16;
+                }
+            }
+
+            _unwindInfo->UnwindCodes[0] = PackUwop(UnwindOperation.AllocLarge, unwindInfo.PrologueSize, 1);
+            _unwindInfo->UnwindCodes[1] = (ushort)(unwindInfo.FixedAllocSize >> 0);
+            _unwindInfo->UnwindCodes[2] = (ushort)(unwindInfo.FixedAllocSize >> 16);
+
+            codeIndex += 3;
+
+            for (int index = unwindInfo.PushEntries.Length - 1; index >= 0; index--)
+            {
+                var entry = unwindInfo.PushEntries[index];
+
+                if (entry.Type == RegisterType.Integer)
+                {
+                    ushort uwop = PackUwop(UnwindOperation.PushNonvol, entry.StreamEndOffset, entry.Index);
+
+                    _unwindInfo->UnwindCodes[codeIndex++] = uwop;
                 }
             }