JITs: Expand max spill slot size to 32 bytes

This will be necessary to handle spilling 256-bit vectors.

External/FEXCore/Source/Interface/Core/JIT/Arm64/JIT.cpp

@@ -783,10 +783,12 @@ void *Arm64JITCore::CompileCode(uint64_t Entry,
   SpillSlots = RAData->SpillSlots();
 
   if (SpillSlots) {
-    if (IsImmAddSub(SpillSlots * 16)) {
-      sub(sp, sp, SpillSlots * 16);
+    const auto TotalSpillSlotsSize = SpillSlots * MaxSpillSlotSize;
+
+    if (IsImmAddSub(TotalSpillSlotsSize)) {
+      sub(sp, sp, TotalSpillSlotsSize);
     } else {
-      LoadConstant(x0, SpillSlots * 16);
+      LoadConstant(x0, TotalSpillSlotsSize);
       sub(sp, sp, x0);
     }
   }
@@ -854,14 +856,17 @@ void *Arm64JITCore::CompileCode(uint64_t Entry,
 }
 
 void Arm64JITCore::ResetStack() {
-  if (SpillSlots == 0)
+  if (SpillSlots == 0) {
     return;
+  }
 
-  if (IsImmAddSub(SpillSlots * 16)) {
-    add(sp, sp, SpillSlots * 16);
+  const auto TotalSpillSlotsSize = SpillSlots * MaxSpillSlotSize;
+
+  if (IsImmAddSub(TotalSpillSlotsSize)) {
+    add(sp, sp, TotalSpillSlotsSize);
   } else {
    // Too big to fit in a 12bit immediate
-   LoadConstant(x0, SpillSlots * 16);
+   LoadConstant(x0, TotalSpillSlotsSize);
    add(sp, sp, x0);
   }
 }

External/FEXCore/Source/Interface/Core/JIT/Arm64/MemoryOps.cpp

@@ -446,9 +446,9 @@ DEF_OP(StoreContextIndexed) {
 }
 
 DEF_OP(SpillRegister) {
-  auto Op = IROp->C<IR::IROp_SpillRegister>();
+  const auto Op = IROp->C<IR::IROp_SpillRegister>();
   const uint8_t OpSize = IROp->Size;
-  const uint32_t SlotOffset = Op->Slot * 16;
+  const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;
 
   if (Op->Class == FEXCore::IR::GPRClass) {
     switch (OpSize) {
@@ -492,9 +492,9 @@ DEF_OP(SpillRegister) {
 }
 
 DEF_OP(FillRegister) {
-  auto Op = IROp->C<IR::IROp_FillRegister>();
-  uint8_t OpSize = IROp->Size;
-  uint32_t SlotOffset = Op->Slot * 16;
+  const auto Op = IROp->C<IR::IROp_FillRegister>();
+  const uint8_t OpSize = IROp->Size;
+  const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;
 
   if (Op->Class == FEXCore::IR::GPRClass) {
     switch (OpSize) {

External/FEXCore/Source/Interface/Core/JIT/x86_64/BranchOps.cpp

@@ -33,7 +33,7 @@ namespace FEXCore::CPU {
 DEF_OP(SignalReturn) {
   // Adjust the stack first for a regular return
   if (SpillSlots) {
-    add(rsp, SpillSlots * 16); // + 8 to consume return address
+    add(rsp, SpillSlots * MaxSpillSlotSize); // + 8 to consume return address
   }
 
   jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SignalReturnHandler)]);
@@ -42,7 +42,7 @@ DEF_OP(SignalReturn) {
 DEF_OP(CallbackReturn) {
   // Adjust the stack first for a regular return
   if (SpillSlots) {
-    add(rsp, SpillSlots * 16); // + 8 to consume return address
+    add(rsp, SpillSlots * MaxSpillSlotSize); // + 8 to consume return address
   }
 
   // Make sure to adjust the refcounter so we don't clear the cache now
@@ -71,7 +71,7 @@ DEF_OP(ExitFunction) {
 
 
   if (SpillSlots) {
-    add(rsp, SpillSlots * 16);
+    add(rsp, SpillSlots * MaxSpillSlotSize);
   }
 
   uint64_t NewRIP;

External/FEXCore/Source/Interface/Core/JIT/x86_64/JIT.cpp

@@ -599,7 +599,7 @@ void *X86JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRLi
   SpillSlots = RAData->SpillSlots();
 
   if (SpillSlots) {
-    sub(rsp, SpillSlots * 16);
+    sub(rsp, SpillSlots * MaxSpillSlotSize);
   }
 
 #ifdef BLOCKSTATS

External/FEXCore/Source/Interface/Core/JIT/x86_64/MemoryOps.cpp

@@ -333,10 +333,10 @@ DEF_OP(StoreContextIndexed) {
 }
 
 DEF_OP(SpillRegister) {
-  auto Op = IROp->C<IR::IROp_SpillRegister>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_SpillRegister>();
+  const uint8_t OpSize = IROp->Size;
+  const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;
 
-  uint32_t SlotOffset = Op->Slot * 16;
   if (Op->Class == FEXCore::IR::GPRClass) {
     switch (OpSize) {
       case 1: {
@@ -381,10 +381,10 @@ DEF_OP(SpillRegister) {
 }
 
 DEF_OP(FillRegister) {
-  auto Op = IROp->C<IR::IROp_FillRegister>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_FillRegister>();
+  const uint8_t OpSize = IROp->Size;
+  const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;
 
-  uint32_t SlotOffset = Op->Slot * 16;
   if (Op->Class == FEXCore::IR::GPRClass) {
     switch (OpSize) {
       case 1: {

External/FEXCore/Source/Interface/Core/JIT/x86_64/MiscOps.cpp

@@ -47,7 +47,7 @@ DEF_OP(Break) {
   auto Op = IROp->C<IR::IROp_Break>();
 
   if (SpillSlots) {
-    add(rsp, SpillSlots * 16);
+    add(rsp, SpillSlots * MaxSpillSlotSize);
   }
 
   mov(byte [STATE + offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData.FaultToTopAndGeneratedException)], 1);

External/FEXCore/include/FEXCore/Core/CPUBackend.h

@@ -122,6 +122,10 @@ namespace CPU {
     bool IsAddressInCodeBuffer(uintptr_t Address) const;
 
   protected:
+    // Max spill slot size in bytes. We need at most 32 bytes
+    // to be able to handle a 256-bit vector store to a slot.
+    constexpr static uint32_t MaxSpillSlotSize = 32;
+
     FEXCore::Core::InternalThreadState *ThreadState;
 
     size_t InitialCodeSize, MaxCodeSize;