Return address handling. Not happy with the design, but it (sorta) works.

src/alloy/backend/ivm/ivm_function.cc

@@ -34,7 +34,7 @@ void IVMFunction::Setup(TranslationContext& ctx) {
   intcodes_ = (IntCode*)ctx.intcode_arena->CloneContents();
 }
 
-int IVMFunction::CallImpl(ThreadState* thread_state) {
+int IVMFunction::CallImpl(ThreadState* thread_state, uint64_t return_address) {
   // Setup register file on stack.
   size_t register_file_size = register_count_ * sizeof(Register);
   Register* register_file = (Register*)alloca(register_file_size);
@@ -46,7 +46,8 @@ int IVMFunction::CallImpl(ThreadState* thread_state) {
   ics.did_carry = 0;
   ics.access_callbacks = thread_state->runtime()->access_callbacks();
   ics.thread_state = thread_state;
-  ics.return_address = 0xBEBEBEBE;
+  ics.return_address = return_address;
+  ics.call_return_address = 0;
 
   // TODO(benvanik): DID_CARRY -- need HIR to set a OPCODE_FLAG_SET_CARRY
   //                 or something so the fns can set an ics flag.

src/alloy/backend/ivm/ivm_function.h

@@ -29,7 +29,8 @@ public:
   void Setup(TranslationContext& ctx);
 
 protected:
-  virtual int CallImpl(runtime::ThreadState* thread_state);
+  virtual int CallImpl(runtime::ThreadState* thread_state,
+                       uint64_t return_address);
 
 private:
 

src/alloy/backend/ivm/ivm_intcode.cc

@@ -523,10 +523,13 @@ int Translate_TRAP_TRUE(TranslationContext& ctx, Instr* i) {
 
 uint32_t IntCode_CALL_XX(IntCodeState& ics, const IntCode* i, uint32_t reg) {
   FunctionInfo* symbol_info = (FunctionInfo*)ics.rf[reg].u64;
-  Function* fn;
+  Function* fn = NULL;
   ics.thread_state->runtime()->ResolveFunction(symbol_info->address(), &fn);
+  XEASSERTNOTNULL(fn);
   // TODO(benvanik): proper tail call support, somehow.
-  fn->Call(ics.thread_state);
+  uint64_t return_address =
+      (i->flags & CALL_TAIL) ? ics.return_address : ics.call_return_address;
+  fn->Call(ics.thread_state, return_address);
   if (i->flags & CALL_TAIL) {
     return IA_RETURN;
   }
@@ -589,16 +592,21 @@ int Translate_CALL_TRUE(TranslationContext& ctx, Instr* i) {
 }
 
 uint32_t IntCode_CALL_INDIRECT_XX(IntCodeState& ics, const IntCode* i, uint32_t reg) {
+  uint64_t target = ics.rf[reg].u64;
+
   // Check if return address - if so, return.
-  /*if (ics.rf[reg].u64 == ics.return_address) {
+  if (target == ics.return_address) {
     return IA_RETURN;
-  }*/
+  }
 
   // Real call.
-  Function* fn;
-  ics.thread_state->runtime()->ResolveFunction(ics.rf[reg].u64, &fn);
+  Function* fn = NULL;
+  ics.thread_state->runtime()->ResolveFunction(target, &fn);
+  XEASSERTNOTNULL(fn);
   // TODO(benvanik): proper tail call support, somehow.
-  fn->Call(ics.thread_state);
+  uint64_t return_address =
+      (i->flags & CALL_TAIL) ? ics.return_address : ics.call_return_address;
+  fn->Call(ics.thread_state, return_address);
   if (i->flags & CALL_TAIL) {
     return IA_RETURN;
   }
@@ -667,6 +675,14 @@ int Translate_RETURN(TranslationContext& ctx, Instr* i) {
   return DispatchToC(ctx, i, IntCode_RETURN);
 }
 
+uint32_t IntCode_SET_RETURN_ADDRESS(IntCodeState& ics, const IntCode* i) {
+  ics.call_return_address = ics.rf[i->src1_reg].u32;
+  return IA_NEXT;
+}
+int Translate_SET_RETURN_ADDRESS(TranslationContext& ctx, Instr* i) {
+  return DispatchToC(ctx, i, IntCode_SET_RETURN_ADDRESS);
+}
+
 uint32_t IntCode_BRANCH_XX(IntCodeState& ics, const IntCode* i, uint32_t reg) {
   return ics.rf[reg].u32;
 }
@@ -1167,6 +1183,7 @@ uint32_t IntCode_LOAD_I32(IntCodeState& ics, const IntCode* i) {
   if (DYNAMIC_REGISTER_ACCESS_CHECK(address)) {
     return IntCode_LOAD_REGISTER_I32_DYNAMIC(ics, i);
   }
+  DFLUSH();
   DPRINT("%d (%X) = load.i32 %.8X\n",
          *((int32_t*)(ics.membase + address)),
          *((uint32_t*)(ics.membase + address)),
@@ -2827,6 +2844,7 @@ static const TranslateFn dispatch_table[] = {
   Translate_CALL_INDIRECT,
   Translate_CALL_INDIRECT_TRUE,
   Translate_RETURN,
+  Translate_SET_RETURN_ADDRESS,
 
   Translate_BRANCH,
   Translate_BRANCH_IF,

src/alloy/backend/ivm/ivm_intcode.h

@@ -47,6 +47,7 @@ typedef struct {
   runtime::RegisterAccessCallbacks* access_callbacks;
   runtime::ThreadState* thread_state;
   uint64_t      return_address;
+  uint64_t      call_return_address;
 } IntCodeState;
 
 

src/alloy/frontend/ppc/ppc_emit_control.cc

@@ -32,10 +32,12 @@ int InstrEmit_branch(
   // The docs say always, though...
   // Note that we do the update before we branch/call as we need it to
   // be correct for returns.
+  Value* return_address = f.LoadConstant(cia + 4);
+  f.SetReturnAddress(return_address);
   if (lk) {
-    f.StoreLR(f.LoadConstant(cia + 4));
+    f.StoreLR(return_address);
   }
-  
+
   if (!lk) {
     // If LR is not set this call will never return here.
     call_flags |= CALL_TAIL;
@@ -301,8 +303,8 @@ XEEMITTER(bclrx,        0x4C000020, XL )(PPCFunctionBuilder& f, InstrData& i) {
   if (!i.XL.LK && !ok) {
     // Return (most likely).
     // TODO(benvanik): test? ReturnCheck()?
-    f.Return();
-    return 0;
+    //f.Return();
+    //return 0;
   }
 
   return InstrEmit_branch(

src/alloy/hir/function_builder.cc

@@ -459,6 +459,12 @@ void FunctionBuilder::Return() {
   EndBlock();
 }
 
+void FunctionBuilder::SetReturnAddress(Value* value) {
+  Instr* i = AppendInstr(OPCODE_SET_RETURN_ADDRESS_info, 0);
+  i->set_src1(value);
+  i->src2.value = i->src3.value = NULL;
+}
+
 void FunctionBuilder::Branch(Label* label, uint32_t branch_flags) {
   Instr* i = AppendInstr(OPCODE_BRANCH_info, branch_flags);
   i->src1.label = label;

src/alloy/hir/function_builder.h

@@ -70,6 +70,7 @@ public:
   void CallIndirect(Value* value, uint32_t call_flags = 0);
   void CallIndirectTrue(Value* cond, Value* value, uint32_t call_flags = 0);
   void Return();
+  void SetReturnAddress(Value* value);
 
   void Branch(Label* label, uint32_t branch_flags = 0);
   void BranchIf(Value* cond, Label* true_label, Label* false_label,

src/alloy/hir/opcodes.h

@@ -75,6 +75,7 @@ enum Opcode {
   OPCODE_CALL_INDIRECT,
   OPCODE_CALL_INDIRECT_TRUE,
   OPCODE_RETURN,
+  OPCODE_SET_RETURN_ADDRESS,
 
   OPCODE_BRANCH,
   OPCODE_BRANCH_IF,

src/alloy/hir/opcodes.inl

@@ -74,6 +74,12 @@ DEFINE_OPCODE(
     OPCODE_SIG_X,
     OPCODE_FLAG_BRANCH);
 
+DEFINE_OPCODE(
+    OPCODE_SET_RETURN_ADDRESS,
+    "set_return_address",
+    OPCODE_SIG_X_V,
+    0);
+
 DEFINE_OPCODE(
     OPCODE_BRANCH,
     "branch",

src/alloy/runtime/function.cc

@@ -23,12 +23,12 @@ Function::Function(Type type, uint64_t address) :
 Function::~Function() {
 }
 
-int Function::Call(ThreadState* thread_state) {
+int Function::Call(ThreadState* thread_state, uint64_t return_address) {
   ThreadState* original_thread_state = ThreadState::Get();
   if (original_thread_state != thread_state) {
     ThreadState::Bind(thread_state);
   }
-  int result = CallImpl(thread_state);
+  int result = CallImpl(thread_state, return_address);
   if (original_thread_state != thread_state) {
     ThreadState::Bind(original_thread_state);
   }
@@ -44,7 +44,8 @@ ExternFunction::ExternFunction(
 ExternFunction::~ExternFunction() {
 }
 
-int ExternFunction::CallImpl(ThreadState* thread_state) {
+int ExternFunction::CallImpl(ThreadState* thread_state,
+                             uint64_t return_address) {
   if (!handler_) {
     XELOGW("undefined extern call to %.8X", address());
     return 0;

src/alloy/runtime/function.h

@@ -34,10 +34,10 @@ public:
   Type type() const { return type_; }
   uint64_t address() const { return address_; }
 
-  int Call(ThreadState* thread_state);
+  int Call(ThreadState* thread_state, uint64_t return_address);
 
 protected:
-  virtual int CallImpl(ThreadState* thread_state) = 0;
+  virtual int CallImpl(ThreadState* thread_state, uint64_t return_address) = 0;
 
 protected:
   Type        type_;
@@ -57,7 +57,7 @@ public:
   void* arg1() const { return arg1_; }
 
 protected:
-  virtual int CallImpl(ThreadState* thread_state);
+  virtual int CallImpl(ThreadState* thread_state, uint64_t return_address);
 
 protected:
   Handler handler_;

src/xenia/cpu/processor.cc

@@ -117,7 +117,7 @@ int Processor::Execute(XenonThreadState* thread_state, uint64_t address) {
   context->lr = lr;
 
   // Execute the function.
-  fn->Call(thread_state);
+  fn->Call(thread_state, lr);
   return 0;
 }
 

tools/alloy-sandbox/alloy-sandbox.cc

@@ -47,7 +47,7 @@ int alloy_sandbox(int argc, xechar_t** argv) {
 
   Function* fn;
   runtime->ResolveFunction(0x82000000, &fn);
-  fn->Call(thread_state);
+  fn->Call(thread_state, 0xBEBEBEBE);
 
   delete thread_state;