Removing SET_RETURN_ADDRESS - hopefully it'll never be needed again.

src/alloy/backend/ivm/ivm_function.cc

@@ -104,7 +104,7 @@ void IVMFunction::OnBreakpointHit(ThreadState* thread_state, IntCode* i) {
 
 #undef TRACE_SOURCE_OFFSET
 
-int IVMFunction::CallImpl(ThreadState* thread_state, uint64_t return_address) {
+int IVMFunction::CallImpl(ThreadState* thread_state) {
   // Setup register file on stack.
   auto stack = (IVMStack*)thread_state->backend_data();
   auto register_file = (Register*)stack->Alloc(register_count_);
@@ -120,8 +120,6 @@ int IVMFunction::CallImpl(ThreadState* thread_state, uint64_t return_address) {
   ics.did_saturate = 0;
   ics.access_callbacks = thread_state->runtime()->access_callbacks();
   ics.thread_state = thread_state;
-  ics.return_address = return_address;
-  ics.call_return_address = 0;
 
   volatile int* suspend_flag_address = thread_state->suspend_flag_address();
 

src/alloy/backend/ivm/ivm_function.h

@@ -31,8 +31,7 @@ public:
 protected:
   virtual int AddBreakpointImpl(runtime::Breakpoint* breakpoint);
   virtual int RemoveBreakpointImpl(runtime::Breakpoint* breakpoint);
-  virtual int CallImpl(runtime::ThreadState* thread_state,
+  virtual int CallImpl(runtime::ThreadState* thread_state);
-                       uint64_t return_address);
 
 private:
   IntCode* GetIntCodeAtSourceOffset(uint64_t offset);

src/alloy/backend/ivm/ivm_intcode.cc

@@ -580,9 +580,7 @@ uint32_t IntCode_CALL_XX(IntCodeState& ics, const IntCode* i, uint32_t reg) {
   ics.thread_state->runtime()->ResolveFunction(symbol_info->address(), &fn);
   XEASSERTNOTNULL(fn);
   // TODO(benvanik): proper tail call support, somehow.
-  uint64_t return_address =
+  fn->Call(ics.thread_state);
-      (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;
   }
@@ -647,19 +645,12 @@ 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].u32;
 
-  // Check if return address - if so, return.
-  if (target == ics.return_address) {
-    return IA_RETURN;
-  }
-
   // Real call.
   Function* fn = NULL;
   ics.thread_state->runtime()->ResolveFunction(target, &fn);
   XEASSERTNOTNULL(fn);
   // TODO(benvanik): proper tail call support, somehow.
-  uint64_t return_address =
+  fn->Call(ics.thread_state);
-      (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;
   }
@@ -777,14 +768,6 @@ int Translate_RETURN_TRUE(TranslationContext& ctx, Instr* i) {
   return DispatchToC(ctx, i, fns[i->src1.value->type]);
 }
 
-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;
 }
@@ -4028,7 +4011,6 @@ static const TranslateFn dispatch_table[] = {
   Translate_CALL_INDIRECT_TRUE,
   Translate_RETURN,
   Translate_RETURN_TRUE,
-  Translate_SET_RETURN_ADDRESS,
 
   Translate_BRANCH,
   Translate_BRANCH_TRUE,

src/alloy/backend/ivm/ivm_intcode.h

@@ -48,8 +48,6 @@ typedef struct {
   int8_t        did_saturate;
   runtime::RegisterAccessCallbacks* access_callbacks;
   runtime::ThreadState* thread_state;
-  uint64_t      return_address;
-  uint64_t      call_return_address;
 } IntCodeState;
 
 

src/alloy/backend/x64/lowering/lowering_sequences.cc

@@ -11,11 +11,15 @@
 
 #include <alloy/backend/x64/x64_emitter.h>
 #include <alloy/backend/x64/lowering/lowering_table.h>
+#include <alloy/runtime/symbol_info.h>
+#include <alloy/runtime/runtime.h>
+#include <alloy/runtime/thread_state.h>
 
 using namespace alloy;
 using namespace alloy::backend::x64;
 using namespace alloy::backend::x64::lowering;
 using namespace alloy::hir;
+using namespace alloy::runtime;
 
 using namespace Xbyak;
 
@@ -31,6 +35,49 @@ void Dummy() {
   //
 }
 
+// TODO(benvanik): fancy stuff.
+void CallThunk(void* raw_context, uint8_t* membase,
+               FunctionInfo* symbol_info) {
+  // TODO(benvanik): generate this thunk at runtime? or a shim?
+  auto thread_state = *((ThreadState**)raw_context);
+
+  Function* fn = NULL;
+  thread_state->runtime()->ResolveFunction(symbol_info->address(), &fn);
+  XEASSERTNOTNULL(fn);
+  fn->Call(thread_state);
+}
+void IssueCall(X64Emitter& e, FunctionInfo* symbol_info, uint32_t flags) {
+  e.mov(e.r8, (uint64_t)symbol_info);
+  e.mov(e.rax, (uint64_t)CallThunk);
+  if (flags & CALL_TAIL) {
+    e.jmp(e.rax);
+  } else {
+    e.call(e.rax);
+  }
+}
+
+void IndirectCallThunk(void* raw_context, uint8_t* membase,
+                       uint64_t target_address) {
+  // TODO(benvanik): generate this thunk at runtime? or a shim?
+  auto thread_state = *((ThreadState**)raw_context);
+}
+void IssueCallIndirect(X64Emitter& e, Value* target, uint32_t flags) {
+  Reg64 r;
+  e.BeginOp(target, r, 0);
+  if (r != e.r8) {
+    e.mov(e.r8, r);
+  }
+  e.EndOp(r);
+  e.mov(e.rax, (uint64_t)IndirectCallThunk);
+  if (flags & CALL_TAIL) {
+    e.jmp(e.rax);
+  } else {
+    e.sub(e.rsp, 0x20);
+    e.call(e.rax);
+    e.add(e.rsp, 0x20);
+  }
+}
+
 // Sets EFLAGs with zf for the given value.
 void CheckBoolean(X64Emitter& e, Value* v) {
   if (v->IsConstant()) {
@@ -259,9 +306,7 @@ void alloy::backend::x64::lowering::RegisterSequences(LoweringTable* table) {
   // --------------------------------------------------------------------------
 
   table->AddSequence(OPCODE_CALL, [](X64Emitter& e, Instr*& i) {
-    e.mov(e.rax, (uint64_t)Dummy);
+    IssueCall(e, i->src1.symbol_info, i->flags);
-    e.call(e.rax);
-    UNIMPLEMENTED_SEQ();
     i = e.Advance(i);
     return true;
   });
@@ -270,10 +315,7 @@ void alloy::backend::x64::lowering::RegisterSequences(LoweringTable* table) {
     e.inLocalLabel();
     CheckBoolean(e, i->src1.value);
     e.jne(".x", e.T_SHORT);
-    // TODO(benvanik): call
+    IssueCall(e, i->src2.symbol_info, i->flags);
-    e.mov(e.rax, (uint64_t)Dummy);
-    e.call(e.rax);
-    UNIMPLEMENTED_SEQ();
     e.L(".x");
     e.outLocalLabel();
     i = e.Advance(i);
@@ -281,9 +323,7 @@ void alloy::backend::x64::lowering::RegisterSequences(LoweringTable* table) {
   });
 
   table->AddSequence(OPCODE_CALL_INDIRECT, [](X64Emitter& e, Instr*& i) {
-    e.mov(e.rax, (uint64_t)Dummy);
+    IssueCallIndirect(e, i->src1.value, i->flags);
-    e.call(e.rax);
-    UNIMPLEMENTED_SEQ();
     i = e.Advance(i);
     return true;
   });
@@ -292,10 +332,7 @@ void alloy::backend::x64::lowering::RegisterSequences(LoweringTable* table) {
     e.inLocalLabel();
     CheckBoolean(e, i->src1.value);
     e.jne(".x", e.T_SHORT);
-    // TODO(benvanik): call
+    IssueCallIndirect(e, i->src2.value, i->flags);
-    e.mov(e.rax, (uint64_t)Dummy);
-    e.call(e.rax);
-    UNIMPLEMENTED_SEQ();
     e.L(".x");
     e.outLocalLabel();
     i = e.Advance(i);
@@ -319,12 +356,6 @@ void alloy::backend::x64::lowering::RegisterSequences(LoweringTable* table) {
     return true;
   });
 
-  table->AddSequence(OPCODE_SET_RETURN_ADDRESS, [](X64Emitter& e, Instr*& i) {
-    //UNIMPLEMENTED_SEQ();
-    i = e.Advance(i);
-    return true;
-  });
-
   // --------------------------------------------------------------------------
   // Branches
   // --------------------------------------------------------------------------

src/alloy/backend/x64/x64_function.cc

@@ -41,10 +41,8 @@ int X64Function::RemoveBreakpointImpl(Breakpoint* breakpoint) {
   return 0;
 }
 
-int X64Function::CallImpl(ThreadState* thread_state, uint64_t return_address) {
+int X64Function::CallImpl(ThreadState* thread_state) {
-  //typedef void(*call_t)(ThreadState* thread_state, uint64_t return_address);
+  typedef void(*call_t)(void* raw_context, uint8_t* membase);
-  //((call_t)machine_code_)(thread_state, return_address);
+  ((call_t)machine_code_)(thread_state->raw_context(), thread_state->memory()->membase());
-  typedef void(*call_t)(ThreadState* thread_state, uint8_t* membase);
-  ((call_t)machine_code_)(thread_state, thread_state->memory()->membase());
   return 0;
 }

src/alloy/backend/x64/x64_function.h

@@ -30,8 +30,7 @@ public:
 protected:
   virtual int AddBreakpointImpl(runtime::Breakpoint* breakpoint);
   virtual int RemoveBreakpointImpl(runtime::Breakpoint* breakpoint);
-  virtual int CallImpl(runtime::ThreadState* thread_state,
+  virtual int CallImpl(runtime::ThreadState* thread_state);
-                       uint64_t return_address);
 
 private:
   void*   machine_code_;

src/alloy/frontend/context_info.cc

@@ -13,8 +13,9 @@ using namespace alloy;
 using namespace alloy::frontend;
 
 
-ContextInfo::ContextInfo(size_t size) :
+ContextInfo::ContextInfo(size_t size, uintptr_t thread_state_offset) :
-    size_(size) {
+    size_(size),
+    thread_state_offset_(thread_state_offset) {
 }
 
 ContextInfo::~ContextInfo() {

src/alloy/frontend/context_info.h

@@ -19,13 +19,16 @@ namespace frontend {
 
 class ContextInfo {
 public:
-  ContextInfo(size_t size);
+  ContextInfo(size_t size, uintptr_t thread_state_offset);
   ~ContextInfo();
 
   size_t size() const { return size_; }
 
+  uintptr_t thread_state_offset() const { return thread_state_offset_; }
+
 private:
   size_t size_;
+  uintptr_t thread_state_offset_;
 };
 
 

src/alloy/frontend/ppc/ppc_context.h

@@ -64,6 +64,10 @@ typedef union {
 
 #pragma pack(push, 4)
 typedef struct XECACHEALIGN64 PPCContext_s {
+  // Must be stored at 0x0 for now.
+  // TODO(benvanik): find a nice way to describe this to the JIT.
+  runtime::ThreadState* thread_state;
+
   // Most frequently used registers first.
   uint64_t    r[32];              // General purpose registers
   uint64_t    lr;                 // Link register
@@ -194,7 +198,6 @@ typedef struct XECACHEALIGN64 PPCContext_s {
   // current runtime and its data.
   uint8_t*              membase;
   runtime::Runtime*     runtime;
-  runtime::ThreadState* thread_state;
   volatile int          suspend_flag;
 
   void SetRegFromString(const char* name, const char* value);

src/alloy/frontend/ppc/ppc_emit_control.cc

@@ -35,7 +35,6 @@ int InstrEmit_branch(
   // be correct for returns.
   if (lk) {
     Value* return_address = f.LoadConstant(cia + 4);
-    f.SetReturnAddress(return_address);
     f.StoreLR(return_address);
   }
 

src/alloy/frontend/ppc/ppc_frontend.cc

@@ -50,7 +50,9 @@ PPCFrontend::PPCFrontend(Runtime* runtime) :
     Frontend(runtime) {
   InitializeIfNeeded();
 
-  ContextInfo* info = new ContextInfo(sizeof(PPCContext));
+  ContextInfo* info = new ContextInfo(
+      sizeof(PPCContext),
+      offsetof(PPCContext, thread_state));
   // Add fields/etc.
   context_info_ = info;
 }

src/alloy/hir/hir_builder.cc

@@ -578,12 +578,6 @@ void HIRBuilder::ReturnTrue(Value* cond) {
   EndBlock();
 }
 
-void HIRBuilder::SetReturnAddress(Value* value) {
-  Instr* i = AppendInstr(OPCODE_SET_RETURN_ADDRESS_info, 0);
-  i->set_src1(value);
-  i->src2.value = i->src3.value = NULL;
-}
-
 void HIRBuilder::Branch(Label* label, uint32_t branch_flags) {
   Instr* i = AppendInstr(OPCODE_BRANCH_info, branch_flags);
   i->src1.label = label;

src/alloy/hir/hir_builder.h

@@ -76,7 +76,6 @@ public:
   void CallIndirectTrue(Value* cond, Value* value, uint32_t call_flags = 0);
   void Return();
   void ReturnTrue(Value* cond);
-  void SetReturnAddress(Value* value);
 
   void Branch(Label* label, uint32_t branch_flags = 0);
   void Branch(Block* block, uint32_t branch_flags = 0);

src/alloy/hir/opcodes.h

@@ -96,7 +96,6 @@ enum Opcode {
   OPCODE_CALL_INDIRECT_TRUE,
   OPCODE_RETURN,
   OPCODE_RETURN_TRUE,
-  OPCODE_SET_RETURN_ADDRESS,
 
   OPCODE_BRANCH,
   OPCODE_BRANCH_TRUE,

src/alloy/hir/opcodes.inl

@@ -86,12 +86,6 @@ DEFINE_OPCODE(
     OPCODE_SIG_X_V,
     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

@@ -72,12 +72,12 @@ Breakpoint* Function::FindBreakpoint(uint64_t address) {
   return result;
 }
 
-int Function::Call(ThreadState* thread_state, uint64_t return_address) {
+int Function::Call(ThreadState* thread_state) {
   ThreadState* original_thread_state = ThreadState::Get();
   if (original_thread_state != thread_state) {
     ThreadState::Bind(thread_state);
   }
-  int result = CallImpl(thread_state, return_address);
+  int result = CallImpl(thread_state);
   if (original_thread_state != thread_state) {
     ThreadState::Bind(original_thread_state);
   }
@@ -101,8 +101,7 @@ void ExternFunction::set_name(const char* name) {
   name_ = xestrdupa(name);
 }
 
-int ExternFunction::CallImpl(ThreadState* thread_state,
+int ExternFunction::CallImpl(ThreadState* thread_state) {
-                             uint64_t return_address) {
   if (!handler_) {
     XELOGW("undefined extern call to %.8X %s", address(), name());
     return 0;

src/alloy/runtime/function.h

@@ -42,13 +42,13 @@ public:
   int AddBreakpoint(Breakpoint* breakpoint);
   int RemoveBreakpoint(Breakpoint* breakpoint);
 
-  int Call(ThreadState* thread_state, uint64_t return_address);
+  int Call(ThreadState* thread_state);
 
 protected:
   Breakpoint* FindBreakpoint(uint64_t address);
   virtual int AddBreakpointImpl(Breakpoint* breakpoint) { return 0; }
   virtual int RemoveBreakpointImpl(Breakpoint* breakpoint) { return 0; }
-  virtual int CallImpl(ThreadState* thread_state, uint64_t return_address) = 0;
+  virtual int CallImpl(ThreadState* thread_state) = 0;
 
 protected:
   Type        type_;
@@ -76,7 +76,7 @@ public:
   void* arg1() const { return arg1_; }
 
 protected:
-  virtual int CallImpl(ThreadState* thread_state, uint64_t return_address);
+  virtual int CallImpl(ThreadState* thread_state);
 
 protected:
   char* name_;

src/xenia/cpu/processor.cc

@@ -165,7 +165,7 @@ int Processor::Execute(XenonThreadState* thread_state, uint64_t address) {
   context->lr = lr;
 
   // Execute the function.
-  fn->Call(thread_state, lr);
+  fn->Call(thread_state);
   return 0;
 }
 

tools/alloy-sandbox/alloy-sandbox.cc

@@ -49,7 +49,7 @@ int alloy_sandbox(int argc, xechar_t** argv) {
   ctx->lr = 0xBEBEBEBE;
   ctx->r[5] = 10;
   ctx->r[25] = 25;
-  fn->Call(thread_state, 0xBEBEBEBE);
+  fn->Call(thread_state);
   auto result = ctx->r[11];
 
   delete thread_state;