Fixing tail calls.

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

@@ -10,6 +10,7 @@
 #include <alloy/backend/x64/lowering/lowering_sequences.h>
 
 #include <alloy/backend/x64/x64_emitter.h>
+#include <alloy/backend/x64/x64_function.h>
 #include <alloy/backend/x64/lowering/lowering_table.h>
 #include <alloy/runtime/symbol_info.h>
 #include <alloy/runtime/runtime.h>
@@ -35,7 +36,7 @@ void Dummy() {
   //
 }
 
-void PrintString(void* raw_context, uint8_t* membase, const char* str) {
+void PrintString(void* raw_context, const char* str) {
   // TODO(benvanik): generate this thunk at runtime? or a shim?
   auto thread_state = *((ThreadState**)raw_context);
   fprintf(stdout, "XE[t] :%d: %s\n", thread_state->GetThreadID(), str);
@@ -43,48 +44,74 @@ void PrintString(void* raw_context, uint8_t* membase, const char* str) {
 }
 
 // TODO(benvanik): fancy stuff.
-void CallThunk(void* raw_context, uint8_t* membase,
-               FunctionInfo* symbol_info) {
+void* ResolveFunctionSymbol(void* raw_context, 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);
+  XEASSERT(fn->type() == Function::USER_FUNCTION);
+  auto x64_fn = (X64Function*)fn;
+  return x64_fn->machine_code();
+}
+void* ResolveFunctionAddress(void* raw_context, uint64_t target_address) {
+  // TODO(benvanik): generate this thunk at runtime? or a shim?
+  auto thread_state = *((ThreadState**)raw_context);
+
+  Function* fn = NULL;
+  thread_state->runtime()->ResolveFunction(target_address, &fn);
+  XEASSERTNOTNULL(fn);
+  XEASSERTALWAYS();
+  //fn->Call(thread_state);
+  return 0;
 }
 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 we are an extern function, we can directly insert a call.
+  auto fn = symbol_info->function();
+  if (fn && fn->type() == Function::EXTERN_FUNCTION) {
+    auto extern_fn = (ExternFunction*)fn;
+    e.mov(e.rdx, (uint64_t)extern_fn->arg0());
+    e.mov(e.r8, (uint64_t)extern_fn->arg1());
+    e.mov(e.rax, (uint64_t)extern_fn->handler());
+  } else {
+    // Generic call, resolve address.
+    // TODO(benvanik): caching/etc. For now this makes debugging easier.
+    e.mov(e.rdx, (uint64_t)symbol_info);
+    e.mov(e.rax, (uint64_t)ResolveFunctionSymbol);
+    e.call(e.rax);
+    e.mov(e.rcx, e.qword[e.rsp + 0]);
+    e.mov(e.rdx, e.qword[e.rcx + 8]); // membase
+  }
   if (flags & CALL_TAIL) {
+    // TODO(benvanik): adjust stack?
+    e.add(e.rsp, 0x40);
     e.jmp(e.rax);
   } else {
     e.call(e.rax);
-    e.mov(e.rdx, e.qword[e.rsp + 8]);
     e.mov(e.rcx, e.qword[e.rsp + 0]);
+    e.mov(e.rdx, e.qword[e.rcx + 8]); // membase
   }
 }
-
-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);
-  XEASSERTALWAYS();
-}
 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);
+  if (r != e.rdx) {
+    e.mov(e.rdx, r);
   }
   e.EndOp(r);
-  e.mov(e.rax, (uint64_t)IndirectCallThunk);
+  e.mov(e.rax, (uint64_t)ResolveFunctionAddress);
+  e.call(e.rax);
+  e.mov(e.rcx, e.qword[e.rsp + 0]);
+  e.mov(e.rdx, e.qword[e.rcx + 8]); // membase
   if (flags & CALL_TAIL) {
+    // TODO(benvanik): adjust stack?
+    e.add(e.rsp, 0x40);
     e.jmp(e.rax);
   } else {
-    e.sub(e.rsp, 0x20);
     e.call(e.rax);
-    e.add(e.rsp, 0x20);
+    e.mov(e.rcx, e.qword[e.rsp + 0]);
+    e.mov(e.rdx, e.qword[e.rcx + 8]); // membase
   }
 }
 
@@ -514,11 +541,11 @@ void alloy::backend::x64::lowering::RegisterSequences(LoweringTable* table) {
     // TODO(benvanik): pass through.
     auto str = (const char*)i->src1.offset;
     auto str_copy = xestrdupa(str);
-    e.mov(e.r8, (uint64_t)str_copy);
+    e.mov(e.rdx, (uint64_t)str_copy);
     e.mov(e.rax, (uint64_t)PrintString);
     e.call(e.rax);
-    e.mov(e.rdx, e.qword[e.rsp + 8]);
     e.mov(e.rcx, e.qword[e.rsp + 0]);
+    e.mov(e.rdx, e.qword[e.rcx + 8]); // membase
     i = e.Advance(i);
     return true;
   });

src/alloy/backend/x64/x64_emitter.cc

@@ -128,7 +128,6 @@ int X64Emitter::Emit(HIRBuilder* builder) {
   const bool emit_prolog = true;
   const size_t stack_size = 64;
   if (emit_prolog) {
-    mov(qword[rsp + 16], rdx);
     mov(qword[rsp + 8], rcx);
     sub(rsp, stack_size);
     mov(qword[rsp + 8 * 0], rbx);
@@ -138,6 +137,10 @@ int X64Emitter::Emit(HIRBuilder* builder) {
     mov(qword[rsp + 8 * 4], r15);
   }
 
+  // membase stays in rdx. If we evict it (like on function calls) we
+  // must put it back.
+  mov(rdx, qword[rcx + 8]);
+
   auto lowering_table = backend_->lowering_table();
 
   // Body.

src/alloy/backend/x64/x64_function.h

@@ -25,6 +25,9 @@ public:
   X64Function(runtime::FunctionInfo* symbol_info);
   virtual ~X64Function();
 
+  void* machine_code() const { return machine_code_; }
+  size_t code_size() const { return code_size_; }
+
   void Setup(void* machine_code, size_t code_size);
 
 protected:

src/alloy/frontend/ppc/ppc_context.h

@@ -67,6 +67,8 @@ 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;
+  // TODO(benvanik): this is getting nasty. Must be here.
+  uint8_t*              membase;
 
   // Most frequently used registers first.
   uint64_t    r[32];              // General purpose registers
@@ -196,7 +198,6 @@ typedef struct XECACHEALIGN64 PPCContext_s {
 
   // Runtime-specific data pointer. Used on callbacks to get access to the
   // current runtime and its data.
-  uint8_t*              membase;
   runtime::Runtime*     runtime;
   volatile int          suspend_flag;