Move ret pointer before closed vars in proc calls

The previous call convention we had the following before the function
call:

[ PROC_ADDR | CLOSED_VARS | ARGS | RP ]

This was made to be compatible with the `call` instruction and the
`cdecl` calling convention. This commit changes the structure to:

[ RP | PROC_ADDR | CLOSED_VARS | ARGS ]

With this convention, the stack frame can be discarded in its entirety
if we have a tail optimized procedure.

1 files changed, 14 insertions, 6 deletions

diff --git a/src/compiler.h b/src/compiler.h
index d1952ed..1ef81e3 100644
--- a/src/compiler.h
+++ b/src/compiler.h
@@ -343,6 +343,13 @@ compile_cdr(Object *obj) {
 void
 compile_call(Object *obj) {
     context_printf("    ;; --> compile_call\n");
+
+    // Prepare return pointer.
+    char *lab_end = generate_label("BDLL");
+    context_printf("    lea     rcx, [%s]\n", lab_end);
+    context_printf("    push    rcx\n");
+
+    // Compile operator.
     compile_object(obj->head);
     context_printf("    pop     rax\n");
     context_printf("    mov     rcx, PTR_MASK\n");
@@ -379,10 +386,7 @@ compile_call(Object *obj) {
         compile_object(obj->head);
     }
 
-    char *lab_end = generate_label("BDLL");
     context_printf("    mov     rax, [rsp + %zu]\n", 8 * offset);
-    context_printf("    lea     rcx, [%s]\n", lab_end);
-    context_printf("    push    rcx\n");
     context_printf("    jmp     rax\n");
     context_printf("%s:\n", lab_end);
 
@@ -535,7 +539,11 @@ compile_lambda(Object *obj) {
     context_printf("    pop     rax\n");
 
     // Restore the previous call frame.
-    context_printf("    mov     rdi, [rbp + 8]\n");
+    size_t n_locals = array_size(current_env->locals);
+    size_t n_params = array_size(current_env->params);
+    size_t n_captured = array_size(current_env->captured);
+    size_t offset = 8 * (n_locals + n_params + n_captured + 2);
+    context_printf("    mov     rdi, [rbp + %zu]\n", offset);
     context_printf("    mov     rsp, rbp\n");
     context_printf("    add     rsp, %zu\n", 8 * array_size(current_env->locals));
     context_printf("    pop     rbp\n");
@@ -602,7 +610,7 @@ compile_symbol(Object *obj) {
         size_t n_locals = array_size(current_env->locals);
         size_t n_params = array_size(current_env->params);
         size_t n_cap = array_size(current_env->captured);
-        size_t offset = 8 * (n_locals + n_params + n_cap - idx + 1);
+        size_t offset = 8 * (n_locals + n_params + n_cap - idx);
         context_printf("    mov     rcx, [rbp + %ld]\n", offset);
         context_printf("    mov     rax, [rcx]\n");
         context_printf("    push    rax\n");
@@ -624,7 +632,7 @@ compile_symbol(Object *obj) {
     if (idx != -1) {
         size_t n_locals = array_size(current_env->locals);
         size_t n_params = array_size(current_env->params);
-        size_t offset = 8 * (n_locals + n_params - idx + 1);
+        size_t offset = 8 * (n_locals + n_params - idx);
         context_printf("    mov     rax, [rbp + %ld]\n", offset);
         context_printf("    push    rax\n");
         context_printf("    ;; <-- compile_symbol\n");