vm_invoke_builtin_delegate with start index.

opt_invokebuiltin_delegate and opt_invokebuiltin_delegate_leave
invokes builtin functions with same parameters of the method.
This technique eliminate stack push operations. However, delegation
parameters should be completely same as given parameters.
(e.g. `def foo(a, b, c) __builtin_foo(a, b, c)` is okay, but
__builtin_foo(b, c) is not allowed)

This patch relaxes this restriction. ISeq has a local variables
table which includes parameters. For example, the method defined
as `def foo(a, b, c) x=y=nil`, then local variables table contains
[a, b, c, x, y]. If calling builtin-function with arguments which
are sub-array of the lvar table, use opt_invokebuiltin_delegate
instruction with start index. For example, `__builtin_foo(b, c)`,
`__builtin_bar(c, x, y)` is okay, and so on.

compile.c

@@ -6762,39 +6762,65 @@ iseq_builtin_function_name(ID mid)
 }
 
 static int
-delegate_call_p(const rb_iseq_t *iseq, unsigned int argc, const LINK_ANCHOR *args)
+delegate_call_p(const rb_iseq_t *iseq, unsigned int argc, const LINK_ANCHOR *args, unsigned int *pstart_index)
 {
+
     if (argc == 0) {
+        *pstart_index = 0;
         return TRUE;
     }
-    else if (argc == iseq->body->param.size) {
+    else if (argc <= iseq->body->local_table_size) {
+        unsigned int start=0;
+
+        // local_table: [p1, p2, p3, l1, l2, l3]
+        // arguments:           [p3, l1, l2]     -> 2
+        for (start = 0;
+             argc + start <= iseq->body->local_table_size;
+             start++) {
             const LINK_ELEMENT *elem = FIRST_ELEMENT(args);
 
-        for (unsigned int i=0; i<argc; i++) {
+            for (unsigned int i=start; i-start<argc; i++) {
                 if (elem->type == ISEQ_ELEMENT_INSN &&
                     INSN_OF(elem) == BIN(getlocal)) {
                     int local_index = FIX2INT(OPERAND_AT(elem, 0));
                     int local_level = FIX2INT(OPERAND_AT(elem, 1));
+
                     if (local_level == 0) {
                         unsigned int index = iseq->body->local_table_size - (local_index - VM_ENV_DATA_SIZE + 1);
-#if 0
-                    ID param_id = iseq->body->local_table[i];
-                    fprintf(stderr, "param_id:%s (%d), id:%s (%d) local_index:%d, local_size:%d\n",
-                            rb_id2name(param_id), i,
+                        if (0) { // for debug
+                            fprintf(stderr, "lvar:%s (%d), id:%s (%d) local_index:%d, local_size:%d\n",
+                                    rb_id2name(iseq->body->local_table[i]),     i,
                                     rb_id2name(iseq->body->local_table[index]), index,
                                     local_index, (int)iseq->body->local_table_size);
-#endif
+                        }
                         if (i == index) {
                             elem = elem->next;
                             continue; /* for */
                         }
+                        else {
+                            goto next;
                         }
                     }
+                    else {
+                        goto fail; // level != 0 is unsupport
+                    }
+                }
+                else {
+                    goto fail; // insn is not a getlocal
+                }
+            }
+            goto success;
+          next:;
+        }
+      fail:
         return FALSE;
-        }
+      success:
+        *pstart_index = start;
         return TRUE;
     }
+    else {
         return FALSE;
+    }
 }
 
 static int
@@ -6924,8 +6950,9 @@ compile_call(rb_iseq_t *iseq, LINK_ANCHOR *const ret, const NODE *const node, in
                 return COMPILE_NG;
             }
 
-            if (delegate_call_p(iseq, FIX2INT(argc), args)) {
-                ADD_INSN1(ret, line, opt_invokebuiltin_delegate, bf);
+            unsigned int start_index;
+            if (delegate_call_p(iseq, FIX2INT(argc), args, &start_index)) {
+                ADD_INSN2(ret, line, opt_invokebuiltin_delegate, bf, INT2FIX(start_index));
             }
             else {
                 ADD_SEQ(ret, args);

insns.def

@@ -1492,26 +1492,26 @@ invokebuiltin
 /* call specific function with args (same parameters) */
 DEFINE_INSN
 opt_invokebuiltin_delegate
-(RB_BUILTIN bf)
+(RB_BUILTIN bf, rb_num_t index)
 ()
 (VALUE ret)
 // attr bool leaf = false; /* anything can happen inside */
 {
-    ret = vm_invoke_builtin_delegate(ec, reg_cfp, bf);
+    ret = vm_invoke_builtin_delegate(ec, reg_cfp, bf, index);
 }
 
 /* call specific function with args (same parameters) and leave */
 DEFINE_INSN
 opt_invokebuiltin_delegate_leave
-(RB_BUILTIN bf)
+(RB_BUILTIN bf, rb_num_t index)
 ()
 (VALUE val)
 // attr bool leaf = false; /* anything can happen inside */
 {
-    val = vm_invoke_builtin_delegate(ec, reg_cfp, bf);
+    val = vm_invoke_builtin_delegate(ec, reg_cfp, bf, index);
 
     /* leave fastpath */
-    /* TracePoint/return should fallback this insn to invokecfuncwparam */
+    /* TracePoint/return should fallback this insn to opt_invokebuiltin_delegate */
     if (vm_pop_frame(ec, GET_CFP(), GET_EP())) {
 #if OPT_CALL_THREADED_CODE
         rb_ec_thread_ptr(ec)->retval = val;

mini_builtin.c

@@ -21,6 +21,11 @@ rb_load_with_builtin_functions(const char *feature_name, const struct rb_builtin
 
     rb_ast_dispose(ast);
 
+    // for debug
+    if (0 && strcmp("prelude", feature_name) == 0) {
+        rb_io_write(rb_stdout, rb_iseq_disasm((const rb_iseq_t *)iseq));
+    }
+
     // register (loaded iseq will not be freed)
     st_insert(loaded_builtin_table, (st_data_t)feature_name, (st_data_t)iseq);
     rb_gc_register_mark_object((VALUE)iseq);

vm_insnhelper.c

@@ -4989,10 +4989,18 @@ vm_invoke_builtin(rb_execution_context_t *ec, rb_control_frame_t *cfp, const str
 }
 
 static VALUE
-vm_invoke_builtin_delegate(rb_execution_context_t *ec, rb_control_frame_t *cfp, const struct rb_builtin_function *bf)
+vm_invoke_builtin_delegate(rb_execution_context_t *ec, rb_control_frame_t *cfp, const struct rb_builtin_function *bf, int start_index)
 {
-    const VALUE *argv = cfp->ep - cfp->iseq->body->local_table_size - VM_ENV_DATA_SIZE + 1;
-    // fprintf(stderr, "%s %s(%d):%p\n", __func__, bf->name, bf->argc, bf->func_ptr);
+    if (0) { // debug print
+        fprintf(stderr, "vm_invoke_builtin_delegate: passing -> ");
+        for (int i=0; i<bf->argc; i++) {
+            fprintf(stderr, ":%s ", rb_id2name(cfp->iseq->body->local_table[i+start_index]));
+        }
+        fprintf(stderr, "\n");
+        fprintf(stderr, "%s %s(%d):%p\n", __func__, bf->name, bf->argc, bf->func_ptr);
+    }
+
+    const VALUE *argv = cfp->ep - cfp->iseq->body->local_table_size - VM_ENV_DATA_SIZE + 1 + start_index;
     return invoke_bf(ec, cfp, bf, argv);
 }