use inline cache for refinements

From Ruby 3.0, refined method invocations are slow because
resolved methods are not cached by inline cache because of
conservertive strategy. However, `using` clears all caches
so that it seems safe to cache resolved method entries.

This patch caches resolved method entries in inline cache
and clear all of inline method caches when `using` is called.

fix [Bug #18572]

```ruby
 # without refinements

class C
  def foo = :C
end

N = 1_000_000

obj = C.new
require 'benchmark'
Benchmark.bm{|x|
  x.report{N.times{
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
  }}
}

_END__
              user     system      total        real
master    0.362859   0.002544   0.365403 (  0.365424)
modified  0.357251   0.000000   0.357251 (  0.357258)
```

```ruby
 # with refinment but without using

class C
  def foo = :C
end

module R
  refine C do
    def foo = :R
  end
end

N = 1_000_000

obj = C.new
require 'benchmark'
Benchmark.bm{|x|
  x.report{N.times{
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
  }}
}
__END__
               user     system      total        real
master     0.957182   0.000000   0.957182 (  0.957212)
modified   0.359228   0.000000   0.359228 (  0.359238)
```

```ruby
 # with using

class C
  def foo = :C
end

module R
  refine C do
    def foo = :R
  end
end

N = 1_000_000

using R

obj = C.new
require 'benchmark'
Benchmark.bm{|x|
  x.report{N.times{
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
    obj.foo; obj.foo; obj.foo; obj.foo; obj.foo;
  }}
}

eval.c

@@ -1341,7 +1341,7 @@ rb_using_module(const rb_cref_t *cref, VALUE module)
 {
     Check_Type(module, T_MODULE);
     using_module_recursive(cref, module);
-    rb_clear_method_cache_all();
+    rb_clear_all_refinement_method_cache();
 }
 
 /*

gc.c

@@ -3187,6 +3187,8 @@ vm_ccs_free(struct rb_class_cc_entries *ccs, int alive, rb_objspace_t *objspace,
                     asan_poison_object((VALUE)cc);
                 }
             }
+
+            VM_ASSERT(!vm_cc_super_p(cc) && !vm_cc_refinement_p(cc));
             vm_cc_invalidate(cc);
         }
         ruby_xfree(ccs->entries);

method.h

@@ -249,6 +249,6 @@ void rb_scope_visibility_set(rb_method_visibility_t);
 VALUE rb_unnamed_parameters(int arity);
 
 void rb_clear_method_cache(VALUE klass_or_module, ID mid);
-void rb_clear_method_cache_all(void);
+void rb_clear_all_refinement_method_cache(void);
 
 #endif /* RUBY_METHOD_H */

test/ruby/test_refinement.rb

@@ -2626,6 +2626,24 @@ class TestRefinement < Test::Unit::TestCase
     assert_equal([], Refinement.used_modules)
   end
 
+  def test_inlinecache
+    assert_separately([], <<-"end;")
+      module R
+        refine String do
+          def to_s = :R
+        end
+      end
+
+      2.times{|i|
+        s = ''.to_s
+        assert_equal '', s if i == 0
+        assert_equal :R, s if i == 1
+        using R            if i == 0
+        assert_equal :R, ''.to_s
+      }
+    end;
+  end
+
   private
 
   def eval_using(mod, s)

vm_callinfo.h

@@ -299,10 +299,12 @@ struct rb_callcache {
 #define VM_CALLCACHE_IVAR       IMEMO_FL_USER0
 #define VM_CALLCACHE_BF         IMEMO_FL_USER1
 #define VM_CALLCACHE_SUPER      IMEMO_FL_USER2
+#define VM_CALLCACHE_REFINEMENT IMEMO_FL_USER3
 
 enum vm_cc_type {
     cc_type_normal, // chained from ccs
     cc_type_super,
+    cc_type_refinement,
 };
 
 extern const struct rb_callcache *rb_vm_empty_cc(void);
@@ -332,6 +334,9 @@ vm_cc_new(VALUE klass,
       case cc_type_super:
         *(VALUE *)&cc->flags |= VM_CALLCACHE_SUPER;
         break;
+      case cc_type_refinement:
+        *(VALUE *)&cc->flags |= VM_CALLCACHE_REFINEMENT;
+        break;
     }
 
     vm_cc_attr_index_initialize(cc, INVALID_SHAPE_ID);
@@ -345,6 +350,12 @@ vm_cc_super_p(const struct rb_callcache *cc)
     return (cc->flags & VM_CALLCACHE_SUPER) != 0;
 }
 
+static inline bool
+vm_cc_refinement_p(const struct rb_callcache *cc)
+{
+    return (cc->flags & VM_CALLCACHE_REFINEMENT) != 0;
+}
+
 #define VM_CC_ON_STACK(clazz, call, aux, cme) \
     (struct rb_callcache) {                   \
         .flags = T_IMEMO |                    \

vm_insnhelper.c

@@ -2003,6 +2003,8 @@ vm_ccs_verify(struct rb_class_cc_entries *ccs, ID mid, VALUE klass)
         VM_ASSERT(IMEMO_TYPE_P(cc, imemo_callcache));
         VM_ASSERT(vm_cc_class_check(cc, klass));
         VM_ASSERT(vm_cc_check_cme(cc, ccs->cme));
+        VM_ASSERT(!vm_cc_super_p(cc));
+        VM_ASSERT(!vm_cc_refinement_p(cc));
     }
     return TRUE;
 }
@@ -4193,13 +4195,20 @@ search_refined_method(rb_execution_context_t *ec, rb_control_frame_t *cfp, struc
 static VALUE
 vm_call_refined(rb_execution_context_t *ec, rb_control_frame_t *cfp, struct rb_calling_info *calling)
 {
-    struct rb_callcache *ref_cc =  &VM_CC_ON_STACK(Qundef, vm_call_general, {{ 0 }},
+    const rb_callable_method_entry_t *ref_cme = search_refined_method(ec, cfp, calling);
-                                                   search_refined_method(ec, cfp, calling));
 
-    if (vm_cc_cme(ref_cc)) {
+    if (ref_cme) {
+        if (calling->cd->cc) {
+            const struct rb_callcache *cc = calling->cc = vm_cc_new(vm_cc_cme(calling->cc)->defined_class, ref_cme, vm_call_general, cc_type_refinement);
+            RB_OBJ_WRITE(cfp->iseq, &calling->cd->cc, cc);
+            return vm_call_method(ec, cfp, calling);
+        }
+        else {
+            struct rb_callcache *ref_cc =  &VM_CC_ON_STACK(Qundef, vm_call_general, {{ 0 }}, ref_cme);
             calling->cc= ref_cc;
             return vm_call_method(ec, cfp, calling);
         }
+    }
     else {
         return vm_call_method_nome(ec, cfp, calling);
     }

vm_method.c

@@ -307,19 +307,19 @@ rb_clear_method_cache(VALUE klass_or_module, ID mid)
 void rb_cc_table_free(VALUE klass);
 
 static int
-invalidate_all_cc(void *vstart, void *vend, size_t stride, void *data)
+invalidate_all_refinement_cc(void *vstart, void *vend, size_t stride, void *data)
 {
     VALUE v = (VALUE)vstart;
     for (; v != (VALUE)vend; v += stride) {
         void *ptr = asan_poisoned_object_p(v);
         asan_unpoison_object(v, false);
+
         if (RBASIC(v)->flags) { // liveness check
-            if (RB_TYPE_P(v, T_CLASS) ||
+            if (imemo_type_p(v, imemo_callcache)) {
-                RB_TYPE_P(v, T_ICLASS)) {
+                const struct rb_callcache *cc = (const struct rb_callcache *)v;
-                if (RCLASS_CC_TBL(v)) {
+                if (vm_cc_refinement_p(cc) && cc->klass) {
-                    rb_cc_table_free(v);
+                    vm_cc_invalidate(cc);
                 }
-                RCLASS_CC_TBL(v) = NULL;
             }
         }
 
@@ -331,10 +331,9 @@ invalidate_all_cc(void *vstart, void *vend, size_t stride, void *data)
 }
 
 void
-rb_clear_method_cache_all(void)
+rb_clear_all_refinement_method_cache(void)
 {
-    rb_objspace_each_objects(invalidate_all_cc, NULL);
+    rb_objspace_each_objects(invalidate_all_refinement_cc, NULL);
-
     rb_yjit_invalidate_all_method_lookup_assumptions();
 }