* vm_insnhelper.c: introduce new call handler for simple ISeqs.

  vm_call_iseq_setup_normal_0start() is simple, however it has
  some loops/conditions depends on ISeq::param.size and
  ISeq::local_size (in vm_push_frame(), inlined into this function).

  There are many simple methods which has a few parameters and local
  variables. So that this patch introduces several special functions
  generated in vm_call_iseq_optimized.inc by
  tool/mk_call_iseq_optimized.rb.

  This script makes
    vm_call_iseq_setup_normal_0start_Xparams_Ylocals()
  where X is 0 to 3 and Y is 1 to 6 (as current setting).
  In this case, X * Y = 24 functions are created.

  These functions creates fast method dispatch by inlining
  vm_push_frame() with immediate params/locals sizes.

  On my laptop, we can have the following results.

  vm2_method*       1.083 (8.3% faster)
  vm2_poly_method*  0.961 (3.4% slower)

  It shows 8.3% faster for inner loop method dispatch (hit inline
  cache), but 3.4% slower when inline cache miss because we need
  to find a suitable call handler.

* common.mk: add a rule for vm_call_iseq_optimized.inc.

* tool/mk_call_iseq_optimized.rb: added.

* vm.c: include vm_call_iseq_optimized.inc.




git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@52254 b2dd03c8-39d4-4d8f-98ff-823fe69b080e

1 files changed, 46 insertions, 38 deletions

diff --git a/vm_insnhelper.c b/vm_insnhelper.c
index ae636a964e..cec303767f 100644
--- a/vm_insnhelper.c
+++ b/vm_insnhelper.c
@@ -1235,17 +1235,17 @@ vm_base_ptr(rb_control_frame_t *cfp)
 
 #include "vm_args.c"
 
-static inline VALUE vm_call_iseq_setup_2(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc, int opt_pc);
-static inline VALUE vm_call_iseq_setup_normal(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc, int opt_pc);
+static inline VALUE vm_call_iseq_setup_2(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc, int opt_pc, int param_size, int local_size);
+static inline VALUE vm_call_iseq_setup_normal(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc, int opt_pc, int param_size, int local_size);
 static inline VALUE vm_call_iseq_setup_tailcall(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc, int opt_pc);
-static VALUE vm_call_iseq_setup_normal_0start(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc);
-static VALUE vm_call_iseq_setup_tailcall_0start(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc);
-
 static VALUE vm_call_super_method(rb_thread_t *th, rb_control_frame_t *reg_cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc);
 static VALUE vm_call_method_nome(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc);
 static VALUE vm_call_method_each_type(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc);
 static inline VALUE vm_call_method(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc);
 
+typedef VALUE (*vm_call_handler)(struct rb_thread_struct *th, struct rb_control_frame_struct *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc);
+static vm_call_handler vm_call_iseq_setup_func(const struct rb_call_info *ci, const int param_size, const int local_size);
+
 static rb_method_definition_t *method_definition_create(rb_method_type_t type, ID mid);
 static void method_definition_set(const rb_method_entry_t *me, rb_method_definition_t *def, void *opts);
 static int rb_method_definition_eq(const rb_method_definition_t *d1, const rb_method_definition_t *d2);
@@ -1332,8 +1332,33 @@ vm_callee_setup_block_arg(rb_thread_t *th, struct rb_calling_info *calling, cons
     }
 }
 
+static const rb_iseq_t *
+def_iseq_ptr(rb_method_definition_t *def)
+{
+#if VM_CHECK_MODE > 0
+    if (def->type != VM_METHOD_TYPE_ISEQ) rb_bug("def_iseq_ptr: not iseq (%d)", def->type);
+#endif
+    return def->body.iseq.iseqptr;
+}
+
+static VALUE
+vm_call_iseq_setup_tailcall_0start(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc)
+{
+    return vm_call_iseq_setup_tailcall(th, cfp, calling, ci, cc, 0);
+}
+
+static VALUE
+vm_call_iseq_setup_normal_0start(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc)
+{
+    const rb_iseq_t *iseq = def_iseq_ptr(cc->me->def);
+    int param = iseq->body->param.size;
+    int local = iseq->body->local_size;
+    return vm_call_iseq_setup_normal(th, cfp, calling, ci, cc, 0, param, local);
+}
+
 static inline int
-vm_callee_setup_arg(rb_thread_t *th, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc, const rb_iseq_t *iseq, VALUE *argv)
+vm_callee_setup_arg(rb_thread_t *th, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc,
+		    const rb_iseq_t *iseq, VALUE *argv, int param_size, int local_size)
 {
     if (LIKELY(simple_iseq_p(iseq))) {
 	rb_control_frame_t *cfp = th->cfp;
@@ -1344,9 +1369,7 @@ vm_callee_setup_arg(rb_thread_t *th, struct rb_calling_info *calling, const stru
 	    argument_arity_error(th, iseq, calling->argc, iseq->body->param.lead_num, iseq->body->param.lead_num);
 	}
 
-	CI_SET_FASTPATH(cc,
-			(UNLIKELY(ci->flag & VM_CALL_TAILCALL) ? vm_call_iseq_setup_tailcall_0start :
-			                                         vm_call_iseq_setup_normal_0start),
+	CI_SET_FASTPATH(cc, vm_call_iseq_setup_func(ci, param_size, local_size),
 			(!IS_ARGS_SPLAT(ci) && !IS_ARGS_KEYWORD(ci) &&
 			 !(METHOD_ENTRY_VISI(cc->me) == METHOD_VISI_PROTECTED)));
 	return 0;
@@ -1356,27 +1379,22 @@ vm_callee_setup_arg(rb_thread_t *th, struct rb_calling_info *calling, const stru
     }
 }
 
-static const rb_iseq_t *
-def_iseq_ptr(rb_method_definition_t *def)
-{
-#if VM_CHECK_MODE > 0
-    if (def->type != VM_METHOD_TYPE_ISEQ) rb_bug("def_iseq_ptr: not iseq (%d)", def->type);
-#endif
-    return def->body.iseq.iseqptr;
-}
-
 static VALUE
 vm_call_iseq_setup(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc)
 {
-    int opt_pc = vm_callee_setup_arg(th, calling, ci, cc, def_iseq_ptr(cc->me->def), cfp->sp - calling->argc);
-    return vm_call_iseq_setup_2(th, cfp, calling, ci, cc, opt_pc);
+    const rb_iseq_t *iseq = def_iseq_ptr(cc->me->def);
+    const int param_size = iseq->body->param.size;
+    const int local_size = iseq->body->local_size;
+    const int opt_pc = vm_callee_setup_arg(th, calling, ci, cc, def_iseq_ptr(cc->me->def), cfp->sp - calling->argc, param_size, local_size);
+    return vm_call_iseq_setup_2(th, cfp, calling, ci, cc, opt_pc, param_size, local_size);
 }
 
 static inline VALUE
-vm_call_iseq_setup_2(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc, int opt_pc)
+vm_call_iseq_setup_2(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc,
+		     int opt_pc, int param_size, int local_size)
 {
     if (LIKELY(!(ci->flag & VM_CALL_TAILCALL))) {
-	return vm_call_iseq_setup_normal(th, cfp, calling, ci, cc, opt_pc);
+	return vm_call_iseq_setup_normal(th, cfp, calling, ci, cc, opt_pc, param_size, local_size);
     }
     else {
 	return vm_call_iseq_setup_tailcall(th, cfp, calling, ci, cc, opt_pc);
@@ -1384,24 +1402,26 @@ vm_call_iseq_setup_2(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling
 }
 
 static inline VALUE
-vm_call_iseq_setup_normal(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc, int opt_pc)
+vm_call_iseq_setup_normal(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc,
+			  int opt_pc, int param_size, int local_size)
 {
     const rb_callable_method_entry_t *me = cc->me;
     const rb_iseq_t *iseq = def_iseq_ptr(me->def);
     VALUE *argv = cfp->sp - calling->argc;
-    VALUE *sp = argv + iseq->body->param.size;
+    VALUE *sp = argv + param_size;
     cfp->sp = argv - 1 /* recv */;
 
     vm_push_frame(th, iseq, VM_FRAME_MAGIC_METHOD, calling->recv,
 		  VM_ENVVAL_BLOCK_PTR(calling->blockptr), (VALUE)me,
 		  iseq->body->iseq_encoded + opt_pc, sp,
-		  iseq->body->local_size - iseq->body->param.size,
+		  local_size - param_size,
 		  iseq->body->stack_max);
     return Qundef;
 }
 
 static inline VALUE
-vm_call_iseq_setup_tailcall(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc, int opt_pc)
+vm_call_iseq_setup_tailcall(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc,
+			    int opt_pc)
 {
     unsigned int i;
     VALUE *argv = cfp->sp - calling->argc;
@@ -1437,18 +1457,6 @@ vm_call_iseq_setup_tailcall(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_
 }
 
 static VALUE
-vm_call_iseq_setup_normal_0start(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc)
-{
-    return vm_call_iseq_setup_normal(th, cfp, calling, ci, cc, 0);
-}
-
-static VALUE
-vm_call_iseq_setup_tailcall_0start(rb_thread_t *th, rb_control_frame_t *cfp, struct rb_calling_info *calling, const struct rb_call_info *ci, struct rb_call_cache *cc)
-{
-    return vm_call_iseq_setup_tailcall(th, cfp, calling, ci, cc, 0);
-}
-
-static VALUE
 call_cfunc_m2(VALUE (*func)(ANYARGS), VALUE recv, int argc, const VALUE *argv)
 {
     return (*func)(recv, rb_ary_new4(argc, argv));