mjit_compile.c: reduce sp motion on JIT

This retries r62655, which was reverted at r63863 for r63763.

tool/ruby_vm/views/_mjit_compile_insn.erb: revert the revert.
tool/ruby_vm/views/_mjit_compile_insn_body.erb: ditto.
tool/ruby_vm/views/_mjit_compile_pc_and_sp.erb: ditto.
tool/ruby_vm/views/_mjit_compile_send.erb: ditto.
tool/ruby_vm/views/mjit_compile.inc.erb: ditto.

tool/ruby_vm/views/_insn_entry.erb: revert half of r63763. The commit
  was originally reverted since changing pc motion was bad for tracing,
  but changing sp motion was totally fine. For JIT, I wanna resurrect
  the sp motion change in r62051.
tool/ruby_vm/models/bare_instructions.rb: ditto.
insns.def: ditto.
vm_insnhelper.c: ditto.
vm_insnhelper.h: ditto.

* benchmark

$ benchmark-driver benchmark.yml --rbenv 'before;after;before --jit;after --jit' --repeat-count 12 -v
before: ruby 2.6.0dev (2018-07-19 trunk 63998) [x86_64-linux]
after: ruby 2.6.0dev (2018-07-19 add-sp 63998) [x86_64-linux]
last_commit=mjit_compile.c: reduce sp motion on JIT
before --jit: ruby 2.6.0dev (2018-07-19 trunk 63998) +JIT [x86_64-linux]
after --jit: ruby 2.6.0dev (2018-07-19 add-sp 63998) +JIT [x86_64-linux]
last_commit=mjit_compile.c: reduce sp motion on JIT
Calculating -------------------------------------
                             before       after  before --jit  after --jit
Optcarrot Lan_Master.nes     51.354      50.238        70.010       72.139 fps

Comparison:
             Optcarrot Lan_Master.nes
             after --jit:        72.1 fps
            before --jit:        70.0 fps - 1.03x  slower
                  before:        51.4 fps - 1.40x  slower
                   after:        50.2 fps - 1.44x  slower

git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@63999 b2dd03c8-39d4-4d8f-98ff-823fe69b080e

1 files changed, 15 insertions, 53 deletions

diff --git a/insns.def b/insns.def
index f048cac252..da5a771de0 100644
--- a/insns.def
+++ b/insns.def
@@ -43,6 +43,8 @@
        * sp_inc: Used to dynamically calculate sp increase in
          `insn_stack_increase`.
 
+       * handles_frame: If it is true, VM deals with sp in the insn.
+
    - Attributes can access operands, but not stack (push/pop) variables.
 
    - An instruction's body is a pure C block, copied verbatimly into
@@ -364,7 +366,6 @@ concatstrings
 // attr rb_snum_t sp_inc = 1 - num;
 {
     val = rb_str_concat_literals(num, STACK_ADDR_FROM_TOP(num));
-    POPN(num);
 }
 
 /* push the result of to_s. */
@@ -400,7 +401,6 @@ toregexp
     const VALUE ary = rb_ary_tmp_new_from_values(0, cnt, STACK_ADDR_FROM_TOP(cnt));
     val = rb_reg_new_ary(ary, (int)opt);
     rb_ary_clear(ary);
-    POPN(cnt);
 }
 
 /* intern str to Symbol and push it. */
@@ -422,7 +422,6 @@ newarray
 // attr rb_snum_t sp_inc = 1 - num;
 {
     val = rb_ary_new4(num, STACK_ADDR_FROM_TOP(num));
-    POPN(num);
 }
 
 /* dup array */
@@ -451,7 +450,7 @@ expandarray
 (...)
 // attr rb_snum_t sp_inc = num - 1 + (flag & 1 ? 1 : 0);
 {
-    vm_expandarray(GET_CFP(), ary, num, (int)flag);
+    vm_expandarray(GET_SP(), ary, num, (int)flag);
 }
 
 /* concat two arrays */
@@ -488,7 +487,6 @@ newhash
 
     if (num) {
         rb_hash_bulk_insert(num, STACK_ADDR_FROM_TOP(num), val);
-        POPN(num);
     }
 }
 
@@ -538,7 +536,6 @@ dupn
     void *dst = GET_SP();
     void *src = STACK_ADDR_FROM_TOP(n);
 
-    INC_SP(n); /* alloca */
     MEMCPY(dst, src, VALUE, n);
 }
 
@@ -601,7 +598,7 @@ setn
 (VALUE val)
 // attr rb_snum_t sp_inc = 0;
 {
-    TOPN(n - 1) = val;
+    TOPN(n) = val;
 }
 
 /* empty current stack */
@@ -612,7 +609,7 @@ adjuststack
 (...)
 // attr rb_snum_t sp_inc = -(rb_snum_t)n;
 {
-    POPN(n);
+    /* none */
 }
 
 /**********************************************************/
@@ -690,6 +687,7 @@ defineclass
 (ID id, ISEQ class_iseq, rb_num_t flags)
 (VALUE cbase, VALUE super)
 (VALUE val)
+// attr bool handles_frame = true;
 {
     VALUE klass = vm_find_or_create_class_by_id(id, flags, cbase, super);
 
@@ -716,6 +714,7 @@ send
 (CALL_INFO ci, CALL_CACHE cc, ISEQ blockiseq)
 (...)
 (VALUE val)
+// attr bool handles_frame = true;
 // attr rb_snum_t sp_inc = - (int)(ci->orig_argc + ((ci->flag & VM_CALL_ARGS_BLOCKARG) ? 1 : 0));
 {
     struct rb_calling_info calling;
@@ -751,7 +750,6 @@ opt_newarray_max
 // attr rb_snum_t sp_inc = 1 - num;
 {
     val = vm_opt_newarray_max(num, STACK_ADDR_FROM_TOP(num));
-    POPN(num);
 }
 
 DEFINE_INSN
@@ -762,7 +760,6 @@ opt_newarray_min
 // attr rb_snum_t sp_inc = 1 - num;
 {
     val = vm_opt_newarray_min(num, STACK_ADDR_FROM_TOP(num));
-    POPN(num);
 }
 
 /* Invoke method without block */
@@ -771,6 +768,7 @@ opt_send_without_block
 (CALL_INFO ci, CALL_CACHE cc)
 (...)
 (VALUE val)
+// attr bool handles_frame = true;
 // attr rb_snum_t sp_inc = -ci->orig_argc;
 {
     struct rb_calling_info calling;
@@ -785,6 +783,7 @@ invokesuper
 (CALL_INFO ci, CALL_CACHE cc, ISEQ blockiseq)
 (...)
 (VALUE val)
+// attr bool handles_frame = true;
 // attr rb_snum_t sp_inc = - (int)(ci->orig_argc + ((ci->flag & VM_CALL_ARGS_BLOCKARG) ? 1 : 0));
 {
     struct rb_calling_info calling;
@@ -802,6 +801,7 @@ invokeblock
 (CALL_INFO ci)
 (...)
 (VALUE val)
+// attr bool handles_frame = true;
 // attr rb_snum_t sp_inc = 1 - ci->orig_argc;
 {
     struct rb_calling_info calling;
@@ -828,6 +828,7 @@ leave
 ()
 (VALUE val)
 (VALUE val)
+// attr bool handles_frame = true;
 {
     if (OPT_CHECKED_RUN) {
 	const VALUE *const bp = vm_base_ptr(reg_cfp);
@@ -989,8 +990,6 @@ opt_plus
     val = vm_opt_plus(recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1008,8 +1007,6 @@ opt_minus
     val = vm_opt_minus(recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1027,8 +1024,6 @@ opt_mult
     val = vm_opt_mult(recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1046,8 +1041,6 @@ opt_div
     val = vm_opt_div(recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1065,8 +1058,6 @@ opt_mod
     val = vm_opt_mod(recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1084,8 +1075,6 @@ opt_eq
     val = opt_eq_func(recv, obj, ci, cc);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1103,8 +1092,6 @@ opt_neq
     val = vm_opt_neq(ci, cc, ci_eq, cc_eq, recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1122,8 +1109,6 @@ opt_lt
     val = vm_opt_lt(recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1141,8 +1126,6 @@ opt_le
     val = vm_opt_le(recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1160,8 +1143,6 @@ opt_gt
     val = vm_opt_gt(recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1179,8 +1160,6 @@ opt_ge
     val = vm_opt_ge(recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1198,8 +1177,6 @@ opt_ltlt
     val = vm_opt_ltlt(recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1217,8 +1194,6 @@ opt_aref
     val = vm_opt_aref(recv, obj);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1236,9 +1211,6 @@ opt_aset
     val = vm_opt_aset(recv, obj, set);
 
     if (val == Qundef) {
-        PUSH(recv);
-        PUSH(obj);
-        PUSH(set);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1259,12 +1231,9 @@ opt_aset_with
 	val = tmp;
     }
     else {
-        PUSH(recv);
-#ifndef MJIT_HEADER
-        PUSH(rb_str_resurrect(key));
-#endif
-        PUSH(val);
 #ifndef MJIT_HEADER
+	TOPN(0) = rb_str_resurrect(key);
+	PUSH(val);
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
 	DISPATCH_ORIGINAL_INSN(opt_send_without_block);
@@ -1281,9 +1250,8 @@ opt_aref_with
     val = vm_opt_aref_with(recv, key);
 
     if (val == Qundef) {
-        PUSH(recv);
 #ifndef MJIT_HEADER
-        PUSH(rb_str_resurrect(key));
+	PUSH(rb_str_resurrect(key));
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
 	DISPATCH_ORIGINAL_INSN(opt_send_without_block);
@@ -1300,7 +1268,6 @@ opt_length
     val = vm_opt_length(recv, BOP_LENGTH);
 
     if (val == Qundef) {
-        PUSH(recv);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1318,7 +1285,6 @@ opt_size
     val = vm_opt_length(recv, BOP_SIZE);
 
     if (val == Qundef) {
-        PUSH(recv);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1336,7 +1302,6 @@ opt_empty_p
     val = vm_opt_empty_p(recv);
 
     if (val == Qundef) {
-        PUSH(recv);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1354,7 +1319,6 @@ opt_succ
     val = vm_opt_succ(recv);
 
     if (val == Qundef) {
-        PUSH(recv);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1372,7 +1336,6 @@ opt_not
     val = vm_opt_not(ci, cc, recv);
 
     if (val == Qundef) {
-        PUSH(recv);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1400,8 +1363,6 @@ opt_regexpmatch2
     val = vm_opt_regexpmatch2(obj2, obj1);
 
     if (val == Qundef) {
-        PUSH(obj2);
-        PUSH(obj1);
 #ifndef MJIT_HEADER
         ADD_PC(-WIDTH_OF_opt_send_without_block);
 #endif
@@ -1415,6 +1376,7 @@ opt_call_c_function
 (rb_insn_func_t funcptr)
 ()
 ()
+// attr bool handles_frame = true;
 {
     reg_cfp = (funcptr)(ec, reg_cfp);