* compile.c, vm_macro.def: support tail call optimization

  (on default, this feature is not enabled).
* iseq.c, compile.c, vm_opts.h: add "tailcall_optimization"
  option.
* sample/test.rb (test_ok): fix to adjust tailcall stack layout.
* insns.def, vm.c, compile.c, yarvcore.c, yarvcore.h:
  add opt_gt, opt_le instructions.



git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@12304 b2dd03c8-39d4-4d8f-98ff-823fe69b080e

12 files changed, 174 insertions, 37 deletions

diff --git a/ChangeLog b/ChangeLog
index 0a8d9fe02b..5a3afa1652 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,16 @@
+Mon May 21 13:40:00 2007  Koichi Sasada  <ko1@atdot.net>
+
+	* compile.c, vm_macro.def: support tail call optimization
+	  (on default, this feature is not enabled).
+
+	* iseq.c, compile.c, vm_opts.h: add "tailcall_optimization"
+	  option.
+
+	* sample/test.rb (test_ok): fix to adjust tailcall stack layout.
+
+	* insns.def, vm.c, compile.c, yarvcore.c, yarvcore.h:
+	  add opt_gt, opt_le instructions.
+
 Mon May 21 03:34:06 2007  Minero Aoki  <aamine@loveruby.net>
 
 	* lib/net/smtp.rb: CRAM-MD5 authentication did not work.
diff --git a/compile.c b/compile.c
index 568c792b44..c539e93050 100644
--- a/compile.c
+++ b/compile.c
@@ -1347,7 +1347,7 @@ get_prev_insn(INSN *iobj)
 }
 
 static int
-iseq_peephole_optimize(rb_iseq_t *iseq, LINK_ELEMENT *list)
+iseq_peephole_optimize(rb_iseq_t *iseq, LINK_ELEMENT *list, const int do_tailcallopt)
 {
     INSN *iobj = (INSN *)list;
   again:
@@ -1367,6 +1367,12 @@ iseq_peephole_optimize(rb_iseq_t *iseq, LINK_ELEMENT *list)
 	niobj = (INSN *)get_next_insn(iobj);
 
 	if (diobj == niobj) {
+	    /*
+	     *   jump LABEL
+	     *  LABEL:
+	     * =>
+	     *   LABEL:
+	     */
 	    REMOVE_ELEM(&iobj->link);
 	}
 	else if (iobj != diobj && diobj->insn_id == BIN(jump)) {
@@ -1374,11 +1380,23 @@ iseq_peephole_optimize(rb_iseq_t *iseq, LINK_ELEMENT *list)
 	    goto again;
 	}
 	else if (diobj->insn_id == BIN(leave)) {
+	    /*
+	     *  jump LABEL
+	     *  ...
+	     * LABEL:
+	     *  leave
+	     * =>
+	     *  leave
+	     *  ...
+	     * LABEL:
+	     *  leave
+	     */
 	    INSN *eiobj = new_insn_core(iseq, iobj->line_no, BIN(leave),
 					diobj->operand_size,
 					diobj->operands);
 	    /* replace */
 	    REPLACE_ELEM((LINK_ELEMENT *)iobj, (LINK_ELEMENT *)eiobj);
+	    iobj = eiobj;
 	}
 	/*
 	 * useless jump elimination (if/unless destination):
@@ -1405,6 +1423,7 @@ iseq_peephole_optimize(rb_iseq_t *iseq, LINK_ELEMENT *list)
 	    }
 	}
     }
+
     if (iobj->insn_id == BIN(branchif) ||
 	iobj->insn_id == BIN(branchunless)) {
 	/*
@@ -1421,14 +1440,20 @@ iseq_peephole_optimize(rb_iseq_t *iseq, LINK_ELEMENT *list)
 	}
     }
 
-    if (iobj->insn_id == BIN(leave)) {
+    if (do_tailcallopt && iobj->insn_id == BIN(leave)) {
+	/*
+	 *  send ...
+	 *  leave
+	 * =>
+	 *  send ..., ... | VM_CALL_TAILCALL_BIT, ...
+	 *  leave # unreachable
+	 */
 	INSN *piobj = (INSN *)get_prev_insn((INSN *)list);
-	if (piobj->insn_id == BIN(send)) {
-	    /* TODO: tail call optimization */
-	    if (piobj->operands[2] == 0) {
-		/* piobj->operands[3] = INT2FIX(FIX2INT(piobj->operands[3]) | VM_CALL_TAILCALL_BIT); */
-		/* piobj->operands[3] = INT2FIX(FIX2INT(piobj->operands[3]) | VM_CALL_TAILRECURSION_BIT); */
-	    }
+
+	if (piobj->insn_id == BIN(send) &&
+	    piobj->operands[2] == 0 /* block */
+	    ) {
+	    piobj->operands[3] = INT2FIX(FIX2INT(piobj->operands[3]) | VM_CALL_TAILCALL_BIT);
 	}
     }
     return COMPILE_OK;
@@ -1489,6 +1514,12 @@ iseq_specialized_instruction(rb_iseq_t *iseq, INSN *iobj)
 		else if (mid == idLE) {
 		    insn_set_specialized_instruction(iobj, BIN(opt_le));
 		}
+		else if (mid == idGT) {
+		    insn_set_specialized_instruction(iobj, BIN(opt_gt));
+		}
+		else if (mid == idGE) {
+		    insn_set_specialized_instruction(iobj, BIN(opt_ge));
+		}
 		else if (mid == idLTLT) {
 		    insn_set_specialized_instruction(iobj, BIN(opt_ltlt));
 		}
@@ -1510,15 +1541,16 @@ static int
 iseq_optimize(rb_iseq_t *iseq, LINK_ANCHOR *anchor)
 {
     LINK_ELEMENT *list;
-    const int do_peephole = iseq->compile_data->option->peephole_optimization;
+    const int do_peepholeopt = iseq->compile_data->option->peephole_optimization;
+    const int do_tailcallopt = iseq->compile_data->option->tailcall_optimization;
     const int do_si = iseq->compile_data->option->specialized_instruction;
     const int do_ou = iseq->compile_data->option->operands_unification;
     list = FIRST_ELEMENT(anchor);
     
     while (list) {
 	if (list->type == ISEQ_ELEMENT_INSN) {
-	    if (do_peephole) {
-		iseq_peephole_optimize(iseq, list);
+	    if (do_peepholeopt) {
+		iseq_peephole_optimize(iseq, list, do_tailcallopt);
 	    }
 	    if (do_si) {
 		iseq_specialized_instruction(iseq, (INSN *)list);
diff --git a/insns.def b/insns.def
index 3631ac4ebd..8272f4aac1 100644
--- a/insns.def
+++ b/insns.def
@@ -2139,6 +2139,76 @@ opt_le
 
 /**
   @c optimize
+  @e optimized X>Y.
+  @j 最適化された X>Y。
+ */
+DEFINE_INSN
+opt_gt
+()
+(VALUE recv, VALUE obj)
+(VALUE val)
+{
+    if (FIXNUM_2_P(recv, obj) &&
+	BASIC_OP_UNREDEFINED_P(BOP_GT)) {
+	long a = FIX2LONG(recv), b = FIX2LONG(obj);
+
+	if (a > b) {
+	    val = Qtrue;
+	}
+	else {
+	    val = Qfalse;
+	}
+    }
+    else {
+	/* other */
+#ifdef YARV_AOT_COMPILED
+	val = rb_funcall(recv, idGT, 1, obj);
+#else
+	PUSH(recv);
+	PUSH(obj);
+	tmp_id = idGT;
+	goto LABEL_IS_SC(start_init_in_send_for_opt_1);
+#endif
+    }
+}
+
+/**
+  @c optimize
+  @e optimized X>=Y.
+  @j 最適化された X>=Y。
+ */
+DEFINE_INSN
+opt_ge
+()
+(VALUE recv, VALUE obj)
+(VALUE val)
+{
+    if (FIXNUM_2_P(recv, obj) &&
+	BASIC_OP_UNREDEFINED_P(BOP_GE)) {
+	long a = FIX2LONG(recv), b = FIX2LONG(obj);
+
+	if (a >= b) {
+	    val = Qtrue;
+	}
+	else {
+	    val = Qfalse;
+	}
+    }
+    else {
+	/* other */
+#ifdef YARV_AOT_COMPILED
+	val = rb_funcall(recv, idGE, 1, obj);
+#else
+	PUSH(recv);
+	PUSH(obj);
+	tmp_id = idGE;
+	goto LABEL_IS_SC(start_init_in_send_for_opt_1);
+#endif
+    }
+}
+
+/**
+  @c optimize
   @e <<
   @j <<
  */
diff --git a/iseq.c b/iseq.c
index 11d5b9cbaa..e8c75d5625 100644
--- a/iseq.c
+++ b/iseq.c
@@ -196,11 +196,12 @@ cleanup_iseq_build(rb_iseq_t *iseq)
 static rb_compile_option_t COMPILE_OPTION_DEFAULT = {
     OPT_INLINE_CONST_CACHE, /* int inline_const_cache; */
     OPT_PEEPHOLE_OPTIMIZATION, /* int peephole_optimization; */
+    OPT_TAILCALL_OPTIMIZATION, /* int tailcall_optimization */
     OPT_SPECIALISED_INSTRUCTION, /* int specialized_instruction; */
     OPT_OPERANDS_UNIFICATION, /* int operands_unification; */
     OPT_INSTRUCTIONS_UNIFICATION, /* int instructions_unification; */
     OPT_STACK_CACHING, /* int stack_caching; */
-    OPT_TRACE_INSTRUCTION,
+    OPT_TRACE_INSTRUCTION, /* int trace_instruction */
 };
 static const rb_compile_option_t COMPILE_OPTION_FALSE;
 
@@ -217,13 +218,16 @@ make_compile_option(rb_compile_option_t *option, VALUE opt)
 	memset(option, 1, sizeof(rb_compile_option_t));
     }
     else if (CLASS_OF(opt) == rb_cHash) {
+	*option = COMPILE_OPTION_DEFAULT;
+
 #define SET_COMPILE_OPTION(o, h, mem) \
-  { VALUE flag = rb_hash_aref(h, ID2SYM(rb_intern(#mem))); dp(flag);\
+  { VALUE flag = rb_hash_aref(h, ID2SYM(rb_intern(#mem))); \
       if (flag == Qtrue)  { o->mem = 1; } \
-      if (flag == Qfalse) { o->mem = 0; } \
+      else if (flag == Qfalse)  { o->mem = 0; } \
   }
 	SET_COMPILE_OPTION(option, opt, inline_const_cache);
 	SET_COMPILE_OPTION(option, opt, peephole_optimization);
+	SET_COMPILE_OPTION(option, opt, tailcall_optimization);
 	SET_COMPILE_OPTION(option, opt, specialized_instruction);
 	SET_COMPILE_OPTION(option, opt, operands_unification);
 	SET_COMPILE_OPTION(option, opt, instructions_unification);
@@ -245,6 +249,7 @@ make_compile_option_value(rb_compile_option_t *option)
     {
 	SET_COMPILE_OPTION(option, opt, inline_const_cache);
 	SET_COMPILE_OPTION(option, opt, peephole_optimization);
+	SET_COMPILE_OPTION(option, opt, tailcall_optimization);
 	SET_COMPILE_OPTION(option, opt, specialized_instruction);
 	SET_COMPILE_OPTION(option, opt, operands_unification);
 	SET_COMPILE_OPTION(option, opt, instructions_unification);
diff --git a/sample/test.rb b/sample/test.rb
index 7200ad41bf..da44b180f3 100644
--- a/sample/test.rb
+++ b/sample/test.rb
@@ -17,7 +17,7 @@ def test_ok(cond,n=1)
   if cond
     printf "ok %d\n", $testnum
   else
-    where = caller(n)[0]
+    where = (st = caller(n)) ? st[0] : "caller error! (n=#{n}, trace=#{caller(0).join(', ')}"
     printf "not ok %s %d -- %s\n", $what, $testnum, where
     $failed+=1 
   end
diff --git a/version.h b/version.h
index 11754656c6..ca5979f19f 100644
--- a/version.h
+++ b/version.h
@@ -1,7 +1,7 @@
 #define RUBY_VERSION "1.9.0"
-#define RUBY_RELEASE_DATE "2007-05-19"
+#define RUBY_RELEASE_DATE "2007-05-21"
 #define RUBY_VERSION_CODE 190
-#define RUBY_RELEASE_CODE 20070519
+#define RUBY_RELEASE_CODE 20070521
 #define RUBY_PATCHLEVEL 0
 
 #define RUBY_VERSION_MAJOR 1
@@ -9,7 +9,7 @@
 #define RUBY_VERSION_TEENY 0
 #define RUBY_RELEASE_YEAR 2007
 #define RUBY_RELEASE_MONTH 5
-#define RUBY_RELEASE_DAY 19
+#define RUBY_RELEASE_DAY 21
 
 #ifdef RUBY_EXTERN
 RUBY_EXTERN const char ruby_version[];
diff --git a/vm.c b/vm.c
index d1a0b34e37..3337117f2c 100644
--- a/vm.c
+++ b/vm.c
@@ -1480,6 +1480,8 @@ yarv_init_redefined_flag(void)
 	idASET, BOP_ASET, rb_cArray, rb_cHash, 0,
 	idLength, BOP_LENGTH, rb_cArray, rb_cString, rb_cHash, 0,
 	idSucc, BOP_SUCC, rb_cFixnum, rb_cString, rb_cTime, 0,
+	idGT, BOP_GT, rb_cFixnum, 0,
+	idGE, BOP_GE, rb_cFixnum, 0,
 	0,
     };
     VALUE *ptr = register_info;
diff --git a/vm.h b/vm.h
index 930947086d..457922ceba 100644
--- a/vm.h
+++ b/vm.h
@@ -283,5 +283,7 @@ default:                        \
 #define BOP_ASET    0x400
 #define BOP_LENGTH  0x800
 #define BOP_SUCC   0x1000
+#define BOP_GT     0x2000
+#define BOP_GE     0x4000
 
 #endif /* _VM_H_INCLUDED_ */
diff --git a/vm_macro.def b/vm_macro.def
index dff6b98c9c..adeaa52ac6 100644
--- a/vm_macro.def
+++ b/vm_macro.def
@@ -198,33 +198,38 @@ MACRO macro_eval_invoke_func(niseqval, recv, klass, blockptr, num)
 	rb_exc_raise(sysstack_error);
     }
 
-    for (i = 0; i < clear_local_size; i++) {
-	*sp++ = Qnil;
-    }
+    if (flag & VM_CALL_TAILCALL_BIT) {
+	/* copy arguments */
+	VALUE *p_rsp, *p_sp;
+	reg_cfp = ++th->cfp;
+	p_rsp = th->cfp->sp;
+
+	for (i=0; i < (sp - rsp); i++) {
+	    p_rsp[i] = rsp[i];
+	}
+
+	sp -= rsp - p_rsp;
 
-    if (0 && (flag & VM_CALL_TAILCALL_BIT)) {
-	th->cfp++;
-	push_frame(th, niseq, FRAME_MAGIC_METHOD,
-		   recv, (VALUE) blockptr,
+	for (i = 0; i < clear_local_size; i++) {
+	    *sp++ = Qnil;
+	}
+
+	push_frame(th, niseq,
+		   FRAME_MAGIC_METHOD, recv, (VALUE) blockptr,
 		   niseq->iseq_encoded + opt_pc, sp, 0, 0);
     }
-    else if (0 &&
-	     (flag & VM_CALL_TAILRECURSION_BIT) && niseq == GET_ISEQ()) {
-	/* do nothing */
-	GET_CFP()->self = recv;
-	SET_LFP(sp);
-	SET_DFP(sp);
-	*sp++ = (VALUE) blockptr;
-	reg_cfp->sp = sp;
-	reg_cfp->bp = sp;
-	SET_PC(niseq->iseq_encoded + opt_pc);
-    }
     else {
+	for (i = 0; i < clear_local_size; i++) {
+	    *sp++ = Qnil;
+	}
+
 	push_frame(th, niseq,
 		   FRAME_MAGIC_METHOD, recv, (VALUE) blockptr,
 		   niseq->iseq_encoded + opt_pc, sp, 0, 0);
+
 	reg_cfp->sp = rsp;
     }
+
     RESTORE_REGS();
 }
 
diff --git a/vm_opts.h b/vm_opts.h
index 041cef7faa..32b00ac79a 100644
--- a/vm_opts.h
+++ b/vm_opts.h
@@ -22,10 +22,11 @@
 
 /* VM running option */
 #define OPT_CHECKED_RUN              1
-#define OPT_TRACE_INSTRUCTION        1
+#define OPT_TRACE_INSTRUCTION        0 /* default value */
 
 /* at compile */
 #define OPT_INLINE_CONST_CACHE       1
+#define OPT_TAILCALL_OPTIMIZATION    0 /* default value */
 #define OPT_PEEPHOLE_OPTIMIZATION    1
 #define OPT_SPECIALISED_INSTRUCTION  1
 
diff --git a/yarvcore.c b/yarvcore.c
index cb66e67892..0d2dcecd81 100644
--- a/yarvcore.c
+++ b/yarvcore.c
@@ -31,6 +31,8 @@ ID idMOD;
 ID idLT;
 ID idLTLT;
 ID idLE;
+ID idGT;
+ID idGE;
 ID idEq;
 ID idEqq;
 ID idBackquote;
@@ -495,6 +497,8 @@ Init_VM(void)
     idLT = rb_intern("<");
     idLTLT = rb_intern("<<");
     idLE = rb_intern("<=");
+    idGT = rb_intern(">");
+    idGE = rb_intern(">=");
     idEq = rb_intern("==");
     idEqq = rb_intern("===");
     idBackquote = rb_intern("`");
diff --git a/yarvcore.h b/yarvcore.h
index 0ff3e9b011..61bd53764b 100644
--- a/yarvcore.h
+++ b/yarvcore.h
@@ -109,6 +109,8 @@ extern ID idMOD;
 extern ID idLT;
 extern ID idLTLT;
 extern ID idLE;
+extern ID idGT;
+extern ID idGE;
 extern ID idEq;
 extern ID idEqq;
 extern ID idBackquote;
@@ -182,6 +184,7 @@ struct iseq_compile_data_ensure_node_stack;
 typedef struct rb_compile_option_struct {
     int inline_const_cache;
     int peephole_optimization;
+    int tailcall_optimization;
     int specialized_instruction;
     int operands_unification;
     int instructions_unification;