1 /*
2 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "incls/_precompiled.incl"
26 #include "incls/_templateTable_x86_64.cpp.incl"
27
28 #ifndef CC_INTERP
29
30 #define __ _masm->
31
32 // Platform-dependent initialization
33
34 void TemplateTable::pd_initialize() {
35 // No amd64 specific initialization
36 }
37
38 // Address computation: local variables
39
40 static inline Address iaddress(int n) {
41 return Address(r14, Interpreter::local_offset_in_bytes(n));
42 }
43
44 static inline Address laddress(int n) {
45 return iaddress(n + 1);
46 }
47
48 static inline Address faddress(int n) {
49 return iaddress(n);
50 }
51
52 static inline Address daddress(int n) {
53 return laddress(n);
54 }
55
56 static inline Address aaddress(int n) {
57 return iaddress(n);
58 }
59
60 static inline Address iaddress(Register r) {
61 return Address(r14, r, Address::times_8);
62 }
63
64 static inline Address laddress(Register r) {
65 return Address(r14, r, Address::times_8, Interpreter::local_offset_in_bytes(1));
66 }
67
68 static inline Address faddress(Register r) {
69 return iaddress(r);
70 }
71
72 static inline Address daddress(Register r) {
73 return laddress(r);
74 }
75
76 static inline Address aaddress(Register r) {
77 return iaddress(r);
78 }
79
80 static inline Address at_rsp() {
81 return Address(rsp, 0);
82 }
83
84 // At top of Java expression stack which may be different than esp(). It
85 // isn't for category 1 objects.
86 static inline Address at_tos () {
87 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
88 }
89
90 static inline Address at_tos_p1() {
91 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
92 }
93
94 static inline Address at_tos_p2() {
95 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
96 }
97
98 static inline Address at_tos_p3() {
99 return Address(rsp, Interpreter::expr_offset_in_bytes(3));
100 }
101
102 // Condition conversion
103 static Assembler::Condition j_not(TemplateTable::Condition cc) {
104 switch (cc) {
105 case TemplateTable::equal : return Assembler::notEqual;
106 case TemplateTable::not_equal : return Assembler::equal;
107 case TemplateTable::less : return Assembler::greaterEqual;
108 case TemplateTable::less_equal : return Assembler::greater;
109 case TemplateTable::greater : return Assembler::lessEqual;
110 case TemplateTable::greater_equal: return Assembler::less;
111 }
112 ShouldNotReachHere();
113 return Assembler::zero;
114 }
115
116
117 // Miscelaneous helper routines
118 // Store an oop (or NULL) at the address described by obj.
119 // If val == noreg this means store a NULL
120
121 static void do_oop_store(InterpreterMacroAssembler* _masm,
122 Address obj,
123 Register val,
124 BarrierSet::Name barrier,
125 bool precise) {
126 assert(val == noreg || val == rax, "parameter is just for looks");
127 switch (barrier) {
128 #ifndef SERIALGC
129 case BarrierSet::G1SATBCT:
130 case BarrierSet::G1SATBCTLogging:
131 {
132 // flatten object address if needed
133 if (obj.index() == noreg && obj.disp() == 0) {
134 if (obj.base() != rdx) {
135 __ movq(rdx, obj.base());
136 }
137 } else {
138 __ leaq(rdx, obj);
139 }
140 __ g1_write_barrier_pre(rdx, r8, rbx, val != noreg);
141 if (val == noreg) {
142 __ store_heap_oop_null(Address(rdx, 0));
143 } else {
144 __ store_heap_oop(Address(rdx, 0), val);
145 __ g1_write_barrier_post(rdx, val, r8, rbx);
146 }
147
148 }
149 break;
150 #endif // SERIALGC
151 case BarrierSet::CardTableModRef:
152 case BarrierSet::CardTableExtension:
153 {
154 if (val == noreg) {
155 __ store_heap_oop_null(obj);
156 } else {
157 __ store_heap_oop(obj, val);
158 // flatten object address if needed
159 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
160 __ store_check(obj.base());
161 } else {
162 __ leaq(rdx, obj);
163 __ store_check(rdx);
164 }
165 }
166 }
167 break;
168 case BarrierSet::ModRef:
169 case BarrierSet::Other:
170 if (val == noreg) {
171 __ store_heap_oop_null(obj);
172 } else {
173 __ store_heap_oop(obj, val);
174 }
175 break;
176 default :
177 ShouldNotReachHere();
178
179 }
180 }
181
182 Address TemplateTable::at_bcp(int offset) {
183 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
184 return Address(r13, offset);
185 }
186
187 void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
188 Register scratch,
189 bool load_bc_into_scratch/*=true*/) {
190 if (!RewriteBytecodes) {
191 return;
192 }
193 // the pair bytecodes have already done the load.
194 if (load_bc_into_scratch) {
195 __ movl(bc, bytecode);
196 }
197 Label patch_done;
198 if (JvmtiExport::can_post_breakpoint()) {
199 Label fast_patch;
200 // if a breakpoint is present we can't rewrite the stream directly
201 __ movzbl(scratch, at_bcp(0));
202 __ cmpl(scratch, Bytecodes::_breakpoint);
203 __ jcc(Assembler::notEqual, fast_patch);
204 __ get_method(scratch);
205 // Let breakpoint table handling rewrite to quicker bytecode
206 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, r13, bc);
207 #ifndef ASSERT
208 __ jmpb(patch_done);
209 #else
210 __ jmp(patch_done);
211 #endif
212 __ bind(fast_patch);
213 }
214 #ifdef ASSERT
215 Label okay;
216 __ load_unsigned_byte(scratch, at_bcp(0));
217 __ cmpl(scratch, (int) Bytecodes::java_code(bytecode));
218 __ jcc(Assembler::equal, okay);
219 __ cmpl(scratch, bc);
220 __ jcc(Assembler::equal, okay);
221 __ stop("patching the wrong bytecode");
222 __ bind(okay);
223 #endif
224 // patch bytecode
225 __ movb(at_bcp(0), bc);
226 __ bind(patch_done);
227 }
228
229
230 // Individual instructions
231
232 void TemplateTable::nop() {
233 transition(vtos, vtos);
234 // nothing to do
235 }
236
237 void TemplateTable::shouldnotreachhere() {
238 transition(vtos, vtos);
239 __ stop("shouldnotreachhere bytecode");
240 }
241
242 void TemplateTable::aconst_null() {
243 transition(vtos, atos);
244 __ xorl(rax, rax);
245 }
246
247 void TemplateTable::iconst(int value) {
248 transition(vtos, itos);
249 if (value == 0) {
250 __ xorl(rax, rax);
251 } else {
252 __ movl(rax, value);
253 }
254 }
255
256 void TemplateTable::lconst(int value) {
257 transition(vtos, ltos);
258 if (value == 0) {
259 __ xorl(rax, rax);
260 } else {
261 __ movl(rax, value);
262 }
263 }
264
265 void TemplateTable::fconst(int value) {
266 transition(vtos, ftos);
267 static float one = 1.0f, two = 2.0f;
268 switch (value) {
269 case 0:
270 __ xorps(xmm0, xmm0);
271 break;
272 case 1:
273 __ movflt(xmm0, ExternalAddress((address) &one));
274 break;
275 case 2:
276 __ movflt(xmm0, ExternalAddress((address) &two));
277 break;
278 default:
279 ShouldNotReachHere();
280 break;
281 }
282 }
283
284 void TemplateTable::dconst(int value) {
285 transition(vtos, dtos);
286 static double one = 1.0;
287 switch (value) {
288 case 0:
289 __ xorpd(xmm0, xmm0);
290 break;
291 case 1:
292 __ movdbl(xmm0, ExternalAddress((address) &one));
293 break;
294 default:
295 ShouldNotReachHere();
296 break;
297 }
298 }
299
300 void TemplateTable::bipush() {
301 transition(vtos, itos);
302 __ load_signed_byte(rax, at_bcp(1));
303 }
304
305 void TemplateTable::sipush() {
306 transition(vtos, itos);
307 __ load_unsigned_short(rax, at_bcp(1));
308 __ bswapl(rax);
309 __ sarl(rax, 16);
310 }
311
312 void TemplateTable::ldc(bool wide) {
313 transition(vtos, vtos);
314 Label call_ldc, notFloat, notClass, Done;
315
316 if (wide) {
317 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
318 } else {
319 __ load_unsigned_byte(rbx, at_bcp(1));
320 }
321
322 __ get_cpool_and_tags(rcx, rax);
323 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
324 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
325
326 // get type
327 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
328
329 // unresolved string - get the resolved string
330 __ cmpl(rdx, JVM_CONSTANT_UnresolvedString);
331 __ jccb(Assembler::equal, call_ldc);
332
333 // unresolved class - get the resolved class
334 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
335 __ jccb(Assembler::equal, call_ldc);
336
337 // unresolved class in error state - call into runtime to throw the error
338 // from the first resolution attempt
339 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
340 __ jccb(Assembler::equal, call_ldc);
341
342 // resolved class - need to call vm to get java mirror of the class
343 __ cmpl(rdx, JVM_CONSTANT_Class);
344 __ jcc(Assembler::notEqual, notClass);
345
346 __ bind(call_ldc);
347 __ movl(c_rarg1, wide);
348 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
349 __ push_ptr(rax);
350 __ verify_oop(rax);
351 __ jmp(Done);
352
353 __ bind(notClass);
354 __ cmpl(rdx, JVM_CONSTANT_Float);
355 __ jccb(Assembler::notEqual, notFloat);
356 // ftos
357 __ movflt(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
358 __ push_f();
359 __ jmp(Done);
360
361 __ bind(notFloat);
362 #ifdef ASSERT
363 {
364 Label L;
365 __ cmpl(rdx, JVM_CONSTANT_Integer);
366 __ jcc(Assembler::equal, L);
367 __ cmpl(rdx, JVM_CONSTANT_String);
368 __ jcc(Assembler::equal, L);
369 __ stop("unexpected tag type in ldc");
370 __ bind(L);
371 }
372 #endif
373 // atos and itos
374 Label isOop;
375 __ cmpl(rdx, JVM_CONSTANT_Integer);
376 __ jcc(Assembler::notEqual, isOop);
377 __ movl(rax, Address(rcx, rbx, Address::times_8, base_offset));
378 __ push_i(rax);
379 __ jmp(Done);
380
381 __ bind(isOop);
382 __ movptr(rax, Address(rcx, rbx, Address::times_8, base_offset));
383 __ push_ptr(rax);
384
385 if (VerifyOops) {
386 __ verify_oop(rax);
387 }
388
389 __ bind(Done);
390 }
391
392 // Fast path for caching oop constants.
393 // %%% We should use this to handle Class and String constants also.
394 // %%% It will simplify the ldc/primitive path considerably.
395 void TemplateTable::fast_aldc(bool wide) {
396 transition(vtos, atos);
397
398 if (!EnableMethodHandles) {
399 // We should not encounter this bytecode if !EnableMethodHandles.
400 // The verifier will stop it. However, if we get past the verifier,
401 // this will stop the thread in a reasonable way, without crashing the JVM.
402 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
403 InterpreterRuntime::throw_IncompatibleClassChangeError));
404 // the call_VM checks for exception, so we should never return here.
405 __ should_not_reach_here();
406 return;
407 }
408
409 const Register cache = rcx;
410 const Register index = rdx;
411
412 resolve_cache_and_index(f1_oop, rax, cache, index, wide ? sizeof(u2) : sizeof(u1));
413 if (VerifyOops) {
414 __ verify_oop(rax);
415 }
416 }
417
418 void TemplateTable::ldc2_w() {
419 transition(vtos, vtos);
420 Label Long, Done;
421 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
422
423 __ get_cpool_and_tags(rcx, rax);
424 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
425 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
426
427 // get type
428 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
429 JVM_CONSTANT_Double);
430 __ jccb(Assembler::notEqual, Long);
431 // dtos
432 __ movdbl(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
433 __ push_d();
434 __ jmpb(Done);
435
436 __ bind(Long);
437 // ltos
438 __ movq(rax, Address(rcx, rbx, Address::times_8, base_offset));
439 __ push_l();
440
441 __ bind(Done);
442 }
443
444 void TemplateTable::locals_index(Register reg, int offset) {
445 __ load_unsigned_byte(reg, at_bcp(offset));
446 __ negptr(reg);
447 }
448
449 void TemplateTable::iload() {
450 transition(vtos, itos);
451 if (RewriteFrequentPairs) {
452 Label rewrite, done;
453 const Register bc = c_rarg3;
454 assert(rbx != bc, "register damaged");
455
456 // get next byte
457 __ load_unsigned_byte(rbx,
458 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
459 // if _iload, wait to rewrite to iload2. We only want to rewrite the
460 // last two iloads in a pair. Comparing against fast_iload means that
461 // the next bytecode is neither an iload or a caload, and therefore
462 // an iload pair.
463 __ cmpl(rbx, Bytecodes::_iload);
464 __ jcc(Assembler::equal, done);
465
466 __ cmpl(rbx, Bytecodes::_fast_iload);
467 __ movl(bc, Bytecodes::_fast_iload2);
468 __ jccb(Assembler::equal, rewrite);
469
470 // if _caload, rewrite to fast_icaload
471 __ cmpl(rbx, Bytecodes::_caload);
472 __ movl(bc, Bytecodes::_fast_icaload);
473 __ jccb(Assembler::equal, rewrite);
474
475 // rewrite so iload doesn't check again.
476 __ movl(bc, Bytecodes::_fast_iload);
477
478 // rewrite
479 // bc: fast bytecode
480 __ bind(rewrite);
481 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
482 __ bind(done);
483 }
484
485 // Get the local value into tos
486 locals_index(rbx);
487 __ movl(rax, iaddress(rbx));
488 }
489
490 void TemplateTable::fast_iload2() {
491 transition(vtos, itos);
492 locals_index(rbx);
493 __ movl(rax, iaddress(rbx));
494 __ push(itos);
495 locals_index(rbx, 3);
496 __ movl(rax, iaddress(rbx));
497 }
498
499 void TemplateTable::fast_iload() {
500 transition(vtos, itos);
501 locals_index(rbx);
502 __ movl(rax, iaddress(rbx));
503 }
504
505 void TemplateTable::lload() {
506 transition(vtos, ltos);
507 locals_index(rbx);
508 __ movq(rax, laddress(rbx));
509 }
510
511 void TemplateTable::fload() {
512 transition(vtos, ftos);
513 locals_index(rbx);
514 __ movflt(xmm0, faddress(rbx));
515 }
516
517 void TemplateTable::dload() {
518 transition(vtos, dtos);
519 locals_index(rbx);
520 __ movdbl(xmm0, daddress(rbx));
521 }
522
523 void TemplateTable::aload() {
524 transition(vtos, atos);
525 locals_index(rbx);
526 __ movptr(rax, aaddress(rbx));
527 }
528
529 void TemplateTable::locals_index_wide(Register reg) {
530 __ movl(reg, at_bcp(2));
531 __ bswapl(reg);
532 __ shrl(reg, 16);
533 __ negptr(reg);
534 }
535
536 void TemplateTable::wide_iload() {
537 transition(vtos, itos);
538 locals_index_wide(rbx);
539 __ movl(rax, iaddress(rbx));
540 }
541
542 void TemplateTable::wide_lload() {
543 transition(vtos, ltos);
544 locals_index_wide(rbx);
545 __ movq(rax, laddress(rbx));
546 }
547
548 void TemplateTable::wide_fload() {
549 transition(vtos, ftos);
550 locals_index_wide(rbx);
551 __ movflt(xmm0, faddress(rbx));
552 }
553
554 void TemplateTable::wide_dload() {
555 transition(vtos, dtos);
556 locals_index_wide(rbx);
557 __ movdbl(xmm0, daddress(rbx));
558 }
559
560 void TemplateTable::wide_aload() {
561 transition(vtos, atos);
562 locals_index_wide(rbx);
563 __ movptr(rax, aaddress(rbx));
564 }
565
566 void TemplateTable::index_check(Register array, Register index) {
567 // destroys rbx
568 // check array
569 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
570 // sign extend index for use by indexed load
571 __ movl2ptr(index, index);
572 // check index
573 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
574 if (index != rbx) {
575 // ??? convention: move aberrant index into ebx for exception message
576 assert(rbx != array, "different registers");
577 __ movl(rbx, index);
578 }
579 __ jump_cc(Assembler::aboveEqual,
580 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
581 }
582
583 void TemplateTable::iaload() {
584 transition(itos, itos);
585 __ pop_ptr(rdx);
586 // eax: index
587 // rdx: array
588 index_check(rdx, rax); // kills rbx
589 __ movl(rax, Address(rdx, rax,
590 Address::times_4,
591 arrayOopDesc::base_offset_in_bytes(T_INT)));
592 }
593
594 void TemplateTable::laload() {
595 transition(itos, ltos);
596 __ pop_ptr(rdx);
597 // eax: index
598 // rdx: array
599 index_check(rdx, rax); // kills rbx
600 __ movq(rax, Address(rdx, rbx,
601 Address::times_8,
602 arrayOopDesc::base_offset_in_bytes(T_LONG)));
603 }
604
605 void TemplateTable::faload() {
606 transition(itos, ftos);
607 __ pop_ptr(rdx);
608 // eax: index
609 // rdx: array
610 index_check(rdx, rax); // kills rbx
611 __ movflt(xmm0, Address(rdx, rax,
612 Address::times_4,
613 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
614 }
615
616 void TemplateTable::daload() {
617 transition(itos, dtos);
618 __ pop_ptr(rdx);
619 // eax: index
620 // rdx: array
621 index_check(rdx, rax); // kills rbx
622 __ movdbl(xmm0, Address(rdx, rax,
623 Address::times_8,
624 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
625 }
626
627 void TemplateTable::aaload() {
628 transition(itos, atos);
629 __ pop_ptr(rdx);
630 // eax: index
631 // rdx: array
632 index_check(rdx, rax); // kills rbx
633 __ load_heap_oop(rax, Address(rdx, rax,
634 UseCompressedOops ? Address::times_4 : Address::times_8,
635 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
636 }
637
638 void TemplateTable::baload() {
639 transition(itos, itos);
640 __ pop_ptr(rdx);
641 // eax: index
642 // rdx: array
643 index_check(rdx, rax); // kills rbx
644 __ load_signed_byte(rax,
645 Address(rdx, rax,
646 Address::times_1,
647 arrayOopDesc::base_offset_in_bytes(T_BYTE)));
648 }
649
650 void TemplateTable::caload() {
651 transition(itos, itos);
652 __ pop_ptr(rdx);
653 // eax: index
654 // rdx: array
655 index_check(rdx, rax); // kills rbx
656 __ load_unsigned_short(rax,
657 Address(rdx, rax,
658 Address::times_2,
659 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
660 }
661
662 // iload followed by caload frequent pair
663 void TemplateTable::fast_icaload() {
664 transition(vtos, itos);
665 // load index out of locals
666 locals_index(rbx);
667 __ movl(rax, iaddress(rbx));
668
669 // eax: index
670 // rdx: array
671 __ pop_ptr(rdx);
672 index_check(rdx, rax); // kills rbx
673 __ load_unsigned_short(rax,
674 Address(rdx, rax,
675 Address::times_2,
676 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
677 }
678
679 void TemplateTable::saload() {
680 transition(itos, itos);
681 __ pop_ptr(rdx);
682 // eax: index
683 // rdx: array
684 index_check(rdx, rax); // kills rbx
685 __ load_signed_short(rax,
686 Address(rdx, rax,
687 Address::times_2,
688 arrayOopDesc::base_offset_in_bytes(T_SHORT)));
689 }
690
691 void TemplateTable::iload(int n) {
692 transition(vtos, itos);
693 __ movl(rax, iaddress(n));
694 }
695
696 void TemplateTable::lload(int n) {
697 transition(vtos, ltos);
698 __ movq(rax, laddress(n));
699 }
700
701 void TemplateTable::fload(int n) {
702 transition(vtos, ftos);
703 __ movflt(xmm0, faddress(n));
704 }
705
706 void TemplateTable::dload(int n) {
707 transition(vtos, dtos);
708 __ movdbl(xmm0, daddress(n));
709 }
710
711 void TemplateTable::aload(int n) {
712 transition(vtos, atos);
713 __ movptr(rax, aaddress(n));
714 }
715
716 void TemplateTable::aload_0() {
717 transition(vtos, atos);
718 // According to bytecode histograms, the pairs:
719 //
720 // _aload_0, _fast_igetfield
721 // _aload_0, _fast_agetfield
722 // _aload_0, _fast_fgetfield
723 //
724 // occur frequently. If RewriteFrequentPairs is set, the (slow)
725 // _aload_0 bytecode checks if the next bytecode is either
726 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
727 // rewrites the current bytecode into a pair bytecode; otherwise it
728 // rewrites the current bytecode into _fast_aload_0 that doesn't do
729 // the pair check anymore.
730 //
731 // Note: If the next bytecode is _getfield, the rewrite must be
732 // delayed, otherwise we may miss an opportunity for a pair.
733 //
734 // Also rewrite frequent pairs
735 // aload_0, aload_1
736 // aload_0, iload_1
737 // These bytecodes with a small amount of code are most profitable
738 // to rewrite
739 if (RewriteFrequentPairs) {
740 Label rewrite, done;
741 const Register bc = c_rarg3;
742 assert(rbx != bc, "register damaged");
743 // get next byte
744 __ load_unsigned_byte(rbx,
745 at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
746
747 // do actual aload_0
748 aload(0);
749
750 // if _getfield then wait with rewrite
751 __ cmpl(rbx, Bytecodes::_getfield);
752 __ jcc(Assembler::equal, done);
753
754 // if _igetfield then reqrite to _fast_iaccess_0
755 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) ==
756 Bytecodes::_aload_0,
757 "fix bytecode definition");
758 __ cmpl(rbx, Bytecodes::_fast_igetfield);
759 __ movl(bc, Bytecodes::_fast_iaccess_0);
760 __ jccb(Assembler::equal, rewrite);
761
762 // if _agetfield then reqrite to _fast_aaccess_0
763 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) ==
764 Bytecodes::_aload_0,
765 "fix bytecode definition");
766 __ cmpl(rbx, Bytecodes::_fast_agetfield);
767 __ movl(bc, Bytecodes::_fast_aaccess_0);
768 __ jccb(Assembler::equal, rewrite);
769
770 // if _fgetfield then reqrite to _fast_faccess_0
771 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) ==
772 Bytecodes::_aload_0,
773 "fix bytecode definition");
774 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
775 __ movl(bc, Bytecodes::_fast_faccess_0);
776 __ jccb(Assembler::equal, rewrite);
777
778 // else rewrite to _fast_aload0
779 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) ==
780 Bytecodes::_aload_0,
781 "fix bytecode definition");
782 __ movl(bc, Bytecodes::_fast_aload_0);
783
784 // rewrite
785 // bc: fast bytecode
786 __ bind(rewrite);
787 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
788
789 __ bind(done);
790 } else {
791 aload(0);
792 }
793 }
794
795 void TemplateTable::istore() {
796 transition(itos, vtos);
797 locals_index(rbx);
798 __ movl(iaddress(rbx), rax);
799 }
800
801 void TemplateTable::lstore() {
802 transition(ltos, vtos);
803 locals_index(rbx);
804 __ movq(laddress(rbx), rax);
805 }
806
807 void TemplateTable::fstore() {
808 transition(ftos, vtos);
809 locals_index(rbx);
810 __ movflt(faddress(rbx), xmm0);
811 }
812
813 void TemplateTable::dstore() {
814 transition(dtos, vtos);
815 locals_index(rbx);
816 __ movdbl(daddress(rbx), xmm0);
817 }
818
819 void TemplateTable::astore() {
820 transition(vtos, vtos);
821 __ pop_ptr(rax);
822 locals_index(rbx);
823 __ movptr(aaddress(rbx), rax);
824 }
825
826 void TemplateTable::wide_istore() {
827 transition(vtos, vtos);
828 __ pop_i();
829 locals_index_wide(rbx);
830 __ movl(iaddress(rbx), rax);
831 }
832
833 void TemplateTable::wide_lstore() {
834 transition(vtos, vtos);
835 __ pop_l();
836 locals_index_wide(rbx);
837 __ movq(laddress(rbx), rax);
838 }
839
840 void TemplateTable::wide_fstore() {
841 transition(vtos, vtos);
842 __ pop_f();
843 locals_index_wide(rbx);
844 __ movflt(faddress(rbx), xmm0);
845 }
846
847 void TemplateTable::wide_dstore() {
848 transition(vtos, vtos);
849 __ pop_d();
850 locals_index_wide(rbx);
851 __ movdbl(daddress(rbx), xmm0);
852 }
853
854 void TemplateTable::wide_astore() {
855 transition(vtos, vtos);
856 __ pop_ptr(rax);
857 locals_index_wide(rbx);
858 __ movptr(aaddress(rbx), rax);
859 }
860
861 void TemplateTable::iastore() {
862 transition(itos, vtos);
863 __ pop_i(rbx);
864 __ pop_ptr(rdx);
865 // eax: value
866 // ebx: index
867 // rdx: array
868 index_check(rdx, rbx); // prefer index in ebx
869 __ movl(Address(rdx, rbx,
870 Address::times_4,
871 arrayOopDesc::base_offset_in_bytes(T_INT)),
872 rax);
873 }
874
875 void TemplateTable::lastore() {
876 transition(ltos, vtos);
877 __ pop_i(rbx);
878 __ pop_ptr(rdx);
879 // rax: value
880 // ebx: index
881 // rdx: array
882 index_check(rdx, rbx); // prefer index in ebx
883 __ movq(Address(rdx, rbx,
884 Address::times_8,
885 arrayOopDesc::base_offset_in_bytes(T_LONG)),
886 rax);
887 }
888
889 void TemplateTable::fastore() {
890 transition(ftos, vtos);
891 __ pop_i(rbx);
892 __ pop_ptr(rdx);
893 // xmm0: value
894 // ebx: index
895 // rdx: array
896 index_check(rdx, rbx); // prefer index in ebx
897 __ movflt(Address(rdx, rbx,
898 Address::times_4,
899 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
900 xmm0);
901 }
902
903 void TemplateTable::dastore() {
904 transition(dtos, vtos);
905 __ pop_i(rbx);
906 __ pop_ptr(rdx);
907 // xmm0: value
908 // ebx: index
909 // rdx: array
910 index_check(rdx, rbx); // prefer index in ebx
911 __ movdbl(Address(rdx, rbx,
912 Address::times_8,
913 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
914 xmm0);
915 }
916
917 void TemplateTable::aastore() {
918 Label is_null, ok_is_subtype, done;
919 transition(vtos, vtos);
920 // stack: ..., array, index, value
921 __ movptr(rax, at_tos()); // value
922 __ movl(rcx, at_tos_p1()); // index
923 __ movptr(rdx, at_tos_p2()); // array
924
925 Address element_address(rdx, rcx,
926 UseCompressedOops? Address::times_4 : Address::times_8,
927 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
928
929 index_check(rdx, rcx); // kills rbx
930 // do array store check - check for NULL value first
931 __ testptr(rax, rax);
932 __ jcc(Assembler::zero, is_null);
933
934 // Move subklass into rbx
935 __ load_klass(rbx, rax);
936 // Move superklass into rax
937 __ load_klass(rax, rdx);
938 __ movptr(rax, Address(rax,
939 sizeof(oopDesc) +
940 objArrayKlass::element_klass_offset_in_bytes()));
941 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
942 __ lea(rdx, element_address);
943
944 // Generate subtype check. Blows rcx, rdi
945 // Superklass in rax. Subklass in rbx.
946 __ gen_subtype_check(rbx, ok_is_subtype);
947
948 // Come here on failure
949 // object is at TOS
950 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
951
952 // Come here on success
953 __ bind(ok_is_subtype);
954
955 // Get the value we will store
956 __ movptr(rax, at_tos());
957 // Now store using the appropriate barrier
958 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
959 __ jmp(done);
960
961 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
962 __ bind(is_null);
963 __ profile_null_seen(rbx);
964
965 // Store a NULL
966 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
967
968 // Pop stack arguments
969 __ bind(done);
970 __ addptr(rsp, 3 * Interpreter::stackElementSize);
971 }
972
973 void TemplateTable::bastore() {
974 transition(itos, vtos);
975 __ pop_i(rbx);
976 __ pop_ptr(rdx);
977 // eax: value
978 // ebx: index
979 // rdx: array
980 index_check(rdx, rbx); // prefer index in ebx
981 __ movb(Address(rdx, rbx,
982 Address::times_1,
983 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
984 rax);
985 }
986
987 void TemplateTable::castore() {
988 transition(itos, vtos);
989 __ pop_i(rbx);
990 __ pop_ptr(rdx);
991 // eax: value
992 // ebx: index
993 // rdx: array
994 index_check(rdx, rbx); // prefer index in ebx
995 __ movw(Address(rdx, rbx,
996 Address::times_2,
997 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
998 rax);
999 }
1000
1001 void TemplateTable::sastore() {
1002 castore();
1003 }
1004
1005 void TemplateTable::istore(int n) {
1006 transition(itos, vtos);
1007 __ movl(iaddress(n), rax);
1008 }
1009
1010 void TemplateTable::lstore(int n) {
1011 transition(ltos, vtos);
1012 __ movq(laddress(n), rax);
1013 }
1014
1015 void TemplateTable::fstore(int n) {
1016 transition(ftos, vtos);
1017 __ movflt(faddress(n), xmm0);
1018 }
1019
1020 void TemplateTable::dstore(int n) {
1021 transition(dtos, vtos);
1022 __ movdbl(daddress(n), xmm0);
1023 }
1024
1025 void TemplateTable::astore(int n) {
1026 transition(vtos, vtos);
1027 __ pop_ptr(rax);
1028 __ movptr(aaddress(n), rax);
1029 }
1030
1031 void TemplateTable::pop() {
1032 transition(vtos, vtos);
1033 __ addptr(rsp, Interpreter::stackElementSize);
1034 }
1035
1036 void TemplateTable::pop2() {
1037 transition(vtos, vtos);
1038 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1039 }
1040
1041 void TemplateTable::dup() {
1042 transition(vtos, vtos);
1043 __ load_ptr(0, rax);
1044 __ push_ptr(rax);
1045 // stack: ..., a, a
1046 }
1047
1048 void TemplateTable::dup_x1() {
1049 transition(vtos, vtos);
1050 // stack: ..., a, b
1051 __ load_ptr( 0, rax); // load b
1052 __ load_ptr( 1, rcx); // load a
1053 __ store_ptr(1, rax); // store b
1054 __ store_ptr(0, rcx); // store a
1055 __ push_ptr(rax); // push b
1056 // stack: ..., b, a, b
1057 }
1058
1059 void TemplateTable::dup_x2() {
1060 transition(vtos, vtos);
1061 // stack: ..., a, b, c
1062 __ load_ptr( 0, rax); // load c
1063 __ load_ptr( 2, rcx); // load a
1064 __ store_ptr(2, rax); // store c in a
1065 __ push_ptr(rax); // push c
1066 // stack: ..., c, b, c, c
1067 __ load_ptr( 2, rax); // load b
1068 __ store_ptr(2, rcx); // store a in b
1069 // stack: ..., c, a, c, c
1070 __ store_ptr(1, rax); // store b in c
1071 // stack: ..., c, a, b, c
1072 }
1073
1074 void TemplateTable::dup2() {
1075 transition(vtos, vtos);
1076 // stack: ..., a, b
1077 __ load_ptr(1, rax); // load a
1078 __ push_ptr(rax); // push a
1079 __ load_ptr(1, rax); // load b
1080 __ push_ptr(rax); // push b
1081 // stack: ..., a, b, a, b
1082 }
1083
1084 void TemplateTable::dup2_x1() {
1085 transition(vtos, vtos);
1086 // stack: ..., a, b, c
1087 __ load_ptr( 0, rcx); // load c
1088 __ load_ptr( 1, rax); // load b
1089 __ push_ptr(rax); // push b
1090 __ push_ptr(rcx); // push c
1091 // stack: ..., a, b, c, b, c
1092 __ store_ptr(3, rcx); // store c in b
1093 // stack: ..., a, c, c, b, c
1094 __ load_ptr( 4, rcx); // load a
1095 __ store_ptr(2, rcx); // store a in 2nd c
1096 // stack: ..., a, c, a, b, c
1097 __ store_ptr(4, rax); // store b in a
1098 // stack: ..., b, c, a, b, c
1099 }
1100
1101 void TemplateTable::dup2_x2() {
1102 transition(vtos, vtos);
1103 // stack: ..., a, b, c, d
1104 __ load_ptr( 0, rcx); // load d
1105 __ load_ptr( 1, rax); // load c
1106 __ push_ptr(rax); // push c
1107 __ push_ptr(rcx); // push d
1108 // stack: ..., a, b, c, d, c, d
1109 __ load_ptr( 4, rax); // load b
1110 __ store_ptr(2, rax); // store b in d
1111 __ store_ptr(4, rcx); // store d in b
1112 // stack: ..., a, d, c, b, c, d
1113 __ load_ptr( 5, rcx); // load a
1114 __ load_ptr( 3, rax); // load c
1115 __ store_ptr(3, rcx); // store a in c
1116 __ store_ptr(5, rax); // store c in a
1117 // stack: ..., c, d, a, b, c, d
1118 }
1119
1120 void TemplateTable::swap() {
1121 transition(vtos, vtos);
1122 // stack: ..., a, b
1123 __ load_ptr( 1, rcx); // load a
1124 __ load_ptr( 0, rax); // load b
1125 __ store_ptr(0, rcx); // store a in b
1126 __ store_ptr(1, rax); // store b in a
1127 // stack: ..., b, a
1128 }
1129
1130 void TemplateTable::iop2(Operation op) {
1131 transition(itos, itos);
1132 switch (op) {
1133 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1134 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1135 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1136 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1137 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1138 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1139 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1140 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1141 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1142 default : ShouldNotReachHere();
1143 }
1144 }
1145
1146 void TemplateTable::lop2(Operation op) {
1147 transition(ltos, ltos);
1148 switch (op) {
1149 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1150 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1151 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1152 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1153 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1154 default : ShouldNotReachHere();
1155 }
1156 }
1157
1158 void TemplateTable::idiv() {
1159 transition(itos, itos);
1160 __ movl(rcx, rax);
1161 __ pop_i(rax);
1162 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1163 // they are not equal, one could do a normal division (no correction
1164 // needed), which may speed up this implementation for the common case.
1165 // (see also JVM spec., p.243 & p.271)
1166 __ corrected_idivl(rcx);
1167 }
1168
1169 void TemplateTable::irem() {
1170 transition(itos, itos);
1171 __ movl(rcx, rax);
1172 __ pop_i(rax);
1173 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1174 // they are not equal, one could do a normal division (no correction
1175 // needed), which may speed up this implementation for the common case.
1176 // (see also JVM spec., p.243 & p.271)
1177 __ corrected_idivl(rcx);
1178 __ movl(rax, rdx);
1179 }
1180
1181 void TemplateTable::lmul() {
1182 transition(ltos, ltos);
1183 __ pop_l(rdx);
1184 __ imulq(rax, rdx);
1185 }
1186
1187 void TemplateTable::ldiv() {
1188 transition(ltos, ltos);
1189 __ mov(rcx, rax);
1190 __ pop_l(rax);
1191 // generate explicit div0 check
1192 __ testq(rcx, rcx);
1193 __ jump_cc(Assembler::zero,
1194 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1195 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1196 // they are not equal, one could do a normal division (no correction
1197 // needed), which may speed up this implementation for the common case.
1198 // (see also JVM spec., p.243 & p.271)
1199 __ corrected_idivq(rcx); // kills rbx
1200 }
1201
1202 void TemplateTable::lrem() {
1203 transition(ltos, ltos);
1204 __ mov(rcx, rax);
1205 __ pop_l(rax);
1206 __ testq(rcx, rcx);
1207 __ jump_cc(Assembler::zero,
1208 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1209 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1210 // they are not equal, one could do a normal division (no correction
1211 // needed), which may speed up this implementation for the common case.
1212 // (see also JVM spec., p.243 & p.271)
1213 __ corrected_idivq(rcx); // kills rbx
1214 __ mov(rax, rdx);
1215 }
1216
1217 void TemplateTable::lshl() {
1218 transition(itos, ltos);
1219 __ movl(rcx, rax); // get shift count
1220 __ pop_l(rax); // get shift value
1221 __ shlq(rax);
1222 }
1223
1224 void TemplateTable::lshr() {
1225 transition(itos, ltos);
1226 __ movl(rcx, rax); // get shift count
1227 __ pop_l(rax); // get shift value
1228 __ sarq(rax);
1229 }
1230
1231 void TemplateTable::lushr() {
1232 transition(itos, ltos);
1233 __ movl(rcx, rax); // get shift count
1234 __ pop_l(rax); // get shift value
1235 __ shrq(rax);
1236 }
1237
1238 void TemplateTable::fop2(Operation op) {
1239 transition(ftos, ftos);
1240 switch (op) {
1241 case add:
1242 __ addss(xmm0, at_rsp());
1243 __ addptr(rsp, Interpreter::stackElementSize);
1244 break;
1245 case sub:
1246 __ movflt(xmm1, xmm0);
1247 __ pop_f(xmm0);
1248 __ subss(xmm0, xmm1);
1249 break;
1250 case mul:
1251 __ mulss(xmm0, at_rsp());
1252 __ addptr(rsp, Interpreter::stackElementSize);
1253 break;
1254 case div:
1255 __ movflt(xmm1, xmm0);
1256 __ pop_f(xmm0);
1257 __ divss(xmm0, xmm1);
1258 break;
1259 case rem:
1260 __ movflt(xmm1, xmm0);
1261 __ pop_f(xmm0);
1262 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1263 break;
1264 default:
1265 ShouldNotReachHere();
1266 break;
1267 }
1268 }
1269
1270 void TemplateTable::dop2(Operation op) {
1271 transition(dtos, dtos);
1272 switch (op) {
1273 case add:
1274 __ addsd(xmm0, at_rsp());
1275 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1276 break;
1277 case sub:
1278 __ movdbl(xmm1, xmm0);
1279 __ pop_d(xmm0);
1280 __ subsd(xmm0, xmm1);
1281 break;
1282 case mul:
1283 __ mulsd(xmm0, at_rsp());
1284 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1285 break;
1286 case div:
1287 __ movdbl(xmm1, xmm0);
1288 __ pop_d(xmm0);
1289 __ divsd(xmm0, xmm1);
1290 break;
1291 case rem:
1292 __ movdbl(xmm1, xmm0);
1293 __ pop_d(xmm0);
1294 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1295 break;
1296 default:
1297 ShouldNotReachHere();
1298 break;
1299 }
1300 }
1301
1302 void TemplateTable::ineg() {
1303 transition(itos, itos);
1304 __ negl(rax);
1305 }
1306
1307 void TemplateTable::lneg() {
1308 transition(ltos, ltos);
1309 __ negq(rax);
1310 }
1311
1312 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1313 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1314 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1315 // of 128-bits operands for SSE instructions.
1316 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1317 // Store the value to a 128-bits operand.
1318 operand[0] = lo;
1319 operand[1] = hi;
1320 return operand;
1321 }
1322
1323 // Buffer for 128-bits masks used by SSE instructions.
1324 static jlong float_signflip_pool[2*2];
1325 static jlong double_signflip_pool[2*2];
1326
1327 void TemplateTable::fneg() {
1328 transition(ftos, ftos);
1329 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], 0x8000000080000000, 0x8000000080000000);
1330 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1331 }
1332
1333 void TemplateTable::dneg() {
1334 transition(dtos, dtos);
1335 static jlong *double_signflip = double_quadword(&double_signflip_pool[1], 0x8000000000000000, 0x8000000000000000);
1336 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1337 }
1338
1339 void TemplateTable::iinc() {
1340 transition(vtos, vtos);
1341 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1342 locals_index(rbx);
1343 __ addl(iaddress(rbx), rdx);
1344 }
1345
1346 void TemplateTable::wide_iinc() {
1347 transition(vtos, vtos);
1348 __ movl(rdx, at_bcp(4)); // get constant
1349 locals_index_wide(rbx);
1350 __ bswapl(rdx); // swap bytes & sign-extend constant
1351 __ sarl(rdx, 16);
1352 __ addl(iaddress(rbx), rdx);
1353 // Note: should probably use only one movl to get both
1354 // the index and the constant -> fix this
1355 }
1356
1357 void TemplateTable::convert() {
1358 // Checking
1359 #ifdef ASSERT
1360 {
1361 TosState tos_in = ilgl;
1362 TosState tos_out = ilgl;
1363 switch (bytecode()) {
1364 case Bytecodes::_i2l: // fall through
1365 case Bytecodes::_i2f: // fall through
1366 case Bytecodes::_i2d: // fall through
1367 case Bytecodes::_i2b: // fall through
1368 case Bytecodes::_i2c: // fall through
1369 case Bytecodes::_i2s: tos_in = itos; break;
1370 case Bytecodes::_l2i: // fall through
1371 case Bytecodes::_l2f: // fall through
1372 case Bytecodes::_l2d: tos_in = ltos; break;
1373 case Bytecodes::_f2i: // fall through
1374 case Bytecodes::_f2l: // fall through
1375 case Bytecodes::_f2d: tos_in = ftos; break;
1376 case Bytecodes::_d2i: // fall through
1377 case Bytecodes::_d2l: // fall through
1378 case Bytecodes::_d2f: tos_in = dtos; break;
1379 default : ShouldNotReachHere();
1380 }
1381 switch (bytecode()) {
1382 case Bytecodes::_l2i: // fall through
1383 case Bytecodes::_f2i: // fall through
1384 case Bytecodes::_d2i: // fall through
1385 case Bytecodes::_i2b: // fall through
1386 case Bytecodes::_i2c: // fall through
1387 case Bytecodes::_i2s: tos_out = itos; break;
1388 case Bytecodes::_i2l: // fall through
1389 case Bytecodes::_f2l: // fall through
1390 case Bytecodes::_d2l: tos_out = ltos; break;
1391 case Bytecodes::_i2f: // fall through
1392 case Bytecodes::_l2f: // fall through
1393 case Bytecodes::_d2f: tos_out = ftos; break;
1394 case Bytecodes::_i2d: // fall through
1395 case Bytecodes::_l2d: // fall through
1396 case Bytecodes::_f2d: tos_out = dtos; break;
1397 default : ShouldNotReachHere();
1398 }
1399 transition(tos_in, tos_out);
1400 }
1401 #endif // ASSERT
1402
1403 static const int64_t is_nan = 0x8000000000000000L;
1404
1405 // Conversion
1406 switch (bytecode()) {
1407 case Bytecodes::_i2l:
1408 __ movslq(rax, rax);
1409 break;
1410 case Bytecodes::_i2f:
1411 __ cvtsi2ssl(xmm0, rax);
1412 break;
1413 case Bytecodes::_i2d:
1414 __ cvtsi2sdl(xmm0, rax);
1415 break;
1416 case Bytecodes::_i2b:
1417 __ movsbl(rax, rax);
1418 break;
1419 case Bytecodes::_i2c:
1420 __ movzwl(rax, rax);
1421 break;
1422 case Bytecodes::_i2s:
1423 __ movswl(rax, rax);
1424 break;
1425 case Bytecodes::_l2i:
1426 __ movl(rax, rax);
1427 break;
1428 case Bytecodes::_l2f:
1429 __ cvtsi2ssq(xmm0, rax);
1430 break;
1431 case Bytecodes::_l2d:
1432 __ cvtsi2sdq(xmm0, rax);
1433 break;
1434 case Bytecodes::_f2i:
1435 {
1436 Label L;
1437 __ cvttss2sil(rax, xmm0);
1438 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1439 __ jcc(Assembler::notEqual, L);
1440 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1441 __ bind(L);
1442 }
1443 break;
1444 case Bytecodes::_f2l:
1445 {
1446 Label L;
1447 __ cvttss2siq(rax, xmm0);
1448 // NaN or overflow/underflow?
1449 __ cmp64(rax, ExternalAddress((address) &is_nan));
1450 __ jcc(Assembler::notEqual, L);
1451 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1452 __ bind(L);
1453 }
1454 break;
1455 case Bytecodes::_f2d:
1456 __ cvtss2sd(xmm0, xmm0);
1457 break;
1458 case Bytecodes::_d2i:
1459 {
1460 Label L;
1461 __ cvttsd2sil(rax, xmm0);
1462 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1463 __ jcc(Assembler::notEqual, L);
1464 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1465 __ bind(L);
1466 }
1467 break;
1468 case Bytecodes::_d2l:
1469 {
1470 Label L;
1471 __ cvttsd2siq(rax, xmm0);
1472 // NaN or overflow/underflow?
1473 __ cmp64(rax, ExternalAddress((address) &is_nan));
1474 __ jcc(Assembler::notEqual, L);
1475 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1476 __ bind(L);
1477 }
1478 break;
1479 case Bytecodes::_d2f:
1480 __ cvtsd2ss(xmm0, xmm0);
1481 break;
1482 default:
1483 ShouldNotReachHere();
1484 }
1485 }
1486
1487 void TemplateTable::lcmp() {
1488 transition(ltos, itos);
1489 Label done;
1490 __ pop_l(rdx);
1491 __ cmpq(rdx, rax);
1492 __ movl(rax, -1);
1493 __ jccb(Assembler::less, done);
1494 __ setb(Assembler::notEqual, rax);
1495 __ movzbl(rax, rax);
1496 __ bind(done);
1497 }
1498
1499 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1500 Label done;
1501 if (is_float) {
1502 // XXX get rid of pop here, use ... reg, mem32
1503 __ pop_f(xmm1);
1504 __ ucomiss(xmm1, xmm0);
1505 } else {
1506 // XXX get rid of pop here, use ... reg, mem64
1507 __ pop_d(xmm1);
1508 __ ucomisd(xmm1, xmm0);
1509 }
1510 if (unordered_result < 0) {
1511 __ movl(rax, -1);
1512 __ jccb(Assembler::parity, done);
1513 __ jccb(Assembler::below, done);
1514 __ setb(Assembler::notEqual, rdx);
1515 __ movzbl(rax, rdx);
1516 } else {
1517 __ movl(rax, 1);
1518 __ jccb(Assembler::parity, done);
1519 __ jccb(Assembler::above, done);
1520 __ movl(rax, 0);
1521 __ jccb(Assembler::equal, done);
1522 __ decrementl(rax);
1523 }
1524 __ bind(done);
1525 }
1526
1527 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1528 __ get_method(rcx); // rcx holds method
1529 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
1530 // holds bumped taken count
1531
1532 const ByteSize be_offset = methodOopDesc::backedge_counter_offset() +
1533 InvocationCounter::counter_offset();
1534 const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() +
1535 InvocationCounter::counter_offset();
1536 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1537
1538 // Load up edx with the branch displacement
1539 __ movl(rdx, at_bcp(1));
1540 __ bswapl(rdx);
1541
1542 if (!is_wide) {
1543 __ sarl(rdx, 16);
1544 }
1545 __ movl2ptr(rdx, rdx);
1546
1547 // Handle all the JSR stuff here, then exit.
1548 // It's much shorter and cleaner than intermingling with the non-JSR
1549 // normal-branch stuff occurring below.
1550 if (is_jsr) {
1551 // Pre-load the next target bytecode into rbx
1552 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1, 0));
1553
1554 // compute return address as bci in rax
1555 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
1556 in_bytes(constMethodOopDesc::codes_offset())));
1557 __ subptr(rax, Address(rcx, methodOopDesc::const_offset()));
1558 // Adjust the bcp in r13 by the displacement in rdx
1559 __ addptr(r13, rdx);
1560 // jsr returns atos that is not an oop
1561 __ push_i(rax);
1562 __ dispatch_only(vtos);
1563 return;
1564 }
1565
1566 // Normal (non-jsr) branch handling
1567
1568 // Adjust the bcp in r13 by the displacement in rdx
1569 __ addptr(r13, rdx);
1570
1571 assert(UseLoopCounter || !UseOnStackReplacement,
1572 "on-stack-replacement requires loop counters");
1573 Label backedge_counter_overflow;
1574 Label profile_method;
1575 Label dispatch;
1576 if (UseLoopCounter) {
1577 // increment backedge counter for backward branches
1578 // rax: MDO
1579 // ebx: MDO bumped taken-count
1580 // rcx: method
1581 // rdx: target offset
1582 // r13: target bcp
1583 // r14: locals pointer
1584 __ testl(rdx, rdx); // check if forward or backward branch
1585 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1586 if (TieredCompilation) {
1587 Label no_mdo;
1588 int increment = InvocationCounter::count_increment;
1589 int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1590 if (ProfileInterpreter) {
1591 // Are we profiling?
1592 __ movptr(rbx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1593 __ testptr(rbx, rbx);
1594 __ jccb(Assembler::zero, no_mdo);
1595 // Increment the MDO backedge counter
1596 const Address mdo_backedge_counter(rbx, in_bytes(methodDataOopDesc::backedge_counter_offset()) +
1597 in_bytes(InvocationCounter::counter_offset()));
1598 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1599 rax, false, Assembler::zero, &backedge_counter_overflow);
1600 __ jmp(dispatch);
1601 }
1602 __ bind(no_mdo);
1603 // Increment backedge counter in methodOop
1604 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
1605 rax, false, Assembler::zero, &backedge_counter_overflow);
1606 } else {
1607 // increment counter
1608 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1609 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1610 __ movl(Address(rcx, be_offset), rax); // store counter
1611
1612 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1613 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1614 __ addl(rax, Address(rcx, be_offset)); // add both counters
1615
1616 if (ProfileInterpreter) {
1617 // Test to see if we should create a method data oop
1618 __ cmp32(rax,
1619 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1620 __ jcc(Assembler::less, dispatch);
1621
1622 // if no method data exists, go to profile method
1623 __ test_method_data_pointer(rax, profile_method);
1624
1625 if (UseOnStackReplacement) {
1626 // check for overflow against ebx which is the MDO taken count
1627 __ cmp32(rbx,
1628 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1629 __ jcc(Assembler::below, dispatch);
1630
1631 // When ProfileInterpreter is on, the backedge_count comes
1632 // from the methodDataOop, which value does not get reset on
1633 // the call to frequency_counter_overflow(). To avoid
1634 // excessive calls to the overflow routine while the method is
1635 // being compiled, add a second test to make sure the overflow
1636 // function is called only once every overflow_frequency.
1637 const int overflow_frequency = 1024;
1638 __ andl(rbx, overflow_frequency - 1);
1639 __ jcc(Assembler::zero, backedge_counter_overflow);
1640
1641 }
1642 } else {
1643 if (UseOnStackReplacement) {
1644 // check for overflow against eax, which is the sum of the
1645 // counters
1646 __ cmp32(rax,
1647 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1648 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1649
1650 }
1651 }
1652 }
1653 __ bind(dispatch);
1654 }
1655
1656 // Pre-load the next target bytecode into rbx
1657 __ load_unsigned_byte(rbx, Address(r13, 0));
1658
1659 // continue with the bytecode @ target
1660 // eax: return bci for jsr's, unused otherwise
1661 // ebx: target bytecode
1662 // r13: target bcp
1663 __ dispatch_only(vtos);
1664
1665 if (UseLoopCounter) {
1666 if (ProfileInterpreter) {
1667 // Out-of-line code to allocate method data oop.
1668 __ bind(profile_method);
1669 __ call_VM(noreg,
1670 CAST_FROM_FN_PTR(address,
1671 InterpreterRuntime::profile_method), r13);
1672 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1673 __ movptr(rcx, Address(rbp, method_offset));
1674 __ movptr(rcx, Address(rcx,
1675 in_bytes(methodOopDesc::method_data_offset())));
1676 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
1677 rcx);
1678 __ test_method_data_pointer(rcx, dispatch);
1679 // offset non-null mdp by MDO::data_offset() + IR::profile_method()
1680 __ addptr(rcx, in_bytes(methodDataOopDesc::data_offset()));
1681 __ addptr(rcx, rax);
1682 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
1683 rcx);
1684 __ jmp(dispatch);
1685 }
1686
1687 if (UseOnStackReplacement) {
1688 // invocation counter overflow
1689 __ bind(backedge_counter_overflow);
1690 __ negptr(rdx);
1691 __ addptr(rdx, r13); // branch bcp
1692 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1693 __ call_VM(noreg,
1694 CAST_FROM_FN_PTR(address,
1695 InterpreterRuntime::frequency_counter_overflow),
1696 rdx);
1697 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1698
1699 // rax: osr nmethod (osr ok) or NULL (osr not possible)
1700 // ebx: target bytecode
1701 // rdx: scratch
1702 // r14: locals pointer
1703 // r13: bcp
1704 __ testptr(rax, rax); // test result
1705 __ jcc(Assembler::zero, dispatch); // no osr if null
1706 // nmethod may have been invalidated (VM may block upon call_VM return)
1707 __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1708 __ cmpl(rcx, InvalidOSREntryBci);
1709 __ jcc(Assembler::equal, dispatch);
1710
1711 // We have the address of an on stack replacement routine in eax
1712 // We need to prepare to execute the OSR method. First we must
1713 // migrate the locals and monitors off of the stack.
1714
1715 __ mov(r13, rax); // save the nmethod
1716
1717 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1718
1719 // eax is OSR buffer, move it to expected parameter location
1720 __ mov(j_rarg0, rax);
1721
1722 // We use j_rarg definitions here so that registers don't conflict as parameter
1723 // registers change across platforms as we are in the midst of a calling
1724 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
1725
1726 const Register retaddr = j_rarg2;
1727 const Register sender_sp = j_rarg1;
1728
1729 // pop the interpreter frame
1730 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1731 __ leave(); // remove frame anchor
1732 __ pop(retaddr); // get return address
1733 __ mov(rsp, sender_sp); // set sp to sender sp
1734 // Ensure compiled code always sees stack at proper alignment
1735 __ andptr(rsp, -(StackAlignmentInBytes));
1736
1737 // unlike x86 we need no specialized return from compiled code
1738 // to the interpreter or the call stub.
1739
1740 // push the return address
1741 __ push(retaddr);
1742
1743 // and begin the OSR nmethod
1744 __ jmp(Address(r13, nmethod::osr_entry_point_offset()));
1745 }
1746 }
1747 }
1748
1749
1750 void TemplateTable::if_0cmp(Condition cc) {
1751 transition(itos, vtos);
1752 // assume branch is more often taken than not (loops use backward branches)
1753 Label not_taken;
1754 __ testl(rax, rax);
1755 __ jcc(j_not(cc), not_taken);
1756 branch(false, false);
1757 __ bind(not_taken);
1758 __ profile_not_taken_branch(rax);
1759 }
1760
1761 void TemplateTable::if_icmp(Condition cc) {
1762 transition(itos, vtos);
1763 // assume branch is more often taken than not (loops use backward branches)
1764 Label not_taken;
1765 __ pop_i(rdx);
1766 __ cmpl(rdx, rax);
1767 __ jcc(j_not(cc), not_taken);
1768 branch(false, false);
1769 __ bind(not_taken);
1770 __ profile_not_taken_branch(rax);
1771 }
1772
1773 void TemplateTable::if_nullcmp(Condition cc) {
1774 transition(atos, vtos);
1775 // assume branch is more often taken than not (loops use backward branches)
1776 Label not_taken;
1777 __ testptr(rax, rax);
1778 __ jcc(j_not(cc), not_taken);
1779 branch(false, false);
1780 __ bind(not_taken);
1781 __ profile_not_taken_branch(rax);
1782 }
1783
1784 void TemplateTable::if_acmp(Condition cc) {
1785 transition(atos, vtos);
1786 // assume branch is more often taken than not (loops use backward branches)
1787 Label not_taken;
1788 __ pop_ptr(rdx);
1789 __ cmpptr(rdx, rax);
1790 __ jcc(j_not(cc), not_taken);
1791 branch(false, false);
1792 __ bind(not_taken);
1793 __ profile_not_taken_branch(rax);
1794 }
1795
1796 void TemplateTable::ret() {
1797 transition(vtos, vtos);
1798 locals_index(rbx);
1799 __ movslq(rbx, iaddress(rbx)); // get return bci, compute return bcp
1800 __ profile_ret(rbx, rcx);
1801 __ get_method(rax);
1802 __ movptr(r13, Address(rax, methodOopDesc::const_offset()));
1803 __ lea(r13, Address(r13, rbx, Address::times_1,
1804 constMethodOopDesc::codes_offset()));
1805 __ dispatch_next(vtos);
1806 }
1807
1808 void TemplateTable::wide_ret() {
1809 transition(vtos, vtos);
1810 locals_index_wide(rbx);
1811 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
1812 __ profile_ret(rbx, rcx);
1813 __ get_method(rax);
1814 __ movptr(r13, Address(rax, methodOopDesc::const_offset()));
1815 __ lea(r13, Address(r13, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1816 __ dispatch_next(vtos);
1817 }
1818
1819 void TemplateTable::tableswitch() {
1820 Label default_case, continue_execution;
1821 transition(itos, vtos);
1822 // align r13
1823 __ lea(rbx, at_bcp(BytesPerInt));
1824 __ andptr(rbx, -BytesPerInt);
1825 // load lo & hi
1826 __ movl(rcx, Address(rbx, BytesPerInt));
1827 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
1828 __ bswapl(rcx);
1829 __ bswapl(rdx);
1830 // check against lo & hi
1831 __ cmpl(rax, rcx);
1832 __ jcc(Assembler::less, default_case);
1833 __ cmpl(rax, rdx);
1834 __ jcc(Assembler::greater, default_case);
1835 // lookup dispatch offset
1836 __ subl(rax, rcx);
1837 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1838 __ profile_switch_case(rax, rbx, rcx);
1839 // continue execution
1840 __ bind(continue_execution);
1841 __ bswapl(rdx);
1842 __ movl2ptr(rdx, rdx);
1843 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1844 __ addptr(r13, rdx);
1845 __ dispatch_only(vtos);
1846 // handle default
1847 __ bind(default_case);
1848 __ profile_switch_default(rax);
1849 __ movl(rdx, Address(rbx, 0));
1850 __ jmp(continue_execution);
1851 }
1852
1853 void TemplateTable::lookupswitch() {
1854 transition(itos, itos);
1855 __ stop("lookupswitch bytecode should have been rewritten");
1856 }
1857
1858 void TemplateTable::fast_linearswitch() {
1859 transition(itos, vtos);
1860 Label loop_entry, loop, found, continue_execution;
1861 // bswap rax so we can avoid bswapping the table entries
1862 __ bswapl(rax);
1863 // align r13
1864 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
1865 // this instruction (change offsets
1866 // below)
1867 __ andptr(rbx, -BytesPerInt);
1868 // set counter
1869 __ movl(rcx, Address(rbx, BytesPerInt));
1870 __ bswapl(rcx);
1871 __ jmpb(loop_entry);
1872 // table search
1873 __ bind(loop);
1874 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
1875 __ jcc(Assembler::equal, found);
1876 __ bind(loop_entry);
1877 __ decrementl(rcx);
1878 __ jcc(Assembler::greaterEqual, loop);
1879 // default case
1880 __ profile_switch_default(rax);
1881 __ movl(rdx, Address(rbx, 0));
1882 __ jmp(continue_execution);
1883 // entry found -> get offset
1884 __ bind(found);
1885 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
1886 __ profile_switch_case(rcx, rax, rbx);
1887 // continue execution
1888 __ bind(continue_execution);
1889 __ bswapl(rdx);
1890 __ movl2ptr(rdx, rdx);
1891 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1892 __ addptr(r13, rdx);
1893 __ dispatch_only(vtos);
1894 }
1895
1896 void TemplateTable::fast_binaryswitch() {
1897 transition(itos, vtos);
1898 // Implementation using the following core algorithm:
1899 //
1900 // int binary_search(int key, LookupswitchPair* array, int n) {
1901 // // Binary search according to "Methodik des Programmierens" by
1902 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1903 // int i = 0;
1904 // int j = n;
1905 // while (i+1 < j) {
1906 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1907 // // with Q: for all i: 0 <= i < n: key < a[i]
1908 // // where a stands for the array and assuming that the (inexisting)
1909 // // element a[n] is infinitely big.
1910 // int h = (i + j) >> 1;
1911 // // i < h < j
1912 // if (key < array[h].fast_match()) {
1913 // j = h;
1914 // } else {
1915 // i = h;
1916 // }
1917 // }
1918 // // R: a[i] <= key < a[i+1] or Q
1919 // // (i.e., if key is within array, i is the correct index)
1920 // return i;
1921 // }
1922
1923 // Register allocation
1924 const Register key = rax; // already set (tosca)
1925 const Register array = rbx;
1926 const Register i = rcx;
1927 const Register j = rdx;
1928 const Register h = rdi;
1929 const Register temp = rsi;
1930
1931 // Find array start
1932 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
1933 // get rid of this
1934 // instruction (change
1935 // offsets below)
1936 __ andptr(array, -BytesPerInt);
1937
1938 // Initialize i & j
1939 __ xorl(i, i); // i = 0;
1940 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
1941
1942 // Convert j into native byteordering
1943 __ bswapl(j);
1944
1945 // And start
1946 Label entry;
1947 __ jmp(entry);
1948
1949 // binary search loop
1950 {
1951 Label loop;
1952 __ bind(loop);
1953 // int h = (i + j) >> 1;
1954 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1955 __ sarl(h, 1); // h = (i + j) >> 1;
1956 // if (key < array[h].fast_match()) {
1957 // j = h;
1958 // } else {
1959 // i = h;
1960 // }
1961 // Convert array[h].match to native byte-ordering before compare
1962 __ movl(temp, Address(array, h, Address::times_8));
1963 __ bswapl(temp);
1964 __ cmpl(key, temp);
1965 // j = h if (key < array[h].fast_match())
1966 __ cmovl(Assembler::less, j, h);
1967 // i = h if (key >= array[h].fast_match())
1968 __ cmovl(Assembler::greaterEqual, i, h);
1969 // while (i+1 < j)
1970 __ bind(entry);
1971 __ leal(h, Address(i, 1)); // i+1
1972 __ cmpl(h, j); // i+1 < j
1973 __ jcc(Assembler::less, loop);
1974 }
1975
1976 // end of binary search, result index is i (must check again!)
1977 Label default_case;
1978 // Convert array[i].match to native byte-ordering before compare
1979 __ movl(temp, Address(array, i, Address::times_8));
1980 __ bswapl(temp);
1981 __ cmpl(key, temp);
1982 __ jcc(Assembler::notEqual, default_case);
1983
1984 // entry found -> j = offset
1985 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
1986 __ profile_switch_case(i, key, array);
1987 __ bswapl(j);
1988 __ movl2ptr(j, j);
1989 __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
1990 __ addptr(r13, j);
1991 __ dispatch_only(vtos);
1992
1993 // default case -> j = default offset
1994 __ bind(default_case);
1995 __ profile_switch_default(i);
1996 __ movl(j, Address(array, -2 * BytesPerInt));
1997 __ bswapl(j);
1998 __ movl2ptr(j, j);
1999 __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
2000 __ addptr(r13, j);
2001 __ dispatch_only(vtos);
2002 }
2003
2004
2005 void TemplateTable::_return(TosState state) {
2006 transition(state, state);
2007 assert(_desc->calls_vm(),
2008 "inconsistent calls_vm information"); // call in remove_activation
2009
2010 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2011 assert(state == vtos, "only valid state");
2012 __ movptr(c_rarg1, aaddress(0));
2013 __ load_klass(rdi, c_rarg1);
2014 __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
2015 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2016 Label skip_register_finalizer;
2017 __ jcc(Assembler::zero, skip_register_finalizer);
2018
2019 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2020
2021 __ bind(skip_register_finalizer);
2022 }
2023
2024 __ remove_activation(state, r13);
2025 __ jmp(r13);
2026 }
2027
2028 // ----------------------------------------------------------------------------
2029 // Volatile variables demand their effects be made known to all CPU's
2030 // in order. Store buffers on most chips allow reads & writes to
2031 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2032 // without some kind of memory barrier (i.e., it's not sufficient that
2033 // the interpreter does not reorder volatile references, the hardware
2034 // also must not reorder them).
2035 //
2036 // According to the new Java Memory Model (JMM):
2037 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2038 // writes act as aquire & release, so:
2039 // (2) A read cannot let unrelated NON-volatile memory refs that
2040 // happen after the read float up to before the read. It's OK for
2041 // non-volatile memory refs that happen before the volatile read to
2042 // float down below it.
2043 // (3) Similar a volatile write cannot let unrelated NON-volatile
2044 // memory refs that happen BEFORE the write float down to after the
2045 // write. It's OK for non-volatile memory refs that happen after the
2046 // volatile write to float up before it.
2047 //
2048 // We only put in barriers around volatile refs (they are expensive),
2049 // not _between_ memory refs (that would require us to track the
2050 // flavor of the previous memory refs). Requirements (2) and (3)
2051 // require some barriers before volatile stores and after volatile
2052 // loads. These nearly cover requirement (1) but miss the
2053 // volatile-store-volatile-load case. This final case is placed after
2054 // volatile-stores although it could just as well go before
2055 // volatile-loads.
2056 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits
2057 order_constraint) {
2058 // Helper function to insert a is-volatile test and memory barrier
2059 if (os::is_MP()) { // Not needed on single CPU
2060 __ membar(order_constraint);
2061 }
2062 }
2063
2064 void TemplateTable::resolve_cache_and_index(int byte_no,
2065 Register result,
2066 Register Rcache,
2067 Register index,
2068 size_t index_size) {
2069 const Register temp = rbx;
2070 assert_different_registers(result, Rcache, index, temp);
2071
2072 Label resolved;
2073 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2074 if (byte_no == f1_oop) {
2075 // We are resolved if the f1 field contains a non-null object (CallSite, etc.)
2076 // This kind of CP cache entry does not need to match the flags byte, because
2077 // there is a 1-1 relation between bytecode type and CP entry type.
2078 assert(result != noreg, ""); //else do cmpptr(Address(...), (int32_t) NULL_WORD)
2079 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2080 __ testptr(result, result);
2081 __ jcc(Assembler::notEqual, resolved);
2082 } else {
2083 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2084 assert(result == noreg, ""); //else change code for setting result
2085 const int shift_count = (1 + byte_no) * BitsPerByte;
2086 __ movl(temp, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
2087 __ shrl(temp, shift_count);
2088 // have we resolved this bytecode?
2089 __ andl(temp, 0xFF);
2090 __ cmpl(temp, (int) bytecode());
2091 __ jcc(Assembler::equal, resolved);
2092 }
2093
2094 // resolve first time through
2095 address entry;
2096 switch (bytecode()) {
2097 case Bytecodes::_getstatic:
2098 case Bytecodes::_putstatic:
2099 case Bytecodes::_getfield:
2100 case Bytecodes::_putfield:
2101 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
2102 break;
2103 case Bytecodes::_invokevirtual:
2104 case Bytecodes::_invokespecial:
2105 case Bytecodes::_invokestatic:
2106 case Bytecodes::_invokeinterface:
2107 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
2108 break;
2109 case Bytecodes::_invokedynamic:
2110 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
2111 break;
2112 case Bytecodes::_fast_aldc:
2113 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
2114 break;
2115 case Bytecodes::_fast_aldc_w:
2116 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
2117 break;
2118 default:
2119 ShouldNotReachHere();
2120 break;
2121 }
2122 __ movl(temp, (int) bytecode());
2123 __ call_VM(noreg, entry, temp);
2124
2125 // Update registers with resolved info
2126 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2127 if (result != noreg)
2128 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2129 __ bind(resolved);
2130 }
2131
2132 // The Rcache and index registers must be set before call
2133 void TemplateTable::load_field_cp_cache_entry(Register obj,
2134 Register cache,
2135 Register index,
2136 Register off,
2137 Register flags,
2138 bool is_static = false) {
2139 assert_different_registers(cache, index, flags, off);
2140
2141 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2142 // Field offset
2143 __ movptr(off, Address(cache, index, Address::times_8,
2144 in_bytes(cp_base_offset +
2145 ConstantPoolCacheEntry::f2_offset())));
2146 // Flags
2147 __ movl(flags, Address(cache, index, Address::times_8,
2148 in_bytes(cp_base_offset +
2149 ConstantPoolCacheEntry::flags_offset())));
2150
2151 // klass overwrite register
2152 if (is_static) {
2153 __ movptr(obj, Address(cache, index, Address::times_8,
2154 in_bytes(cp_base_offset +
2155 ConstantPoolCacheEntry::f1_offset())));
2156 }
2157 }
2158
2159 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2160 Register method,
2161 Register itable_index,
2162 Register flags,
2163 bool is_invokevirtual,
2164 bool is_invokevfinal, /*unused*/
2165 bool is_invokedynamic) {
2166 // setup registers
2167 const Register cache = rcx;
2168 const Register index = rdx;
2169 assert_different_registers(method, flags);
2170 assert_different_registers(method, cache, index);
2171 assert_different_registers(itable_index, flags);
2172 assert_different_registers(itable_index, cache, index);
2173 // determine constant pool cache field offsets
2174 const int method_offset = in_bytes(
2175 constantPoolCacheOopDesc::base_offset() +
2176 (is_invokevirtual
2177 ? ConstantPoolCacheEntry::f2_offset()
2178 : ConstantPoolCacheEntry::f1_offset()));
2179 const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2180 ConstantPoolCacheEntry::flags_offset());
2181 // access constant pool cache fields
2182 const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2183 ConstantPoolCacheEntry::f2_offset());
2184
2185 if (byte_no == f1_oop) {
2186 // Resolved f1_oop goes directly into 'method' register.
2187 assert(is_invokedynamic, "");
2188 resolve_cache_and_index(byte_no, method, cache, index, sizeof(u4));
2189 } else {
2190 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2191 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2192 }
2193 if (itable_index != noreg) {
2194 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2195 }
2196 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2197 }
2198
2199
2200 // The registers cache and index expected to be set before call.
2201 // Correct values of the cache and index registers are preserved.
2202 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2203 bool is_static, bool has_tos) {
2204 // do the JVMTI work here to avoid disturbing the register state below
2205 // We use c_rarg registers here because we want to use the register used in
2206 // the call to the VM
2207 if (JvmtiExport::can_post_field_access()) {
2208 // Check to see if a field access watch has been set before we
2209 // take the time to call into the VM.
2210 Label L1;
2211 assert_different_registers(cache, index, rax);
2212 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2213 __ testl(rax, rax);
2214 __ jcc(Assembler::zero, L1);
2215
2216 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2217
2218 // cache entry pointer
2219 __ addptr(c_rarg2, in_bytes(constantPoolCacheOopDesc::base_offset()));
2220 __ shll(c_rarg3, LogBytesPerWord);
2221 __ addptr(c_rarg2, c_rarg3);
2222 if (is_static) {
2223 __ xorl(c_rarg1, c_rarg1); // NULL object reference
2224 } else {
2225 __ movptr(c_rarg1, at_tos()); // get object pointer without popping it
2226 __ verify_oop(c_rarg1);
2227 }
2228 // c_rarg1: object pointer or NULL
2229 // c_rarg2: cache entry pointer
2230 // c_rarg3: jvalue object on the stack
2231 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2232 InterpreterRuntime::post_field_access),
2233 c_rarg1, c_rarg2, c_rarg3);
2234 __ get_cache_and_index_at_bcp(cache, index, 1);
2235 __ bind(L1);
2236 }
2237 }
2238
2239 void TemplateTable::pop_and_check_object(Register r) {
2240 __ pop_ptr(r);
2241 __ null_check(r); // for field access must check obj.
2242 __ verify_oop(r);
2243 }
2244
2245 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2246 transition(vtos, vtos);
2247
2248 const Register cache = rcx;
2249 const Register index = rdx;
2250 const Register obj = c_rarg3;
2251 const Register off = rbx;
2252 const Register flags = rax;
2253 const Register bc = c_rarg3; // uses same reg as obj, so don't mix them
2254
2255 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2256 jvmti_post_field_access(cache, index, is_static, false);
2257 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2258
2259 if (!is_static) {
2260 // obj is on the stack
2261 pop_and_check_object(obj);
2262 }
2263
2264 const Address field(obj, off, Address::times_1);
2265
2266 Label Done, notByte, notInt, notShort, notChar,
2267 notLong, notFloat, notObj, notDouble;
2268
2269 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2270 assert(btos == 0, "change code, btos != 0");
2271
2272 __ andl(flags, 0x0F);
2273 __ jcc(Assembler::notZero, notByte);
2274 // btos
2275 __ load_signed_byte(rax, field);
2276 __ push(btos);
2277 // Rewrite bytecode to be faster
2278 if (!is_static) {
2279 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2280 }
2281 __ jmp(Done);
2282
2283 __ bind(notByte);
2284 __ cmpl(flags, atos);
2285 __ jcc(Assembler::notEqual, notObj);
2286 // atos
2287 __ load_heap_oop(rax, field);
2288 __ push(atos);
2289 if (!is_static) {
2290 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2291 }
2292 __ jmp(Done);
2293
2294 __ bind(notObj);
2295 __ cmpl(flags, itos);
2296 __ jcc(Assembler::notEqual, notInt);
2297 // itos
2298 __ movl(rax, field);
2299 __ push(itos);
2300 // Rewrite bytecode to be faster
2301 if (!is_static) {
2302 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2303 }
2304 __ jmp(Done);
2305
2306 __ bind(notInt);
2307 __ cmpl(flags, ctos);
2308 __ jcc(Assembler::notEqual, notChar);
2309 // ctos
2310 __ load_unsigned_short(rax, field);
2311 __ push(ctos);
2312 // Rewrite bytecode to be faster
2313 if (!is_static) {
2314 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2315 }
2316 __ jmp(Done);
2317
2318 __ bind(notChar);
2319 __ cmpl(flags, stos);
2320 __ jcc(Assembler::notEqual, notShort);
2321 // stos
2322 __ load_signed_short(rax, field);
2323 __ push(stos);
2324 // Rewrite bytecode to be faster
2325 if (!is_static) {
2326 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2327 }
2328 __ jmp(Done);
2329
2330 __ bind(notShort);
2331 __ cmpl(flags, ltos);
2332 __ jcc(Assembler::notEqual, notLong);
2333 // ltos
2334 __ movq(rax, field);
2335 __ push(ltos);
2336 // Rewrite bytecode to be faster
2337 if (!is_static) {
2338 patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx);
2339 }
2340 __ jmp(Done);
2341
2342 __ bind(notLong);
2343 __ cmpl(flags, ftos);
2344 __ jcc(Assembler::notEqual, notFloat);
2345 // ftos
2346 __ movflt(xmm0, field);
2347 __ push(ftos);
2348 // Rewrite bytecode to be faster
2349 if (!is_static) {
2350 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2351 }
2352 __ jmp(Done);
2353
2354 __ bind(notFloat);
2355 #ifdef ASSERT
2356 __ cmpl(flags, dtos);
2357 __ jcc(Assembler::notEqual, notDouble);
2358 #endif
2359 // dtos
2360 __ movdbl(xmm0, field);
2361 __ push(dtos);
2362 // Rewrite bytecode to be faster
2363 if (!is_static) {
2364 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2365 }
2366 #ifdef ASSERT
2367 __ jmp(Done);
2368
2369 __ bind(notDouble);
2370 __ stop("Bad state");
2371 #endif
2372
2373 __ bind(Done);
2374 // [jk] not needed currently
2375 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2376 // Assembler::LoadStore));
2377 }
2378
2379
2380 void TemplateTable::getfield(int byte_no) {
2381 getfield_or_static(byte_no, false);
2382 }
2383
2384 void TemplateTable::getstatic(int byte_no) {
2385 getfield_or_static(byte_no, true);
2386 }
2387
2388 // The registers cache and index expected to be set before call.
2389 // The function may destroy various registers, just not the cache and index registers.
2390 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2391 transition(vtos, vtos);
2392
2393 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2394
2395 if (JvmtiExport::can_post_field_modification()) {
2396 // Check to see if a field modification watch has been set before
2397 // we take the time to call into the VM.
2398 Label L1;
2399 assert_different_registers(cache, index, rax);
2400 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2401 __ testl(rax, rax);
2402 __ jcc(Assembler::zero, L1);
2403
2404 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2405
2406 if (is_static) {
2407 // Life is simple. Null out the object pointer.
2408 __ xorl(c_rarg1, c_rarg1);
2409 } else {
2410 // Life is harder. The stack holds the value on top, followed by
2411 // the object. We don't know the size of the value, though; it
2412 // could be one or two words depending on its type. As a result,
2413 // we must find the type to determine where the object is.
2414 __ movl(c_rarg3, Address(c_rarg2, rscratch1,
2415 Address::times_8,
2416 in_bytes(cp_base_offset +
2417 ConstantPoolCacheEntry::flags_offset())));
2418 __ shrl(c_rarg3, ConstantPoolCacheEntry::tosBits);
2419 // Make sure we don't need to mask rcx for tosBits after the
2420 // above shift
2421 ConstantPoolCacheEntry::verify_tosBits();
2422 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2423 __ cmpl(c_rarg3, ltos);
2424 __ cmovptr(Assembler::equal,
2425 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2426 __ cmpl(c_rarg3, dtos);
2427 __ cmovptr(Assembler::equal,
2428 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2429 }
2430 // cache entry pointer
2431 __ addptr(c_rarg2, in_bytes(cp_base_offset));
2432 __ shll(rscratch1, LogBytesPerWord);
2433 __ addptr(c_rarg2, rscratch1);
2434 // object (tos)
2435 __ mov(c_rarg3, rsp);
2436 // c_rarg1: object pointer set up above (NULL if static)
2437 // c_rarg2: cache entry pointer
2438 // c_rarg3: jvalue object on the stack
2439 __ call_VM(noreg,
2440 CAST_FROM_FN_PTR(address,
2441 InterpreterRuntime::post_field_modification),
2442 c_rarg1, c_rarg2, c_rarg3);
2443 __ get_cache_and_index_at_bcp(cache, index, 1);
2444 __ bind(L1);
2445 }
2446 }
2447
2448 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2449 transition(vtos, vtos);
2450
2451 const Register cache = rcx;
2452 const Register index = rdx;
2453 const Register obj = rcx;
2454 const Register off = rbx;
2455 const Register flags = rax;
2456 const Register bc = c_rarg3;
2457
2458 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2459 jvmti_post_field_mod(cache, index, is_static);
2460 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2461
2462 // [jk] not needed currently
2463 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2464 // Assembler::StoreStore));
2465
2466 Label notVolatile, Done;
2467 __ movl(rdx, flags);
2468 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2469 __ andl(rdx, 0x1);
2470
2471 // field address
2472 const Address field(obj, off, Address::times_1);
2473
2474 Label notByte, notInt, notShort, notChar,
2475 notLong, notFloat, notObj, notDouble;
2476
2477 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2478
2479 assert(btos == 0, "change code, btos != 0");
2480 __ andl(flags, 0x0f);
2481 __ jcc(Assembler::notZero, notByte);
2482 // btos
2483 __ pop(btos);
2484 if (!is_static) pop_and_check_object(obj);
2485 __ movb(field, rax);
2486 if (!is_static) {
2487 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx);
2488 }
2489 __ jmp(Done);
2490
2491 __ bind(notByte);
2492 __ cmpl(flags, atos);
2493 __ jcc(Assembler::notEqual, notObj);
2494 // atos
2495 __ pop(atos);
2496 if (!is_static) pop_and_check_object(obj);
2497
2498 // Store into the field
2499 do_oop_store(_masm, field, rax, _bs->kind(), false);
2500
2501 if (!is_static) {
2502 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx);
2503 }
2504 __ jmp(Done);
2505
2506 __ bind(notObj);
2507 __ cmpl(flags, itos);
2508 __ jcc(Assembler::notEqual, notInt);
2509 // itos
2510 __ pop(itos);
2511 if (!is_static) pop_and_check_object(obj);
2512 __ movl(field, rax);
2513 if (!is_static) {
2514 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx);
2515 }
2516 __ jmp(Done);
2517
2518 __ bind(notInt);
2519 __ cmpl(flags, ctos);
2520 __ jcc(Assembler::notEqual, notChar);
2521 // ctos
2522 __ pop(ctos);
2523 if (!is_static) pop_and_check_object(obj);
2524 __ movw(field, rax);
2525 if (!is_static) {
2526 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx);
2527 }
2528 __ jmp(Done);
2529
2530 __ bind(notChar);
2531 __ cmpl(flags, stos);
2532 __ jcc(Assembler::notEqual, notShort);
2533 // stos
2534 __ pop(stos);
2535 if (!is_static) pop_and_check_object(obj);
2536 __ movw(field, rax);
2537 if (!is_static) {
2538 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx);
2539 }
2540 __ jmp(Done);
2541
2542 __ bind(notShort);
2543 __ cmpl(flags, ltos);
2544 __ jcc(Assembler::notEqual, notLong);
2545 // ltos
2546 __ pop(ltos);
2547 if (!is_static) pop_and_check_object(obj);
2548 __ movq(field, rax);
2549 if (!is_static) {
2550 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx);
2551 }
2552 __ jmp(Done);
2553
2554 __ bind(notLong);
2555 __ cmpl(flags, ftos);
2556 __ jcc(Assembler::notEqual, notFloat);
2557 // ftos
2558 __ pop(ftos);
2559 if (!is_static) pop_and_check_object(obj);
2560 __ movflt(field, xmm0);
2561 if (!is_static) {
2562 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx);
2563 }
2564 __ jmp(Done);
2565
2566 __ bind(notFloat);
2567 #ifdef ASSERT
2568 __ cmpl(flags, dtos);
2569 __ jcc(Assembler::notEqual, notDouble);
2570 #endif
2571 // dtos
2572 __ pop(dtos);
2573 if (!is_static) pop_and_check_object(obj);
2574 __ movdbl(field, xmm0);
2575 if (!is_static) {
2576 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx);
2577 }
2578
2579 #ifdef ASSERT
2580 __ jmp(Done);
2581
2582 __ bind(notDouble);
2583 __ stop("Bad state");
2584 #endif
2585
2586 __ bind(Done);
2587 // Check for volatile store
2588 __ testl(rdx, rdx);
2589 __ jcc(Assembler::zero, notVolatile);
2590 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2591 Assembler::StoreStore));
2592
2593 __ bind(notVolatile);
2594 }
2595
2596 void TemplateTable::putfield(int byte_no) {
2597 putfield_or_static(byte_no, false);
2598 }
2599
2600 void TemplateTable::putstatic(int byte_no) {
2601 putfield_or_static(byte_no, true);
2602 }
2603
2604 void TemplateTable::jvmti_post_fast_field_mod() {
2605 if (JvmtiExport::can_post_field_modification()) {
2606 // Check to see if a field modification watch has been set before
2607 // we take the time to call into the VM.
2608 Label L2;
2609 __ mov32(c_rarg3, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2610 __ testl(c_rarg3, c_rarg3);
2611 __ jcc(Assembler::zero, L2);
2612 __ pop_ptr(rbx); // copy the object pointer from tos
2613 __ verify_oop(rbx);
2614 __ push_ptr(rbx); // put the object pointer back on tos
2615 __ subptr(rsp, sizeof(jvalue)); // add space for a jvalue object
2616 __ mov(c_rarg3, rsp);
2617 const Address field(c_rarg3, 0);
2618
2619 switch (bytecode()) { // load values into the jvalue object
2620 case Bytecodes::_fast_aputfield: __ movq(field, rax); break;
2621 case Bytecodes::_fast_lputfield: __ movq(field, rax); break;
2622 case Bytecodes::_fast_iputfield: __ movl(field, rax); break;
2623 case Bytecodes::_fast_bputfield: __ movb(field, rax); break;
2624 case Bytecodes::_fast_sputfield: // fall through
2625 case Bytecodes::_fast_cputfield: __ movw(field, rax); break;
2626 case Bytecodes::_fast_fputfield: __ movflt(field, xmm0); break;
2627 case Bytecodes::_fast_dputfield: __ movdbl(field, xmm0); break;
2628 default:
2629 ShouldNotReachHere();
2630 }
2631
2632 // Save rax because call_VM() will clobber it, then use it for
2633 // JVMTI purposes
2634 __ push(rax);
2635 // access constant pool cache entry
2636 __ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1);
2637 __ verify_oop(rbx);
2638 // rbx: object pointer copied above
2639 // c_rarg2: cache entry pointer
2640 // c_rarg3: jvalue object on the stack
2641 __ call_VM(noreg,
2642 CAST_FROM_FN_PTR(address,
2643 InterpreterRuntime::post_field_modification),
2644 rbx, c_rarg2, c_rarg3);
2645 __ pop(rax); // restore lower value
2646 __ addptr(rsp, sizeof(jvalue)); // release jvalue object space
2647 __ bind(L2);
2648 }
2649 }
2650
2651 void TemplateTable::fast_storefield(TosState state) {
2652 transition(state, vtos);
2653
2654 ByteSize base = constantPoolCacheOopDesc::base_offset();
2655
2656 jvmti_post_fast_field_mod();
2657
2658 // access constant pool cache
2659 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2660
2661 // test for volatile with rdx
2662 __ movl(rdx, Address(rcx, rbx, Address::times_8,
2663 in_bytes(base +
2664 ConstantPoolCacheEntry::flags_offset())));
2665
2666 // replace index with field offset from cache entry
2667 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2668 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2669
2670 // [jk] not needed currently
2671 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2672 // Assembler::StoreStore));
2673
2674 Label notVolatile;
2675 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2676 __ andl(rdx, 0x1);
2677
2678 // Get object from stack
2679 pop_and_check_object(rcx);
2680
2681 // field address
2682 const Address field(rcx, rbx, Address::times_1);
2683
2684 // access field
2685 switch (bytecode()) {
2686 case Bytecodes::_fast_aputfield:
2687 do_oop_store(_masm, field, rax, _bs->kind(), false);
2688 break;
2689 case Bytecodes::_fast_lputfield:
2690 __ movq(field, rax);
2691 break;
2692 case Bytecodes::_fast_iputfield:
2693 __ movl(field, rax);
2694 break;
2695 case Bytecodes::_fast_bputfield:
2696 __ movb(field, rax);
2697 break;
2698 case Bytecodes::_fast_sputfield:
2699 // fall through
2700 case Bytecodes::_fast_cputfield:
2701 __ movw(field, rax);
2702 break;
2703 case Bytecodes::_fast_fputfield:
2704 __ movflt(field, xmm0);
2705 break;
2706 case Bytecodes::_fast_dputfield:
2707 __ movdbl(field, xmm0);
2708 break;
2709 default:
2710 ShouldNotReachHere();
2711 }
2712
2713 // Check for volatile store
2714 __ testl(rdx, rdx);
2715 __ jcc(Assembler::zero, notVolatile);
2716 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2717 Assembler::StoreStore));
2718 __ bind(notVolatile);
2719 }
2720
2721
2722 void TemplateTable::fast_accessfield(TosState state) {
2723 transition(atos, state);
2724
2725 // Do the JVMTI work here to avoid disturbing the register state below
2726 if (JvmtiExport::can_post_field_access()) {
2727 // Check to see if a field access watch has been set before we
2728 // take the time to call into the VM.
2729 Label L1;
2730 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2731 __ testl(rcx, rcx);
2732 __ jcc(Assembler::zero, L1);
2733 // access constant pool cache entry
2734 __ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1);
2735 __ verify_oop(rax);
2736 __ mov(r12, rax); // save object pointer before call_VM() clobbers it
2737 __ mov(c_rarg1, rax);
2738 // c_rarg1: object pointer copied above
2739 // c_rarg2: cache entry pointer
2740 __ call_VM(noreg,
2741 CAST_FROM_FN_PTR(address,
2742 InterpreterRuntime::post_field_access),
2743 c_rarg1, c_rarg2);
2744 __ mov(rax, r12); // restore object pointer
2745 __ reinit_heapbase();
2746 __ bind(L1);
2747 }
2748
2749 // access constant pool cache
2750 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2751 // replace index with field offset from cache entry
2752 // [jk] not needed currently
2753 // if (os::is_MP()) {
2754 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
2755 // in_bytes(constantPoolCacheOopDesc::base_offset() +
2756 // ConstantPoolCacheEntry::flags_offset())));
2757 // __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2758 // __ andl(rdx, 0x1);
2759 // }
2760 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2761 in_bytes(constantPoolCacheOopDesc::base_offset() +
2762 ConstantPoolCacheEntry::f2_offset())));
2763
2764 // rax: object
2765 __ verify_oop(rax);
2766 __ null_check(rax);
2767 Address field(rax, rbx, Address::times_1);
2768
2769 // access field
2770 switch (bytecode()) {
2771 case Bytecodes::_fast_agetfield:
2772 __ load_heap_oop(rax, field);
2773 __ verify_oop(rax);
2774 break;
2775 case Bytecodes::_fast_lgetfield:
2776 __ movq(rax, field);
2777 break;
2778 case Bytecodes::_fast_igetfield:
2779 __ movl(rax, field);
2780 break;
2781 case Bytecodes::_fast_bgetfield:
2782 __ movsbl(rax, field);
2783 break;
2784 case Bytecodes::_fast_sgetfield:
2785 __ load_signed_short(rax, field);
2786 break;
2787 case Bytecodes::_fast_cgetfield:
2788 __ load_unsigned_short(rax, field);
2789 break;
2790 case Bytecodes::_fast_fgetfield:
2791 __ movflt(xmm0, field);
2792 break;
2793 case Bytecodes::_fast_dgetfield:
2794 __ movdbl(xmm0, field);
2795 break;
2796 default:
2797 ShouldNotReachHere();
2798 }
2799 // [jk] not needed currently
2800 // if (os::is_MP()) {
2801 // Label notVolatile;
2802 // __ testl(rdx, rdx);
2803 // __ jcc(Assembler::zero, notVolatile);
2804 // __ membar(Assembler::LoadLoad);
2805 // __ bind(notVolatile);
2806 //};
2807 }
2808
2809 void TemplateTable::fast_xaccess(TosState state) {
2810 transition(vtos, state);
2811
2812 // get receiver
2813 __ movptr(rax, aaddress(0));
2814 // access constant pool cache
2815 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2816 __ movptr(rbx,
2817 Address(rcx, rdx, Address::times_8,
2818 in_bytes(constantPoolCacheOopDesc::base_offset() +
2819 ConstantPoolCacheEntry::f2_offset())));
2820 // make sure exception is reported in correct bcp range (getfield is
2821 // next instruction)
2822 __ increment(r13);
2823 __ null_check(rax);
2824 switch (state) {
2825 case itos:
2826 __ movl(rax, Address(rax, rbx, Address::times_1));
2827 break;
2828 case atos:
2829 __ load_heap_oop(rax, Address(rax, rbx, Address::times_1));
2830 __ verify_oop(rax);
2831 break;
2832 case ftos:
2833 __ movflt(xmm0, Address(rax, rbx, Address::times_1));
2834 break;
2835 default:
2836 ShouldNotReachHere();
2837 }
2838
2839 // [jk] not needed currently
2840 // if (os::is_MP()) {
2841 // Label notVolatile;
2842 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
2843 // in_bytes(constantPoolCacheOopDesc::base_offset() +
2844 // ConstantPoolCacheEntry::flags_offset())));
2845 // __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2846 // __ testl(rdx, 0x1);
2847 // __ jcc(Assembler::zero, notVolatile);
2848 // __ membar(Assembler::LoadLoad);
2849 // __ bind(notVolatile);
2850 // }
2851
2852 __ decrement(r13);
2853 }
2854
2855
2856
2857 //-----------------------------------------------------------------------------
2858 // Calls
2859
2860 void TemplateTable::count_calls(Register method, Register temp) {
2861 // implemented elsewhere
2862 ShouldNotReachHere();
2863 }
2864
2865 void TemplateTable::prepare_invoke(Register method, Register index, int byte_no) {
2866 // determine flags
2867 Bytecodes::Code code = bytecode();
2868 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2869 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
2870 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2871 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2872 const bool load_receiver = (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic);
2873 const bool receiver_null_check = is_invokespecial;
2874 const bool save_flags = is_invokeinterface || is_invokevirtual;
2875 // setup registers & access constant pool cache
2876 const Register recv = rcx;
2877 const Register flags = rdx;
2878 assert_different_registers(method, index, recv, flags);
2879
2880 // save 'interpreter return address'
2881 __ save_bcp();
2882
2883 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2884
2885 // load receiver if needed (note: no return address pushed yet)
2886 if (load_receiver) {
2887 assert(!is_invokedynamic, "");
2888 __ movl(recv, flags);
2889 __ andl(recv, 0xFF);
2890 Address recv_addr(rsp, recv, Address::times_8, -Interpreter::expr_offset_in_bytes(1));
2891 __ movptr(recv, recv_addr);
2892 __ verify_oop(recv);
2893 }
2894
2895 // do null check if needed
2896 if (receiver_null_check) {
2897 __ null_check(recv);
2898 }
2899
2900 if (save_flags) {
2901 __ movl(r13, flags);
2902 }
2903
2904 // compute return type
2905 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2906 // Make sure we don't need to mask flags for tosBits after the above shift
2907 ConstantPoolCacheEntry::verify_tosBits();
2908 // load return address
2909 {
2910 address table_addr;
2911 if (is_invokeinterface || is_invokedynamic)
2912 table_addr = (address)Interpreter::return_5_addrs_by_index_table();
2913 else
2914 table_addr = (address)Interpreter::return_3_addrs_by_index_table();
2915 ExternalAddress table(table_addr);
2916 __ lea(rscratch1, table);
2917 __ movptr(flags, Address(rscratch1, flags, Address::times_ptr));
2918 }
2919
2920 // push return address
2921 __ push(flags);
2922
2923 // Restore flag field from the constant pool cache, and restore esi
2924 // for later null checks. r13 is the bytecode pointer
2925 if (save_flags) {
2926 __ movl(flags, r13);
2927 __ restore_bcp();
2928 }
2929 }
2930
2931
2932 void TemplateTable::invokevirtual_helper(Register index,
2933 Register recv,
2934 Register flags) {
2935 // Uses temporary registers rax, rdx
2936 assert_different_registers(index, recv, rax, rdx);
2937
2938 // Test for an invoke of a final method
2939 Label notFinal;
2940 __ movl(rax, flags);
2941 __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
2942 __ jcc(Assembler::zero, notFinal);
2943
2944 const Register method = index; // method must be rbx
2945 assert(method == rbx,
2946 "methodOop must be rbx for interpreter calling convention");
2947
2948 // do the call - the index is actually the method to call
2949 __ verify_oop(method);
2950
2951 // It's final, need a null check here!
2952 __ null_check(recv);
2953
2954 // profile this call
2955 __ profile_final_call(rax);
2956
2957 __ jump_from_interpreted(method, rax);
2958
2959 __ bind(notFinal);
2960
2961 // get receiver klass
2962 __ null_check(recv, oopDesc::klass_offset_in_bytes());
2963 __ load_klass(rax, recv);
2964
2965 __ verify_oop(rax);
2966
2967 // profile this call
2968 __ profile_virtual_call(rax, r14, rdx);
2969
2970 // get target methodOop & entry point
2971 const int base = instanceKlass::vtable_start_offset() * wordSize;
2972 assert(vtableEntry::size() * wordSize == 8,
2973 "adjust the scaling in the code below");
2974 __ movptr(method, Address(rax, index,
2975 Address::times_8,
2976 base + vtableEntry::method_offset_in_bytes()));
2977 __ movptr(rdx, Address(method, methodOopDesc::interpreter_entry_offset()));
2978 __ jump_from_interpreted(method, rdx);
2979 }
2980
2981
2982 void TemplateTable::invokevirtual(int byte_no) {
2983 transition(vtos, vtos);
2984 assert(byte_no == f2_byte, "use this argument");
2985 prepare_invoke(rbx, noreg, byte_no);
2986
2987 // rbx: index
2988 // rcx: receiver
2989 // rdx: flags
2990
2991 invokevirtual_helper(rbx, rcx, rdx);
2992 }
2993
2994
2995 void TemplateTable::invokespecial(int byte_no) {
2996 transition(vtos, vtos);
2997 assert(byte_no == f1_byte, "use this argument");
2998 prepare_invoke(rbx, noreg, byte_no);
2999 // do the call
3000 __ verify_oop(rbx);
3001 __ profile_call(rax);
3002 __ jump_from_interpreted(rbx, rax);
3003 }
3004
3005
3006 void TemplateTable::invokestatic(int byte_no) {
3007 transition(vtos, vtos);
3008 assert(byte_no == f1_byte, "use this argument");
3009 prepare_invoke(rbx, noreg, byte_no);
3010 // do the call
3011 __ verify_oop(rbx);
3012 __ profile_call(rax);
3013 __ jump_from_interpreted(rbx, rax);
3014 }
3015
3016 void TemplateTable::fast_invokevfinal(int byte_no) {
3017 transition(vtos, vtos);
3018 assert(byte_no == f2_byte, "use this argument");
3019 __ stop("fast_invokevfinal not used on amd64");
3020 }
3021
3022 void TemplateTable::invokeinterface(int byte_no) {
3023 transition(vtos, vtos);
3024 assert(byte_no == f1_byte, "use this argument");
3025 prepare_invoke(rax, rbx, byte_no);
3026
3027 // rax: Interface
3028 // rbx: index
3029 // rcx: receiver
3030 // rdx: flags
3031
3032 // Special case of invokeinterface called for virtual method of
3033 // java.lang.Object. See cpCacheOop.cpp for details.
3034 // This code isn't produced by javac, but could be produced by
3035 // another compliant java compiler.
3036 Label notMethod;
3037 __ movl(r14, rdx);
3038 __ andl(r14, (1 << ConstantPoolCacheEntry::methodInterface));
3039 __ jcc(Assembler::zero, notMethod);
3040
3041 invokevirtual_helper(rbx, rcx, rdx);
3042 __ bind(notMethod);
3043
3044 // Get receiver klass into rdx - also a null check
3045 __ restore_locals(); // restore r14
3046 __ load_klass(rdx, rcx);
3047 __ verify_oop(rdx);
3048
3049 // profile this call
3050 __ profile_virtual_call(rdx, r13, r14);
3051
3052 Label no_such_interface, no_such_method;
3053
3054 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3055 rdx, rax, rbx,
3056 // outputs: method, scan temp. reg
3057 rbx, r13,
3058 no_such_interface);
3059
3060 // rbx,: methodOop to call
3061 // rcx: receiver
3062 // Check for abstract method error
3063 // Note: This should be done more efficiently via a throw_abstract_method_error
3064 // interpreter entry point and a conditional jump to it in case of a null
3065 // method.
3066 __ testptr(rbx, rbx);
3067 __ jcc(Assembler::zero, no_such_method);
3068
3069 // do the call
3070 // rcx: receiver
3071 // rbx,: methodOop
3072 __ jump_from_interpreted(rbx, rdx);
3073 __ should_not_reach_here();
3074
3075 // exception handling code follows...
3076 // note: must restore interpreter registers to canonical
3077 // state for exception handling to work correctly!
3078
3079 __ bind(no_such_method);
3080 // throw exception
3081 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3082 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3083 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3084 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3085 // the call_VM checks for exception, so we should never return here.
3086 __ should_not_reach_here();
3087
3088 __ bind(no_such_interface);
3089 // throw exception
3090 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3091 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3092 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3093 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3094 InterpreterRuntime::throw_IncompatibleClassChangeError));
3095 // the call_VM checks for exception, so we should never return here.
3096 __ should_not_reach_here();
3097 return;
3098 }
3099
3100 void TemplateTable::invokedynamic(int byte_no) {
3101 transition(vtos, vtos);
3102 assert(byte_no == f1_oop, "use this argument");
3103
3104 if (!EnableInvokeDynamic) {
3105 // We should not encounter this bytecode if !EnableInvokeDynamic.
3106 // The verifier will stop it. However, if we get past the verifier,
3107 // this will stop the thread in a reasonable way, without crashing the JVM.
3108 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3109 InterpreterRuntime::throw_IncompatibleClassChangeError));
3110 // the call_VM checks for exception, so we should never return here.
3111 __ should_not_reach_here();
3112 return;
3113 }
3114
3115 assert(byte_no == f1_oop, "use this argument");
3116 prepare_invoke(rax, rbx, byte_no);
3117
3118 // rax: CallSite object (f1)
3119 // rbx: unused (f2)
3120 // rcx: receiver address
3121 // rdx: flags (unused)
3122
3123 Register rax_callsite = rax;
3124 Register rcx_method_handle = rcx;
3125
3126 if (ProfileInterpreter) {
3127 // %%% should make a type profile for any invokedynamic that takes a ref argument
3128 // profile this call
3129 __ profile_call(r13);
3130 }
3131
3132 __ load_heap_oop(rcx_method_handle, Address(rax_callsite, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx)));
3133 __ null_check(rcx_method_handle);
3134 __ prepare_to_jump_from_interpreted();
3135 __ jump_to_method_handle_entry(rcx_method_handle, rdx);
3136 }
3137
3138
3139 //-----------------------------------------------------------------------------
3140 // Allocation
3141
3142 void TemplateTable::_new() {
3143 transition(vtos, atos);
3144 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3145 Label slow_case;
3146 Label done;
3147 Label initialize_header;
3148 Label initialize_object; // including clearing the fields
3149 Label allocate_shared;
3150
3151 __ get_cpool_and_tags(rsi, rax);
3152 // Make sure the class we're about to instantiate has been resolved.
3153 // This is done before loading instanceKlass to be consistent with the order
3154 // how Constant Pool is updated (see constantPoolOopDesc::klass_at_put)
3155 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
3156 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset),
3157 JVM_CONSTANT_Class);
3158 __ jcc(Assembler::notEqual, slow_case);
3159
3160 // get instanceKlass
3161 __ movptr(rsi, Address(rsi, rdx,
3162 Address::times_8, sizeof(constantPoolOopDesc)));
3163
3164 // make sure klass is initialized & doesn't have finalizer
3165 // make sure klass is fully initialized
3166 __ cmpl(Address(rsi,
3167 instanceKlass::init_state_offset_in_bytes() +
3168 sizeof(oopDesc)),
3169 instanceKlass::fully_initialized);
3170 __ jcc(Assembler::notEqual, slow_case);
3171
3172 // get instance_size in instanceKlass (scaled to a count of bytes)
3173 __ movl(rdx,
3174 Address(rsi,
3175 Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
3176 // test to see if it has a finalizer or is malformed in some way
3177 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3178 __ jcc(Assembler::notZero, slow_case);
3179
3180 // Allocate the instance
3181 // 1) Try to allocate in the TLAB
3182 // 2) if fail and the object is large allocate in the shared Eden
3183 // 3) if the above fails (or is not applicable), go to a slow case
3184 // (creates a new TLAB, etc.)
3185
3186 const bool allow_shared_alloc =
3187 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3188
3189 if (UseTLAB) {
3190 __ movptr(rax, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
3191 __ lea(rbx, Address(rax, rdx, Address::times_1));
3192 __ cmpptr(rbx, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
3193 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3194 __ movptr(Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3195 if (ZeroTLAB) {
3196 // the fields have been already cleared
3197 __ jmp(initialize_header);
3198 } else {
3199 // initialize both the header and fields
3200 __ jmp(initialize_object);
3201 }
3202 }
3203
3204 // Allocation in the shared Eden, if allowed.
3205 //
3206 // rdx: instance size in bytes
3207 if (allow_shared_alloc) {
3208 __ bind(allocate_shared);
3209
3210 ExternalAddress top((address)Universe::heap()->top_addr());
3211 ExternalAddress end((address)Universe::heap()->end_addr());
3212
3213 const Register RtopAddr = rscratch1;
3214 const Register RendAddr = rscratch2;
3215
3216 __ lea(RtopAddr, top);
3217 __ lea(RendAddr, end);
3218 __ movptr(rax, Address(RtopAddr, 0));
3219
3220 // For retries rax gets set by cmpxchgq
3221 Label retry;
3222 __ bind(retry);
3223 __ lea(rbx, Address(rax, rdx, Address::times_1));
3224 __ cmpptr(rbx, Address(RendAddr, 0));
3225 __ jcc(Assembler::above, slow_case);
3226
3227 // Compare rax with the top addr, and if still equal, store the new
3228 // top addr in rbx at the address of the top addr pointer. Sets ZF if was
3229 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3230 //
3231 // rax: object begin
3232 // rbx: object end
3233 // rdx: instance size in bytes
3234 if (os::is_MP()) {
3235 __ lock();
3236 }
3237 __ cmpxchgptr(rbx, Address(RtopAddr, 0));
3238
3239 // if someone beat us on the allocation, try again, otherwise continue
3240 __ jcc(Assembler::notEqual, retry);
3241 }
3242
3243 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3244 // The object is initialized before the header. If the object size is
3245 // zero, go directly to the header initialization.
3246 __ bind(initialize_object);
3247 __ decrementl(rdx, sizeof(oopDesc));
3248 __ jcc(Assembler::zero, initialize_header);
3249
3250 // Initialize object fields
3251 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3252 __ shrl(rdx, LogBytesPerLong); // divide by oopSize to simplify the loop
3253 {
3254 Label loop;
3255 __ bind(loop);
3256 __ movq(Address(rax, rdx, Address::times_8,
3257 sizeof(oopDesc) - oopSize),
3258 rcx);
3259 __ decrementl(rdx);
3260 __ jcc(Assembler::notZero, loop);
3261 }
3262
3263 // initialize object header only.
3264 __ bind(initialize_header);
3265 if (UseBiasedLocking) {
3266 __ movptr(rscratch1, Address(rsi, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
3267 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()), rscratch1);
3268 } else {
3269 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()),
3270 (intptr_t) markOopDesc::prototype()); // header (address 0x1)
3271 }
3272 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3273 __ store_klass_gap(rax, rcx); // zero klass gap for compressed oops
3274 __ store_klass(rax, rsi); // store klass last
3275
3276 {
3277 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3278 // Trigger dtrace event for fastpath
3279 __ push(atos); // save the return value
3280 __ call_VM_leaf(
3281 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3282 __ pop(atos); // restore the return value
3283
3284 }
3285 __ jmp(done);
3286 }
3287
3288
3289 // slow case
3290 __ bind(slow_case);
3291 __ get_constant_pool(c_rarg1);
3292 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3293 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3294 __ verify_oop(rax);
3295
3296 // continue
3297 __ bind(done);
3298 }
3299
3300 void TemplateTable::newarray() {
3301 transition(itos, atos);
3302 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3303 __ movl(c_rarg2, rax);
3304 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3305 c_rarg1, c_rarg2);
3306 }
3307
3308 void TemplateTable::anewarray() {
3309 transition(itos, atos);
3310 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3311 __ get_constant_pool(c_rarg1);
3312 __ movl(c_rarg3, rax);
3313 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3314 c_rarg1, c_rarg2, c_rarg3);
3315 }
3316
3317 void TemplateTable::arraylength() {
3318 transition(atos, itos);
3319 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3320 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3321 }
3322
3323 void TemplateTable::checkcast() {
3324 transition(atos, atos);
3325 Label done, is_null, ok_is_subtype, quicked, resolved;
3326 __ testptr(rax, rax); // object is in rax
3327 __ jcc(Assembler::zero, is_null);
3328
3329 // Get cpool & tags index
3330 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3331 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3332 // See if bytecode has already been quicked
3333 __ cmpb(Address(rdx, rbx,
3334 Address::times_1,
3335 typeArrayOopDesc::header_size(T_BYTE) * wordSize),
3336 JVM_CONSTANT_Class);
3337 __ jcc(Assembler::equal, quicked);
3338 __ push(atos); // save receiver for result, and for GC
3339 __ mov(r12, rcx); // save rcx XXX
3340 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3341 __ movq(rcx, r12); // restore rcx XXX
3342 __ reinit_heapbase();
3343 __ pop_ptr(rdx); // restore receiver
3344 __ jmpb(resolved);
3345
3346 // Get superklass in rax and subklass in rbx
3347 __ bind(quicked);
3348 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
3349 __ movptr(rax, Address(rcx, rbx,
3350 Address::times_8, sizeof(constantPoolOopDesc)));
3351
3352 __ bind(resolved);
3353 __ load_klass(rbx, rdx);
3354
3355 // Generate subtype check. Blows rcx, rdi. Object in rdx.
3356 // Superklass in rax. Subklass in rbx.
3357 __ gen_subtype_check(rbx, ok_is_subtype);
3358
3359 // Come here on failure
3360 __ push_ptr(rdx);
3361 // object is at TOS
3362 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3363
3364 // Come here on success
3365 __ bind(ok_is_subtype);
3366 __ mov(rax, rdx); // Restore object in rdx
3367
3368 // Collect counts on whether this check-cast sees NULLs a lot or not.
3369 if (ProfileInterpreter) {
3370 __ jmp(done);
3371 __ bind(is_null);
3372 __ profile_null_seen(rcx);
3373 } else {
3374 __ bind(is_null); // same as 'done'
3375 }
3376 __ bind(done);
3377 }
3378
3379 void TemplateTable::instanceof() {
3380 transition(atos, itos);
3381 Label done, is_null, ok_is_subtype, quicked, resolved;
3382 __ testptr(rax, rax);
3383 __ jcc(Assembler::zero, is_null);
3384
3385 // Get cpool & tags index
3386 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3387 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3388 // See if bytecode has already been quicked
3389 __ cmpb(Address(rdx, rbx,
3390 Address::times_1,
3391 typeArrayOopDesc::header_size(T_BYTE) * wordSize),
3392 JVM_CONSTANT_Class);
3393 __ jcc(Assembler::equal, quicked);
3394
3395 __ push(atos); // save receiver for result, and for GC
3396 __ mov(r12, rcx); // save rcx
3397 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3398 __ movq(rcx, r12); // restore rcx
3399 __ reinit_heapbase();
3400 __ pop_ptr(rdx); // restore receiver
3401 __ load_klass(rdx, rdx);
3402 __ jmpb(resolved);
3403
3404 // Get superklass in rax and subklass in rdx
3405 __ bind(quicked);
3406 __ load_klass(rdx, rax);
3407 __ movptr(rax, Address(rcx, rbx,
3408 Address::times_8, sizeof(constantPoolOopDesc)));
3409
3410 __ bind(resolved);
3411
3412 // Generate subtype check. Blows rcx, rdi
3413 // Superklass in rax. Subklass in rdx.
3414 __ gen_subtype_check(rdx, ok_is_subtype);
3415
3416 // Come here on failure
3417 __ xorl(rax, rax);
3418 __ jmpb(done);
3419 // Come here on success
3420 __ bind(ok_is_subtype);
3421 __ movl(rax, 1);
3422
3423 // Collect counts on whether this test sees NULLs a lot or not.
3424 if (ProfileInterpreter) {
3425 __ jmp(done);
3426 __ bind(is_null);
3427 __ profile_null_seen(rcx);
3428 } else {
3429 __ bind(is_null); // same as 'done'
3430 }
3431 __ bind(done);
3432 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
3433 // rax = 1: obj != NULL and obj is an instanceof the specified klass
3434 }
3435
3436 //-----------------------------------------------------------------------------
3437 // Breakpoints
3438 void TemplateTable::_breakpoint() {
3439 // Note: We get here even if we are single stepping..
3440 // jbug inists on setting breakpoints at every bytecode
3441 // even if we are in single step mode.
3442
3443 transition(vtos, vtos);
3444
3445 // get the unpatched byte code
3446 __ get_method(c_rarg1);
3447 __ call_VM(noreg,
3448 CAST_FROM_FN_PTR(address,
3449 InterpreterRuntime::get_original_bytecode_at),
3450 c_rarg1, r13);
3451 __ mov(rbx, rax);
3452
3453 // post the breakpoint event
3454 __ get_method(c_rarg1);
3455 __ call_VM(noreg,
3456 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3457 c_rarg1, r13);
3458
3459 // complete the execution of original bytecode
3460 __ dispatch_only_normal(vtos);
3461 }
3462
3463 //-----------------------------------------------------------------------------
3464 // Exceptions
3465
3466 void TemplateTable::athrow() {
3467 transition(atos, vtos);
3468 __ null_check(rax);
3469 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3470 }
3471
3472 //-----------------------------------------------------------------------------
3473 // Synchronization
3474 //
3475 // Note: monitorenter & exit are symmetric routines; which is reflected
3476 // in the assembly code structure as well
3477 //
3478 // Stack layout:
3479 //
3480 // [expressions ] <--- rsp = expression stack top
3481 // ..
3482 // [expressions ]
3483 // [monitor entry] <--- monitor block top = expression stack bot
3484 // ..
3485 // [monitor entry]
3486 // [frame data ] <--- monitor block bot
3487 // ...
3488 // [saved rbp ] <--- rbp
3489 void TemplateTable::monitorenter() {
3490 transition(atos, vtos);
3491
3492 // check for NULL object
3493 __ null_check(rax);
3494
3495 const Address monitor_block_top(
3496 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3497 const Address monitor_block_bot(
3498 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3499 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3500
3501 Label allocated;
3502
3503 // initialize entry pointer
3504 __ xorl(c_rarg1, c_rarg1); // points to free slot or NULL
3505
3506 // find a free slot in the monitor block (result in c_rarg1)
3507 {
3508 Label entry, loop, exit;
3509 __ movptr(c_rarg3, monitor_block_top); // points to current entry,
3510 // starting with top-most entry
3511 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3512 // of monitor block
3513 __ jmpb(entry);
3514
3515 __ bind(loop);
3516 // check if current entry is used
3517 __ cmpptr(Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
3518 // if not used then remember entry in c_rarg1
3519 __ cmov(Assembler::equal, c_rarg1, c_rarg3);
3520 // check if current entry is for same object
3521 __ cmpptr(rax, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3522 // if same object then stop searching
3523 __ jccb(Assembler::equal, exit);
3524 // otherwise advance to next entry
3525 __ addptr(c_rarg3, entry_size);
3526 __ bind(entry);
3527 // check if bottom reached
3528 __ cmpptr(c_rarg3, c_rarg2);
3529 // if not at bottom then check this entry
3530 __ jcc(Assembler::notEqual, loop);
3531 __ bind(exit);
3532 }
3533
3534 __ testptr(c_rarg1, c_rarg1); // check if a slot has been found
3535 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
3536
3537 // allocate one if there's no free slot
3538 {
3539 Label entry, loop;
3540 // 1. compute new pointers // rsp: old expression stack top
3541 __ movptr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
3542 __ subptr(rsp, entry_size); // move expression stack top
3543 __ subptr(c_rarg1, entry_size); // move expression stack bottom
3544 __ mov(c_rarg3, rsp); // set start value for copy loop
3545 __ movptr(monitor_block_bot, c_rarg1); // set new monitor block bottom
3546 __ jmp(entry);
3547 // 2. move expression stack contents
3548 __ bind(loop);
3549 __ movptr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3550 // word from old location
3551 __ movptr(Address(c_rarg3, 0), c_rarg2); // and store it at new location
3552 __ addptr(c_rarg3, wordSize); // advance to next word
3553 __ bind(entry);
3554 __ cmpptr(c_rarg3, c_rarg1); // check if bottom reached
3555 __ jcc(Assembler::notEqual, loop); // if not at bottom then
3556 // copy next word
3557 }
3558
3559 // call run-time routine
3560 // c_rarg1: points to monitor entry
3561 __ bind(allocated);
3562
3563 // Increment bcp to point to the next bytecode, so exception
3564 // handling for async. exceptions work correctly.
3565 // The object has already been poped from the stack, so the
3566 // expression stack looks correct.
3567 __ increment(r13);
3568
3569 // store object
3570 __ movptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), rax);
3571 __ lock_object(c_rarg1);
3572
3573 // check to make sure this monitor doesn't cause stack overflow after locking
3574 __ save_bcp(); // in case of exception
3575 __ generate_stack_overflow_check(0);
3576
3577 // The bcp has already been incremented. Just need to dispatch to
3578 // next instruction.
3579 __ dispatch_next(vtos);
3580 }
3581
3582
3583 void TemplateTable::monitorexit() {
3584 transition(atos, vtos);
3585
3586 // check for NULL object
3587 __ null_check(rax);
3588
3589 const Address monitor_block_top(
3590 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3591 const Address monitor_block_bot(
3592 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3593 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3594
3595 Label found;
3596
3597 // find matching slot
3598 {
3599 Label entry, loop;
3600 __ movptr(c_rarg1, monitor_block_top); // points to current entry,
3601 // starting with top-most entry
3602 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3603 // of monitor block
3604 __ jmpb(entry);
3605
3606 __ bind(loop);
3607 // check if current entry is for same object
3608 __ cmpptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3609 // if same object then stop searching
3610 __ jcc(Assembler::equal, found);
3611 // otherwise advance to next entry
3612 __ addptr(c_rarg1, entry_size);
3613 __ bind(entry);
3614 // check if bottom reached
3615 __ cmpptr(c_rarg1, c_rarg2);
3616 // if not at bottom then check this entry
3617 __ jcc(Assembler::notEqual, loop);
3618 }
3619
3620 // error handling. Unlocking was not block-structured
3621 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3622 InterpreterRuntime::throw_illegal_monitor_state_exception));
3623 __ should_not_reach_here();
3624
3625 // call run-time routine
3626 // rsi: points to monitor entry
3627 __ bind(found);
3628 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3629 __ unlock_object(c_rarg1);
3630 __ pop_ptr(rax); // discard object
3631 }
3632
3633
3634 // Wide instructions
3635 void TemplateTable::wide() {
3636 transition(vtos, vtos);
3637 __ load_unsigned_byte(rbx, at_bcp(1));
3638 __ lea(rscratch1, ExternalAddress((address)Interpreter::_wentry_point));
3639 __ jmp(Address(rscratch1, rbx, Address::times_8));
3640 // Note: the r13 increment step is part of the individual wide
3641 // bytecode implementations
3642 }
3643
3644
3645 // Multi arrays
3646 void TemplateTable::multianewarray() {
3647 transition(vtos, atos);
3648 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3649 // last dim is on top of stack; we want address of first one:
3650 // first_addr = last_addr + (ndims - 1) * wordSize
3651 __ lea(c_rarg1, Address(rsp, rax, Address::times_8, -wordSize));
3652 call_VM(rax,
3653 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3654 c_rarg1);
3655 __ load_unsigned_byte(rbx, at_bcp(3));
3656 __ lea(rsp, Address(rsp, rbx, Address::times_8));
3657 }
3658 #endif // !CC_INTERP