1 /*
2 * Copyright 2003-2009 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any 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, Interpreter::value_offset_in_bytes());
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 void TemplateTable::ldc2_w() {
393 transition(vtos, vtos);
394 Label Long, Done;
395 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
396
397 __ get_cpool_and_tags(rcx, rax);
398 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
399 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
400
401 // get type
402 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
403 JVM_CONSTANT_Double);
404 __ jccb(Assembler::notEqual, Long);
405 // dtos
406 __ movdbl(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
407 __ push_d();
408 __ jmpb(Done);
409
410 __ bind(Long);
411 // ltos
412 __ movq(rax, Address(rcx, rbx, Address::times_8, base_offset));
413 __ push_l();
414
415 __ bind(Done);
416 }
417
418 void TemplateTable::locals_index(Register reg, int offset) {
419 __ load_unsigned_byte(reg, at_bcp(offset));
420 __ negptr(reg);
421 if (TaggedStackInterpreter) __ shlptr(reg, 1); // index = index*2
422 }
423
424 void TemplateTable::iload() {
425 transition(vtos, itos);
426 if (RewriteFrequentPairs) {
427 Label rewrite, done;
428 const Register bc = c_rarg3;
429 assert(rbx != bc, "register damaged");
430
431 // get next byte
432 __ load_unsigned_byte(rbx,
433 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
434 // if _iload, wait to rewrite to iload2. We only want to rewrite the
435 // last two iloads in a pair. Comparing against fast_iload means that
436 // the next bytecode is neither an iload or a caload, and therefore
437 // an iload pair.
438 __ cmpl(rbx, Bytecodes::_iload);
439 __ jcc(Assembler::equal, done);
440
441 __ cmpl(rbx, Bytecodes::_fast_iload);
442 __ movl(bc, Bytecodes::_fast_iload2);
443 __ jccb(Assembler::equal, rewrite);
444
445 // if _caload, rewrite to fast_icaload
446 __ cmpl(rbx, Bytecodes::_caload);
447 __ movl(bc, Bytecodes::_fast_icaload);
448 __ jccb(Assembler::equal, rewrite);
449
450 // rewrite so iload doesn't check again.
451 __ movl(bc, Bytecodes::_fast_iload);
452
453 // rewrite
454 // bc: fast bytecode
455 __ bind(rewrite);
456 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
457 __ bind(done);
458 }
459
460 // Get the local value into tos
461 locals_index(rbx);
462 __ movl(rax, iaddress(rbx));
463 debug_only(__ verify_local_tag(frame::TagValue, rbx));
464 }
465
466 void TemplateTable::fast_iload2() {
467 transition(vtos, itos);
468 locals_index(rbx);
469 __ movl(rax, iaddress(rbx));
470 debug_only(__ verify_local_tag(frame::TagValue, rbx));
471 __ push(itos);
472 locals_index(rbx, 3);
473 __ movl(rax, iaddress(rbx));
474 debug_only(__ verify_local_tag(frame::TagValue, rbx));
475 }
476
477 void TemplateTable::fast_iload() {
478 transition(vtos, itos);
479 locals_index(rbx);
480 __ movl(rax, iaddress(rbx));
481 debug_only(__ verify_local_tag(frame::TagValue, rbx));
482 }
483
484 void TemplateTable::lload() {
485 transition(vtos, ltos);
486 locals_index(rbx);
487 __ movq(rax, laddress(rbx));
488 debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
489 }
490
491 void TemplateTable::fload() {
492 transition(vtos, ftos);
493 locals_index(rbx);
494 __ movflt(xmm0, faddress(rbx));
495 debug_only(__ verify_local_tag(frame::TagValue, rbx));
496 }
497
498 void TemplateTable::dload() {
499 transition(vtos, dtos);
500 locals_index(rbx);
501 __ movdbl(xmm0, daddress(rbx));
502 debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
503 }
504
505 void TemplateTable::aload() {
506 transition(vtos, atos);
507 locals_index(rbx);
508 __ movptr(rax, aaddress(rbx));
509 debug_only(__ verify_local_tag(frame::TagReference, rbx));
510 }
511
512 void TemplateTable::locals_index_wide(Register reg) {
513 __ movl(reg, at_bcp(2));
514 __ bswapl(reg);
515 __ shrl(reg, 16);
516 __ negptr(reg);
517 if (TaggedStackInterpreter) __ shlptr(reg, 1); // index = index*2
518 }
519
520 void TemplateTable::wide_iload() {
521 transition(vtos, itos);
522 locals_index_wide(rbx);
523 __ movl(rax, iaddress(rbx));
524 debug_only(__ verify_local_tag(frame::TagValue, rbx));
525 }
526
527 void TemplateTable::wide_lload() {
528 transition(vtos, ltos);
529 locals_index_wide(rbx);
530 __ movq(rax, laddress(rbx));
531 debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
532 }
533
534 void TemplateTable::wide_fload() {
535 transition(vtos, ftos);
536 locals_index_wide(rbx);
537 __ movflt(xmm0, faddress(rbx));
538 debug_only(__ verify_local_tag(frame::TagValue, rbx));
539 }
540
541 void TemplateTable::wide_dload() {
542 transition(vtos, dtos);
543 locals_index_wide(rbx);
544 __ movdbl(xmm0, daddress(rbx));
545 debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
546 }
547
548 void TemplateTable::wide_aload() {
549 transition(vtos, atos);
550 locals_index_wide(rbx);
551 __ movptr(rax, aaddress(rbx));
552 debug_only(__ verify_local_tag(frame::TagReference, rbx));
553 }
554
555 void TemplateTable::index_check(Register array, Register index) {
556 // destroys rbx
557 // check array
558 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
559 // sign extend index for use by indexed load
560 __ movl2ptr(index, index);
561 // check index
562 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
563 if (index != rbx) {
564 // ??? convention: move aberrant index into ebx for exception message
565 assert(rbx != array, "different registers");
566 __ movl(rbx, index);
567 }
568 __ jump_cc(Assembler::aboveEqual,
569 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
570 }
571
572 void TemplateTable::iaload() {
573 transition(itos, itos);
574 __ pop_ptr(rdx);
575 // eax: index
576 // rdx: array
577 index_check(rdx, rax); // kills rbx
578 __ movl(rax, Address(rdx, rax,
579 Address::times_4,
580 arrayOopDesc::base_offset_in_bytes(T_INT)));
581 }
582
583 void TemplateTable::laload() {
584 transition(itos, ltos);
585 __ pop_ptr(rdx);
586 // eax: index
587 // rdx: array
588 index_check(rdx, rax); // kills rbx
589 __ movq(rax, Address(rdx, rbx,
590 Address::times_8,
591 arrayOopDesc::base_offset_in_bytes(T_LONG)));
592 }
593
594 void TemplateTable::faload() {
595 transition(itos, ftos);
596 __ pop_ptr(rdx);
597 // eax: index
598 // rdx: array
599 index_check(rdx, rax); // kills rbx
600 __ movflt(xmm0, Address(rdx, rax,
601 Address::times_4,
602 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
603 }
604
605 void TemplateTable::daload() {
606 transition(itos, dtos);
607 __ pop_ptr(rdx);
608 // eax: index
609 // rdx: array
610 index_check(rdx, rax); // kills rbx
611 __ movdbl(xmm0, Address(rdx, rax,
612 Address::times_8,
613 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
614 }
615
616 void TemplateTable::aaload() {
617 transition(itos, atos);
618 __ pop_ptr(rdx);
619 // eax: index
620 // rdx: array
621 index_check(rdx, rax); // kills rbx
622 __ load_heap_oop(rax, Address(rdx, rax,
623 UseCompressedOops ? Address::times_4 : Address::times_8,
624 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
625 }
626
627 void TemplateTable::baload() {
628 transition(itos, itos);
629 __ pop_ptr(rdx);
630 // eax: index
631 // rdx: array
632 index_check(rdx, rax); // kills rbx
633 __ load_signed_byte(rax,
634 Address(rdx, rax,
635 Address::times_1,
636 arrayOopDesc::base_offset_in_bytes(T_BYTE)));
637 }
638
639 void TemplateTable::caload() {
640 transition(itos, itos);
641 __ pop_ptr(rdx);
642 // eax: index
643 // rdx: array
644 index_check(rdx, rax); // kills rbx
645 __ load_unsigned_short(rax,
646 Address(rdx, rax,
647 Address::times_2,
648 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
649 }
650
651 // iload followed by caload frequent pair
652 void TemplateTable::fast_icaload() {
653 transition(vtos, itos);
654 // load index out of locals
655 locals_index(rbx);
656 __ movl(rax, iaddress(rbx));
657 debug_only(__ verify_local_tag(frame::TagValue, rbx));
658
659 // eax: index
660 // rdx: array
661 __ pop_ptr(rdx);
662 index_check(rdx, rax); // kills rbx
663 __ load_unsigned_short(rax,
664 Address(rdx, rax,
665 Address::times_2,
666 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
667 }
668
669 void TemplateTable::saload() {
670 transition(itos, itos);
671 __ pop_ptr(rdx);
672 // eax: index
673 // rdx: array
674 index_check(rdx, rax); // kills rbx
675 __ load_signed_short(rax,
676 Address(rdx, rax,
677 Address::times_2,
678 arrayOopDesc::base_offset_in_bytes(T_SHORT)));
679 }
680
681 void TemplateTable::iload(int n) {
682 transition(vtos, itos);
683 __ movl(rax, iaddress(n));
684 debug_only(__ verify_local_tag(frame::TagValue, n));
685 }
686
687 void TemplateTable::lload(int n) {
688 transition(vtos, ltos);
689 __ movq(rax, laddress(n));
690 debug_only(__ verify_local_tag(frame::TagCategory2, n));
691 }
692
693 void TemplateTable::fload(int n) {
694 transition(vtos, ftos);
695 __ movflt(xmm0, faddress(n));
696 debug_only(__ verify_local_tag(frame::TagValue, n));
697 }
698
699 void TemplateTable::dload(int n) {
700 transition(vtos, dtos);
701 __ movdbl(xmm0, daddress(n));
702 debug_only(__ verify_local_tag(frame::TagCategory2, n));
703 }
704
705 void TemplateTable::aload(int n) {
706 transition(vtos, atos);
707 __ movptr(rax, aaddress(n));
708 debug_only(__ verify_local_tag(frame::TagReference, n));
709 }
710
711 void TemplateTable::aload_0() {
712 transition(vtos, atos);
713 // According to bytecode histograms, the pairs:
714 //
715 // _aload_0, _fast_igetfield
716 // _aload_0, _fast_agetfield
717 // _aload_0, _fast_fgetfield
718 //
719 // occur frequently. If RewriteFrequentPairs is set, the (slow)
720 // _aload_0 bytecode checks if the next bytecode is either
721 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
722 // rewrites the current bytecode into a pair bytecode; otherwise it
723 // rewrites the current bytecode into _fast_aload_0 that doesn't do
724 // the pair check anymore.
725 //
726 // Note: If the next bytecode is _getfield, the rewrite must be
727 // delayed, otherwise we may miss an opportunity for a pair.
728 //
729 // Also rewrite frequent pairs
730 // aload_0, aload_1
731 // aload_0, iload_1
732 // These bytecodes with a small amount of code are most profitable
733 // to rewrite
734 if (RewriteFrequentPairs) {
735 Label rewrite, done;
736 const Register bc = c_rarg3;
737 assert(rbx != bc, "register damaged");
738 // get next byte
739 __ load_unsigned_byte(rbx,
740 at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
741
742 // do actual aload_0
743 aload(0);
744
745 // if _getfield then wait with rewrite
746 __ cmpl(rbx, Bytecodes::_getfield);
747 __ jcc(Assembler::equal, done);
748
749 // if _igetfield then reqrite to _fast_iaccess_0
750 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) ==
751 Bytecodes::_aload_0,
752 "fix bytecode definition");
753 __ cmpl(rbx, Bytecodes::_fast_igetfield);
754 __ movl(bc, Bytecodes::_fast_iaccess_0);
755 __ jccb(Assembler::equal, rewrite);
756
757 // if _agetfield then reqrite to _fast_aaccess_0
758 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) ==
759 Bytecodes::_aload_0,
760 "fix bytecode definition");
761 __ cmpl(rbx, Bytecodes::_fast_agetfield);
762 __ movl(bc, Bytecodes::_fast_aaccess_0);
763 __ jccb(Assembler::equal, rewrite);
764
765 // if _fgetfield then reqrite to _fast_faccess_0
766 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) ==
767 Bytecodes::_aload_0,
768 "fix bytecode definition");
769 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
770 __ movl(bc, Bytecodes::_fast_faccess_0);
771 __ jccb(Assembler::equal, rewrite);
772
773 // else rewrite to _fast_aload0
774 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) ==
775 Bytecodes::_aload_0,
776 "fix bytecode definition");
777 __ movl(bc, Bytecodes::_fast_aload_0);
778
779 // rewrite
780 // bc: fast bytecode
781 __ bind(rewrite);
782 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
783
784 __ bind(done);
785 } else {
786 aload(0);
787 }
788 }
789
790 void TemplateTable::istore() {
791 transition(itos, vtos);
792 locals_index(rbx);
793 __ movl(iaddress(rbx), rax);
794 __ tag_local(frame::TagValue, rbx);
795 }
796
797 void TemplateTable::lstore() {
798 transition(ltos, vtos);
799 locals_index(rbx);
800 __ movq(laddress(rbx), rax);
801 __ tag_local(frame::TagCategory2, rbx);
802 }
803
804 void TemplateTable::fstore() {
805 transition(ftos, vtos);
806 locals_index(rbx);
807 __ movflt(faddress(rbx), xmm0);
808 __ tag_local(frame::TagValue, rbx);
809 }
810
811 void TemplateTable::dstore() {
812 transition(dtos, vtos);
813 locals_index(rbx);
814 __ movdbl(daddress(rbx), xmm0);
815 __ tag_local(frame::TagCategory2, rbx);
816 }
817
818 void TemplateTable::astore() {
819 transition(vtos, vtos);
820 __ pop_ptr(rax, rdx); // will need to pop tag too
821 locals_index(rbx);
822 __ movptr(aaddress(rbx), rax);
823 __ tag_local(rdx, rbx); // store tag from stack, might be returnAddr
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 __ tag_local(frame::TagValue, rbx);
832 }
833
834 void TemplateTable::wide_lstore() {
835 transition(vtos, vtos);
836 __ pop_l();
837 locals_index_wide(rbx);
838 __ movq(laddress(rbx), rax);
839 __ tag_local(frame::TagCategory2, rbx);
840 }
841
842 void TemplateTable::wide_fstore() {
843 transition(vtos, vtos);
844 __ pop_f();
845 locals_index_wide(rbx);
846 __ movflt(faddress(rbx), xmm0);
847 __ tag_local(frame::TagValue, rbx);
848 }
849
850 void TemplateTable::wide_dstore() {
851 transition(vtos, vtos);
852 __ pop_d();
853 locals_index_wide(rbx);
854 __ movdbl(daddress(rbx), xmm0);
855 __ tag_local(frame::TagCategory2, rbx);
856 }
857
858 void TemplateTable::wide_astore() {
859 transition(vtos, vtos);
860 __ pop_ptr(rax, rdx); // will need to pop tag too
861 locals_index_wide(rbx);
862 __ movptr(aaddress(rbx), rax);
863 __ tag_local(rdx, rbx); // store tag from stack, might be returnAddr
864 }
865
866 void TemplateTable::iastore() {
867 transition(itos, vtos);
868 __ pop_i(rbx);
869 __ pop_ptr(rdx);
870 // eax: value
871 // ebx: index
872 // rdx: array
873 index_check(rdx, rbx); // prefer index in ebx
874 __ movl(Address(rdx, rbx,
875 Address::times_4,
876 arrayOopDesc::base_offset_in_bytes(T_INT)),
877 rax);
878 }
879
880 void TemplateTable::lastore() {
881 transition(ltos, vtos);
882 __ pop_i(rbx);
883 __ pop_ptr(rdx);
884 // rax: value
885 // ebx: index
886 // rdx: array
887 index_check(rdx, rbx); // prefer index in ebx
888 __ movq(Address(rdx, rbx,
889 Address::times_8,
890 arrayOopDesc::base_offset_in_bytes(T_LONG)),
891 rax);
892 }
893
894 void TemplateTable::fastore() {
895 transition(ftos, vtos);
896 __ pop_i(rbx);
897 __ pop_ptr(rdx);
898 // xmm0: value
899 // ebx: index
900 // rdx: array
901 index_check(rdx, rbx); // prefer index in ebx
902 __ movflt(Address(rdx, rbx,
903 Address::times_4,
904 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
905 xmm0);
906 }
907
908 void TemplateTable::dastore() {
909 transition(dtos, vtos);
910 __ pop_i(rbx);
911 __ pop_ptr(rdx);
912 // xmm0: value
913 // ebx: index
914 // rdx: array
915 index_check(rdx, rbx); // prefer index in ebx
916 __ movdbl(Address(rdx, rbx,
917 Address::times_8,
918 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
919 xmm0);
920 }
921
922 void TemplateTable::aastore() {
923 Label is_null, ok_is_subtype, done;
924 transition(vtos, vtos);
925 // stack: ..., array, index, value
926 __ movptr(rax, at_tos()); // value
927 __ movl(rcx, at_tos_p1()); // index
928 __ movptr(rdx, at_tos_p2()); // array
929
930 Address element_address(rdx, rcx,
931 UseCompressedOops? Address::times_4 : Address::times_8,
932 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
933
934 index_check(rdx, rcx); // kills rbx
935 // do array store check - check for NULL value first
936 __ testptr(rax, rax);
937 __ jcc(Assembler::zero, is_null);
938
939 // Move subklass into rbx
940 __ load_klass(rbx, rax);
941 // Move superklass into rax
942 __ load_klass(rax, rdx);
943 __ movptr(rax, Address(rax,
944 sizeof(oopDesc) +
945 objArrayKlass::element_klass_offset_in_bytes()));
946 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
947 __ lea(rdx, element_address);
948
949 // Generate subtype check. Blows rcx, rdi
950 // Superklass in rax. Subklass in rbx.
951 __ gen_subtype_check(rbx, ok_is_subtype);
952
953 // Come here on failure
954 // object is at TOS
955 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
956
957 // Come here on success
958 __ bind(ok_is_subtype);
959
960 // Get the value we will store
961 __ movptr(rax, at_tos());
962 // Now store using the appropriate barrier
963 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
964 __ jmp(done);
965
966 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
967 __ bind(is_null);
968 __ profile_null_seen(rbx);
969
970 // Store a NULL
971 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
972
973 // Pop stack arguments
974 __ bind(done);
975 __ addptr(rsp, 3 * Interpreter::stackElementSize());
976 }
977
978 void TemplateTable::bastore() {
979 transition(itos, vtos);
980 __ pop_i(rbx);
981 __ pop_ptr(rdx);
982 // eax: value
983 // ebx: index
984 // rdx: array
985 index_check(rdx, rbx); // prefer index in ebx
986 __ movb(Address(rdx, rbx,
987 Address::times_1,
988 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
989 rax);
990 }
991
992 void TemplateTable::castore() {
993 transition(itos, vtos);
994 __ pop_i(rbx);
995 __ pop_ptr(rdx);
996 // eax: value
997 // ebx: index
998 // rdx: array
999 index_check(rdx, rbx); // prefer index in ebx
1000 __ movw(Address(rdx, rbx,
1001 Address::times_2,
1002 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1003 rax);
1004 }
1005
1006 void TemplateTable::sastore() {
1007 castore();
1008 }
1009
1010 void TemplateTable::istore(int n) {
1011 transition(itos, vtos);
1012 __ movl(iaddress(n), rax);
1013 __ tag_local(frame::TagValue, n);
1014 }
1015
1016 void TemplateTable::lstore(int n) {
1017 transition(ltos, vtos);
1018 __ movq(laddress(n), rax);
1019 __ tag_local(frame::TagCategory2, n);
1020 }
1021
1022 void TemplateTable::fstore(int n) {
1023 transition(ftos, vtos);
1024 __ movflt(faddress(n), xmm0);
1025 __ tag_local(frame::TagValue, n);
1026 }
1027
1028 void TemplateTable::dstore(int n) {
1029 transition(dtos, vtos);
1030 __ movdbl(daddress(n), xmm0);
1031 __ tag_local(frame::TagCategory2, n);
1032 }
1033
1034 void TemplateTable::astore(int n) {
1035 transition(vtos, vtos);
1036 __ pop_ptr(rax, rdx);
1037 __ movptr(aaddress(n), rax);
1038 __ tag_local(rdx, n);
1039 }
1040
1041 void TemplateTable::pop() {
1042 transition(vtos, vtos);
1043 __ addptr(rsp, Interpreter::stackElementSize());
1044 }
1045
1046 void TemplateTable::pop2() {
1047 transition(vtos, vtos);
1048 __ addptr(rsp, 2 * Interpreter::stackElementSize());
1049 }
1050
1051 void TemplateTable::dup() {
1052 transition(vtos, vtos);
1053 __ load_ptr_and_tag(0, rax, rdx);
1054 __ push_ptr(rax, rdx);
1055 // stack: ..., a, a
1056 }
1057
1058 void TemplateTable::dup_x1() {
1059 transition(vtos, vtos);
1060 // stack: ..., a, b
1061 __ load_ptr_and_tag(0, rax, rdx); // load b
1062 __ load_ptr_and_tag(1, rcx, rbx); // load a
1063 __ store_ptr_and_tag(1, rax, rdx); // store b
1064 __ store_ptr_and_tag(0, rcx, rbx); // store a
1065 __ push_ptr(rax, rdx); // push b
1066 // stack: ..., b, a, b
1067 }
1068
1069 void TemplateTable::dup_x2() {
1070 transition(vtos, vtos);
1071 // stack: ..., a, b, c
1072 __ load_ptr_and_tag(0, rax, rdx); // load c
1073 __ load_ptr_and_tag(2, rcx, rbx); // load a
1074 __ store_ptr_and_tag(2, rax, rdx); // store c in a
1075 __ push_ptr(rax, rdx); // push c
1076 // stack: ..., c, b, c, c
1077 __ load_ptr_and_tag(2, rax, rdx); // load b
1078 __ store_ptr_and_tag(2, rcx, rbx); // store a in b
1079 // stack: ..., c, a, c, c
1080 __ store_ptr_and_tag(1, rax, rdx); // store b in c
1081 // stack: ..., c, a, b, c
1082 }
1083
1084 void TemplateTable::dup2() {
1085 transition(vtos, vtos);
1086 // stack: ..., a, b
1087 __ load_ptr_and_tag(1, rax, rdx); // load a
1088 __ push_ptr(rax, rdx); // push a
1089 __ load_ptr_and_tag(1, rax, rdx); // load b
1090 __ push_ptr(rax, rdx); // push b
1091 // stack: ..., a, b, a, b
1092 }
1093
1094 void TemplateTable::dup2_x1() {
1095 transition(vtos, vtos);
1096 // stack: ..., a, b, c
1097 __ load_ptr_and_tag(0, rcx, rbx); // load c
1098 __ load_ptr_and_tag(1, rax, rdx); // load b
1099 __ push_ptr(rax, rdx); // push b
1100 __ push_ptr(rcx, rbx); // push c
1101 // stack: ..., a, b, c, b, c
1102 __ store_ptr_and_tag(3, rcx, rbx); // store c in b
1103 // stack: ..., a, c, c, b, c
1104 __ load_ptr_and_tag(4, rcx, rbx); // load a
1105 __ store_ptr_and_tag(2, rcx, rbx); // store a in 2nd c
1106 // stack: ..., a, c, a, b, c
1107 __ store_ptr_and_tag(4, rax, rdx); // store b in a
1108 // stack: ..., b, c, a, b, c
1109 }
1110
1111 void TemplateTable::dup2_x2() {
1112 transition(vtos, vtos);
1113 // stack: ..., a, b, c, d
1114 __ load_ptr_and_tag(0, rcx, rbx); // load d
1115 __ load_ptr_and_tag(1, rax, rdx); // load c
1116 __ push_ptr(rax, rdx); // push c
1117 __ push_ptr(rcx, rbx); // push d
1118 // stack: ..., a, b, c, d, c, d
1119 __ load_ptr_and_tag(4, rax, rdx); // load b
1120 __ store_ptr_and_tag(2, rax, rdx); // store b in d
1121 __ store_ptr_and_tag(4, rcx, rbx); // store d in b
1122 // stack: ..., a, d, c, b, c, d
1123 __ load_ptr_and_tag(5, rcx, rbx); // load a
1124 __ load_ptr_and_tag(3, rax, rdx); // load c
1125 __ store_ptr_and_tag(3, rcx, rbx); // store a in c
1126 __ store_ptr_and_tag(5, rax, rdx); // store c in a
1127 // stack: ..., c, d, a, b, c, d
1128 }
1129
1130 void TemplateTable::swap() {
1131 transition(vtos, vtos);
1132 // stack: ..., a, b
1133 __ load_ptr_and_tag(1, rcx, rbx); // load a
1134 __ load_ptr_and_tag(0, rax, rdx); // load b
1135 __ store_ptr_and_tag(0, rcx, rbx); // store a in b
1136 __ store_ptr_and_tag(1, rax, rdx); // store b in a
1137 // stack: ..., b, a
1138 }
1139
1140 void TemplateTable::iop2(Operation op) {
1141 transition(itos, itos);
1142 switch (op) {
1143 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1144 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1145 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1146 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1147 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1148 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1149 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1150 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1151 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1152 default : ShouldNotReachHere();
1153 }
1154 }
1155
1156 void TemplateTable::lop2(Operation op) {
1157 transition(ltos, ltos);
1158 switch (op) {
1159 case add : __ pop_l(rdx); __ addptr (rax, rdx); break;
1160 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr (rax, rdx); break;
1161 case _and : __ pop_l(rdx); __ andptr (rax, rdx); break;
1162 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1163 case _xor : __ pop_l(rdx); __ xorptr (rax, rdx); break;
1164 default : ShouldNotReachHere();
1165 }
1166 }
1167
1168 void TemplateTable::idiv() {
1169 transition(itos, itos);
1170 __ movl(rcx, rax);
1171 __ pop_i(rax);
1172 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1173 // they are not equal, one could do a normal division (no correction
1174 // needed), which may speed up this implementation for the common case.
1175 // (see also JVM spec., p.243 & p.271)
1176 __ corrected_idivl(rcx);
1177 }
1178
1179 void TemplateTable::irem() {
1180 transition(itos, itos);
1181 __ movl(rcx, rax);
1182 __ pop_i(rax);
1183 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1184 // they are not equal, one could do a normal division (no correction
1185 // needed), which may speed up this implementation for the common case.
1186 // (see also JVM spec., p.243 & p.271)
1187 __ corrected_idivl(rcx);
1188 __ movl(rax, rdx);
1189 }
1190
1191 void TemplateTable::lmul() {
1192 transition(ltos, ltos);
1193 __ pop_l(rdx);
1194 __ imulq(rax, rdx);
1195 }
1196
1197 void TemplateTable::ldiv() {
1198 transition(ltos, ltos);
1199 __ mov(rcx, rax);
1200 __ pop_l(rax);
1201 // generate explicit div0 check
1202 __ testq(rcx, rcx);
1203 __ jump_cc(Assembler::zero,
1204 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1205 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1206 // they are not equal, one could do a normal division (no correction
1207 // needed), which may speed up this implementation for the common case.
1208 // (see also JVM spec., p.243 & p.271)
1209 __ corrected_idivq(rcx); // kills rbx
1210 }
1211
1212 void TemplateTable::lrem() {
1213 transition(ltos, ltos);
1214 __ mov(rcx, rax);
1215 __ pop_l(rax);
1216 __ testq(rcx, rcx);
1217 __ jump_cc(Assembler::zero,
1218 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1219 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1220 // they are not equal, one could do a normal division (no correction
1221 // needed), which may speed up this implementation for the common case.
1222 // (see also JVM spec., p.243 & p.271)
1223 __ corrected_idivq(rcx); // kills rbx
1224 __ mov(rax, rdx);
1225 }
1226
1227 void TemplateTable::lshl() {
1228 transition(itos, ltos);
1229 __ movl(rcx, rax); // get shift count
1230 __ pop_l(rax); // get shift value
1231 __ shlq(rax);
1232 }
1233
1234 void TemplateTable::lshr() {
1235 transition(itos, ltos);
1236 __ movl(rcx, rax); // get shift count
1237 __ pop_l(rax); // get shift value
1238 __ sarq(rax);
1239 }
1240
1241 void TemplateTable::lushr() {
1242 transition(itos, ltos);
1243 __ movl(rcx, rax); // get shift count
1244 __ pop_l(rax); // get shift value
1245 __ shrq(rax);
1246 }
1247
1248 void TemplateTable::fop2(Operation op) {
1249 transition(ftos, ftos);
1250 switch (op) {
1251 case add:
1252 __ addss(xmm0, at_rsp());
1253 __ addptr(rsp, Interpreter::stackElementSize());
1254 break;
1255 case sub:
1256 __ movflt(xmm1, xmm0);
1257 __ pop_f(xmm0);
1258 __ subss(xmm0, xmm1);
1259 break;
1260 case mul:
1261 __ mulss(xmm0, at_rsp());
1262 __ addptr(rsp, Interpreter::stackElementSize());
1263 break;
1264 case div:
1265 __ movflt(xmm1, xmm0);
1266 __ pop_f(xmm0);
1267 __ divss(xmm0, xmm1);
1268 break;
1269 case rem:
1270 __ movflt(xmm1, xmm0);
1271 __ pop_f(xmm0);
1272 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1273 break;
1274 default:
1275 ShouldNotReachHere();
1276 break;
1277 }
1278 }
1279
1280 void TemplateTable::dop2(Operation op) {
1281 transition(dtos, dtos);
1282 switch (op) {
1283 case add:
1284 __ addsd(xmm0, at_rsp());
1285 __ addptr(rsp, 2 * Interpreter::stackElementSize());
1286 break;
1287 case sub:
1288 __ movdbl(xmm1, xmm0);
1289 __ pop_d(xmm0);
1290 __ subsd(xmm0, xmm1);
1291 break;
1292 case mul:
1293 __ mulsd(xmm0, at_rsp());
1294 __ addptr(rsp, 2 * Interpreter::stackElementSize());
1295 break;
1296 case div:
1297 __ movdbl(xmm1, xmm0);
1298 __ pop_d(xmm0);
1299 __ divsd(xmm0, xmm1);
1300 break;
1301 case rem:
1302 __ movdbl(xmm1, xmm0);
1303 __ pop_d(xmm0);
1304 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1305 break;
1306 default:
1307 ShouldNotReachHere();
1308 break;
1309 }
1310 }
1311
1312 void TemplateTable::ineg() {
1313 transition(itos, itos);
1314 __ negl(rax);
1315 }
1316
1317 void TemplateTable::lneg() {
1318 transition(ltos, ltos);
1319 __ negq(rax);
1320 }
1321
1322 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1323 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1324 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1325 // of 128-bits operands for SSE instructions.
1326 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1327 // Store the value to a 128-bits operand.
1328 operand[0] = lo;
1329 operand[1] = hi;
1330 return operand;
1331 }
1332
1333 // Buffer for 128-bits masks used by SSE instructions.
1334 static jlong float_signflip_pool[2*2];
1335 static jlong double_signflip_pool[2*2];
1336
1337 void TemplateTable::fneg() {
1338 transition(ftos, ftos);
1339 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], 0x8000000080000000, 0x8000000080000000);
1340 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1341 }
1342
1343 void TemplateTable::dneg() {
1344 transition(dtos, dtos);
1345 static jlong *double_signflip = double_quadword(&double_signflip_pool[1], 0x8000000000000000, 0x8000000000000000);
1346 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1347 }
1348
1349 void TemplateTable::iinc() {
1350 transition(vtos, vtos);
1351 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1352 locals_index(rbx);
1353 __ addl(iaddress(rbx), rdx);
1354 }
1355
1356 void TemplateTable::wide_iinc() {
1357 transition(vtos, vtos);
1358 __ movl(rdx, at_bcp(4)); // get constant
1359 locals_index_wide(rbx);
1360 __ bswapl(rdx); // swap bytes & sign-extend constant
1361 __ sarl(rdx, 16);
1362 __ addl(iaddress(rbx), rdx);
1363 // Note: should probably use only one movl to get both
1364 // the index and the constant -> fix this
1365 }
1366
1367 void TemplateTable::convert() {
1368 // Checking
1369 #ifdef ASSERT
1370 {
1371 TosState tos_in = ilgl;
1372 TosState tos_out = ilgl;
1373 switch (bytecode()) {
1374 case Bytecodes::_i2l: // fall through
1375 case Bytecodes::_i2f: // fall through
1376 case Bytecodes::_i2d: // fall through
1377 case Bytecodes::_i2b: // fall through
1378 case Bytecodes::_i2c: // fall through
1379 case Bytecodes::_i2s: tos_in = itos; break;
1380 case Bytecodes::_l2i: // fall through
1381 case Bytecodes::_l2f: // fall through
1382 case Bytecodes::_l2d: tos_in = ltos; break;
1383 case Bytecodes::_f2i: // fall through
1384 case Bytecodes::_f2l: // fall through
1385 case Bytecodes::_f2d: tos_in = ftos; break;
1386 case Bytecodes::_d2i: // fall through
1387 case Bytecodes::_d2l: // fall through
1388 case Bytecodes::_d2f: tos_in = dtos; break;
1389 default : ShouldNotReachHere();
1390 }
1391 switch (bytecode()) {
1392 case Bytecodes::_l2i: // fall through
1393 case Bytecodes::_f2i: // fall through
1394 case Bytecodes::_d2i: // fall through
1395 case Bytecodes::_i2b: // fall through
1396 case Bytecodes::_i2c: // fall through
1397 case Bytecodes::_i2s: tos_out = itos; break;
1398 case Bytecodes::_i2l: // fall through
1399 case Bytecodes::_f2l: // fall through
1400 case Bytecodes::_d2l: tos_out = ltos; break;
1401 case Bytecodes::_i2f: // fall through
1402 case Bytecodes::_l2f: // fall through
1403 case Bytecodes::_d2f: tos_out = ftos; break;
1404 case Bytecodes::_i2d: // fall through
1405 case Bytecodes::_l2d: // fall through
1406 case Bytecodes::_f2d: tos_out = dtos; break;
1407 default : ShouldNotReachHere();
1408 }
1409 transition(tos_in, tos_out);
1410 }
1411 #endif // ASSERT
1412
1413 static const int64_t is_nan = 0x8000000000000000L;
1414
1415 // Conversion
1416 switch (bytecode()) {
1417 case Bytecodes::_i2l:
1418 __ movslq(rax, rax);
1419 break;
1420 case Bytecodes::_i2f:
1421 __ cvtsi2ssl(xmm0, rax);
1422 break;
1423 case Bytecodes::_i2d:
1424 __ cvtsi2sdl(xmm0, rax);
1425 break;
1426 case Bytecodes::_i2b:
1427 __ movsbl(rax, rax);
1428 break;
1429 case Bytecodes::_i2c:
1430 __ movzwl(rax, rax);
1431 break;
1432 case Bytecodes::_i2s:
1433 __ movswl(rax, rax);
1434 break;
1435 case Bytecodes::_l2i:
1436 __ movl(rax, rax);
1437 break;
1438 case Bytecodes::_l2f:
1439 __ cvtsi2ssq(xmm0, rax);
1440 break;
1441 case Bytecodes::_l2d:
1442 __ cvtsi2sdq(xmm0, rax);
1443 break;
1444 case Bytecodes::_f2i:
1445 {
1446 Label L;
1447 __ cvttss2sil(rax, xmm0);
1448 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1449 __ jcc(Assembler::notEqual, L);
1450 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1451 __ bind(L);
1452 }
1453 break;
1454 case Bytecodes::_f2l:
1455 {
1456 Label L;
1457 __ cvttss2siq(rax, xmm0);
1458 // NaN or overflow/underflow?
1459 __ cmp64(rax, ExternalAddress((address) &is_nan));
1460 __ jcc(Assembler::notEqual, L);
1461 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1462 __ bind(L);
1463 }
1464 break;
1465 case Bytecodes::_f2d:
1466 __ cvtss2sd(xmm0, xmm0);
1467 break;
1468 case Bytecodes::_d2i:
1469 {
1470 Label L;
1471 __ cvttsd2sil(rax, xmm0);
1472 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1473 __ jcc(Assembler::notEqual, L);
1474 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1475 __ bind(L);
1476 }
1477 break;
1478 case Bytecodes::_d2l:
1479 {
1480 Label L;
1481 __ cvttsd2siq(rax, xmm0);
1482 // NaN or overflow/underflow?
1483 __ cmp64(rax, ExternalAddress((address) &is_nan));
1484 __ jcc(Assembler::notEqual, L);
1485 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1486 __ bind(L);
1487 }
1488 break;
1489 case Bytecodes::_d2f:
1490 __ cvtsd2ss(xmm0, xmm0);
1491 break;
1492 default:
1493 ShouldNotReachHere();
1494 }
1495 }
1496
1497 void TemplateTable::lcmp() {
1498 transition(ltos, itos);
1499 Label done;
1500 __ pop_l(rdx);
1501 __ cmpq(rdx, rax);
1502 __ movl(rax, -1);
1503 __ jccb(Assembler::less, done);
1504 __ setb(Assembler::notEqual, rax);
1505 __ movzbl(rax, rax);
1506 __ bind(done);
1507 }
1508
1509 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1510 Label done;
1511 if (is_float) {
1512 // XXX get rid of pop here, use ... reg, mem32
1513 __ pop_f(xmm1);
1514 __ ucomiss(xmm1, xmm0);
1515 } else {
1516 // XXX get rid of pop here, use ... reg, mem64
1517 __ pop_d(xmm1);
1518 __ ucomisd(xmm1, xmm0);
1519 }
1520 if (unordered_result < 0) {
1521 __ movl(rax, -1);
1522 __ jccb(Assembler::parity, done);
1523 __ jccb(Assembler::below, done);
1524 __ setb(Assembler::notEqual, rdx);
1525 __ movzbl(rax, rdx);
1526 } else {
1527 __ movl(rax, 1);
1528 __ jccb(Assembler::parity, done);
1529 __ jccb(Assembler::above, done);
1530 __ movl(rax, 0);
1531 __ jccb(Assembler::equal, done);
1532 __ decrementl(rax);
1533 }
1534 __ bind(done);
1535 }
1536
1537 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1538 __ get_method(rcx); // rcx holds method
1539 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
1540 // holds bumped taken count
1541
1542 const ByteSize be_offset = methodOopDesc::backedge_counter_offset() +
1543 InvocationCounter::counter_offset();
1544 const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() +
1545 InvocationCounter::counter_offset();
1546 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1547
1548 // Load up edx with the branch displacement
1549 __ movl(rdx, at_bcp(1));
1550 __ bswapl(rdx);
1551
1552 if (!is_wide) {
1553 __ sarl(rdx, 16);
1554 }
1555 __ movl2ptr(rdx, rdx);
1556
1557 // Handle all the JSR stuff here, then exit.
1558 // It's much shorter and cleaner than intermingling with the non-JSR
1559 // normal-branch stuff occurring below.
1560 if (is_jsr) {
1561 // Pre-load the next target bytecode into rbx
1562 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1, 0));
1563
1564 // compute return address as bci in rax
1565 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
1566 in_bytes(constMethodOopDesc::codes_offset())));
1567 __ subptr(rax, Address(rcx, methodOopDesc::const_offset()));
1568 // Adjust the bcp in r13 by the displacement in rdx
1569 __ addptr(r13, rdx);
1570 // jsr returns atos that is not an oop
1571 __ push_i(rax);
1572 __ dispatch_only(vtos);
1573 return;
1574 }
1575
1576 // Normal (non-jsr) branch handling
1577
1578 // Adjust the bcp in r13 by the displacement in rdx
1579 __ addptr(r13, rdx);
1580
1581 assert(UseLoopCounter || !UseOnStackReplacement,
1582 "on-stack-replacement requires loop counters");
1583 Label backedge_counter_overflow;
1584 Label profile_method;
1585 Label dispatch;
1586 if (UseLoopCounter) {
1587 // increment backedge counter for backward branches
1588 // rax: MDO
1589 // ebx: MDO bumped taken-count
1590 // rcx: method
1591 // rdx: target offset
1592 // r13: target bcp
1593 // r14: locals pointer
1594 __ testl(rdx, rdx); // check if forward or backward branch
1595 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1596
1597 // increment counter
1598 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1599 __ incrementl(rax, InvocationCounter::count_increment); // increment
1600 // counter
1601 __ movl(Address(rcx, be_offset), rax); // store counter
1602
1603 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1604 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1605 __ addl(rax, Address(rcx, be_offset)); // add both counters
1606
1607 if (ProfileInterpreter) {
1608 // Test to see if we should create a method data oop
1609 __ cmp32(rax,
1610 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1611 __ jcc(Assembler::less, dispatch);
1612
1613 // if no method data exists, go to profile method
1614 __ test_method_data_pointer(rax, profile_method);
1615
1616 if (UseOnStackReplacement) {
1617 // check for overflow against ebx which is the MDO taken count
1618 __ cmp32(rbx,
1619 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1620 __ jcc(Assembler::below, dispatch);
1621
1622 // When ProfileInterpreter is on, the backedge_count comes
1623 // from the methodDataOop, which value does not get reset on
1624 // the call to frequency_counter_overflow(). To avoid
1625 // excessive calls to the overflow routine while the method is
1626 // being compiled, add a second test to make sure the overflow
1627 // function is called only once every overflow_frequency.
1628 const int overflow_frequency = 1024;
1629 __ andl(rbx, overflow_frequency - 1);
1630 __ jcc(Assembler::zero, backedge_counter_overflow);
1631
1632 }
1633 } else {
1634 if (UseOnStackReplacement) {
1635 // check for overflow against eax, which is the sum of the
1636 // counters
1637 __ cmp32(rax,
1638 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1639 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1640
1641 }
1642 }
1643 __ bind(dispatch);
1644 }
1645
1646 // Pre-load the next target bytecode into rbx
1647 __ load_unsigned_byte(rbx, Address(r13, 0));
1648
1649 // continue with the bytecode @ target
1650 // eax: return bci for jsr's, unused otherwise
1651 // ebx: target bytecode
1652 // r13: target bcp
1653 __ dispatch_only(vtos);
1654
1655 if (UseLoopCounter) {
1656 if (ProfileInterpreter) {
1657 // Out-of-line code to allocate method data oop.
1658 __ bind(profile_method);
1659 __ call_VM(noreg,
1660 CAST_FROM_FN_PTR(address,
1661 InterpreterRuntime::profile_method), r13);
1662 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1663 __ movptr(rcx, Address(rbp, method_offset));
1664 __ movptr(rcx, Address(rcx,
1665 in_bytes(methodOopDesc::method_data_offset())));
1666 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
1667 rcx);
1668 __ test_method_data_pointer(rcx, dispatch);
1669 // offset non-null mdp by MDO::data_offset() + IR::profile_method()
1670 __ addptr(rcx, in_bytes(methodDataOopDesc::data_offset()));
1671 __ addptr(rcx, rax);
1672 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize),
1673 rcx);
1674 __ jmp(dispatch);
1675 }
1676
1677 if (UseOnStackReplacement) {
1678 // invocation counter overflow
1679 __ bind(backedge_counter_overflow);
1680 __ negptr(rdx);
1681 __ addptr(rdx, r13); // branch bcp
1682 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1683 __ call_VM(noreg,
1684 CAST_FROM_FN_PTR(address,
1685 InterpreterRuntime::frequency_counter_overflow),
1686 rdx);
1687 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1688
1689 // rax: osr nmethod (osr ok) or NULL (osr not possible)
1690 // ebx: target bytecode
1691 // rdx: scratch
1692 // r14: locals pointer
1693 // r13: bcp
1694 __ testptr(rax, rax); // test result
1695 __ jcc(Assembler::zero, dispatch); // no osr if null
1696 // nmethod may have been invalidated (VM may block upon call_VM return)
1697 __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1698 __ cmpl(rcx, InvalidOSREntryBci);
1699 __ jcc(Assembler::equal, dispatch);
1700
1701 // We have the address of an on stack replacement routine in eax
1702 // We need to prepare to execute the OSR method. First we must
1703 // migrate the locals and monitors off of the stack.
1704
1705 __ mov(r13, rax); // save the nmethod
1706
1707 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1708
1709 // eax is OSR buffer, move it to expected parameter location
1710 __ mov(j_rarg0, rax);
1711
1712 // We use j_rarg definitions here so that registers don't conflict as parameter
1713 // registers change across platforms as we are in the midst of a calling
1714 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
1715
1716 const Register retaddr = j_rarg2;
1717 const Register sender_sp = j_rarg1;
1718
1719 // pop the interpreter frame
1720 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1721 __ leave(); // remove frame anchor
1722 __ pop(retaddr); // get return address
1723 __ mov(rsp, sender_sp); // set sp to sender sp
1724 // Ensure compiled code always sees stack at proper alignment
1725 __ andptr(rsp, -(StackAlignmentInBytes));
1726
1727 // unlike x86 we need no specialized return from compiled code
1728 // to the interpreter or the call stub.
1729
1730 // push the return address
1731 __ push(retaddr);
1732
1733 // and begin the OSR nmethod
1734 __ jmp(Address(r13, nmethod::osr_entry_point_offset()));
1735 }
1736 }
1737 }
1738
1739
1740 void TemplateTable::if_0cmp(Condition cc) {
1741 transition(itos, vtos);
1742 // assume branch is more often taken than not (loops use backward branches)
1743 Label not_taken;
1744 __ testl(rax, rax);
1745 __ jcc(j_not(cc), not_taken);
1746 branch(false, false);
1747 __ bind(not_taken);
1748 __ profile_not_taken_branch(rax);
1749 }
1750
1751 void TemplateTable::if_icmp(Condition cc) {
1752 transition(itos, vtos);
1753 // assume branch is more often taken than not (loops use backward branches)
1754 Label not_taken;
1755 __ pop_i(rdx);
1756 __ cmpl(rdx, rax);
1757 __ jcc(j_not(cc), not_taken);
1758 branch(false, false);
1759 __ bind(not_taken);
1760 __ profile_not_taken_branch(rax);
1761 }
1762
1763 void TemplateTable::if_nullcmp(Condition cc) {
1764 transition(atos, vtos);
1765 // assume branch is more often taken than not (loops use backward branches)
1766 Label not_taken;
1767 __ testptr(rax, rax);
1768 __ jcc(j_not(cc), not_taken);
1769 branch(false, false);
1770 __ bind(not_taken);
1771 __ profile_not_taken_branch(rax);
1772 }
1773
1774 void TemplateTable::if_acmp(Condition cc) {
1775 transition(atos, vtos);
1776 // assume branch is more often taken than not (loops use backward branches)
1777 Label not_taken;
1778 __ pop_ptr(rdx);
1779 __ cmpptr(rdx, rax);
1780 __ jcc(j_not(cc), not_taken);
1781 branch(false, false);
1782 __ bind(not_taken);
1783 __ profile_not_taken_branch(rax);
1784 }
1785
1786 void TemplateTable::ret() {
1787 transition(vtos, vtos);
1788 locals_index(rbx);
1789 __ movslq(rbx, iaddress(rbx)); // get return bci, compute return bcp
1790 __ profile_ret(rbx, rcx);
1791 __ get_method(rax);
1792 __ movptr(r13, Address(rax, methodOopDesc::const_offset()));
1793 __ lea(r13, Address(r13, rbx, Address::times_1,
1794 constMethodOopDesc::codes_offset()));
1795 __ dispatch_next(vtos);
1796 }
1797
1798 void TemplateTable::wide_ret() {
1799 transition(vtos, vtos);
1800 locals_index_wide(rbx);
1801 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
1802 __ profile_ret(rbx, rcx);
1803 __ get_method(rax);
1804 __ movptr(r13, Address(rax, methodOopDesc::const_offset()));
1805 __ lea(r13, Address(r13, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1806 __ dispatch_next(vtos);
1807 }
1808
1809 void TemplateTable::tableswitch() {
1810 Label default_case, continue_execution;
1811 transition(itos, vtos);
1812 // align r13
1813 __ lea(rbx, at_bcp(BytesPerInt));
1814 __ andptr(rbx, -BytesPerInt);
1815 // load lo & hi
1816 __ movl(rcx, Address(rbx, BytesPerInt));
1817 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
1818 __ bswapl(rcx);
1819 __ bswapl(rdx);
1820 // check against lo & hi
1821 __ cmpl(rax, rcx);
1822 __ jcc(Assembler::less, default_case);
1823 __ cmpl(rax, rdx);
1824 __ jcc(Assembler::greater, default_case);
1825 // lookup dispatch offset
1826 __ subl(rax, rcx);
1827 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1828 __ profile_switch_case(rax, rbx, rcx);
1829 // continue execution
1830 __ bind(continue_execution);
1831 __ bswapl(rdx);
1832 __ movl2ptr(rdx, rdx);
1833 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1834 __ addptr(r13, rdx);
1835 __ dispatch_only(vtos);
1836 // handle default
1837 __ bind(default_case);
1838 __ profile_switch_default(rax);
1839 __ movl(rdx, Address(rbx, 0));
1840 __ jmp(continue_execution);
1841 }
1842
1843 void TemplateTable::lookupswitch() {
1844 transition(itos, itos);
1845 __ stop("lookupswitch bytecode should have been rewritten");
1846 }
1847
1848 void TemplateTable::fast_linearswitch() {
1849 transition(itos, vtos);
1850 Label loop_entry, loop, found, continue_execution;
1851 // bswap rax so we can avoid bswapping the table entries
1852 __ bswapl(rax);
1853 // align r13
1854 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
1855 // this instruction (change offsets
1856 // below)
1857 __ andptr(rbx, -BytesPerInt);
1858 // set counter
1859 __ movl(rcx, Address(rbx, BytesPerInt));
1860 __ bswapl(rcx);
1861 __ jmpb(loop_entry);
1862 // table search
1863 __ bind(loop);
1864 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
1865 __ jcc(Assembler::equal, found);
1866 __ bind(loop_entry);
1867 __ decrementl(rcx);
1868 __ jcc(Assembler::greaterEqual, loop);
1869 // default case
1870 __ profile_switch_default(rax);
1871 __ movl(rdx, Address(rbx, 0));
1872 __ jmp(continue_execution);
1873 // entry found -> get offset
1874 __ bind(found);
1875 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
1876 __ profile_switch_case(rcx, rax, rbx);
1877 // continue execution
1878 __ bind(continue_execution);
1879 __ bswapl(rdx);
1880 __ movl2ptr(rdx, rdx);
1881 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1882 __ addptr(r13, rdx);
1883 __ dispatch_only(vtos);
1884 }
1885
1886 void TemplateTable::fast_binaryswitch() {
1887 transition(itos, vtos);
1888 // Implementation using the following core algorithm:
1889 //
1890 // int binary_search(int key, LookupswitchPair* array, int n) {
1891 // // Binary search according to "Methodik des Programmierens" by
1892 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1893 // int i = 0;
1894 // int j = n;
1895 // while (i+1 < j) {
1896 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1897 // // with Q: for all i: 0 <= i < n: key < a[i]
1898 // // where a stands for the array and assuming that the (inexisting)
1899 // // element a[n] is infinitely big.
1900 // int h = (i + j) >> 1;
1901 // // i < h < j
1902 // if (key < array[h].fast_match()) {
1903 // j = h;
1904 // } else {
1905 // i = h;
1906 // }
1907 // }
1908 // // R: a[i] <= key < a[i+1] or Q
1909 // // (i.e., if key is within array, i is the correct index)
1910 // return i;
1911 // }
1912
1913 // Register allocation
1914 const Register key = rax; // already set (tosca)
1915 const Register array = rbx;
1916 const Register i = rcx;
1917 const Register j = rdx;
1918 const Register h = rdi;
1919 const Register temp = rsi;
1920
1921 // Find array start
1922 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
1923 // get rid of this
1924 // instruction (change
1925 // offsets below)
1926 __ andptr(array, -BytesPerInt);
1927
1928 // Initialize i & j
1929 __ xorl(i, i); // i = 0;
1930 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
1931
1932 // Convert j into native byteordering
1933 __ bswapl(j);
1934
1935 // And start
1936 Label entry;
1937 __ jmp(entry);
1938
1939 // binary search loop
1940 {
1941 Label loop;
1942 __ bind(loop);
1943 // int h = (i + j) >> 1;
1944 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1945 __ sarl(h, 1); // h = (i + j) >> 1;
1946 // if (key < array[h].fast_match()) {
1947 // j = h;
1948 // } else {
1949 // i = h;
1950 // }
1951 // Convert array[h].match to native byte-ordering before compare
1952 __ movl(temp, Address(array, h, Address::times_8));
1953 __ bswapl(temp);
1954 __ cmpl(key, temp);
1955 // j = h if (key < array[h].fast_match())
1956 __ cmovl(Assembler::less, j, h);
1957 // i = h if (key >= array[h].fast_match())
1958 __ cmovl(Assembler::greaterEqual, i, h);
1959 // while (i+1 < j)
1960 __ bind(entry);
1961 __ leal(h, Address(i, 1)); // i+1
1962 __ cmpl(h, j); // i+1 < j
1963 __ jcc(Assembler::less, loop);
1964 }
1965
1966 // end of binary search, result index is i (must check again!)
1967 Label default_case;
1968 // Convert array[i].match to native byte-ordering before compare
1969 __ movl(temp, Address(array, i, Address::times_8));
1970 __ bswapl(temp);
1971 __ cmpl(key, temp);
1972 __ jcc(Assembler::notEqual, default_case);
1973
1974 // entry found -> j = offset
1975 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
1976 __ profile_switch_case(i, key, array);
1977 __ bswapl(j);
1978 __ movl2ptr(j, j);
1979 __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
1980 __ addptr(r13, j);
1981 __ dispatch_only(vtos);
1982
1983 // default case -> j = default offset
1984 __ bind(default_case);
1985 __ profile_switch_default(i);
1986 __ movl(j, Address(array, -2 * BytesPerInt));
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
1994
1995 void TemplateTable::_return(TosState state) {
1996 transition(state, state);
1997 assert(_desc->calls_vm(),
1998 "inconsistent calls_vm information"); // call in remove_activation
1999
2000 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2001 assert(state == vtos, "only valid state");
2002 __ movptr(c_rarg1, aaddress(0));
2003 __ load_klass(rdi, c_rarg1);
2004 __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
2005 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2006 Label skip_register_finalizer;
2007 __ jcc(Assembler::zero, skip_register_finalizer);
2008
2009 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2010
2011 __ bind(skip_register_finalizer);
2012 }
2013
2014 __ remove_activation(state, r13);
2015 __ jmp(r13);
2016 }
2017
2018 // ----------------------------------------------------------------------------
2019 // Volatile variables demand their effects be made known to all CPU's
2020 // in order. Store buffers on most chips allow reads & writes to
2021 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2022 // without some kind of memory barrier (i.e., it's not sufficient that
2023 // the interpreter does not reorder volatile references, the hardware
2024 // also must not reorder them).
2025 //
2026 // According to the new Java Memory Model (JMM):
2027 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2028 // writes act as aquire & release, so:
2029 // (2) A read cannot let unrelated NON-volatile memory refs that
2030 // happen after the read float up to before the read. It's OK for
2031 // non-volatile memory refs that happen before the volatile read to
2032 // float down below it.
2033 // (3) Similar a volatile write cannot let unrelated NON-volatile
2034 // memory refs that happen BEFORE the write float down to after the
2035 // write. It's OK for non-volatile memory refs that happen after the
2036 // volatile write to float up before it.
2037 //
2038 // We only put in barriers around volatile refs (they are expensive),
2039 // not _between_ memory refs (that would require us to track the
2040 // flavor of the previous memory refs). Requirements (2) and (3)
2041 // require some barriers before volatile stores and after volatile
2042 // loads. These nearly cover requirement (1) but miss the
2043 // volatile-store-volatile-load case. This final case is placed after
2044 // volatile-stores although it could just as well go before
2045 // volatile-loads.
2046 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits
2047 order_constraint) {
2048 // Helper function to insert a is-volatile test and memory barrier
2049 if (os::is_MP()) { // Not needed on single CPU
2050 __ membar(order_constraint);
2051 }
2052 }
2053
2054 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register index) {
2055 assert(byte_no == 1 || byte_no == 2, "byte_no out of range");
2056 bool is_invokedynamic = (bytecode() == Bytecodes::_invokedynamic);
2057
2058 const Register temp = rbx;
2059 assert_different_registers(Rcache, index, temp);
2060
2061 const int shift_count = (1 + byte_no) * BitsPerByte;
2062 Label resolved;
2063 __ get_cache_and_index_at_bcp(Rcache, index, 1, is_invokedynamic);
2064 if (is_invokedynamic) {
2065 // we are resolved if the f1 field contains a non-null CallSite object
2066 __ cmpptr(Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()), (int32_t) NULL_WORD);
2067 __ jcc(Assembler::notEqual, resolved);
2068 } else {
2069 __ movl(temp, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
2070 __ shrl(temp, shift_count);
2071 // have we resolved this bytecode?
2072 __ andl(temp, 0xFF);
2073 __ cmpl(temp, (int) bytecode());
2074 __ jcc(Assembler::equal, resolved);
2075 }
2076
2077 // resolve first time through
2078 address entry;
2079 switch (bytecode()) {
2080 case Bytecodes::_getstatic:
2081 case Bytecodes::_putstatic:
2082 case Bytecodes::_getfield:
2083 case Bytecodes::_putfield:
2084 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
2085 break;
2086 case Bytecodes::_invokevirtual:
2087 case Bytecodes::_invokespecial:
2088 case Bytecodes::_invokestatic:
2089 case Bytecodes::_invokeinterface:
2090 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
2091 break;
2092 case Bytecodes::_invokedynamic:
2093 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
2094 break;
2095 default:
2096 ShouldNotReachHere();
2097 break;
2098 }
2099 __ movl(temp, (int) bytecode());
2100 __ call_VM(noreg, entry, temp);
2101
2102 // Update registers with resolved info
2103 __ get_cache_and_index_at_bcp(Rcache, index, 1, is_invokedynamic);
2104 __ bind(resolved);
2105 }
2106
2107 // The Rcache and index registers must be set before call
2108 void TemplateTable::load_field_cp_cache_entry(Register obj,
2109 Register cache,
2110 Register index,
2111 Register off,
2112 Register flags,
2113 bool is_static = false) {
2114 assert_different_registers(cache, index, flags, off);
2115
2116 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2117 // Field offset
2118 __ movptr(off, Address(cache, index, Address::times_8,
2119 in_bytes(cp_base_offset +
2120 ConstantPoolCacheEntry::f2_offset())));
2121 // Flags
2122 __ movl(flags, Address(cache, index, Address::times_8,
2123 in_bytes(cp_base_offset +
2124 ConstantPoolCacheEntry::flags_offset())));
2125
2126 // klass overwrite register
2127 if (is_static) {
2128 __ movptr(obj, Address(cache, index, Address::times_8,
2129 in_bytes(cp_base_offset +
2130 ConstantPoolCacheEntry::f1_offset())));
2131 }
2132 }
2133
2134 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2135 Register method,
2136 Register itable_index,
2137 Register flags,
2138 bool is_invokevirtual,
2139 bool is_invokevfinal /*unused*/) {
2140 // setup registers
2141 const Register cache = rcx;
2142 const Register index = rdx;
2143 assert_different_registers(method, flags);
2144 assert_different_registers(method, cache, index);
2145 assert_different_registers(itable_index, flags);
2146 assert_different_registers(itable_index, cache, index);
2147 // determine constant pool cache field offsets
2148 const int method_offset = in_bytes(
2149 constantPoolCacheOopDesc::base_offset() +
2150 (is_invokevirtual
2151 ? ConstantPoolCacheEntry::f2_offset()
2152 : ConstantPoolCacheEntry::f1_offset()));
2153 const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2154 ConstantPoolCacheEntry::flags_offset());
2155 // access constant pool cache fields
2156 const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2157 ConstantPoolCacheEntry::f2_offset());
2158
2159 resolve_cache_and_index(byte_no, cache, index);
2160
2161 assert(wordSize == 8, "adjust code below");
2162 __ movptr(method, Address(cache, index, Address::times_8, method_offset));
2163 if (itable_index != noreg) {
2164 __ movptr(itable_index,
2165 Address(cache, index, Address::times_8, index_offset));
2166 }
2167 __ movl(flags , Address(cache, index, Address::times_8, flags_offset));
2168 }
2169
2170
2171 // The registers cache and index expected to be set before call.
2172 // Correct values of the cache and index registers are preserved.
2173 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2174 bool is_static, bool has_tos) {
2175 // do the JVMTI work here to avoid disturbing the register state below
2176 // We use c_rarg registers here because we want to use the register used in
2177 // the call to the VM
2178 if (JvmtiExport::can_post_field_access()) {
2179 // Check to see if a field access watch has been set before we
2180 // take the time to call into the VM.
2181 Label L1;
2182 assert_different_registers(cache, index, rax);
2183 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2184 __ testl(rax, rax);
2185 __ jcc(Assembler::zero, L1);
2186
2187 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2188
2189 // cache entry pointer
2190 __ addptr(c_rarg2, in_bytes(constantPoolCacheOopDesc::base_offset()));
2191 __ shll(c_rarg3, LogBytesPerWord);
2192 __ addptr(c_rarg2, c_rarg3);
2193 if (is_static) {
2194 __ xorl(c_rarg1, c_rarg1); // NULL object reference
2195 } else {
2196 __ movptr(c_rarg1, at_tos()); // get object pointer without popping it
2197 __ verify_oop(c_rarg1);
2198 }
2199 // c_rarg1: object pointer or NULL
2200 // c_rarg2: cache entry pointer
2201 // c_rarg3: jvalue object on the stack
2202 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2203 InterpreterRuntime::post_field_access),
2204 c_rarg1, c_rarg2, c_rarg3);
2205 __ get_cache_and_index_at_bcp(cache, index, 1);
2206 __ bind(L1);
2207 }
2208 }
2209
2210 void TemplateTable::pop_and_check_object(Register r) {
2211 __ pop_ptr(r);
2212 __ null_check(r); // for field access must check obj.
2213 __ verify_oop(r);
2214 }
2215
2216 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2217 transition(vtos, vtos);
2218
2219 const Register cache = rcx;
2220 const Register index = rdx;
2221 const Register obj = c_rarg3;
2222 const Register off = rbx;
2223 const Register flags = rax;
2224 const Register bc = c_rarg3; // uses same reg as obj, so don't mix them
2225
2226 resolve_cache_and_index(byte_no, cache, index);
2227 jvmti_post_field_access(cache, index, is_static, false);
2228 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2229
2230 if (!is_static) {
2231 // obj is on the stack
2232 pop_and_check_object(obj);
2233 }
2234
2235 const Address field(obj, off, Address::times_1);
2236
2237 Label Done, notByte, notInt, notShort, notChar,
2238 notLong, notFloat, notObj, notDouble;
2239
2240 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2241 assert(btos == 0, "change code, btos != 0");
2242
2243 __ andl(flags, 0x0F);
2244 __ jcc(Assembler::notZero, notByte);
2245 // btos
2246 __ load_signed_byte(rax, field);
2247 __ push(btos);
2248 // Rewrite bytecode to be faster
2249 if (!is_static) {
2250 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2251 }
2252 __ jmp(Done);
2253
2254 __ bind(notByte);
2255 __ cmpl(flags, atos);
2256 __ jcc(Assembler::notEqual, notObj);
2257 // atos
2258 __ load_heap_oop(rax, field);
2259 __ push(atos);
2260 if (!is_static) {
2261 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2262 }
2263 __ jmp(Done);
2264
2265 __ bind(notObj);
2266 __ cmpl(flags, itos);
2267 __ jcc(Assembler::notEqual, notInt);
2268 // itos
2269 __ movl(rax, field);
2270 __ push(itos);
2271 // Rewrite bytecode to be faster
2272 if (!is_static) {
2273 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2274 }
2275 __ jmp(Done);
2276
2277 __ bind(notInt);
2278 __ cmpl(flags, ctos);
2279 __ jcc(Assembler::notEqual, notChar);
2280 // ctos
2281 __ load_unsigned_short(rax, field);
2282 __ push(ctos);
2283 // Rewrite bytecode to be faster
2284 if (!is_static) {
2285 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2286 }
2287 __ jmp(Done);
2288
2289 __ bind(notChar);
2290 __ cmpl(flags, stos);
2291 __ jcc(Assembler::notEqual, notShort);
2292 // stos
2293 __ load_signed_short(rax, field);
2294 __ push(stos);
2295 // Rewrite bytecode to be faster
2296 if (!is_static) {
2297 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2298 }
2299 __ jmp(Done);
2300
2301 __ bind(notShort);
2302 __ cmpl(flags, ltos);
2303 __ jcc(Assembler::notEqual, notLong);
2304 // ltos
2305 __ movq(rax, field);
2306 __ push(ltos);
2307 // Rewrite bytecode to be faster
2308 if (!is_static) {
2309 patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx);
2310 }
2311 __ jmp(Done);
2312
2313 __ bind(notLong);
2314 __ cmpl(flags, ftos);
2315 __ jcc(Assembler::notEqual, notFloat);
2316 // ftos
2317 __ movflt(xmm0, field);
2318 __ push(ftos);
2319 // Rewrite bytecode to be faster
2320 if (!is_static) {
2321 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2322 }
2323 __ jmp(Done);
2324
2325 __ bind(notFloat);
2326 #ifdef ASSERT
2327 __ cmpl(flags, dtos);
2328 __ jcc(Assembler::notEqual, notDouble);
2329 #endif
2330 // dtos
2331 __ movdbl(xmm0, field);
2332 __ push(dtos);
2333 // Rewrite bytecode to be faster
2334 if (!is_static) {
2335 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2336 }
2337 #ifdef ASSERT
2338 __ jmp(Done);
2339
2340 __ bind(notDouble);
2341 __ stop("Bad state");
2342 #endif
2343
2344 __ bind(Done);
2345 // [jk] not needed currently
2346 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2347 // Assembler::LoadStore));
2348 }
2349
2350
2351 void TemplateTable::getfield(int byte_no) {
2352 getfield_or_static(byte_no, false);
2353 }
2354
2355 void TemplateTable::getstatic(int byte_no) {
2356 getfield_or_static(byte_no, true);
2357 }
2358
2359 // The registers cache and index expected to be set before call.
2360 // The function may destroy various registers, just not the cache and index registers.
2361 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2362 transition(vtos, vtos);
2363
2364 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2365
2366 if (JvmtiExport::can_post_field_modification()) {
2367 // Check to see if a field modification watch has been set before
2368 // we take the time to call into the VM.
2369 Label L1;
2370 assert_different_registers(cache, index, rax);
2371 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2372 __ testl(rax, rax);
2373 __ jcc(Assembler::zero, L1);
2374
2375 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2376
2377 if (is_static) {
2378 // Life is simple. Null out the object pointer.
2379 __ xorl(c_rarg1, c_rarg1);
2380 } else {
2381 // Life is harder. The stack holds the value on top, followed by
2382 // the object. We don't know the size of the value, though; it
2383 // could be one or two words depending on its type. As a result,
2384 // we must find the type to determine where the object is.
2385 __ movl(c_rarg3, Address(c_rarg2, rscratch1,
2386 Address::times_8,
2387 in_bytes(cp_base_offset +
2388 ConstantPoolCacheEntry::flags_offset())));
2389 __ shrl(c_rarg3, ConstantPoolCacheEntry::tosBits);
2390 // Make sure we don't need to mask rcx for tosBits after the
2391 // above shift
2392 ConstantPoolCacheEntry::verify_tosBits();
2393 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2394 __ cmpl(c_rarg3, ltos);
2395 __ cmovptr(Assembler::equal,
2396 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2397 __ cmpl(c_rarg3, dtos);
2398 __ cmovptr(Assembler::equal,
2399 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2400 }
2401 // cache entry pointer
2402 __ addptr(c_rarg2, in_bytes(cp_base_offset));
2403 __ shll(rscratch1, LogBytesPerWord);
2404 __ addptr(c_rarg2, rscratch1);
2405 // object (tos)
2406 __ mov(c_rarg3, rsp);
2407 // c_rarg1: object pointer set up above (NULL if static)
2408 // c_rarg2: cache entry pointer
2409 // c_rarg3: jvalue object on the stack
2410 __ call_VM(noreg,
2411 CAST_FROM_FN_PTR(address,
2412 InterpreterRuntime::post_field_modification),
2413 c_rarg1, c_rarg2, c_rarg3);
2414 __ get_cache_and_index_at_bcp(cache, index, 1);
2415 __ bind(L1);
2416 }
2417 }
2418
2419 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2420 transition(vtos, vtos);
2421
2422 const Register cache = rcx;
2423 const Register index = rdx;
2424 const Register obj = rcx;
2425 const Register off = rbx;
2426 const Register flags = rax;
2427 const Register bc = c_rarg3;
2428
2429 resolve_cache_and_index(byte_no, cache, index);
2430 jvmti_post_field_mod(cache, index, is_static);
2431 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2432
2433 // [jk] not needed currently
2434 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2435 // Assembler::StoreStore));
2436
2437 Label notVolatile, Done;
2438 __ movl(rdx, flags);
2439 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2440 __ andl(rdx, 0x1);
2441
2442 // field address
2443 const Address field(obj, off, Address::times_1);
2444
2445 Label notByte, notInt, notShort, notChar,
2446 notLong, notFloat, notObj, notDouble;
2447
2448 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2449
2450 assert(btos == 0, "change code, btos != 0");
2451 __ andl(flags, 0x0f);
2452 __ jcc(Assembler::notZero, notByte);
2453 // btos
2454 __ pop(btos);
2455 if (!is_static) pop_and_check_object(obj);
2456 __ movb(field, rax);
2457 if (!is_static) {
2458 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx);
2459 }
2460 __ jmp(Done);
2461
2462 __ bind(notByte);
2463 __ cmpl(flags, atos);
2464 __ jcc(Assembler::notEqual, notObj);
2465 // atos
2466 __ pop(atos);
2467 if (!is_static) pop_and_check_object(obj);
2468
2469 // Store into the field
2470 do_oop_store(_masm, field, rax, _bs->kind(), false);
2471
2472 if (!is_static) {
2473 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx);
2474 }
2475 __ jmp(Done);
2476
2477 __ bind(notObj);
2478 __ cmpl(flags, itos);
2479 __ jcc(Assembler::notEqual, notInt);
2480 // itos
2481 __ pop(itos);
2482 if (!is_static) pop_and_check_object(obj);
2483 __ movl(field, rax);
2484 if (!is_static) {
2485 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx);
2486 }
2487 __ jmp(Done);
2488
2489 __ bind(notInt);
2490 __ cmpl(flags, ctos);
2491 __ jcc(Assembler::notEqual, notChar);
2492 // ctos
2493 __ pop(ctos);
2494 if (!is_static) pop_and_check_object(obj);
2495 __ movw(field, rax);
2496 if (!is_static) {
2497 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx);
2498 }
2499 __ jmp(Done);
2500
2501 __ bind(notChar);
2502 __ cmpl(flags, stos);
2503 __ jcc(Assembler::notEqual, notShort);
2504 // stos
2505 __ pop(stos);
2506 if (!is_static) pop_and_check_object(obj);
2507 __ movw(field, rax);
2508 if (!is_static) {
2509 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx);
2510 }
2511 __ jmp(Done);
2512
2513 __ bind(notShort);
2514 __ cmpl(flags, ltos);
2515 __ jcc(Assembler::notEqual, notLong);
2516 // ltos
2517 __ pop(ltos);
2518 if (!is_static) pop_and_check_object(obj);
2519 __ movq(field, rax);
2520 if (!is_static) {
2521 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx);
2522 }
2523 __ jmp(Done);
2524
2525 __ bind(notLong);
2526 __ cmpl(flags, ftos);
2527 __ jcc(Assembler::notEqual, notFloat);
2528 // ftos
2529 __ pop(ftos);
2530 if (!is_static) pop_and_check_object(obj);
2531 __ movflt(field, xmm0);
2532 if (!is_static) {
2533 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx);
2534 }
2535 __ jmp(Done);
2536
2537 __ bind(notFloat);
2538 #ifdef ASSERT
2539 __ cmpl(flags, dtos);
2540 __ jcc(Assembler::notEqual, notDouble);
2541 #endif
2542 // dtos
2543 __ pop(dtos);
2544 if (!is_static) pop_and_check_object(obj);
2545 __ movdbl(field, xmm0);
2546 if (!is_static) {
2547 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx);
2548 }
2549
2550 #ifdef ASSERT
2551 __ jmp(Done);
2552
2553 __ bind(notDouble);
2554 __ stop("Bad state");
2555 #endif
2556
2557 __ bind(Done);
2558 // Check for volatile store
2559 __ testl(rdx, rdx);
2560 __ jcc(Assembler::zero, notVolatile);
2561 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2562 Assembler::StoreStore));
2563
2564 __ bind(notVolatile);
2565 }
2566
2567 void TemplateTable::putfield(int byte_no) {
2568 putfield_or_static(byte_no, false);
2569 }
2570
2571 void TemplateTable::putstatic(int byte_no) {
2572 putfield_or_static(byte_no, true);
2573 }
2574
2575 void TemplateTable::jvmti_post_fast_field_mod() {
2576 if (JvmtiExport::can_post_field_modification()) {
2577 // Check to see if a field modification watch has been set before
2578 // we take the time to call into the VM.
2579 Label L2;
2580 __ mov32(c_rarg3, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2581 __ testl(c_rarg3, c_rarg3);
2582 __ jcc(Assembler::zero, L2);
2583 __ pop_ptr(rbx); // copy the object pointer from tos
2584 __ verify_oop(rbx);
2585 __ push_ptr(rbx); // put the object pointer back on tos
2586 __ subptr(rsp, sizeof(jvalue)); // add space for a jvalue object
2587 __ mov(c_rarg3, rsp);
2588 const Address field(c_rarg3, 0);
2589
2590 switch (bytecode()) { // load values into the jvalue object
2591 case Bytecodes::_fast_aputfield: __ movq(field, rax); break;
2592 case Bytecodes::_fast_lputfield: __ movq(field, rax); break;
2593 case Bytecodes::_fast_iputfield: __ movl(field, rax); break;
2594 case Bytecodes::_fast_bputfield: __ movb(field, rax); break;
2595 case Bytecodes::_fast_sputfield: // fall through
2596 case Bytecodes::_fast_cputfield: __ movw(field, rax); break;
2597 case Bytecodes::_fast_fputfield: __ movflt(field, xmm0); break;
2598 case Bytecodes::_fast_dputfield: __ movdbl(field, xmm0); break;
2599 default:
2600 ShouldNotReachHere();
2601 }
2602
2603 // Save rax because call_VM() will clobber it, then use it for
2604 // JVMTI purposes
2605 __ push(rax);
2606 // access constant pool cache entry
2607 __ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1);
2608 __ verify_oop(rbx);
2609 // rbx: object pointer copied above
2610 // c_rarg2: cache entry pointer
2611 // c_rarg3: jvalue object on the stack
2612 __ call_VM(noreg,
2613 CAST_FROM_FN_PTR(address,
2614 InterpreterRuntime::post_field_modification),
2615 rbx, c_rarg2, c_rarg3);
2616 __ pop(rax); // restore lower value
2617 __ addptr(rsp, sizeof(jvalue)); // release jvalue object space
2618 __ bind(L2);
2619 }
2620 }
2621
2622 void TemplateTable::fast_storefield(TosState state) {
2623 transition(state, vtos);
2624
2625 ByteSize base = constantPoolCacheOopDesc::base_offset();
2626
2627 jvmti_post_fast_field_mod();
2628
2629 // access constant pool cache
2630 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2631
2632 // test for volatile with rdx
2633 __ movl(rdx, Address(rcx, rbx, Address::times_8,
2634 in_bytes(base +
2635 ConstantPoolCacheEntry::flags_offset())));
2636
2637 // replace index with field offset from cache entry
2638 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2639 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2640
2641 // [jk] not needed currently
2642 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2643 // Assembler::StoreStore));
2644
2645 Label notVolatile;
2646 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2647 __ andl(rdx, 0x1);
2648
2649 // Get object from stack
2650 pop_and_check_object(rcx);
2651
2652 // field address
2653 const Address field(rcx, rbx, Address::times_1);
2654
2655 // access field
2656 switch (bytecode()) {
2657 case Bytecodes::_fast_aputfield:
2658 do_oop_store(_masm, field, rax, _bs->kind(), false);
2659 break;
2660 case Bytecodes::_fast_lputfield:
2661 __ movq(field, rax);
2662 break;
2663 case Bytecodes::_fast_iputfield:
2664 __ movl(field, rax);
2665 break;
2666 case Bytecodes::_fast_bputfield:
2667 __ movb(field, rax);
2668 break;
2669 case Bytecodes::_fast_sputfield:
2670 // fall through
2671 case Bytecodes::_fast_cputfield:
2672 __ movw(field, rax);
2673 break;
2674 case Bytecodes::_fast_fputfield:
2675 __ movflt(field, xmm0);
2676 break;
2677 case Bytecodes::_fast_dputfield:
2678 __ movdbl(field, xmm0);
2679 break;
2680 default:
2681 ShouldNotReachHere();
2682 }
2683
2684 // Check for volatile store
2685 __ testl(rdx, rdx);
2686 __ jcc(Assembler::zero, notVolatile);
2687 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2688 Assembler::StoreStore));
2689 __ bind(notVolatile);
2690 }
2691
2692
2693 void TemplateTable::fast_accessfield(TosState state) {
2694 transition(atos, state);
2695
2696 // Do the JVMTI work here to avoid disturbing the register state below
2697 if (JvmtiExport::can_post_field_access()) {
2698 // Check to see if a field access watch has been set before we
2699 // take the time to call into the VM.
2700 Label L1;
2701 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2702 __ testl(rcx, rcx);
2703 __ jcc(Assembler::zero, L1);
2704 // access constant pool cache entry
2705 __ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1);
2706 __ verify_oop(rax);
2707 __ mov(r12, rax); // save object pointer before call_VM() clobbers it
2708 __ mov(c_rarg1, rax);
2709 // c_rarg1: object pointer copied above
2710 // c_rarg2: cache entry pointer
2711 __ call_VM(noreg,
2712 CAST_FROM_FN_PTR(address,
2713 InterpreterRuntime::post_field_access),
2714 c_rarg1, c_rarg2);
2715 __ mov(rax, r12); // restore object pointer
2716 __ reinit_heapbase();
2717 __ bind(L1);
2718 }
2719
2720 // access constant pool cache
2721 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2722 // replace index with field offset from cache entry
2723 // [jk] not needed currently
2724 // if (os::is_MP()) {
2725 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
2726 // in_bytes(constantPoolCacheOopDesc::base_offset() +
2727 // ConstantPoolCacheEntry::flags_offset())));
2728 // __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2729 // __ andl(rdx, 0x1);
2730 // }
2731 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2732 in_bytes(constantPoolCacheOopDesc::base_offset() +
2733 ConstantPoolCacheEntry::f2_offset())));
2734
2735 // rax: object
2736 __ verify_oop(rax);
2737 __ null_check(rax);
2738 Address field(rax, rbx, Address::times_1);
2739
2740 // access field
2741 switch (bytecode()) {
2742 case Bytecodes::_fast_agetfield:
2743 __ load_heap_oop(rax, field);
2744 __ verify_oop(rax);
2745 break;
2746 case Bytecodes::_fast_lgetfield:
2747 __ movq(rax, field);
2748 break;
2749 case Bytecodes::_fast_igetfield:
2750 __ movl(rax, field);
2751 break;
2752 case Bytecodes::_fast_bgetfield:
2753 __ movsbl(rax, field);
2754 break;
2755 case Bytecodes::_fast_sgetfield:
2756 __ load_signed_short(rax, field);
2757 break;
2758 case Bytecodes::_fast_cgetfield:
2759 __ load_unsigned_short(rax, field);
2760 break;
2761 case Bytecodes::_fast_fgetfield:
2762 __ movflt(xmm0, field);
2763 break;
2764 case Bytecodes::_fast_dgetfield:
2765 __ movdbl(xmm0, field);
2766 break;
2767 default:
2768 ShouldNotReachHere();
2769 }
2770 // [jk] not needed currently
2771 // if (os::is_MP()) {
2772 // Label notVolatile;
2773 // __ testl(rdx, rdx);
2774 // __ jcc(Assembler::zero, notVolatile);
2775 // __ membar(Assembler::LoadLoad);
2776 // __ bind(notVolatile);
2777 //};
2778 }
2779
2780 void TemplateTable::fast_xaccess(TosState state) {
2781 transition(vtos, state);
2782
2783 // get receiver
2784 __ movptr(rax, aaddress(0));
2785 debug_only(__ verify_local_tag(frame::TagReference, 0));
2786 // access constant pool cache
2787 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2788 __ movptr(rbx,
2789 Address(rcx, rdx, Address::times_8,
2790 in_bytes(constantPoolCacheOopDesc::base_offset() +
2791 ConstantPoolCacheEntry::f2_offset())));
2792 // make sure exception is reported in correct bcp range (getfield is
2793 // next instruction)
2794 __ increment(r13);
2795 __ null_check(rax);
2796 switch (state) {
2797 case itos:
2798 __ movl(rax, Address(rax, rbx, Address::times_1));
2799 break;
2800 case atos:
2801 __ load_heap_oop(rax, Address(rax, rbx, Address::times_1));
2802 __ verify_oop(rax);
2803 break;
2804 case ftos:
2805 __ movflt(xmm0, Address(rax, rbx, Address::times_1));
2806 break;
2807 default:
2808 ShouldNotReachHere();
2809 }
2810
2811 // [jk] not needed currently
2812 // if (os::is_MP()) {
2813 // Label notVolatile;
2814 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
2815 // in_bytes(constantPoolCacheOopDesc::base_offset() +
2816 // ConstantPoolCacheEntry::flags_offset())));
2817 // __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2818 // __ testl(rdx, 0x1);
2819 // __ jcc(Assembler::zero, notVolatile);
2820 // __ membar(Assembler::LoadLoad);
2821 // __ bind(notVolatile);
2822 // }
2823
2824 __ decrement(r13);
2825 }
2826
2827
2828
2829 //-----------------------------------------------------------------------------
2830 // Calls
2831
2832 void TemplateTable::count_calls(Register method, Register temp) {
2833 // implemented elsewhere
2834 ShouldNotReachHere();
2835 }
2836
2837 void TemplateTable::prepare_invoke(Register method, Register index, int byte_no) {
2838 // determine flags
2839 Bytecodes::Code code = bytecode();
2840 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2841 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
2842 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2843 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2844 const bool load_receiver = (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic);
2845 const bool receiver_null_check = is_invokespecial;
2846 const bool save_flags = is_invokeinterface || is_invokevirtual;
2847 // setup registers & access constant pool cache
2848 const Register recv = rcx;
2849 const Register flags = rdx;
2850 assert_different_registers(method, index, recv, flags);
2851
2852 // save 'interpreter return address'
2853 __ save_bcp();
2854
2855 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual);
2856
2857 // load receiver if needed (note: no return address pushed yet)
2858 if (load_receiver) {
2859 __ movl(recv, flags);
2860 __ andl(recv, 0xFF);
2861 if (TaggedStackInterpreter) __ shll(recv, 1); // index*2
2862 Address recv_addr(rsp, recv, Address::times_8, -Interpreter::expr_offset_in_bytes(1));
2863 if (is_invokedynamic) {
2864 __ lea(recv, recv_addr);
2865 } else {
2866 __ movptr(recv, recv_addr);
2867 __ verify_oop(recv);
2868 }
2869 }
2870
2871 // do null check if needed
2872 if (receiver_null_check) {
2873 __ null_check(recv);
2874 }
2875
2876 if (save_flags) {
2877 __ movl(r13, flags);
2878 }
2879
2880 // compute return type
2881 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2882 // Make sure we don't need to mask flags for tosBits after the above shift
2883 ConstantPoolCacheEntry::verify_tosBits();
2884 // load return address
2885 {
2886 address table_addr;
2887 if (is_invokeinterface || is_invokedynamic)
2888 table_addr = (address)Interpreter::return_5_addrs_by_index_table();
2889 else
2890 table_addr = (address)Interpreter::return_3_addrs_by_index_table();
2891 ExternalAddress table(table_addr);
2892 __ lea(rscratch1, table);
2893 __ movptr(flags, Address(rscratch1, flags, Address::times_ptr));
2894 }
2895
2896 // push return address
2897 __ push(flags);
2898
2899 // Restore flag field from the constant pool cache, and restore esi
2900 // for later null checks. r13 is the bytecode pointer
2901 if (save_flags) {
2902 __ movl(flags, r13);
2903 __ restore_bcp();
2904 }
2905 }
2906
2907
2908 void TemplateTable::invokevirtual_helper(Register index,
2909 Register recv,
2910 Register flags) {
2911 // Uses temporary registers rax, rdx assert_different_registers(index, recv, rax, rdx);
2912
2913 // Test for an invoke of a final method
2914 Label notFinal;
2915 __ movl(rax, flags);
2916 __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
2917 __ jcc(Assembler::zero, notFinal);
2918
2919 const Register method = index; // method must be rbx
2920 assert(method == rbx,
2921 "methodOop must be rbx for interpreter calling convention");
2922
2923 // do the call - the index is actually the method to call
2924 __ verify_oop(method);
2925
2926 // It's final, need a null check here!
2927 __ null_check(recv);
2928
2929 // profile this call
2930 __ profile_final_call(rax);
2931
2932 __ jump_from_interpreted(method, rax);
2933
2934 __ bind(notFinal);
2935
2936 // get receiver klass
2937 __ null_check(recv, oopDesc::klass_offset_in_bytes());
2938 __ load_klass(rax, recv);
2939
2940 __ verify_oop(rax);
2941
2942 // profile this call
2943 __ profile_virtual_call(rax, r14, rdx);
2944
2945 // get target methodOop & entry point
2946 const int base = instanceKlass::vtable_start_offset() * wordSize;
2947 assert(vtableEntry::size() * wordSize == 8,
2948 "adjust the scaling in the code below");
2949 __ movptr(method, Address(rax, index,
2950 Address::times_8,
2951 base + vtableEntry::method_offset_in_bytes()));
2952 __ movptr(rdx, Address(method, methodOopDesc::interpreter_entry_offset()));
2953 __ jump_from_interpreted(method, rdx);
2954 }
2955
2956
2957 void TemplateTable::invokevirtual(int byte_no) {
2958 transition(vtos, vtos);
2959 prepare_invoke(rbx, noreg, byte_no);
2960
2961 // rbx: index
2962 // rcx: receiver
2963 // rdx: flags
2964
2965 invokevirtual_helper(rbx, rcx, rdx);
2966 }
2967
2968
2969 void TemplateTable::invokespecial(int byte_no) {
2970 transition(vtos, vtos);
2971 prepare_invoke(rbx, noreg, byte_no);
2972 // do the call
2973 __ verify_oop(rbx);
2974 __ profile_call(rax);
2975 __ jump_from_interpreted(rbx, rax);
2976 }
2977
2978
2979 void TemplateTable::invokestatic(int byte_no) {
2980 transition(vtos, vtos);
2981 prepare_invoke(rbx, noreg, byte_no);
2982 // do the call
2983 __ verify_oop(rbx);
2984 __ profile_call(rax);
2985 __ jump_from_interpreted(rbx, rax);
2986 }
2987
2988 void TemplateTable::fast_invokevfinal(int byte_no) {
2989 transition(vtos, vtos);
2990 __ stop("fast_invokevfinal not used on amd64");
2991 }
2992
2993 void TemplateTable::invokeinterface(int byte_no) {
2994 transition(vtos, vtos);
2995 prepare_invoke(rax, rbx, byte_no);
2996
2997 // rax: Interface
2998 // rbx: index
2999 // rcx: receiver
3000 // rdx: flags
3001
3002 // Special case of invokeinterface called for virtual method of
3003 // java.lang.Object. See cpCacheOop.cpp for details.
3004 // This code isn't produced by javac, but could be produced by
3005 // another compliant java compiler.
3006 Label notMethod;
3007 __ movl(r14, rdx);
3008 __ andl(r14, (1 << ConstantPoolCacheEntry::methodInterface));
3009 __ jcc(Assembler::zero, notMethod);
3010
3011 invokevirtual_helper(rbx, rcx, rdx);
3012 __ bind(notMethod);
3013
3014 // Get receiver klass into rdx - also a null check
3015 __ restore_locals(); // restore r14
3016 __ load_klass(rdx, rcx);
3017 __ verify_oop(rdx);
3018
3019 // profile this call
3020 __ profile_virtual_call(rdx, r13, r14);
3021
3022 Label no_such_interface, no_such_method;
3023
3024 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3025 rdx, rax, rbx,
3026 // outputs: method, scan temp. reg
3027 rbx, r13,
3028 no_such_interface);
3029
3030 // rbx,: methodOop to call
3031 // rcx: receiver
3032 // Check for abstract method error
3033 // Note: This should be done more efficiently via a throw_abstract_method_error
3034 // interpreter entry point and a conditional jump to it in case of a null
3035 // method.
3036 __ testptr(rbx, rbx);
3037 __ jcc(Assembler::zero, no_such_method);
3038
3039 // do the call
3040 // rcx: receiver
3041 // rbx,: methodOop
3042 __ jump_from_interpreted(rbx, rdx);
3043 __ should_not_reach_here();
3044
3045 // exception handling code follows...
3046 // note: must restore interpreter registers to canonical
3047 // state for exception handling to work correctly!
3048
3049 __ bind(no_such_method);
3050 // throw exception
3051 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3052 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3053 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3054 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3055 // the call_VM checks for exception, so we should never return here.
3056 __ should_not_reach_here();
3057
3058 __ bind(no_such_interface);
3059 // throw exception
3060 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3061 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3062 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3063 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3064 InterpreterRuntime::throw_IncompatibleClassChangeError));
3065 // the call_VM checks for exception, so we should never return here.
3066 __ should_not_reach_here();
3067 return;
3068 }
3069
3070 void TemplateTable::invokedynamic(int byte_no) {
3071 transition(vtos, vtos);
3072
3073 if (!EnableInvokeDynamic) {
3074 // We should not encounter this bytecode if !EnableInvokeDynamic.
3075 // The verifier will stop it. However, if we get past the verifier,
3076 // this will stop the thread in a reasonable way, without crashing the JVM.
3077 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3078 InterpreterRuntime::throw_IncompatibleClassChangeError));
3079 // the call_VM checks for exception, so we should never return here.
3080 __ should_not_reach_here();
3081 return;
3082 }
3083
3084 prepare_invoke(rax, rbx, byte_no);
3085
3086 // rax: CallSite object (f1)
3087 // rbx: unused (f2)
3088 // rcx: receiver address
3089 // rdx: flags (unused)
3090
3091 if (ProfileInterpreter) {
3092 Label L;
3093 // %%% should make a type profile for any invokedynamic that takes a ref argument
3094 // profile this call
3095 __ profile_call(r13);
3096 }
3097
3098 __ movptr(rcx, Address(rax, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx)));
3099 __ null_check(rcx);
3100 __ prepare_to_jump_from_interpreted();
3101 __ jump_to_method_handle_entry(rcx, rdx);
3102 }
3103
3104
3105 //-----------------------------------------------------------------------------
3106 // Allocation
3107
3108 void TemplateTable::_new() {
3109 transition(vtos, atos);
3110 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3111 Label slow_case;
3112 Label done;
3113 Label initialize_header;
3114 Label initialize_object; // including clearing the fields
3115 Label allocate_shared;
3116
3117 __ get_cpool_and_tags(rsi, rax);
3118 // get instanceKlass
3119 __ movptr(rsi, Address(rsi, rdx,
3120 Address::times_8, sizeof(constantPoolOopDesc)));
3121
3122 // make sure the class we're about to instantiate has been
3123 // resolved. Note: slow_case does a pop of stack, which is why we
3124 // loaded class/pushed above
3125 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
3126 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset),
3127 JVM_CONSTANT_Class);
3128 __ jcc(Assembler::notEqual, slow_case);
3129
3130 // make sure klass is initialized & doesn't have finalizer
3131 // make sure klass is fully initialized
3132 __ cmpl(Address(rsi,
3133 instanceKlass::init_state_offset_in_bytes() +
3134 sizeof(oopDesc)),
3135 instanceKlass::fully_initialized);
3136 __ jcc(Assembler::notEqual, slow_case);
3137
3138 // get instance_size in instanceKlass (scaled to a count of bytes)
3139 __ movl(rdx,
3140 Address(rsi,
3141 Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
3142 // test to see if it has a finalizer or is malformed in some way
3143 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3144 __ jcc(Assembler::notZero, slow_case);
3145
3146 // Allocate the instance
3147 // 1) Try to allocate in the TLAB
3148 // 2) if fail and the object is large allocate in the shared Eden
3149 // 3) if the above fails (or is not applicable), go to a slow case
3150 // (creates a new TLAB, etc.)
3151
3152 const bool allow_shared_alloc =
3153 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3154
3155 if (UseTLAB) {
3156 __ movptr(rax, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
3157 __ lea(rbx, Address(rax, rdx, Address::times_1));
3158 __ cmpptr(rbx, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
3159 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3160 __ movptr(Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3161 if (ZeroTLAB) {
3162 // the fields have been already cleared
3163 __ jmp(initialize_header);
3164 } else {
3165 // initialize both the header and fields
3166 __ jmp(initialize_object);
3167 }
3168 }
3169
3170 // Allocation in the shared Eden, if allowed.
3171 //
3172 // rdx: instance size in bytes
3173 if (allow_shared_alloc) {
3174 __ bind(allocate_shared);
3175
3176 ExternalAddress top((address)Universe::heap()->top_addr());
3177 ExternalAddress end((address)Universe::heap()->end_addr());
3178
3179 const Register RtopAddr = rscratch1;
3180 const Register RendAddr = rscratch2;
3181
3182 __ lea(RtopAddr, top);
3183 __ lea(RendAddr, end);
3184 __ movptr(rax, Address(RtopAddr, 0));
3185
3186 // For retries rax gets set by cmpxchgq
3187 Label retry;
3188 __ bind(retry);
3189 __ lea(rbx, Address(rax, rdx, Address::times_1));
3190 __ cmpptr(rbx, Address(RendAddr, 0));
3191 __ jcc(Assembler::above, slow_case);
3192
3193 // Compare rax with the top addr, and if still equal, store the new
3194 // top addr in rbx at the address of the top addr pointer. Sets ZF if was
3195 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3196 //
3197 // rax: object begin
3198 // rbx: object end
3199 // rdx: instance size in bytes
3200 if (os::is_MP()) {
3201 __ lock();
3202 }
3203 __ cmpxchgptr(rbx, Address(RtopAddr, 0));
3204
3205 // if someone beat us on the allocation, try again, otherwise continue
3206 __ jcc(Assembler::notEqual, retry);
3207 }
3208
3209 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3210 // The object is initialized before the header. If the object size is
3211 // zero, go directly to the header initialization.
3212 __ bind(initialize_object);
3213 __ decrementl(rdx, sizeof(oopDesc));
3214 __ jcc(Assembler::zero, initialize_header);
3215
3216 // Initialize object fields
3217 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3218 __ shrl(rdx, LogBytesPerLong); // divide by oopSize to simplify the loop
3219 {
3220 Label loop;
3221 __ bind(loop);
3222 __ movq(Address(rax, rdx, Address::times_8,
3223 sizeof(oopDesc) - oopSize),
3224 rcx);
3225 __ decrementl(rdx);
3226 __ jcc(Assembler::notZero, loop);
3227 }
3228
3229 // initialize object header only.
3230 __ bind(initialize_header);
3231 if (UseBiasedLocking) {
3232 __ movptr(rscratch1, Address(rsi, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
3233 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()), rscratch1);
3234 } else {
3235 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()),
3236 (intptr_t) markOopDesc::prototype()); // header (address 0x1)
3237 }
3238 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3239 __ store_klass_gap(rax, rcx); // zero klass gap for compressed oops
3240 __ store_klass(rax, rsi); // store klass last
3241 __ jmp(done);
3242 }
3243
3244 {
3245 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3246 // Trigger dtrace event for fastpath
3247 __ push(atos); // save the return value
3248 __ call_VM_leaf(
3249 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3250 __ pop(atos); // restore the return value
3251 }
3252
3253 // slow case
3254 __ bind(slow_case);
3255 __ get_constant_pool(c_rarg1);
3256 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3257 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3258 __ verify_oop(rax);
3259
3260 // continue
3261 __ bind(done);
3262 }
3263
3264 void TemplateTable::newarray() {
3265 transition(itos, atos);
3266 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3267 __ movl(c_rarg2, rax);
3268 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3269 c_rarg1, c_rarg2);
3270 }
3271
3272 void TemplateTable::anewarray() {
3273 transition(itos, atos);
3274 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3275 __ get_constant_pool(c_rarg1);
3276 __ movl(c_rarg3, rax);
3277 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3278 c_rarg1, c_rarg2, c_rarg3);
3279 }
3280
3281 void TemplateTable::arraylength() {
3282 transition(atos, itos);
3283 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3284 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3285 }
3286
3287 void TemplateTable::checkcast() {
3288 transition(atos, atos);
3289 Label done, is_null, ok_is_subtype, quicked, resolved;
3290 __ testptr(rax, rax); // object is in rax
3291 __ jcc(Assembler::zero, is_null);
3292
3293 // Get cpool & tags index
3294 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3295 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3296 // See if bytecode has already been quicked
3297 __ cmpb(Address(rdx, rbx,
3298 Address::times_1,
3299 typeArrayOopDesc::header_size(T_BYTE) * wordSize),
3300 JVM_CONSTANT_Class);
3301 __ jcc(Assembler::equal, quicked);
3302 __ push(atos); // save receiver for result, and for GC
3303 __ mov(r12, rcx); // save rcx XXX
3304 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3305 __ movq(rcx, r12); // restore rcx XXX
3306 __ reinit_heapbase();
3307 __ pop_ptr(rdx); // restore receiver
3308 __ jmpb(resolved);
3309
3310 // Get superklass in rax and subklass in rbx
3311 __ bind(quicked);
3312 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
3313 __ movptr(rax, Address(rcx, rbx,
3314 Address::times_8, sizeof(constantPoolOopDesc)));
3315
3316 __ bind(resolved);
3317 __ load_klass(rbx, rdx);
3318
3319 // Generate subtype check. Blows rcx, rdi. Object in rdx.
3320 // Superklass in rax. Subklass in rbx.
3321 __ gen_subtype_check(rbx, ok_is_subtype);
3322
3323 // Come here on failure
3324 __ push_ptr(rdx);
3325 // object is at TOS
3326 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3327
3328 // Come here on success
3329 __ bind(ok_is_subtype);
3330 __ mov(rax, rdx); // Restore object in rdx
3331
3332 // Collect counts on whether this check-cast sees NULLs a lot or not.
3333 if (ProfileInterpreter) {
3334 __ jmp(done);
3335 __ bind(is_null);
3336 __ profile_null_seen(rcx);
3337 } else {
3338 __ bind(is_null); // same as 'done'
3339 }
3340 __ bind(done);
3341 }
3342
3343 void TemplateTable::instanceof() {
3344 transition(atos, itos);
3345 Label done, is_null, ok_is_subtype, quicked, resolved;
3346 __ testptr(rax, rax);
3347 __ jcc(Assembler::zero, is_null);
3348
3349 // Get cpool & tags index
3350 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3351 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3352 // See if bytecode has already been quicked
3353 __ cmpb(Address(rdx, rbx,
3354 Address::times_1,
3355 typeArrayOopDesc::header_size(T_BYTE) * wordSize),
3356 JVM_CONSTANT_Class);
3357 __ jcc(Assembler::equal, quicked);
3358
3359 __ push(atos); // save receiver for result, and for GC
3360 __ mov(r12, rcx); // save rcx
3361 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3362 __ movq(rcx, r12); // restore rcx
3363 __ reinit_heapbase();
3364 __ pop_ptr(rdx); // restore receiver
3365 __ load_klass(rdx, rdx);
3366 __ jmpb(resolved);
3367
3368 // Get superklass in rax and subklass in rdx
3369 __ bind(quicked);
3370 __ load_klass(rdx, rax);
3371 __ movptr(rax, Address(rcx, rbx,
3372 Address::times_8, sizeof(constantPoolOopDesc)));
3373
3374 __ bind(resolved);
3375
3376 // Generate subtype check. Blows rcx, rdi
3377 // Superklass in rax. Subklass in rdx.
3378 __ gen_subtype_check(rdx, ok_is_subtype);
3379
3380 // Come here on failure
3381 __ xorl(rax, rax);
3382 __ jmpb(done);
3383 // Come here on success
3384 __ bind(ok_is_subtype);
3385 __ movl(rax, 1);
3386
3387 // Collect counts on whether this test sees NULLs a lot or not.
3388 if (ProfileInterpreter) {
3389 __ jmp(done);
3390 __ bind(is_null);
3391 __ profile_null_seen(rcx);
3392 } else {
3393 __ bind(is_null); // same as 'done'
3394 }
3395 __ bind(done);
3396 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
3397 // rax = 1: obj != NULL and obj is an instanceof the specified klass
3398 }
3399
3400 //-----------------------------------------------------------------------------
3401 // Breakpoints
3402 void TemplateTable::_breakpoint() {
3403 // Note: We get here even if we are single stepping..
3404 // jbug inists on setting breakpoints at every bytecode
3405 // even if we are in single step mode.
3406
3407 transition(vtos, vtos);
3408
3409 // get the unpatched byte code
3410 __ get_method(c_rarg1);
3411 __ call_VM(noreg,
3412 CAST_FROM_FN_PTR(address,
3413 InterpreterRuntime::get_original_bytecode_at),
3414 c_rarg1, r13);
3415 __ mov(rbx, rax);
3416
3417 // post the breakpoint event
3418 __ get_method(c_rarg1);
3419 __ call_VM(noreg,
3420 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3421 c_rarg1, r13);
3422
3423 // complete the execution of original bytecode
3424 __ dispatch_only_normal(vtos);
3425 }
3426
3427 //-----------------------------------------------------------------------------
3428 // Exceptions
3429
3430 void TemplateTable::athrow() {
3431 transition(atos, vtos);
3432 __ null_check(rax);
3433 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3434 }
3435
3436 //-----------------------------------------------------------------------------
3437 // Synchronization
3438 //
3439 // Note: monitorenter & exit are symmetric routines; which is reflected
3440 // in the assembly code structure as well
3441 //
3442 // Stack layout:
3443 //
3444 // [expressions ] <--- rsp = expression stack top
3445 // ..
3446 // [expressions ]
3447 // [monitor entry] <--- monitor block top = expression stack bot
3448 // ..
3449 // [monitor entry]
3450 // [frame data ] <--- monitor block bot
3451 // ...
3452 // [saved rbp ] <--- rbp
3453 void TemplateTable::monitorenter() {
3454 transition(atos, vtos);
3455
3456 // check for NULL object
3457 __ null_check(rax);
3458
3459 const Address monitor_block_top(
3460 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3461 const Address monitor_block_bot(
3462 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3463 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3464
3465 Label allocated;
3466
3467 // initialize entry pointer
3468 __ xorl(c_rarg1, c_rarg1); // points to free slot or NULL
3469
3470 // find a free slot in the monitor block (result in c_rarg1)
3471 {
3472 Label entry, loop, exit;
3473 __ movptr(c_rarg3, monitor_block_top); // points to current entry,
3474 // starting with top-most entry
3475 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3476 // of monitor block
3477 __ jmpb(entry);
3478
3479 __ bind(loop);
3480 // check if current entry is used
3481 __ cmpptr(Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
3482 // if not used then remember entry in c_rarg1
3483 __ cmov(Assembler::equal, c_rarg1, c_rarg3);
3484 // check if current entry is for same object
3485 __ cmpptr(rax, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3486 // if same object then stop searching
3487 __ jccb(Assembler::equal, exit);
3488 // otherwise advance to next entry
3489 __ addptr(c_rarg3, entry_size);
3490 __ bind(entry);
3491 // check if bottom reached
3492 __ cmpptr(c_rarg3, c_rarg2);
3493 // if not at bottom then check this entry
3494 __ jcc(Assembler::notEqual, loop);
3495 __ bind(exit);
3496 }
3497
3498 __ testptr(c_rarg1, c_rarg1); // check if a slot has been found
3499 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
3500
3501 // allocate one if there's no free slot
3502 {
3503 Label entry, loop;
3504 // 1. compute new pointers // rsp: old expression stack top
3505 __ movptr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
3506 __ subptr(rsp, entry_size); // move expression stack top
3507 __ subptr(c_rarg1, entry_size); // move expression stack bottom
3508 __ mov(c_rarg3, rsp); // set start value for copy loop
3509 __ movptr(monitor_block_bot, c_rarg1); // set new monitor block bottom
3510 __ jmp(entry);
3511 // 2. move expression stack contents
3512 __ bind(loop);
3513 __ movptr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3514 // word from old location
3515 __ movptr(Address(c_rarg3, 0), c_rarg2); // and store it at new location
3516 __ addptr(c_rarg3, wordSize); // advance to next word
3517 __ bind(entry);
3518 __ cmpptr(c_rarg3, c_rarg1); // check if bottom reached
3519 __ jcc(Assembler::notEqual, loop); // if not at bottom then
3520 // copy next word
3521 }
3522
3523 // call run-time routine
3524 // c_rarg1: points to monitor entry
3525 __ bind(allocated);
3526
3527 // Increment bcp to point to the next bytecode, so exception
3528 // handling for async. exceptions work correctly.
3529 // The object has already been poped from the stack, so the
3530 // expression stack looks correct.
3531 __ increment(r13);
3532
3533 // store object
3534 __ movptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), rax);
3535 __ lock_object(c_rarg1);
3536
3537 // check to make sure this monitor doesn't cause stack overflow after locking
3538 __ save_bcp(); // in case of exception
3539 __ generate_stack_overflow_check(0);
3540
3541 // The bcp has already been incremented. Just need to dispatch to
3542 // next instruction.
3543 __ dispatch_next(vtos);
3544 }
3545
3546
3547 void TemplateTable::monitorexit() {
3548 transition(atos, vtos);
3549
3550 // check for NULL object
3551 __ null_check(rax);
3552
3553 const Address monitor_block_top(
3554 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3555 const Address monitor_block_bot(
3556 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3557 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3558
3559 Label found;
3560
3561 // find matching slot
3562 {
3563 Label entry, loop;
3564 __ movptr(c_rarg1, monitor_block_top); // points to current entry,
3565 // starting with top-most entry
3566 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3567 // of monitor block
3568 __ jmpb(entry);
3569
3570 __ bind(loop);
3571 // check if current entry is for same object
3572 __ cmpptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3573 // if same object then stop searching
3574 __ jcc(Assembler::equal, found);
3575 // otherwise advance to next entry
3576 __ addptr(c_rarg1, entry_size);
3577 __ bind(entry);
3578 // check if bottom reached
3579 __ cmpptr(c_rarg1, c_rarg2);
3580 // if not at bottom then check this entry
3581 __ jcc(Assembler::notEqual, loop);
3582 }
3583
3584 // error handling. Unlocking was not block-structured
3585 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3586 InterpreterRuntime::throw_illegal_monitor_state_exception));
3587 __ should_not_reach_here();
3588
3589 // call run-time routine
3590 // rsi: points to monitor entry
3591 __ bind(found);
3592 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3593 __ unlock_object(c_rarg1);
3594 __ pop_ptr(rax); // discard object
3595 }
3596
3597
3598 // Wide instructions
3599 void TemplateTable::wide() {
3600 transition(vtos, vtos);
3601 __ load_unsigned_byte(rbx, at_bcp(1));
3602 __ lea(rscratch1, ExternalAddress((address)Interpreter::_wentry_point));
3603 __ jmp(Address(rscratch1, rbx, Address::times_8));
3604 // Note: the r13 increment step is part of the individual wide
3605 // bytecode implementations
3606 }
3607
3608
3609 // Multi arrays
3610 void TemplateTable::multianewarray() {
3611 transition(vtos, atos);
3612 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3613 // last dim is on top of stack; we want address of first one:
3614 // first_addr = last_addr + (ndims - 1) * wordSize
3615 if (TaggedStackInterpreter) __ shll(rax, 1); // index*2
3616 __ lea(c_rarg1, Address(rsp, rax, Address::times_8, -wordSize));
3617 call_VM(rax,
3618 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3619 c_rarg1);
3620 __ load_unsigned_byte(rbx, at_bcp(3));
3621 if (TaggedStackInterpreter) __ shll(rbx, 1); // index*2
3622 __ lea(rsp, Address(rsp, rbx, Address::times_8));
3623 }
3624 #endif // !CC_INTERP