1 /*
2 * Copyright (c) 2003, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.inline.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #ifndef CC_INTERP
42
43 #define __ _masm->
44
45 // Platform-dependent initialization
46
47 void TemplateTable::pd_initialize() {
48 // No amd64 specific initialization
49 }
50
51 // Address computation: local variables
52
53 static inline Address iaddress(int n) {
54 return Address(r14, Interpreter::local_offset_in_bytes(n));
55 }
56
57 static inline Address laddress(int n) {
58 return iaddress(n + 1);
59 }
60
61 static inline Address faddress(int n) {
62 return iaddress(n);
63 }
64
65 static inline Address daddress(int n) {
66 return laddress(n);
67 }
68
69 static inline Address aaddress(int n) {
70 return iaddress(n);
71 }
72
73 static inline Address iaddress(Register r) {
74 return Address(r14, r, Address::times_8);
75 }
76
77 static inline Address laddress(Register r) {
78 return Address(r14, r, Address::times_8, Interpreter::local_offset_in_bytes(1));
79 }
80
81 static inline Address faddress(Register r) {
82 return iaddress(r);
83 }
84
85 static inline Address daddress(Register r) {
86 return laddress(r);
87 }
88
89 static inline Address aaddress(Register r) {
90 return iaddress(r);
91 }
92
93 static inline Address at_rsp() {
94 return Address(rsp, 0);
95 }
96
97 // At top of Java expression stack which may be different than esp(). It
98 // isn't for category 1 objects.
99 static inline Address at_tos () {
100 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
101 }
102
103 static inline Address at_tos_p1() {
104 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
105 }
106
107 static inline Address at_tos_p2() {
108 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
109 }
110
111 // Condition conversion
112 static Assembler::Condition j_not(TemplateTable::Condition cc) {
113 switch (cc) {
114 case TemplateTable::equal : return Assembler::notEqual;
115 case TemplateTable::not_equal : return Assembler::equal;
116 case TemplateTable::less : return Assembler::greaterEqual;
117 case TemplateTable::less_equal : return Assembler::greater;
118 case TemplateTable::greater : return Assembler::lessEqual;
119 case TemplateTable::greater_equal: return Assembler::less;
120 }
121 ShouldNotReachHere();
122 return Assembler::zero;
123 }
124
125
126 // Miscelaneous helper routines
127 // Store an oop (or NULL) at the address described by obj.
128 // If val == noreg this means store a NULL
129
130 static void do_oop_store(InterpreterMacroAssembler* _masm,
131 Address obj,
132 Register val,
133 BarrierSet::Name barrier,
134 bool precise) {
135 assert(val == noreg || val == rax, "parameter is just for looks");
136 switch (barrier) {
137 #if INCLUDE_ALL_GCS
138 case BarrierSet::G1SATBCT:
139 case BarrierSet::G1SATBCTLogging:
140 {
141 // flatten object address if needed
142 if (obj.index() == noreg && obj.disp() == 0) {
143 if (obj.base() != rdx) {
144 __ movq(rdx, obj.base());
145 }
146 } else {
147 __ leaq(rdx, obj);
148 }
149 __ g1_write_barrier_pre(rdx /* obj */,
150 rbx /* pre_val */,
151 r15_thread /* thread */,
152 r8 /* tmp */,
153 val != noreg /* tosca_live */,
154 false /* expand_call */);
155 if (val == noreg) {
156 __ store_heap_oop_null(Address(rdx, 0));
157 } else {
158 // G1 barrier needs uncompressed oop for region cross check.
159 Register new_val = val;
160 if (UseCompressedOops) {
161 new_val = rbx;
162 __ movptr(new_val, val);
163 }
164 __ store_heap_oop(Address(rdx, 0), val);
165 __ g1_write_barrier_post(rdx /* store_adr */,
166 new_val /* new_val */,
167 r15_thread /* thread */,
168 r8 /* tmp */,
169 rbx /* tmp2 */);
170 }
171 }
172 break;
173 #endif // INCLUDE_ALL_GCS
174 case BarrierSet::CardTableModRef:
175 case BarrierSet::CardTableExtension:
176 {
177 if (val == noreg) {
178 __ store_heap_oop_null(obj);
179 } else {
180 __ store_heap_oop(obj, val);
181 // flatten object address if needed
182 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
183 __ store_check(obj.base());
184 } else {
185 __ leaq(rdx, obj);
186 __ store_check(rdx);
187 }
188 }
189 }
190 break;
191 case BarrierSet::ModRef:
192 if (val == noreg) {
193 __ store_heap_oop_null(obj);
194 } else {
195 __ store_heap_oop(obj, val);
196 }
197 break;
198 default :
199 ShouldNotReachHere();
200
201 }
202 }
203
204 Address TemplateTable::at_bcp(int offset) {
205 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
206 return Address(r13, offset);
207 }
208
209 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
210 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
211 int byte_no) {
212 if (!RewriteBytecodes) return;
213 Label L_patch_done;
214
215 switch (bc) {
216 case Bytecodes::_fast_aputfield:
217 case Bytecodes::_fast_bputfield:
218 case Bytecodes::_fast_cputfield:
219 case Bytecodes::_fast_dputfield:
220 case Bytecodes::_fast_fputfield:
221 case Bytecodes::_fast_iputfield:
222 case Bytecodes::_fast_lputfield:
223 case Bytecodes::_fast_sputfield:
224 {
225 // We skip bytecode quickening for putfield instructions when
226 // the put_code written to the constant pool cache is zero.
227 // This is required so that every execution of this instruction
228 // calls out to InterpreterRuntime::resolve_get_put to do
229 // additional, required work.
230 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
231 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
232 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
233 __ movl(bc_reg, bc);
234 __ cmpl(temp_reg, (int) 0);
235 __ jcc(Assembler::zero, L_patch_done); // don't patch
236 }
237 break;
238 default:
239 assert(byte_no == -1, "sanity");
240 // the pair bytecodes have already done the load.
241 if (load_bc_into_bc_reg) {
242 __ movl(bc_reg, bc);
243 }
244 }
245
246 if (JvmtiExport::can_post_breakpoint()) {
247 Label L_fast_patch;
248 // if a breakpoint is present we can't rewrite the stream directly
249 __ movzbl(temp_reg, at_bcp(0));
250 __ cmpl(temp_reg, Bytecodes::_breakpoint);
251 __ jcc(Assembler::notEqual, L_fast_patch);
252 __ get_method(temp_reg);
253 // Let breakpoint table handling rewrite to quicker bytecode
254 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, r13, bc_reg);
255 #ifndef ASSERT
256 __ jmpb(L_patch_done);
257 #else
258 __ jmp(L_patch_done);
259 #endif
260 __ bind(L_fast_patch);
261 }
262
263 #ifdef ASSERT
264 Label L_okay;
265 __ load_unsigned_byte(temp_reg, at_bcp(0));
266 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
267 __ jcc(Assembler::equal, L_okay);
268 __ cmpl(temp_reg, bc_reg);
269 __ jcc(Assembler::equal, L_okay);
270 __ stop("patching the wrong bytecode");
271 __ bind(L_okay);
272 #endif
273
274 // patch bytecode
275 __ movb(at_bcp(0), bc_reg);
276 __ bind(L_patch_done);
277 }
278
279
280 // Individual instructions
281
282 void TemplateTable::nop() {
283 transition(vtos, vtos);
284 // nothing to do
285 }
286
287 void TemplateTable::shouldnotreachhere() {
288 transition(vtos, vtos);
289 __ stop("shouldnotreachhere bytecode");
290 }
291
292 void TemplateTable::aconst_null() {
293 transition(vtos, atos);
294 __ xorl(rax, rax);
295 }
296
297 void TemplateTable::iconst(int value) {
298 transition(vtos, itos);
299 if (value == 0) {
300 __ xorl(rax, rax);
301 } else {
302 __ movl(rax, value);
303 }
304 }
305
306 void TemplateTable::lconst(int value) {
307 transition(vtos, ltos);
308 if (value == 0) {
309 __ xorl(rax, rax);
310 } else {
311 __ movl(rax, value);
312 }
313 }
314
315 void TemplateTable::fconst(int value) {
316 transition(vtos, ftos);
317 static float one = 1.0f, two = 2.0f;
318 switch (value) {
319 case 0:
320 __ xorps(xmm0, xmm0);
321 break;
322 case 1:
323 __ movflt(xmm0, ExternalAddress((address) &one));
324 break;
325 case 2:
326 __ movflt(xmm0, ExternalAddress((address) &two));
327 break;
328 default:
329 ShouldNotReachHere();
330 break;
331 }
332 }
333
334 void TemplateTable::dconst(int value) {
335 transition(vtos, dtos);
336 static double one = 1.0;
337 switch (value) {
338 case 0:
339 __ xorpd(xmm0, xmm0);
340 break;
341 case 1:
342 __ movdbl(xmm0, ExternalAddress((address) &one));
343 break;
344 default:
345 ShouldNotReachHere();
346 break;
347 }
348 }
349
350 void TemplateTable::bipush() {
351 transition(vtos, itos);
352 __ load_signed_byte(rax, at_bcp(1));
353 }
354
355 void TemplateTable::sipush() {
356 transition(vtos, itos);
357 __ load_unsigned_short(rax, at_bcp(1));
358 __ bswapl(rax);
359 __ sarl(rax, 16);
360 }
361
362 void TemplateTable::ldc(bool wide) {
363 transition(vtos, vtos);
364 Label call_ldc, notFloat, notClass, Done;
365
366 if (wide) {
367 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
368 } else {
369 __ load_unsigned_byte(rbx, at_bcp(1));
370 }
371
372 __ get_cpool_and_tags(rcx, rax);
373 const int base_offset = ConstantPool::header_size() * wordSize;
374 const int tags_offset = Array<u1>::base_offset_in_bytes();
375
376 // get type
377 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
378
379 // unresolved class - get the resolved class
380 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
381 __ jccb(Assembler::equal, call_ldc);
382
383 // unresolved class in error state - call into runtime to throw the error
384 // from the first resolution attempt
385 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
386 __ jccb(Assembler::equal, call_ldc);
387
388 // resolved class - need to call vm to get java mirror of the class
389 __ cmpl(rdx, JVM_CONSTANT_Class);
390 __ jcc(Assembler::notEqual, notClass);
391
392 __ bind(call_ldc);
393 __ movl(c_rarg1, wide);
394 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
395 __ push_ptr(rax);
396 __ verify_oop(rax);
397 __ jmp(Done);
398
399 __ bind(notClass);
400 __ cmpl(rdx, JVM_CONSTANT_Float);
401 __ jccb(Assembler::notEqual, notFloat);
402 // ftos
403 __ movflt(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
404 __ push_f();
405 __ jmp(Done);
406
407 __ bind(notFloat);
408 #ifdef ASSERT
409 {
410 Label L;
411 __ cmpl(rdx, JVM_CONSTANT_Integer);
412 __ jcc(Assembler::equal, L);
413 // String and Object are rewritten to fast_aldc
414 __ stop("unexpected tag type in ldc");
415 __ bind(L);
416 }
417 #endif
418 // itos JVM_CONSTANT_Integer only
419 __ movl(rax, Address(rcx, rbx, Address::times_8, base_offset));
420 __ push_i(rax);
421 __ bind(Done);
422 }
423
424 // Fast path for caching oop constants.
425 void TemplateTable::fast_aldc(bool wide) {
426 transition(vtos, atos);
427
428 Register result = rax;
429 Register tmp = rdx;
430 int index_size = wide ? sizeof(u2) : sizeof(u1);
431
432 Label resolved;
433
434 // We are resolved if the resolved reference cache entry contains a
435 // non-null object (String, MethodType, etc.)
436 assert_different_registers(result, tmp);
437 __ get_cache_index_at_bcp(tmp, 1, index_size);
438 __ load_resolved_reference_at_index(result, tmp);
439 __ testl(result, result);
440 __ jcc(Assembler::notZero, resolved);
441
442 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
443
444 // first time invocation - must resolve first
445 __ movl(tmp, (int)bytecode());
446 __ call_VM(result, entry, tmp);
447
448 __ bind(resolved);
449
450 if (VerifyOops) {
451 __ verify_oop(result);
452 }
453 }
454
455 void TemplateTable::ldc2_w() {
456 transition(vtos, vtos);
457 Label Long, Done;
458 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
459
460 __ get_cpool_and_tags(rcx, rax);
461 const int base_offset = ConstantPool::header_size() * wordSize;
462 const int tags_offset = Array<u1>::base_offset_in_bytes();
463
464 // get type
465 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
466 JVM_CONSTANT_Double);
467 __ jccb(Assembler::notEqual, Long);
468 // dtos
469 __ movdbl(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
470 __ push_d();
471 __ jmpb(Done);
472
473 __ bind(Long);
474 // ltos
475 __ movq(rax, Address(rcx, rbx, Address::times_8, base_offset));
476 __ push_l();
477
478 __ bind(Done);
479 }
480
481 void TemplateTable::locals_index(Register reg, int offset) {
482 __ load_unsigned_byte(reg, at_bcp(offset));
483 __ negptr(reg);
484 }
485
486 void TemplateTable::iload() {
487 transition(vtos, itos);
488 if (RewriteFrequentPairs) {
489 Label rewrite, done;
490 const Register bc = c_rarg3;
491 assert(rbx != bc, "register damaged");
492
493 // get next byte
494 __ load_unsigned_byte(rbx,
495 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
496 // if _iload, wait to rewrite to iload2. We only want to rewrite the
497 // last two iloads in a pair. Comparing against fast_iload means that
498 // the next bytecode is neither an iload or a caload, and therefore
499 // an iload pair.
500 __ cmpl(rbx, Bytecodes::_iload);
501 __ jcc(Assembler::equal, done);
502
503 __ cmpl(rbx, Bytecodes::_fast_iload);
504 __ movl(bc, Bytecodes::_fast_iload2);
505 __ jccb(Assembler::equal, rewrite);
506
507 // if _caload, rewrite to fast_icaload
508 __ cmpl(rbx, Bytecodes::_caload);
509 __ movl(bc, Bytecodes::_fast_icaload);
510 __ jccb(Assembler::equal, rewrite);
511
512 // rewrite so iload doesn't check again.
513 __ movl(bc, Bytecodes::_fast_iload);
514
515 // rewrite
516 // bc: fast bytecode
517 __ bind(rewrite);
518 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
519 __ bind(done);
520 }
521
522 // Get the local value into tos
523 locals_index(rbx);
524 __ movl(rax, iaddress(rbx));
525 }
526
527 void TemplateTable::fast_iload2() {
528 transition(vtos, itos);
529 locals_index(rbx);
530 __ movl(rax, iaddress(rbx));
531 __ push(itos);
532 locals_index(rbx, 3);
533 __ movl(rax, iaddress(rbx));
534 }
535
536 void TemplateTable::fast_iload() {
537 transition(vtos, itos);
538 locals_index(rbx);
539 __ movl(rax, iaddress(rbx));
540 }
541
542 void TemplateTable::lload() {
543 transition(vtos, ltos);
544 locals_index(rbx);
545 __ movq(rax, laddress(rbx));
546 }
547
548 void TemplateTable::fload() {
549 transition(vtos, ftos);
550 locals_index(rbx);
551 __ movflt(xmm0, faddress(rbx));
552 }
553
554 void TemplateTable::dload() {
555 transition(vtos, dtos);
556 locals_index(rbx);
557 __ movdbl(xmm0, daddress(rbx));
558 }
559
560 void TemplateTable::aload() {
561 transition(vtos, atos);
562 locals_index(rbx);
563 __ movptr(rax, aaddress(rbx));
564 }
565
566 void TemplateTable::locals_index_wide(Register reg) {
567 __ load_unsigned_short(reg, at_bcp(2));
568 __ bswapl(reg);
569 __ shrl(reg, 16);
570 __ negptr(reg);
571 }
572
573 void TemplateTable::wide_iload() {
574 transition(vtos, itos);
575 locals_index_wide(rbx);
576 __ movl(rax, iaddress(rbx));
577 }
578
579 void TemplateTable::wide_lload() {
580 transition(vtos, ltos);
581 locals_index_wide(rbx);
582 __ movq(rax, laddress(rbx));
583 }
584
585 void TemplateTable::wide_fload() {
586 transition(vtos, ftos);
587 locals_index_wide(rbx);
588 __ movflt(xmm0, faddress(rbx));
589 }
590
591 void TemplateTable::wide_dload() {
592 transition(vtos, dtos);
593 locals_index_wide(rbx);
594 __ movdbl(xmm0, daddress(rbx));
595 }
596
597 void TemplateTable::wide_aload() {
598 transition(vtos, atos);
599 locals_index_wide(rbx);
600 __ movptr(rax, aaddress(rbx));
601 }
602
603 void TemplateTable::index_check(Register array, Register index) {
604 // destroys rbx
605 // check array
606 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
607 // sign extend index for use by indexed load
608 __ movl2ptr(index, index);
609 // check index
610 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
611 if (index != rbx) {
612 // ??? convention: move aberrant index into ebx for exception message
613 assert(rbx != array, "different registers");
614 __ movl(rbx, index);
615 }
616 __ jump_cc(Assembler::aboveEqual,
617 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
618 }
619
620 void TemplateTable::iaload() {
621 transition(itos, itos);
622 __ pop_ptr(rdx);
623 // eax: index
624 // rdx: array
625 index_check(rdx, rax); // kills rbx
626 __ movl(rax, Address(rdx, rax,
627 Address::times_4,
628 arrayOopDesc::base_offset_in_bytes(T_INT)));
629 }
630
631 void TemplateTable::laload() {
632 transition(itos, ltos);
633 __ pop_ptr(rdx);
634 // eax: index
635 // rdx: array
636 index_check(rdx, rax); // kills rbx
637 __ movq(rax, Address(rdx, rbx,
638 Address::times_8,
639 arrayOopDesc::base_offset_in_bytes(T_LONG)));
640 }
641
642 void TemplateTable::faload() {
643 transition(itos, ftos);
644 __ pop_ptr(rdx);
645 // eax: index
646 // rdx: array
647 index_check(rdx, rax); // kills rbx
648 __ movflt(xmm0, Address(rdx, rax,
649 Address::times_4,
650 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
651 }
652
653 void TemplateTable::daload() {
654 transition(itos, dtos);
655 __ pop_ptr(rdx);
656 // eax: index
657 // rdx: array
658 index_check(rdx, rax); // kills rbx
659 __ movdbl(xmm0, Address(rdx, rax,
660 Address::times_8,
661 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
662 }
663
664 void TemplateTable::aaload() {
665 transition(itos, atos);
666 __ pop_ptr(rdx);
667 // eax: index
668 // rdx: array
669 index_check(rdx, rax); // kills rbx
670 __ load_heap_oop(rax, Address(rdx, rax,
671 UseCompressedOops ? Address::times_4 : Address::times_8,
672 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
673 }
674
675 void TemplateTable::baload() {
676 transition(itos, itos);
677 __ pop_ptr(rdx);
678 // eax: index
679 // rdx: array
680 index_check(rdx, rax); // kills rbx
681 __ load_signed_byte(rax,
682 Address(rdx, rax,
683 Address::times_1,
684 arrayOopDesc::base_offset_in_bytes(T_BYTE)));
685 }
686
687 void TemplateTable::caload() {
688 transition(itos, itos);
689 __ pop_ptr(rdx);
690 // eax: index
691 // rdx: array
692 index_check(rdx, rax); // kills rbx
693 __ load_unsigned_short(rax,
694 Address(rdx, rax,
695 Address::times_2,
696 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
697 }
698
699 // iload followed by caload frequent pair
700 void TemplateTable::fast_icaload() {
701 transition(vtos, itos);
702 // load index out of locals
703 locals_index(rbx);
704 __ movl(rax, iaddress(rbx));
705
706 // eax: index
707 // rdx: array
708 __ pop_ptr(rdx);
709 index_check(rdx, rax); // kills rbx
710 __ load_unsigned_short(rax,
711 Address(rdx, rax,
712 Address::times_2,
713 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
714 }
715
716 void TemplateTable::saload() {
717 transition(itos, itos);
718 __ pop_ptr(rdx);
719 // eax: index
720 // rdx: array
721 index_check(rdx, rax); // kills rbx
722 __ load_signed_short(rax,
723 Address(rdx, rax,
724 Address::times_2,
725 arrayOopDesc::base_offset_in_bytes(T_SHORT)));
726 }
727
728 void TemplateTable::iload(int n) {
729 transition(vtos, itos);
730 __ movl(rax, iaddress(n));
731 }
732
733 void TemplateTable::lload(int n) {
734 transition(vtos, ltos);
735 __ movq(rax, laddress(n));
736 }
737
738 void TemplateTable::fload(int n) {
739 transition(vtos, ftos);
740 __ movflt(xmm0, faddress(n));
741 }
742
743 void TemplateTable::dload(int n) {
744 transition(vtos, dtos);
745 __ movdbl(xmm0, daddress(n));
746 }
747
748 void TemplateTable::aload(int n) {
749 transition(vtos, atos);
750 __ movptr(rax, aaddress(n));
751 }
752
753 void TemplateTable::aload_0() {
754 transition(vtos, atos);
755 // According to bytecode histograms, the pairs:
756 //
757 // _aload_0, _fast_igetfield
758 // _aload_0, _fast_agetfield
759 // _aload_0, _fast_fgetfield
760 //
761 // occur frequently. If RewriteFrequentPairs is set, the (slow)
762 // _aload_0 bytecode checks if the next bytecode is either
763 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
764 // rewrites the current bytecode into a pair bytecode; otherwise it
765 // rewrites the current bytecode into _fast_aload_0 that doesn't do
766 // the pair check anymore.
767 //
768 // Note: If the next bytecode is _getfield, the rewrite must be
769 // delayed, otherwise we may miss an opportunity for a pair.
770 //
771 // Also rewrite frequent pairs
772 // aload_0, aload_1
773 // aload_0, iload_1
774 // These bytecodes with a small amount of code are most profitable
775 // to rewrite
776 if (RewriteFrequentPairs) {
777 Label rewrite, done;
778 const Register bc = c_rarg3;
779 assert(rbx != bc, "register damaged");
780 // get next byte
781 __ load_unsigned_byte(rbx,
782 at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
783
784 // do actual aload_0
785 aload(0);
786
787 // if _getfield then wait with rewrite
788 __ cmpl(rbx, Bytecodes::_getfield);
789 __ jcc(Assembler::equal, done);
790
791 // if _igetfield then reqrite to _fast_iaccess_0
792 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) ==
793 Bytecodes::_aload_0,
794 "fix bytecode definition");
795 __ cmpl(rbx, Bytecodes::_fast_igetfield);
796 __ movl(bc, Bytecodes::_fast_iaccess_0);
797 __ jccb(Assembler::equal, rewrite);
798
799 // if _agetfield then reqrite to _fast_aaccess_0
800 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) ==
801 Bytecodes::_aload_0,
802 "fix bytecode definition");
803 __ cmpl(rbx, Bytecodes::_fast_agetfield);
804 __ movl(bc, Bytecodes::_fast_aaccess_0);
805 __ jccb(Assembler::equal, rewrite);
806
807 // if _fgetfield then reqrite to _fast_faccess_0
808 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) ==
809 Bytecodes::_aload_0,
810 "fix bytecode definition");
811 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
812 __ movl(bc, Bytecodes::_fast_faccess_0);
813 __ jccb(Assembler::equal, rewrite);
814
815 // else rewrite to _fast_aload0
816 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) ==
817 Bytecodes::_aload_0,
818 "fix bytecode definition");
819 __ movl(bc, Bytecodes::_fast_aload_0);
820
821 // rewrite
822 // bc: fast bytecode
823 __ bind(rewrite);
824 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
825
826 __ bind(done);
827 } else {
828 aload(0);
829 }
830 }
831
832 void TemplateTable::istore() {
833 transition(itos, vtos);
834 locals_index(rbx);
835 __ movl(iaddress(rbx), rax);
836 }
837
838 void TemplateTable::lstore() {
839 transition(ltos, vtos);
840 locals_index(rbx);
841 __ movq(laddress(rbx), rax);
842 }
843
844 void TemplateTable::fstore() {
845 transition(ftos, vtos);
846 locals_index(rbx);
847 __ movflt(faddress(rbx), xmm0);
848 }
849
850 void TemplateTable::dstore() {
851 transition(dtos, vtos);
852 locals_index(rbx);
853 __ movdbl(daddress(rbx), xmm0);
854 }
855
856 void TemplateTable::astore() {
857 transition(vtos, vtos);
858 __ pop_ptr(rax);
859 locals_index(rbx);
860 __ movptr(aaddress(rbx), rax);
861 }
862
863 void TemplateTable::wide_istore() {
864 transition(vtos, vtos);
865 __ pop_i();
866 locals_index_wide(rbx);
867 __ movl(iaddress(rbx), rax);
868 }
869
870 void TemplateTable::wide_lstore() {
871 transition(vtos, vtos);
872 __ pop_l();
873 locals_index_wide(rbx);
874 __ movq(laddress(rbx), rax);
875 }
876
877 void TemplateTable::wide_fstore() {
878 transition(vtos, vtos);
879 __ pop_f();
880 locals_index_wide(rbx);
881 __ movflt(faddress(rbx), xmm0);
882 }
883
884 void TemplateTable::wide_dstore() {
885 transition(vtos, vtos);
886 __ pop_d();
887 locals_index_wide(rbx);
888 __ movdbl(daddress(rbx), xmm0);
889 }
890
891 void TemplateTable::wide_astore() {
892 transition(vtos, vtos);
893 __ pop_ptr(rax);
894 locals_index_wide(rbx);
895 __ movptr(aaddress(rbx), rax);
896 }
897
898 void TemplateTable::iastore() {
899 transition(itos, vtos);
900 __ pop_i(rbx);
901 __ pop_ptr(rdx);
902 // eax: value
903 // ebx: index
904 // rdx: array
905 index_check(rdx, rbx); // prefer index in ebx
906 __ movl(Address(rdx, rbx,
907 Address::times_4,
908 arrayOopDesc::base_offset_in_bytes(T_INT)),
909 rax);
910 }
911
912 void TemplateTable::lastore() {
913 transition(ltos, vtos);
914 __ pop_i(rbx);
915 __ pop_ptr(rdx);
916 // rax: value
917 // ebx: index
918 // rdx: array
919 index_check(rdx, rbx); // prefer index in ebx
920 __ movq(Address(rdx, rbx,
921 Address::times_8,
922 arrayOopDesc::base_offset_in_bytes(T_LONG)),
923 rax);
924 }
925
926 void TemplateTable::fastore() {
927 transition(ftos, vtos);
928 __ pop_i(rbx);
929 __ pop_ptr(rdx);
930 // xmm0: value
931 // ebx: index
932 // rdx: array
933 index_check(rdx, rbx); // prefer index in ebx
934 __ movflt(Address(rdx, rbx,
935 Address::times_4,
936 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
937 xmm0);
938 }
939
940 void TemplateTable::dastore() {
941 transition(dtos, vtos);
942 __ pop_i(rbx);
943 __ pop_ptr(rdx);
944 // xmm0: value
945 // ebx: index
946 // rdx: array
947 index_check(rdx, rbx); // prefer index in ebx
948 __ movdbl(Address(rdx, rbx,
949 Address::times_8,
950 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
951 xmm0);
952 }
953
954 void TemplateTable::aastore() {
955 Label is_null, ok_is_subtype, done;
956 transition(vtos, vtos);
957 // stack: ..., array, index, value
958 __ movptr(rax, at_tos()); // value
959 __ movl(rcx, at_tos_p1()); // index
960 __ movptr(rdx, at_tos_p2()); // array
961
962 Address element_address(rdx, rcx,
963 UseCompressedOops? Address::times_4 : Address::times_8,
964 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
965
966 index_check(rdx, rcx); // kills rbx
967 // do array store check - check for NULL value first
968 __ testptr(rax, rax);
969 __ jcc(Assembler::zero, is_null);
970
971 // Move subklass into rbx
972 __ load_klass(rbx, rax);
973 // Move superklass into rax
974 __ load_klass(rax, rdx);
975 __ movptr(rax, Address(rax,
976 ObjArrayKlass::element_klass_offset()));
977 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
978 __ lea(rdx, element_address);
979
980 // Generate subtype check. Blows rcx, rdi
981 // Superklass in rax. Subklass in rbx.
982 __ gen_subtype_check(rbx, ok_is_subtype);
983
984 // Come here on failure
985 // object is at TOS
986 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
987
988 // Come here on success
989 __ bind(ok_is_subtype);
990
991 // Get the value we will store
992 __ movptr(rax, at_tos());
993 // Now store using the appropriate barrier
994 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
995 __ jmp(done);
996
997 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
998 __ bind(is_null);
999 __ profile_null_seen(rbx);
1000
1001 // Store a NULL
1002 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1003
1004 // Pop stack arguments
1005 __ bind(done);
1006 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1007 }
1008
1009 void TemplateTable::bastore() {
1010 transition(itos, vtos);
1011 __ pop_i(rbx);
1012 __ pop_ptr(rdx);
1013 // eax: value
1014 // ebx: index
1015 // rdx: array
1016 index_check(rdx, rbx); // prefer index in ebx
1017 __ movb(Address(rdx, rbx,
1018 Address::times_1,
1019 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1020 rax);
1021 }
1022
1023 void TemplateTable::castore() {
1024 transition(itos, vtos);
1025 __ pop_i(rbx);
1026 __ pop_ptr(rdx);
1027 // eax: value
1028 // ebx: index
1029 // rdx: array
1030 index_check(rdx, rbx); // prefer index in ebx
1031 __ movw(Address(rdx, rbx,
1032 Address::times_2,
1033 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1034 rax);
1035 }
1036
1037 void TemplateTable::sastore() {
1038 castore();
1039 }
1040
1041 void TemplateTable::istore(int n) {
1042 transition(itos, vtos);
1043 __ movl(iaddress(n), rax);
1044 }
1045
1046 void TemplateTable::lstore(int n) {
1047 transition(ltos, vtos);
1048 __ movq(laddress(n), rax);
1049 }
1050
1051 void TemplateTable::fstore(int n) {
1052 transition(ftos, vtos);
1053 __ movflt(faddress(n), xmm0);
1054 }
1055
1056 void TemplateTable::dstore(int n) {
1057 transition(dtos, vtos);
1058 __ movdbl(daddress(n), xmm0);
1059 }
1060
1061 void TemplateTable::astore(int n) {
1062 transition(vtos, vtos);
1063 __ pop_ptr(rax);
1064 __ movptr(aaddress(n), rax);
1065 }
1066
1067 void TemplateTable::pop() {
1068 transition(vtos, vtos);
1069 __ addptr(rsp, Interpreter::stackElementSize);
1070 }
1071
1072 void TemplateTable::pop2() {
1073 transition(vtos, vtos);
1074 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1075 }
1076
1077 void TemplateTable::dup() {
1078 transition(vtos, vtos);
1079 __ load_ptr(0, rax);
1080 __ push_ptr(rax);
1081 // stack: ..., a, a
1082 }
1083
1084 void TemplateTable::dup_x1() {
1085 transition(vtos, vtos);
1086 // stack: ..., a, b
1087 __ load_ptr( 0, rax); // load b
1088 __ load_ptr( 1, rcx); // load a
1089 __ store_ptr(1, rax); // store b
1090 __ store_ptr(0, rcx); // store a
1091 __ push_ptr(rax); // push b
1092 // stack: ..., b, a, b
1093 }
1094
1095 void TemplateTable::dup_x2() {
1096 transition(vtos, vtos);
1097 // stack: ..., a, b, c
1098 __ load_ptr( 0, rax); // load c
1099 __ load_ptr( 2, rcx); // load a
1100 __ store_ptr(2, rax); // store c in a
1101 __ push_ptr(rax); // push c
1102 // stack: ..., c, b, c, c
1103 __ load_ptr( 2, rax); // load b
1104 __ store_ptr(2, rcx); // store a in b
1105 // stack: ..., c, a, c, c
1106 __ store_ptr(1, rax); // store b in c
1107 // stack: ..., c, a, b, c
1108 }
1109
1110 void TemplateTable::dup2() {
1111 transition(vtos, vtos);
1112 // stack: ..., a, b
1113 __ load_ptr(1, rax); // load a
1114 __ push_ptr(rax); // push a
1115 __ load_ptr(1, rax); // load b
1116 __ push_ptr(rax); // push b
1117 // stack: ..., a, b, a, b
1118 }
1119
1120 void TemplateTable::dup2_x1() {
1121 transition(vtos, vtos);
1122 // stack: ..., a, b, c
1123 __ load_ptr( 0, rcx); // load c
1124 __ load_ptr( 1, rax); // load b
1125 __ push_ptr(rax); // push b
1126 __ push_ptr(rcx); // push c
1127 // stack: ..., a, b, c, b, c
1128 __ store_ptr(3, rcx); // store c in b
1129 // stack: ..., a, c, c, b, c
1130 __ load_ptr( 4, rcx); // load a
1131 __ store_ptr(2, rcx); // store a in 2nd c
1132 // stack: ..., a, c, a, b, c
1133 __ store_ptr(4, rax); // store b in a
1134 // stack: ..., b, c, a, b, c
1135 }
1136
1137 void TemplateTable::dup2_x2() {
1138 transition(vtos, vtos);
1139 // stack: ..., a, b, c, d
1140 __ load_ptr( 0, rcx); // load d
1141 __ load_ptr( 1, rax); // load c
1142 __ push_ptr(rax); // push c
1143 __ push_ptr(rcx); // push d
1144 // stack: ..., a, b, c, d, c, d
1145 __ load_ptr( 4, rax); // load b
1146 __ store_ptr(2, rax); // store b in d
1147 __ store_ptr(4, rcx); // store d in b
1148 // stack: ..., a, d, c, b, c, d
1149 __ load_ptr( 5, rcx); // load a
1150 __ load_ptr( 3, rax); // load c
1151 __ store_ptr(3, rcx); // store a in c
1152 __ store_ptr(5, rax); // store c in a
1153 // stack: ..., c, d, a, b, c, d
1154 }
1155
1156 void TemplateTable::swap() {
1157 transition(vtos, vtos);
1158 // stack: ..., a, b
1159 __ load_ptr( 1, rcx); // load a
1160 __ load_ptr( 0, rax); // load b
1161 __ store_ptr(0, rcx); // store a in b
1162 __ store_ptr(1, rax); // store b in a
1163 // stack: ..., b, a
1164 }
1165
1166 void TemplateTable::iop2(Operation op) {
1167 transition(itos, itos);
1168 switch (op) {
1169 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1170 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1171 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1172 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1173 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1174 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1175 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1176 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1177 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1178 default : ShouldNotReachHere();
1179 }
1180 }
1181
1182 void TemplateTable::lop2(Operation op) {
1183 transition(ltos, ltos);
1184 switch (op) {
1185 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1186 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1187 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1188 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1189 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1190 default : ShouldNotReachHere();
1191 }
1192 }
1193
1194 void TemplateTable::idiv() {
1195 transition(itos, itos);
1196 __ movl(rcx, rax);
1197 __ pop_i(rax);
1198 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1199 // they are not equal, one could do a normal division (no correction
1200 // needed), which may speed up this implementation for the common case.
1201 // (see also JVM spec., p.243 & p.271)
1202 __ corrected_idivl(rcx);
1203 }
1204
1205 void TemplateTable::irem() {
1206 transition(itos, itos);
1207 __ movl(rcx, rax);
1208 __ pop_i(rax);
1209 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1210 // they are not equal, one could do a normal division (no correction
1211 // needed), which may speed up this implementation for the common case.
1212 // (see also JVM spec., p.243 & p.271)
1213 __ corrected_idivl(rcx);
1214 __ movl(rax, rdx);
1215 }
1216
1217 void TemplateTable::lmul() {
1218 transition(ltos, ltos);
1219 __ pop_l(rdx);
1220 __ imulq(rax, rdx);
1221 }
1222
1223 void TemplateTable::ldiv() {
1224 transition(ltos, ltos);
1225 __ mov(rcx, rax);
1226 __ pop_l(rax);
1227 // generate explicit div0 check
1228 __ testq(rcx, rcx);
1229 __ jump_cc(Assembler::zero,
1230 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1231 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1232 // they are not equal, one could do a normal division (no correction
1233 // needed), which may speed up this implementation for the common case.
1234 // (see also JVM spec., p.243 & p.271)
1235 __ corrected_idivq(rcx); // kills rbx
1236 }
1237
1238 void TemplateTable::lrem() {
1239 transition(ltos, ltos);
1240 __ mov(rcx, rax);
1241 __ pop_l(rax);
1242 __ testq(rcx, rcx);
1243 __ jump_cc(Assembler::zero,
1244 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1245 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1246 // they are not equal, one could do a normal division (no correction
1247 // needed), which may speed up this implementation for the common case.
1248 // (see also JVM spec., p.243 & p.271)
1249 __ corrected_idivq(rcx); // kills rbx
1250 __ mov(rax, rdx);
1251 }
1252
1253 void TemplateTable::lshl() {
1254 transition(itos, ltos);
1255 __ movl(rcx, rax); // get shift count
1256 __ pop_l(rax); // get shift value
1257 __ shlq(rax);
1258 }
1259
1260 void TemplateTable::lshr() {
1261 transition(itos, ltos);
1262 __ movl(rcx, rax); // get shift count
1263 __ pop_l(rax); // get shift value
1264 __ sarq(rax);
1265 }
1266
1267 void TemplateTable::lushr() {
1268 transition(itos, ltos);
1269 __ movl(rcx, rax); // get shift count
1270 __ pop_l(rax); // get shift value
1271 __ shrq(rax);
1272 }
1273
1274 void TemplateTable::fop2(Operation op) {
1275 transition(ftos, ftos);
1276 switch (op) {
1277 case add:
1278 __ addss(xmm0, at_rsp());
1279 __ addptr(rsp, Interpreter::stackElementSize);
1280 break;
1281 case sub:
1282 __ movflt(xmm1, xmm0);
1283 __ pop_f(xmm0);
1284 __ subss(xmm0, xmm1);
1285 break;
1286 case mul:
1287 __ mulss(xmm0, at_rsp());
1288 __ addptr(rsp, Interpreter::stackElementSize);
1289 break;
1290 case div:
1291 __ movflt(xmm1, xmm0);
1292 __ pop_f(xmm0);
1293 __ divss(xmm0, xmm1);
1294 break;
1295 case rem:
1296 __ movflt(xmm1, xmm0);
1297 __ pop_f(xmm0);
1298 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1299 break;
1300 default:
1301 ShouldNotReachHere();
1302 break;
1303 }
1304 }
1305
1306 void TemplateTable::dop2(Operation op) {
1307 transition(dtos, dtos);
1308 switch (op) {
1309 case add:
1310 __ addsd(xmm0, at_rsp());
1311 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1312 break;
1313 case sub:
1314 __ movdbl(xmm1, xmm0);
1315 __ pop_d(xmm0);
1316 __ subsd(xmm0, xmm1);
1317 break;
1318 case mul:
1319 __ mulsd(xmm0, at_rsp());
1320 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1321 break;
1322 case div:
1323 __ movdbl(xmm1, xmm0);
1324 __ pop_d(xmm0);
1325 __ divsd(xmm0, xmm1);
1326 break;
1327 case rem:
1328 __ movdbl(xmm1, xmm0);
1329 __ pop_d(xmm0);
1330 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1331 break;
1332 default:
1333 ShouldNotReachHere();
1334 break;
1335 }
1336 }
1337
1338 void TemplateTable::ineg() {
1339 transition(itos, itos);
1340 __ negl(rax);
1341 }
1342
1343 void TemplateTable::lneg() {
1344 transition(ltos, ltos);
1345 __ negq(rax);
1346 }
1347
1348 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1349 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1350 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1351 // of 128-bits operands for SSE instructions.
1352 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1353 // Store the value to a 128-bits operand.
1354 operand[0] = lo;
1355 operand[1] = hi;
1356 return operand;
1357 }
1358
1359 // Buffer for 128-bits masks used by SSE instructions.
1360 static jlong float_signflip_pool[2*2];
1361 static jlong double_signflip_pool[2*2];
1362
1363 void TemplateTable::fneg() {
1364 transition(ftos, ftos);
1365 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], 0x8000000080000000, 0x8000000080000000);
1366 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1367 }
1368
1369 void TemplateTable::dneg() {
1370 transition(dtos, dtos);
1371 static jlong *double_signflip = double_quadword(&double_signflip_pool[1], 0x8000000000000000, 0x8000000000000000);
1372 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1373 }
1374
1375 void TemplateTable::iinc() {
1376 transition(vtos, vtos);
1377 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1378 locals_index(rbx);
1379 __ addl(iaddress(rbx), rdx);
1380 }
1381
1382 void TemplateTable::wide_iinc() {
1383 transition(vtos, vtos);
1384 __ movl(rdx, at_bcp(4)); // get constant
1385 locals_index_wide(rbx);
1386 __ bswapl(rdx); // swap bytes & sign-extend constant
1387 __ sarl(rdx, 16);
1388 __ addl(iaddress(rbx), rdx);
1389 // Note: should probably use only one movl to get both
1390 // the index and the constant -> fix this
1391 }
1392
1393 void TemplateTable::convert() {
1394 // Checking
1395 #ifdef ASSERT
1396 {
1397 TosState tos_in = ilgl;
1398 TosState tos_out = ilgl;
1399 switch (bytecode()) {
1400 case Bytecodes::_i2l: // fall through
1401 case Bytecodes::_i2f: // fall through
1402 case Bytecodes::_i2d: // fall through
1403 case Bytecodes::_i2b: // fall through
1404 case Bytecodes::_i2c: // fall through
1405 case Bytecodes::_i2s: tos_in = itos; break;
1406 case Bytecodes::_l2i: // fall through
1407 case Bytecodes::_l2f: // fall through
1408 case Bytecodes::_l2d: tos_in = ltos; break;
1409 case Bytecodes::_f2i: // fall through
1410 case Bytecodes::_f2l: // fall through
1411 case Bytecodes::_f2d: tos_in = ftos; break;
1412 case Bytecodes::_d2i: // fall through
1413 case Bytecodes::_d2l: // fall through
1414 case Bytecodes::_d2f: tos_in = dtos; break;
1415 default : ShouldNotReachHere();
1416 }
1417 switch (bytecode()) {
1418 case Bytecodes::_l2i: // fall through
1419 case Bytecodes::_f2i: // fall through
1420 case Bytecodes::_d2i: // fall through
1421 case Bytecodes::_i2b: // fall through
1422 case Bytecodes::_i2c: // fall through
1423 case Bytecodes::_i2s: tos_out = itos; break;
1424 case Bytecodes::_i2l: // fall through
1425 case Bytecodes::_f2l: // fall through
1426 case Bytecodes::_d2l: tos_out = ltos; break;
1427 case Bytecodes::_i2f: // fall through
1428 case Bytecodes::_l2f: // fall through
1429 case Bytecodes::_d2f: tos_out = ftos; break;
1430 case Bytecodes::_i2d: // fall through
1431 case Bytecodes::_l2d: // fall through
1432 case Bytecodes::_f2d: tos_out = dtos; break;
1433 default : ShouldNotReachHere();
1434 }
1435 transition(tos_in, tos_out);
1436 }
1437 #endif // ASSERT
1438
1439 static const int64_t is_nan = 0x8000000000000000L;
1440
1441 // Conversion
1442 switch (bytecode()) {
1443 case Bytecodes::_i2l:
1444 __ movslq(rax, rax);
1445 break;
1446 case Bytecodes::_i2f:
1447 __ cvtsi2ssl(xmm0, rax);
1448 break;
1449 case Bytecodes::_i2d:
1450 __ cvtsi2sdl(xmm0, rax);
1451 break;
1452 case Bytecodes::_i2b:
1453 __ movsbl(rax, rax);
1454 break;
1455 case Bytecodes::_i2c:
1456 __ movzwl(rax, rax);
1457 break;
1458 case Bytecodes::_i2s:
1459 __ movswl(rax, rax);
1460 break;
1461 case Bytecodes::_l2i:
1462 __ movl(rax, rax);
1463 break;
1464 case Bytecodes::_l2f:
1465 __ cvtsi2ssq(xmm0, rax);
1466 break;
1467 case Bytecodes::_l2d:
1468 __ cvtsi2sdq(xmm0, rax);
1469 break;
1470 case Bytecodes::_f2i:
1471 {
1472 Label L;
1473 __ cvttss2sil(rax, xmm0);
1474 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1475 __ jcc(Assembler::notEqual, L);
1476 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1477 __ bind(L);
1478 }
1479 break;
1480 case Bytecodes::_f2l:
1481 {
1482 Label L;
1483 __ cvttss2siq(rax, xmm0);
1484 // NaN or overflow/underflow?
1485 __ cmp64(rax, ExternalAddress((address) &is_nan));
1486 __ jcc(Assembler::notEqual, L);
1487 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1488 __ bind(L);
1489 }
1490 break;
1491 case Bytecodes::_f2d:
1492 __ cvtss2sd(xmm0, xmm0);
1493 break;
1494 case Bytecodes::_d2i:
1495 {
1496 Label L;
1497 __ cvttsd2sil(rax, xmm0);
1498 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1499 __ jcc(Assembler::notEqual, L);
1500 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1501 __ bind(L);
1502 }
1503 break;
1504 case Bytecodes::_d2l:
1505 {
1506 Label L;
1507 __ cvttsd2siq(rax, xmm0);
1508 // NaN or overflow/underflow?
1509 __ cmp64(rax, ExternalAddress((address) &is_nan));
1510 __ jcc(Assembler::notEqual, L);
1511 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1512 __ bind(L);
1513 }
1514 break;
1515 case Bytecodes::_d2f:
1516 __ cvtsd2ss(xmm0, xmm0);
1517 break;
1518 default:
1519 ShouldNotReachHere();
1520 }
1521 }
1522
1523 void TemplateTable::lcmp() {
1524 transition(ltos, itos);
1525 Label done;
1526 __ pop_l(rdx);
1527 __ cmpq(rdx, rax);
1528 __ movl(rax, -1);
1529 __ jccb(Assembler::less, done);
1530 __ setb(Assembler::notEqual, rax);
1531 __ movzbl(rax, rax);
1532 __ bind(done);
1533 }
1534
1535 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1536 Label done;
1537 if (is_float) {
1538 // XXX get rid of pop here, use ... reg, mem32
1539 __ pop_f(xmm1);
1540 __ ucomiss(xmm1, xmm0);
1541 } else {
1542 // XXX get rid of pop here, use ... reg, mem64
1543 __ pop_d(xmm1);
1544 __ ucomisd(xmm1, xmm0);
1545 }
1546 if (unordered_result < 0) {
1547 __ movl(rax, -1);
1548 __ jccb(Assembler::parity, done);
1549 __ jccb(Assembler::below, done);
1550 __ setb(Assembler::notEqual, rdx);
1551 __ movzbl(rax, rdx);
1552 } else {
1553 __ movl(rax, 1);
1554 __ jccb(Assembler::parity, done);
1555 __ jccb(Assembler::above, done);
1556 __ movl(rax, 0);
1557 __ jccb(Assembler::equal, done);
1558 __ decrementl(rax);
1559 }
1560 __ bind(done);
1561 }
1562
1563 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1564 __ get_method(rcx); // rcx holds method
1565 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
1566 // holds bumped taken count
1567
1568 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1569 InvocationCounter::counter_offset();
1570 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1571 InvocationCounter::counter_offset();
1572
1573 // Load up edx with the branch displacement
1574 if (is_wide) {
1575 __ movl(rdx, at_bcp(1));
1576 } else {
1577 __ load_signed_short(rdx, at_bcp(1));
1578 }
1579 __ bswapl(rdx);
1580
1581 if (!is_wide) {
1582 __ sarl(rdx, 16);
1583 }
1584 __ movl2ptr(rdx, rdx);
1585
1586 // Handle all the JSR stuff here, then exit.
1587 // It's much shorter and cleaner than intermingling with the non-JSR
1588 // normal-branch stuff occurring below.
1589 if (is_jsr) {
1590 // Pre-load the next target bytecode into rbx
1591 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1, 0));
1592
1593 // compute return address as bci in rax
1594 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
1595 in_bytes(ConstMethod::codes_offset())));
1596 __ subptr(rax, Address(rcx, Method::const_offset()));
1597 // Adjust the bcp in r13 by the displacement in rdx
1598 __ addptr(r13, rdx);
1599 // jsr returns atos that is not an oop
1600 __ push_i(rax);
1601 __ dispatch_only(vtos);
1602 return;
1603 }
1604
1605 // Normal (non-jsr) branch handling
1606
1607 // Adjust the bcp in r13 by the displacement in rdx
1608 __ addptr(r13, rdx);
1609
1610 assert(UseLoopCounter || !UseOnStackReplacement,
1611 "on-stack-replacement requires loop counters");
1612 Label backedge_counter_overflow;
1613 Label profile_method;
1614 Label dispatch;
1615 if (UseLoopCounter) {
1616 // increment backedge counter for backward branches
1617 // rax: MDO
1618 // ebx: MDO bumped taken-count
1619 // rcx: method
1620 // rdx: target offset
1621 // r13: target bcp
1622 // r14: locals pointer
1623 __ testl(rdx, rdx); // check if forward or backward branch
1624 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1625
1626 // check if MethodCounters exists
1627 Label has_counters;
1628 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
1629 __ testptr(rax, rax);
1630 __ jcc(Assembler::notZero, has_counters);
1631 __ push(rdx);
1632 __ push(rcx);
1633 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
1634 rcx);
1635 __ pop(rcx);
1636 __ pop(rdx);
1637 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
1638 __ jcc(Assembler::zero, dispatch);
1639 __ bind(has_counters);
1640
1641 if (TieredCompilation) {
1642 Label no_mdo;
1643 int increment = InvocationCounter::count_increment;
1644 if (ProfileInterpreter) {
1645 // Are we profiling?
1646 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
1647 __ testptr(rbx, rbx);
1648 __ jccb(Assembler::zero, no_mdo);
1649 // Increment the MDO backedge counter
1650 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
1651 in_bytes(InvocationCounter::counter_offset()));
1652 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
1653 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1654 rax, false, Assembler::zero, &backedge_counter_overflow);
1655 __ jmp(dispatch);
1656 }
1657 __ bind(no_mdo);
1658 // Increment backedge counter in MethodCounters*
1659 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
1660 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
1661 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
1662 rax, false, Assembler::zero, &backedge_counter_overflow);
1663 } else { // not TieredCompilation
1664 // increment counter
1665 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
1666 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1667 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1668 __ movl(Address(rcx, be_offset), rax); // store counter
1669
1670 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1671
1672 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1673 __ addl(rax, Address(rcx, be_offset)); // add both counters
1674
1675 if (ProfileInterpreter) {
1676 // Test to see if we should create a method data oop
1677 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
1678 __ jcc(Assembler::less, dispatch);
1679
1680 // if no method data exists, go to profile method
1681 __ test_method_data_pointer(rax, profile_method);
1682
1683 if (UseOnStackReplacement) {
1684 // check for overflow against ebx which is the MDO taken count
1685 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1686 __ jcc(Assembler::below, dispatch);
1687
1688 // When ProfileInterpreter is on, the backedge_count comes
1689 // from the MethodData*, which value does not get reset on
1690 // the call to frequency_counter_overflow(). To avoid
1691 // excessive calls to the overflow routine while the method is
1692 // being compiled, add a second test to make sure the overflow
1693 // function is called only once every overflow_frequency.
1694 const int overflow_frequency = 1024;
1695 __ andl(rbx, overflow_frequency - 1);
1696 __ jcc(Assembler::zero, backedge_counter_overflow);
1697
1698 }
1699 } else {
1700 if (UseOnStackReplacement) {
1701 // check for overflow against eax, which is the sum of the
1702 // counters
1703 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1704 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1705
1706 }
1707 }
1708 }
1709 __ bind(dispatch);
1710 }
1711
1712 // Pre-load the next target bytecode into rbx
1713 __ load_unsigned_byte(rbx, Address(r13, 0));
1714
1715 // continue with the bytecode @ target
1716 // eax: return bci for jsr's, unused otherwise
1717 // ebx: target bytecode
1718 // r13: target bcp
1719 __ dispatch_only(vtos);
1720
1721 if (UseLoopCounter) {
1722 if (ProfileInterpreter) {
1723 // Out-of-line code to allocate method data oop.
1724 __ bind(profile_method);
1725 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1726 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1727 __ set_method_data_pointer_for_bcp();
1728 __ jmp(dispatch);
1729 }
1730
1731 if (UseOnStackReplacement) {
1732 // invocation counter overflow
1733 __ bind(backedge_counter_overflow);
1734 __ negptr(rdx);
1735 __ addptr(rdx, r13); // branch bcp
1736 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1737 __ call_VM(noreg,
1738 CAST_FROM_FN_PTR(address,
1739 InterpreterRuntime::frequency_counter_overflow),
1740 rdx);
1741 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1742
1743 // rax: osr nmethod (osr ok) or NULL (osr not possible)
1744 // ebx: target bytecode
1745 // rdx: scratch
1746 // r14: locals pointer
1747 // r13: bcp
1748 __ testptr(rax, rax); // test result
1749 __ jcc(Assembler::zero, dispatch); // no osr if null
1750 // nmethod may have been invalidated (VM may block upon call_VM return)
1751 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
1752 __ jcc(Assembler::notEqual, dispatch);
1753
1754 // We have the address of an on stack replacement routine in eax
1755 // We need to prepare to execute the OSR method. First we must
1756 // migrate the locals and monitors off of the stack.
1757
1758 __ mov(r13, rax); // save the nmethod
1759
1760 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1761
1762 // eax is OSR buffer, move it to expected parameter location
1763 __ mov(j_rarg0, rax);
1764
1765 // We use j_rarg definitions here so that registers don't conflict as parameter
1766 // registers change across platforms as we are in the midst of a calling
1767 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
1768
1769 const Register retaddr = j_rarg2;
1770 const Register sender_sp = j_rarg1;
1771
1772 // pop the interpreter frame
1773 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1774 __ leave(); // remove frame anchor
1775 __ pop(retaddr); // get return address
1776 __ mov(rsp, sender_sp); // set sp to sender sp
1777 // Ensure compiled code always sees stack at proper alignment
1778 __ andptr(rsp, -(StackAlignmentInBytes));
1779
1780 // unlike x86 we need no specialized return from compiled code
1781 // to the interpreter or the call stub.
1782
1783 // push the return address
1784 __ push(retaddr);
1785
1786 // and begin the OSR nmethod
1787 __ jmp(Address(r13, nmethod::osr_entry_point_offset()));
1788 }
1789 }
1790 }
1791
1792
1793 void TemplateTable::if_0cmp(Condition cc) {
1794 transition(itos, vtos);
1795 // assume branch is more often taken than not (loops use backward branches)
1796 Label not_taken;
1797 __ testl(rax, rax);
1798 __ jcc(j_not(cc), not_taken);
1799 branch(false, false);
1800 __ bind(not_taken);
1801 __ profile_not_taken_branch(rax);
1802 }
1803
1804 void TemplateTable::if_icmp(Condition cc) {
1805 transition(itos, vtos);
1806 // assume branch is more often taken than not (loops use backward branches)
1807 Label not_taken;
1808 __ pop_i(rdx);
1809 __ cmpl(rdx, rax);
1810 __ jcc(j_not(cc), not_taken);
1811 branch(false, false);
1812 __ bind(not_taken);
1813 __ profile_not_taken_branch(rax);
1814 }
1815
1816 void TemplateTable::if_nullcmp(Condition cc) {
1817 transition(atos, vtos);
1818 // assume branch is more often taken than not (loops use backward branches)
1819 Label not_taken;
1820 __ testptr(rax, rax);
1821 __ jcc(j_not(cc), not_taken);
1822 branch(false, false);
1823 __ bind(not_taken);
1824 __ profile_not_taken_branch(rax);
1825 }
1826
1827 void TemplateTable::if_acmp(Condition cc) {
1828 transition(atos, vtos);
1829 // assume branch is more often taken than not (loops use backward branches)
1830 Label not_taken;
1831 __ pop_ptr(rdx);
1832 __ cmpptr(rdx, rax);
1833 __ jcc(j_not(cc), not_taken);
1834 branch(false, false);
1835 __ bind(not_taken);
1836 __ profile_not_taken_branch(rax);
1837 }
1838
1839 void TemplateTable::ret() {
1840 transition(vtos, vtos);
1841 locals_index(rbx);
1842 __ movslq(rbx, iaddress(rbx)); // get return bci, compute return bcp
1843 __ profile_ret(rbx, rcx);
1844 __ get_method(rax);
1845 __ movptr(r13, Address(rax, Method::const_offset()));
1846 __ lea(r13, Address(r13, rbx, Address::times_1,
1847 ConstMethod::codes_offset()));
1848 __ dispatch_next(vtos);
1849 }
1850
1851 void TemplateTable::wide_ret() {
1852 transition(vtos, vtos);
1853 locals_index_wide(rbx);
1854 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
1855 __ profile_ret(rbx, rcx);
1856 __ get_method(rax);
1857 __ movptr(r13, Address(rax, Method::const_offset()));
1858 __ lea(r13, Address(r13, rbx, Address::times_1, ConstMethod::codes_offset()));
1859 __ dispatch_next(vtos);
1860 }
1861
1862 void TemplateTable::tableswitch() {
1863 Label default_case, continue_execution;
1864 transition(itos, vtos);
1865 // align r13
1866 __ lea(rbx, at_bcp(BytesPerInt));
1867 __ andptr(rbx, -BytesPerInt);
1868 // load lo & hi
1869 __ movl(rcx, Address(rbx, BytesPerInt));
1870 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
1871 __ bswapl(rcx);
1872 __ bswapl(rdx);
1873 // check against lo & hi
1874 __ cmpl(rax, rcx);
1875 __ jcc(Assembler::less, default_case);
1876 __ cmpl(rax, rdx);
1877 __ jcc(Assembler::greater, default_case);
1878 // lookup dispatch offset
1879 __ subl(rax, rcx);
1880 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1881 __ profile_switch_case(rax, rbx, rcx);
1882 // continue execution
1883 __ bind(continue_execution);
1884 __ bswapl(rdx);
1885 __ movl2ptr(rdx, rdx);
1886 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1887 __ addptr(r13, rdx);
1888 __ dispatch_only(vtos);
1889 // handle default
1890 __ bind(default_case);
1891 __ profile_switch_default(rax);
1892 __ movl(rdx, Address(rbx, 0));
1893 __ jmp(continue_execution);
1894 }
1895
1896 void TemplateTable::lookupswitch() {
1897 transition(itos, itos);
1898 __ stop("lookupswitch bytecode should have been rewritten");
1899 }
1900
1901 void TemplateTable::fast_linearswitch() {
1902 transition(itos, vtos);
1903 Label loop_entry, loop, found, continue_execution;
1904 // bswap rax so we can avoid bswapping the table entries
1905 __ bswapl(rax);
1906 // align r13
1907 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
1908 // this instruction (change offsets
1909 // below)
1910 __ andptr(rbx, -BytesPerInt);
1911 // set counter
1912 __ movl(rcx, Address(rbx, BytesPerInt));
1913 __ bswapl(rcx);
1914 __ jmpb(loop_entry);
1915 // table search
1916 __ bind(loop);
1917 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
1918 __ jcc(Assembler::equal, found);
1919 __ bind(loop_entry);
1920 __ decrementl(rcx);
1921 __ jcc(Assembler::greaterEqual, loop);
1922 // default case
1923 __ profile_switch_default(rax);
1924 __ movl(rdx, Address(rbx, 0));
1925 __ jmp(continue_execution);
1926 // entry found -> get offset
1927 __ bind(found);
1928 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
1929 __ profile_switch_case(rcx, rax, rbx);
1930 // continue execution
1931 __ bind(continue_execution);
1932 __ bswapl(rdx);
1933 __ movl2ptr(rdx, rdx);
1934 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1935 __ addptr(r13, rdx);
1936 __ dispatch_only(vtos);
1937 }
1938
1939 void TemplateTable::fast_binaryswitch() {
1940 transition(itos, vtos);
1941 // Implementation using the following core algorithm:
1942 //
1943 // int binary_search(int key, LookupswitchPair* array, int n) {
1944 // // Binary search according to "Methodik des Programmierens" by
1945 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1946 // int i = 0;
1947 // int j = n;
1948 // while (i+1 < j) {
1949 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1950 // // with Q: for all i: 0 <= i < n: key < a[i]
1951 // // where a stands for the array and assuming that the (inexisting)
1952 // // element a[n] is infinitely big.
1953 // int h = (i + j) >> 1;
1954 // // i < h < j
1955 // if (key < array[h].fast_match()) {
1956 // j = h;
1957 // } else {
1958 // i = h;
1959 // }
1960 // }
1961 // // R: a[i] <= key < a[i+1] or Q
1962 // // (i.e., if key is within array, i is the correct index)
1963 // return i;
1964 // }
1965
1966 // Register allocation
1967 const Register key = rax; // already set (tosca)
1968 const Register array = rbx;
1969 const Register i = rcx;
1970 const Register j = rdx;
1971 const Register h = rdi;
1972 const Register temp = rsi;
1973
1974 // Find array start
1975 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
1976 // get rid of this
1977 // instruction (change
1978 // offsets below)
1979 __ andptr(array, -BytesPerInt);
1980
1981 // Initialize i & j
1982 __ xorl(i, i); // i = 0;
1983 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
1984
1985 // Convert j into native byteordering
1986 __ bswapl(j);
1987
1988 // And start
1989 Label entry;
1990 __ jmp(entry);
1991
1992 // binary search loop
1993 {
1994 Label loop;
1995 __ bind(loop);
1996 // int h = (i + j) >> 1;
1997 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1998 __ sarl(h, 1); // h = (i + j) >> 1;
1999 // if (key < array[h].fast_match()) {
2000 // j = h;
2001 // } else {
2002 // i = h;
2003 // }
2004 // Convert array[h].match to native byte-ordering before compare
2005 __ movl(temp, Address(array, h, Address::times_8));
2006 __ bswapl(temp);
2007 __ cmpl(key, temp);
2008 // j = h if (key < array[h].fast_match())
2009 __ cmovl(Assembler::less, j, h);
2010 // i = h if (key >= array[h].fast_match())
2011 __ cmovl(Assembler::greaterEqual, i, h);
2012 // while (i+1 < j)
2013 __ bind(entry);
2014 __ leal(h, Address(i, 1)); // i+1
2015 __ cmpl(h, j); // i+1 < j
2016 __ jcc(Assembler::less, loop);
2017 }
2018
2019 // end of binary search, result index is i (must check again!)
2020 Label default_case;
2021 // Convert array[i].match to native byte-ordering before compare
2022 __ movl(temp, Address(array, i, Address::times_8));
2023 __ bswapl(temp);
2024 __ cmpl(key, temp);
2025 __ jcc(Assembler::notEqual, default_case);
2026
2027 // entry found -> j = offset
2028 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2029 __ profile_switch_case(i, key, array);
2030 __ bswapl(j);
2031 __ movl2ptr(j, j);
2032 __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
2033 __ addptr(r13, j);
2034 __ dispatch_only(vtos);
2035
2036 // default case -> j = default offset
2037 __ bind(default_case);
2038 __ profile_switch_default(i);
2039 __ movl(j, Address(array, -2 * BytesPerInt));
2040 __ bswapl(j);
2041 __ movl2ptr(j, j);
2042 __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
2043 __ addptr(r13, j);
2044 __ dispatch_only(vtos);
2045 }
2046
2047
2048 void TemplateTable::_return(TosState state) {
2049 transition(state, state);
2050 assert(_desc->calls_vm(),
2051 "inconsistent calls_vm information"); // call in remove_activation
2052
2053 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2054 assert(state == vtos, "only valid state");
2055 __ movptr(c_rarg1, aaddress(0));
2056 __ load_klass(rdi, c_rarg1);
2057 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2058 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2059 Label skip_register_finalizer;
2060 __ jcc(Assembler::zero, skip_register_finalizer);
2061
2062 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2063
2064 __ bind(skip_register_finalizer);
2065 }
2066
2067 __ remove_activation(state, r13);
2068 __ jmp(r13);
2069 }
2070
2071 // ----------------------------------------------------------------------------
2072 // Volatile variables demand their effects be made known to all CPU's
2073 // in order. Store buffers on most chips allow reads & writes to
2074 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2075 // without some kind of memory barrier (i.e., it's not sufficient that
2076 // the interpreter does not reorder volatile references, the hardware
2077 // also must not reorder them).
2078 //
2079 // According to the new Java Memory Model (JMM):
2080 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2081 // writes act as aquire & release, so:
2082 // (2) A read cannot let unrelated NON-volatile memory refs that
2083 // happen after the read float up to before the read. It's OK for
2084 // non-volatile memory refs that happen before the volatile read to
2085 // float down below it.
2086 // (3) Similar a volatile write cannot let unrelated NON-volatile
2087 // memory refs that happen BEFORE the write float down to after the
2088 // write. It's OK for non-volatile memory refs that happen after the
2089 // volatile write to float up before it.
2090 //
2091 // We only put in barriers around volatile refs (they are expensive),
2092 // not _between_ memory refs (that would require us to track the
2093 // flavor of the previous memory refs). Requirements (2) and (3)
2094 // require some barriers before volatile stores and after volatile
2095 // loads. These nearly cover requirement (1) but miss the
2096 // volatile-store-volatile-load case. This final case is placed after
2097 // volatile-stores although it could just as well go before
2098 // volatile-loads.
2099 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits
2100 order_constraint) {
2101 // Helper function to insert a is-volatile test and memory barrier
2102 if (os::is_MP()) { // Not needed on single CPU
2103 __ membar(order_constraint);
2104 }
2105 }
2106
2107 void TemplateTable::resolve_cache_and_index(int byte_no,
2108 Register Rcache,
2109 Register index,
2110 size_t index_size) {
2111 const Register temp = rbx;
2112 assert_different_registers(Rcache, index, temp);
2113
2114 Label resolved;
2115 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2116 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2117 __ cmpl(temp, (int) bytecode()); // have we resolved this bytecode?
2118 __ jcc(Assembler::equal, resolved);
2119
2120 // resolve first time through
2121 address entry;
2122 switch (bytecode()) {
2123 case Bytecodes::_getstatic:
2124 case Bytecodes::_putstatic:
2125 case Bytecodes::_getfield:
2126 case Bytecodes::_putfield:
2127 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
2128 break;
2129 case Bytecodes::_invokevirtual:
2130 case Bytecodes::_invokespecial:
2131 case Bytecodes::_invokestatic:
2132 case Bytecodes::_invokeinterface:
2133 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
2134 break;
2135 case Bytecodes::_invokehandle:
2136 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);
2137 break;
2138 case Bytecodes::_invokedynamic:
2139 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
2140 break;
2141 default:
2142 fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2143 break;
2144 }
2145 __ movl(temp, (int) bytecode());
2146 __ call_VM(noreg, entry, temp);
2147
2148 // Update registers with resolved info
2149 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2150 __ bind(resolved);
2151 }
2152
2153 // The cache and index registers must be set before call
2154 void TemplateTable::load_field_cp_cache_entry(Register obj,
2155 Register cache,
2156 Register index,
2157 Register off,
2158 Register flags,
2159 bool is_static = false) {
2160 assert_different_registers(cache, index, flags, off);
2161
2162 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2163 // Field offset
2164 __ movptr(off, Address(cache, index, Address::times_ptr,
2165 in_bytes(cp_base_offset +
2166 ConstantPoolCacheEntry::f2_offset())));
2167 // Flags
2168 __ movl(flags, Address(cache, index, Address::times_ptr,
2169 in_bytes(cp_base_offset +
2170 ConstantPoolCacheEntry::flags_offset())));
2171
2172 // klass overwrite register
2173 if (is_static) {
2174 __ movptr(obj, Address(cache, index, Address::times_ptr,
2175 in_bytes(cp_base_offset +
2176 ConstantPoolCacheEntry::f1_offset())));
2177 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2178 __ movptr(obj, Address(obj, mirror_offset));
2179 }
2180 }
2181
2182 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2183 Register method,
2184 Register itable_index,
2185 Register flags,
2186 bool is_invokevirtual,
2187 bool is_invokevfinal, /*unused*/
2188 bool is_invokedynamic) {
2189 // setup registers
2190 const Register cache = rcx;
2191 const Register index = rdx;
2192 assert_different_registers(method, flags);
2193 assert_different_registers(method, cache, index);
2194 assert_different_registers(itable_index, flags);
2195 assert_different_registers(itable_index, cache, index);
2196 // determine constant pool cache field offsets
2197 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2198 const int method_offset = in_bytes(
2199 ConstantPoolCache::base_offset() +
2200 ((byte_no == f2_byte)
2201 ? ConstantPoolCacheEntry::f2_offset()
2202 : ConstantPoolCacheEntry::f1_offset()));
2203 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2204 ConstantPoolCacheEntry::flags_offset());
2205 // access constant pool cache fields
2206 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2207 ConstantPoolCacheEntry::f2_offset());
2208
2209 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2210 resolve_cache_and_index(byte_no, cache, index, index_size);
2211 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2212
2213 if (itable_index != noreg) {
2214 // pick up itable or appendix index from f2 also:
2215 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2216 }
2217 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2218 }
2219
2220 // Correct values of the cache and index registers are preserved.
2221 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2222 bool is_static, bool has_tos) {
2223 // do the JVMTI work here to avoid disturbing the register state below
2224 // We use c_rarg registers here because we want to use the register used in
2225 // the call to the VM
2226 if (JvmtiExport::can_post_field_access()) {
2227 // Check to see if a field access watch has been set before we
2228 // take the time to call into the VM.
2229 Label L1;
2230 assert_different_registers(cache, index, rax);
2231 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2232 __ testl(rax, rax);
2233 __ jcc(Assembler::zero, L1);
2234
2235 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2236
2237 // cache entry pointer
2238 __ addptr(c_rarg2, in_bytes(ConstantPoolCache::base_offset()));
2239 __ shll(c_rarg3, LogBytesPerWord);
2240 __ addptr(c_rarg2, c_rarg3);
2241 if (is_static) {
2242 __ xorl(c_rarg1, c_rarg1); // NULL object reference
2243 } else {
2244 __ movptr(c_rarg1, at_tos()); // get object pointer without popping it
2245 __ verify_oop(c_rarg1);
2246 }
2247 // c_rarg1: object pointer or NULL
2248 // c_rarg2: cache entry pointer
2249 // c_rarg3: jvalue object on the stack
2250 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2251 InterpreterRuntime::post_field_access),
2252 c_rarg1, c_rarg2, c_rarg3);
2253 __ get_cache_and_index_at_bcp(cache, index, 1);
2254 __ bind(L1);
2255 }
2256 }
2257
2258 void TemplateTable::pop_and_check_object(Register r) {
2259 __ pop_ptr(r);
2260 __ null_check(r); // for field access must check obj.
2261 __ verify_oop(r);
2262 }
2263
2264 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2265 transition(vtos, vtos);
2266
2267 const Register cache = rcx;
2268 const Register index = rdx;
2269 const Register obj = c_rarg3;
2270 const Register off = rbx;
2271 const Register flags = rax;
2272 const Register bc = c_rarg3; // uses same reg as obj, so don't mix them
2273
2274 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2275 jvmti_post_field_access(cache, index, is_static, false);
2276 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2277
2278 if (!is_static) {
2279 // obj is on the stack
2280 pop_and_check_object(obj);
2281 }
2282
2283 const Address field(obj, off, Address::times_1);
2284
2285 Label Done, notByte, notInt, notShort, notChar,
2286 notLong, notFloat, notObj, notDouble;
2287
2288 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2289 // Make sure we don't need to mask edx after the above shift
2290 assert(btos == 0, "change code, btos != 0");
2291
2292 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2293 __ jcc(Assembler::notZero, notByte);
2294 // btos
2295 __ load_signed_byte(rax, field);
2296 __ push(btos);
2297 // Rewrite bytecode to be faster
2298 if (!is_static) {
2299 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2300 }
2301 __ jmp(Done);
2302
2303 __ bind(notByte);
2304 __ cmpl(flags, atos);
2305 __ jcc(Assembler::notEqual, notObj);
2306 // atos
2307 __ load_heap_oop(rax, field);
2308 __ push(atos);
2309 if (!is_static) {
2310 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2311 }
2312 __ jmp(Done);
2313
2314 __ bind(notObj);
2315 __ cmpl(flags, itos);
2316 __ jcc(Assembler::notEqual, notInt);
2317 // itos
2318 __ movl(rax, field);
2319 __ push(itos);
2320 // Rewrite bytecode to be faster
2321 if (!is_static) {
2322 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2323 }
2324 __ jmp(Done);
2325
2326 __ bind(notInt);
2327 __ cmpl(flags, ctos);
2328 __ jcc(Assembler::notEqual, notChar);
2329 // ctos
2330 __ load_unsigned_short(rax, field);
2331 __ push(ctos);
2332 // Rewrite bytecode to be faster
2333 if (!is_static) {
2334 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2335 }
2336 __ jmp(Done);
2337
2338 __ bind(notChar);
2339 __ cmpl(flags, stos);
2340 __ jcc(Assembler::notEqual, notShort);
2341 // stos
2342 __ load_signed_short(rax, field);
2343 __ push(stos);
2344 // Rewrite bytecode to be faster
2345 if (!is_static) {
2346 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2347 }
2348 __ jmp(Done);
2349
2350 __ bind(notShort);
2351 __ cmpl(flags, ltos);
2352 __ jcc(Assembler::notEqual, notLong);
2353 // ltos
2354 __ movq(rax, field);
2355 __ push(ltos);
2356 // Rewrite bytecode to be faster
2357 if (!is_static) {
2358 patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx);
2359 }
2360 __ jmp(Done);
2361
2362 __ bind(notLong);
2363 __ cmpl(flags, ftos);
2364 __ jcc(Assembler::notEqual, notFloat);
2365 // ftos
2366 __ movflt(xmm0, field);
2367 __ push(ftos);
2368 // Rewrite bytecode to be faster
2369 if (!is_static) {
2370 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2371 }
2372 __ jmp(Done);
2373
2374 __ bind(notFloat);
2375 #ifdef ASSERT
2376 __ cmpl(flags, dtos);
2377 __ jcc(Assembler::notEqual, notDouble);
2378 #endif
2379 // dtos
2380 __ movdbl(xmm0, field);
2381 __ push(dtos);
2382 // Rewrite bytecode to be faster
2383 if (!is_static) {
2384 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2385 }
2386 #ifdef ASSERT
2387 __ jmp(Done);
2388
2389 __ bind(notDouble);
2390 __ stop("Bad state");
2391 #endif
2392
2393 __ bind(Done);
2394 // [jk] not needed currently
2395 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2396 // Assembler::LoadStore));
2397 }
2398
2399
2400 void TemplateTable::getfield(int byte_no) {
2401 getfield_or_static(byte_no, false);
2402 }
2403
2404 void TemplateTable::getstatic(int byte_no) {
2405 getfield_or_static(byte_no, true);
2406 }
2407
2408 // The registers cache and index expected to be set before call.
2409 // The function may destroy various registers, just not the cache and index registers.
2410 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2411 transition(vtos, vtos);
2412
2413 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2414
2415 if (JvmtiExport::can_post_field_modification()) {
2416 // Check to see if a field modification watch has been set before
2417 // we take the time to call into the VM.
2418 Label L1;
2419 assert_different_registers(cache, index, rax);
2420 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2421 __ testl(rax, rax);
2422 __ jcc(Assembler::zero, L1);
2423
2424 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2425
2426 if (is_static) {
2427 // Life is simple. Null out the object pointer.
2428 __ xorl(c_rarg1, c_rarg1);
2429 } else {
2430 // Life is harder. The stack holds the value on top, followed by
2431 // the object. We don't know the size of the value, though; it
2432 // could be one or two words depending on its type. As a result,
2433 // we must find the type to determine where the object is.
2434 __ movl(c_rarg3, Address(c_rarg2, rscratch1,
2435 Address::times_8,
2436 in_bytes(cp_base_offset +
2437 ConstantPoolCacheEntry::flags_offset())));
2438 __ shrl(c_rarg3, ConstantPoolCacheEntry::tos_state_shift);
2439 // Make sure we don't need to mask rcx after the above shift
2440 ConstantPoolCacheEntry::verify_tos_state_shift();
2441 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2442 __ cmpl(c_rarg3, ltos);
2443 __ cmovptr(Assembler::equal,
2444 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2445 __ cmpl(c_rarg3, dtos);
2446 __ cmovptr(Assembler::equal,
2447 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2448 }
2449 // cache entry pointer
2450 __ addptr(c_rarg2, in_bytes(cp_base_offset));
2451 __ shll(rscratch1, LogBytesPerWord);
2452 __ addptr(c_rarg2, rscratch1);
2453 // object (tos)
2454 __ mov(c_rarg3, rsp);
2455 // c_rarg1: object pointer set up above (NULL if static)
2456 // c_rarg2: cache entry pointer
2457 // c_rarg3: jvalue object on the stack
2458 __ call_VM(noreg,
2459 CAST_FROM_FN_PTR(address,
2460 InterpreterRuntime::post_field_modification),
2461 c_rarg1, c_rarg2, c_rarg3);
2462 __ get_cache_and_index_at_bcp(cache, index, 1);
2463 __ bind(L1);
2464 }
2465 }
2466
2467 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2468 transition(vtos, vtos);
2469
2470 const Register cache = rcx;
2471 const Register index = rdx;
2472 const Register obj = rcx;
2473 const Register off = rbx;
2474 const Register flags = rax;
2475 const Register bc = c_rarg3;
2476
2477 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2478 jvmti_post_field_mod(cache, index, is_static);
2479 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2480
2481 // [jk] not needed currently
2482 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2483 // Assembler::StoreStore));
2484
2485 Label notVolatile, Done;
2486 __ movl(rdx, flags);
2487 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2488 __ andl(rdx, 0x1);
2489
2490 // field address
2491 const Address field(obj, off, Address::times_1);
2492
2493 Label notByte, notInt, notShort, notChar,
2494 notLong, notFloat, notObj, notDouble;
2495
2496 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2497
2498 assert(btos == 0, "change code, btos != 0");
2499 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2500 __ jcc(Assembler::notZero, notByte);
2501
2502 // btos
2503 {
2504 __ pop(btos);
2505 if (!is_static) pop_and_check_object(obj);
2506 __ movb(field, rax);
2507 if (!is_static) {
2508 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
2509 }
2510 __ jmp(Done);
2511 }
2512
2513 __ bind(notByte);
2514 __ cmpl(flags, atos);
2515 __ jcc(Assembler::notEqual, notObj);
2516
2517 // atos
2518 {
2519 __ pop(atos);
2520 if (!is_static) pop_and_check_object(obj);
2521 // Store into the field
2522 do_oop_store(_masm, field, rax, _bs->kind(), false);
2523 if (!is_static) {
2524 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
2525 }
2526 __ jmp(Done);
2527 }
2528
2529 __ bind(notObj);
2530 __ cmpl(flags, itos);
2531 __ jcc(Assembler::notEqual, notInt);
2532
2533 // itos
2534 {
2535 __ pop(itos);
2536 if (!is_static) pop_and_check_object(obj);
2537 __ movl(field, rax);
2538 if (!is_static) {
2539 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
2540 }
2541 __ jmp(Done);
2542 }
2543
2544 __ bind(notInt);
2545 __ cmpl(flags, ctos);
2546 __ jcc(Assembler::notEqual, notChar);
2547
2548 // ctos
2549 {
2550 __ pop(ctos);
2551 if (!is_static) pop_and_check_object(obj);
2552 __ movw(field, rax);
2553 if (!is_static) {
2554 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
2555 }
2556 __ jmp(Done);
2557 }
2558
2559 __ bind(notChar);
2560 __ cmpl(flags, stos);
2561 __ jcc(Assembler::notEqual, notShort);
2562
2563 // stos
2564 {
2565 __ pop(stos);
2566 if (!is_static) pop_and_check_object(obj);
2567 __ movw(field, rax);
2568 if (!is_static) {
2569 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
2570 }
2571 __ jmp(Done);
2572 }
2573
2574 __ bind(notShort);
2575 __ cmpl(flags, ltos);
2576 __ jcc(Assembler::notEqual, notLong);
2577
2578 // ltos
2579 {
2580 __ pop(ltos);
2581 if (!is_static) pop_and_check_object(obj);
2582 __ movq(field, rax);
2583 if (!is_static) {
2584 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
2585 }
2586 __ jmp(Done);
2587 }
2588
2589 __ bind(notLong);
2590 __ cmpl(flags, ftos);
2591 __ jcc(Assembler::notEqual, notFloat);
2592
2593 // ftos
2594 {
2595 __ pop(ftos);
2596 if (!is_static) pop_and_check_object(obj);
2597 __ movflt(field, xmm0);
2598 if (!is_static) {
2599 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
2600 }
2601 __ jmp(Done);
2602 }
2603
2604 __ bind(notFloat);
2605 #ifdef ASSERT
2606 __ cmpl(flags, dtos);
2607 __ jcc(Assembler::notEqual, notDouble);
2608 #endif
2609
2610 // dtos
2611 {
2612 __ pop(dtos);
2613 if (!is_static) pop_and_check_object(obj);
2614 __ movdbl(field, xmm0);
2615 if (!is_static) {
2616 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
2617 }
2618 }
2619
2620 #ifdef ASSERT
2621 __ jmp(Done);
2622
2623 __ bind(notDouble);
2624 __ stop("Bad state");
2625 #endif
2626
2627 __ bind(Done);
2628
2629 // Check for volatile store
2630 __ testl(rdx, rdx);
2631 __ jcc(Assembler::zero, notVolatile);
2632 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2633 Assembler::StoreStore));
2634 __ bind(notVolatile);
2635 }
2636
2637 void TemplateTable::putfield(int byte_no) {
2638 putfield_or_static(byte_no, false);
2639 }
2640
2641 void TemplateTable::putstatic(int byte_no) {
2642 putfield_or_static(byte_no, true);
2643 }
2644
2645 void TemplateTable::jvmti_post_fast_field_mod() {
2646 if (JvmtiExport::can_post_field_modification()) {
2647 // Check to see if a field modification watch has been set before
2648 // we take the time to call into the VM.
2649 Label L2;
2650 __ mov32(c_rarg3, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2651 __ testl(c_rarg3, c_rarg3);
2652 __ jcc(Assembler::zero, L2);
2653 __ pop_ptr(rbx); // copy the object pointer from tos
2654 __ verify_oop(rbx);
2655 __ push_ptr(rbx); // put the object pointer back on tos
2656 // Save tos values before call_VM() clobbers them. Since we have
2657 // to do it for every data type, we use the saved values as the
2658 // jvalue object.
2659 switch (bytecode()) { // load values into the jvalue object
2660 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
2661 case Bytecodes::_fast_bputfield: // fall through
2662 case Bytecodes::_fast_sputfield: // fall through
2663 case Bytecodes::_fast_cputfield: // fall through
2664 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
2665 case Bytecodes::_fast_dputfield: __ push_d(); break;
2666 case Bytecodes::_fast_fputfield: __ push_f(); break;
2667 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
2668
2669 default:
2670 ShouldNotReachHere();
2671 }
2672 __ mov(c_rarg3, rsp); // points to jvalue on the stack
2673 // access constant pool cache entry
2674 __ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1);
2675 __ verify_oop(rbx);
2676 // rbx: object pointer copied above
2677 // c_rarg2: cache entry pointer
2678 // c_rarg3: jvalue object on the stack
2679 __ call_VM(noreg,
2680 CAST_FROM_FN_PTR(address,
2681 InterpreterRuntime::post_field_modification),
2682 rbx, c_rarg2, c_rarg3);
2683
2684 switch (bytecode()) { // restore tos values
2685 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
2686 case Bytecodes::_fast_bputfield: // fall through
2687 case Bytecodes::_fast_sputfield: // fall through
2688 case Bytecodes::_fast_cputfield: // fall through
2689 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
2690 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2691 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2692 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
2693 }
2694 __ bind(L2);
2695 }
2696 }
2697
2698 void TemplateTable::fast_storefield(TosState state) {
2699 transition(state, vtos);
2700
2701 ByteSize base = ConstantPoolCache::base_offset();
2702
2703 jvmti_post_fast_field_mod();
2704
2705 // access constant pool cache
2706 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2707
2708 // test for volatile with rdx
2709 __ movl(rdx, Address(rcx, rbx, Address::times_8,
2710 in_bytes(base +
2711 ConstantPoolCacheEntry::flags_offset())));
2712
2713 // replace index with field offset from cache entry
2714 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2715 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2716
2717 // [jk] not needed currently
2718 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2719 // Assembler::StoreStore));
2720
2721 Label notVolatile;
2722 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2723 __ andl(rdx, 0x1);
2724
2725 // Get object from stack
2726 pop_and_check_object(rcx);
2727
2728 // field address
2729 const Address field(rcx, rbx, Address::times_1);
2730
2731 // access field
2732 switch (bytecode()) {
2733 case Bytecodes::_fast_aputfield:
2734 do_oop_store(_masm, field, rax, _bs->kind(), false);
2735 break;
2736 case Bytecodes::_fast_lputfield:
2737 __ movq(field, rax);
2738 break;
2739 case Bytecodes::_fast_iputfield:
2740 __ movl(field, rax);
2741 break;
2742 case Bytecodes::_fast_bputfield:
2743 __ movb(field, rax);
2744 break;
2745 case Bytecodes::_fast_sputfield:
2746 // fall through
2747 case Bytecodes::_fast_cputfield:
2748 __ movw(field, rax);
2749 break;
2750 case Bytecodes::_fast_fputfield:
2751 __ movflt(field, xmm0);
2752 break;
2753 case Bytecodes::_fast_dputfield:
2754 __ movdbl(field, xmm0);
2755 break;
2756 default:
2757 ShouldNotReachHere();
2758 }
2759
2760 // Check for volatile store
2761 __ testl(rdx, rdx);
2762 __ jcc(Assembler::zero, notVolatile);
2763 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2764 Assembler::StoreStore));
2765 __ bind(notVolatile);
2766 }
2767
2768
2769 void TemplateTable::fast_accessfield(TosState state) {
2770 transition(atos, state);
2771
2772 // Do the JVMTI work here to avoid disturbing the register state below
2773 if (JvmtiExport::can_post_field_access()) {
2774 // Check to see if a field access watch has been set before we
2775 // take the time to call into the VM.
2776 Label L1;
2777 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2778 __ testl(rcx, rcx);
2779 __ jcc(Assembler::zero, L1);
2780 // access constant pool cache entry
2781 __ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1);
2782 __ verify_oop(rax);
2783 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
2784 __ mov(c_rarg1, rax);
2785 // c_rarg1: object pointer copied above
2786 // c_rarg2: cache entry pointer
2787 __ call_VM(noreg,
2788 CAST_FROM_FN_PTR(address,
2789 InterpreterRuntime::post_field_access),
2790 c_rarg1, c_rarg2);
2791 __ pop_ptr(rax); // restore object pointer
2792 __ bind(L1);
2793 }
2794
2795 // access constant pool cache
2796 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2797 // replace index with field offset from cache entry
2798 // [jk] not needed currently
2799 // if (os::is_MP()) {
2800 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
2801 // in_bytes(ConstantPoolCache::base_offset() +
2802 // ConstantPoolCacheEntry::flags_offset())));
2803 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2804 // __ andl(rdx, 0x1);
2805 // }
2806 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2807 in_bytes(ConstantPoolCache::base_offset() +
2808 ConstantPoolCacheEntry::f2_offset())));
2809
2810 // rax: object
2811 __ verify_oop(rax);
2812 __ null_check(rax);
2813 Address field(rax, rbx, Address::times_1);
2814
2815 // access field
2816 switch (bytecode()) {
2817 case Bytecodes::_fast_agetfield:
2818 __ load_heap_oop(rax, field);
2819 __ verify_oop(rax);
2820 break;
2821 case Bytecodes::_fast_lgetfield:
2822 __ movq(rax, field);
2823 break;
2824 case Bytecodes::_fast_igetfield:
2825 __ movl(rax, field);
2826 break;
2827 case Bytecodes::_fast_bgetfield:
2828 __ movsbl(rax, field);
2829 break;
2830 case Bytecodes::_fast_sgetfield:
2831 __ load_signed_short(rax, field);
2832 break;
2833 case Bytecodes::_fast_cgetfield:
2834 __ load_unsigned_short(rax, field);
2835 break;
2836 case Bytecodes::_fast_fgetfield:
2837 __ movflt(xmm0, field);
2838 break;
2839 case Bytecodes::_fast_dgetfield:
2840 __ movdbl(xmm0, field);
2841 break;
2842 default:
2843 ShouldNotReachHere();
2844 }
2845 // [jk] not needed currently
2846 // if (os::is_MP()) {
2847 // Label notVolatile;
2848 // __ testl(rdx, rdx);
2849 // __ jcc(Assembler::zero, notVolatile);
2850 // __ membar(Assembler::LoadLoad);
2851 // __ bind(notVolatile);
2852 //};
2853 }
2854
2855 void TemplateTable::fast_xaccess(TosState state) {
2856 transition(vtos, state);
2857
2858 // get receiver
2859 __ movptr(rax, aaddress(0));
2860 // access constant pool cache
2861 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2862 __ movptr(rbx,
2863 Address(rcx, rdx, Address::times_8,
2864 in_bytes(ConstantPoolCache::base_offset() +
2865 ConstantPoolCacheEntry::f2_offset())));
2866 // make sure exception is reported in correct bcp range (getfield is
2867 // next instruction)
2868 __ increment(r13);
2869 __ null_check(rax);
2870 switch (state) {
2871 case itos:
2872 __ movl(rax, Address(rax, rbx, Address::times_1));
2873 break;
2874 case atos:
2875 __ load_heap_oop(rax, Address(rax, rbx, Address::times_1));
2876 __ verify_oop(rax);
2877 break;
2878 case ftos:
2879 __ movflt(xmm0, Address(rax, rbx, Address::times_1));
2880 break;
2881 default:
2882 ShouldNotReachHere();
2883 }
2884
2885 // [jk] not needed currently
2886 // if (os::is_MP()) {
2887 // Label notVolatile;
2888 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
2889 // in_bytes(ConstantPoolCache::base_offset() +
2890 // ConstantPoolCacheEntry::flags_offset())));
2891 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2892 // __ testl(rdx, 0x1);
2893 // __ jcc(Assembler::zero, notVolatile);
2894 // __ membar(Assembler::LoadLoad);
2895 // __ bind(notVolatile);
2896 // }
2897
2898 __ decrement(r13);
2899 }
2900
2901
2902
2903 //-----------------------------------------------------------------------------
2904 // Calls
2905
2906 void TemplateTable::count_calls(Register method, Register temp) {
2907 // implemented elsewhere
2908 ShouldNotReachHere();
2909 }
2910
2911 void TemplateTable::prepare_invoke(int byte_no,
2912 Register method, // linked method (or i-klass)
2913 Register index, // itable index, MethodType, etc.
2914 Register recv, // if caller wants to see it
2915 Register flags // if caller wants to test it
2916 ) {
2917 // determine flags
2918 const Bytecodes::Code code = bytecode();
2919 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2920 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
2921 const bool is_invokehandle = code == Bytecodes::_invokehandle;
2922 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2923 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2924 const bool load_receiver = (recv != noreg);
2925 const bool save_flags = (flags != noreg);
2926 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
2927 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
2928 assert(flags == noreg || flags == rdx, "");
2929 assert(recv == noreg || recv == rcx, "");
2930
2931 // setup registers & access constant pool cache
2932 if (recv == noreg) recv = rcx;
2933 if (flags == noreg) flags = rdx;
2934 assert_different_registers(method, index, recv, flags);
2935
2936 // save 'interpreter return address'
2937 __ save_bcp();
2938
2939 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2940
2941 // maybe push appendix to arguments (just before return address)
2942 if (is_invokedynamic || is_invokehandle) {
2943 Label L_no_push;
2944 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
2945 __ jcc(Assembler::zero, L_no_push);
2946 // Push the appendix as a trailing parameter.
2947 // This must be done before we get the receiver,
2948 // since the parameter_size includes it.
2949 __ push(rbx);
2950 __ mov(rbx, index);
2951 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
2952 __ load_resolved_reference_at_index(index, rbx);
2953 __ pop(rbx);
2954 __ push(index); // push appendix (MethodType, CallSite, etc.)
2955 __ bind(L_no_push);
2956 }
2957
2958 // load receiver if needed (after appendix is pushed so parameter size is correct)
2959 // Note: no return address pushed yet
2960 if (load_receiver) {
2961 __ movl(recv, flags);
2962 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
2963 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
2964 const int receiver_is_at_end = -1; // back off one slot to get receiver
2965 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
2966 __ movptr(recv, recv_addr);
2967 __ verify_oop(recv);
2968 }
2969
2970 if (save_flags) {
2971 __ movl(r13, flags);
2972 }
2973
2974 // compute return type
2975 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2976 // Make sure we don't need to mask flags after the above shift
2977 ConstantPoolCacheEntry::verify_tos_state_shift();
2978 // load return address
2979 {
2980 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2981 ExternalAddress table(table_addr);
2982 __ lea(rscratch1, table);
2983 __ movptr(flags, Address(rscratch1, flags, Address::times_ptr));
2984 }
2985
2986 // push return address
2987 __ push(flags);
2988
2989 // Restore flags value from the constant pool cache, and restore rsi
2990 // for later null checks. r13 is the bytecode pointer
2991 if (save_flags) {
2992 __ movl(flags, r13);
2993 __ restore_bcp();
2994 }
2995 }
2996
2997
2998 void TemplateTable::invokevirtual_helper(Register index,
2999 Register recv,
3000 Register flags) {
3001 // Uses temporary registers rax, rdx
3002 assert_different_registers(index, recv, rax, rdx);
3003 assert(index == rbx, "");
3004 assert(recv == rcx, "");
3005
3006 // Test for an invoke of a final method
3007 Label notFinal;
3008 __ movl(rax, flags);
3009 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3010 __ jcc(Assembler::zero, notFinal);
3011
3012 const Register method = index; // method must be rbx
3013 assert(method == rbx,
3014 "Method* must be rbx for interpreter calling convention");
3015
3016 // do the call - the index is actually the method to call
3017 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3018
3019 // It's final, need a null check here!
3020 __ null_check(recv);
3021
3022 // profile this call
3023 __ profile_final_call(rax);
3024 __ profile_arguments_type(rax, method, r13, true);
3025
3026 __ jump_from_interpreted(method, rax);
3027
3028 __ bind(notFinal);
3029
3030 // get receiver klass
3031 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3032 __ load_klass(rax, recv);
3033
3034 // profile this call
3035 __ profile_virtual_call(rax, r14, rdx);
3036
3037 // get target Method* & entry point
3038 __ lookup_virtual_method(rax, index, method);
3039 __ profile_arguments_type(rdx, method, r13, true);
3040 __ jump_from_interpreted(method, rdx);
3041 }
3042
3043
3044 void TemplateTable::invokevirtual(int byte_no) {
3045 transition(vtos, vtos);
3046 assert(byte_no == f2_byte, "use this argument");
3047 prepare_invoke(byte_no,
3048 rbx, // method or vtable index
3049 noreg, // unused itable index
3050 rcx, rdx); // recv, flags
3051
3052 // rbx: index
3053 // rcx: receiver
3054 // rdx: flags
3055
3056 invokevirtual_helper(rbx, rcx, rdx);
3057 }
3058
3059
3060 void TemplateTable::invokespecial(int byte_no) {
3061 transition(vtos, vtos);
3062 assert(byte_no == f1_byte, "use this argument");
3063 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3064 rcx); // get receiver also for null check
3065 __ verify_oop(rcx);
3066 __ null_check(rcx);
3067 // do the call
3068 __ profile_call(rax);
3069 __ profile_arguments_type(rax, rbx, r13, false);
3070 __ jump_from_interpreted(rbx, rax);
3071 }
3072
3073
3074 void TemplateTable::invokestatic(int byte_no) {
3075 transition(vtos, vtos);
3076 assert(byte_no == f1_byte, "use this argument");
3077 prepare_invoke(byte_no, rbx); // get f1 Method*
3078 // do the call
3079 __ profile_call(rax);
3080 __ profile_arguments_type(rax, rbx, r13, false);
3081 __ jump_from_interpreted(rbx, rax);
3082 }
3083
3084 void TemplateTable::fast_invokevfinal(int byte_no) {
3085 transition(vtos, vtos);
3086 assert(byte_no == f2_byte, "use this argument");
3087 __ stop("fast_invokevfinal not used on amd64");
3088 }
3089
3090 void TemplateTable::invokeinterface(int byte_no) {
3091 transition(vtos, vtos);
3092 assert(byte_no == f1_byte, "use this argument");
3093 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3094 rcx, rdx); // recv, flags
3095
3096 // rax: interface klass (from f1)
3097 // rbx: itable index (from f2)
3098 // rcx: receiver
3099 // rdx: flags
3100
3101 // Special case of invokeinterface called for virtual method of
3102 // java.lang.Object. See cpCacheOop.cpp for details.
3103 // This code isn't produced by javac, but could be produced by
3104 // another compliant java compiler.
3105 Label notMethod;
3106 __ movl(r14, rdx);
3107 __ andl(r14, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3108 __ jcc(Assembler::zero, notMethod);
3109
3110 invokevirtual_helper(rbx, rcx, rdx);
3111 __ bind(notMethod);
3112
3113 // Get receiver klass into rdx - also a null check
3114 __ restore_locals(); // restore r14
3115 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3116 __ load_klass(rdx, rcx);
3117
3118 // profile this call
3119 __ profile_virtual_call(rdx, r13, r14);
3120
3121 Label no_such_interface, no_such_method;
3122
3123 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3124 rdx, rax, rbx,
3125 // outputs: method, scan temp. reg
3126 rbx, r13,
3127 no_such_interface);
3128
3129 // rbx: Method* to call
3130 // rcx: receiver
3131 // Check for abstract method error
3132 // Note: This should be done more efficiently via a throw_abstract_method_error
3133 // interpreter entry point and a conditional jump to it in case of a null
3134 // method.
3135 __ testptr(rbx, rbx);
3136 __ jcc(Assembler::zero, no_such_method);
3137
3138 __ profile_arguments_type(rdx, rbx, r13, true);
3139
3140 // do the call
3141 // rcx: receiver
3142 // rbx,: Method*
3143 __ jump_from_interpreted(rbx, rdx);
3144 __ should_not_reach_here();
3145
3146 // exception handling code follows...
3147 // note: must restore interpreter registers to canonical
3148 // state for exception handling to work correctly!
3149
3150 __ bind(no_such_method);
3151 // throw exception
3152 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3153 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3154 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3155 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3156 // the call_VM checks for exception, so we should never return here.
3157 __ should_not_reach_here();
3158
3159 __ bind(no_such_interface);
3160 // throw exception
3161 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3162 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3163 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3164 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3165 InterpreterRuntime::throw_IncompatibleClassChangeError));
3166 // the call_VM checks for exception, so we should never return here.
3167 __ should_not_reach_here();
3168 }
3169
3170
3171 void TemplateTable::invokehandle(int byte_no) {
3172 transition(vtos, vtos);
3173 assert(byte_no == f1_byte, "use this argument");
3174 const Register rbx_method = rbx;
3175 const Register rax_mtype = rax;
3176 const Register rcx_recv = rcx;
3177 const Register rdx_flags = rdx;
3178
3179 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3180 __ verify_method_ptr(rbx_method);
3181 __ verify_oop(rcx_recv);
3182 __ null_check(rcx_recv);
3183
3184 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3185 // rbx: MH.invokeExact_MT method (from f2)
3186
3187 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3188
3189 // FIXME: profile the LambdaForm also
3190 __ profile_final_call(rax);
3191 __ profile_arguments_type(rdx, rbx_method, r13, true);
3192
3193 __ jump_from_interpreted(rbx_method, rdx);
3194 }
3195
3196
3197 void TemplateTable::invokedynamic(int byte_no) {
3198 transition(vtos, vtos);
3199 assert(byte_no == f1_byte, "use this argument");
3200
3201 const Register rbx_method = rbx;
3202 const Register rax_callsite = rax;
3203
3204 prepare_invoke(byte_no, rbx_method, rax_callsite);
3205
3206 // rax: CallSite object (from cpool->resolved_references[f1])
3207 // rbx: MH.linkToCallSite method (from f2)
3208
3209 // Note: rax_callsite is already pushed by prepare_invoke
3210
3211 // %%% should make a type profile for any invokedynamic that takes a ref argument
3212 // profile this call
3213 __ profile_call(r13);
3214 __ profile_arguments_type(rdx, rbx_method, r13, false);
3215
3216 __ verify_oop(rax_callsite);
3217
3218 __ jump_from_interpreted(rbx_method, rdx);
3219 }
3220
3221
3222 //-----------------------------------------------------------------------------
3223 // Allocation
3224
3225 void TemplateTable::_new() {
3226 transition(vtos, atos);
3227 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3228 Label slow_case;
3229 Label done;
3230 Label initialize_header;
3231 Label initialize_object; // including clearing the fields
3232 Label allocate_shared;
3233
3234 __ get_cpool_and_tags(rsi, rax);
3235 // Make sure the class we're about to instantiate has been resolved.
3236 // This is done before loading InstanceKlass to be consistent with the order
3237 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3238 const int tags_offset = Array<u1>::base_offset_in_bytes();
3239 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset),
3240 JVM_CONSTANT_Class);
3241 __ jcc(Assembler::notEqual, slow_case);
3242
3243 // get InstanceKlass
3244 __ movptr(rsi, Address(rsi, rdx,
3245 Address::times_8, sizeof(ConstantPool)));
3246
3247 // make sure klass is initialized & doesn't have finalizer
3248 // make sure klass is fully initialized
3249 __ cmpb(Address(rsi,
3250 InstanceKlass::init_state_offset()),
3251 InstanceKlass::fully_initialized);
3252 __ jcc(Assembler::notEqual, slow_case);
3253
3254 // get instance_size in InstanceKlass (scaled to a count of bytes)
3255 __ movl(rdx,
3256 Address(rsi,
3257 Klass::layout_helper_offset()));
3258 // test to see if it has a finalizer or is malformed in some way
3259 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3260 __ jcc(Assembler::notZero, slow_case);
3261
3262 // Allocate the instance
3263 // 1) Try to allocate in the TLAB
3264 // 2) if fail and the object is large allocate in the shared Eden
3265 // 3) if the above fails (or is not applicable), go to a slow case
3266 // (creates a new TLAB, etc.)
3267
3268 const bool allow_shared_alloc =
3269 Universe::heap()->supports_inline_contig_alloc();
3270
3271 if (UseTLAB) {
3272 __ movptr(rax, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
3273 __ lea(rbx, Address(rax, rdx, Address::times_1));
3274 __ cmpptr(rbx, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
3275 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3276 __ movptr(Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3277 if (ZeroTLAB) {
3278 // the fields have been already cleared
3279 __ jmp(initialize_header);
3280 } else {
3281 // initialize both the header and fields
3282 __ jmp(initialize_object);
3283 }
3284 }
3285
3286 // Allocation in the shared Eden, if allowed.
3287 //
3288 // rdx: instance size in bytes
3289 if (allow_shared_alloc) {
3290 __ bind(allocate_shared);
3291
3292 ExternalAddress top((address)Universe::heap()->top_addr());
3293 ExternalAddress end((address)Universe::heap()->end_addr());
3294
3295 const Register RtopAddr = rscratch1;
3296 const Register RendAddr = rscratch2;
3297
3298 __ lea(RtopAddr, top);
3299 __ lea(RendAddr, end);
3300 __ movptr(rax, Address(RtopAddr, 0));
3301
3302 // For retries rax gets set by cmpxchgq
3303 Label retry;
3304 __ bind(retry);
3305 __ lea(rbx, Address(rax, rdx, Address::times_1));
3306 __ cmpptr(rbx, Address(RendAddr, 0));
3307 __ jcc(Assembler::above, slow_case);
3308
3309 // Compare rax with the top addr, and if still equal, store the new
3310 // top addr in rbx at the address of the top addr pointer. Sets ZF if was
3311 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3312 //
3313 // rax: object begin
3314 // rbx: object end
3315 // rdx: instance size in bytes
3316 if (os::is_MP()) {
3317 __ lock();
3318 }
3319 __ cmpxchgptr(rbx, Address(RtopAddr, 0));
3320
3321 // if someone beat us on the allocation, try again, otherwise continue
3322 __ jcc(Assembler::notEqual, retry);
3323
3324 __ incr_allocated_bytes(r15_thread, rdx, 0);
3325 }
3326
3327 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3328 // The object is initialized before the header. If the object size is
3329 // zero, go directly to the header initialization.
3330 __ bind(initialize_object);
3331 __ decrementl(rdx, sizeof(oopDesc));
3332 __ jcc(Assembler::zero, initialize_header);
3333
3334 // Initialize object fields
3335 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3336 __ shrl(rdx, LogBytesPerLong); // divide by oopSize to simplify the loop
3337 {
3338 Label loop;
3339 __ bind(loop);
3340 __ movq(Address(rax, rdx, Address::times_8,
3341 sizeof(oopDesc) - oopSize),
3342 rcx);
3343 __ decrementl(rdx);
3344 __ jcc(Assembler::notZero, loop);
3345 }
3346
3347 // initialize object header only.
3348 __ bind(initialize_header);
3349 if (UseBiasedLocking) {
3350 __ movptr(rscratch1, Address(rsi, Klass::prototype_header_offset()));
3351 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()), rscratch1);
3352 } else {
3353 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()),
3354 (intptr_t) markOopDesc::prototype()); // header (address 0x1)
3355 }
3356 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3357 __ store_klass_gap(rax, rcx); // zero klass gap for compressed oops
3358 __ store_klass(rax, rsi); // store klass last
3359
3360 {
3361 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3362 // Trigger dtrace event for fastpath
3363 __ push(atos); // save the return value
3364 __ call_VM_leaf(
3365 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3366 __ pop(atos); // restore the return value
3367
3368 }
3369 __ jmp(done);
3370 }
3371
3372
3373 // slow case
3374 __ bind(slow_case);
3375 __ get_constant_pool(c_rarg1);
3376 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3377 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3378 __ verify_oop(rax);
3379
3380 // continue
3381 __ bind(done);
3382 }
3383
3384 void TemplateTable::newarray() {
3385 transition(itos, atos);
3386 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3387 __ movl(c_rarg2, rax);
3388 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3389 c_rarg1, c_rarg2);
3390 }
3391
3392 void TemplateTable::anewarray() {
3393 transition(itos, atos);
3394 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3395 __ get_constant_pool(c_rarg1);
3396 __ movl(c_rarg3, rax);
3397 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3398 c_rarg1, c_rarg2, c_rarg3);
3399 }
3400
3401 void TemplateTable::arraylength() {
3402 transition(atos, itos);
3403 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3404 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3405 }
3406
3407 void TemplateTable::checkcast() {
3408 transition(atos, atos);
3409 Label done, is_null, ok_is_subtype, quicked, resolved;
3410 __ testptr(rax, rax); // object is in rax
3411 __ jcc(Assembler::zero, is_null);
3412
3413 // Get cpool & tags index
3414 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3415 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3416 // See if bytecode has already been quicked
3417 __ cmpb(Address(rdx, rbx,
3418 Address::times_1,
3419 Array<u1>::base_offset_in_bytes()),
3420 JVM_CONSTANT_Class);
3421 __ jcc(Assembler::equal, quicked);
3422 __ push(atos); // save receiver for result, and for GC
3423 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3424 // vm_result_2 has metadata result
3425 __ get_vm_result_2(rax, r15_thread);
3426 __ pop_ptr(rdx); // restore receiver
3427 __ jmpb(resolved);
3428
3429 // Get superklass in rax and subklass in rbx
3430 __ bind(quicked);
3431 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
3432 __ movptr(rax, Address(rcx, rbx,
3433 Address::times_8, sizeof(ConstantPool)));
3434
3435 __ bind(resolved);
3436 __ load_klass(rbx, rdx);
3437
3438 // Generate subtype check. Blows rcx, rdi. Object in rdx.
3439 // Superklass in rax. Subklass in rbx.
3440 __ gen_subtype_check(rbx, ok_is_subtype);
3441
3442 // Come here on failure
3443 __ push_ptr(rdx);
3444 // object is at TOS
3445 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3446
3447 // Come here on success
3448 __ bind(ok_is_subtype);
3449 __ mov(rax, rdx); // Restore object in rdx
3450
3451 // Collect counts on whether this check-cast sees NULLs a lot or not.
3452 if (ProfileInterpreter) {
3453 __ jmp(done);
3454 __ bind(is_null);
3455 __ profile_null_seen(rcx);
3456 } else {
3457 __ bind(is_null); // same as 'done'
3458 }
3459 __ bind(done);
3460 }
3461
3462 void TemplateTable::instanceof() {
3463 transition(atos, itos);
3464 Label done, is_null, ok_is_subtype, quicked, resolved;
3465 __ testptr(rax, rax);
3466 __ jcc(Assembler::zero, is_null);
3467
3468 // Get cpool & tags index
3469 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3470 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3471 // See if bytecode has already been quicked
3472 __ cmpb(Address(rdx, rbx,
3473 Address::times_1,
3474 Array<u1>::base_offset_in_bytes()),
3475 JVM_CONSTANT_Class);
3476 __ jcc(Assembler::equal, quicked);
3477
3478 __ push(atos); // save receiver for result, and for GC
3479 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3480 // vm_result_2 has metadata result
3481 __ get_vm_result_2(rax, r15_thread);
3482 __ pop_ptr(rdx); // restore receiver
3483 __ verify_oop(rdx);
3484 __ load_klass(rdx, rdx);
3485 __ jmpb(resolved);
3486
3487 // Get superklass in rax and subklass in rdx
3488 __ bind(quicked);
3489 __ load_klass(rdx, rax);
3490 __ movptr(rax, Address(rcx, rbx,
3491 Address::times_8, sizeof(ConstantPool)));
3492
3493 __ bind(resolved);
3494
3495 // Generate subtype check. Blows rcx, rdi
3496 // Superklass in rax. Subklass in rdx.
3497 __ gen_subtype_check(rdx, ok_is_subtype);
3498
3499 // Come here on failure
3500 __ xorl(rax, rax);
3501 __ jmpb(done);
3502 // Come here on success
3503 __ bind(ok_is_subtype);
3504 __ movl(rax, 1);
3505
3506 // Collect counts on whether this test sees NULLs a lot or not.
3507 if (ProfileInterpreter) {
3508 __ jmp(done);
3509 __ bind(is_null);
3510 __ profile_null_seen(rcx);
3511 } else {
3512 __ bind(is_null); // same as 'done'
3513 }
3514 __ bind(done);
3515 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
3516 // rax = 1: obj != NULL and obj is an instanceof the specified klass
3517 }
3518
3519 //-----------------------------------------------------------------------------
3520 // Breakpoints
3521 void TemplateTable::_breakpoint() {
3522 // Note: We get here even if we are single stepping..
3523 // jbug inists on setting breakpoints at every bytecode
3524 // even if we are in single step mode.
3525
3526 transition(vtos, vtos);
3527
3528 // get the unpatched byte code
3529 __ get_method(c_rarg1);
3530 __ call_VM(noreg,
3531 CAST_FROM_FN_PTR(address,
3532 InterpreterRuntime::get_original_bytecode_at),
3533 c_rarg1, r13);
3534 __ mov(rbx, rax);
3535
3536 // post the breakpoint event
3537 __ get_method(c_rarg1);
3538 __ call_VM(noreg,
3539 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3540 c_rarg1, r13);
3541
3542 // complete the execution of original bytecode
3543 __ dispatch_only_normal(vtos);
3544 }
3545
3546 //-----------------------------------------------------------------------------
3547 // Exceptions
3548
3549 void TemplateTable::athrow() {
3550 transition(atos, vtos);
3551 __ null_check(rax);
3552 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3553 }
3554
3555 //-----------------------------------------------------------------------------
3556 // Synchronization
3557 //
3558 // Note: monitorenter & exit are symmetric routines; which is reflected
3559 // in the assembly code structure as well
3560 //
3561 // Stack layout:
3562 //
3563 // [expressions ] <--- rsp = expression stack top
3564 // ..
3565 // [expressions ]
3566 // [monitor entry] <--- monitor block top = expression stack bot
3567 // ..
3568 // [monitor entry]
3569 // [frame data ] <--- monitor block bot
3570 // ...
3571 // [saved rbp ] <--- rbp
3572 void TemplateTable::monitorenter() {
3573 transition(atos, vtos);
3574
3575 // check for NULL object
3576 __ null_check(rax);
3577
3578 const Address monitor_block_top(
3579 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3580 const Address monitor_block_bot(
3581 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3582 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3583
3584 Label allocated;
3585
3586 // initialize entry pointer
3587 __ xorl(c_rarg1, c_rarg1); // points to free slot or NULL
3588
3589 // find a free slot in the monitor block (result in c_rarg1)
3590 {
3591 Label entry, loop, exit;
3592 __ movptr(c_rarg3, monitor_block_top); // points to current entry,
3593 // starting with top-most entry
3594 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3595 // of monitor block
3596 __ jmpb(entry);
3597
3598 __ bind(loop);
3599 // check if current entry is used
3600 __ cmpptr(Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
3601 // if not used then remember entry in c_rarg1
3602 __ cmov(Assembler::equal, c_rarg1, c_rarg3);
3603 // check if current entry is for same object
3604 __ cmpptr(rax, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3605 // if same object then stop searching
3606 __ jccb(Assembler::equal, exit);
3607 // otherwise advance to next entry
3608 __ addptr(c_rarg3, entry_size);
3609 __ bind(entry);
3610 // check if bottom reached
3611 __ cmpptr(c_rarg3, c_rarg2);
3612 // if not at bottom then check this entry
3613 __ jcc(Assembler::notEqual, loop);
3614 __ bind(exit);
3615 }
3616
3617 __ testptr(c_rarg1, c_rarg1); // check if a slot has been found
3618 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
3619
3620 // allocate one if there's no free slot
3621 {
3622 Label entry, loop;
3623 // 1. compute new pointers // rsp: old expression stack top
3624 __ movptr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
3625 __ subptr(rsp, entry_size); // move expression stack top
3626 __ subptr(c_rarg1, entry_size); // move expression stack bottom
3627 __ mov(c_rarg3, rsp); // set start value for copy loop
3628 __ movptr(monitor_block_bot, c_rarg1); // set new monitor block bottom
3629 __ jmp(entry);
3630 // 2. move expression stack contents
3631 __ bind(loop);
3632 __ movptr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3633 // word from old location
3634 __ movptr(Address(c_rarg3, 0), c_rarg2); // and store it at new location
3635 __ addptr(c_rarg3, wordSize); // advance to next word
3636 __ bind(entry);
3637 __ cmpptr(c_rarg3, c_rarg1); // check if bottom reached
3638 __ jcc(Assembler::notEqual, loop); // if not at bottom then
3639 // copy next word
3640 }
3641
3642 // call run-time routine
3643 // c_rarg1: points to monitor entry
3644 __ bind(allocated);
3645
3646 // Increment bcp to point to the next bytecode, so exception
3647 // handling for async. exceptions work correctly.
3648 // The object has already been poped from the stack, so the
3649 // expression stack looks correct.
3650 __ increment(r13);
3651
3652 // store object
3653 __ movptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), rax);
3654 __ lock_object(c_rarg1);
3655
3656 // check to make sure this monitor doesn't cause stack overflow after locking
3657 __ save_bcp(); // in case of exception
3658 __ generate_stack_overflow_check(0);
3659
3660 // The bcp has already been incremented. Just need to dispatch to
3661 // next instruction.
3662 __ dispatch_next(vtos);
3663 }
3664
3665
3666 void TemplateTable::monitorexit() {
3667 transition(atos, vtos);
3668
3669 // check for NULL object
3670 __ null_check(rax);
3671
3672 const Address monitor_block_top(
3673 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3674 const Address monitor_block_bot(
3675 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3676 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3677
3678 Label found;
3679
3680 // find matching slot
3681 {
3682 Label entry, loop;
3683 __ movptr(c_rarg1, monitor_block_top); // points to current entry,
3684 // starting with top-most entry
3685 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3686 // of monitor block
3687 __ jmpb(entry);
3688
3689 __ bind(loop);
3690 // check if current entry is for same object
3691 __ cmpptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3692 // if same object then stop searching
3693 __ jcc(Assembler::equal, found);
3694 // otherwise advance to next entry
3695 __ addptr(c_rarg1, entry_size);
3696 __ bind(entry);
3697 // check if bottom reached
3698 __ cmpptr(c_rarg1, c_rarg2);
3699 // if not at bottom then check this entry
3700 __ jcc(Assembler::notEqual, loop);
3701 }
3702
3703 // error handling. Unlocking was not block-structured
3704 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3705 InterpreterRuntime::throw_illegal_monitor_state_exception));
3706 __ should_not_reach_here();
3707
3708 // call run-time routine
3709 // rsi: points to monitor entry
3710 __ bind(found);
3711 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3712 __ unlock_object(c_rarg1);
3713 __ pop_ptr(rax); // discard object
3714 }
3715
3716
3717 // Wide instructions
3718 void TemplateTable::wide() {
3719 transition(vtos, vtos);
3720 __ load_unsigned_byte(rbx, at_bcp(1));
3721 __ lea(rscratch1, ExternalAddress((address)Interpreter::_wentry_point));
3722 __ jmp(Address(rscratch1, rbx, Address::times_8));
3723 // Note: the r13 increment step is part of the individual wide
3724 // bytecode implementations
3725 }
3726
3727
3728 // Multi arrays
3729 void TemplateTable::multianewarray() {
3730 transition(vtos, atos);
3731 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3732 // last dim is on top of stack; we want address of first one:
3733 // first_addr = last_addr + (ndims - 1) * wordSize
3734 __ lea(c_rarg1, Address(rsp, rax, Address::times_8, -wordSize));
3735 call_VM(rax,
3736 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3737 c_rarg1);
3738 __ load_unsigned_byte(rbx, at_bcp(3));
3739 __ lea(rsp, Address(rsp, rbx, Address::times_8));
3740 }
3741 #endif // !CC_INTERP