1 /*
2 * Copyright (c) 2003, 2011, 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 "interpreter/interpreter.hpp"
27 #include "interpreter/interpreterRuntime.hpp"
28 #include "interpreter/templateTable.hpp"
29 #include "memory/universe.inline.hpp"
30 #include "oops/methodDataOop.hpp"
31 #include "oops/objArrayKlass.hpp"
32 #include "oops/oop.inline.hpp"
33 #include "prims/methodHandles.hpp"
34 #include "runtime/sharedRuntime.hpp"
35 #include "runtime/stubRoutines.hpp"
36 #include "runtime/synchronizer.hpp"
37
38 #ifndef CC_INTERP
39
40 #define __ _masm->
41
42 // Platform-dependent initialization
43
44 void TemplateTable::pd_initialize() {
45 // No amd64 specific initialization
46 }
47
48 // Address computation: local variables
49
50 static inline Address iaddress(int n) {
51 return Address(r14, Interpreter::local_offset_in_bytes(n));
52 }
53
54 static inline Address laddress(int n) {
55 return iaddress(n + 1);
56 }
57
58 static inline Address faddress(int n) {
59 return iaddress(n);
60 }
61
62 static inline Address daddress(int n) {
63 return laddress(n);
64 }
65
66 static inline Address aaddress(int n) {
67 return iaddress(n);
68 }
69
70 static inline Address iaddress(Register r) {
71 return Address(r14, r, Address::times_8);
72 }
73
74 static inline Address laddress(Register r) {
75 return Address(r14, r, Address::times_8, Interpreter::local_offset_in_bytes(1));
76 }
77
78 static inline Address faddress(Register r) {
79 return iaddress(r);
80 }
81
82 static inline Address daddress(Register r) {
83 return laddress(r);
84 }
85
86 static inline Address aaddress(Register r) {
87 return iaddress(r);
88 }
89
90 static inline Address at_rsp() {
91 return Address(rsp, 0);
92 }
93
94 // At top of Java expression stack which may be different than esp(). It
95 // isn't for category 1 objects.
96 static inline Address at_tos () {
97 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
98 }
99
100 static inline Address at_tos_p1() {
101 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
102 }
103
104 static inline Address at_tos_p2() {
105 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
106 }
107
108 static inline Address at_tos_p3() {
109 return Address(rsp, Interpreter::expr_offset_in_bytes(3));
110 }
111
112 // Condition conversion
113 static Assembler::Condition j_not(TemplateTable::Condition cc) {
114 switch (cc) {
115 case TemplateTable::equal : return Assembler::notEqual;
116 case TemplateTable::not_equal : return Assembler::equal;
117 case TemplateTable::less : return Assembler::greaterEqual;
118 case TemplateTable::less_equal : return Assembler::greater;
119 case TemplateTable::greater : return Assembler::lessEqual;
120 case TemplateTable::greater_equal: return Assembler::less;
121 }
122 ShouldNotReachHere();
123 return Assembler::zero;
124 }
125
126
127 // Miscelaneous helper routines
128 // Store an oop (or NULL) at the address described by obj.
129 // If val == noreg this means store a NULL
130
131 static void do_oop_store(InterpreterMacroAssembler* _masm,
132 Address obj,
133 Register val,
134 BarrierSet::Name barrier,
135 bool precise) {
136 assert(val == noreg || val == rax, "parameter is just for looks");
137 switch (barrier) {
138 #ifndef SERIALGC
139 case BarrierSet::G1SATBCT:
140 case BarrierSet::G1SATBCTLogging:
141 {
142 // flatten object address if needed
143 if (obj.index() == noreg && obj.disp() == 0) {
144 if (obj.base() != rdx) {
145 __ movq(rdx, obj.base());
146 }
147 } else {
148 __ leaq(rdx, obj);
149 }
150 __ g1_write_barrier_pre(rdx, r8, rbx, val != noreg);
151 if (val == noreg) {
152 __ store_heap_oop_null(Address(rdx, 0));
153 } else {
154 __ store_heap_oop(Address(rdx, 0), val);
155 __ g1_write_barrier_post(rdx, val, r8, rbx);
156 }
157
158 }
159 break;
160 #endif // SERIALGC
161 case BarrierSet::CardTableModRef:
162 case BarrierSet::CardTableExtension:
163 {
164 if (val == noreg) {
165 __ store_heap_oop_null(obj);
166 } else {
167 __ store_heap_oop(obj, val);
168 // flatten object address if needed
169 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
170 __ store_check(obj.base());
171 } else {
172 __ leaq(rdx, obj);
173 __ store_check(rdx);
174 }
175 }
176 }
177 break;
178 case BarrierSet::ModRef:
179 case BarrierSet::Other:
180 if (val == noreg) {
181 __ store_heap_oop_null(obj);
182 } else {
183 __ store_heap_oop(obj, val);
184 }
185 break;
186 default :
187 ShouldNotReachHere();
188
189 }
190 }
191
192 Address TemplateTable::at_bcp(int offset) {
193 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
194 return Address(r13, offset);
195 }
196
197 void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
198 Register scratch,
199 bool load_bc_into_scratch/*=true*/) {
200 if (!RewriteBytecodes) {
201 return;
202 }
203 // the pair bytecodes have already done the load.
204 if (load_bc_into_scratch) {
205 __ movl(bc, bytecode);
206 }
207 Label patch_done;
208 if (JvmtiExport::can_post_breakpoint()) {
209 Label fast_patch;
210 // if a breakpoint is present we can't rewrite the stream directly
211 __ movzbl(scratch, at_bcp(0));
212 __ cmpl(scratch, Bytecodes::_breakpoint);
213 __ jcc(Assembler::notEqual, fast_patch);
214 __ get_method(scratch);
215 // Let breakpoint table handling rewrite to quicker bytecode
216 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, r13, bc);
217 #ifndef ASSERT
218 __ jmpb(patch_done);
219 #else
220 __ jmp(patch_done);
221 #endif
222 __ bind(fast_patch);
223 }
224 #ifdef ASSERT
225 Label okay;
226 __ load_unsigned_byte(scratch, at_bcp(0));
227 __ cmpl(scratch, (int) Bytecodes::java_code(bytecode));
228 __ jcc(Assembler::equal, okay);
229 __ cmpl(scratch, bc);
230 __ jcc(Assembler::equal, okay);
231 __ stop("patching the wrong bytecode");
232 __ bind(okay);
233 #endif
234 // patch bytecode
235 __ movb(at_bcp(0), bc);
236 __ bind(patch_done);
237 }
238
239
240 // Individual instructions
241
242 void TemplateTable::nop() {
243 transition(vtos, vtos);
244 // nothing to do
245 }
246
247 void TemplateTable::shouldnotreachhere() {
248 transition(vtos, vtos);
249 __ stop("shouldnotreachhere bytecode");
250 }
251
252 void TemplateTable::aconst_null() {
253 transition(vtos, atos);
254 __ xorl(rax, rax);
255 }
256
257 void TemplateTable::iconst(int value) {
258 transition(vtos, itos);
259 if (value == 0) {
260 __ xorl(rax, rax);
261 } else {
262 __ movl(rax, value);
263 }
264 }
265
266 void TemplateTable::lconst(int value) {
267 transition(vtos, ltos);
268 if (value == 0) {
269 __ xorl(rax, rax);
270 } else {
271 __ movl(rax, value);
272 }
273 }
274
275 void TemplateTable::fconst(int value) {
276 transition(vtos, ftos);
277 static float one = 1.0f, two = 2.0f;
278 switch (value) {
279 case 0:
280 __ xorps(xmm0, xmm0);
281 break;
282 case 1:
283 __ movflt(xmm0, ExternalAddress((address) &one));
284 break;
285 case 2:
286 __ movflt(xmm0, ExternalAddress((address) &two));
287 break;
288 default:
289 ShouldNotReachHere();
290 break;
291 }
292 }
293
294 void TemplateTable::dconst(int value) {
295 transition(vtos, dtos);
296 static double one = 1.0;
297 switch (value) {
298 case 0:
299 __ xorpd(xmm0, xmm0);
300 break;
301 case 1:
302 __ movdbl(xmm0, ExternalAddress((address) &one));
303 break;
304 default:
305 ShouldNotReachHere();
306 break;
307 }
308 }
309
310 void TemplateTable::bipush() {
311 transition(vtos, itos);
312 __ load_signed_byte(rax, at_bcp(1));
313 }
314
315 void TemplateTable::sipush() {
316 transition(vtos, itos);
317 __ load_unsigned_short(rax, at_bcp(1));
318 __ bswapl(rax);
319 __ sarl(rax, 16);
320 }
321
322 void TemplateTable::ldc(bool wide) {
323 transition(vtos, vtos);
324 Label call_ldc, notFloat, notClass, Done;
325
326 if (wide) {
327 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
328 } else {
329 __ load_unsigned_byte(rbx, at_bcp(1));
330 }
331
332 __ get_cpool_and_tags(rcx, rax);
333 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
334 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
335
336 // get type
337 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
338
339 // unresolved string - get the resolved string
340 __ cmpl(rdx, JVM_CONSTANT_UnresolvedString);
341 __ jccb(Assembler::equal, call_ldc);
342
343 // unresolved class - get the resolved class
344 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
345 __ jccb(Assembler::equal, call_ldc);
346
347 // unresolved class in error state - call into runtime to throw the error
348 // from the first resolution attempt
349 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
350 __ jccb(Assembler::equal, call_ldc);
351
352 // resolved class - need to call vm to get java mirror of the class
353 __ cmpl(rdx, JVM_CONSTANT_Class);
354 __ jcc(Assembler::notEqual, notClass);
355
356 __ bind(call_ldc);
357 __ movl(c_rarg1, wide);
358 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
359 __ push_ptr(rax);
360 __ verify_oop(rax);
361 __ jmp(Done);
362
363 __ bind(notClass);
364 __ cmpl(rdx, JVM_CONSTANT_Float);
365 __ jccb(Assembler::notEqual, notFloat);
366 // ftos
367 __ movflt(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
368 __ push_f();
369 __ jmp(Done);
370
371 __ bind(notFloat);
372 #ifdef ASSERT
373 {
374 Label L;
375 __ cmpl(rdx, JVM_CONSTANT_Integer);
376 __ jcc(Assembler::equal, L);
377 __ cmpl(rdx, JVM_CONSTANT_String);
378 __ jcc(Assembler::equal, L);
379 __ stop("unexpected tag type in ldc");
380 __ bind(L);
381 }
382 #endif
383 // atos and itos
384 Label isOop;
385 __ cmpl(rdx, JVM_CONSTANT_Integer);
386 __ jcc(Assembler::notEqual, isOop);
387 __ movl(rax, Address(rcx, rbx, Address::times_8, base_offset));
388 __ push_i(rax);
389 __ jmp(Done);
390
391 __ bind(isOop);
392 __ movptr(rax, Address(rcx, rbx, Address::times_8, base_offset));
393 __ push_ptr(rax);
394
395 if (VerifyOops) {
396 __ verify_oop(rax);
397 }
398
399 __ bind(Done);
400 }
401
402 // Fast path for caching oop constants.
403 // %%% We should use this to handle Class and String constants also.
404 // %%% It will simplify the ldc/primitive path considerably.
405 void TemplateTable::fast_aldc(bool wide) {
406 transition(vtos, atos);
407
408 if (!EnableInvokeDynamic) {
409 // We should not encounter this bytecode if !EnableInvokeDynamic.
410 // The verifier will stop it. However, if we get past the verifier,
411 // this will stop the thread in a reasonable way, without crashing the JVM.
412 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
413 InterpreterRuntime::throw_IncompatibleClassChangeError));
414 // the call_VM checks for exception, so we should never return here.
415 __ should_not_reach_here();
416 return;
417 }
418
419 const Register cache = rcx;
420 const Register index = rdx;
421
422 resolve_cache_and_index(f1_oop, rax, cache, index, wide ? sizeof(u2) : sizeof(u1));
423 if (VerifyOops) {
424 __ verify_oop(rax);
425 }
426
427 Label L_done, L_throw_exception;
428 const Register con_klass_temp = rcx; // same as cache
429 const Register array_klass_temp = rdx; // same as index
430 __ load_klass(con_klass_temp, rax);
431 __ lea(array_klass_temp, ExternalAddress((address)Universe::systemObjArrayKlassObj_addr()));
432 __ cmpptr(con_klass_temp, Address(array_klass_temp, 0));
433 __ jcc(Assembler::notEqual, L_done);
434 __ cmpl(Address(rax, arrayOopDesc::length_offset_in_bytes()), 0);
435 __ jcc(Assembler::notEqual, L_throw_exception);
436 __ xorptr(rax, rax);
437 __ jmp(L_done);
438
439 // Load the exception from the system-array which wraps it:
440 __ bind(L_throw_exception);
441 __ load_heap_oop(rax, Address(rax, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
442 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
443
444 __ bind(L_done);
445 }
446
447 void TemplateTable::ldc2_w() {
448 transition(vtos, vtos);
449 Label Long, Done;
450 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
451
452 __ get_cpool_and_tags(rcx, rax);
453 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
454 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
455
456 // get type
457 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
458 JVM_CONSTANT_Double);
459 __ jccb(Assembler::notEqual, Long);
460 // dtos
461 __ movdbl(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
462 __ push_d();
463 __ jmpb(Done);
464
465 __ bind(Long);
466 // ltos
467 __ movq(rax, Address(rcx, rbx, Address::times_8, base_offset));
468 __ push_l();
469
470 __ bind(Done);
471 }
472
473 void TemplateTable::locals_index(Register reg, int offset) {
474 __ load_unsigned_byte(reg, at_bcp(offset));
475 __ negptr(reg);
476 }
477
478 void TemplateTable::iload() {
479 transition(vtos, itos);
480 if (RewriteFrequentPairs) {
481 Label rewrite, done;
482 const Register bc = c_rarg3;
483 assert(rbx != bc, "register damaged");
484
485 // get next byte
486 __ load_unsigned_byte(rbx,
487 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
488 // if _iload, wait to rewrite to iload2. We only want to rewrite the
489 // last two iloads in a pair. Comparing against fast_iload means that
490 // the next bytecode is neither an iload or a caload, and therefore
491 // an iload pair.
492 __ cmpl(rbx, Bytecodes::_iload);
493 __ jcc(Assembler::equal, done);
494
495 __ cmpl(rbx, Bytecodes::_fast_iload);
496 __ movl(bc, Bytecodes::_fast_iload2);
497 __ jccb(Assembler::equal, rewrite);
498
499 // if _caload, rewrite to fast_icaload
500 __ cmpl(rbx, Bytecodes::_caload);
501 __ movl(bc, Bytecodes::_fast_icaload);
502 __ jccb(Assembler::equal, rewrite);
503
504 // rewrite so iload doesn't check again.
505 __ movl(bc, Bytecodes::_fast_iload);
506
507 // rewrite
508 // bc: fast bytecode
509 __ bind(rewrite);
510 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
511 __ bind(done);
512 }
513
514 // Get the local value into tos
515 locals_index(rbx);
516 __ movl(rax, iaddress(rbx));
517 }
518
519 void TemplateTable::fast_iload2() {
520 transition(vtos, itos);
521 locals_index(rbx);
522 __ movl(rax, iaddress(rbx));
523 __ push(itos);
524 locals_index(rbx, 3);
525 __ movl(rax, iaddress(rbx));
526 }
527
528 void TemplateTable::fast_iload() {
529 transition(vtos, itos);
530 locals_index(rbx);
531 __ movl(rax, iaddress(rbx));
532 }
533
534 void TemplateTable::lload() {
535 transition(vtos, ltos);
536 locals_index(rbx);
537 __ movq(rax, laddress(rbx));
538 }
539
540 void TemplateTable::fload() {
541 transition(vtos, ftos);
542 locals_index(rbx);
543 __ movflt(xmm0, faddress(rbx));
544 }
545
546 void TemplateTable::dload() {
547 transition(vtos, dtos);
548 locals_index(rbx);
549 __ movdbl(xmm0, daddress(rbx));
550 }
551
552 void TemplateTable::aload() {
553 transition(vtos, atos);
554 locals_index(rbx);
555 __ movptr(rax, aaddress(rbx));
556 }
557
558 void TemplateTable::locals_index_wide(Register reg) {
559 __ movl(reg, at_bcp(2));
560 __ bswapl(reg);
561 __ shrl(reg, 16);
562 __ negptr(reg);
563 }
564
565 void TemplateTable::wide_iload() {
566 transition(vtos, itos);
567 locals_index_wide(rbx);
568 __ movl(rax, iaddress(rbx));
569 }
570
571 void TemplateTable::wide_lload() {
572 transition(vtos, ltos);
573 locals_index_wide(rbx);
574 __ movq(rax, laddress(rbx));
575 }
576
577 void TemplateTable::wide_fload() {
578 transition(vtos, ftos);
579 locals_index_wide(rbx);
580 __ movflt(xmm0, faddress(rbx));
581 }
582
583 void TemplateTable::wide_dload() {
584 transition(vtos, dtos);
585 locals_index_wide(rbx);
586 __ movdbl(xmm0, daddress(rbx));
587 }
588
589 void TemplateTable::wide_aload() {
590 transition(vtos, atos);
591 locals_index_wide(rbx);
592 __ movptr(rax, aaddress(rbx));
593 }
594
595 void TemplateTable::index_check(Register array, Register index) {
596 // destroys rbx
597 // check array
598 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
599 // sign extend index for use by indexed load
600 __ movl2ptr(index, index);
601 // check index
602 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
603 if (index != rbx) {
604 // ??? convention: move aberrant index into ebx for exception message
605 assert(rbx != array, "different registers");
606 __ movl(rbx, index);
607 }
608 __ jump_cc(Assembler::aboveEqual,
609 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
610 }
611
612 void TemplateTable::iaload() {
613 transition(itos, itos);
614 __ pop_ptr(rdx);
615 // eax: index
616 // rdx: array
617 index_check(rdx, rax); // kills rbx
618 __ movl(rax, Address(rdx, rax,
619 Address::times_4,
620 arrayOopDesc::base_offset_in_bytes(T_INT)));
621 }
622
623 void TemplateTable::laload() {
624 transition(itos, ltos);
625 __ pop_ptr(rdx);
626 // eax: index
627 // rdx: array
628 index_check(rdx, rax); // kills rbx
629 __ movq(rax, Address(rdx, rbx,
630 Address::times_8,
631 arrayOopDesc::base_offset_in_bytes(T_LONG)));
632 }
633
634 void TemplateTable::faload() {
635 transition(itos, ftos);
636 __ pop_ptr(rdx);
637 // eax: index
638 // rdx: array
639 index_check(rdx, rax); // kills rbx
640 __ movflt(xmm0, Address(rdx, rax,
641 Address::times_4,
642 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
643 }
644
645 void TemplateTable::daload() {
646 transition(itos, dtos);
647 __ pop_ptr(rdx);
648 // eax: index
649 // rdx: array
650 index_check(rdx, rax); // kills rbx
651 __ movdbl(xmm0, Address(rdx, rax,
652 Address::times_8,
653 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
654 }
655
656 void TemplateTable::aaload() {
657 transition(itos, atos);
658 __ pop_ptr(rdx);
659 // eax: index
660 // rdx: array
661 index_check(rdx, rax); // kills rbx
662 __ load_heap_oop(rax, Address(rdx, rax,
663 UseCompressedOops ? Address::times_4 : Address::times_8,
664 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
665 }
666
667 void TemplateTable::baload() {
668 transition(itos, itos);
669 __ pop_ptr(rdx);
670 // eax: index
671 // rdx: array
672 index_check(rdx, rax); // kills rbx
673 __ load_signed_byte(rax,
674 Address(rdx, rax,
675 Address::times_1,
676 arrayOopDesc::base_offset_in_bytes(T_BYTE)));
677 }
678
679 void TemplateTable::caload() {
680 transition(itos, itos);
681 __ pop_ptr(rdx);
682 // eax: index
683 // rdx: array
684 index_check(rdx, rax); // kills rbx
685 __ load_unsigned_short(rax,
686 Address(rdx, rax,
687 Address::times_2,
688 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
689 }
690
691 // iload followed by caload frequent pair
692 void TemplateTable::fast_icaload() {
693 transition(vtos, itos);
694 // load index out of locals
695 locals_index(rbx);
696 __ movl(rax, iaddress(rbx));
697
698 // eax: index
699 // rdx: array
700 __ pop_ptr(rdx);
701 index_check(rdx, rax); // kills rbx
702 __ load_unsigned_short(rax,
703 Address(rdx, rax,
704 Address::times_2,
705 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
706 }
707
708 void TemplateTable::saload() {
709 transition(itos, itos);
710 __ pop_ptr(rdx);
711 // eax: index
712 // rdx: array
713 index_check(rdx, rax); // kills rbx
714 __ load_signed_short(rax,
715 Address(rdx, rax,
716 Address::times_2,
717 arrayOopDesc::base_offset_in_bytes(T_SHORT)));
718 }
719
720 void TemplateTable::iload(int n) {
721 transition(vtos, itos);
722 __ movl(rax, iaddress(n));
723 }
724
725 void TemplateTable::lload(int n) {
726 transition(vtos, ltos);
727 __ movq(rax, laddress(n));
728 }
729
730 void TemplateTable::fload(int n) {
731 transition(vtos, ftos);
732 __ movflt(xmm0, faddress(n));
733 }
734
735 void TemplateTable::dload(int n) {
736 transition(vtos, dtos);
737 __ movdbl(xmm0, daddress(n));
738 }
739
740 void TemplateTable::aload(int n) {
741 transition(vtos, atos);
742 __ movptr(rax, aaddress(n));
743 }
744
745 void TemplateTable::aload_0() {
746 transition(vtos, atos);
747 // According to bytecode histograms, the pairs:
748 //
749 // _aload_0, _fast_igetfield
750 // _aload_0, _fast_agetfield
751 // _aload_0, _fast_fgetfield
752 //
753 // occur frequently. If RewriteFrequentPairs is set, the (slow)
754 // _aload_0 bytecode checks if the next bytecode is either
755 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
756 // rewrites the current bytecode into a pair bytecode; otherwise it
757 // rewrites the current bytecode into _fast_aload_0 that doesn't do
758 // the pair check anymore.
759 //
760 // Note: If the next bytecode is _getfield, the rewrite must be
761 // delayed, otherwise we may miss an opportunity for a pair.
762 //
763 // Also rewrite frequent pairs
764 // aload_0, aload_1
765 // aload_0, iload_1
766 // These bytecodes with a small amount of code are most profitable
767 // to rewrite
768 if (RewriteFrequentPairs) {
769 Label rewrite, done;
770 const Register bc = c_rarg3;
771 assert(rbx != bc, "register damaged");
772 // get next byte
773 __ load_unsigned_byte(rbx,
774 at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
775
776 // do actual aload_0
777 aload(0);
778
779 // if _getfield then wait with rewrite
780 __ cmpl(rbx, Bytecodes::_getfield);
781 __ jcc(Assembler::equal, done);
782
783 // if _igetfield then reqrite to _fast_iaccess_0
784 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) ==
785 Bytecodes::_aload_0,
786 "fix bytecode definition");
787 __ cmpl(rbx, Bytecodes::_fast_igetfield);
788 __ movl(bc, Bytecodes::_fast_iaccess_0);
789 __ jccb(Assembler::equal, rewrite);
790
791 // if _agetfield then reqrite to _fast_aaccess_0
792 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) ==
793 Bytecodes::_aload_0,
794 "fix bytecode definition");
795 __ cmpl(rbx, Bytecodes::_fast_agetfield);
796 __ movl(bc, Bytecodes::_fast_aaccess_0);
797 __ jccb(Assembler::equal, rewrite);
798
799 // if _fgetfield then reqrite to _fast_faccess_0
800 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) ==
801 Bytecodes::_aload_0,
802 "fix bytecode definition");
803 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
804 __ movl(bc, Bytecodes::_fast_faccess_0);
805 __ jccb(Assembler::equal, rewrite);
806
807 // else rewrite to _fast_aload0
808 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) ==
809 Bytecodes::_aload_0,
810 "fix bytecode definition");
811 __ movl(bc, Bytecodes::_fast_aload_0);
812
813 // rewrite
814 // bc: fast bytecode
815 __ bind(rewrite);
816 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
817
818 __ bind(done);
819 } else {
820 aload(0);
821 }
822 }
823
824 void TemplateTable::istore() {
825 transition(itos, vtos);
826 locals_index(rbx);
827 __ movl(iaddress(rbx), rax);
828 }
829
830 void TemplateTable::lstore() {
831 transition(ltos, vtos);
832 locals_index(rbx);
833 __ movq(laddress(rbx), rax);
834 }
835
836 void TemplateTable::fstore() {
837 transition(ftos, vtos);
838 locals_index(rbx);
839 __ movflt(faddress(rbx), xmm0);
840 }
841
842 void TemplateTable::dstore() {
843 transition(dtos, vtos);
844 locals_index(rbx);
845 __ movdbl(daddress(rbx), xmm0);
846 }
847
848 void TemplateTable::astore() {
849 transition(vtos, vtos);
850 __ pop_ptr(rax);
851 locals_index(rbx);
852 __ movptr(aaddress(rbx), rax);
853 }
854
855 void TemplateTable::wide_istore() {
856 transition(vtos, vtos);
857 __ pop_i();
858 locals_index_wide(rbx);
859 __ movl(iaddress(rbx), rax);
860 }
861
862 void TemplateTable::wide_lstore() {
863 transition(vtos, vtos);
864 __ pop_l();
865 locals_index_wide(rbx);
866 __ movq(laddress(rbx), rax);
867 }
868
869 void TemplateTable::wide_fstore() {
870 transition(vtos, vtos);
871 __ pop_f();
872 locals_index_wide(rbx);
873 __ movflt(faddress(rbx), xmm0);
874 }
875
876 void TemplateTable::wide_dstore() {
877 transition(vtos, vtos);
878 __ pop_d();
879 locals_index_wide(rbx);
880 __ movdbl(daddress(rbx), xmm0);
881 }
882
883 void TemplateTable::wide_astore() {
884 transition(vtos, vtos);
885 __ pop_ptr(rax);
886 locals_index_wide(rbx);
887 __ movptr(aaddress(rbx), rax);
888 }
889
890 void TemplateTable::iastore() {
891 transition(itos, vtos);
892 __ pop_i(rbx);
893 __ pop_ptr(rdx);
894 // eax: value
895 // ebx: index
896 // rdx: array
897 index_check(rdx, rbx); // prefer index in ebx
898 __ movl(Address(rdx, rbx,
899 Address::times_4,
900 arrayOopDesc::base_offset_in_bytes(T_INT)),
901 rax);
902 }
903
904 void TemplateTable::lastore() {
905 transition(ltos, vtos);
906 __ pop_i(rbx);
907 __ pop_ptr(rdx);
908 // rax: value
909 // ebx: index
910 // rdx: array
911 index_check(rdx, rbx); // prefer index in ebx
912 __ movq(Address(rdx, rbx,
913 Address::times_8,
914 arrayOopDesc::base_offset_in_bytes(T_LONG)),
915 rax);
916 }
917
918 void TemplateTable::fastore() {
919 transition(ftos, vtos);
920 __ pop_i(rbx);
921 __ pop_ptr(rdx);
922 // xmm0: value
923 // ebx: index
924 // rdx: array
925 index_check(rdx, rbx); // prefer index in ebx
926 __ movflt(Address(rdx, rbx,
927 Address::times_4,
928 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
929 xmm0);
930 }
931
932 void TemplateTable::dastore() {
933 transition(dtos, vtos);
934 __ pop_i(rbx);
935 __ pop_ptr(rdx);
936 // xmm0: value
937 // ebx: index
938 // rdx: array
939 index_check(rdx, rbx); // prefer index in ebx
940 __ movdbl(Address(rdx, rbx,
941 Address::times_8,
942 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
943 xmm0);
944 }
945
946 void TemplateTable::aastore() {
947 Label is_null, ok_is_subtype, done;
948 transition(vtos, vtos);
949 // stack: ..., array, index, value
950 __ movptr(rax, at_tos()); // value
951 __ movl(rcx, at_tos_p1()); // index
952 __ movptr(rdx, at_tos_p2()); // array
953
954 Address element_address(rdx, rcx,
955 UseCompressedOops? Address::times_4 : Address::times_8,
956 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
957
958 index_check(rdx, rcx); // kills rbx
959 // do array store check - check for NULL value first
960 __ testptr(rax, rax);
961 __ jcc(Assembler::zero, is_null);
962
963 // Move subklass into rbx
964 __ load_klass(rbx, rax);
965 // Move superklass into rax
966 __ load_klass(rax, rdx);
967 __ movptr(rax, Address(rax,
968 sizeof(oopDesc) +
969 objArrayKlass::element_klass_offset_in_bytes()));
970 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
971 __ lea(rdx, element_address);
972
973 // Generate subtype check. Blows rcx, rdi
974 // Superklass in rax. Subklass in rbx.
975 __ gen_subtype_check(rbx, ok_is_subtype);
976
977 // Come here on failure
978 // object is at TOS
979 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
980
981 // Come here on success
982 __ bind(ok_is_subtype);
983
984 // Get the value we will store
985 __ movptr(rax, at_tos());
986 // Now store using the appropriate barrier
987 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
988 __ jmp(done);
989
990 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
991 __ bind(is_null);
992 __ profile_null_seen(rbx);
993
994 // Store a NULL
995 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
996
997 // Pop stack arguments
998 __ bind(done);
999 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1000 }
1001
1002 void TemplateTable::bastore() {
1003 transition(itos, vtos);
1004 __ pop_i(rbx);
1005 __ pop_ptr(rdx);
1006 // eax: value
1007 // ebx: index
1008 // rdx: array
1009 index_check(rdx, rbx); // prefer index in ebx
1010 __ movb(Address(rdx, rbx,
1011 Address::times_1,
1012 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1013 rax);
1014 }
1015
1016 void TemplateTable::castore() {
1017 transition(itos, vtos);
1018 __ pop_i(rbx);
1019 __ pop_ptr(rdx);
1020 // eax: value
1021 // ebx: index
1022 // rdx: array
1023 index_check(rdx, rbx); // prefer index in ebx
1024 __ movw(Address(rdx, rbx,
1025 Address::times_2,
1026 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1027 rax);
1028 }
1029
1030 void TemplateTable::sastore() {
1031 castore();
1032 }
1033
1034 void TemplateTable::istore(int n) {
1035 transition(itos, vtos);
1036 __ movl(iaddress(n), rax);
1037 }
1038
1039 void TemplateTable::lstore(int n) {
1040 transition(ltos, vtos);
1041 __ movq(laddress(n), rax);
1042 }
1043
1044 void TemplateTable::fstore(int n) {
1045 transition(ftos, vtos);
1046 __ movflt(faddress(n), xmm0);
1047 }
1048
1049 void TemplateTable::dstore(int n) {
1050 transition(dtos, vtos);
1051 __ movdbl(daddress(n), xmm0);
1052 }
1053
1054 void TemplateTable::astore(int n) {
1055 transition(vtos, vtos);
1056 __ pop_ptr(rax);
1057 __ movptr(aaddress(n), rax);
1058 }
1059
1060 void TemplateTable::pop() {
1061 transition(vtos, vtos);
1062 __ addptr(rsp, Interpreter::stackElementSize);
1063 }
1064
1065 void TemplateTable::pop2() {
1066 transition(vtos, vtos);
1067 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1068 }
1069
1070 void TemplateTable::dup() {
1071 transition(vtos, vtos);
1072 __ load_ptr(0, rax);
1073 __ push_ptr(rax);
1074 // stack: ..., a, a
1075 }
1076
1077 void TemplateTable::dup_x1() {
1078 transition(vtos, vtos);
1079 // stack: ..., a, b
1080 __ load_ptr( 0, rax); // load b
1081 __ load_ptr( 1, rcx); // load a
1082 __ store_ptr(1, rax); // store b
1083 __ store_ptr(0, rcx); // store a
1084 __ push_ptr(rax); // push b
1085 // stack: ..., b, a, b
1086 }
1087
1088 void TemplateTable::dup_x2() {
1089 transition(vtos, vtos);
1090 // stack: ..., a, b, c
1091 __ load_ptr( 0, rax); // load c
1092 __ load_ptr( 2, rcx); // load a
1093 __ store_ptr(2, rax); // store c in a
1094 __ push_ptr(rax); // push c
1095 // stack: ..., c, b, c, c
1096 __ load_ptr( 2, rax); // load b
1097 __ store_ptr(2, rcx); // store a in b
1098 // stack: ..., c, a, c, c
1099 __ store_ptr(1, rax); // store b in c
1100 // stack: ..., c, a, b, c
1101 }
1102
1103 void TemplateTable::dup2() {
1104 transition(vtos, vtos);
1105 // stack: ..., a, b
1106 __ load_ptr(1, rax); // load a
1107 __ push_ptr(rax); // push a
1108 __ load_ptr(1, rax); // load b
1109 __ push_ptr(rax); // push b
1110 // stack: ..., a, b, a, b
1111 }
1112
1113 void TemplateTable::dup2_x1() {
1114 transition(vtos, vtos);
1115 // stack: ..., a, b, c
1116 __ load_ptr( 0, rcx); // load c
1117 __ load_ptr( 1, rax); // load b
1118 __ push_ptr(rax); // push b
1119 __ push_ptr(rcx); // push c
1120 // stack: ..., a, b, c, b, c
1121 __ store_ptr(3, rcx); // store c in b
1122 // stack: ..., a, c, c, b, c
1123 __ load_ptr( 4, rcx); // load a
1124 __ store_ptr(2, rcx); // store a in 2nd c
1125 // stack: ..., a, c, a, b, c
1126 __ store_ptr(4, rax); // store b in a
1127 // stack: ..., b, c, a, b, c
1128 }
1129
1130 void TemplateTable::dup2_x2() {
1131 transition(vtos, vtos);
1132 // stack: ..., a, b, c, d
1133 __ load_ptr( 0, rcx); // load d
1134 __ load_ptr( 1, rax); // load c
1135 __ push_ptr(rax); // push c
1136 __ push_ptr(rcx); // push d
1137 // stack: ..., a, b, c, d, c, d
1138 __ load_ptr( 4, rax); // load b
1139 __ store_ptr(2, rax); // store b in d
1140 __ store_ptr(4, rcx); // store d in b
1141 // stack: ..., a, d, c, b, c, d
1142 __ load_ptr( 5, rcx); // load a
1143 __ load_ptr( 3, rax); // load c
1144 __ store_ptr(3, rcx); // store a in c
1145 __ store_ptr(5, rax); // store c in a
1146 // stack: ..., c, d, a, b, c, d
1147 }
1148
1149 void TemplateTable::swap() {
1150 transition(vtos, vtos);
1151 // stack: ..., a, b
1152 __ load_ptr( 1, rcx); // load a
1153 __ load_ptr( 0, rax); // load b
1154 __ store_ptr(0, rcx); // store a in b
1155 __ store_ptr(1, rax); // store b in a
1156 // stack: ..., b, a
1157 }
1158
1159 void TemplateTable::iop2(Operation op) {
1160 transition(itos, itos);
1161 switch (op) {
1162 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1163 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1164 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1165 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1166 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1167 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1168 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1169 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1170 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1171 default : ShouldNotReachHere();
1172 }
1173 }
1174
1175 void TemplateTable::lop2(Operation op) {
1176 transition(ltos, ltos);
1177 switch (op) {
1178 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1179 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1180 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1181 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1182 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1183 default : ShouldNotReachHere();
1184 }
1185 }
1186
1187 void TemplateTable::idiv() {
1188 transition(itos, itos);
1189 __ movl(rcx, rax);
1190 __ pop_i(rax);
1191 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1192 // they are not equal, one could do a normal division (no correction
1193 // needed), which may speed up this implementation for the common case.
1194 // (see also JVM spec., p.243 & p.271)
1195 __ corrected_idivl(rcx);
1196 }
1197
1198 void TemplateTable::irem() {
1199 transition(itos, itos);
1200 __ movl(rcx, rax);
1201 __ pop_i(rax);
1202 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1203 // they are not equal, one could do a normal division (no correction
1204 // needed), which may speed up this implementation for the common case.
1205 // (see also JVM spec., p.243 & p.271)
1206 __ corrected_idivl(rcx);
1207 __ movl(rax, rdx);
1208 }
1209
1210 void TemplateTable::lmul() {
1211 transition(ltos, ltos);
1212 __ pop_l(rdx);
1213 __ imulq(rax, rdx);
1214 }
1215
1216 void TemplateTable::ldiv() {
1217 transition(ltos, ltos);
1218 __ mov(rcx, rax);
1219 __ pop_l(rax);
1220 // generate explicit div0 check
1221 __ testq(rcx, rcx);
1222 __ jump_cc(Assembler::zero,
1223 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1224 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1225 // they are not equal, one could do a normal division (no correction
1226 // needed), which may speed up this implementation for the common case.
1227 // (see also JVM spec., p.243 & p.271)
1228 __ corrected_idivq(rcx); // kills rbx
1229 }
1230
1231 void TemplateTable::lrem() {
1232 transition(ltos, ltos);
1233 __ mov(rcx, rax);
1234 __ pop_l(rax);
1235 __ testq(rcx, rcx);
1236 __ jump_cc(Assembler::zero,
1237 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1238 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1239 // they are not equal, one could do a normal division (no correction
1240 // needed), which may speed up this implementation for the common case.
1241 // (see also JVM spec., p.243 & p.271)
1242 __ corrected_idivq(rcx); // kills rbx
1243 __ mov(rax, rdx);
1244 }
1245
1246 void TemplateTable::lshl() {
1247 transition(itos, ltos);
1248 __ movl(rcx, rax); // get shift count
1249 __ pop_l(rax); // get shift value
1250 __ shlq(rax);
1251 }
1252
1253 void TemplateTable::lshr() {
1254 transition(itos, ltos);
1255 __ movl(rcx, rax); // get shift count
1256 __ pop_l(rax); // get shift value
1257 __ sarq(rax);
1258 }
1259
1260 void TemplateTable::lushr() {
1261 transition(itos, ltos);
1262 __ movl(rcx, rax); // get shift count
1263 __ pop_l(rax); // get shift value
1264 __ shrq(rax);
1265 }
1266
1267 void TemplateTable::fop2(Operation op) {
1268 transition(ftos, ftos);
1269 switch (op) {
1270 case add:
1271 __ addss(xmm0, at_rsp());
1272 __ addptr(rsp, Interpreter::stackElementSize);
1273 break;
1274 case sub:
1275 __ movflt(xmm1, xmm0);
1276 __ pop_f(xmm0);
1277 __ subss(xmm0, xmm1);
1278 break;
1279 case mul:
1280 __ mulss(xmm0, at_rsp());
1281 __ addptr(rsp, Interpreter::stackElementSize);
1282 break;
1283 case div:
1284 __ movflt(xmm1, xmm0);
1285 __ pop_f(xmm0);
1286 __ divss(xmm0, xmm1);
1287 break;
1288 case rem:
1289 __ movflt(xmm1, xmm0);
1290 __ pop_f(xmm0);
1291 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1292 break;
1293 default:
1294 ShouldNotReachHere();
1295 break;
1296 }
1297 }
1298
1299 void TemplateTable::dop2(Operation op) {
1300 transition(dtos, dtos);
1301 switch (op) {
1302 case add:
1303 __ addsd(xmm0, at_rsp());
1304 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1305 break;
1306 case sub:
1307 __ movdbl(xmm1, xmm0);
1308 __ pop_d(xmm0);
1309 __ subsd(xmm0, xmm1);
1310 break;
1311 case mul:
1312 __ mulsd(xmm0, at_rsp());
1313 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1314 break;
1315 case div:
1316 __ movdbl(xmm1, xmm0);
1317 __ pop_d(xmm0);
1318 __ divsd(xmm0, xmm1);
1319 break;
1320 case rem:
1321 __ movdbl(xmm1, xmm0);
1322 __ pop_d(xmm0);
1323 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1324 break;
1325 default:
1326 ShouldNotReachHere();
1327 break;
1328 }
1329 }
1330
1331 void TemplateTable::ineg() {
1332 transition(itos, itos);
1333 __ negl(rax);
1334 }
1335
1336 void TemplateTable::lneg() {
1337 transition(ltos, ltos);
1338 __ negq(rax);
1339 }
1340
1341 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1342 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1343 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1344 // of 128-bits operands for SSE instructions.
1345 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1346 // Store the value to a 128-bits operand.
1347 operand[0] = lo;
1348 operand[1] = hi;
1349 return operand;
1350 }
1351
1352 // Buffer for 128-bits masks used by SSE instructions.
1353 static jlong float_signflip_pool[2*2];
1354 static jlong double_signflip_pool[2*2];
1355
1356 void TemplateTable::fneg() {
1357 transition(ftos, ftos);
1358 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], 0x8000000080000000, 0x8000000080000000);
1359 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1360 }
1361
1362 void TemplateTable::dneg() {
1363 transition(dtos, dtos);
1364 static jlong *double_signflip = double_quadword(&double_signflip_pool[1], 0x8000000000000000, 0x8000000000000000);
1365 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1366 }
1367
1368 void TemplateTable::iinc() {
1369 transition(vtos, vtos);
1370 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1371 locals_index(rbx);
1372 __ addl(iaddress(rbx), rdx);
1373 }
1374
1375 void TemplateTable::wide_iinc() {
1376 transition(vtos, vtos);
1377 __ movl(rdx, at_bcp(4)); // get constant
1378 locals_index_wide(rbx);
1379 __ bswapl(rdx); // swap bytes & sign-extend constant
1380 __ sarl(rdx, 16);
1381 __ addl(iaddress(rbx), rdx);
1382 // Note: should probably use only one movl to get both
1383 // the index and the constant -> fix this
1384 }
1385
1386 void TemplateTable::convert() {
1387 // Checking
1388 #ifdef ASSERT
1389 {
1390 TosState tos_in = ilgl;
1391 TosState tos_out = ilgl;
1392 switch (bytecode()) {
1393 case Bytecodes::_i2l: // fall through
1394 case Bytecodes::_i2f: // fall through
1395 case Bytecodes::_i2d: // fall through
1396 case Bytecodes::_i2b: // fall through
1397 case Bytecodes::_i2c: // fall through
1398 case Bytecodes::_i2s: tos_in = itos; break;
1399 case Bytecodes::_l2i: // fall through
1400 case Bytecodes::_l2f: // fall through
1401 case Bytecodes::_l2d: tos_in = ltos; break;
1402 case Bytecodes::_f2i: // fall through
1403 case Bytecodes::_f2l: // fall through
1404 case Bytecodes::_f2d: tos_in = ftos; break;
1405 case Bytecodes::_d2i: // fall through
1406 case Bytecodes::_d2l: // fall through
1407 case Bytecodes::_d2f: tos_in = dtos; break;
1408 default : ShouldNotReachHere();
1409 }
1410 switch (bytecode()) {
1411 case Bytecodes::_l2i: // fall through
1412 case Bytecodes::_f2i: // fall through
1413 case Bytecodes::_d2i: // fall through
1414 case Bytecodes::_i2b: // fall through
1415 case Bytecodes::_i2c: // fall through
1416 case Bytecodes::_i2s: tos_out = itos; break;
1417 case Bytecodes::_i2l: // fall through
1418 case Bytecodes::_f2l: // fall through
1419 case Bytecodes::_d2l: tos_out = ltos; break;
1420 case Bytecodes::_i2f: // fall through
1421 case Bytecodes::_l2f: // fall through
1422 case Bytecodes::_d2f: tos_out = ftos; break;
1423 case Bytecodes::_i2d: // fall through
1424 case Bytecodes::_l2d: // fall through
1425 case Bytecodes::_f2d: tos_out = dtos; break;
1426 default : ShouldNotReachHere();
1427 }
1428 transition(tos_in, tos_out);
1429 }
1430 #endif // ASSERT
1431
1432 static const int64_t is_nan = 0x8000000000000000L;
1433
1434 // Conversion
1435 switch (bytecode()) {
1436 case Bytecodes::_i2l:
1437 __ movslq(rax, rax);
1438 break;
1439 case Bytecodes::_i2f:
1440 __ cvtsi2ssl(xmm0, rax);
1441 break;
1442 case Bytecodes::_i2d:
1443 __ cvtsi2sdl(xmm0, rax);
1444 break;
1445 case Bytecodes::_i2b:
1446 __ movsbl(rax, rax);
1447 break;
1448 case Bytecodes::_i2c:
1449 __ movzwl(rax, rax);
1450 break;
1451 case Bytecodes::_i2s:
1452 __ movswl(rax, rax);
1453 break;
1454 case Bytecodes::_l2i:
1455 __ movl(rax, rax);
1456 break;
1457 case Bytecodes::_l2f:
1458 __ cvtsi2ssq(xmm0, rax);
1459 break;
1460 case Bytecodes::_l2d:
1461 __ cvtsi2sdq(xmm0, rax);
1462 break;
1463 case Bytecodes::_f2i:
1464 {
1465 Label L;
1466 __ cvttss2sil(rax, xmm0);
1467 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1468 __ jcc(Assembler::notEqual, L);
1469 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1470 __ bind(L);
1471 }
1472 break;
1473 case Bytecodes::_f2l:
1474 {
1475 Label L;
1476 __ cvttss2siq(rax, xmm0);
1477 // NaN or overflow/underflow?
1478 __ cmp64(rax, ExternalAddress((address) &is_nan));
1479 __ jcc(Assembler::notEqual, L);
1480 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1481 __ bind(L);
1482 }
1483 break;
1484 case Bytecodes::_f2d:
1485 __ cvtss2sd(xmm0, xmm0);
1486 break;
1487 case Bytecodes::_d2i:
1488 {
1489 Label L;
1490 __ cvttsd2sil(rax, xmm0);
1491 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1492 __ jcc(Assembler::notEqual, L);
1493 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1494 __ bind(L);
1495 }
1496 break;
1497 case Bytecodes::_d2l:
1498 {
1499 Label L;
1500 __ cvttsd2siq(rax, xmm0);
1501 // NaN or overflow/underflow?
1502 __ cmp64(rax, ExternalAddress((address) &is_nan));
1503 __ jcc(Assembler::notEqual, L);
1504 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1505 __ bind(L);
1506 }
1507 break;
1508 case Bytecodes::_d2f:
1509 __ cvtsd2ss(xmm0, xmm0);
1510 break;
1511 default:
1512 ShouldNotReachHere();
1513 }
1514 }
1515
1516 void TemplateTable::lcmp() {
1517 transition(ltos, itos);
1518 Label done;
1519 __ pop_l(rdx);
1520 __ cmpq(rdx, rax);
1521 __ movl(rax, -1);
1522 __ jccb(Assembler::less, done);
1523 __ setb(Assembler::notEqual, rax);
1524 __ movzbl(rax, rax);
1525 __ bind(done);
1526 }
1527
1528 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1529 Label done;
1530 if (is_float) {
1531 // XXX get rid of pop here, use ... reg, mem32
1532 __ pop_f(xmm1);
1533 __ ucomiss(xmm1, xmm0);
1534 } else {
1535 // XXX get rid of pop here, use ... reg, mem64
1536 __ pop_d(xmm1);
1537 __ ucomisd(xmm1, xmm0);
1538 }
1539 if (unordered_result < 0) {
1540 __ movl(rax, -1);
1541 __ jccb(Assembler::parity, done);
1542 __ jccb(Assembler::below, done);
1543 __ setb(Assembler::notEqual, rdx);
1544 __ movzbl(rax, rdx);
1545 } else {
1546 __ movl(rax, 1);
1547 __ jccb(Assembler::parity, done);
1548 __ jccb(Assembler::above, done);
1549 __ movl(rax, 0);
1550 __ jccb(Assembler::equal, done);
1551 __ decrementl(rax);
1552 }
1553 __ bind(done);
1554 }
1555
1556 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1557 __ get_method(rcx); // rcx holds method
1558 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
1559 // holds bumped taken count
1560
1561 const ByteSize be_offset = methodOopDesc::backedge_counter_offset() +
1562 InvocationCounter::counter_offset();
1563 const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() +
1564 InvocationCounter::counter_offset();
1565 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1566
1567 // Load up edx with the branch displacement
1568 __ movl(rdx, at_bcp(1));
1569 __ bswapl(rdx);
1570
1571 if (!is_wide) {
1572 __ sarl(rdx, 16);
1573 }
1574 __ movl2ptr(rdx, rdx);
1575
1576 // Handle all the JSR stuff here, then exit.
1577 // It's much shorter and cleaner than intermingling with the non-JSR
1578 // normal-branch stuff occurring below.
1579 if (is_jsr) {
1580 // Pre-load the next target bytecode into rbx
1581 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1, 0));
1582
1583 // compute return address as bci in rax
1584 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
1585 in_bytes(constMethodOopDesc::codes_offset())));
1586 __ subptr(rax, Address(rcx, methodOopDesc::const_offset()));
1587 // Adjust the bcp in r13 by the displacement in rdx
1588 __ addptr(r13, rdx);
1589 // jsr returns atos that is not an oop
1590 __ push_i(rax);
1591 __ dispatch_only(vtos);
1592 return;
1593 }
1594
1595 // Normal (non-jsr) branch handling
1596
1597 // Adjust the bcp in r13 by the displacement in rdx
1598 __ addptr(r13, rdx);
1599
1600 assert(UseLoopCounter || !UseOnStackReplacement,
1601 "on-stack-replacement requires loop counters");
1602 Label backedge_counter_overflow;
1603 Label profile_method;
1604 Label dispatch;
1605 if (UseLoopCounter) {
1606 // increment backedge counter for backward branches
1607 // rax: MDO
1608 // ebx: MDO bumped taken-count
1609 // rcx: method
1610 // rdx: target offset
1611 // r13: target bcp
1612 // r14: locals pointer
1613 __ testl(rdx, rdx); // check if forward or backward branch
1614 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1615 if (TieredCompilation) {
1616 Label no_mdo;
1617 int increment = InvocationCounter::count_increment;
1618 int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1619 if (ProfileInterpreter) {
1620 // Are we profiling?
1621 __ movptr(rbx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1622 __ testptr(rbx, rbx);
1623 __ jccb(Assembler::zero, no_mdo);
1624 // Increment the MDO backedge counter
1625 const Address mdo_backedge_counter(rbx, in_bytes(methodDataOopDesc::backedge_counter_offset()) +
1626 in_bytes(InvocationCounter::counter_offset()));
1627 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1628 rax, false, Assembler::zero, &backedge_counter_overflow);
1629 __ jmp(dispatch);
1630 }
1631 __ bind(no_mdo);
1632 // Increment backedge counter in methodOop
1633 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
1634 rax, false, Assembler::zero, &backedge_counter_overflow);
1635 } else {
1636 // increment counter
1637 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1638 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1639 __ movl(Address(rcx, be_offset), rax); // store counter
1640
1641 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1642 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1643 __ addl(rax, Address(rcx, be_offset)); // add both counters
1644
1645 if (ProfileInterpreter) {
1646 // Test to see if we should create a method data oop
1647 __ cmp32(rax,
1648 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1649 __ jcc(Assembler::less, dispatch);
1650
1651 // if no method data exists, go to profile method
1652 __ test_method_data_pointer(rax, profile_method);
1653
1654 if (UseOnStackReplacement) {
1655 // check for overflow against ebx which is the MDO taken count
1656 __ cmp32(rbx,
1657 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1658 __ jcc(Assembler::below, dispatch);
1659
1660 // When ProfileInterpreter is on, the backedge_count comes
1661 // from the methodDataOop, which value does not get reset on
1662 // the call to frequency_counter_overflow(). To avoid
1663 // excessive calls to the overflow routine while the method is
1664 // being compiled, add a second test to make sure the overflow
1665 // function is called only once every overflow_frequency.
1666 const int overflow_frequency = 1024;
1667 __ andl(rbx, overflow_frequency - 1);
1668 __ jcc(Assembler::zero, backedge_counter_overflow);
1669
1670 }
1671 } else {
1672 if (UseOnStackReplacement) {
1673 // check for overflow against eax, which is the sum of the
1674 // counters
1675 __ cmp32(rax,
1676 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1677 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1678
1679 }
1680 }
1681 }
1682 __ bind(dispatch);
1683 }
1684
1685 // Pre-load the next target bytecode into rbx
1686 __ load_unsigned_byte(rbx, Address(r13, 0));
1687
1688 // continue with the bytecode @ target
1689 // eax: return bci for jsr's, unused otherwise
1690 // ebx: target bytecode
1691 // r13: target bcp
1692 __ dispatch_only(vtos);
1693
1694 if (UseLoopCounter) {
1695 if (ProfileInterpreter) {
1696 // Out-of-line code to allocate method data oop.
1697 __ bind(profile_method);
1698 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1699 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1700 __ set_method_data_pointer_for_bcp();
1701 __ jmp(dispatch);
1702 }
1703
1704 if (UseOnStackReplacement) {
1705 // invocation counter overflow
1706 __ bind(backedge_counter_overflow);
1707 __ negptr(rdx);
1708 __ addptr(rdx, r13); // branch bcp
1709 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1710 __ call_VM(noreg,
1711 CAST_FROM_FN_PTR(address,
1712 InterpreterRuntime::frequency_counter_overflow),
1713 rdx);
1714 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1715
1716 // rax: osr nmethod (osr ok) or NULL (osr not possible)
1717 // ebx: target bytecode
1718 // rdx: scratch
1719 // r14: locals pointer
1720 // r13: bcp
1721 __ testptr(rax, rax); // test result
1722 __ jcc(Assembler::zero, dispatch); // no osr if null
1723 // nmethod may have been invalidated (VM may block upon call_VM return)
1724 __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1725 __ cmpl(rcx, InvalidOSREntryBci);
1726 __ jcc(Assembler::equal, dispatch);
1727
1728 // We have the address of an on stack replacement routine in eax
1729 // We need to prepare to execute the OSR method. First we must
1730 // migrate the locals and monitors off of the stack.
1731
1732 __ mov(r13, rax); // save the nmethod
1733
1734 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1735
1736 // eax is OSR buffer, move it to expected parameter location
1737 __ mov(j_rarg0, rax);
1738
1739 // We use j_rarg definitions here so that registers don't conflict as parameter
1740 // registers change across platforms as we are in the midst of a calling
1741 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
1742
1743 const Register retaddr = j_rarg2;
1744 const Register sender_sp = j_rarg1;
1745
1746 // pop the interpreter frame
1747 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1748 __ leave(); // remove frame anchor
1749 __ pop(retaddr); // get return address
1750 __ mov(rsp, sender_sp); // set sp to sender sp
1751 // Ensure compiled code always sees stack at proper alignment
1752 __ andptr(rsp, -(StackAlignmentInBytes));
1753
1754 // unlike x86 we need no specialized return from compiled code
1755 // to the interpreter or the call stub.
1756
1757 // push the return address
1758 __ push(retaddr);
1759
1760 // and begin the OSR nmethod
1761 __ jmp(Address(r13, nmethod::osr_entry_point_offset()));
1762 }
1763 }
1764 }
1765
1766
1767 void TemplateTable::if_0cmp(Condition cc) {
1768 transition(itos, vtos);
1769 // assume branch is more often taken than not (loops use backward branches)
1770 Label not_taken;
1771 __ testl(rax, rax);
1772 __ jcc(j_not(cc), not_taken);
1773 branch(false, false);
1774 __ bind(not_taken);
1775 __ profile_not_taken_branch(rax);
1776 }
1777
1778 void TemplateTable::if_icmp(Condition cc) {
1779 transition(itos, vtos);
1780 // assume branch is more often taken than not (loops use backward branches)
1781 Label not_taken;
1782 __ pop_i(rdx);
1783 __ cmpl(rdx, rax);
1784 __ jcc(j_not(cc), not_taken);
1785 branch(false, false);
1786 __ bind(not_taken);
1787 __ profile_not_taken_branch(rax);
1788 }
1789
1790 void TemplateTable::if_nullcmp(Condition cc) {
1791 transition(atos, vtos);
1792 // assume branch is more often taken than not (loops use backward branches)
1793 Label not_taken;
1794 __ testptr(rax, rax);
1795 __ jcc(j_not(cc), not_taken);
1796 branch(false, false);
1797 __ bind(not_taken);
1798 __ profile_not_taken_branch(rax);
1799 }
1800
1801 void TemplateTable::if_acmp(Condition cc) {
1802 transition(atos, vtos);
1803 // assume branch is more often taken than not (loops use backward branches)
1804 Label not_taken;
1805 __ pop_ptr(rdx);
1806 __ cmpptr(rdx, rax);
1807 __ jcc(j_not(cc), not_taken);
1808 branch(false, false);
1809 __ bind(not_taken);
1810 __ profile_not_taken_branch(rax);
1811 }
1812
1813 void TemplateTable::ret() {
1814 transition(vtos, vtos);
1815 locals_index(rbx);
1816 __ movslq(rbx, iaddress(rbx)); // get return bci, compute return bcp
1817 __ profile_ret(rbx, rcx);
1818 __ get_method(rax);
1819 __ movptr(r13, Address(rax, methodOopDesc::const_offset()));
1820 __ lea(r13, Address(r13, rbx, Address::times_1,
1821 constMethodOopDesc::codes_offset()));
1822 __ dispatch_next(vtos);
1823 }
1824
1825 void TemplateTable::wide_ret() {
1826 transition(vtos, vtos);
1827 locals_index_wide(rbx);
1828 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
1829 __ profile_ret(rbx, rcx);
1830 __ get_method(rax);
1831 __ movptr(r13, Address(rax, methodOopDesc::const_offset()));
1832 __ lea(r13, Address(r13, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1833 __ dispatch_next(vtos);
1834 }
1835
1836 void TemplateTable::tableswitch() {
1837 Label default_case, continue_execution;
1838 transition(itos, vtos);
1839 // align r13
1840 __ lea(rbx, at_bcp(BytesPerInt));
1841 __ andptr(rbx, -BytesPerInt);
1842 // load lo & hi
1843 __ movl(rcx, Address(rbx, BytesPerInt));
1844 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
1845 __ bswapl(rcx);
1846 __ bswapl(rdx);
1847 // check against lo & hi
1848 __ cmpl(rax, rcx);
1849 __ jcc(Assembler::less, default_case);
1850 __ cmpl(rax, rdx);
1851 __ jcc(Assembler::greater, default_case);
1852 // lookup dispatch offset
1853 __ subl(rax, rcx);
1854 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1855 __ profile_switch_case(rax, rbx, rcx);
1856 // continue execution
1857 __ bind(continue_execution);
1858 __ bswapl(rdx);
1859 __ movl2ptr(rdx, rdx);
1860 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1861 __ addptr(r13, rdx);
1862 __ dispatch_only(vtos);
1863 // handle default
1864 __ bind(default_case);
1865 __ profile_switch_default(rax);
1866 __ movl(rdx, Address(rbx, 0));
1867 __ jmp(continue_execution);
1868 }
1869
1870 void TemplateTable::lookupswitch() {
1871 transition(itos, itos);
1872 __ stop("lookupswitch bytecode should have been rewritten");
1873 }
1874
1875 void TemplateTable::fast_linearswitch() {
1876 transition(itos, vtos);
1877 Label loop_entry, loop, found, continue_execution;
1878 // bswap rax so we can avoid bswapping the table entries
1879 __ bswapl(rax);
1880 // align r13
1881 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
1882 // this instruction (change offsets
1883 // below)
1884 __ andptr(rbx, -BytesPerInt);
1885 // set counter
1886 __ movl(rcx, Address(rbx, BytesPerInt));
1887 __ bswapl(rcx);
1888 __ jmpb(loop_entry);
1889 // table search
1890 __ bind(loop);
1891 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
1892 __ jcc(Assembler::equal, found);
1893 __ bind(loop_entry);
1894 __ decrementl(rcx);
1895 __ jcc(Assembler::greaterEqual, loop);
1896 // default case
1897 __ profile_switch_default(rax);
1898 __ movl(rdx, Address(rbx, 0));
1899 __ jmp(continue_execution);
1900 // entry found -> get offset
1901 __ bind(found);
1902 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
1903 __ profile_switch_case(rcx, rax, rbx);
1904 // continue execution
1905 __ bind(continue_execution);
1906 __ bswapl(rdx);
1907 __ movl2ptr(rdx, rdx);
1908 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1909 __ addptr(r13, rdx);
1910 __ dispatch_only(vtos);
1911 }
1912
1913 void TemplateTable::fast_binaryswitch() {
1914 transition(itos, vtos);
1915 // Implementation using the following core algorithm:
1916 //
1917 // int binary_search(int key, LookupswitchPair* array, int n) {
1918 // // Binary search according to "Methodik des Programmierens" by
1919 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1920 // int i = 0;
1921 // int j = n;
1922 // while (i+1 < j) {
1923 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1924 // // with Q: for all i: 0 <= i < n: key < a[i]
1925 // // where a stands for the array and assuming that the (inexisting)
1926 // // element a[n] is infinitely big.
1927 // int h = (i + j) >> 1;
1928 // // i < h < j
1929 // if (key < array[h].fast_match()) {
1930 // j = h;
1931 // } else {
1932 // i = h;
1933 // }
1934 // }
1935 // // R: a[i] <= key < a[i+1] or Q
1936 // // (i.e., if key is within array, i is the correct index)
1937 // return i;
1938 // }
1939
1940 // Register allocation
1941 const Register key = rax; // already set (tosca)
1942 const Register array = rbx;
1943 const Register i = rcx;
1944 const Register j = rdx;
1945 const Register h = rdi;
1946 const Register temp = rsi;
1947
1948 // Find array start
1949 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
1950 // get rid of this
1951 // instruction (change
1952 // offsets below)
1953 __ andptr(array, -BytesPerInt);
1954
1955 // Initialize i & j
1956 __ xorl(i, i); // i = 0;
1957 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
1958
1959 // Convert j into native byteordering
1960 __ bswapl(j);
1961
1962 // And start
1963 Label entry;
1964 __ jmp(entry);
1965
1966 // binary search loop
1967 {
1968 Label loop;
1969 __ bind(loop);
1970 // int h = (i + j) >> 1;
1971 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1972 __ sarl(h, 1); // h = (i + j) >> 1;
1973 // if (key < array[h].fast_match()) {
1974 // j = h;
1975 // } else {
1976 // i = h;
1977 // }
1978 // Convert array[h].match to native byte-ordering before compare
1979 __ movl(temp, Address(array, h, Address::times_8));
1980 __ bswapl(temp);
1981 __ cmpl(key, temp);
1982 // j = h if (key < array[h].fast_match())
1983 __ cmovl(Assembler::less, j, h);
1984 // i = h if (key >= array[h].fast_match())
1985 __ cmovl(Assembler::greaterEqual, i, h);
1986 // while (i+1 < j)
1987 __ bind(entry);
1988 __ leal(h, Address(i, 1)); // i+1
1989 __ cmpl(h, j); // i+1 < j
1990 __ jcc(Assembler::less, loop);
1991 }
1992
1993 // end of binary search, result index is i (must check again!)
1994 Label default_case;
1995 // Convert array[i].match to native byte-ordering before compare
1996 __ movl(temp, Address(array, i, Address::times_8));
1997 __ bswapl(temp);
1998 __ cmpl(key, temp);
1999 __ jcc(Assembler::notEqual, default_case);
2000
2001 // entry found -> j = offset
2002 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2003 __ profile_switch_case(i, key, array);
2004 __ bswapl(j);
2005 __ movl2ptr(j, j);
2006 __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
2007 __ addptr(r13, j);
2008 __ dispatch_only(vtos);
2009
2010 // default case -> j = default offset
2011 __ bind(default_case);
2012 __ profile_switch_default(i);
2013 __ movl(j, Address(array, -2 * BytesPerInt));
2014 __ bswapl(j);
2015 __ movl2ptr(j, j);
2016 __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
2017 __ addptr(r13, j);
2018 __ dispatch_only(vtos);
2019 }
2020
2021
2022 void TemplateTable::_return(TosState state) {
2023 transition(state, state);
2024 assert(_desc->calls_vm(),
2025 "inconsistent calls_vm information"); // call in remove_activation
2026
2027 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2028 assert(state == vtos, "only valid state");
2029 __ movptr(c_rarg1, aaddress(0));
2030 __ load_klass(rdi, c_rarg1);
2031 __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
2032 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2033 Label skip_register_finalizer;
2034 __ jcc(Assembler::zero, skip_register_finalizer);
2035
2036 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2037
2038 __ bind(skip_register_finalizer);
2039 }
2040
2041 __ remove_activation(state, r13);
2042 __ jmp(r13);
2043 }
2044
2045 // ----------------------------------------------------------------------------
2046 // Volatile variables demand their effects be made known to all CPU's
2047 // in order. Store buffers on most chips allow reads & writes to
2048 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2049 // without some kind of memory barrier (i.e., it's not sufficient that
2050 // the interpreter does not reorder volatile references, the hardware
2051 // also must not reorder them).
2052 //
2053 // According to the new Java Memory Model (JMM):
2054 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2055 // writes act as aquire & release, so:
2056 // (2) A read cannot let unrelated NON-volatile memory refs that
2057 // happen after the read float up to before the read. It's OK for
2058 // non-volatile memory refs that happen before the volatile read to
2059 // float down below it.
2060 // (3) Similar a volatile write cannot let unrelated NON-volatile
2061 // memory refs that happen BEFORE the write float down to after the
2062 // write. It's OK for non-volatile memory refs that happen after the
2063 // volatile write to float up before it.
2064 //
2065 // We only put in barriers around volatile refs (they are expensive),
2066 // not _between_ memory refs (that would require us to track the
2067 // flavor of the previous memory refs). Requirements (2) and (3)
2068 // require some barriers before volatile stores and after volatile
2069 // loads. These nearly cover requirement (1) but miss the
2070 // volatile-store-volatile-load case. This final case is placed after
2071 // volatile-stores although it could just as well go before
2072 // volatile-loads.
2073 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits
2074 order_constraint) {
2075 // Helper function to insert a is-volatile test and memory barrier
2076 if (os::is_MP()) { // Not needed on single CPU
2077 __ membar(order_constraint);
2078 }
2079 }
2080
2081 void TemplateTable::resolve_cache_and_index(int byte_no,
2082 Register result,
2083 Register Rcache,
2084 Register index,
2085 size_t index_size) {
2086 const Register temp = rbx;
2087 assert_different_registers(result, Rcache, index, temp);
2088
2089 Label resolved;
2090 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2091 if (byte_no == f1_oop) {
2092 // We are resolved if the f1 field contains a non-null object (CallSite, etc.)
2093 // This kind of CP cache entry does not need to match the flags byte, because
2094 // there is a 1-1 relation between bytecode type and CP entry type.
2095 assert(result != noreg, ""); //else do cmpptr(Address(...), (int32_t) NULL_WORD)
2096 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2097 __ testptr(result, result);
2098 __ jcc(Assembler::notEqual, resolved);
2099 } else {
2100 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2101 assert(result == noreg, ""); //else change code for setting result
2102 const int shift_count = (1 + byte_no) * BitsPerByte;
2103 __ movl(temp, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
2104 __ shrl(temp, shift_count);
2105 // have we resolved this bytecode?
2106 __ andl(temp, 0xFF);
2107 __ cmpl(temp, (int) bytecode());
2108 __ jcc(Assembler::equal, resolved);
2109 }
2110
2111 // resolve first time through
2112 address entry;
2113 switch (bytecode()) {
2114 case Bytecodes::_getstatic:
2115 case Bytecodes::_putstatic:
2116 case Bytecodes::_getfield:
2117 case Bytecodes::_putfield:
2118 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
2119 break;
2120 case Bytecodes::_invokevirtual:
2121 case Bytecodes::_invokespecial:
2122 case Bytecodes::_invokestatic:
2123 case Bytecodes::_invokeinterface:
2124 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
2125 break;
2126 case Bytecodes::_invokedynamic:
2127 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
2128 break;
2129 case Bytecodes::_fast_aldc:
2130 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
2131 break;
2132 case Bytecodes::_fast_aldc_w:
2133 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
2134 break;
2135 default:
2136 ShouldNotReachHere();
2137 break;
2138 }
2139 __ movl(temp, (int) bytecode());
2140 __ call_VM(noreg, entry, temp);
2141
2142 // Update registers with resolved info
2143 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2144 if (result != noreg)
2145 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2146 __ bind(resolved);
2147 }
2148
2149 // The Rcache and index registers must be set before call
2150 void TemplateTable::load_field_cp_cache_entry(Register obj,
2151 Register cache,
2152 Register index,
2153 Register off,
2154 Register flags,
2155 bool is_static = false) {
2156 assert_different_registers(cache, index, flags, off);
2157
2158 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2159 // Field offset
2160 __ movptr(off, Address(cache, index, Address::times_8,
2161 in_bytes(cp_base_offset +
2162 ConstantPoolCacheEntry::f2_offset())));
2163 // Flags
2164 __ movl(flags, Address(cache, index, Address::times_8,
2165 in_bytes(cp_base_offset +
2166 ConstantPoolCacheEntry::flags_offset())));
2167
2168 // klass overwrite register
2169 if (is_static) {
2170 __ movptr(obj, Address(cache, index, Address::times_8,
2171 in_bytes(cp_base_offset +
2172 ConstantPoolCacheEntry::f1_offset())));
2173 }
2174 }
2175
2176 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2177 Register method,
2178 Register itable_index,
2179 Register flags,
2180 bool is_invokevirtual,
2181 bool is_invokevfinal, /*unused*/
2182 bool is_invokedynamic) {
2183 // setup registers
2184 const Register cache = rcx;
2185 const Register index = rdx;
2186 assert_different_registers(method, flags);
2187 assert_different_registers(method, cache, index);
2188 assert_different_registers(itable_index, flags);
2189 assert_different_registers(itable_index, cache, index);
2190 // determine constant pool cache field offsets
2191 const int method_offset = in_bytes(
2192 constantPoolCacheOopDesc::base_offset() +
2193 (is_invokevirtual
2194 ? ConstantPoolCacheEntry::f2_offset()
2195 : ConstantPoolCacheEntry::f1_offset()));
2196 const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2197 ConstantPoolCacheEntry::flags_offset());
2198 // access constant pool cache fields
2199 const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2200 ConstantPoolCacheEntry::f2_offset());
2201
2202 if (byte_no == f1_oop) {
2203 // Resolved f1_oop goes directly into 'method' register.
2204 assert(is_invokedynamic, "");
2205 resolve_cache_and_index(byte_no, method, cache, index, sizeof(u4));
2206 } else {
2207 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2208 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2209 }
2210 if (itable_index != noreg) {
2211 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2212 }
2213 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2214 }
2215
2216
2217 // The registers cache and index expected to be set before call.
2218 // Correct values of the cache and index registers are preserved.
2219 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2220 bool is_static, bool has_tos) {
2221 // do the JVMTI work here to avoid disturbing the register state below
2222 // We use c_rarg registers here because we want to use the register used in
2223 // the call to the VM
2224 if (JvmtiExport::can_post_field_access()) {
2225 // Check to see if a field access watch has been set before we
2226 // take the time to call into the VM.
2227 Label L1;
2228 assert_different_registers(cache, index, rax);
2229 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2230 __ testl(rax, rax);
2231 __ jcc(Assembler::zero, L1);
2232
2233 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2234
2235 // cache entry pointer
2236 __ addptr(c_rarg2, in_bytes(constantPoolCacheOopDesc::base_offset()));
2237 __ shll(c_rarg3, LogBytesPerWord);
2238 __ addptr(c_rarg2, c_rarg3);
2239 if (is_static) {
2240 __ xorl(c_rarg1, c_rarg1); // NULL object reference
2241 } else {
2242 __ movptr(c_rarg1, at_tos()); // get object pointer without popping it
2243 __ verify_oop(c_rarg1);
2244 }
2245 // c_rarg1: object pointer or NULL
2246 // c_rarg2: cache entry pointer
2247 // c_rarg3: jvalue object on the stack
2248 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2249 InterpreterRuntime::post_field_access),
2250 c_rarg1, c_rarg2, c_rarg3);
2251 __ get_cache_and_index_at_bcp(cache, index, 1);
2252 __ bind(L1);
2253 }
2254 }
2255
2256 void TemplateTable::pop_and_check_object(Register r) {
2257 __ pop_ptr(r);
2258 __ null_check(r); // for field access must check obj.
2259 __ verify_oop(r);
2260 }
2261
2262 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2263 transition(vtos, vtos);
2264
2265 const Register cache = rcx;
2266 const Register index = rdx;
2267 const Register obj = c_rarg3;
2268 const Register off = rbx;
2269 const Register flags = rax;
2270 const Register bc = c_rarg3; // uses same reg as obj, so don't mix them
2271
2272 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2273 jvmti_post_field_access(cache, index, is_static, false);
2274 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2275
2276 if (!is_static) {
2277 // obj is on the stack
2278 pop_and_check_object(obj);
2279 }
2280
2281 const Address field(obj, off, Address::times_1);
2282
2283 Label Done, notByte, notInt, notShort, notChar,
2284 notLong, notFloat, notObj, notDouble;
2285
2286 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2287 assert(btos == 0, "change code, btos != 0");
2288
2289 __ andl(flags, 0x0F);
2290 __ jcc(Assembler::notZero, notByte);
2291 // btos
2292 __ load_signed_byte(rax, field);
2293 __ push(btos);
2294 // Rewrite bytecode to be faster
2295 if (!is_static) {
2296 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2297 }
2298 __ jmp(Done);
2299
2300 __ bind(notByte);
2301 __ cmpl(flags, atos);
2302 __ jcc(Assembler::notEqual, notObj);
2303 // atos
2304 __ load_heap_oop(rax, field);
2305 __ push(atos);
2306 if (!is_static) {
2307 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2308 }
2309 __ jmp(Done);
2310
2311 __ bind(notObj);
2312 __ cmpl(flags, itos);
2313 __ jcc(Assembler::notEqual, notInt);
2314 // itos
2315 __ movl(rax, field);
2316 __ push(itos);
2317 // Rewrite bytecode to be faster
2318 if (!is_static) {
2319 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2320 }
2321 __ jmp(Done);
2322
2323 __ bind(notInt);
2324 __ cmpl(flags, ctos);
2325 __ jcc(Assembler::notEqual, notChar);
2326 // ctos
2327 __ load_unsigned_short(rax, field);
2328 __ push(ctos);
2329 // Rewrite bytecode to be faster
2330 if (!is_static) {
2331 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2332 }
2333 __ jmp(Done);
2334
2335 __ bind(notChar);
2336 __ cmpl(flags, stos);
2337 __ jcc(Assembler::notEqual, notShort);
2338 // stos
2339 __ load_signed_short(rax, field);
2340 __ push(stos);
2341 // Rewrite bytecode to be faster
2342 if (!is_static) {
2343 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2344 }
2345 __ jmp(Done);
2346
2347 __ bind(notShort);
2348 __ cmpl(flags, ltos);
2349 __ jcc(Assembler::notEqual, notLong);
2350 // ltos
2351 __ movq(rax, field);
2352 __ push(ltos);
2353 // Rewrite bytecode to be faster
2354 if (!is_static) {
2355 patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx);
2356 }
2357 __ jmp(Done);
2358
2359 __ bind(notLong);
2360 __ cmpl(flags, ftos);
2361 __ jcc(Assembler::notEqual, notFloat);
2362 // ftos
2363 __ movflt(xmm0, field);
2364 __ push(ftos);
2365 // Rewrite bytecode to be faster
2366 if (!is_static) {
2367 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2368 }
2369 __ jmp(Done);
2370
2371 __ bind(notFloat);
2372 #ifdef ASSERT
2373 __ cmpl(flags, dtos);
2374 __ jcc(Assembler::notEqual, notDouble);
2375 #endif
2376 // dtos
2377 __ movdbl(xmm0, field);
2378 __ push(dtos);
2379 // Rewrite bytecode to be faster
2380 if (!is_static) {
2381 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2382 }
2383 #ifdef ASSERT
2384 __ jmp(Done);
2385
2386 __ bind(notDouble);
2387 __ stop("Bad state");
2388 #endif
2389
2390 __ bind(Done);
2391 // [jk] not needed currently
2392 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2393 // Assembler::LoadStore));
2394 }
2395
2396
2397 void TemplateTable::getfield(int byte_no) {
2398 getfield_or_static(byte_no, false);
2399 }
2400
2401 void TemplateTable::getstatic(int byte_no) {
2402 getfield_or_static(byte_no, true);
2403 }
2404
2405 // The registers cache and index expected to be set before call.
2406 // The function may destroy various registers, just not the cache and index registers.
2407 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2408 transition(vtos, vtos);
2409
2410 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2411
2412 if (JvmtiExport::can_post_field_modification()) {
2413 // Check to see if a field modification watch has been set before
2414 // we take the time to call into the VM.
2415 Label L1;
2416 assert_different_registers(cache, index, rax);
2417 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2418 __ testl(rax, rax);
2419 __ jcc(Assembler::zero, L1);
2420
2421 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2422
2423 if (is_static) {
2424 // Life is simple. Null out the object pointer.
2425 __ xorl(c_rarg1, c_rarg1);
2426 } else {
2427 // Life is harder. The stack holds the value on top, followed by
2428 // the object. We don't know the size of the value, though; it
2429 // could be one or two words depending on its type. As a result,
2430 // we must find the type to determine where the object is.
2431 __ movl(c_rarg3, Address(c_rarg2, rscratch1,
2432 Address::times_8,
2433 in_bytes(cp_base_offset +
2434 ConstantPoolCacheEntry::flags_offset())));
2435 __ shrl(c_rarg3, ConstantPoolCacheEntry::tosBits);
2436 // Make sure we don't need to mask rcx for tosBits after the
2437 // above shift
2438 ConstantPoolCacheEntry::verify_tosBits();
2439 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2440 __ cmpl(c_rarg3, ltos);
2441 __ cmovptr(Assembler::equal,
2442 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2443 __ cmpl(c_rarg3, dtos);
2444 __ cmovptr(Assembler::equal,
2445 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2446 }
2447 // cache entry pointer
2448 __ addptr(c_rarg2, in_bytes(cp_base_offset));
2449 __ shll(rscratch1, LogBytesPerWord);
2450 __ addptr(c_rarg2, rscratch1);
2451 // object (tos)
2452 __ mov(c_rarg3, rsp);
2453 // c_rarg1: object pointer set up above (NULL if static)
2454 // c_rarg2: cache entry pointer
2455 // c_rarg3: jvalue object on the stack
2456 __ call_VM(noreg,
2457 CAST_FROM_FN_PTR(address,
2458 InterpreterRuntime::post_field_modification),
2459 c_rarg1, c_rarg2, c_rarg3);
2460 __ get_cache_and_index_at_bcp(cache, index, 1);
2461 __ bind(L1);
2462 }
2463 }
2464
2465 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2466 transition(vtos, vtos);
2467
2468 const Register cache = rcx;
2469 const Register index = rdx;
2470 const Register obj = rcx;
2471 const Register off = rbx;
2472 const Register flags = rax;
2473 const Register bc = c_rarg3;
2474
2475 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2476 jvmti_post_field_mod(cache, index, is_static);
2477 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2478
2479 // [jk] not needed currently
2480 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2481 // Assembler::StoreStore));
2482
2483 Label notVolatile, Done;
2484 __ movl(rdx, flags);
2485 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2486 __ andl(rdx, 0x1);
2487
2488 // field address
2489 const Address field(obj, off, Address::times_1);
2490
2491 Label notByte, notInt, notShort, notChar,
2492 notLong, notFloat, notObj, notDouble;
2493
2494 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2495
2496 assert(btos == 0, "change code, btos != 0");
2497 __ andl(flags, 0x0f);
2498 __ jcc(Assembler::notZero, notByte);
2499 // btos
2500 __ pop(btos);
2501 if (!is_static) pop_and_check_object(obj);
2502 __ movb(field, rax);
2503 if (!is_static) {
2504 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx);
2505 }
2506 __ jmp(Done);
2507
2508 __ bind(notByte);
2509 __ cmpl(flags, atos);
2510 __ jcc(Assembler::notEqual, notObj);
2511 // atos
2512 __ pop(atos);
2513 if (!is_static) pop_and_check_object(obj);
2514
2515 // Store into the field
2516 do_oop_store(_masm, field, rax, _bs->kind(), false);
2517
2518 if (!is_static) {
2519 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx);
2520 }
2521 __ jmp(Done);
2522
2523 __ bind(notObj);
2524 __ cmpl(flags, itos);
2525 __ jcc(Assembler::notEqual, notInt);
2526 // itos
2527 __ pop(itos);
2528 if (!is_static) pop_and_check_object(obj);
2529 __ movl(field, rax);
2530 if (!is_static) {
2531 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx);
2532 }
2533 __ jmp(Done);
2534
2535 __ bind(notInt);
2536 __ cmpl(flags, ctos);
2537 __ jcc(Assembler::notEqual, notChar);
2538 // ctos
2539 __ pop(ctos);
2540 if (!is_static) pop_and_check_object(obj);
2541 __ movw(field, rax);
2542 if (!is_static) {
2543 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx);
2544 }
2545 __ jmp(Done);
2546
2547 __ bind(notChar);
2548 __ cmpl(flags, stos);
2549 __ jcc(Assembler::notEqual, notShort);
2550 // stos
2551 __ pop(stos);
2552 if (!is_static) pop_and_check_object(obj);
2553 __ movw(field, rax);
2554 if (!is_static) {
2555 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx);
2556 }
2557 __ jmp(Done);
2558
2559 __ bind(notShort);
2560 __ cmpl(flags, ltos);
2561 __ jcc(Assembler::notEqual, notLong);
2562 // ltos
2563 __ pop(ltos);
2564 if (!is_static) pop_and_check_object(obj);
2565 __ movq(field, rax);
2566 if (!is_static) {
2567 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx);
2568 }
2569 __ jmp(Done);
2570
2571 __ bind(notLong);
2572 __ cmpl(flags, ftos);
2573 __ jcc(Assembler::notEqual, notFloat);
2574 // ftos
2575 __ pop(ftos);
2576 if (!is_static) pop_and_check_object(obj);
2577 __ movflt(field, xmm0);
2578 if (!is_static) {
2579 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx);
2580 }
2581 __ jmp(Done);
2582
2583 __ bind(notFloat);
2584 #ifdef ASSERT
2585 __ cmpl(flags, dtos);
2586 __ jcc(Assembler::notEqual, notDouble);
2587 #endif
2588 // dtos
2589 __ pop(dtos);
2590 if (!is_static) pop_and_check_object(obj);
2591 __ movdbl(field, xmm0);
2592 if (!is_static) {
2593 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx);
2594 }
2595
2596 #ifdef ASSERT
2597 __ jmp(Done);
2598
2599 __ bind(notDouble);
2600 __ stop("Bad state");
2601 #endif
2602
2603 __ bind(Done);
2604 // Check for volatile store
2605 __ testl(rdx, rdx);
2606 __ jcc(Assembler::zero, notVolatile);
2607 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2608 Assembler::StoreStore));
2609
2610 __ bind(notVolatile);
2611 }
2612
2613 void TemplateTable::putfield(int byte_no) {
2614 putfield_or_static(byte_no, false);
2615 }
2616
2617 void TemplateTable::putstatic(int byte_no) {
2618 putfield_or_static(byte_no, true);
2619 }
2620
2621 void TemplateTable::jvmti_post_fast_field_mod() {
2622 if (JvmtiExport::can_post_field_modification()) {
2623 // Check to see if a field modification watch has been set before
2624 // we take the time to call into the VM.
2625 Label L2;
2626 __ mov32(c_rarg3, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2627 __ testl(c_rarg3, c_rarg3);
2628 __ jcc(Assembler::zero, L2);
2629 __ pop_ptr(rbx); // copy the object pointer from tos
2630 __ verify_oop(rbx);
2631 __ push_ptr(rbx); // put the object pointer back on tos
2632 __ subptr(rsp, sizeof(jvalue)); // add space for a jvalue object
2633 __ mov(c_rarg3, rsp);
2634 const Address field(c_rarg3, 0);
2635
2636 switch (bytecode()) { // load values into the jvalue object
2637 case Bytecodes::_fast_aputfield: __ movq(field, rax); break;
2638 case Bytecodes::_fast_lputfield: __ movq(field, rax); break;
2639 case Bytecodes::_fast_iputfield: __ movl(field, rax); break;
2640 case Bytecodes::_fast_bputfield: __ movb(field, rax); break;
2641 case Bytecodes::_fast_sputfield: // fall through
2642 case Bytecodes::_fast_cputfield: __ movw(field, rax); break;
2643 case Bytecodes::_fast_fputfield: __ movflt(field, xmm0); break;
2644 case Bytecodes::_fast_dputfield: __ movdbl(field, xmm0); break;
2645 default:
2646 ShouldNotReachHere();
2647 }
2648
2649 // Save rax because call_VM() will clobber it, then use it for
2650 // JVMTI purposes
2651 __ push(rax);
2652 // access constant pool cache entry
2653 __ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1);
2654 __ verify_oop(rbx);
2655 // rbx: object pointer copied above
2656 // c_rarg2: cache entry pointer
2657 // c_rarg3: jvalue object on the stack
2658 __ call_VM(noreg,
2659 CAST_FROM_FN_PTR(address,
2660 InterpreterRuntime::post_field_modification),
2661 rbx, c_rarg2, c_rarg3);
2662 __ pop(rax); // restore lower value
2663 __ addptr(rsp, sizeof(jvalue)); // release jvalue object space
2664 __ bind(L2);
2665 }
2666 }
2667
2668 void TemplateTable::fast_storefield(TosState state) {
2669 transition(state, vtos);
2670
2671 ByteSize base = constantPoolCacheOopDesc::base_offset();
2672
2673 jvmti_post_fast_field_mod();
2674
2675 // access constant pool cache
2676 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2677
2678 // test for volatile with rdx
2679 __ movl(rdx, Address(rcx, rbx, Address::times_8,
2680 in_bytes(base +
2681 ConstantPoolCacheEntry::flags_offset())));
2682
2683 // replace index with field offset from cache entry
2684 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2685 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2686
2687 // [jk] not needed currently
2688 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2689 // Assembler::StoreStore));
2690
2691 Label notVolatile;
2692 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2693 __ andl(rdx, 0x1);
2694
2695 // Get object from stack
2696 pop_and_check_object(rcx);
2697
2698 // field address
2699 const Address field(rcx, rbx, Address::times_1);
2700
2701 // access field
2702 switch (bytecode()) {
2703 case Bytecodes::_fast_aputfield:
2704 do_oop_store(_masm, field, rax, _bs->kind(), false);
2705 break;
2706 case Bytecodes::_fast_lputfield:
2707 __ movq(field, rax);
2708 break;
2709 case Bytecodes::_fast_iputfield:
2710 __ movl(field, rax);
2711 break;
2712 case Bytecodes::_fast_bputfield:
2713 __ movb(field, rax);
2714 break;
2715 case Bytecodes::_fast_sputfield:
2716 // fall through
2717 case Bytecodes::_fast_cputfield:
2718 __ movw(field, rax);
2719 break;
2720 case Bytecodes::_fast_fputfield:
2721 __ movflt(field, xmm0);
2722 break;
2723 case Bytecodes::_fast_dputfield:
2724 __ movdbl(field, xmm0);
2725 break;
2726 default:
2727 ShouldNotReachHere();
2728 }
2729
2730 // Check for volatile store
2731 __ testl(rdx, rdx);
2732 __ jcc(Assembler::zero, notVolatile);
2733 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2734 Assembler::StoreStore));
2735 __ bind(notVolatile);
2736 }
2737
2738
2739 void TemplateTable::fast_accessfield(TosState state) {
2740 transition(atos, state);
2741
2742 // Do the JVMTI work here to avoid disturbing the register state below
2743 if (JvmtiExport::can_post_field_access()) {
2744 // Check to see if a field access watch has been set before we
2745 // take the time to call into the VM.
2746 Label L1;
2747 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2748 __ testl(rcx, rcx);
2749 __ jcc(Assembler::zero, L1);
2750 // access constant pool cache entry
2751 __ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1);
2752 __ verify_oop(rax);
2753 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
2754 __ mov(c_rarg1, rax);
2755 // c_rarg1: object pointer copied above
2756 // c_rarg2: cache entry pointer
2757 __ call_VM(noreg,
2758 CAST_FROM_FN_PTR(address,
2759 InterpreterRuntime::post_field_access),
2760 c_rarg1, c_rarg2);
2761 __ pop_ptr(rax); // restore object pointer
2762 __ bind(L1);
2763 }
2764
2765 // access constant pool cache
2766 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2767 // replace index with field offset from cache entry
2768 // [jk] not needed currently
2769 // if (os::is_MP()) {
2770 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
2771 // in_bytes(constantPoolCacheOopDesc::base_offset() +
2772 // ConstantPoolCacheEntry::flags_offset())));
2773 // __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2774 // __ andl(rdx, 0x1);
2775 // }
2776 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2777 in_bytes(constantPoolCacheOopDesc::base_offset() +
2778 ConstantPoolCacheEntry::f2_offset())));
2779
2780 // rax: object
2781 __ verify_oop(rax);
2782 __ null_check(rax);
2783 Address field(rax, rbx, Address::times_1);
2784
2785 // access field
2786 switch (bytecode()) {
2787 case Bytecodes::_fast_agetfield:
2788 __ load_heap_oop(rax, field);
2789 __ verify_oop(rax);
2790 break;
2791 case Bytecodes::_fast_lgetfield:
2792 __ movq(rax, field);
2793 break;
2794 case Bytecodes::_fast_igetfield:
2795 __ movl(rax, field);
2796 break;
2797 case Bytecodes::_fast_bgetfield:
2798 __ movsbl(rax, field);
2799 break;
2800 case Bytecodes::_fast_sgetfield:
2801 __ load_signed_short(rax, field);
2802 break;
2803 case Bytecodes::_fast_cgetfield:
2804 __ load_unsigned_short(rax, field);
2805 break;
2806 case Bytecodes::_fast_fgetfield:
2807 __ movflt(xmm0, field);
2808 break;
2809 case Bytecodes::_fast_dgetfield:
2810 __ movdbl(xmm0, field);
2811 break;
2812 default:
2813 ShouldNotReachHere();
2814 }
2815 // [jk] not needed currently
2816 // if (os::is_MP()) {
2817 // Label notVolatile;
2818 // __ testl(rdx, rdx);
2819 // __ jcc(Assembler::zero, notVolatile);
2820 // __ membar(Assembler::LoadLoad);
2821 // __ bind(notVolatile);
2822 //};
2823 }
2824
2825 void TemplateTable::fast_xaccess(TosState state) {
2826 transition(vtos, state);
2827
2828 // get receiver
2829 __ movptr(rax, aaddress(0));
2830 // access constant pool cache
2831 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2832 __ movptr(rbx,
2833 Address(rcx, rdx, Address::times_8,
2834 in_bytes(constantPoolCacheOopDesc::base_offset() +
2835 ConstantPoolCacheEntry::f2_offset())));
2836 // make sure exception is reported in correct bcp range (getfield is
2837 // next instruction)
2838 __ increment(r13);
2839 __ null_check(rax);
2840 switch (state) {
2841 case itos:
2842 __ movl(rax, Address(rax, rbx, Address::times_1));
2843 break;
2844 case atos:
2845 __ load_heap_oop(rax, Address(rax, rbx, Address::times_1));
2846 __ verify_oop(rax);
2847 break;
2848 case ftos:
2849 __ movflt(xmm0, Address(rax, rbx, Address::times_1));
2850 break;
2851 default:
2852 ShouldNotReachHere();
2853 }
2854
2855 // [jk] not needed currently
2856 // if (os::is_MP()) {
2857 // Label notVolatile;
2858 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
2859 // in_bytes(constantPoolCacheOopDesc::base_offset() +
2860 // ConstantPoolCacheEntry::flags_offset())));
2861 // __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2862 // __ testl(rdx, 0x1);
2863 // __ jcc(Assembler::zero, notVolatile);
2864 // __ membar(Assembler::LoadLoad);
2865 // __ bind(notVolatile);
2866 // }
2867
2868 __ decrement(r13);
2869 }
2870
2871
2872
2873 //-----------------------------------------------------------------------------
2874 // Calls
2875
2876 void TemplateTable::count_calls(Register method, Register temp) {
2877 // implemented elsewhere
2878 ShouldNotReachHere();
2879 }
2880
2881 void TemplateTable::prepare_invoke(Register method, Register index, int byte_no) {
2882 // determine flags
2883 Bytecodes::Code code = bytecode();
2884 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2885 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
2886 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2887 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2888 const bool load_receiver = (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic);
2889 const bool receiver_null_check = is_invokespecial;
2890 const bool save_flags = is_invokeinterface || is_invokevirtual;
2891 // setup registers & access constant pool cache
2892 const Register recv = rcx;
2893 const Register flags = rdx;
2894 assert_different_registers(method, index, recv, flags);
2895
2896 // save 'interpreter return address'
2897 __ save_bcp();
2898
2899 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2900
2901 // load receiver if needed (note: no return address pushed yet)
2902 if (load_receiver) {
2903 assert(!is_invokedynamic, "");
2904 __ movl(recv, flags);
2905 __ andl(recv, 0xFF);
2906 Address recv_addr(rsp, recv, Address::times_8, -Interpreter::expr_offset_in_bytes(1));
2907 __ movptr(recv, recv_addr);
2908 __ verify_oop(recv);
2909 }
2910
2911 // do null check if needed
2912 if (receiver_null_check) {
2913 __ null_check(recv);
2914 }
2915
2916 if (save_flags) {
2917 __ movl(r13, flags);
2918 }
2919
2920 // compute return type
2921 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2922 // Make sure we don't need to mask flags for tosBits after the above shift
2923 ConstantPoolCacheEntry::verify_tosBits();
2924 // load return address
2925 {
2926 address table_addr;
2927 if (is_invokeinterface || is_invokedynamic)
2928 table_addr = (address)Interpreter::return_5_addrs_by_index_table();
2929 else
2930 table_addr = (address)Interpreter::return_3_addrs_by_index_table();
2931 ExternalAddress table(table_addr);
2932 __ lea(rscratch1, table);
2933 __ movptr(flags, Address(rscratch1, flags, Address::times_ptr));
2934 }
2935
2936 // push return address
2937 __ push(flags);
2938
2939 // Restore flag field from the constant pool cache, and restore esi
2940 // for later null checks. r13 is the bytecode pointer
2941 if (save_flags) {
2942 __ movl(flags, r13);
2943 __ restore_bcp();
2944 }
2945 }
2946
2947
2948 void TemplateTable::invokevirtual_helper(Register index,
2949 Register recv,
2950 Register flags) {
2951 // Uses temporary registers rax, rdx
2952 assert_different_registers(index, recv, rax, rdx);
2953
2954 // Test for an invoke of a final method
2955 Label notFinal;
2956 __ movl(rax, flags);
2957 __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
2958 __ jcc(Assembler::zero, notFinal);
2959
2960 const Register method = index; // method must be rbx
2961 assert(method == rbx,
2962 "methodOop must be rbx for interpreter calling convention");
2963
2964 // do the call - the index is actually the method to call
2965 __ verify_oop(method);
2966
2967 // It's final, need a null check here!
2968 __ null_check(recv);
2969
2970 // profile this call
2971 __ profile_final_call(rax);
2972
2973 __ jump_from_interpreted(method, rax);
2974
2975 __ bind(notFinal);
2976
2977 // get receiver klass
2978 __ null_check(recv, oopDesc::klass_offset_in_bytes());
2979 __ load_klass(rax, recv);
2980
2981 __ verify_oop(rax);
2982
2983 // profile this call
2984 __ profile_virtual_call(rax, r14, rdx);
2985
2986 // get target methodOop & entry point
2987 const int base = instanceKlass::vtable_start_offset() * wordSize;
2988 assert(vtableEntry::size() * wordSize == 8,
2989 "adjust the scaling in the code below");
2990 __ movptr(method, Address(rax, index,
2991 Address::times_8,
2992 base + vtableEntry::method_offset_in_bytes()));
2993 __ movptr(rdx, Address(method, methodOopDesc::interpreter_entry_offset()));
2994 __ jump_from_interpreted(method, rdx);
2995 }
2996
2997
2998 void TemplateTable::invokevirtual(int byte_no) {
2999 transition(vtos, vtos);
3000 assert(byte_no == f2_byte, "use this argument");
3001 prepare_invoke(rbx, noreg, byte_no);
3002
3003 // rbx: index
3004 // rcx: receiver
3005 // rdx: flags
3006
3007 invokevirtual_helper(rbx, rcx, rdx);
3008 }
3009
3010
3011 void TemplateTable::invokespecial(int byte_no) {
3012 transition(vtos, vtos);
3013 assert(byte_no == f1_byte, "use this argument");
3014 prepare_invoke(rbx, noreg, byte_no);
3015 // do the call
3016 __ verify_oop(rbx);
3017 __ profile_call(rax);
3018 __ jump_from_interpreted(rbx, rax);
3019 }
3020
3021
3022 void TemplateTable::invokestatic(int byte_no) {
3023 transition(vtos, vtos);
3024 assert(byte_no == f1_byte, "use this argument");
3025 prepare_invoke(rbx, noreg, byte_no);
3026 // do the call
3027 __ verify_oop(rbx);
3028 __ profile_call(rax);
3029 __ jump_from_interpreted(rbx, rax);
3030 }
3031
3032 void TemplateTable::fast_invokevfinal(int byte_no) {
3033 transition(vtos, vtos);
3034 assert(byte_no == f2_byte, "use this argument");
3035 __ stop("fast_invokevfinal not used on amd64");
3036 }
3037
3038 void TemplateTable::invokeinterface(int byte_no) {
3039 transition(vtos, vtos);
3040 assert(byte_no == f1_byte, "use this argument");
3041 prepare_invoke(rax, rbx, byte_no);
3042
3043 // rax: Interface
3044 // rbx: index
3045 // rcx: receiver
3046 // rdx: flags
3047
3048 // Special case of invokeinterface called for virtual method of
3049 // java.lang.Object. See cpCacheOop.cpp for details.
3050 // This code isn't produced by javac, but could be produced by
3051 // another compliant java compiler.
3052 Label notMethod;
3053 __ movl(r14, rdx);
3054 __ andl(r14, (1 << ConstantPoolCacheEntry::methodInterface));
3055 __ jcc(Assembler::zero, notMethod);
3056
3057 invokevirtual_helper(rbx, rcx, rdx);
3058 __ bind(notMethod);
3059
3060 // Get receiver klass into rdx - also a null check
3061 __ restore_locals(); // restore r14
3062 __ load_klass(rdx, rcx);
3063 __ verify_oop(rdx);
3064
3065 // profile this call
3066 __ profile_virtual_call(rdx, r13, r14);
3067
3068 Label no_such_interface, no_such_method;
3069
3070 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3071 rdx, rax, rbx,
3072 // outputs: method, scan temp. reg
3073 rbx, r13,
3074 no_such_interface);
3075
3076 // rbx,: methodOop to call
3077 // rcx: receiver
3078 // Check for abstract method error
3079 // Note: This should be done more efficiently via a throw_abstract_method_error
3080 // interpreter entry point and a conditional jump to it in case of a null
3081 // method.
3082 __ testptr(rbx, rbx);
3083 __ jcc(Assembler::zero, no_such_method);
3084
3085 // do the call
3086 // rcx: receiver
3087 // rbx,: methodOop
3088 __ jump_from_interpreted(rbx, rdx);
3089 __ should_not_reach_here();
3090
3091 // exception handling code follows...
3092 // note: must restore interpreter registers to canonical
3093 // state for exception handling to work correctly!
3094
3095 __ bind(no_such_method);
3096 // throw exception
3097 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3098 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3099 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3100 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3101 // the call_VM checks for exception, so we should never return here.
3102 __ should_not_reach_here();
3103
3104 __ bind(no_such_interface);
3105 // throw exception
3106 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3107 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3108 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3109 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3110 InterpreterRuntime::throw_IncompatibleClassChangeError));
3111 // the call_VM checks for exception, so we should never return here.
3112 __ should_not_reach_here();
3113 return;
3114 }
3115
3116 void TemplateTable::invokedynamic(int byte_no) {
3117 transition(vtos, vtos);
3118 assert(byte_no == f1_oop, "use this argument");
3119
3120 if (!EnableInvokeDynamic) {
3121 // We should not encounter this bytecode if !EnableInvokeDynamic.
3122 // The verifier will stop it. However, if we get past the verifier,
3123 // this will stop the thread in a reasonable way, without crashing the JVM.
3124 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3125 InterpreterRuntime::throw_IncompatibleClassChangeError));
3126 // the call_VM checks for exception, so we should never return here.
3127 __ should_not_reach_here();
3128 return;
3129 }
3130
3131 prepare_invoke(rax, rbx, byte_no);
3132
3133 // rax: CallSite object (f1)
3134 // rbx: unused (f2)
3135 // rcx: receiver address
3136 // rdx: flags (unused)
3137
3138 Register rax_callsite = rax;
3139 Register rcx_method_handle = rcx;
3140
3141 // %%% should make a type profile for any invokedynamic that takes a ref argument
3142 // profile this call
3143 __ profile_call(r13);
3144
3145 __ verify_oop(rax_callsite);
3146 __ load_heap_oop(rcx_method_handle, Address(rax_callsite, __ delayed_value(java_lang_invoke_CallSite::target_offset_in_bytes, rdx)));
3147 __ null_check(rcx_method_handle);
3148 __ verify_oop(rcx_method_handle);
3149 __ prepare_to_jump_from_interpreted();
3150 __ jump_to_method_handle_entry(rcx_method_handle, rdx);
3151 }
3152
3153
3154 //-----------------------------------------------------------------------------
3155 // Allocation
3156
3157 void TemplateTable::_new() {
3158 transition(vtos, atos);
3159 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3160 Label slow_case;
3161 Label done;
3162 Label initialize_header;
3163 Label initialize_object; // including clearing the fields
3164 Label allocate_shared;
3165
3166 __ get_cpool_and_tags(rsi, rax);
3167 // Make sure the class we're about to instantiate has been resolved.
3168 // This is done before loading instanceKlass to be consistent with the order
3169 // how Constant Pool is updated (see constantPoolOopDesc::klass_at_put)
3170 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
3171 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset),
3172 JVM_CONSTANT_Class);
3173 __ jcc(Assembler::notEqual, slow_case);
3174
3175 // get instanceKlass
3176 __ movptr(rsi, Address(rsi, rdx,
3177 Address::times_8, sizeof(constantPoolOopDesc)));
3178
3179 // make sure klass is initialized & doesn't have finalizer
3180 // make sure klass is fully initialized
3181 __ cmpl(Address(rsi,
3182 instanceKlass::init_state_offset_in_bytes() +
3183 sizeof(oopDesc)),
3184 instanceKlass::fully_initialized);
3185 __ jcc(Assembler::notEqual, slow_case);
3186
3187 // get instance_size in instanceKlass (scaled to a count of bytes)
3188 __ movl(rdx,
3189 Address(rsi,
3190 Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
3191 // test to see if it has a finalizer or is malformed in some way
3192 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3193 __ jcc(Assembler::notZero, slow_case);
3194
3195 // Allocate the instance
3196 // 1) Try to allocate in the TLAB
3197 // 2) if fail and the object is large allocate in the shared Eden
3198 // 3) if the above fails (or is not applicable), go to a slow case
3199 // (creates a new TLAB, etc.)
3200
3201 const bool allow_shared_alloc =
3202 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3203
3204 if (UseTLAB) {
3205 __ movptr(rax, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
3206 __ lea(rbx, Address(rax, rdx, Address::times_1));
3207 __ cmpptr(rbx, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
3208 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3209 __ movptr(Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3210 if (ZeroTLAB) {
3211 // the fields have been already cleared
3212 __ jmp(initialize_header);
3213 } else {
3214 // initialize both the header and fields
3215 __ jmp(initialize_object);
3216 }
3217 }
3218
3219 // Allocation in the shared Eden, if allowed.
3220 //
3221 // rdx: instance size in bytes
3222 if (allow_shared_alloc) {
3223 __ bind(allocate_shared);
3224
3225 ExternalAddress top((address)Universe::heap()->top_addr());
3226 ExternalAddress end((address)Universe::heap()->end_addr());
3227
3228 const Register RtopAddr = rscratch1;
3229 const Register RendAddr = rscratch2;
3230
3231 __ lea(RtopAddr, top);
3232 __ lea(RendAddr, end);
3233 __ movptr(rax, Address(RtopAddr, 0));
3234
3235 // For retries rax gets set by cmpxchgq
3236 Label retry;
3237 __ bind(retry);
3238 __ lea(rbx, Address(rax, rdx, Address::times_1));
3239 __ cmpptr(rbx, Address(RendAddr, 0));
3240 __ jcc(Assembler::above, slow_case);
3241
3242 // Compare rax with the top addr, and if still equal, store the new
3243 // top addr in rbx at the address of the top addr pointer. Sets ZF if was
3244 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3245 //
3246 // rax: object begin
3247 // rbx: object end
3248 // rdx: instance size in bytes
3249 if (os::is_MP()) {
3250 __ lock();
3251 }
3252 __ cmpxchgptr(rbx, Address(RtopAddr, 0));
3253
3254 // if someone beat us on the allocation, try again, otherwise continue
3255 __ jcc(Assembler::notEqual, retry);
3256
3257 __ incr_allocated_bytes(r15_thread, rdx, 0);
3258 }
3259
3260 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3261 // The object is initialized before the header. If the object size is
3262 // zero, go directly to the header initialization.
3263 __ bind(initialize_object);
3264 __ decrementl(rdx, sizeof(oopDesc));
3265 __ jcc(Assembler::zero, initialize_header);
3266
3267 // Initialize object fields
3268 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3269 __ shrl(rdx, LogBytesPerLong); // divide by oopSize to simplify the loop
3270 {
3271 Label loop;
3272 __ bind(loop);
3273 __ movq(Address(rax, rdx, Address::times_8,
3274 sizeof(oopDesc) - oopSize),
3275 rcx);
3276 __ decrementl(rdx);
3277 __ jcc(Assembler::notZero, loop);
3278 }
3279
3280 // initialize object header only.
3281 __ bind(initialize_header);
3282 if (UseBiasedLocking) {
3283 __ movptr(rscratch1, Address(rsi, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
3284 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()), rscratch1);
3285 } else {
3286 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()),
3287 (intptr_t) markOopDesc::prototype()); // header (address 0x1)
3288 }
3289 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3290 __ store_klass_gap(rax, rcx); // zero klass gap for compressed oops
3291 __ store_klass(rax, rsi); // store klass last
3292
3293 {
3294 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3295 // Trigger dtrace event for fastpath
3296 __ push(atos); // save the return value
3297 __ call_VM_leaf(
3298 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3299 __ pop(atos); // restore the return value
3300
3301 }
3302 __ jmp(done);
3303 }
3304
3305
3306 // slow case
3307 __ bind(slow_case);
3308 __ get_constant_pool(c_rarg1);
3309 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3310 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3311 __ verify_oop(rax);
3312
3313 // continue
3314 __ bind(done);
3315 }
3316
3317 void TemplateTable::newarray() {
3318 transition(itos, atos);
3319 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3320 __ movl(c_rarg2, rax);
3321 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3322 c_rarg1, c_rarg2);
3323 }
3324
3325 void TemplateTable::anewarray() {
3326 transition(itos, atos);
3327 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3328 __ get_constant_pool(c_rarg1);
3329 __ movl(c_rarg3, rax);
3330 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3331 c_rarg1, c_rarg2, c_rarg3);
3332 }
3333
3334 void TemplateTable::arraylength() {
3335 transition(atos, itos);
3336 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3337 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3338 }
3339
3340 void TemplateTable::checkcast() {
3341 transition(atos, atos);
3342 Label done, is_null, ok_is_subtype, quicked, resolved;
3343 __ testptr(rax, rax); // object is in rax
3344 __ jcc(Assembler::zero, is_null);
3345
3346 // Get cpool & tags index
3347 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3348 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3349 // See if bytecode has already been quicked
3350 __ cmpb(Address(rdx, rbx,
3351 Address::times_1,
3352 typeArrayOopDesc::header_size(T_BYTE) * wordSize),
3353 JVM_CONSTANT_Class);
3354 __ jcc(Assembler::equal, quicked);
3355 __ push(atos); // save receiver for result, and for GC
3356 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3357 __ pop_ptr(rdx); // restore receiver
3358 __ jmpb(resolved);
3359
3360 // Get superklass in rax and subklass in rbx
3361 __ bind(quicked);
3362 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
3363 __ movptr(rax, Address(rcx, rbx,
3364 Address::times_8, sizeof(constantPoolOopDesc)));
3365
3366 __ bind(resolved);
3367 __ load_klass(rbx, rdx);
3368
3369 // Generate subtype check. Blows rcx, rdi. Object in rdx.
3370 // Superklass in rax. Subklass in rbx.
3371 __ gen_subtype_check(rbx, ok_is_subtype);
3372
3373 // Come here on failure
3374 __ push_ptr(rdx);
3375 // object is at TOS
3376 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3377
3378 // Come here on success
3379 __ bind(ok_is_subtype);
3380 __ mov(rax, rdx); // Restore object in rdx
3381
3382 // Collect counts on whether this check-cast sees NULLs a lot or not.
3383 if (ProfileInterpreter) {
3384 __ jmp(done);
3385 __ bind(is_null);
3386 __ profile_null_seen(rcx);
3387 } else {
3388 __ bind(is_null); // same as 'done'
3389 }
3390 __ bind(done);
3391 }
3392
3393 void TemplateTable::instanceof() {
3394 transition(atos, itos);
3395 Label done, is_null, ok_is_subtype, quicked, resolved;
3396 __ testptr(rax, rax);
3397 __ jcc(Assembler::zero, is_null);
3398
3399 // Get cpool & tags index
3400 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3401 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3402 // See if bytecode has already been quicked
3403 __ cmpb(Address(rdx, rbx,
3404 Address::times_1,
3405 typeArrayOopDesc::header_size(T_BYTE) * wordSize),
3406 JVM_CONSTANT_Class);
3407 __ jcc(Assembler::equal, quicked);
3408
3409 __ push(atos); // save receiver for result, and for GC
3410 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3411 __ pop_ptr(rdx); // restore receiver
3412 __ verify_oop(rdx);
3413 __ load_klass(rdx, rdx);
3414 __ jmpb(resolved);
3415
3416 // Get superklass in rax and subklass in rdx
3417 __ bind(quicked);
3418 __ load_klass(rdx, rax);
3419 __ movptr(rax, Address(rcx, rbx,
3420 Address::times_8, sizeof(constantPoolOopDesc)));
3421
3422 __ bind(resolved);
3423
3424 // Generate subtype check. Blows rcx, rdi
3425 // Superklass in rax. Subklass in rdx.
3426 __ gen_subtype_check(rdx, ok_is_subtype);
3427
3428 // Come here on failure
3429 __ xorl(rax, rax);
3430 __ jmpb(done);
3431 // Come here on success
3432 __ bind(ok_is_subtype);
3433 __ movl(rax, 1);
3434
3435 // Collect counts on whether this test sees NULLs a lot or not.
3436 if (ProfileInterpreter) {
3437 __ jmp(done);
3438 __ bind(is_null);
3439 __ profile_null_seen(rcx);
3440 } else {
3441 __ bind(is_null); // same as 'done'
3442 }
3443 __ bind(done);
3444 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
3445 // rax = 1: obj != NULL and obj is an instanceof the specified klass
3446 }
3447
3448 //-----------------------------------------------------------------------------
3449 // Breakpoints
3450 void TemplateTable::_breakpoint() {
3451 // Note: We get here even if we are single stepping..
3452 // jbug inists on setting breakpoints at every bytecode
3453 // even if we are in single step mode.
3454
3455 transition(vtos, vtos);
3456
3457 // get the unpatched byte code
3458 __ get_method(c_rarg1);
3459 __ call_VM(noreg,
3460 CAST_FROM_FN_PTR(address,
3461 InterpreterRuntime::get_original_bytecode_at),
3462 c_rarg1, r13);
3463 __ mov(rbx, rax);
3464
3465 // post the breakpoint event
3466 __ get_method(c_rarg1);
3467 __ call_VM(noreg,
3468 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3469 c_rarg1, r13);
3470
3471 // complete the execution of original bytecode
3472 __ dispatch_only_normal(vtos);
3473 }
3474
3475 //-----------------------------------------------------------------------------
3476 // Exceptions
3477
3478 void TemplateTable::athrow() {
3479 transition(atos, vtos);
3480 __ null_check(rax);
3481 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3482 }
3483
3484 //-----------------------------------------------------------------------------
3485 // Synchronization
3486 //
3487 // Note: monitorenter & exit are symmetric routines; which is reflected
3488 // in the assembly code structure as well
3489 //
3490 // Stack layout:
3491 //
3492 // [expressions ] <--- rsp = expression stack top
3493 // ..
3494 // [expressions ]
3495 // [monitor entry] <--- monitor block top = expression stack bot
3496 // ..
3497 // [monitor entry]
3498 // [frame data ] <--- monitor block bot
3499 // ...
3500 // [saved rbp ] <--- rbp
3501 void TemplateTable::monitorenter() {
3502 transition(atos, vtos);
3503
3504 // check for NULL object
3505 __ null_check(rax);
3506
3507 const Address monitor_block_top(
3508 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3509 const Address monitor_block_bot(
3510 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3511 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3512
3513 Label allocated;
3514
3515 // initialize entry pointer
3516 __ xorl(c_rarg1, c_rarg1); // points to free slot or NULL
3517
3518 // find a free slot in the monitor block (result in c_rarg1)
3519 {
3520 Label entry, loop, exit;
3521 __ movptr(c_rarg3, monitor_block_top); // points to current entry,
3522 // starting with top-most entry
3523 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3524 // of monitor block
3525 __ jmpb(entry);
3526
3527 __ bind(loop);
3528 // check if current entry is used
3529 __ cmpptr(Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
3530 // if not used then remember entry in c_rarg1
3531 __ cmov(Assembler::equal, c_rarg1, c_rarg3);
3532 // check if current entry is for same object
3533 __ cmpptr(rax, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3534 // if same object then stop searching
3535 __ jccb(Assembler::equal, exit);
3536 // otherwise advance to next entry
3537 __ addptr(c_rarg3, entry_size);
3538 __ bind(entry);
3539 // check if bottom reached
3540 __ cmpptr(c_rarg3, c_rarg2);
3541 // if not at bottom then check this entry
3542 __ jcc(Assembler::notEqual, loop);
3543 __ bind(exit);
3544 }
3545
3546 __ testptr(c_rarg1, c_rarg1); // check if a slot has been found
3547 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
3548
3549 // allocate one if there's no free slot
3550 {
3551 Label entry, loop;
3552 // 1. compute new pointers // rsp: old expression stack top
3553 __ movptr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
3554 __ subptr(rsp, entry_size); // move expression stack top
3555 __ subptr(c_rarg1, entry_size); // move expression stack bottom
3556 __ mov(c_rarg3, rsp); // set start value for copy loop
3557 __ movptr(monitor_block_bot, c_rarg1); // set new monitor block bottom
3558 __ jmp(entry);
3559 // 2. move expression stack contents
3560 __ bind(loop);
3561 __ movptr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3562 // word from old location
3563 __ movptr(Address(c_rarg3, 0), c_rarg2); // and store it at new location
3564 __ addptr(c_rarg3, wordSize); // advance to next word
3565 __ bind(entry);
3566 __ cmpptr(c_rarg3, c_rarg1); // check if bottom reached
3567 __ jcc(Assembler::notEqual, loop); // if not at bottom then
3568 // copy next word
3569 }
3570
3571 // call run-time routine
3572 // c_rarg1: points to monitor entry
3573 __ bind(allocated);
3574
3575 // Increment bcp to point to the next bytecode, so exception
3576 // handling for async. exceptions work correctly.
3577 // The object has already been poped from the stack, so the
3578 // expression stack looks correct.
3579 __ increment(r13);
3580
3581 // store object
3582 __ movptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), rax);
3583 __ lock_object(c_rarg1);
3584
3585 // check to make sure this monitor doesn't cause stack overflow after locking
3586 __ save_bcp(); // in case of exception
3587 __ generate_stack_overflow_check(0);
3588
3589 // The bcp has already been incremented. Just need to dispatch to
3590 // next instruction.
3591 __ dispatch_next(vtos);
3592 }
3593
3594
3595 void TemplateTable::monitorexit() {
3596 transition(atos, vtos);
3597
3598 // check for NULL object
3599 __ null_check(rax);
3600
3601 const Address monitor_block_top(
3602 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3603 const Address monitor_block_bot(
3604 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3605 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3606
3607 Label found;
3608
3609 // find matching slot
3610 {
3611 Label entry, loop;
3612 __ movptr(c_rarg1, monitor_block_top); // points to current entry,
3613 // starting with top-most entry
3614 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3615 // of monitor block
3616 __ jmpb(entry);
3617
3618 __ bind(loop);
3619 // check if current entry is for same object
3620 __ cmpptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3621 // if same object then stop searching
3622 __ jcc(Assembler::equal, found);
3623 // otherwise advance to next entry
3624 __ addptr(c_rarg1, entry_size);
3625 __ bind(entry);
3626 // check if bottom reached
3627 __ cmpptr(c_rarg1, c_rarg2);
3628 // if not at bottom then check this entry
3629 __ jcc(Assembler::notEqual, loop);
3630 }
3631
3632 // error handling. Unlocking was not block-structured
3633 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3634 InterpreterRuntime::throw_illegal_monitor_state_exception));
3635 __ should_not_reach_here();
3636
3637 // call run-time routine
3638 // rsi: points to monitor entry
3639 __ bind(found);
3640 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3641 __ unlock_object(c_rarg1);
3642 __ pop_ptr(rax); // discard object
3643 }
3644
3645
3646 // Wide instructions
3647 void TemplateTable::wide() {
3648 transition(vtos, vtos);
3649 __ load_unsigned_byte(rbx, at_bcp(1));
3650 __ lea(rscratch1, ExternalAddress((address)Interpreter::_wentry_point));
3651 __ jmp(Address(rscratch1, rbx, Address::times_8));
3652 // Note: the r13 increment step is part of the individual wide
3653 // bytecode implementations
3654 }
3655
3656
3657 // Multi arrays
3658 void TemplateTable::multianewarray() {
3659 transition(vtos, atos);
3660 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3661 // last dim is on top of stack; we want address of first one:
3662 // first_addr = last_addr + (ndims - 1) * wordSize
3663 __ lea(c_rarg1, Address(rsp, rax, Address::times_8, -wordSize));
3664 call_VM(rax,
3665 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3666 c_rarg1);
3667 __ load_unsigned_byte(rbx, at_bcp(3));
3668 __ lea(rsp, Address(rsp, rbx, Address::times_8));
3669 }
3670 #endif // !CC_INTERP