1 /*
2 * Copyright (c) 2003, 2014, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.inline.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #ifndef CC_INTERP
42
43 #define __ _masm->
44
45 // Platform-dependent initialization
46
47 void TemplateTable::pd_initialize() {
48 // No amd64 specific initialization
49 }
50
51 // Address computation: local variables
52
53 static inline Address iaddress(int n) {
54 return Address(r14, Interpreter::local_offset_in_bytes(n));
55 }
56
57 static inline Address laddress(int n) {
58 return iaddress(n + 1);
59 }
60
61 static inline Address faddress(int n) {
62 return iaddress(n);
63 }
64
65 static inline Address daddress(int n) {
66 return laddress(n);
67 }
68
69 static inline Address aaddress(int n) {
70 return iaddress(n);
71 }
72
73 static inline Address iaddress(Register r) {
74 return Address(r14, r, Address::times_8);
75 }
76
77 static inline Address laddress(Register r) {
78 return Address(r14, r, Address::times_8, Interpreter::local_offset_in_bytes(1));
79 }
80
81 static inline Address faddress(Register r) {
82 return iaddress(r);
83 }
84
85 static inline Address daddress(Register r) {
86 return laddress(r);
87 }
88
89 static inline Address aaddress(Register r) {
90 return iaddress(r);
91 }
92
93 static inline Address at_rsp() {
94 return Address(rsp, 0);
95 }
96
97 // At top of Java expression stack which may be different than esp(). It
98 // isn't for category 1 objects.
99 static inline Address at_tos () {
100 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
101 }
102
103 static inline Address at_tos_p1() {
104 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
105 }
106
107 static inline Address at_tos_p2() {
108 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
109 }
110
111 // Condition conversion
112 static Assembler::Condition j_not(TemplateTable::Condition cc) {
113 switch (cc) {
114 case TemplateTable::equal : return Assembler::notEqual;
115 case TemplateTable::not_equal : return Assembler::equal;
116 case TemplateTable::less : return Assembler::greaterEqual;
117 case TemplateTable::less_equal : return Assembler::greater;
118 case TemplateTable::greater : return Assembler::lessEqual;
119 case TemplateTable::greater_equal: return Assembler::less;
120 }
121 ShouldNotReachHere();
122 return Assembler::zero;
123 }
124
125
126 // Miscelaneous helper routines
127 // Store an oop (or NULL) at the address described by obj.
128 // If val == noreg this means store a NULL
129
130 static void do_oop_store(InterpreterMacroAssembler* _masm,
131 Address obj,
132 Register val,
133 BarrierSet::Name barrier,
134 bool precise) {
135 assert(val == noreg || val == rax, "parameter is just for looks");
136 switch (barrier) {
137 #if INCLUDE_ALL_GCS
138 case BarrierSet::G1SATBCT:
139 case BarrierSet::G1SATBCTLogging:
140 {
141 // flatten object address if needed
142 if (obj.index() == noreg && obj.disp() == 0) {
143 if (obj.base() != rdx) {
144 __ movq(rdx, obj.base());
145 }
146 } else {
147 __ leaq(rdx, obj);
148 }
149 __ g1_write_barrier_pre(rdx /* obj */,
150 rbx /* pre_val */,
151 r15_thread /* thread */,
152 r8 /* tmp */,
153 val != noreg /* tosca_live */,
154 false /* expand_call */);
155 if (val == noreg) {
156 __ store_heap_oop_null(Address(rdx, 0));
157 } else {
158 // G1 barrier needs uncompressed oop for region cross check.
159 Register new_val = val;
160 if (UseCompressedOops) {
161 new_val = rbx;
162 __ movptr(new_val, val);
163 }
164 __ store_heap_oop(Address(rdx, 0), val);
165 __ g1_write_barrier_post(rdx /* store_adr */,
166 new_val /* new_val */,
167 r15_thread /* thread */,
168 r8 /* tmp */,
169 rbx /* tmp2 */);
170 }
171 }
172 break;
173 #endif // INCLUDE_ALL_GCS
174 case BarrierSet::CardTableModRef:
175 case BarrierSet::CardTableExtension:
176 {
177 if (val == noreg) {
178 __ store_heap_oop_null(obj);
179 } else {
180 __ store_heap_oop(obj, val);
181 // flatten object address if needed
182 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
183 __ store_check(obj.base());
184 } else {
185 __ leaq(rdx, obj);
186 __ store_check(rdx);
187 }
188 }
189 }
190 break;
191 case BarrierSet::ModRef:
192 case BarrierSet::Other:
193 if (val == noreg) {
194 __ store_heap_oop_null(obj);
195 } else {
196 __ store_heap_oop(obj, val);
197 }
198 break;
199 default :
200 ShouldNotReachHere();
201
202 }
203 }
204
205 Address TemplateTable::at_bcp(int offset) {
206 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
207 return Address(r13, offset);
208 }
209
210 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
211 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
212 int byte_no) {
213 if (!RewriteBytecodes) return;
214 Label L_patch_done;
215
216 switch (bc) {
217 case Bytecodes::_fast_aputfield:
218 case Bytecodes::_fast_bputfield:
219 case Bytecodes::_fast_cputfield:
220 case Bytecodes::_fast_dputfield:
221 case Bytecodes::_fast_fputfield:
222 case Bytecodes::_fast_iputfield:
223 case Bytecodes::_fast_lputfield:
224 case Bytecodes::_fast_sputfield:
225 {
226 // We skip bytecode quickening for putfield instructions when
227 // the put_code written to the constant pool cache is zero.
228 // This is required so that every execution of this instruction
229 // calls out to InterpreterRuntime::resolve_get_put to do
230 // additional, required work.
231 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
232 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
233 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
234 __ movl(bc_reg, bc);
235 __ cmpl(temp_reg, (int) 0);
236 __ jcc(Assembler::zero, L_patch_done); // don't patch
237 }
238 break;
239 default:
240 assert(byte_no == -1, "sanity");
241 // the pair bytecodes have already done the load.
242 if (load_bc_into_bc_reg) {
243 __ movl(bc_reg, bc);
244 }
245 }
246
247 if (JvmtiExport::can_post_breakpoint()) {
248 Label L_fast_patch;
249 // if a breakpoint is present we can't rewrite the stream directly
250 __ movzbl(temp_reg, at_bcp(0));
251 __ cmpl(temp_reg, Bytecodes::_breakpoint);
252 __ jcc(Assembler::notEqual, L_fast_patch);
253 __ get_method(temp_reg);
254 // Let breakpoint table handling rewrite to quicker bytecode
255 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, r13, bc_reg);
256 #ifndef ASSERT
257 __ jmpb(L_patch_done);
258 #else
259 __ jmp(L_patch_done);
260 #endif
261 __ bind(L_fast_patch);
262 }
263
264 #ifdef ASSERT
265 Label L_okay;
266 __ load_unsigned_byte(temp_reg, at_bcp(0));
267 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
268 __ jcc(Assembler::equal, L_okay);
269 __ cmpl(temp_reg, bc_reg);
270 __ jcc(Assembler::equal, L_okay);
271 __ stop("patching the wrong bytecode");
272 __ bind(L_okay);
273 #endif
274
275 // patch bytecode
276 __ movb(at_bcp(0), bc_reg);
277 __ bind(L_patch_done);
278 }
279
280
281 // Individual instructions
282
283 void TemplateTable::nop() {
284 transition(vtos, vtos);
285 // nothing to do
286 }
287
288 void TemplateTable::shouldnotreachhere() {
289 transition(vtos, vtos);
290 __ stop("shouldnotreachhere bytecode");
291 }
292
293 void TemplateTable::aconst_null() {
294 transition(vtos, atos);
295 __ xorl(rax, rax);
296 }
297
298 void TemplateTable::iconst(int value) {
299 transition(vtos, itos);
300 if (value == 0) {
301 __ xorl(rax, rax);
302 } else {
303 __ movl(rax, value);
304 }
305 }
306
307 void TemplateTable::lconst(int value) {
308 transition(vtos, ltos);
309 if (value == 0) {
310 __ xorl(rax, rax);
311 } else {
312 __ movl(rax, value);
313 }
314 }
315
316 void TemplateTable::fconst(int value) {
317 transition(vtos, ftos);
318 static float one = 1.0f, two = 2.0f;
319 switch (value) {
320 case 0:
321 __ xorps(xmm0, xmm0);
322 break;
323 case 1:
324 __ movflt(xmm0, ExternalAddress((address) &one));
325 break;
326 case 2:
327 __ movflt(xmm0, ExternalAddress((address) &two));
328 break;
329 default:
330 ShouldNotReachHere();
331 break;
332 }
333 }
334
335 void TemplateTable::dconst(int value) {
336 transition(vtos, dtos);
337 static double one = 1.0;
338 switch (value) {
339 case 0:
340 __ xorpd(xmm0, xmm0);
341 break;
342 case 1:
343 __ movdbl(xmm0, ExternalAddress((address) &one));
344 break;
345 default:
346 ShouldNotReachHere();
347 break;
348 }
349 }
350
351 void TemplateTable::bipush() {
352 transition(vtos, itos);
353 __ load_signed_byte(rax, at_bcp(1));
354 }
355
356 void TemplateTable::sipush() {
357 transition(vtos, itos);
358 __ load_unsigned_short(rax, at_bcp(1));
359 __ bswapl(rax);
360 __ sarl(rax, 16);
361 }
362
363 void TemplateTable::ldc(bool wide) {
364 transition(vtos, vtos);
365 Label call_ldc, notFloat, notClass, Done;
366
367 if (wide) {
368 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
369 } else {
370 __ load_unsigned_byte(rbx, at_bcp(1));
371 }
372
373 __ get_cpool_and_tags(rcx, rax);
374 const int base_offset = ConstantPool::header_size() * wordSize;
375 const int tags_offset = Array<u1>::base_offset_in_bytes();
376
377 // get type
378 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
379
380 // unresolved class - get the resolved class
381 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
382 __ jccb(Assembler::equal, call_ldc);
383
384 // unresolved class in error state - call into runtime to throw the error
385 // from the first resolution attempt
386 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
387 __ jccb(Assembler::equal, call_ldc);
388
389 // resolved class - need to call vm to get java mirror of the class
390 __ cmpl(rdx, JVM_CONSTANT_Class);
391 __ jcc(Assembler::notEqual, notClass);
392
393 __ bind(call_ldc);
394 __ movl(c_rarg1, wide);
395 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
396 __ push_ptr(rax);
397 __ verify_oop(rax);
398 __ jmp(Done);
399
400 __ bind(notClass);
401 __ cmpl(rdx, JVM_CONSTANT_Float);
402 __ jccb(Assembler::notEqual, notFloat);
403 // ftos
404 __ movflt(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
405 __ push_f();
406 __ jmp(Done);
407
408 __ bind(notFloat);
409 #ifdef ASSERT
410 {
411 Label L;
412 __ cmpl(rdx, JVM_CONSTANT_Integer);
413 __ jcc(Assembler::equal, L);
414 // String and Object are rewritten to fast_aldc
415 __ stop("unexpected tag type in ldc");
416 __ bind(L);
417 }
418 #endif
419 // itos JVM_CONSTANT_Integer only
420 __ movl(rax, Address(rcx, rbx, Address::times_8, base_offset));
421 __ push_i(rax);
422 __ bind(Done);
423 }
424
425 // Fast path for caching oop constants.
426 void TemplateTable::fast_aldc(bool wide) {
427 transition(vtos, atos);
428
429 Register result = rax;
430 Register tmp = rdx;
431 int index_size = wide ? sizeof(u2) : sizeof(u1);
432
433 Label resolved;
434
435 // We are resolved if the resolved reference cache entry contains a
436 // non-null object (String, MethodType, etc.)
437 assert_different_registers(result, tmp);
438 __ get_cache_index_at_bcp(tmp, 1, index_size);
439 __ load_resolved_reference_at_index(result, tmp);
440 __ testl(result, result);
441 __ jcc(Assembler::notZero, resolved);
442
443 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
444
445 // first time invocation - must resolve first
446 __ movl(tmp, (int)bytecode());
447 __ call_VM(result, entry, tmp);
448
449 __ bind(resolved);
450
451 if (VerifyOops) {
452 __ verify_oop(result);
453 }
454 }
455
456 void TemplateTable::ldc2_w() {
457 transition(vtos, vtos);
458 Label Long, Done;
459 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
460
461 __ get_cpool_and_tags(rcx, rax);
462 const int base_offset = ConstantPool::header_size() * wordSize;
463 const int tags_offset = Array<u1>::base_offset_in_bytes();
464
465 // get type
466 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
467 JVM_CONSTANT_Double);
468 __ jccb(Assembler::notEqual, Long);
469 // dtos
470 __ movdbl(xmm0, Address(rcx, rbx, Address::times_8, base_offset));
471 __ push_d();
472 __ jmpb(Done);
473
474 __ bind(Long);
475 // ltos
476 __ movq(rax, Address(rcx, rbx, Address::times_8, base_offset));
477 __ push_l();
478
479 __ bind(Done);
480 }
481
482 void TemplateTable::locals_index(Register reg, int offset) {
483 __ load_unsigned_byte(reg, at_bcp(offset));
484 __ negptr(reg);
485 }
486
487 void TemplateTable::iload() {
488 transition(vtos, itos);
489 if (RewriteFrequentPairs) {
490 Label rewrite, done;
491 const Register bc = c_rarg3;
492 assert(rbx != bc, "register damaged");
493
494 // get next byte
495 __ load_unsigned_byte(rbx,
496 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
497 // if _iload, wait to rewrite to iload2. We only want to rewrite the
498 // last two iloads in a pair. Comparing against fast_iload means that
499 // the next bytecode is neither an iload or a caload, and therefore
500 // an iload pair.
501 __ cmpl(rbx, Bytecodes::_iload);
502 __ jcc(Assembler::equal, done);
503
504 __ cmpl(rbx, Bytecodes::_fast_iload);
505 __ movl(bc, Bytecodes::_fast_iload2);
506 __ jccb(Assembler::equal, rewrite);
507
508 // if _caload, rewrite to fast_icaload
509 __ cmpl(rbx, Bytecodes::_caload);
510 __ movl(bc, Bytecodes::_fast_icaload);
511 __ jccb(Assembler::equal, rewrite);
512
513 // rewrite so iload doesn't check again.
514 __ movl(bc, Bytecodes::_fast_iload);
515
516 // rewrite
517 // bc: fast bytecode
518 __ bind(rewrite);
519 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
520 __ bind(done);
521 }
522
523 // Get the local value into tos
524 locals_index(rbx);
525 __ movl(rax, iaddress(rbx));
526 }
527
528 void TemplateTable::fast_iload2() {
529 transition(vtos, itos);
530 locals_index(rbx);
531 __ movl(rax, iaddress(rbx));
532 __ push(itos);
533 locals_index(rbx, 3);
534 __ movl(rax, iaddress(rbx));
535 }
536
537 void TemplateTable::fast_iload() {
538 transition(vtos, itos);
539 locals_index(rbx);
540 __ movl(rax, iaddress(rbx));
541 }
542
543 void TemplateTable::lload() {
544 transition(vtos, ltos);
545 locals_index(rbx);
546 __ movq(rax, laddress(rbx));
547 }
548
549 void TemplateTable::fload() {
550 transition(vtos, ftos);
551 locals_index(rbx);
552 __ movflt(xmm0, faddress(rbx));
553 }
554
555 void TemplateTable::dload() {
556 transition(vtos, dtos);
557 locals_index(rbx);
558 __ movdbl(xmm0, daddress(rbx));
559 }
560
561 void TemplateTable::aload() {
562 transition(vtos, atos);
563 locals_index(rbx);
564 __ movptr(rax, aaddress(rbx));
565 }
566
567 void TemplateTable::locals_index_wide(Register reg) {
568 __ load_unsigned_short(reg, at_bcp(2));
569 __ bswapl(reg);
570 __ shrl(reg, 16);
571 __ negptr(reg);
572 }
573
574 void TemplateTable::wide_iload() {
575 transition(vtos, itos);
576 locals_index_wide(rbx);
577 __ movl(rax, iaddress(rbx));
578 }
579
580 void TemplateTable::wide_lload() {
581 transition(vtos, ltos);
582 locals_index_wide(rbx);
583 __ movq(rax, laddress(rbx));
584 }
585
586 void TemplateTable::wide_fload() {
587 transition(vtos, ftos);
588 locals_index_wide(rbx);
589 __ movflt(xmm0, faddress(rbx));
590 }
591
592 void TemplateTable::wide_dload() {
593 transition(vtos, dtos);
594 locals_index_wide(rbx);
595 __ movdbl(xmm0, daddress(rbx));
596 }
597
598 void TemplateTable::wide_aload() {
599 transition(vtos, atos);
600 locals_index_wide(rbx);
601 __ movptr(rax, aaddress(rbx));
602 }
603
604 void TemplateTable::index_check(Register array, Register index) {
605 // destroys rbx
606 // check array
607 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
608 // sign extend index for use by indexed load
609 __ movl2ptr(index, index);
610 // check index
611 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
612 if (index != rbx) {
613 // ??? convention: move aberrant index into ebx for exception message
614 assert(rbx != array, "different registers");
615 __ movl(rbx, index);
616 }
617 __ jump_cc(Assembler::aboveEqual,
618 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
619 }
620
621 void TemplateTable::iaload() {
622 transition(itos, itos);
623 __ pop_ptr(rdx);
624 // eax: index
625 // rdx: array
626 index_check(rdx, rax); // kills rbx
627 __ movl(rax, Address(rdx, rax,
628 Address::times_4,
629 arrayOopDesc::base_offset_in_bytes(T_INT)));
630 }
631
632 void TemplateTable::laload() {
633 transition(itos, ltos);
634 __ pop_ptr(rdx);
635 // eax: index
636 // rdx: array
637 index_check(rdx, rax); // kills rbx
638 __ movq(rax, Address(rdx, rbx,
639 Address::times_8,
640 arrayOopDesc::base_offset_in_bytes(T_LONG)));
641 }
642
643 void TemplateTable::faload() {
644 transition(itos, ftos);
645 __ pop_ptr(rdx);
646 // eax: index
647 // rdx: array
648 index_check(rdx, rax); // kills rbx
649 __ movflt(xmm0, Address(rdx, rax,
650 Address::times_4,
651 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
652 }
653
654 void TemplateTable::daload() {
655 transition(itos, dtos);
656 __ pop_ptr(rdx);
657 // eax: index
658 // rdx: array
659 index_check(rdx, rax); // kills rbx
660 __ movdbl(xmm0, Address(rdx, rax,
661 Address::times_8,
662 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
663 }
664
665 void TemplateTable::aaload() {
666 transition(itos, atos);
667 __ pop_ptr(rdx);
668 // eax: index
669 // rdx: array
670 index_check(rdx, rax); // kills rbx
671 __ load_heap_oop(rax, Address(rdx, rax,
672 UseCompressedOops ? Address::times_4 : Address::times_8,
673 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
674 }
675
676 void TemplateTable::baload() {
677 transition(itos, itos);
678 __ pop_ptr(rdx);
679 // eax: index
680 // rdx: array
681 index_check(rdx, rax); // kills rbx
682 __ load_signed_byte(rax,
683 Address(rdx, rax,
684 Address::times_1,
685 arrayOopDesc::base_offset_in_bytes(T_BYTE)));
686 }
687
688 void TemplateTable::caload() {
689 transition(itos, itos);
690 __ pop_ptr(rdx);
691 // eax: index
692 // rdx: array
693 index_check(rdx, rax); // kills rbx
694 __ load_unsigned_short(rax,
695 Address(rdx, rax,
696 Address::times_2,
697 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
698 }
699
700 // iload followed by caload frequent pair
701 void TemplateTable::fast_icaload() {
702 transition(vtos, itos);
703 // load index out of locals
704 locals_index(rbx);
705 __ movl(rax, iaddress(rbx));
706
707 // eax: index
708 // rdx: array
709 __ pop_ptr(rdx);
710 index_check(rdx, rax); // kills rbx
711 __ load_unsigned_short(rax,
712 Address(rdx, rax,
713 Address::times_2,
714 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
715 }
716
717 void TemplateTable::saload() {
718 transition(itos, itos);
719 __ pop_ptr(rdx);
720 // eax: index
721 // rdx: array
722 index_check(rdx, rax); // kills rbx
723 __ load_signed_short(rax,
724 Address(rdx, rax,
725 Address::times_2,
726 arrayOopDesc::base_offset_in_bytes(T_SHORT)));
727 }
728
729 void TemplateTable::iload(int n) {
730 transition(vtos, itos);
731 __ movl(rax, iaddress(n));
732 }
733
734 void TemplateTable::lload(int n) {
735 transition(vtos, ltos);
736 __ movq(rax, laddress(n));
737 }
738
739 void TemplateTable::fload(int n) {
740 transition(vtos, ftos);
741 __ movflt(xmm0, faddress(n));
742 }
743
744 void TemplateTable::dload(int n) {
745 transition(vtos, dtos);
746 __ movdbl(xmm0, daddress(n));
747 }
748
749 void TemplateTable::aload(int n) {
750 transition(vtos, atos);
751 __ movptr(rax, aaddress(n));
752 }
753
754 void TemplateTable::aload_0() {
755 transition(vtos, atos);
756 // According to bytecode histograms, the pairs:
757 //
758 // _aload_0, _fast_igetfield
759 // _aload_0, _fast_agetfield
760 // _aload_0, _fast_fgetfield
761 //
762 // occur frequently. If RewriteFrequentPairs is set, the (slow)
763 // _aload_0 bytecode checks if the next bytecode is either
764 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
765 // rewrites the current bytecode into a pair bytecode; otherwise it
766 // rewrites the current bytecode into _fast_aload_0 that doesn't do
767 // the pair check anymore.
768 //
769 // Note: If the next bytecode is _getfield, the rewrite must be
770 // delayed, otherwise we may miss an opportunity for a pair.
771 //
772 // Also rewrite frequent pairs
773 // aload_0, aload_1
774 // aload_0, iload_1
775 // These bytecodes with a small amount of code are most profitable
776 // to rewrite
777 if (RewriteFrequentPairs) {
778 Label rewrite, done;
779 const Register bc = c_rarg3;
780 assert(rbx != bc, "register damaged");
781 // get next byte
782 __ load_unsigned_byte(rbx,
783 at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
784
785 // do actual aload_0
786 aload(0);
787
788 // if _getfield then wait with rewrite
789 __ cmpl(rbx, Bytecodes::_getfield);
790 __ jcc(Assembler::equal, done);
791
792 // if _igetfield then reqrite to _fast_iaccess_0
793 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) ==
794 Bytecodes::_aload_0,
795 "fix bytecode definition");
796 __ cmpl(rbx, Bytecodes::_fast_igetfield);
797 __ movl(bc, Bytecodes::_fast_iaccess_0);
798 __ jccb(Assembler::equal, rewrite);
799
800 // if _agetfield then reqrite to _fast_aaccess_0
801 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) ==
802 Bytecodes::_aload_0,
803 "fix bytecode definition");
804 __ cmpl(rbx, Bytecodes::_fast_agetfield);
805 __ movl(bc, Bytecodes::_fast_aaccess_0);
806 __ jccb(Assembler::equal, rewrite);
807
808 // if _fgetfield then reqrite to _fast_faccess_0
809 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) ==
810 Bytecodes::_aload_0,
811 "fix bytecode definition");
812 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
813 __ movl(bc, Bytecodes::_fast_faccess_0);
814 __ jccb(Assembler::equal, rewrite);
815
816 // else rewrite to _fast_aload0
817 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) ==
818 Bytecodes::_aload_0,
819 "fix bytecode definition");
820 __ movl(bc, Bytecodes::_fast_aload_0);
821
822 // rewrite
823 // bc: fast bytecode
824 __ bind(rewrite);
825 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
826
827 __ bind(done);
828 } else {
829 aload(0);
830 }
831 }
832
833 void TemplateTable::istore() {
834 transition(itos, vtos);
835 locals_index(rbx);
836 __ movl(iaddress(rbx), rax);
837 }
838
839 void TemplateTable::lstore() {
840 transition(ltos, vtos);
841 locals_index(rbx);
842 __ movq(laddress(rbx), rax);
843 }
844
845 void TemplateTable::fstore() {
846 transition(ftos, vtos);
847 locals_index(rbx);
848 __ movflt(faddress(rbx), xmm0);
849 }
850
851 void TemplateTable::dstore() {
852 transition(dtos, vtos);
853 locals_index(rbx);
854 __ movdbl(daddress(rbx), xmm0);
855 }
856
857 void TemplateTable::astore() {
858 transition(vtos, vtos);
859 __ pop_ptr(rax);
860 locals_index(rbx);
861 __ movptr(aaddress(rbx), rax);
862 }
863
864 void TemplateTable::wide_istore() {
865 transition(vtos, vtos);
866 __ pop_i();
867 locals_index_wide(rbx);
868 __ movl(iaddress(rbx), rax);
869 }
870
871 void TemplateTable::wide_lstore() {
872 transition(vtos, vtos);
873 __ pop_l();
874 locals_index_wide(rbx);
875 __ movq(laddress(rbx), rax);
876 }
877
878 void TemplateTable::wide_fstore() {
879 transition(vtos, vtos);
880 __ pop_f();
881 locals_index_wide(rbx);
882 __ movflt(faddress(rbx), xmm0);
883 }
884
885 void TemplateTable::wide_dstore() {
886 transition(vtos, vtos);
887 __ pop_d();
888 locals_index_wide(rbx);
889 __ movdbl(daddress(rbx), xmm0);
890 }
891
892 void TemplateTable::wide_astore() {
893 transition(vtos, vtos);
894 __ pop_ptr(rax);
895 locals_index_wide(rbx);
896 __ movptr(aaddress(rbx), rax);
897 }
898
899 void TemplateTable::iastore() {
900 transition(itos, vtos);
901 __ pop_i(rbx);
902 __ pop_ptr(rdx);
903 // eax: value
904 // ebx: index
905 // rdx: array
906 index_check(rdx, rbx); // prefer index in ebx
907 __ movl(Address(rdx, rbx,
908 Address::times_4,
909 arrayOopDesc::base_offset_in_bytes(T_INT)),
910 rax);
911 }
912
913 void TemplateTable::lastore() {
914 transition(ltos, vtos);
915 __ pop_i(rbx);
916 __ pop_ptr(rdx);
917 // rax: value
918 // ebx: index
919 // rdx: array
920 index_check(rdx, rbx); // prefer index in ebx
921 __ movq(Address(rdx, rbx,
922 Address::times_8,
923 arrayOopDesc::base_offset_in_bytes(T_LONG)),
924 rax);
925 }
926
927 void TemplateTable::fastore() {
928 transition(ftos, vtos);
929 __ pop_i(rbx);
930 __ pop_ptr(rdx);
931 // xmm0: value
932 // ebx: index
933 // rdx: array
934 index_check(rdx, rbx); // prefer index in ebx
935 __ movflt(Address(rdx, rbx,
936 Address::times_4,
937 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
938 xmm0);
939 }
940
941 void TemplateTable::dastore() {
942 transition(dtos, vtos);
943 __ pop_i(rbx);
944 __ pop_ptr(rdx);
945 // xmm0: value
946 // ebx: index
947 // rdx: array
948 index_check(rdx, rbx); // prefer index in ebx
949 __ movdbl(Address(rdx, rbx,
950 Address::times_8,
951 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
952 xmm0);
953 }
954
955 void TemplateTable::aastore() {
956 Label is_null, ok_is_subtype, done;
957 transition(vtos, vtos);
958 // stack: ..., array, index, value
959 __ movptr(rax, at_tos()); // value
960 __ movl(rcx, at_tos_p1()); // index
961 __ movptr(rdx, at_tos_p2()); // array
962
963 Address element_address(rdx, rcx,
964 UseCompressedOops? Address::times_4 : Address::times_8,
965 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
966
967 index_check(rdx, rcx); // kills rbx
968 // do array store check - check for NULL value first
969 __ testptr(rax, rax);
970 __ jcc(Assembler::zero, is_null);
971
972 // Move subklass into rbx
973 __ load_klass(rbx, rax);
974 // Move superklass into rax
975 __ load_klass(rax, rdx);
976 __ movptr(rax, Address(rax,
977 ObjArrayKlass::element_klass_offset()));
978 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
979 __ lea(rdx, element_address);
980
981 // Generate subtype check. Blows rcx, rdi
982 // Superklass in rax. Subklass in rbx.
983 __ gen_subtype_check(rbx, ok_is_subtype);
984
985 // Come here on failure
986 // object is at TOS
987 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
988
989 // Come here on success
990 __ bind(ok_is_subtype);
991
992 // Get the value we will store
993 __ movptr(rax, at_tos());
994 // Now store using the appropriate barrier
995 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
996 __ jmp(done);
997
998 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
999 __ bind(is_null);
1000 __ profile_null_seen(rbx);
1001
1002 // Store a NULL
1003 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1004
1005 // Pop stack arguments
1006 __ bind(done);
1007 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1008 }
1009
1010 void TemplateTable::bastore() {
1011 transition(itos, vtos);
1012 __ pop_i(rbx);
1013 __ pop_ptr(rdx);
1014 // eax: value
1015 // ebx: index
1016 // rdx: array
1017 index_check(rdx, rbx); // prefer index in ebx
1018 __ movb(Address(rdx, rbx,
1019 Address::times_1,
1020 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1021 rax);
1022 }
1023
1024 void TemplateTable::castore() {
1025 transition(itos, vtos);
1026 __ pop_i(rbx);
1027 __ pop_ptr(rdx);
1028 // eax: value
1029 // ebx: index
1030 // rdx: array
1031 index_check(rdx, rbx); // prefer index in ebx
1032 __ movw(Address(rdx, rbx,
1033 Address::times_2,
1034 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1035 rax);
1036 }
1037
1038 void TemplateTable::sastore() {
1039 castore();
1040 }
1041
1042 void TemplateTable::istore(int n) {
1043 transition(itos, vtos);
1044 __ movl(iaddress(n), rax);
1045 }
1046
1047 void TemplateTable::lstore(int n) {
1048 transition(ltos, vtos);
1049 __ movq(laddress(n), rax);
1050 }
1051
1052 void TemplateTable::fstore(int n) {
1053 transition(ftos, vtos);
1054 __ movflt(faddress(n), xmm0);
1055 }
1056
1057 void TemplateTable::dstore(int n) {
1058 transition(dtos, vtos);
1059 __ movdbl(daddress(n), xmm0);
1060 }
1061
1062 void TemplateTable::astore(int n) {
1063 transition(vtos, vtos);
1064 __ pop_ptr(rax);
1065 __ movptr(aaddress(n), rax);
1066 }
1067
1068 void TemplateTable::pop() {
1069 transition(vtos, vtos);
1070 __ addptr(rsp, Interpreter::stackElementSize);
1071 }
1072
1073 void TemplateTable::pop2() {
1074 transition(vtos, vtos);
1075 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1076 }
1077
1078 void TemplateTable::dup() {
1079 transition(vtos, vtos);
1080 __ load_ptr(0, rax);
1081 __ push_ptr(rax);
1082 // stack: ..., a, a
1083 }
1084
1085 void TemplateTable::dup_x1() {
1086 transition(vtos, vtos);
1087 // stack: ..., a, b
1088 __ load_ptr( 0, rax); // load b
1089 __ load_ptr( 1, rcx); // load a
1090 __ store_ptr(1, rax); // store b
1091 __ store_ptr(0, rcx); // store a
1092 __ push_ptr(rax); // push b
1093 // stack: ..., b, a, b
1094 }
1095
1096 void TemplateTable::dup_x2() {
1097 transition(vtos, vtos);
1098 // stack: ..., a, b, c
1099 __ load_ptr( 0, rax); // load c
1100 __ load_ptr( 2, rcx); // load a
1101 __ store_ptr(2, rax); // store c in a
1102 __ push_ptr(rax); // push c
1103 // stack: ..., c, b, c, c
1104 __ load_ptr( 2, rax); // load b
1105 __ store_ptr(2, rcx); // store a in b
1106 // stack: ..., c, a, c, c
1107 __ store_ptr(1, rax); // store b in c
1108 // stack: ..., c, a, b, c
1109 }
1110
1111 void TemplateTable::dup2() {
1112 transition(vtos, vtos);
1113 // stack: ..., a, b
1114 __ load_ptr(1, rax); // load a
1115 __ push_ptr(rax); // push a
1116 __ load_ptr(1, rax); // load b
1117 __ push_ptr(rax); // push b
1118 // stack: ..., a, b, a, b
1119 }
1120
1121 void TemplateTable::dup2_x1() {
1122 transition(vtos, vtos);
1123 // stack: ..., a, b, c
1124 __ load_ptr( 0, rcx); // load c
1125 __ load_ptr( 1, rax); // load b
1126 __ push_ptr(rax); // push b
1127 __ push_ptr(rcx); // push c
1128 // stack: ..., a, b, c, b, c
1129 __ store_ptr(3, rcx); // store c in b
1130 // stack: ..., a, c, c, b, c
1131 __ load_ptr( 4, rcx); // load a
1132 __ store_ptr(2, rcx); // store a in 2nd c
1133 // stack: ..., a, c, a, b, c
1134 __ store_ptr(4, rax); // store b in a
1135 // stack: ..., b, c, a, b, c
1136 }
1137
1138 void TemplateTable::dup2_x2() {
1139 transition(vtos, vtos);
1140 // stack: ..., a, b, c, d
1141 __ load_ptr( 0, rcx); // load d
1142 __ load_ptr( 1, rax); // load c
1143 __ push_ptr(rax); // push c
1144 __ push_ptr(rcx); // push d
1145 // stack: ..., a, b, c, d, c, d
1146 __ load_ptr( 4, rax); // load b
1147 __ store_ptr(2, rax); // store b in d
1148 __ store_ptr(4, rcx); // store d in b
1149 // stack: ..., a, d, c, b, c, d
1150 __ load_ptr( 5, rcx); // load a
1151 __ load_ptr( 3, rax); // load c
1152 __ store_ptr(3, rcx); // store a in c
1153 __ store_ptr(5, rax); // store c in a
1154 // stack: ..., c, d, a, b, c, d
1155 }
1156
1157 void TemplateTable::swap() {
1158 transition(vtos, vtos);
1159 // stack: ..., a, b
1160 __ load_ptr( 1, rcx); // load a
1161 __ load_ptr( 0, rax); // load b
1162 __ store_ptr(0, rcx); // store a in b
1163 __ store_ptr(1, rax); // store b in a
1164 // stack: ..., b, a
1165 }
1166
1167 void TemplateTable::iop2(Operation op) {
1168 transition(itos, itos);
1169 switch (op) {
1170 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1171 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1172 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1173 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1174 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1175 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1176 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1177 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1178 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1179 default : ShouldNotReachHere();
1180 }
1181 }
1182
1183 void TemplateTable::lop2(Operation op) {
1184 transition(ltos, ltos);
1185 switch (op) {
1186 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1187 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1188 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1189 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1190 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1191 default : ShouldNotReachHere();
1192 }
1193 }
1194
1195 void TemplateTable::idiv() {
1196 transition(itos, itos);
1197 __ movl(rcx, rax);
1198 __ pop_i(rax);
1199 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1200 // they are not equal, one could do a normal division (no correction
1201 // needed), which may speed up this implementation for the common case.
1202 // (see also JVM spec., p.243 & p.271)
1203 __ corrected_idivl(rcx);
1204 }
1205
1206 void TemplateTable::irem() {
1207 transition(itos, itos);
1208 __ movl(rcx, rax);
1209 __ pop_i(rax);
1210 // Note: could xor eax and ecx and compare with (-1 ^ min_int). If
1211 // they are not equal, one could do a normal division (no correction
1212 // needed), which may speed up this implementation for the common case.
1213 // (see also JVM spec., p.243 & p.271)
1214 __ corrected_idivl(rcx);
1215 __ movl(rax, rdx);
1216 }
1217
1218 void TemplateTable::lmul() {
1219 transition(ltos, ltos);
1220 __ pop_l(rdx);
1221 __ imulq(rax, rdx);
1222 }
1223
1224 void TemplateTable::ldiv() {
1225 transition(ltos, ltos);
1226 __ mov(rcx, rax);
1227 __ pop_l(rax);
1228 // generate explicit div0 check
1229 __ testq(rcx, rcx);
1230 __ jump_cc(Assembler::zero,
1231 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1232 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1233 // they are not equal, one could do a normal division (no correction
1234 // needed), which may speed up this implementation for the common case.
1235 // (see also JVM spec., p.243 & p.271)
1236 __ corrected_idivq(rcx); // kills rbx
1237 }
1238
1239 void TemplateTable::lrem() {
1240 transition(ltos, ltos);
1241 __ mov(rcx, rax);
1242 __ pop_l(rax);
1243 __ testq(rcx, rcx);
1244 __ jump_cc(Assembler::zero,
1245 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1246 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1247 // they are not equal, one could do a normal division (no correction
1248 // needed), which may speed up this implementation for the common case.
1249 // (see also JVM spec., p.243 & p.271)
1250 __ corrected_idivq(rcx); // kills rbx
1251 __ mov(rax, rdx);
1252 }
1253
1254 void TemplateTable::lshl() {
1255 transition(itos, ltos);
1256 __ movl(rcx, rax); // get shift count
1257 __ pop_l(rax); // get shift value
1258 __ shlq(rax);
1259 }
1260
1261 void TemplateTable::lshr() {
1262 transition(itos, ltos);
1263 __ movl(rcx, rax); // get shift count
1264 __ pop_l(rax); // get shift value
1265 __ sarq(rax);
1266 }
1267
1268 void TemplateTable::lushr() {
1269 transition(itos, ltos);
1270 __ movl(rcx, rax); // get shift count
1271 __ pop_l(rax); // get shift value
1272 __ shrq(rax);
1273 }
1274
1275 void TemplateTable::fop2(Operation op) {
1276 transition(ftos, ftos);
1277 switch (op) {
1278 case add:
1279 __ addss(xmm0, at_rsp());
1280 __ addptr(rsp, Interpreter::stackElementSize);
1281 break;
1282 case sub:
1283 __ movflt(xmm1, xmm0);
1284 __ pop_f(xmm0);
1285 __ subss(xmm0, xmm1);
1286 break;
1287 case mul:
1288 __ mulss(xmm0, at_rsp());
1289 __ addptr(rsp, Interpreter::stackElementSize);
1290 break;
1291 case div:
1292 __ movflt(xmm1, xmm0);
1293 __ pop_f(xmm0);
1294 __ divss(xmm0, xmm1);
1295 break;
1296 case rem:
1297 __ movflt(xmm1, xmm0);
1298 __ pop_f(xmm0);
1299 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1300 break;
1301 default:
1302 ShouldNotReachHere();
1303 break;
1304 }
1305 }
1306
1307 void TemplateTable::dop2(Operation op) {
1308 transition(dtos, dtos);
1309 switch (op) {
1310 case add:
1311 __ addsd(xmm0, at_rsp());
1312 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1313 break;
1314 case sub:
1315 __ movdbl(xmm1, xmm0);
1316 __ pop_d(xmm0);
1317 __ subsd(xmm0, xmm1);
1318 break;
1319 case mul:
1320 __ mulsd(xmm0, at_rsp());
1321 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1322 break;
1323 case div:
1324 __ movdbl(xmm1, xmm0);
1325 __ pop_d(xmm0);
1326 __ divsd(xmm0, xmm1);
1327 break;
1328 case rem:
1329 __ movdbl(xmm1, xmm0);
1330 __ pop_d(xmm0);
1331 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1332 break;
1333 default:
1334 ShouldNotReachHere();
1335 break;
1336 }
1337 }
1338
1339 void TemplateTable::ineg() {
1340 transition(itos, itos);
1341 __ negl(rax);
1342 }
1343
1344 void TemplateTable::lneg() {
1345 transition(ltos, ltos);
1346 __ negq(rax);
1347 }
1348
1349 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1350 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1351 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1352 // of 128-bits operands for SSE instructions.
1353 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1354 // Store the value to a 128-bits operand.
1355 operand[0] = lo;
1356 operand[1] = hi;
1357 return operand;
1358 }
1359
1360 // Buffer for 128-bits masks used by SSE instructions.
1361 static jlong float_signflip_pool[2*2];
1362 static jlong double_signflip_pool[2*2];
1363
1364 void TemplateTable::fneg() {
1365 transition(ftos, ftos);
1366 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], 0x8000000080000000, 0x8000000080000000);
1367 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1368 }
1369
1370 void TemplateTable::dneg() {
1371 transition(dtos, dtos);
1372 static jlong *double_signflip = double_quadword(&double_signflip_pool[1], 0x8000000000000000, 0x8000000000000000);
1373 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1374 }
1375
1376 void TemplateTable::iinc() {
1377 transition(vtos, vtos);
1378 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1379 locals_index(rbx);
1380 __ addl(iaddress(rbx), rdx);
1381 }
1382
1383 void TemplateTable::wide_iinc() {
1384 transition(vtos, vtos);
1385 __ movl(rdx, at_bcp(4)); // get constant
1386 locals_index_wide(rbx);
1387 __ bswapl(rdx); // swap bytes & sign-extend constant
1388 __ sarl(rdx, 16);
1389 __ addl(iaddress(rbx), rdx);
1390 // Note: should probably use only one movl to get both
1391 // the index and the constant -> fix this
1392 }
1393
1394 void TemplateTable::convert() {
1395 // Checking
1396 #ifdef ASSERT
1397 {
1398 TosState tos_in = ilgl;
1399 TosState tos_out = ilgl;
1400 switch (bytecode()) {
1401 case Bytecodes::_i2l: // fall through
1402 case Bytecodes::_i2f: // fall through
1403 case Bytecodes::_i2d: // fall through
1404 case Bytecodes::_i2b: // fall through
1405 case Bytecodes::_i2c: // fall through
1406 case Bytecodes::_i2s: tos_in = itos; break;
1407 case Bytecodes::_l2i: // fall through
1408 case Bytecodes::_l2f: // fall through
1409 case Bytecodes::_l2d: tos_in = ltos; break;
1410 case Bytecodes::_f2i: // fall through
1411 case Bytecodes::_f2l: // fall through
1412 case Bytecodes::_f2d: tos_in = ftos; break;
1413 case Bytecodes::_d2i: // fall through
1414 case Bytecodes::_d2l: // fall through
1415 case Bytecodes::_d2f: tos_in = dtos; break;
1416 default : ShouldNotReachHere();
1417 }
1418 switch (bytecode()) {
1419 case Bytecodes::_l2i: // fall through
1420 case Bytecodes::_f2i: // fall through
1421 case Bytecodes::_d2i: // fall through
1422 case Bytecodes::_i2b: // fall through
1423 case Bytecodes::_i2c: // fall through
1424 case Bytecodes::_i2s: tos_out = itos; break;
1425 case Bytecodes::_i2l: // fall through
1426 case Bytecodes::_f2l: // fall through
1427 case Bytecodes::_d2l: tos_out = ltos; break;
1428 case Bytecodes::_i2f: // fall through
1429 case Bytecodes::_l2f: // fall through
1430 case Bytecodes::_d2f: tos_out = ftos; break;
1431 case Bytecodes::_i2d: // fall through
1432 case Bytecodes::_l2d: // fall through
1433 case Bytecodes::_f2d: tos_out = dtos; break;
1434 default : ShouldNotReachHere();
1435 }
1436 transition(tos_in, tos_out);
1437 }
1438 #endif // ASSERT
1439
1440 static const int64_t is_nan = 0x8000000000000000L;
1441
1442 // Conversion
1443 switch (bytecode()) {
1444 case Bytecodes::_i2l:
1445 __ movslq(rax, rax);
1446 break;
1447 case Bytecodes::_i2f:
1448 __ cvtsi2ssl(xmm0, rax);
1449 break;
1450 case Bytecodes::_i2d:
1451 __ cvtsi2sdl(xmm0, rax);
1452 break;
1453 case Bytecodes::_i2b:
1454 __ movsbl(rax, rax);
1455 break;
1456 case Bytecodes::_i2c:
1457 __ movzwl(rax, rax);
1458 break;
1459 case Bytecodes::_i2s:
1460 __ movswl(rax, rax);
1461 break;
1462 case Bytecodes::_l2i:
1463 __ movl(rax, rax);
1464 break;
1465 case Bytecodes::_l2f:
1466 __ cvtsi2ssq(xmm0, rax);
1467 break;
1468 case Bytecodes::_l2d:
1469 __ cvtsi2sdq(xmm0, rax);
1470 break;
1471 case Bytecodes::_f2i:
1472 {
1473 Label L;
1474 __ cvttss2sil(rax, xmm0);
1475 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1476 __ jcc(Assembler::notEqual, L);
1477 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1478 __ bind(L);
1479 }
1480 break;
1481 case Bytecodes::_f2l:
1482 {
1483 Label L;
1484 __ cvttss2siq(rax, xmm0);
1485 // NaN or overflow/underflow?
1486 __ cmp64(rax, ExternalAddress((address) &is_nan));
1487 __ jcc(Assembler::notEqual, L);
1488 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1489 __ bind(L);
1490 }
1491 break;
1492 case Bytecodes::_f2d:
1493 __ cvtss2sd(xmm0, xmm0);
1494 break;
1495 case Bytecodes::_d2i:
1496 {
1497 Label L;
1498 __ cvttsd2sil(rax, xmm0);
1499 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1500 __ jcc(Assembler::notEqual, L);
1501 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1502 __ bind(L);
1503 }
1504 break;
1505 case Bytecodes::_d2l:
1506 {
1507 Label L;
1508 __ cvttsd2siq(rax, xmm0);
1509 // NaN or overflow/underflow?
1510 __ cmp64(rax, ExternalAddress((address) &is_nan));
1511 __ jcc(Assembler::notEqual, L);
1512 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1513 __ bind(L);
1514 }
1515 break;
1516 case Bytecodes::_d2f:
1517 __ cvtsd2ss(xmm0, xmm0);
1518 break;
1519 default:
1520 ShouldNotReachHere();
1521 }
1522 }
1523
1524 void TemplateTable::lcmp() {
1525 transition(ltos, itos);
1526 Label done;
1527 __ pop_l(rdx);
1528 __ cmpq(rdx, rax);
1529 __ movl(rax, -1);
1530 __ jccb(Assembler::less, done);
1531 __ setb(Assembler::notEqual, rax);
1532 __ movzbl(rax, rax);
1533 __ bind(done);
1534 }
1535
1536 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1537 Label done;
1538 if (is_float) {
1539 // XXX get rid of pop here, use ... reg, mem32
1540 __ pop_f(xmm1);
1541 __ ucomiss(xmm1, xmm0);
1542 } else {
1543 // XXX get rid of pop here, use ... reg, mem64
1544 __ pop_d(xmm1);
1545 __ ucomisd(xmm1, xmm0);
1546 }
1547 if (unordered_result < 0) {
1548 __ movl(rax, -1);
1549 __ jccb(Assembler::parity, done);
1550 __ jccb(Assembler::below, done);
1551 __ setb(Assembler::notEqual, rdx);
1552 __ movzbl(rax, rdx);
1553 } else {
1554 __ movl(rax, 1);
1555 __ jccb(Assembler::parity, done);
1556 __ jccb(Assembler::above, done);
1557 __ movl(rax, 0);
1558 __ jccb(Assembler::equal, done);
1559 __ decrementl(rax);
1560 }
1561 __ bind(done);
1562 }
1563
1564 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1565 __ get_method(rcx); // rcx holds method
1566 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
1567 // holds bumped taken count
1568
1569 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1570 InvocationCounter::counter_offset();
1571 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1572 InvocationCounter::counter_offset();
1573
1574 // Load up edx with the branch displacement
1575 if (is_wide) {
1576 __ movl(rdx, at_bcp(1));
1577 } else {
1578 __ load_signed_short(rdx, at_bcp(1));
1579 }
1580 __ bswapl(rdx);
1581
1582 if (!is_wide) {
1583 __ sarl(rdx, 16);
1584 }
1585 __ movl2ptr(rdx, rdx);
1586
1587 // Handle all the JSR stuff here, then exit.
1588 // It's much shorter and cleaner than intermingling with the non-JSR
1589 // normal-branch stuff occurring below.
1590 if (is_jsr) {
1591 // Pre-load the next target bytecode into rbx
1592 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1, 0));
1593
1594 // compute return address as bci in rax
1595 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
1596 in_bytes(ConstMethod::codes_offset())));
1597 __ subptr(rax, Address(rcx, Method::const_offset()));
1598 // Adjust the bcp in r13 by the displacement in rdx
1599 __ addptr(r13, rdx);
1600 // jsr returns atos that is not an oop
1601 __ push_i(rax);
1602 __ dispatch_only(vtos);
1603 return;
1604 }
1605
1606 // Normal (non-jsr) branch handling
1607
1608 // Adjust the bcp in r13 by the displacement in rdx
1609 __ addptr(r13, rdx);
1610
1611 assert(UseLoopCounter || !UseOnStackReplacement,
1612 "on-stack-replacement requires loop counters");
1613 Label backedge_counter_overflow;
1614 Label profile_method;
1615 Label dispatch;
1616 if (UseLoopCounter) {
1617 // increment backedge counter for backward branches
1618 // rax: MDO
1619 // ebx: MDO bumped taken-count
1620 // rcx: method
1621 // rdx: target offset
1622 // r13: target bcp
1623 // r14: locals pointer
1624 __ testl(rdx, rdx); // check if forward or backward branch
1625 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1626
1627 // check if MethodCounters exists
1628 Label has_counters;
1629 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
1630 __ testptr(rax, rax);
1631 __ jcc(Assembler::notZero, has_counters);
1632 __ push(rdx);
1633 __ push(rcx);
1634 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
1635 rcx);
1636 __ pop(rcx);
1637 __ pop(rdx);
1638 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
1639 __ jcc(Assembler::zero, dispatch);
1640 __ bind(has_counters);
1641
1642 if (TieredCompilation) {
1643 Label no_mdo;
1644 int increment = InvocationCounter::count_increment;
1645 if (ProfileInterpreter) {
1646 // Are we profiling?
1647 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
1648 __ testptr(rbx, rbx);
1649 __ jccb(Assembler::zero, no_mdo);
1650 // Increment the MDO backedge counter
1651 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
1652 in_bytes(InvocationCounter::counter_offset()));
1653 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
1654 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1655 rax, false, Assembler::zero, &backedge_counter_overflow);
1656 __ jmp(dispatch);
1657 }
1658 __ bind(no_mdo);
1659 // Increment backedge counter in MethodCounters*
1660 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
1661 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
1662 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
1663 rax, false, Assembler::zero, &backedge_counter_overflow);
1664 } else { // not TieredCompilation
1665 // increment counter
1666 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
1667 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1668 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1669 __ movl(Address(rcx, be_offset), rax); // store counter
1670
1671 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1672
1673 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1674 __ addl(rax, Address(rcx, be_offset)); // add both counters
1675
1676 if (ProfileInterpreter) {
1677 // Test to see if we should create a method data oop
1678 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
1679 __ jcc(Assembler::less, dispatch);
1680
1681 // if no method data exists, go to profile method
1682 __ test_method_data_pointer(rax, profile_method);
1683
1684 if (UseOnStackReplacement) {
1685 // check for overflow against ebx which is the MDO taken count
1686 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1687 __ jcc(Assembler::below, dispatch);
1688
1689 // When ProfileInterpreter is on, the backedge_count comes
1690 // from the MethodData*, which value does not get reset on
1691 // the call to frequency_counter_overflow(). To avoid
1692 // excessive calls to the overflow routine while the method is
1693 // being compiled, add a second test to make sure the overflow
1694 // function is called only once every overflow_frequency.
1695 const int overflow_frequency = 1024;
1696 __ andl(rbx, overflow_frequency - 1);
1697 __ jcc(Assembler::zero, backedge_counter_overflow);
1698
1699 }
1700 } else {
1701 if (UseOnStackReplacement) {
1702 // check for overflow against eax, which is the sum of the
1703 // counters
1704 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1705 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1706
1707 }
1708 }
1709 }
1710 __ bind(dispatch);
1711 }
1712
1713 // Pre-load the next target bytecode into rbx
1714 __ load_unsigned_byte(rbx, Address(r13, 0));
1715
1716 // continue with the bytecode @ target
1717 // eax: return bci for jsr's, unused otherwise
1718 // ebx: target bytecode
1719 // r13: target bcp
1720 __ dispatch_only(vtos);
1721
1722 if (UseLoopCounter) {
1723 if (ProfileInterpreter) {
1724 // Out-of-line code to allocate method data oop.
1725 __ bind(profile_method);
1726 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1727 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1728 __ set_method_data_pointer_for_bcp();
1729 __ jmp(dispatch);
1730 }
1731
1732 if (UseOnStackReplacement) {
1733 // invocation counter overflow
1734 __ bind(backedge_counter_overflow);
1735 __ negptr(rdx);
1736 __ addptr(rdx, r13); // branch bcp
1737 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1738 __ call_VM(noreg,
1739 CAST_FROM_FN_PTR(address,
1740 InterpreterRuntime::frequency_counter_overflow),
1741 rdx);
1742 __ load_unsigned_byte(rbx, Address(r13, 0)); // restore target bytecode
1743
1744 // rax: osr nmethod (osr ok) or NULL (osr not possible)
1745 // ebx: target bytecode
1746 // rdx: scratch
1747 // r14: locals pointer
1748 // r13: bcp
1749 __ testptr(rax, rax); // test result
1750 __ jcc(Assembler::zero, dispatch); // no osr if null
1751 // nmethod may have been invalidated (VM may block upon call_VM return)
1752 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
1753 __ jcc(Assembler::notEqual, dispatch);
1754
1755 // We have the address of an on stack replacement routine in eax
1756 // We need to prepare to execute the OSR method. First we must
1757 // migrate the locals and monitors off of the stack.
1758
1759 __ mov(r13, rax); // save the nmethod
1760
1761 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1762
1763 // eax is OSR buffer, move it to expected parameter location
1764 __ mov(j_rarg0, rax);
1765
1766 // We use j_rarg definitions here so that registers don't conflict as parameter
1767 // registers change across platforms as we are in the midst of a calling
1768 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
1769
1770 const Register retaddr = j_rarg2;
1771 const Register sender_sp = j_rarg1;
1772
1773 // pop the interpreter frame
1774 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1775 __ leave(); // remove frame anchor
1776 __ pop(retaddr); // get return address
1777 __ mov(rsp, sender_sp); // set sp to sender sp
1778 // Ensure compiled code always sees stack at proper alignment
1779 __ andptr(rsp, -(StackAlignmentInBytes));
1780
1781 // unlike x86 we need no specialized return from compiled code
1782 // to the interpreter or the call stub.
1783
1784 // push the return address
1785 __ push(retaddr);
1786
1787 // and begin the OSR nmethod
1788 __ jmp(Address(r13, nmethod::osr_entry_point_offset()));
1789 }
1790 }
1791 }
1792
1793
1794 void TemplateTable::if_0cmp(Condition cc) {
1795 transition(itos, vtos);
1796 // assume branch is more often taken than not (loops use backward branches)
1797 Label not_taken;
1798 __ testl(rax, rax);
1799 __ jcc(j_not(cc), not_taken);
1800 branch(false, false);
1801 __ bind(not_taken);
1802 __ profile_not_taken_branch(rax);
1803 }
1804
1805 void TemplateTable::if_icmp(Condition cc) {
1806 transition(itos, vtos);
1807 // assume branch is more often taken than not (loops use backward branches)
1808 Label not_taken;
1809 __ pop_i(rdx);
1810 __ cmpl(rdx, rax);
1811 __ jcc(j_not(cc), not_taken);
1812 branch(false, false);
1813 __ bind(not_taken);
1814 __ profile_not_taken_branch(rax);
1815 }
1816
1817 void TemplateTable::if_nullcmp(Condition cc) {
1818 transition(atos, vtos);
1819 // assume branch is more often taken than not (loops use backward branches)
1820 Label not_taken;
1821 __ testptr(rax, rax);
1822 __ jcc(j_not(cc), not_taken);
1823 branch(false, false);
1824 __ bind(not_taken);
1825 __ profile_not_taken_branch(rax);
1826 }
1827
1828 void TemplateTable::if_acmp(Condition cc) {
1829 transition(atos, vtos);
1830 // assume branch is more often taken than not (loops use backward branches)
1831 Label not_taken;
1832 __ pop_ptr(rdx);
1833 __ cmpptr(rdx, rax);
1834 __ jcc(j_not(cc), not_taken);
1835 branch(false, false);
1836 __ bind(not_taken);
1837 __ profile_not_taken_branch(rax);
1838 }
1839
1840 void TemplateTable::ret() {
1841 transition(vtos, vtos);
1842 locals_index(rbx);
1843 __ movslq(rbx, iaddress(rbx)); // get return bci, compute return bcp
1844 __ profile_ret(rbx, rcx);
1845 __ get_method(rax);
1846 __ movptr(r13, Address(rax, Method::const_offset()));
1847 __ lea(r13, Address(r13, rbx, Address::times_1,
1848 ConstMethod::codes_offset()));
1849 __ dispatch_next(vtos);
1850 }
1851
1852 void TemplateTable::wide_ret() {
1853 transition(vtos, vtos);
1854 locals_index_wide(rbx);
1855 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
1856 __ profile_ret(rbx, rcx);
1857 __ get_method(rax);
1858 __ movptr(r13, Address(rax, Method::const_offset()));
1859 __ lea(r13, Address(r13, rbx, Address::times_1, ConstMethod::codes_offset()));
1860 __ dispatch_next(vtos);
1861 }
1862
1863 void TemplateTable::tableswitch() {
1864 Label default_case, continue_execution;
1865 transition(itos, vtos);
1866 // align r13
1867 __ lea(rbx, at_bcp(BytesPerInt));
1868 __ andptr(rbx, -BytesPerInt);
1869 // load lo & hi
1870 __ movl(rcx, Address(rbx, BytesPerInt));
1871 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
1872 __ bswapl(rcx);
1873 __ bswapl(rdx);
1874 // check against lo & hi
1875 __ cmpl(rax, rcx);
1876 __ jcc(Assembler::less, default_case);
1877 __ cmpl(rax, rdx);
1878 __ jcc(Assembler::greater, default_case);
1879 // lookup dispatch offset
1880 __ subl(rax, rcx);
1881 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1882 __ profile_switch_case(rax, rbx, rcx);
1883 // continue execution
1884 __ bind(continue_execution);
1885 __ bswapl(rdx);
1886 __ movl2ptr(rdx, rdx);
1887 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1888 __ addptr(r13, rdx);
1889 __ dispatch_only(vtos);
1890 // handle default
1891 __ bind(default_case);
1892 __ profile_switch_default(rax);
1893 __ movl(rdx, Address(rbx, 0));
1894 __ jmp(continue_execution);
1895 }
1896
1897 void TemplateTable::lookupswitch() {
1898 transition(itos, itos);
1899 __ stop("lookupswitch bytecode should have been rewritten");
1900 }
1901
1902 void TemplateTable::fast_linearswitch() {
1903 transition(itos, vtos);
1904 Label loop_entry, loop, found, continue_execution;
1905 // bswap rax so we can avoid bswapping the table entries
1906 __ bswapl(rax);
1907 // align r13
1908 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
1909 // this instruction (change offsets
1910 // below)
1911 __ andptr(rbx, -BytesPerInt);
1912 // set counter
1913 __ movl(rcx, Address(rbx, BytesPerInt));
1914 __ bswapl(rcx);
1915 __ jmpb(loop_entry);
1916 // table search
1917 __ bind(loop);
1918 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
1919 __ jcc(Assembler::equal, found);
1920 __ bind(loop_entry);
1921 __ decrementl(rcx);
1922 __ jcc(Assembler::greaterEqual, loop);
1923 // default case
1924 __ profile_switch_default(rax);
1925 __ movl(rdx, Address(rbx, 0));
1926 __ jmp(continue_execution);
1927 // entry found -> get offset
1928 __ bind(found);
1929 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
1930 __ profile_switch_case(rcx, rax, rbx);
1931 // continue execution
1932 __ bind(continue_execution);
1933 __ bswapl(rdx);
1934 __ movl2ptr(rdx, rdx);
1935 __ load_unsigned_byte(rbx, Address(r13, rdx, Address::times_1));
1936 __ addptr(r13, rdx);
1937 __ dispatch_only(vtos);
1938 }
1939
1940 void TemplateTable::fast_binaryswitch() {
1941 transition(itos, vtos);
1942 // Implementation using the following core algorithm:
1943 //
1944 // int binary_search(int key, LookupswitchPair* array, int n) {
1945 // // Binary search according to "Methodik des Programmierens" by
1946 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1947 // int i = 0;
1948 // int j = n;
1949 // while (i+1 < j) {
1950 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1951 // // with Q: for all i: 0 <= i < n: key < a[i]
1952 // // where a stands for the array and assuming that the (inexisting)
1953 // // element a[n] is infinitely big.
1954 // int h = (i + j) >> 1;
1955 // // i < h < j
1956 // if (key < array[h].fast_match()) {
1957 // j = h;
1958 // } else {
1959 // i = h;
1960 // }
1961 // }
1962 // // R: a[i] <= key < a[i+1] or Q
1963 // // (i.e., if key is within array, i is the correct index)
1964 // return i;
1965 // }
1966
1967 // Register allocation
1968 const Register key = rax; // already set (tosca)
1969 const Register array = rbx;
1970 const Register i = rcx;
1971 const Register j = rdx;
1972 const Register h = rdi;
1973 const Register temp = rsi;
1974
1975 // Find array start
1976 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
1977 // get rid of this
1978 // instruction (change
1979 // offsets below)
1980 __ andptr(array, -BytesPerInt);
1981
1982 // Initialize i & j
1983 __ xorl(i, i); // i = 0;
1984 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
1985
1986 // Convert j into native byteordering
1987 __ bswapl(j);
1988
1989 // And start
1990 Label entry;
1991 __ jmp(entry);
1992
1993 // binary search loop
1994 {
1995 Label loop;
1996 __ bind(loop);
1997 // int h = (i + j) >> 1;
1998 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1999 __ sarl(h, 1); // h = (i + j) >> 1;
2000 // if (key < array[h].fast_match()) {
2001 // j = h;
2002 // } else {
2003 // i = h;
2004 // }
2005 // Convert array[h].match to native byte-ordering before compare
2006 __ movl(temp, Address(array, h, Address::times_8));
2007 __ bswapl(temp);
2008 __ cmpl(key, temp);
2009 // j = h if (key < array[h].fast_match())
2010 __ cmovl(Assembler::less, j, h);
2011 // i = h if (key >= array[h].fast_match())
2012 __ cmovl(Assembler::greaterEqual, i, h);
2013 // while (i+1 < j)
2014 __ bind(entry);
2015 __ leal(h, Address(i, 1)); // i+1
2016 __ cmpl(h, j); // i+1 < j
2017 __ jcc(Assembler::less, loop);
2018 }
2019
2020 // end of binary search, result index is i (must check again!)
2021 Label default_case;
2022 // Convert array[i].match to native byte-ordering before compare
2023 __ movl(temp, Address(array, i, Address::times_8));
2024 __ bswapl(temp);
2025 __ cmpl(key, temp);
2026 __ jcc(Assembler::notEqual, default_case);
2027
2028 // entry found -> j = offset
2029 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2030 __ profile_switch_case(i, key, array);
2031 __ bswapl(j);
2032 __ movl2ptr(j, j);
2033 __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
2034 __ addptr(r13, j);
2035 __ dispatch_only(vtos);
2036
2037 // default case -> j = default offset
2038 __ bind(default_case);
2039 __ profile_switch_default(i);
2040 __ movl(j, Address(array, -2 * BytesPerInt));
2041 __ bswapl(j);
2042 __ movl2ptr(j, j);
2043 __ load_unsigned_byte(rbx, Address(r13, j, Address::times_1));
2044 __ addptr(r13, j);
2045 __ dispatch_only(vtos);
2046 }
2047
2048
2049 void TemplateTable::_return(TosState state) {
2050 transition(state, state);
2051 assert(_desc->calls_vm(),
2052 "inconsistent calls_vm information"); // call in remove_activation
2053
2054 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2055 assert(state == vtos, "only valid state");
2056 __ movptr(c_rarg1, aaddress(0));
2057 __ load_klass(rdi, c_rarg1);
2058 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2059 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2060 Label skip_register_finalizer;
2061 __ jcc(Assembler::zero, skip_register_finalizer);
2062
2063 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2064
2065 __ bind(skip_register_finalizer);
2066 }
2067
2068 __ remove_activation(state, r13);
2069 __ jmp(r13);
2070 }
2071
2072 // ----------------------------------------------------------------------------
2073 // Volatile variables demand their effects be made known to all CPU's
2074 // in order. Store buffers on most chips allow reads & writes to
2075 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2076 // without some kind of memory barrier (i.e., it's not sufficient that
2077 // the interpreter does not reorder volatile references, the hardware
2078 // also must not reorder them).
2079 //
2080 // According to the new Java Memory Model (JMM):
2081 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2082 // writes act as aquire & release, so:
2083 // (2) A read cannot let unrelated NON-volatile memory refs that
2084 // happen after the read float up to before the read. It's OK for
2085 // non-volatile memory refs that happen before the volatile read to
2086 // float down below it.
2087 // (3) Similar a volatile write cannot let unrelated NON-volatile
2088 // memory refs that happen BEFORE the write float down to after the
2089 // write. It's OK for non-volatile memory refs that happen after the
2090 // volatile write to float up before it.
2091 //
2092 // We only put in barriers around volatile refs (they are expensive),
2093 // not _between_ memory refs (that would require us to track the
2094 // flavor of the previous memory refs). Requirements (2) and (3)
2095 // require some barriers before volatile stores and after volatile
2096 // loads. These nearly cover requirement (1) but miss the
2097 // volatile-store-volatile-load case. This final case is placed after
2098 // volatile-stores although it could just as well go before
2099 // volatile-loads.
2100 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits
2101 order_constraint) {
2102 // Helper function to insert a is-volatile test and memory barrier
2103 if (os::is_MP()) { // Not needed on single CPU
2104 __ membar(order_constraint);
2105 }
2106 }
2107
2108 void TemplateTable::resolve_cache_and_index(int byte_no,
2109 Register Rcache,
2110 Register index,
2111 size_t index_size) {
2112 const Register temp = rbx;
2113 assert_different_registers(Rcache, index, temp);
2114
2115 Label resolved;
2116 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2117 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2118 __ cmpl(temp, (int) bytecode()); // have we resolved this bytecode?
2119 __ jcc(Assembler::equal, resolved);
2120
2121 // resolve first time through
2122 address entry;
2123 switch (bytecode()) {
2124 case Bytecodes::_getstatic:
2125 case Bytecodes::_putstatic:
2126 case Bytecodes::_getfield:
2127 case Bytecodes::_putfield:
2128 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
2129 break;
2130 case Bytecodes::_invokevirtual:
2131 case Bytecodes::_invokespecial:
2132 case Bytecodes::_invokestatic:
2133 case Bytecodes::_invokeinterface:
2134 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
2135 break;
2136 case Bytecodes::_invokehandle:
2137 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);
2138 break;
2139 case Bytecodes::_invokedynamic:
2140 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
2141 break;
2142 default:
2143 fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2144 break;
2145 }
2146 __ movl(temp, (int) bytecode());
2147 __ call_VM(noreg, entry, temp);
2148
2149 // Update registers with resolved info
2150 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2151 __ bind(resolved);
2152 }
2153
2154 // The cache and index registers must be set before call
2155 void TemplateTable::load_field_cp_cache_entry(Register obj,
2156 Register cache,
2157 Register index,
2158 Register off,
2159 Register flags,
2160 bool is_static = false) {
2161 assert_different_registers(cache, index, flags, off);
2162
2163 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2164 // Field offset
2165 __ movptr(off, Address(cache, index, Address::times_ptr,
2166 in_bytes(cp_base_offset +
2167 ConstantPoolCacheEntry::f2_offset())));
2168 // Flags
2169 __ movl(flags, Address(cache, index, Address::times_ptr,
2170 in_bytes(cp_base_offset +
2171 ConstantPoolCacheEntry::flags_offset())));
2172
2173 // klass overwrite register
2174 if (is_static) {
2175 __ movptr(obj, Address(cache, index, Address::times_ptr,
2176 in_bytes(cp_base_offset +
2177 ConstantPoolCacheEntry::f1_offset())));
2178 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2179 __ movptr(obj, Address(obj, mirror_offset));
2180 }
2181 }
2182
2183 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2184 Register method,
2185 Register itable_index,
2186 Register flags,
2187 bool is_invokevirtual,
2188 bool is_invokevfinal, /*unused*/
2189 bool is_invokedynamic) {
2190 // setup registers
2191 const Register cache = rcx;
2192 const Register index = rdx;
2193 assert_different_registers(method, flags);
2194 assert_different_registers(method, cache, index);
2195 assert_different_registers(itable_index, flags);
2196 assert_different_registers(itable_index, cache, index);
2197 // determine constant pool cache field offsets
2198 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2199 const int method_offset = in_bytes(
2200 ConstantPoolCache::base_offset() +
2201 ((byte_no == f2_byte)
2202 ? ConstantPoolCacheEntry::f2_offset()
2203 : ConstantPoolCacheEntry::f1_offset()));
2204 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2205 ConstantPoolCacheEntry::flags_offset());
2206 // access constant pool cache fields
2207 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2208 ConstantPoolCacheEntry::f2_offset());
2209
2210 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2211 resolve_cache_and_index(byte_no, cache, index, index_size);
2212 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2213
2214 if (itable_index != noreg) {
2215 // pick up itable or appendix index from f2 also:
2216 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2217 }
2218 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2219 }
2220
2221 // Correct values of the cache and index registers are preserved.
2222 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2223 bool is_static, bool has_tos) {
2224 // do the JVMTI work here to avoid disturbing the register state below
2225 // We use c_rarg registers here because we want to use the register used in
2226 // the call to the VM
2227 if (JvmtiExport::can_post_field_access()) {
2228 // Check to see if a field access watch has been set before we
2229 // take the time to call into the VM.
2230 Label L1;
2231 assert_different_registers(cache, index, rax);
2232 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2233 __ testl(rax, rax);
2234 __ jcc(Assembler::zero, L1);
2235
2236 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2237
2238 // cache entry pointer
2239 __ addptr(c_rarg2, in_bytes(ConstantPoolCache::base_offset()));
2240 __ shll(c_rarg3, LogBytesPerWord);
2241 __ addptr(c_rarg2, c_rarg3);
2242 if (is_static) {
2243 __ xorl(c_rarg1, c_rarg1); // NULL object reference
2244 } else {
2245 __ movptr(c_rarg1, at_tos()); // get object pointer without popping it
2246 __ verify_oop(c_rarg1);
2247 }
2248 // c_rarg1: object pointer or NULL
2249 // c_rarg2: cache entry pointer
2250 // c_rarg3: jvalue object on the stack
2251 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2252 InterpreterRuntime::post_field_access),
2253 c_rarg1, c_rarg2, c_rarg3);
2254 __ get_cache_and_index_at_bcp(cache, index, 1);
2255 __ bind(L1);
2256 }
2257 }
2258
2259 void TemplateTable::pop_and_check_object(Register r) {
2260 __ pop_ptr(r);
2261 __ null_check(r); // for field access must check obj.
2262 __ verify_oop(r);
2263 }
2264
2265 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2266 transition(vtos, vtos);
2267
2268 const Register cache = rcx;
2269 const Register index = rdx;
2270 const Register obj = c_rarg3;
2271 const Register off = rbx;
2272 const Register flags = rax;
2273 const Register bc = c_rarg3; // uses same reg as obj, so don't mix them
2274
2275 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2276 jvmti_post_field_access(cache, index, is_static, false);
2277 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2278
2279 if (!is_static) {
2280 // obj is on the stack
2281 pop_and_check_object(obj);
2282 }
2283
2284 const Address field(obj, off, Address::times_1);
2285
2286 Label Done, notByte, notInt, notShort, notChar,
2287 notLong, notFloat, notObj, notDouble;
2288
2289 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2290 // Make sure we don't need to mask edx after the above shift
2291 assert(btos == 0, "change code, btos != 0");
2292
2293 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2294 __ jcc(Assembler::notZero, notByte);
2295 // btos
2296 __ load_signed_byte(rax, field);
2297 __ push(btos);
2298 // Rewrite bytecode to be faster
2299 if (!is_static) {
2300 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2301 }
2302 __ jmp(Done);
2303
2304 __ bind(notByte);
2305 __ cmpl(flags, atos);
2306 __ jcc(Assembler::notEqual, notObj);
2307 // atos
2308 __ load_heap_oop(rax, field);
2309 __ push(atos);
2310 if (!is_static) {
2311 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2312 }
2313 __ jmp(Done);
2314
2315 __ bind(notObj);
2316 __ cmpl(flags, itos);
2317 __ jcc(Assembler::notEqual, notInt);
2318 // itos
2319 __ movl(rax, field);
2320 __ push(itos);
2321 // Rewrite bytecode to be faster
2322 if (!is_static) {
2323 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2324 }
2325 __ jmp(Done);
2326
2327 __ bind(notInt);
2328 __ cmpl(flags, ctos);
2329 __ jcc(Assembler::notEqual, notChar);
2330 // ctos
2331 __ load_unsigned_short(rax, field);
2332 __ push(ctos);
2333 // Rewrite bytecode to be faster
2334 if (!is_static) {
2335 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2336 }
2337 __ jmp(Done);
2338
2339 __ bind(notChar);
2340 __ cmpl(flags, stos);
2341 __ jcc(Assembler::notEqual, notShort);
2342 // stos
2343 __ load_signed_short(rax, field);
2344 __ push(stos);
2345 // Rewrite bytecode to be faster
2346 if (!is_static) {
2347 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2348 }
2349 __ jmp(Done);
2350
2351 __ bind(notShort);
2352 __ cmpl(flags, ltos);
2353 __ jcc(Assembler::notEqual, notLong);
2354 // ltos
2355 __ movq(rax, field);
2356 __ push(ltos);
2357 // Rewrite bytecode to be faster
2358 if (!is_static) {
2359 patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx);
2360 }
2361 __ jmp(Done);
2362
2363 __ bind(notLong);
2364 __ cmpl(flags, ftos);
2365 __ jcc(Assembler::notEqual, notFloat);
2366 // ftos
2367 __ movflt(xmm0, field);
2368 __ push(ftos);
2369 // Rewrite bytecode to be faster
2370 if (!is_static) {
2371 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2372 }
2373 __ jmp(Done);
2374
2375 __ bind(notFloat);
2376 #ifdef ASSERT
2377 __ cmpl(flags, dtos);
2378 __ jcc(Assembler::notEqual, notDouble);
2379 #endif
2380 // dtos
2381 __ movdbl(xmm0, field);
2382 __ push(dtos);
2383 // Rewrite bytecode to be faster
2384 if (!is_static) {
2385 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2386 }
2387 #ifdef ASSERT
2388 __ jmp(Done);
2389
2390 __ bind(notDouble);
2391 __ stop("Bad state");
2392 #endif
2393
2394 __ bind(Done);
2395 // [jk] not needed currently
2396 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2397 // Assembler::LoadStore));
2398 }
2399
2400
2401 void TemplateTable::getfield(int byte_no) {
2402 getfield_or_static(byte_no, false);
2403 }
2404
2405 void TemplateTable::getstatic(int byte_no) {
2406 getfield_or_static(byte_no, true);
2407 }
2408
2409 // The registers cache and index expected to be set before call.
2410 // The function may destroy various registers, just not the cache and index registers.
2411 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2412 transition(vtos, vtos);
2413
2414 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2415
2416 if (JvmtiExport::can_post_field_modification()) {
2417 // Check to see if a field modification watch has been set before
2418 // we take the time to call into the VM.
2419 Label L1;
2420 assert_different_registers(cache, index, rax);
2421 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2422 __ testl(rax, rax);
2423 __ jcc(Assembler::zero, L1);
2424
2425 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2426
2427 if (is_static) {
2428 // Life is simple. Null out the object pointer.
2429 __ xorl(c_rarg1, c_rarg1);
2430 } else {
2431 // Life is harder. The stack holds the value on top, followed by
2432 // the object. We don't know the size of the value, though; it
2433 // could be one or two words depending on its type. As a result,
2434 // we must find the type to determine where the object is.
2435 __ movl(c_rarg3, Address(c_rarg2, rscratch1,
2436 Address::times_8,
2437 in_bytes(cp_base_offset +
2438 ConstantPoolCacheEntry::flags_offset())));
2439 __ shrl(c_rarg3, ConstantPoolCacheEntry::tos_state_shift);
2440 // Make sure we don't need to mask rcx after the above shift
2441 ConstantPoolCacheEntry::verify_tos_state_shift();
2442 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2443 __ cmpl(c_rarg3, ltos);
2444 __ cmovptr(Assembler::equal,
2445 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2446 __ cmpl(c_rarg3, dtos);
2447 __ cmovptr(Assembler::equal,
2448 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2449 }
2450 // cache entry pointer
2451 __ addptr(c_rarg2, in_bytes(cp_base_offset));
2452 __ shll(rscratch1, LogBytesPerWord);
2453 __ addptr(c_rarg2, rscratch1);
2454 // object (tos)
2455 __ mov(c_rarg3, rsp);
2456 // c_rarg1: object pointer set up above (NULL if static)
2457 // c_rarg2: cache entry pointer
2458 // c_rarg3: jvalue object on the stack
2459 __ call_VM(noreg,
2460 CAST_FROM_FN_PTR(address,
2461 InterpreterRuntime::post_field_modification),
2462 c_rarg1, c_rarg2, c_rarg3);
2463 __ get_cache_and_index_at_bcp(cache, index, 1);
2464 __ bind(L1);
2465 }
2466 }
2467
2468 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2469 transition(vtos, vtos);
2470
2471 const Register cache = rcx;
2472 const Register index = rdx;
2473 const Register obj = rcx;
2474 const Register off = rbx;
2475 const Register flags = rax;
2476 const Register bc = c_rarg3;
2477
2478 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2479 jvmti_post_field_mod(cache, index, is_static);
2480 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2481
2482 // [jk] not needed currently
2483 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2484 // Assembler::StoreStore));
2485
2486 Label notVolatile, Done;
2487 __ movl(rdx, flags);
2488 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2489 __ andl(rdx, 0x1);
2490
2491 // field address
2492 const Address field(obj, off, Address::times_1);
2493
2494 Label notByte, notInt, notShort, notChar,
2495 notLong, notFloat, notObj, notDouble;
2496
2497 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2498
2499 assert(btos == 0, "change code, btos != 0");
2500 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2501 __ jcc(Assembler::notZero, notByte);
2502
2503 // btos
2504 {
2505 __ pop(btos);
2506 if (!is_static) pop_and_check_object(obj);
2507 __ movb(field, rax);
2508 if (!is_static) {
2509 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
2510 }
2511 __ jmp(Done);
2512 }
2513
2514 __ bind(notByte);
2515 __ cmpl(flags, atos);
2516 __ jcc(Assembler::notEqual, notObj);
2517
2518 // atos
2519 {
2520 __ pop(atos);
2521 if (!is_static) pop_and_check_object(obj);
2522 // Store into the field
2523 do_oop_store(_masm, field, rax, _bs->kind(), false);
2524 if (!is_static) {
2525 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
2526 }
2527 __ jmp(Done);
2528 }
2529
2530 __ bind(notObj);
2531 __ cmpl(flags, itos);
2532 __ jcc(Assembler::notEqual, notInt);
2533
2534 // itos
2535 {
2536 __ pop(itos);
2537 if (!is_static) pop_and_check_object(obj);
2538 __ movl(field, rax);
2539 if (!is_static) {
2540 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
2541 }
2542 __ jmp(Done);
2543 }
2544
2545 __ bind(notInt);
2546 __ cmpl(flags, ctos);
2547 __ jcc(Assembler::notEqual, notChar);
2548
2549 // ctos
2550 {
2551 __ pop(ctos);
2552 if (!is_static) pop_and_check_object(obj);
2553 __ movw(field, rax);
2554 if (!is_static) {
2555 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
2556 }
2557 __ jmp(Done);
2558 }
2559
2560 __ bind(notChar);
2561 __ cmpl(flags, stos);
2562 __ jcc(Assembler::notEqual, notShort);
2563
2564 // stos
2565 {
2566 __ pop(stos);
2567 if (!is_static) pop_and_check_object(obj);
2568 __ movw(field, rax);
2569 if (!is_static) {
2570 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
2571 }
2572 __ jmp(Done);
2573 }
2574
2575 __ bind(notShort);
2576 __ cmpl(flags, ltos);
2577 __ jcc(Assembler::notEqual, notLong);
2578
2579 // ltos
2580 {
2581 __ pop(ltos);
2582 if (!is_static) pop_and_check_object(obj);
2583 __ movq(field, rax);
2584 if (!is_static) {
2585 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
2586 }
2587 __ jmp(Done);
2588 }
2589
2590 __ bind(notLong);
2591 __ cmpl(flags, ftos);
2592 __ jcc(Assembler::notEqual, notFloat);
2593
2594 // ftos
2595 {
2596 __ pop(ftos);
2597 if (!is_static) pop_and_check_object(obj);
2598 __ movflt(field, xmm0);
2599 if (!is_static) {
2600 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
2601 }
2602 __ jmp(Done);
2603 }
2604
2605 __ bind(notFloat);
2606 #ifdef ASSERT
2607 __ cmpl(flags, dtos);
2608 __ jcc(Assembler::notEqual, notDouble);
2609 #endif
2610
2611 // dtos
2612 {
2613 __ pop(dtos);
2614 if (!is_static) pop_and_check_object(obj);
2615 __ movdbl(field, xmm0);
2616 if (!is_static) {
2617 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
2618 }
2619 }
2620
2621 #ifdef ASSERT
2622 __ jmp(Done);
2623
2624 __ bind(notDouble);
2625 __ stop("Bad state");
2626 #endif
2627
2628 __ bind(Done);
2629
2630 // Check for volatile store
2631 __ testl(rdx, rdx);
2632 __ jcc(Assembler::zero, notVolatile);
2633 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2634 Assembler::StoreStore));
2635 __ bind(notVolatile);
2636 }
2637
2638 void TemplateTable::putfield(int byte_no) {
2639 putfield_or_static(byte_no, false);
2640 }
2641
2642 void TemplateTable::putstatic(int byte_no) {
2643 putfield_or_static(byte_no, true);
2644 }
2645
2646 void TemplateTable::jvmti_post_fast_field_mod() {
2647 if (JvmtiExport::can_post_field_modification()) {
2648 // Check to see if a field modification watch has been set before
2649 // we take the time to call into the VM.
2650 Label L2;
2651 __ mov32(c_rarg3, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2652 __ testl(c_rarg3, c_rarg3);
2653 __ jcc(Assembler::zero, L2);
2654 __ pop_ptr(rbx); // copy the object pointer from tos
2655 __ verify_oop(rbx);
2656 __ push_ptr(rbx); // put the object pointer back on tos
2657 // Save tos values before call_VM() clobbers them. Since we have
2658 // to do it for every data type, we use the saved values as the
2659 // jvalue object.
2660 switch (bytecode()) { // load values into the jvalue object
2661 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
2662 case Bytecodes::_fast_bputfield: // fall through
2663 case Bytecodes::_fast_sputfield: // fall through
2664 case Bytecodes::_fast_cputfield: // fall through
2665 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
2666 case Bytecodes::_fast_dputfield: __ push_d(); break;
2667 case Bytecodes::_fast_fputfield: __ push_f(); break;
2668 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
2669
2670 default:
2671 ShouldNotReachHere();
2672 }
2673 __ mov(c_rarg3, rsp); // points to jvalue on the stack
2674 // access constant pool cache entry
2675 __ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1);
2676 __ verify_oop(rbx);
2677 // rbx: object pointer copied above
2678 // c_rarg2: cache entry pointer
2679 // c_rarg3: jvalue object on the stack
2680 __ call_VM(noreg,
2681 CAST_FROM_FN_PTR(address,
2682 InterpreterRuntime::post_field_modification),
2683 rbx, c_rarg2, c_rarg3);
2684
2685 switch (bytecode()) { // restore tos values
2686 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
2687 case Bytecodes::_fast_bputfield: // fall through
2688 case Bytecodes::_fast_sputfield: // fall through
2689 case Bytecodes::_fast_cputfield: // fall through
2690 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
2691 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2692 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2693 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
2694 }
2695 __ bind(L2);
2696 }
2697 }
2698
2699 void TemplateTable::fast_storefield(TosState state) {
2700 transition(state, vtos);
2701
2702 ByteSize base = ConstantPoolCache::base_offset();
2703
2704 jvmti_post_fast_field_mod();
2705
2706 // access constant pool cache
2707 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2708
2709 // test for volatile with rdx
2710 __ movl(rdx, Address(rcx, rbx, Address::times_8,
2711 in_bytes(base +
2712 ConstantPoolCacheEntry::flags_offset())));
2713
2714 // replace index with field offset from cache entry
2715 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2716 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2717
2718 // [jk] not needed currently
2719 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2720 // Assembler::StoreStore));
2721
2722 Label notVolatile;
2723 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2724 __ andl(rdx, 0x1);
2725
2726 // Get object from stack
2727 pop_and_check_object(rcx);
2728
2729 // field address
2730 const Address field(rcx, rbx, Address::times_1);
2731
2732 // access field
2733 switch (bytecode()) {
2734 case Bytecodes::_fast_aputfield:
2735 do_oop_store(_masm, field, rax, _bs->kind(), false);
2736 break;
2737 case Bytecodes::_fast_lputfield:
2738 __ movq(field, rax);
2739 break;
2740 case Bytecodes::_fast_iputfield:
2741 __ movl(field, rax);
2742 break;
2743 case Bytecodes::_fast_bputfield:
2744 __ movb(field, rax);
2745 break;
2746 case Bytecodes::_fast_sputfield:
2747 // fall through
2748 case Bytecodes::_fast_cputfield:
2749 __ movw(field, rax);
2750 break;
2751 case Bytecodes::_fast_fputfield:
2752 __ movflt(field, xmm0);
2753 break;
2754 case Bytecodes::_fast_dputfield:
2755 __ movdbl(field, xmm0);
2756 break;
2757 default:
2758 ShouldNotReachHere();
2759 }
2760
2761 // Check for volatile store
2762 __ testl(rdx, rdx);
2763 __ jcc(Assembler::zero, notVolatile);
2764 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2765 Assembler::StoreStore));
2766 __ bind(notVolatile);
2767 }
2768
2769
2770 void TemplateTable::fast_accessfield(TosState state) {
2771 transition(atos, state);
2772
2773 // Do the JVMTI work here to avoid disturbing the register state below
2774 if (JvmtiExport::can_post_field_access()) {
2775 // Check to see if a field access watch has been set before we
2776 // take the time to call into the VM.
2777 Label L1;
2778 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2779 __ testl(rcx, rcx);
2780 __ jcc(Assembler::zero, L1);
2781 // access constant pool cache entry
2782 __ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1);
2783 __ verify_oop(rax);
2784 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
2785 __ mov(c_rarg1, rax);
2786 // c_rarg1: object pointer copied above
2787 // c_rarg2: cache entry pointer
2788 __ call_VM(noreg,
2789 CAST_FROM_FN_PTR(address,
2790 InterpreterRuntime::post_field_access),
2791 c_rarg1, c_rarg2);
2792 __ pop_ptr(rax); // restore object pointer
2793 __ bind(L1);
2794 }
2795
2796 // access constant pool cache
2797 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2798 // replace index with field offset from cache entry
2799 // [jk] not needed currently
2800 // if (os::is_MP()) {
2801 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
2802 // in_bytes(ConstantPoolCache::base_offset() +
2803 // ConstantPoolCacheEntry::flags_offset())));
2804 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2805 // __ andl(rdx, 0x1);
2806 // }
2807 __ movptr(rbx, Address(rcx, rbx, Address::times_8,
2808 in_bytes(ConstantPoolCache::base_offset() +
2809 ConstantPoolCacheEntry::f2_offset())));
2810
2811 // rax: object
2812 __ verify_oop(rax);
2813 __ null_check(rax);
2814 Address field(rax, rbx, Address::times_1);
2815
2816 // access field
2817 switch (bytecode()) {
2818 case Bytecodes::_fast_agetfield:
2819 __ load_heap_oop(rax, field);
2820 __ verify_oop(rax);
2821 break;
2822 case Bytecodes::_fast_lgetfield:
2823 __ movq(rax, field);
2824 break;
2825 case Bytecodes::_fast_igetfield:
2826 __ movl(rax, field);
2827 break;
2828 case Bytecodes::_fast_bgetfield:
2829 __ movsbl(rax, field);
2830 break;
2831 case Bytecodes::_fast_sgetfield:
2832 __ load_signed_short(rax, field);
2833 break;
2834 case Bytecodes::_fast_cgetfield:
2835 __ load_unsigned_short(rax, field);
2836 break;
2837 case Bytecodes::_fast_fgetfield:
2838 __ movflt(xmm0, field);
2839 break;
2840 case Bytecodes::_fast_dgetfield:
2841 __ movdbl(xmm0, field);
2842 break;
2843 default:
2844 ShouldNotReachHere();
2845 }
2846 // [jk] not needed currently
2847 // if (os::is_MP()) {
2848 // Label notVolatile;
2849 // __ testl(rdx, rdx);
2850 // __ jcc(Assembler::zero, notVolatile);
2851 // __ membar(Assembler::LoadLoad);
2852 // __ bind(notVolatile);
2853 //};
2854 }
2855
2856 void TemplateTable::fast_xaccess(TosState state) {
2857 transition(vtos, state);
2858
2859 // get receiver
2860 __ movptr(rax, aaddress(0));
2861 // access constant pool cache
2862 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2863 __ movptr(rbx,
2864 Address(rcx, rdx, Address::times_8,
2865 in_bytes(ConstantPoolCache::base_offset() +
2866 ConstantPoolCacheEntry::f2_offset())));
2867 // make sure exception is reported in correct bcp range (getfield is
2868 // next instruction)
2869 __ increment(r13);
2870 __ null_check(rax);
2871 switch (state) {
2872 case itos:
2873 __ movl(rax, Address(rax, rbx, Address::times_1));
2874 break;
2875 case atos:
2876 __ load_heap_oop(rax, Address(rax, rbx, Address::times_1));
2877 __ verify_oop(rax);
2878 break;
2879 case ftos:
2880 __ movflt(xmm0, Address(rax, rbx, Address::times_1));
2881 break;
2882 default:
2883 ShouldNotReachHere();
2884 }
2885
2886 // [jk] not needed currently
2887 // if (os::is_MP()) {
2888 // Label notVolatile;
2889 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
2890 // in_bytes(ConstantPoolCache::base_offset() +
2891 // ConstantPoolCacheEntry::flags_offset())));
2892 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2893 // __ testl(rdx, 0x1);
2894 // __ jcc(Assembler::zero, notVolatile);
2895 // __ membar(Assembler::LoadLoad);
2896 // __ bind(notVolatile);
2897 // }
2898
2899 __ decrement(r13);
2900 }
2901
2902
2903
2904 //-----------------------------------------------------------------------------
2905 // Calls
2906
2907 void TemplateTable::count_calls(Register method, Register temp) {
2908 // implemented elsewhere
2909 ShouldNotReachHere();
2910 }
2911
2912 void TemplateTable::prepare_invoke(int byte_no,
2913 Register method, // linked method (or i-klass)
2914 Register index, // itable index, MethodType, etc.
2915 Register recv, // if caller wants to see it
2916 Register flags // if caller wants to test it
2917 ) {
2918 // determine flags
2919 const Bytecodes::Code code = bytecode();
2920 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2921 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
2922 const bool is_invokehandle = code == Bytecodes::_invokehandle;
2923 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2924 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2925 const bool load_receiver = (recv != noreg);
2926 const bool save_flags = (flags != noreg);
2927 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
2928 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
2929 assert(flags == noreg || flags == rdx, "");
2930 assert(recv == noreg || recv == rcx, "");
2931
2932 // setup registers & access constant pool cache
2933 if (recv == noreg) recv = rcx;
2934 if (flags == noreg) flags = rdx;
2935 assert_different_registers(method, index, recv, flags);
2936
2937 // save 'interpreter return address'
2938 __ save_bcp();
2939
2940 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2941
2942 // maybe push appendix to arguments (just before return address)
2943 if (is_invokedynamic || is_invokehandle) {
2944 Label L_no_push;
2945 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
2946 __ jcc(Assembler::zero, L_no_push);
2947 // Push the appendix as a trailing parameter.
2948 // This must be done before we get the receiver,
2949 // since the parameter_size includes it.
2950 __ push(rbx);
2951 __ mov(rbx, index);
2952 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
2953 __ load_resolved_reference_at_index(index, rbx);
2954 __ pop(rbx);
2955 __ push(index); // push appendix (MethodType, CallSite, etc.)
2956 __ bind(L_no_push);
2957 }
2958
2959 // load receiver if needed (after appendix is pushed so parameter size is correct)
2960 // Note: no return address pushed yet
2961 if (load_receiver) {
2962 __ movl(recv, flags);
2963 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
2964 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
2965 const int receiver_is_at_end = -1; // back off one slot to get receiver
2966 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
2967 __ movptr(recv, recv_addr);
2968 __ verify_oop(recv);
2969 }
2970
2971 if (save_flags) {
2972 __ movl(r13, flags);
2973 }
2974
2975 // compute return type
2976 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2977 // Make sure we don't need to mask flags after the above shift
2978 ConstantPoolCacheEntry::verify_tos_state_shift();
2979 // load return address
2980 {
2981 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
2982 ExternalAddress table(table_addr);
2983 __ lea(rscratch1, table);
2984 __ movptr(flags, Address(rscratch1, flags, Address::times_ptr));
2985 }
2986
2987 // push return address
2988 __ push(flags);
2989
2990 // Restore flags value from the constant pool cache, and restore rsi
2991 // for later null checks. r13 is the bytecode pointer
2992 if (save_flags) {
2993 __ movl(flags, r13);
2994 __ restore_bcp();
2995 }
2996 }
2997
2998
2999 void TemplateTable::invokevirtual_helper(Register index,
3000 Register recv,
3001 Register flags) {
3002 // Uses temporary registers rax, rdx
3003 assert_different_registers(index, recv, rax, rdx);
3004 assert(index == rbx, "");
3005 assert(recv == rcx, "");
3006
3007 // Test for an invoke of a final method
3008 Label notFinal;
3009 __ movl(rax, flags);
3010 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3011 __ jcc(Assembler::zero, notFinal);
3012
3013 const Register method = index; // method must be rbx
3014 assert(method == rbx,
3015 "Method* must be rbx for interpreter calling convention");
3016
3017 // do the call - the index is actually the method to call
3018 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3019
3020 // It's final, need a null check here!
3021 __ null_check(recv);
3022
3023 // profile this call
3024 __ profile_final_call(rax);
3025 __ profile_arguments_type(rax, method, r13, true);
3026
3027 __ jump_from_interpreted(method, rax);
3028
3029 __ bind(notFinal);
3030
3031 // get receiver klass
3032 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3033 __ load_klass(rax, recv);
3034
3035 // profile this call
3036 __ profile_virtual_call(rax, r14, rdx);
3037
3038 // get target Method* & entry point
3039 __ lookup_virtual_method(rax, index, method);
3040 __ profile_arguments_type(rdx, method, r13, true);
3041 __ jump_from_interpreted(method, rdx);
3042 }
3043
3044
3045 void TemplateTable::invokevirtual(int byte_no) {
3046 transition(vtos, vtos);
3047 assert(byte_no == f2_byte, "use this argument");
3048 prepare_invoke(byte_no,
3049 rbx, // method or vtable index
3050 noreg, // unused itable index
3051 rcx, rdx); // recv, flags
3052
3053 // rbx: index
3054 // rcx: receiver
3055 // rdx: flags
3056
3057 invokevirtual_helper(rbx, rcx, rdx);
3058 }
3059
3060
3061 void TemplateTable::invokespecial(int byte_no) {
3062 transition(vtos, vtos);
3063 assert(byte_no == f1_byte, "use this argument");
3064 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3065 rcx); // get receiver also for null check
3066 __ verify_oop(rcx);
3067 __ null_check(rcx);
3068 // do the call
3069 __ profile_call(rax);
3070 __ profile_arguments_type(rax, rbx, r13, false);
3071 __ jump_from_interpreted(rbx, rax);
3072 }
3073
3074
3075 void TemplateTable::invokestatic(int byte_no) {
3076 transition(vtos, vtos);
3077 assert(byte_no == f1_byte, "use this argument");
3078 prepare_invoke(byte_no, rbx); // get f1 Method*
3079 // do the call
3080 __ profile_call(rax);
3081 __ profile_arguments_type(rax, rbx, r13, false);
3082 __ jump_from_interpreted(rbx, rax);
3083 }
3084
3085 void TemplateTable::fast_invokevfinal(int byte_no) {
3086 transition(vtos, vtos);
3087 assert(byte_no == f2_byte, "use this argument");
3088 __ stop("fast_invokevfinal not used on amd64");
3089 }
3090
3091 void TemplateTable::invokeinterface(int byte_no) {
3092 transition(vtos, vtos);
3093 assert(byte_no == f1_byte, "use this argument");
3094 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3095 rcx, rdx); // recv, flags
3096
3097 // rax: interface klass (from f1)
3098 // rbx: itable index (from f2)
3099 // rcx: receiver
3100 // rdx: flags
3101
3102 // Special case of invokeinterface called for virtual method of
3103 // java.lang.Object. See cpCacheOop.cpp for details.
3104 // This code isn't produced by javac, but could be produced by
3105 // another compliant java compiler.
3106 Label notMethod;
3107 __ movl(r14, rdx);
3108 __ andl(r14, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3109 __ jcc(Assembler::zero, notMethod);
3110
3111 invokevirtual_helper(rbx, rcx, rdx);
3112 __ bind(notMethod);
3113
3114 // Get receiver klass into rdx - also a null check
3115 __ restore_locals(); // restore r14
3116 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3117 __ load_klass(rdx, rcx);
3118
3119 // profile this call
3120 __ profile_virtual_call(rdx, r13, r14);
3121
3122 Label no_such_interface, no_such_method;
3123
3124 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3125 rdx, rax, rbx,
3126 // outputs: method, scan temp. reg
3127 rbx, r13,
3128 no_such_interface);
3129
3130 // rbx: Method* to call
3131 // rcx: receiver
3132 // Check for abstract method error
3133 // Note: This should be done more efficiently via a throw_abstract_method_error
3134 // interpreter entry point and a conditional jump to it in case of a null
3135 // method.
3136 __ testptr(rbx, rbx);
3137 __ jcc(Assembler::zero, no_such_method);
3138
3139 __ profile_arguments_type(rdx, rbx, r13, true);
3140
3141 // do the call
3142 // rcx: receiver
3143 // rbx,: Method*
3144 __ jump_from_interpreted(rbx, rdx);
3145 __ should_not_reach_here();
3146
3147 // exception handling code follows...
3148 // note: must restore interpreter registers to canonical
3149 // state for exception handling to work correctly!
3150
3151 __ bind(no_such_method);
3152 // throw exception
3153 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3154 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3155 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3156 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3157 // the call_VM checks for exception, so we should never return here.
3158 __ should_not_reach_here();
3159
3160 __ bind(no_such_interface);
3161 // throw exception
3162 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3163 __ restore_bcp(); // r13 must be correct for exception handler (was destroyed)
3164 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3165 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3166 InterpreterRuntime::throw_IncompatibleClassChangeError));
3167 // the call_VM checks for exception, so we should never return here.
3168 __ should_not_reach_here();
3169 }
3170
3171
3172 void TemplateTable::invokehandle(int byte_no) {
3173 transition(vtos, vtos);
3174 assert(byte_no == f1_byte, "use this argument");
3175 const Register rbx_method = rbx;
3176 const Register rax_mtype = rax;
3177 const Register rcx_recv = rcx;
3178 const Register rdx_flags = rdx;
3179
3180 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3181 __ verify_method_ptr(rbx_method);
3182 __ verify_oop(rcx_recv);
3183 __ null_check(rcx_recv);
3184
3185 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3186 // rbx: MH.invokeExact_MT method (from f2)
3187
3188 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3189
3190 // FIXME: profile the LambdaForm also
3191 __ profile_final_call(rax);
3192 __ profile_arguments_type(rdx, rbx_method, r13, true);
3193
3194 __ jump_from_interpreted(rbx_method, rdx);
3195 }
3196
3197
3198 void TemplateTable::invokedynamic(int byte_no) {
3199 transition(vtos, vtos);
3200 assert(byte_no == f1_byte, "use this argument");
3201
3202 const Register rbx_method = rbx;
3203 const Register rax_callsite = rax;
3204
3205 prepare_invoke(byte_no, rbx_method, rax_callsite);
3206
3207 // rax: CallSite object (from cpool->resolved_references[f1])
3208 // rbx: MH.linkToCallSite method (from f2)
3209
3210 // Note: rax_callsite is already pushed by prepare_invoke
3211
3212 // %%% should make a type profile for any invokedynamic that takes a ref argument
3213 // profile this call
3214 __ profile_call(r13);
3215 __ profile_arguments_type(rdx, rbx_method, r13, false);
3216
3217 __ verify_oop(rax_callsite);
3218
3219 __ jump_from_interpreted(rbx_method, rdx);
3220 }
3221
3222
3223 //-----------------------------------------------------------------------------
3224 // Allocation
3225
3226 void TemplateTable::_new() {
3227 transition(vtos, atos);
3228 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3229 Label slow_case;
3230 Label done;
3231 Label initialize_header;
3232 Label initialize_object; // including clearing the fields
3233 Label allocate_shared;
3234
3235 __ get_cpool_and_tags(rsi, rax);
3236 // Make sure the class we're about to instantiate has been resolved.
3237 // This is done before loading InstanceKlass to be consistent with the order
3238 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3239 const int tags_offset = Array<u1>::base_offset_in_bytes();
3240 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset),
3241 JVM_CONSTANT_Class);
3242 __ jcc(Assembler::notEqual, slow_case);
3243
3244 // get InstanceKlass
3245 __ movptr(rsi, Address(rsi, rdx,
3246 Address::times_8, sizeof(ConstantPool)));
3247
3248 // make sure klass is initialized & doesn't have finalizer
3249 // make sure klass is fully initialized
3250 __ cmpb(Address(rsi,
3251 InstanceKlass::init_state_offset()),
3252 InstanceKlass::fully_initialized);
3253 __ jcc(Assembler::notEqual, slow_case);
3254
3255 // get instance_size in InstanceKlass (scaled to a count of bytes)
3256 __ movl(rdx,
3257 Address(rsi,
3258 Klass::layout_helper_offset()));
3259 // test to see if it has a finalizer or is malformed in some way
3260 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3261 __ jcc(Assembler::notZero, slow_case);
3262
3263 // Allocate the instance
3264 // 1) Try to allocate in the TLAB
3265 // 2) if fail and the object is large allocate in the shared Eden
3266 // 3) if the above fails (or is not applicable), go to a slow case
3267 // (creates a new TLAB, etc.)
3268
3269 const bool allow_shared_alloc =
3270 Universe::heap()->supports_inline_contig_alloc();
3271
3272 if (UseTLAB) {
3273 __ movptr(rax, Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())));
3274 __ lea(rbx, Address(rax, rdx, Address::times_1));
3275 __ cmpptr(rbx, Address(r15_thread, in_bytes(JavaThread::tlab_end_offset())));
3276 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3277 __ movptr(Address(r15_thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3278 if (ZeroTLAB) {
3279 // the fields have been already cleared
3280 __ jmp(initialize_header);
3281 } else {
3282 // initialize both the header and fields
3283 __ jmp(initialize_object);
3284 }
3285 }
3286
3287 // Allocation in the shared Eden, if allowed.
3288 //
3289 // rdx: instance size in bytes
3290 if (allow_shared_alloc) {
3291 __ bind(allocate_shared);
3292
3293 ExternalAddress top((address)Universe::heap()->top_addr());
3294 ExternalAddress end((address)Universe::heap()->end_addr());
3295
3296 const Register RtopAddr = rscratch1;
3297 const Register RendAddr = rscratch2;
3298
3299 __ lea(RtopAddr, top);
3300 __ lea(RendAddr, end);
3301 __ movptr(rax, Address(RtopAddr, 0));
3302
3303 // For retries rax gets set by cmpxchgq
3304 Label retry;
3305 __ bind(retry);
3306 __ lea(rbx, Address(rax, rdx, Address::times_1));
3307 __ cmpptr(rbx, Address(RendAddr, 0));
3308 __ jcc(Assembler::above, slow_case);
3309
3310 // Compare rax with the top addr, and if still equal, store the new
3311 // top addr in rbx at the address of the top addr pointer. Sets ZF if was
3312 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3313 //
3314 // rax: object begin
3315 // rbx: object end
3316 // rdx: instance size in bytes
3317 if (os::is_MP()) {
3318 __ lock();
3319 }
3320 __ cmpxchgptr(rbx, Address(RtopAddr, 0));
3321
3322 // if someone beat us on the allocation, try again, otherwise continue
3323 __ jcc(Assembler::notEqual, retry);
3324
3325 __ incr_allocated_bytes(r15_thread, rdx, 0);
3326 }
3327
3328 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3329 // The object is initialized before the header. If the object size is
3330 // zero, go directly to the header initialization.
3331 __ bind(initialize_object);
3332 __ decrementl(rdx, sizeof(oopDesc));
3333 __ jcc(Assembler::zero, initialize_header);
3334
3335 // Initialize object fields
3336 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3337 __ shrl(rdx, LogBytesPerLong); // divide by oopSize to simplify the loop
3338 {
3339 Label loop;
3340 __ bind(loop);
3341 __ movq(Address(rax, rdx, Address::times_8,
3342 sizeof(oopDesc) - oopSize),
3343 rcx);
3344 __ decrementl(rdx);
3345 __ jcc(Assembler::notZero, loop);
3346 }
3347
3348 // initialize object header only.
3349 __ bind(initialize_header);
3350 if (UseBiasedLocking) {
3351 __ movptr(rscratch1, Address(rsi, Klass::prototype_header_offset()));
3352 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()), rscratch1);
3353 } else {
3354 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes()),
3355 (intptr_t) markOopDesc::prototype()); // header (address 0x1)
3356 }
3357 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3358 __ store_klass_gap(rax, rcx); // zero klass gap for compressed oops
3359 __ store_klass(rax, rsi); // store klass last
3360
3361 {
3362 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3363 // Trigger dtrace event for fastpath
3364 __ push(atos); // save the return value
3365 __ call_VM_leaf(
3366 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3367 __ pop(atos); // restore the return value
3368
3369 }
3370 __ jmp(done);
3371 }
3372
3373
3374 // slow case
3375 __ bind(slow_case);
3376 __ get_constant_pool(c_rarg1);
3377 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3378 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3379 __ verify_oop(rax);
3380
3381 // continue
3382 __ bind(done);
3383 }
3384
3385 void TemplateTable::newarray() {
3386 transition(itos, atos);
3387 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3388 __ movl(c_rarg2, rax);
3389 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3390 c_rarg1, c_rarg2);
3391 }
3392
3393 void TemplateTable::anewarray() {
3394 transition(itos, atos);
3395 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3396 __ get_constant_pool(c_rarg1);
3397 __ movl(c_rarg3, rax);
3398 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3399 c_rarg1, c_rarg2, c_rarg3);
3400 }
3401
3402 void TemplateTable::arraylength() {
3403 transition(atos, itos);
3404 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3405 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3406 }
3407
3408 void TemplateTable::checkcast() {
3409 transition(atos, atos);
3410 Label done, is_null, ok_is_subtype, quicked, resolved;
3411 __ testptr(rax, rax); // object is in rax
3412 __ jcc(Assembler::zero, is_null);
3413
3414 // Get cpool & tags index
3415 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3416 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3417 // See if bytecode has already been quicked
3418 __ cmpb(Address(rdx, rbx,
3419 Address::times_1,
3420 Array<u1>::base_offset_in_bytes()),
3421 JVM_CONSTANT_Class);
3422 __ jcc(Assembler::equal, quicked);
3423 __ push(atos); // save receiver for result, and for GC
3424 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3425 // vm_result_2 has metadata result
3426 __ get_vm_result_2(rax, r15_thread);
3427 __ pop_ptr(rdx); // restore receiver
3428 __ jmpb(resolved);
3429
3430 // Get superklass in rax and subklass in rbx
3431 __ bind(quicked);
3432 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
3433 __ movptr(rax, Address(rcx, rbx,
3434 Address::times_8, sizeof(ConstantPool)));
3435
3436 __ bind(resolved);
3437 __ load_klass(rbx, rdx);
3438
3439 // Generate subtype check. Blows rcx, rdi. Object in rdx.
3440 // Superklass in rax. Subklass in rbx.
3441 __ gen_subtype_check(rbx, ok_is_subtype);
3442
3443 // Come here on failure
3444 __ push_ptr(rdx);
3445 // object is at TOS
3446 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3447
3448 // Come here on success
3449 __ bind(ok_is_subtype);
3450 __ mov(rax, rdx); // Restore object in rdx
3451
3452 // Collect counts on whether this check-cast sees NULLs a lot or not.
3453 if (ProfileInterpreter) {
3454 __ jmp(done);
3455 __ bind(is_null);
3456 __ profile_null_seen(rcx);
3457 } else {
3458 __ bind(is_null); // same as 'done'
3459 }
3460 __ bind(done);
3461 }
3462
3463 void TemplateTable::instanceof() {
3464 transition(atos, itos);
3465 Label done, is_null, ok_is_subtype, quicked, resolved;
3466 __ testptr(rax, rax);
3467 __ jcc(Assembler::zero, is_null);
3468
3469 // Get cpool & tags index
3470 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3471 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3472 // See if bytecode has already been quicked
3473 __ cmpb(Address(rdx, rbx,
3474 Address::times_1,
3475 Array<u1>::base_offset_in_bytes()),
3476 JVM_CONSTANT_Class);
3477 __ jcc(Assembler::equal, quicked);
3478
3479 __ push(atos); // save receiver for result, and for GC
3480 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3481 // vm_result_2 has metadata result
3482 __ get_vm_result_2(rax, r15_thread);
3483 __ pop_ptr(rdx); // restore receiver
3484 __ verify_oop(rdx);
3485 __ load_klass(rdx, rdx);
3486 __ jmpb(resolved);
3487
3488 // Get superklass in rax and subklass in rdx
3489 __ bind(quicked);
3490 __ load_klass(rdx, rax);
3491 __ movptr(rax, Address(rcx, rbx,
3492 Address::times_8, sizeof(ConstantPool)));
3493
3494 __ bind(resolved);
3495
3496 // Generate subtype check. Blows rcx, rdi
3497 // Superklass in rax. Subklass in rdx.
3498 __ gen_subtype_check(rdx, ok_is_subtype);
3499
3500 // Come here on failure
3501 __ xorl(rax, rax);
3502 __ jmpb(done);
3503 // Come here on success
3504 __ bind(ok_is_subtype);
3505 __ movl(rax, 1);
3506
3507 // Collect counts on whether this test sees NULLs a lot or not.
3508 if (ProfileInterpreter) {
3509 __ jmp(done);
3510 __ bind(is_null);
3511 __ profile_null_seen(rcx);
3512 } else {
3513 __ bind(is_null); // same as 'done'
3514 }
3515 __ bind(done);
3516 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
3517 // rax = 1: obj != NULL and obj is an instanceof the specified klass
3518 }
3519
3520 //-----------------------------------------------------------------------------
3521 // Breakpoints
3522 void TemplateTable::_breakpoint() {
3523 // Note: We get here even if we are single stepping..
3524 // jbug inists on setting breakpoints at every bytecode
3525 // even if we are in single step mode.
3526
3527 transition(vtos, vtos);
3528
3529 // get the unpatched byte code
3530 __ get_method(c_rarg1);
3531 __ call_VM(noreg,
3532 CAST_FROM_FN_PTR(address,
3533 InterpreterRuntime::get_original_bytecode_at),
3534 c_rarg1, r13);
3535 __ mov(rbx, rax);
3536
3537 // post the breakpoint event
3538 __ get_method(c_rarg1);
3539 __ call_VM(noreg,
3540 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3541 c_rarg1, r13);
3542
3543 // complete the execution of original bytecode
3544 __ dispatch_only_normal(vtos);
3545 }
3546
3547 //-----------------------------------------------------------------------------
3548 // Exceptions
3549
3550 void TemplateTable::athrow() {
3551 transition(atos, vtos);
3552 __ null_check(rax);
3553 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3554 }
3555
3556 //-----------------------------------------------------------------------------
3557 // Synchronization
3558 //
3559 // Note: monitorenter & exit are symmetric routines; which is reflected
3560 // in the assembly code structure as well
3561 //
3562 // Stack layout:
3563 //
3564 // [expressions ] <--- rsp = expression stack top
3565 // ..
3566 // [expressions ]
3567 // [monitor entry] <--- monitor block top = expression stack bot
3568 // ..
3569 // [monitor entry]
3570 // [frame data ] <--- monitor block bot
3571 // ...
3572 // [saved rbp ] <--- rbp
3573 void TemplateTable::monitorenter() {
3574 transition(atos, vtos);
3575
3576 // check for NULL object
3577 __ null_check(rax);
3578
3579 const Address monitor_block_top(
3580 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3581 const Address monitor_block_bot(
3582 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3583 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3584
3585 Label allocated;
3586
3587 // initialize entry pointer
3588 __ xorl(c_rarg1, c_rarg1); // points to free slot or NULL
3589
3590 // find a free slot in the monitor block (result in c_rarg1)
3591 {
3592 Label entry, loop, exit;
3593 __ movptr(c_rarg3, monitor_block_top); // points to current entry,
3594 // starting with top-most entry
3595 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3596 // of monitor block
3597 __ jmpb(entry);
3598
3599 __ bind(loop);
3600 // check if current entry is used
3601 __ cmpptr(Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
3602 // if not used then remember entry in c_rarg1
3603 __ cmov(Assembler::equal, c_rarg1, c_rarg3);
3604 // check if current entry is for same object
3605 __ cmpptr(rax, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3606 // if same object then stop searching
3607 __ jccb(Assembler::equal, exit);
3608 // otherwise advance to next entry
3609 __ addptr(c_rarg3, entry_size);
3610 __ bind(entry);
3611 // check if bottom reached
3612 __ cmpptr(c_rarg3, c_rarg2);
3613 // if not at bottom then check this entry
3614 __ jcc(Assembler::notEqual, loop);
3615 __ bind(exit);
3616 }
3617
3618 __ testptr(c_rarg1, c_rarg1); // check if a slot has been found
3619 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
3620
3621 // allocate one if there's no free slot
3622 {
3623 Label entry, loop;
3624 // 1. compute new pointers // rsp: old expression stack top
3625 __ movptr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
3626 __ subptr(rsp, entry_size); // move expression stack top
3627 __ subptr(c_rarg1, entry_size); // move expression stack bottom
3628 __ mov(c_rarg3, rsp); // set start value for copy loop
3629 __ movptr(monitor_block_bot, c_rarg1); // set new monitor block bottom
3630 __ jmp(entry);
3631 // 2. move expression stack contents
3632 __ bind(loop);
3633 __ movptr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3634 // word from old location
3635 __ movptr(Address(c_rarg3, 0), c_rarg2); // and store it at new location
3636 __ addptr(c_rarg3, wordSize); // advance to next word
3637 __ bind(entry);
3638 __ cmpptr(c_rarg3, c_rarg1); // check if bottom reached
3639 __ jcc(Assembler::notEqual, loop); // if not at bottom then
3640 // copy next word
3641 }
3642
3643 // call run-time routine
3644 // c_rarg1: points to monitor entry
3645 __ bind(allocated);
3646
3647 // Increment bcp to point to the next bytecode, so exception
3648 // handling for async. exceptions work correctly.
3649 // The object has already been poped from the stack, so the
3650 // expression stack looks correct.
3651 __ increment(r13);
3652
3653 // store object
3654 __ movptr(Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()), rax);
3655 __ lock_object(c_rarg1);
3656
3657 // check to make sure this monitor doesn't cause stack overflow after locking
3658 __ save_bcp(); // in case of exception
3659 __ generate_stack_overflow_check(0);
3660
3661 // The bcp has already been incremented. Just need to dispatch to
3662 // next instruction.
3663 __ dispatch_next(vtos);
3664 }
3665
3666
3667 void TemplateTable::monitorexit() {
3668 transition(atos, vtos);
3669
3670 // check for NULL object
3671 __ null_check(rax);
3672
3673 const Address monitor_block_top(
3674 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3675 const Address monitor_block_bot(
3676 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3677 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3678
3679 Label found;
3680
3681 // find matching slot
3682 {
3683 Label entry, loop;
3684 __ movptr(c_rarg1, monitor_block_top); // points to current entry,
3685 // starting with top-most entry
3686 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3687 // of monitor block
3688 __ jmpb(entry);
3689
3690 __ bind(loop);
3691 // check if current entry is for same object
3692 __ cmpptr(rax, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3693 // if same object then stop searching
3694 __ jcc(Assembler::equal, found);
3695 // otherwise advance to next entry
3696 __ addptr(c_rarg1, entry_size);
3697 __ bind(entry);
3698 // check if bottom reached
3699 __ cmpptr(c_rarg1, c_rarg2);
3700 // if not at bottom then check this entry
3701 __ jcc(Assembler::notEqual, loop);
3702 }
3703
3704 // error handling. Unlocking was not block-structured
3705 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3706 InterpreterRuntime::throw_illegal_monitor_state_exception));
3707 __ should_not_reach_here();
3708
3709 // call run-time routine
3710 // rsi: points to monitor entry
3711 __ bind(found);
3712 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3713 __ unlock_object(c_rarg1);
3714 __ pop_ptr(rax); // discard object
3715 }
3716
3717
3718 // Wide instructions
3719 void TemplateTable::wide() {
3720 transition(vtos, vtos);
3721 __ load_unsigned_byte(rbx, at_bcp(1));
3722 __ lea(rscratch1, ExternalAddress((address)Interpreter::_wentry_point));
3723 __ jmp(Address(rscratch1, rbx, Address::times_8));
3724 // Note: the r13 increment step is part of the individual wide
3725 // bytecode implementations
3726 }
3727
3728
3729 // Multi arrays
3730 void TemplateTable::multianewarray() {
3731 transition(vtos, atos);
3732 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3733 // last dim is on top of stack; we want address of first one:
3734 // first_addr = last_addr + (ndims - 1) * wordSize
3735 __ lea(c_rarg1, Address(rsp, rax, Address::times_8, -wordSize));
3736 call_VM(rax,
3737 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3738 c_rarg1);
3739 __ load_unsigned_byte(rbx, at_bcp(3));
3740 __ lea(rsp, Address(rsp, rbx, Address::times_8));
3741 }
3742 #endif // !CC_INTERP