1 /*
2 * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.inline.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #ifndef CC_INTERP
42
43 #define __ _masm->
44
45 // Global Register Names
46 Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
47 Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
48
49 // Platform-dependent initialization
50 void TemplateTable::pd_initialize() {
51 // No x86 specific initialization
52 }
53
54 // Address Computation: local variables
55 static inline Address iaddress(int n) {
56 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
57 }
58
59 static inline Address laddress(int n) {
60 return iaddress(n + 1);
61 }
62
63 #ifndef _LP64
64 static inline Address haddress(int n) {
65 return iaddress(n + 0);
66 }
67 #endif
68
69 static inline Address faddress(int n) {
70 return iaddress(n);
71 }
72
73 static inline Address daddress(int n) {
74 return laddress(n);
75 }
76
77 static inline Address aaddress(int n) {
78 return iaddress(n);
79 }
80
81 static inline Address iaddress(Register r) {
82 return Address(rlocals, r, Address::times_ptr);
83 }
84
85 static inline Address laddress(Register r) {
86 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
87 }
88
89 #ifndef _LP64
90 static inline Address haddress(Register r) {
91 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
92 }
93 #endif
94
95 static inline Address faddress(Register r) {
96 return iaddress(r);
97 }
98
99 static inline Address daddress(Register r) {
100 return laddress(r);
101 }
102
103 static inline Address aaddress(Register r) {
104 return iaddress(r);
105 }
106
107
108 // expression stack
109 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
110 // data beyond the rsp which is potentially unsafe in an MT environment;
111 // an interrupt may overwrite that data.)
112 static inline Address at_rsp () {
113 return Address(rsp, 0);
114 }
115
116 // At top of Java expression stack which may be different than esp(). It
117 // isn't for category 1 objects.
118 static inline Address at_tos () {
119 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
120 }
121
122 static inline Address at_tos_p1() {
123 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
124 }
125
126 static inline Address at_tos_p2() {
127 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
128 }
129
130 // Condition conversion
131 static Assembler::Condition j_not(TemplateTable::Condition cc) {
132 switch (cc) {
133 case TemplateTable::equal : return Assembler::notEqual;
134 case TemplateTable::not_equal : return Assembler::equal;
135 case TemplateTable::less : return Assembler::greaterEqual;
136 case TemplateTable::less_equal : return Assembler::greater;
137 case TemplateTable::greater : return Assembler::lessEqual;
138 case TemplateTable::greater_equal: return Assembler::less;
139 }
140 ShouldNotReachHere();
141 return Assembler::zero;
142 }
143
144
145
146 // Miscelaneous helper routines
147 // Store an oop (or NULL) at the address described by obj.
148 // If val == noreg this means store a NULL
149
150
151 static void do_oop_store(InterpreterMacroAssembler* _masm,
152 Address obj,
153 Register val,
154 BarrierSet::Name barrier,
155 bool precise) {
156 assert(val == noreg || val == rax, "parameter is just for looks");
157 switch (barrier) {
158 #if INCLUDE_ALL_GCS
159 case BarrierSet::G1SATBCT:
160 case BarrierSet::G1SATBCTLogging:
161 {
162 // flatten object address if needed
163 // We do it regardless of precise because we need the registers
164 if (obj.index() == noreg && obj.disp() == 0) {
165 if (obj.base() != rdx) {
166 __ movptr(rdx, obj.base());
167 }
168 } else {
169 __ lea(rdx, obj);
170 }
171
172 Register rtmp = LP64_ONLY(r8) NOT_LP64(rsi);
173 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
174
175 NOT_LP64(__ get_thread(rcx));
176 NOT_LP64(__ save_bcp());
177
178 __ g1_write_barrier_pre(rdx /* obj */,
179 rbx /* pre_val */,
180 rthread /* thread */,
181 rtmp /* tmp */,
182 val != noreg /* tosca_live */,
183 false /* expand_call */);
184 if (val == noreg) {
185 __ store_heap_oop_null(Address(rdx, 0));
186 } else {
187 // G1 barrier needs uncompressed oop for region cross check.
188 Register new_val = val;
189 if (UseCompressedOops) {
190 new_val = rbx;
191 __ movptr(new_val, val);
192 }
193 __ store_heap_oop(Address(rdx, 0), val);
194 __ g1_write_barrier_post(rdx /* store_adr */,
195 new_val /* new_val */,
196 rthread /* thread */,
197 rtmp /* tmp */,
198 rbx /* tmp2 */);
199 }
200 NOT_LP64( __ restore_bcp());
201 }
202 break;
203 #endif // INCLUDE_ALL_GCS
204 case BarrierSet::CardTableModRef:
205 case BarrierSet::CardTableExtension:
206 {
207 if (val == noreg) {
208 __ store_heap_oop_null(obj);
209 } else {
210 __ store_heap_oop(obj, val);
211 // flatten object address if needed
212 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
213 __ store_check(obj.base());
214 } else {
215 __ lea(rdx, obj);
216 __ store_check(rdx);
217 }
218 }
219 }
220 break;
221 case BarrierSet::ModRef:
222 if (val == noreg) {
223 __ store_heap_oop_null(obj);
224 } else {
225 __ store_heap_oop(obj, val);
226 }
227 break;
228 default :
229 ShouldNotReachHere();
230
231 }
232 }
233
234 Address TemplateTable::at_bcp(int offset) {
235 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
236 return Address(rbcp, offset);
237 }
238
239
240 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
241 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
242 int byte_no) {
243 if (!RewriteBytecodes) return;
244 Label L_patch_done;
245
246 switch (bc) {
247 case Bytecodes::_fast_aputfield:
248 case Bytecodes::_fast_bputfield:
249 case Bytecodes::_fast_cputfield:
250 case Bytecodes::_fast_dputfield:
251 case Bytecodes::_fast_fputfield:
252 case Bytecodes::_fast_iputfield:
253 case Bytecodes::_fast_lputfield:
254 case Bytecodes::_fast_sputfield:
255 {
256 // We skip bytecode quickening for putfield instructions when
257 // the put_code written to the constant pool cache is zero.
258 // This is required so that every execution of this instruction
259 // calls out to InterpreterRuntime::resolve_get_put to do
260 // additional, required work.
261 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
262 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
263 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
264 __ movl(bc_reg, bc);
265 __ cmpl(temp_reg, (int) 0);
266 __ jcc(Assembler::zero, L_patch_done); // don't patch
267 }
268 break;
269 default:
270 assert(byte_no == -1, "sanity");
271 // the pair bytecodes have already done the load.
272 if (load_bc_into_bc_reg) {
273 __ movl(bc_reg, bc);
274 }
275 }
276
277 if (JvmtiExport::can_post_breakpoint()) {
278 Label L_fast_patch;
279 // if a breakpoint is present we can't rewrite the stream directly
280 __ movzbl(temp_reg, at_bcp(0));
281 __ cmpl(temp_reg, Bytecodes::_breakpoint);
282 __ jcc(Assembler::notEqual, L_fast_patch);
283 __ get_method(temp_reg);
284 // Let breakpoint table handling rewrite to quicker bytecode
285 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
286 #ifndef ASSERT
287 __ jmpb(L_patch_done);
288 #else
289 __ jmp(L_patch_done);
290 #endif
291 __ bind(L_fast_patch);
292 }
293
294 #ifdef ASSERT
295 Label L_okay;
296 __ load_unsigned_byte(temp_reg, at_bcp(0));
297 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
298 __ jcc(Assembler::equal, L_okay);
299 __ cmpl(temp_reg, bc_reg);
300 __ jcc(Assembler::equal, L_okay);
301 __ stop("patching the wrong bytecode");
302 __ bind(L_okay);
303 #endif
304
305 // patch bytecode
306 __ movb(at_bcp(0), bc_reg);
307 __ bind(L_patch_done);
308 }
309 // Individual instructions
310
311
312 void TemplateTable::nop() {
313 transition(vtos, vtos);
314 // nothing to do
315 }
316
317 void TemplateTable::shouldnotreachhere() {
318 transition(vtos, vtos);
319 __ stop("shouldnotreachhere bytecode");
320 }
321
322 void TemplateTable::aconst_null() {
323 transition(vtos, atos);
324 __ xorl(rax, rax);
325 }
326
327 void TemplateTable::iconst(int value) {
328 transition(vtos, itos);
329 if (value == 0) {
330 __ xorl(rax, rax);
331 } else {
332 __ movl(rax, value);
333 }
334 }
335
336 void TemplateTable::lconst(int value) {
337 transition(vtos, ltos);
338 if (value == 0) {
339 __ xorl(rax, rax);
340 } else {
341 __ movl(rax, value);
342 }
343 #ifndef _LP64
344 assert(value >= 0, "check this code");
345 __ xorptr(rdx, rdx);
346 #endif
347 }
348
349
350
351 void TemplateTable::fconst(int value) {
352 transition(vtos, ftos);
353 #ifdef _LP64
354 static float one = 1.0f, two = 2.0f;
355 switch (value) {
356 case 0:
357 __ xorps(xmm0, xmm0);
358 break;
359 case 1:
360 __ movflt(xmm0, ExternalAddress((address) &one));
361 break;
362 case 2:
363 __ movflt(xmm0, ExternalAddress((address) &two));
364 break;
365 default:
366 ShouldNotReachHere();
367 break;
368 }
369 #else
370 if (value == 0) { __ fldz();
371 } else if (value == 1) { __ fld1();
372 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
373 } else { ShouldNotReachHere();
374 }
375 #endif
376 }
377
378 void TemplateTable::dconst(int value) {
379 transition(vtos, dtos);
380 #ifdef _LP64
381 static double one = 1.0;
382 switch (value) {
383 case 0:
384 __ xorpd(xmm0, xmm0);
385 break;
386 case 1:
387 __ movdbl(xmm0, ExternalAddress((address) &one));
388 break;
389 default:
390 ShouldNotReachHere();
391 break;
392 }
393
394 #else
395 if (value == 0) { __ fldz();
396 } else if (value == 1) { __ fld1();
397 } else { ShouldNotReachHere();
398 }
399 #endif
400 }
401
402 void TemplateTable::bipush() {
403 transition(vtos, itos);
404 __ load_signed_byte(rax, at_bcp(1));
405 }
406
407 void TemplateTable::sipush() {
408 transition(vtos, itos);
409 __ load_unsigned_short(rax, at_bcp(1));
410 __ bswapl(rax);
411 __ sarl(rax, 16);
412 }
413
414 void TemplateTable::ldc(bool wide) {
415 transition(vtos, vtos);
416 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
417 Label call_ldc, notFloat, notClass, Done;
418
419 if (wide) {
420 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
421 } else {
422 __ load_unsigned_byte(rbx, at_bcp(1));
423 }
424
425 __ get_cpool_and_tags(rcx, rax);
426 const int base_offset = ConstantPool::header_size() * wordSize;
427 const int tags_offset = Array<u1>::base_offset_in_bytes();
428
429 // get type
430 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
431
432 // unresolved class - get the resolved class
433 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
434 __ jccb(Assembler::equal, call_ldc);
435
436 // unresolved class in error state - call into runtime to throw the error
437 // from the first resolution attempt
438 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
439 __ jccb(Assembler::equal, call_ldc);
440
441 // resolved class - need to call vm to get java mirror of the class
442 __ cmpl(rdx, JVM_CONSTANT_Class);
443 __ jcc(Assembler::notEqual, notClass);
444
445 __ bind(call_ldc);
446
447 __ movl(rarg, wide);
448 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
449
450 __ push(atos);
451 __ jmp(Done);
452
453 __ bind(notClass);
454 __ cmpl(rdx, JVM_CONSTANT_Float);
455 __ jccb(Assembler::notEqual, notFloat);
456
457 // ftos
458 LP64_ONLY(__ movflt(xmm0, Address(rcx, rbx, Address::times_8, base_offset)));
459 NOT_LP64(__ fld_s( Address(rcx, rbx, Address::times_ptr, base_offset)));
460 __ push(ftos);
461 __ jmp(Done);
462
463 __ bind(notFloat);
464 #ifdef ASSERT
465 {
466 Label L;
467 __ cmpl(rdx, JVM_CONSTANT_Integer);
468 __ jcc(Assembler::equal, L);
469 // String and Object are rewritten to fast_aldc
470 __ stop("unexpected tag type in ldc");
471 __ bind(L);
472 }
473 #endif
474 // itos JVM_CONSTANT_Integer only
475 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
476 __ push(itos);
477 __ bind(Done);
478 }
479
480 // Fast path for caching oop constants.
481 void TemplateTable::fast_aldc(bool wide) {
482 transition(vtos, atos);
483
484 Register result = rax;
485 Register tmp = rdx;
486 int index_size = wide ? sizeof(u2) : sizeof(u1);
487
488 Label resolved;
489
490 // We are resolved if the resolved reference cache entry contains a
491 // non-null object (String, MethodType, etc.)
492 assert_different_registers(result, tmp);
493 __ get_cache_index_at_bcp(tmp, 1, index_size);
494 __ load_resolved_reference_at_index(result, tmp);
495 __ testl(result, result);
496 __ jcc(Assembler::notZero, resolved);
497
498 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
499
500 // first time invocation - must resolve first
501 __ movl(tmp, (int)bytecode());
502 __ call_VM(result, entry, tmp);
503
504 __ bind(resolved);
505
506 if (VerifyOops) {
507 __ verify_oop(result);
508 }
509 }
510
511 void TemplateTable::ldc2_w() {
512 transition(vtos, vtos);
513 Label Long, Done;
514 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
515
516 __ get_cpool_and_tags(rcx, rax);
517 const int base_offset = ConstantPool::header_size() * wordSize;
518 const int tags_offset = Array<u1>::base_offset_in_bytes();
519
520 // get type
521 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
522 JVM_CONSTANT_Double);
523 __ jccb(Assembler::notEqual, Long);
524
525 // dtos
526 LP64_ONLY(__ movdbl(xmm0, Address(rcx, rbx, Address::times_8, base_offset)));
527 NOT_LP64(__ fld_d( Address(rcx, rbx, Address::times_ptr, base_offset)));
528 __ push(dtos);
529
530 __ jmpb(Done);
531 __ bind(Long);
532
533 // ltos
534 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
535 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
536 __ push(ltos);
537
538 __ bind(Done);
539 }
540
541 void TemplateTable::locals_index(Register reg, int offset) {
542 __ load_unsigned_byte(reg, at_bcp(offset));
543 __ negptr(reg);
544 }
545
546 void TemplateTable::iload() {
547 transition(vtos, itos);
548 if (RewriteFrequentPairs) {
549 Label rewrite, done;
550 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
551 LP64_ONLY(assert(rbx != bc, "register damaged"));
552
553 // get next byte
554 __ load_unsigned_byte(rbx,
555 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
556 // if _iload, wait to rewrite to iload2. We only want to rewrite the
557 // last two iloads in a pair. Comparing against fast_iload means that
558 // the next bytecode is neither an iload or a caload, and therefore
559 // an iload pair.
560 __ cmpl(rbx, Bytecodes::_iload);
561 __ jcc(Assembler::equal, done);
562
563 __ cmpl(rbx, Bytecodes::_fast_iload);
564 __ movl(bc, Bytecodes::_fast_iload2);
565
566 __ jccb(Assembler::equal, rewrite);
567
568 // if _caload, rewrite to fast_icaload
569 __ cmpl(rbx, Bytecodes::_caload);
570 __ movl(bc, Bytecodes::_fast_icaload);
571 __ jccb(Assembler::equal, rewrite);
572
573 // rewrite so iload doesn't check again.
574 __ movl(bc, Bytecodes::_fast_iload);
575
576 // rewrite
577 // bc: fast bytecode
578 __ bind(rewrite);
579 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
580 __ bind(done);
581 }
582
583 // Get the local value into tos
584 locals_index(rbx);
585 __ movl(rax, iaddress(rbx));
586 }
587
588 void TemplateTable::fast_iload2() {
589 transition(vtos, itos);
590 locals_index(rbx);
591 __ movl(rax, iaddress(rbx));
592 __ push(itos);
593 locals_index(rbx, 3);
594 __ movl(rax, iaddress(rbx));
595 }
596
597 void TemplateTable::fast_iload() {
598 transition(vtos, itos);
599 locals_index(rbx);
600 __ movl(rax, iaddress(rbx));
601 }
602
603 void TemplateTable::lload() {
604 transition(vtos, ltos);
605 locals_index(rbx);
606 __ movptr(rax, laddress(rbx));
607 NOT_LP64(__ movl(rdx, haddress(rbx)));
608 }
609
610 void TemplateTable::fload() {
611 transition(vtos, ftos);
612 locals_index(rbx);
613 LP64_ONLY(__ movflt(xmm0, faddress(rbx)));
614 NOT_LP64(__ fld_s(faddress(rbx)));
615 }
616
617 void TemplateTable::dload() {
618 transition(vtos, dtos);
619 locals_index(rbx);
620 LP64_ONLY(__ movdbl(xmm0, daddress(rbx)));
621 NOT_LP64(__ fld_d(daddress(rbx)));
622 }
623
624 void TemplateTable::aload() {
625 transition(vtos, atos);
626 locals_index(rbx);
627 __ movptr(rax, aaddress(rbx));
628 }
629
630 void TemplateTable::locals_index_wide(Register reg) {
631 __ load_unsigned_short(reg, at_bcp(2));
632 __ bswapl(reg);
633 __ shrl(reg, 16);
634 __ negptr(reg);
635 }
636
637 void TemplateTable::wide_iload() {
638 transition(vtos, itos);
639 locals_index_wide(rbx);
640 __ movl(rax, iaddress(rbx));
641 }
642
643 void TemplateTable::wide_lload() {
644 transition(vtos, ltos);
645 locals_index_wide(rbx);
646 __ movptr(rax, laddress(rbx));
647 NOT_LP64(__ movl(rdx, haddress(rbx)));
648 }
649
650 void TemplateTable::wide_fload() {
651 transition(vtos, ftos);
652 locals_index_wide(rbx);
653 LP64_ONLY(__ movflt(xmm0, faddress(rbx)));
654 NOT_LP64(__ fld_s(faddress(rbx)));
655 }
656
657 void TemplateTable::wide_dload() {
658 transition(vtos, dtos);
659 locals_index_wide(rbx);
660 LP64_ONLY(__ movdbl(xmm0, daddress(rbx)));
661 NOT_LP64(__ fld_d(daddress(rbx)));
662 }
663
664 void TemplateTable::wide_aload() {
665 transition(vtos, atos);
666 locals_index_wide(rbx);
667 __ movptr(rax, aaddress(rbx));
668 }
669
670 void TemplateTable::index_check(Register array, Register index) {
671 // Pop ptr into array
672 __ pop_ptr(array);
673 index_check_without_pop(array, index);
674 }
675
676 void TemplateTable::index_check_without_pop(Register array, Register index) {
677 // destroys rbx
678 // check array
679 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
680 // sign extend index for use by indexed load
681 __ movl2ptr(index, index);
682 // check index
683 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
684 if (index != rbx) {
685 // ??? convention: move aberrant index into rbx for exception message
686 assert(rbx != array, "different registers");
687 __ movl(rbx, index);
688 }
689 __ jump_cc(Assembler::aboveEqual,
690 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
691 }
692
693
694 void TemplateTable::iaload() {
695 transition(itos, itos);
696 // rax: index
697 // rdx: array
698 index_check(rdx, rax); // kills rbx
699 __ movl(rax, Address(rdx, rax,
700 Address::times_4,
701 arrayOopDesc::base_offset_in_bytes(T_INT)));
702 }
703
704 void TemplateTable::laload() {
705 transition(itos, ltos);
706 // rax: index
707 // rdx: array
708 index_check(rdx, rax); // kills rbx
709 NOT_LP64(__ mov(rbx, rax));
710 // rbx,: index
711 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
712 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
713 }
714
715
716
717 void TemplateTable::faload() {
718 transition(itos, ftos);
719 // rax: index
720 // rdx: array
721 index_check(rdx, rax); // kills rbx
722 LP64_ONLY(__ movflt(xmm0, Address(rdx, rax,
723 Address::times_4,
724 arrayOopDesc::base_offset_in_bytes(T_FLOAT))));
725 NOT_LP64(__ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))));
726 }
727
728 void TemplateTable::daload() {
729 transition(itos, dtos);
730 // rax: index
731 // rdx: array
732 index_check(rdx, rax); // kills rbx
733 LP64_ONLY(__ movdbl(xmm0, Address(rdx, rax,
734 Address::times_8,
735 arrayOopDesc::base_offset_in_bytes(T_DOUBLE))));
736 NOT_LP64(__ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))));
737 }
738
739 void TemplateTable::aaload() {
740 transition(itos, atos);
741 // rax: index
742 // rdx: array
743 index_check(rdx, rax); // kills rbx
744 __ load_heap_oop(rax, Address(rdx, rax,
745 UseCompressedOops ? Address::times_4 : Address::times_ptr,
746 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
747 }
748
749 void TemplateTable::baload() {
750 transition(itos, itos);
751 // rax: index
752 // rdx: array
753 index_check(rdx, rax); // kills rbx
754 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
755 }
756
757 void TemplateTable::caload() {
758 transition(itos, itos);
759 // rax: index
760 // rdx: array
761 index_check(rdx, rax); // kills rbx
762 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
763 }
764
765 // iload followed by caload frequent pair
766 void TemplateTable::fast_icaload() {
767 transition(vtos, itos);
768 // load index out of locals
769 locals_index(rbx);
770 __ movl(rax, iaddress(rbx));
771
772 // rax: index
773 // rdx: array
774 index_check(rdx, rax); // kills rbx
775 __ load_unsigned_short(rax,
776 Address(rdx, rax,
777 Address::times_2,
778 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
779 }
780
781
782 void TemplateTable::saload() {
783 transition(itos, itos);
784 // rax: index
785 // rdx: array
786 index_check(rdx, rax); // kills rbx
787 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
788 }
789
790 void TemplateTable::iload(int n) {
791 transition(vtos, itos);
792 __ movl(rax, iaddress(n));
793 }
794
795 void TemplateTable::lload(int n) {
796 transition(vtos, ltos);
797 __ movptr(rax, laddress(n));
798 NOT_LP64(__ movptr(rdx, haddress(n)));
799 }
800
801 void TemplateTable::fload(int n) {
802 transition(vtos, ftos);
803 LP64_ONLY(__ movflt(xmm0, faddress(n)));
804 NOT_LP64(__ fld_s(faddress(n)));
805 }
806
807 void TemplateTable::dload(int n) {
808 transition(vtos, dtos);
809 LP64_ONLY(__ movdbl(xmm0, daddress(n)));
810 NOT_LP64(__ fld_d(daddress(n)));
811 }
812
813 void TemplateTable::aload(int n) {
814 transition(vtos, atos);
815 __ movptr(rax, aaddress(n));
816 }
817
818 void TemplateTable::aload_0() {
819 transition(vtos, atos);
820 // According to bytecode histograms, the pairs:
821 //
822 // _aload_0, _fast_igetfield
823 // _aload_0, _fast_agetfield
824 // _aload_0, _fast_fgetfield
825 //
826 // occur frequently. If RewriteFrequentPairs is set, the (slow)
827 // _aload_0 bytecode checks if the next bytecode is either
828 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
829 // rewrites the current bytecode into a pair bytecode; otherwise it
830 // rewrites the current bytecode into _fast_aload_0 that doesn't do
831 // the pair check anymore.
832 //
833 // Note: If the next bytecode is _getfield, the rewrite must be
834 // delayed, otherwise we may miss an opportunity for a pair.
835 //
836 // Also rewrite frequent pairs
837 // aload_0, aload_1
838 // aload_0, iload_1
839 // These bytecodes with a small amount of code are most profitable
840 // to rewrite
841 if (RewriteFrequentPairs) {
842 Label rewrite, done;
843
844 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
845 LP64_ONLY(assert(rbx != bc, "register damaged"));
846
847 // get next byte
848 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
849
850 // do actual aload_0
851 aload(0);
852
853 // if _getfield then wait with rewrite
854 __ cmpl(rbx, Bytecodes::_getfield);
855 __ jcc(Assembler::equal, done);
856
857 // if _igetfield then reqrite to _fast_iaccess_0
858 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
859 __ cmpl(rbx, Bytecodes::_fast_igetfield);
860 __ movl(bc, Bytecodes::_fast_iaccess_0);
861 __ jccb(Assembler::equal, rewrite);
862
863 // if _agetfield then reqrite to _fast_aaccess_0
864 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
865 __ cmpl(rbx, Bytecodes::_fast_agetfield);
866 __ movl(bc, Bytecodes::_fast_aaccess_0);
867 __ jccb(Assembler::equal, rewrite);
868
869 // if _fgetfield then reqrite to _fast_faccess_0
870 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
871 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
872 __ movl(bc, Bytecodes::_fast_faccess_0);
873 __ jccb(Assembler::equal, rewrite);
874
875 // else rewrite to _fast_aload0
876 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
877 __ movl(bc, Bytecodes::_fast_aload_0);
878
879 // rewrite
880 // bc: fast bytecode
881 __ bind(rewrite);
882 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
883
884 __ bind(done);
885 } else {
886 aload(0);
887 }
888 }
889
890 void TemplateTable::istore() {
891 transition(itos, vtos);
892 locals_index(rbx);
893 __ movl(iaddress(rbx), rax);
894 }
895
896
897 void TemplateTable::lstore() {
898 transition(ltos, vtos);
899 locals_index(rbx);
900 __ movptr(laddress(rbx), rax);
901 NOT_LP64(__ movptr(haddress(rbx), rdx));
902 }
903
904 void TemplateTable::fstore() {
905 transition(ftos, vtos);
906 locals_index(rbx);
907 LP64_ONLY(__ movflt(faddress(rbx), xmm0));
908 NOT_LP64(__ fstp_s(faddress(rbx)));
909 }
910
911 void TemplateTable::dstore() {
912 transition(dtos, vtos);
913 locals_index(rbx);
914 LP64_ONLY(__ movdbl(daddress(rbx), xmm0));
915 NOT_LP64(__ fstp_d(daddress(rbx)));
916 }
917
918 void TemplateTable::astore() {
919 transition(vtos, vtos);
920 __ pop_ptr(rax);
921 locals_index(rbx);
922 __ movptr(aaddress(rbx), rax);
923 }
924
925 void TemplateTable::wide_istore() {
926 transition(vtos, vtos);
927 __ pop_i();
928 locals_index_wide(rbx);
929 __ movl(iaddress(rbx), rax);
930 }
931
932 void TemplateTable::wide_lstore() {
933 transition(vtos, vtos);
934 NOT_LP64(__ pop_l(rax, rdx));
935 LP64_ONLY(__ pop_l());
936 locals_index_wide(rbx);
937 __ movptr(laddress(rbx), rax);
938 NOT_LP64(__ movl(haddress(rbx), rdx));
939 }
940
941 void TemplateTable::wide_fstore() {
942 #ifdef _LP64
943 transition(vtos, vtos);
944 __ pop_f();
945 locals_index_wide(rbx);
946 __ movflt(faddress(rbx), xmm0);
947 #else
948 wide_istore();
949 #endif
950 }
951
952 void TemplateTable::wide_dstore() {
953 #ifdef _LP64
954 transition(vtos, vtos);
955 __ pop_d();
956 locals_index_wide(rbx);
957 __ movdbl(daddress(rbx), xmm0);
958 #else
959 wide_lstore();
960 #endif
961 }
962
963 void TemplateTable::wide_astore() {
964 transition(vtos, vtos);
965 __ pop_ptr(rax);
966 locals_index_wide(rbx);
967 __ movptr(aaddress(rbx), rax);
968 }
969
970 void TemplateTable::iastore() {
971 transition(itos, vtos);
972 __ pop_i(rbx);
973 // rax: value
974 // rbx: index
975 // rdx: array
976 index_check(rdx, rbx); // prefer index in rbx
977 __ movl(Address(rdx, rbx,
978 Address::times_4,
979 arrayOopDesc::base_offset_in_bytes(T_INT)),
980 rax);
981 }
982
983 void TemplateTable::lastore() {
984 transition(ltos, vtos);
985 __ pop_i(rbx);
986 // rax,: low(value)
987 // rcx: array
988 // rdx: high(value)
989 index_check(rcx, rbx); // prefer index in rbx,
990 // rbx,: index
991 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
992 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
993 }
994
995
996 void TemplateTable::fastore() {
997 transition(ftos, vtos);
998 __ pop_i(rbx);
999 // xmm0: value
1000 // rbx: index
1001 // rdx: array
1002 index_check(rdx, rbx); // prefer index in rbx
1003 LP64_ONLY(__ movflt(Address(rdx, rbx,
1004 Address::times_4,
1005 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
1006 xmm0));
1007 NOT_LP64(__ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))));
1008 }
1009
1010 void TemplateTable::dastore() {
1011 transition(dtos, vtos);
1012 __ pop_i(rbx);
1013 // xmm0: value
1014 // rbx: index
1015 // rdx: array
1016 index_check(rdx, rbx); // prefer index in rbx
1017 LP64_ONLY(__ movdbl(Address(rdx, rbx,
1018 Address::times_8,
1019 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
1020 xmm0));
1021 NOT_LP64(__ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))));
1022 }
1023
1024 void TemplateTable::aastore() {
1025 Label is_null, ok_is_subtype, done;
1026 transition(vtos, vtos);
1027 // stack: ..., array, index, value
1028 __ movptr(rax, at_tos()); // value
1029 __ movl(rcx, at_tos_p1()); // index
1030 __ movptr(rdx, at_tos_p2()); // array
1031
1032 Address element_address(rdx, rcx,
1033 UseCompressedOops? Address::times_4 : Address::times_ptr,
1034 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1035
1036 index_check_without_pop(rdx, rcx); // kills rbx
1037 __ testptr(rax, rax);
1038 __ jcc(Assembler::zero, is_null);
1039
1040 // Move subklass into rbx
1041 __ load_klass(rbx, rax);
1042 // Move superklass into rax
1043 __ load_klass(rax, rdx);
1044 __ movptr(rax, Address(rax,
1045 ObjArrayKlass::element_klass_offset()));
1046 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1047 __ lea(rdx, element_address);
1048
1049 // Generate subtype check. Blows rcx, rdi
1050 // Superklass in rax. Subklass in rbx.
1051 __ gen_subtype_check(rbx, ok_is_subtype);
1052
1053 // Come here on failure
1054 // object is at TOS
1055 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1056
1057 // Come here on success
1058 __ bind(ok_is_subtype);
1059
1060 // Get the value we will store
1061 __ movptr(rax, at_tos());
1062 // Now store using the appropriate barrier
1063 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1064 __ jmp(done);
1065
1066 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1067 __ bind(is_null);
1068 __ profile_null_seen(rbx);
1069
1070 // Store a NULL
1071 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1072
1073 // Pop stack arguments
1074 __ bind(done);
1075 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1076 }
1077
1078 void TemplateTable::bastore() {
1079 transition(itos, vtos);
1080 __ pop_i(rbx);
1081 // rax: value
1082 // rbx: index
1083 // rdx: array
1084 index_check(rdx, rbx); // prefer index in rbx
1085 __ movb(Address(rdx, rbx,
1086 Address::times_1,
1087 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1088 rax);
1089 }
1090
1091 void TemplateTable::castore() {
1092 transition(itos, vtos);
1093 __ pop_i(rbx);
1094 // rax: value
1095 // rbx: index
1096 // rdx: array
1097 index_check(rdx, rbx); // prefer index in rbx
1098 __ movw(Address(rdx, rbx,
1099 Address::times_2,
1100 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1101 rax);
1102 }
1103
1104
1105 void TemplateTable::sastore() {
1106 castore();
1107 }
1108
1109 void TemplateTable::istore(int n) {
1110 transition(itos, vtos);
1111 __ movl(iaddress(n), rax);
1112 }
1113
1114 void TemplateTable::lstore(int n) {
1115 transition(ltos, vtos);
1116 __ movptr(laddress(n), rax);
1117 NOT_LP64(__ movptr(haddress(n), rdx));
1118 }
1119
1120 void TemplateTable::fstore(int n) {
1121 transition(ftos, vtos);
1122 LP64_ONLY(__ movflt(faddress(n), xmm0));
1123 NOT_LP64(__ fstp_s(faddress(n)));
1124 }
1125
1126 void TemplateTable::dstore(int n) {
1127 transition(dtos, vtos);
1128 LP64_ONLY(__ movdbl(daddress(n), xmm0));
1129 NOT_LP64(__ fstp_d(daddress(n)));
1130 }
1131
1132
1133 void TemplateTable::astore(int n) {
1134 transition(vtos, vtos);
1135 __ pop_ptr(rax);
1136 __ movptr(aaddress(n), rax);
1137 }
1138
1139 void TemplateTable::pop() {
1140 transition(vtos, vtos);
1141 __ addptr(rsp, Interpreter::stackElementSize);
1142 }
1143
1144 void TemplateTable::pop2() {
1145 transition(vtos, vtos);
1146 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1147 }
1148
1149
1150 void TemplateTable::dup() {
1151 transition(vtos, vtos);
1152 __ load_ptr(0, rax);
1153 __ push_ptr(rax);
1154 // stack: ..., a, a
1155 }
1156
1157 void TemplateTable::dup_x1() {
1158 transition(vtos, vtos);
1159 // stack: ..., a, b
1160 __ load_ptr( 0, rax); // load b
1161 __ load_ptr( 1, rcx); // load a
1162 __ store_ptr(1, rax); // store b
1163 __ store_ptr(0, rcx); // store a
1164 __ push_ptr(rax); // push b
1165 // stack: ..., b, a, b
1166 }
1167
1168 void TemplateTable::dup_x2() {
1169 transition(vtos, vtos);
1170 // stack: ..., a, b, c
1171 __ load_ptr( 0, rax); // load c
1172 __ load_ptr( 2, rcx); // load a
1173 __ store_ptr(2, rax); // store c in a
1174 __ push_ptr(rax); // push c
1175 // stack: ..., c, b, c, c
1176 __ load_ptr( 2, rax); // load b
1177 __ store_ptr(2, rcx); // store a in b
1178 // stack: ..., c, a, c, c
1179 __ store_ptr(1, rax); // store b in c
1180 // stack: ..., c, a, b, c
1181 }
1182
1183 void TemplateTable::dup2() {
1184 transition(vtos, vtos);
1185 // stack: ..., a, b
1186 __ load_ptr(1, rax); // load a
1187 __ push_ptr(rax); // push a
1188 __ load_ptr(1, rax); // load b
1189 __ push_ptr(rax); // push b
1190 // stack: ..., a, b, a, b
1191 }
1192
1193
1194 void TemplateTable::dup2_x1() {
1195 transition(vtos, vtos);
1196 // stack: ..., a, b, c
1197 __ load_ptr( 0, rcx); // load c
1198 __ load_ptr( 1, rax); // load b
1199 __ push_ptr(rax); // push b
1200 __ push_ptr(rcx); // push c
1201 // stack: ..., a, b, c, b, c
1202 __ store_ptr(3, rcx); // store c in b
1203 // stack: ..., a, c, c, b, c
1204 __ load_ptr( 4, rcx); // load a
1205 __ store_ptr(2, rcx); // store a in 2nd c
1206 // stack: ..., a, c, a, b, c
1207 __ store_ptr(4, rax); // store b in a
1208 // stack: ..., b, c, a, b, c
1209 }
1210
1211 void TemplateTable::dup2_x2() {
1212 transition(vtos, vtos);
1213 // stack: ..., a, b, c, d
1214 __ load_ptr( 0, rcx); // load d
1215 __ load_ptr( 1, rax); // load c
1216 __ push_ptr(rax); // push c
1217 __ push_ptr(rcx); // push d
1218 // stack: ..., a, b, c, d, c, d
1219 __ load_ptr( 4, rax); // load b
1220 __ store_ptr(2, rax); // store b in d
1221 __ store_ptr(4, rcx); // store d in b
1222 // stack: ..., a, d, c, b, c, d
1223 __ load_ptr( 5, rcx); // load a
1224 __ load_ptr( 3, rax); // load c
1225 __ store_ptr(3, rcx); // store a in c
1226 __ store_ptr(5, rax); // store c in a
1227 // stack: ..., c, d, a, b, c, d
1228 }
1229
1230 void TemplateTable::swap() {
1231 transition(vtos, vtos);
1232 // stack: ..., a, b
1233 __ load_ptr( 1, rcx); // load a
1234 __ load_ptr( 0, rax); // load b
1235 __ store_ptr(0, rcx); // store a in b
1236 __ store_ptr(1, rax); // store b in a
1237 // stack: ..., b, a
1238 }
1239
1240 void TemplateTable::iop2(Operation op) {
1241 transition(itos, itos);
1242 switch (op) {
1243 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1244 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1245 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1246 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1247 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1248 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1249 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1250 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1251 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1252 default : ShouldNotReachHere();
1253 }
1254 }
1255
1256 void TemplateTable::lop2(Operation op) {
1257 transition(ltos, ltos);
1258 #ifdef _LP64
1259 switch (op) {
1260 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1261 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1262 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1263 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1264 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1265 default : ShouldNotReachHere();
1266 }
1267 #else
1268 __ pop_l(rbx, rcx);
1269 switch (op) {
1270 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1271 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1272 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1273 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1274 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1275 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1276 default : ShouldNotReachHere();
1277 }
1278 #endif
1279 }
1280
1281 void TemplateTable::idiv() {
1282 transition(itos, itos);
1283 __ movl(rcx, rax);
1284 __ pop_i(rax);
1285 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1286 // they are not equal, one could do a normal division (no correction
1287 // needed), which may speed up this implementation for the common case.
1288 // (see also JVM spec., p.243 & p.271)
1289 __ corrected_idivl(rcx);
1290 }
1291
1292 void TemplateTable::irem() {
1293 transition(itos, itos);
1294 __ movl(rcx, rax);
1295 __ pop_i(rax);
1296 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1297 // they are not equal, one could do a normal division (no correction
1298 // needed), which may speed up this implementation for the common case.
1299 // (see also JVM spec., p.243 & p.271)
1300 __ corrected_idivl(rcx);
1301 __ movl(rax, rdx);
1302 }
1303
1304 void TemplateTable::lmul() {
1305 transition(ltos, ltos);
1306 #ifdef _LP64
1307 __ pop_l(rdx);
1308 __ imulq(rax, rdx);
1309 #else
1310 __ pop_l(rbx, rcx);
1311 __ push(rcx); __ push(rbx);
1312 __ push(rdx); __ push(rax);
1313 __ lmul(2 * wordSize, 0);
1314 __ addptr(rsp, 4 * wordSize); // take off temporaries
1315 #endif
1316 }
1317
1318 void TemplateTable::ldiv() {
1319 transition(ltos, ltos);
1320 #ifdef _LP64
1321 __ mov(rcx, rax);
1322 __ pop_l(rax);
1323 // generate explicit div0 check
1324 __ testq(rcx, rcx);
1325 __ jump_cc(Assembler::zero,
1326 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1327 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1328 // they are not equal, one could do a normal division (no correction
1329 // needed), which may speed up this implementation for the common case.
1330 // (see also JVM spec., p.243 & p.271)
1331 __ corrected_idivq(rcx); // kills rbx
1332 #else
1333 __ pop_l(rbx, rcx);
1334 __ push(rcx); __ push(rbx);
1335 __ push(rdx); __ push(rax);
1336 // check if y = 0
1337 __ orl(rax, rdx);
1338 __ jump_cc(Assembler::zero,
1339 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1340 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1341 __ addptr(rsp, 4 * wordSize); // take off temporaries
1342 #endif
1343 }
1344
1345 void TemplateTable::lrem() {
1346 transition(ltos, ltos);
1347 #ifdef _LP64
1348 __ mov(rcx, rax);
1349 __ pop_l(rax);
1350 __ testq(rcx, rcx);
1351 __ jump_cc(Assembler::zero,
1352 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1353 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1354 // they are not equal, one could do a normal division (no correction
1355 // needed), which may speed up this implementation for the common case.
1356 // (see also JVM spec., p.243 & p.271)
1357 __ corrected_idivq(rcx); // kills rbx
1358 __ mov(rax, rdx);
1359 #else
1360 __ pop_l(rbx, rcx);
1361 __ push(rcx); __ push(rbx);
1362 __ push(rdx); __ push(rax);
1363 // check if y = 0
1364 __ orl(rax, rdx);
1365 __ jump_cc(Assembler::zero,
1366 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1367 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1368 __ addptr(rsp, 4 * wordSize);
1369 #endif
1370 }
1371
1372 void TemplateTable::lshl() {
1373 transition(itos, ltos);
1374 __ movl(rcx, rax); // get shift count
1375 #ifdef _LP64
1376 __ pop_l(rax); // get shift value
1377 __ shlq(rax);
1378 #else
1379 __ pop_l(rax, rdx); // get shift value
1380 __ lshl(rdx, rax);
1381 #endif
1382 }
1383
1384 void TemplateTable::lshr() {
1385 #ifdef _LP64
1386 transition(itos, ltos);
1387 __ movl(rcx, rax); // get shift count
1388 __ pop_l(rax); // get shift value
1389 __ sarq(rax);
1390 #else
1391 transition(itos, ltos);
1392 __ mov(rcx, rax); // get shift count
1393 __ pop_l(rax, rdx); // get shift value
1394 __ lshr(rdx, rax, true);
1395 #endif
1396 }
1397
1398 void TemplateTable::lushr() {
1399 transition(itos, ltos);
1400 #ifdef _LP64
1401 __ movl(rcx, rax); // get shift count
1402 __ pop_l(rax); // get shift value
1403 __ shrq(rax);
1404 #else
1405 __ mov(rcx, rax); // get shift count
1406 __ pop_l(rax, rdx); // get shift value
1407 __ lshr(rdx, rax);
1408 #endif
1409 }
1410
1411 void TemplateTable::fop2(Operation op) {
1412 transition(ftos, ftos);
1413 #ifdef _LP64
1414 switch (op) {
1415 case add:
1416 __ addss(xmm0, at_rsp());
1417 __ addptr(rsp, Interpreter::stackElementSize);
1418 break;
1419 case sub:
1420 __ movflt(xmm1, xmm0);
1421 __ pop_f(xmm0);
1422 __ subss(xmm0, xmm1);
1423 break;
1424 case mul:
1425 __ mulss(xmm0, at_rsp());
1426 __ addptr(rsp, Interpreter::stackElementSize);
1427 break;
1428 case div:
1429 __ movflt(xmm1, xmm0);
1430 __ pop_f(xmm0);
1431 __ divss(xmm0, xmm1);
1432 break;
1433 case rem:
1434 __ movflt(xmm1, xmm0);
1435 __ pop_f(xmm0);
1436 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1437 break;
1438 default:
1439 ShouldNotReachHere();
1440 break;
1441 }
1442 #else
1443 switch (op) {
1444 case add: __ fadd_s (at_rsp()); break;
1445 case sub: __ fsubr_s(at_rsp()); break;
1446 case mul: __ fmul_s (at_rsp()); break;
1447 case div: __ fdivr_s(at_rsp()); break;
1448 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1449 default : ShouldNotReachHere();
1450 }
1451 __ f2ieee();
1452 __ pop(rax); // pop float thing off
1453 #endif
1454 }
1455
1456 void TemplateTable::dop2(Operation op) {
1457 transition(dtos, dtos);
1458 #ifdef _LP64
1459 switch (op) {
1460 case add:
1461 __ addsd(xmm0, at_rsp());
1462 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1463 break;
1464 case sub:
1465 __ movdbl(xmm1, xmm0);
1466 __ pop_d(xmm0);
1467 __ subsd(xmm0, xmm1);
1468 break;
1469 case mul:
1470 __ mulsd(xmm0, at_rsp());
1471 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1472 break;
1473 case div:
1474 __ movdbl(xmm1, xmm0);
1475 __ pop_d(xmm0);
1476 __ divsd(xmm0, xmm1);
1477 break;
1478 case rem:
1479 __ movdbl(xmm1, xmm0);
1480 __ pop_d(xmm0);
1481 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1482 break;
1483 default:
1484 ShouldNotReachHere();
1485 break;
1486 }
1487 #else
1488 switch (op) {
1489 case add: __ fadd_d (at_rsp()); break;
1490 case sub: __ fsubr_d(at_rsp()); break;
1491 case mul: {
1492 Label L_strict;
1493 Label L_join;
1494 const Address access_flags (rcx, Method::access_flags_offset());
1495 __ get_method(rcx);
1496 __ movl(rcx, access_flags);
1497 __ testl(rcx, JVM_ACC_STRICT);
1498 __ jccb(Assembler::notZero, L_strict);
1499 __ fmul_d (at_rsp());
1500 __ jmpb(L_join);
1501 __ bind(L_strict);
1502 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1503 __ fmulp();
1504 __ fmul_d (at_rsp());
1505 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1506 __ fmulp();
1507 __ bind(L_join);
1508 break;
1509 }
1510 case div: {
1511 Label L_strict;
1512 Label L_join;
1513 const Address access_flags (rcx, Method::access_flags_offset());
1514 __ get_method(rcx);
1515 __ movl(rcx, access_flags);
1516 __ testl(rcx, JVM_ACC_STRICT);
1517 __ jccb(Assembler::notZero, L_strict);
1518 __ fdivr_d(at_rsp());
1519 __ jmp(L_join);
1520 __ bind(L_strict);
1521 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1522 __ fmul_d (at_rsp());
1523 __ fdivrp();
1524 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1525 __ fmulp();
1526 __ bind(L_join);
1527 break;
1528 }
1529 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1530 default : ShouldNotReachHere();
1531 }
1532 __ d2ieee();
1533 // Pop double precision number from rsp.
1534 __ pop(rax);
1535 __ pop(rdx);
1536 #endif
1537 }
1538
1539 void TemplateTable::ineg() {
1540 transition(itos, itos);
1541 __ negl(rax);
1542 }
1543
1544 void TemplateTable::lneg() {
1545 transition(ltos, ltos);
1546 LP64_ONLY(__ negq(rax));
1547 NOT_LP64(__ lneg(rdx, rax));
1548 }
1549
1550 #ifdef _LP64
1551 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1552 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1553 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1554 // of 128-bits operands for SSE instructions.
1555 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1556 // Store the value to a 128-bits operand.
1557 operand[0] = lo;
1558 operand[1] = hi;
1559 return operand;
1560 }
1561
1562 // Buffer for 128-bits masks used by SSE instructions.
1563 static jlong float_signflip_pool[2*2];
1564 static jlong double_signflip_pool[2*2];
1565 #endif
1566
1567 void TemplateTable::fneg() {
1568 transition(ftos, ftos);
1569 #ifdef _LP64
1570 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], 0x8000000080000000, 0x8000000080000000);
1571 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1572 #else
1573 __ fchs();
1574 #endif
1575 }
1576
1577 void TemplateTable::dneg() {
1578 transition(dtos, dtos);
1579 #ifdef _LP64
1580 static jlong *double_signflip = double_quadword(&double_signflip_pool[1], 0x8000000000000000, 0x8000000000000000);
1581 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1582 #else
1583 __ fchs();
1584 #endif
1585 }
1586
1587 void TemplateTable::iinc() {
1588 transition(vtos, vtos);
1589 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1590 locals_index(rbx);
1591 __ addl(iaddress(rbx), rdx);
1592 }
1593
1594 void TemplateTable::wide_iinc() {
1595 transition(vtos, vtos);
1596 __ movl(rdx, at_bcp(4)); // get constant
1597 locals_index_wide(rbx);
1598 __ bswapl(rdx); // swap bytes & sign-extend constant
1599 __ sarl(rdx, 16);
1600 __ addl(iaddress(rbx), rdx);
1601 // Note: should probably use only one movl to get both
1602 // the index and the constant -> fix this
1603 }
1604
1605 void TemplateTable::convert() {
1606 #ifdef _LP64
1607 // Checking
1608 #ifdef ASSERT
1609 {
1610 TosState tos_in = ilgl;
1611 TosState tos_out = ilgl;
1612 switch (bytecode()) {
1613 case Bytecodes::_i2l: // fall through
1614 case Bytecodes::_i2f: // fall through
1615 case Bytecodes::_i2d: // fall through
1616 case Bytecodes::_i2b: // fall through
1617 case Bytecodes::_i2c: // fall through
1618 case Bytecodes::_i2s: tos_in = itos; break;
1619 case Bytecodes::_l2i: // fall through
1620 case Bytecodes::_l2f: // fall through
1621 case Bytecodes::_l2d: tos_in = ltos; break;
1622 case Bytecodes::_f2i: // fall through
1623 case Bytecodes::_f2l: // fall through
1624 case Bytecodes::_f2d: tos_in = ftos; break;
1625 case Bytecodes::_d2i: // fall through
1626 case Bytecodes::_d2l: // fall through
1627 case Bytecodes::_d2f: tos_in = dtos; break;
1628 default : ShouldNotReachHere();
1629 }
1630 switch (bytecode()) {
1631 case Bytecodes::_l2i: // fall through
1632 case Bytecodes::_f2i: // fall through
1633 case Bytecodes::_d2i: // fall through
1634 case Bytecodes::_i2b: // fall through
1635 case Bytecodes::_i2c: // fall through
1636 case Bytecodes::_i2s: tos_out = itos; break;
1637 case Bytecodes::_i2l: // fall through
1638 case Bytecodes::_f2l: // fall through
1639 case Bytecodes::_d2l: tos_out = ltos; break;
1640 case Bytecodes::_i2f: // fall through
1641 case Bytecodes::_l2f: // fall through
1642 case Bytecodes::_d2f: tos_out = ftos; break;
1643 case Bytecodes::_i2d: // fall through
1644 case Bytecodes::_l2d: // fall through
1645 case Bytecodes::_f2d: tos_out = dtos; break;
1646 default : ShouldNotReachHere();
1647 }
1648 transition(tos_in, tos_out);
1649 }
1650 #endif // ASSERT
1651
1652 static const int64_t is_nan = 0x8000000000000000L;
1653
1654 // Conversion
1655 switch (bytecode()) {
1656 case Bytecodes::_i2l:
1657 __ movslq(rax, rax);
1658 break;
1659 case Bytecodes::_i2f:
1660 __ cvtsi2ssl(xmm0, rax);
1661 break;
1662 case Bytecodes::_i2d:
1663 __ cvtsi2sdl(xmm0, rax);
1664 break;
1665 case Bytecodes::_i2b:
1666 __ movsbl(rax, rax);
1667 break;
1668 case Bytecodes::_i2c:
1669 __ movzwl(rax, rax);
1670 break;
1671 case Bytecodes::_i2s:
1672 __ movswl(rax, rax);
1673 break;
1674 case Bytecodes::_l2i:
1675 __ movl(rax, rax);
1676 break;
1677 case Bytecodes::_l2f:
1678 __ cvtsi2ssq(xmm0, rax);
1679 break;
1680 case Bytecodes::_l2d:
1681 __ cvtsi2sdq(xmm0, rax);
1682 break;
1683 case Bytecodes::_f2i:
1684 {
1685 Label L;
1686 __ cvttss2sil(rax, xmm0);
1687 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1688 __ jcc(Assembler::notEqual, L);
1689 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1690 __ bind(L);
1691 }
1692 break;
1693 case Bytecodes::_f2l:
1694 {
1695 Label L;
1696 __ cvttss2siq(rax, xmm0);
1697 // NaN or overflow/underflow?
1698 __ cmp64(rax, ExternalAddress((address) &is_nan));
1699 __ jcc(Assembler::notEqual, L);
1700 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1701 __ bind(L);
1702 }
1703 break;
1704 case Bytecodes::_f2d:
1705 __ cvtss2sd(xmm0, xmm0);
1706 break;
1707 case Bytecodes::_d2i:
1708 {
1709 Label L;
1710 __ cvttsd2sil(rax, xmm0);
1711 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1712 __ jcc(Assembler::notEqual, L);
1713 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1714 __ bind(L);
1715 }
1716 break;
1717 case Bytecodes::_d2l:
1718 {
1719 Label L;
1720 __ cvttsd2siq(rax, xmm0);
1721 // NaN or overflow/underflow?
1722 __ cmp64(rax, ExternalAddress((address) &is_nan));
1723 __ jcc(Assembler::notEqual, L);
1724 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1725 __ bind(L);
1726 }
1727 break;
1728 case Bytecodes::_d2f:
1729 __ cvtsd2ss(xmm0, xmm0);
1730 break;
1731 default:
1732 ShouldNotReachHere();
1733 }
1734 #else
1735 // Checking
1736 #ifdef ASSERT
1737 { TosState tos_in = ilgl;
1738 TosState tos_out = ilgl;
1739 switch (bytecode()) {
1740 case Bytecodes::_i2l: // fall through
1741 case Bytecodes::_i2f: // fall through
1742 case Bytecodes::_i2d: // fall through
1743 case Bytecodes::_i2b: // fall through
1744 case Bytecodes::_i2c: // fall through
1745 case Bytecodes::_i2s: tos_in = itos; break;
1746 case Bytecodes::_l2i: // fall through
1747 case Bytecodes::_l2f: // fall through
1748 case Bytecodes::_l2d: tos_in = ltos; break;
1749 case Bytecodes::_f2i: // fall through
1750 case Bytecodes::_f2l: // fall through
1751 case Bytecodes::_f2d: tos_in = ftos; break;
1752 case Bytecodes::_d2i: // fall through
1753 case Bytecodes::_d2l: // fall through
1754 case Bytecodes::_d2f: tos_in = dtos; break;
1755 default : ShouldNotReachHere();
1756 }
1757 switch (bytecode()) {
1758 case Bytecodes::_l2i: // fall through
1759 case Bytecodes::_f2i: // fall through
1760 case Bytecodes::_d2i: // fall through
1761 case Bytecodes::_i2b: // fall through
1762 case Bytecodes::_i2c: // fall through
1763 case Bytecodes::_i2s: tos_out = itos; break;
1764 case Bytecodes::_i2l: // fall through
1765 case Bytecodes::_f2l: // fall through
1766 case Bytecodes::_d2l: tos_out = ltos; break;
1767 case Bytecodes::_i2f: // fall through
1768 case Bytecodes::_l2f: // fall through
1769 case Bytecodes::_d2f: tos_out = ftos; break;
1770 case Bytecodes::_i2d: // fall through
1771 case Bytecodes::_l2d: // fall through
1772 case Bytecodes::_f2d: tos_out = dtos; break;
1773 default : ShouldNotReachHere();
1774 }
1775 transition(tos_in, tos_out);
1776 }
1777 #endif // ASSERT
1778
1779 // Conversion
1780 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1781 switch (bytecode()) {
1782 case Bytecodes::_i2l:
1783 __ extend_sign(rdx, rax);
1784 break;
1785 case Bytecodes::_i2f:
1786 __ push(rax); // store int on tos
1787 __ fild_s(at_rsp()); // load int to ST0
1788 __ f2ieee(); // truncate to float size
1789 __ pop(rcx); // adjust rsp
1790 break;
1791 case Bytecodes::_i2d:
1792 __ push(rax); // add one slot for d2ieee()
1793 __ push(rax); // store int on tos
1794 __ fild_s(at_rsp()); // load int to ST0
1795 __ d2ieee(); // truncate to double size
1796 __ pop(rcx); // adjust rsp
1797 __ pop(rcx);
1798 break;
1799 case Bytecodes::_i2b:
1800 __ shll(rax, 24); // truncate upper 24 bits
1801 __ sarl(rax, 24); // and sign-extend byte
1802 LP64_ONLY(__ movsbl(rax, rax));
1803 break;
1804 case Bytecodes::_i2c:
1805 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1806 LP64_ONLY(__ movzwl(rax, rax));
1807 break;
1808 case Bytecodes::_i2s:
1809 __ shll(rax, 16); // truncate upper 16 bits
1810 __ sarl(rax, 16); // and sign-extend short
1811 LP64_ONLY(__ movswl(rax, rax));
1812 break;
1813 case Bytecodes::_l2i:
1814 /* nothing to do */
1815 break;
1816 case Bytecodes::_l2f:
1817 __ push(rdx); // store long on tos
1818 __ push(rax);
1819 __ fild_d(at_rsp()); // load long to ST0
1820 __ f2ieee(); // truncate to float size
1821 __ pop(rcx); // adjust rsp
1822 __ pop(rcx);
1823 break;
1824 case Bytecodes::_l2d:
1825 __ push(rdx); // store long on tos
1826 __ push(rax);
1827 __ fild_d(at_rsp()); // load long to ST0
1828 __ d2ieee(); // truncate to double size
1829 __ pop(rcx); // adjust rsp
1830 __ pop(rcx);
1831 break;
1832 case Bytecodes::_f2i:
1833 __ push(rcx); // reserve space for argument
1834 __ fstp_s(at_rsp()); // pass float argument on stack
1835 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1836 break;
1837 case Bytecodes::_f2l:
1838 __ push(rcx); // reserve space for argument
1839 __ fstp_s(at_rsp()); // pass float argument on stack
1840 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1841 break;
1842 case Bytecodes::_f2d:
1843 /* nothing to do */
1844 break;
1845 case Bytecodes::_d2i:
1846 __ push(rcx); // reserve space for argument
1847 __ push(rcx);
1848 __ fstp_d(at_rsp()); // pass double argument on stack
1849 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1850 break;
1851 case Bytecodes::_d2l:
1852 __ push(rcx); // reserve space for argument
1853 __ push(rcx);
1854 __ fstp_d(at_rsp()); // pass double argument on stack
1855 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1856 break;
1857 case Bytecodes::_d2f:
1858 __ push(rcx); // reserve space for f2ieee()
1859 __ f2ieee(); // truncate to float size
1860 __ pop(rcx); // adjust rsp
1861 break;
1862 default :
1863 ShouldNotReachHere();
1864 }
1865 #endif
1866 }
1867
1868 void TemplateTable::lcmp() {
1869 transition(ltos, itos);
1870 #ifdef _LP64
1871 Label done;
1872 __ pop_l(rdx);
1873 __ cmpq(rdx, rax);
1874 __ movl(rax, -1);
1875 __ jccb(Assembler::less, done);
1876 __ setb(Assembler::notEqual, rax);
1877 __ movzbl(rax, rax);
1878 __ bind(done);
1879 #else
1880
1881 // y = rdx:rax
1882 __ pop_l(rbx, rcx); // get x = rcx:rbx
1883 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1884 __ mov(rax, rcx);
1885 #endif
1886 }
1887
1888 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1889 #ifdef _LP64
1890 Label done;
1891 if (is_float) {
1892 // XXX get rid of pop here, use ... reg, mem32
1893 __ pop_f(xmm1);
1894 __ ucomiss(xmm1, xmm0);
1895 } else {
1896 // XXX get rid of pop here, use ... reg, mem64
1897 __ pop_d(xmm1);
1898 __ ucomisd(xmm1, xmm0);
1899 }
1900 if (unordered_result < 0) {
1901 __ movl(rax, -1);
1902 __ jccb(Assembler::parity, done);
1903 __ jccb(Assembler::below, done);
1904 __ setb(Assembler::notEqual, rdx);
1905 __ movzbl(rax, rdx);
1906 } else {
1907 __ movl(rax, 1);
1908 __ jccb(Assembler::parity, done);
1909 __ jccb(Assembler::above, done);
1910 __ movl(rax, 0);
1911 __ jccb(Assembler::equal, done);
1912 __ decrementl(rax);
1913 }
1914 __ bind(done);
1915 #else
1916 if (is_float) {
1917 __ fld_s(at_rsp());
1918 } else {
1919 __ fld_d(at_rsp());
1920 __ pop(rdx);
1921 }
1922 __ pop(rcx);
1923 __ fcmp2int(rax, unordered_result < 0);
1924 #endif
1925 }
1926
1927 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1928 __ get_method(rcx); // rcx holds method
1929 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
1930 // holds bumped taken count
1931
1932 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1933 InvocationCounter::counter_offset();
1934 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1935 InvocationCounter::counter_offset();
1936
1937 // Load up edx with the branch displacement
1938 if (is_wide) {
1939 __ movl(rdx, at_bcp(1));
1940 } else {
1941 __ load_signed_short(rdx, at_bcp(1));
1942 }
1943 __ bswapl(rdx);
1944
1945 if (!is_wide) {
1946 __ sarl(rdx, 16);
1947 }
1948 LP64_ONLY(__ movl2ptr(rdx, rdx));
1949
1950 // Handle all the JSR stuff here, then exit.
1951 // It's much shorter and cleaner than intermingling with the non-JSR
1952 // normal-branch stuff occurring below.
1953 if (is_jsr) {
1954 // Pre-load the next target bytecode into rbx
1955 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
1956
1957 // compute return address as bci in rax
1958 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
1959 in_bytes(ConstMethod::codes_offset())));
1960 __ subptr(rax, Address(rcx, Method::const_offset()));
1961 // Adjust the bcp in r13 by the displacement in rdx
1962 __ addptr(rbcp, rdx);
1963 // jsr returns atos that is not an oop
1964 __ push_i(rax);
1965 __ dispatch_only(vtos);
1966 return;
1967 }
1968
1969 // Normal (non-jsr) branch handling
1970
1971 // Adjust the bcp in r13 by the displacement in rdx
1972 __ addptr(rbcp, rdx);
1973
1974 assert(UseLoopCounter || !UseOnStackReplacement,
1975 "on-stack-replacement requires loop counters");
1976 Label backedge_counter_overflow;
1977 Label profile_method;
1978 Label dispatch;
1979 if (UseLoopCounter) {
1980 // increment backedge counter for backward branches
1981 // rax: MDO
1982 // rbx: MDO bumped taken-count
1983 // rcx: method
1984 // rdx: target offset
1985 // r13: target bcp
1986 // r14: locals pointer
1987 __ testl(rdx, rdx); // check if forward or backward branch
1988 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1989
1990 // check if MethodCounters exists
1991 Label has_counters;
1992 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
1993 __ testptr(rax, rax);
1994 __ jcc(Assembler::notZero, has_counters);
1995 __ push(rdx);
1996 __ push(rcx);
1997 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
1998 rcx);
1999 __ pop(rcx);
2000 __ pop(rdx);
2001 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2002 __ jcc(Assembler::zero, dispatch);
2003 __ bind(has_counters);
2004
2005 if (TieredCompilation) {
2006 Label no_mdo;
2007 int increment = InvocationCounter::count_increment;
2008 if (ProfileInterpreter) {
2009 // Are we profiling?
2010 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2011 __ testptr(rbx, rbx);
2012 __ jccb(Assembler::zero, no_mdo);
2013 // Increment the MDO backedge counter
2014 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2015 in_bytes(InvocationCounter::counter_offset()));
2016 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2017 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2018 rax, false, Assembler::zero, &backedge_counter_overflow);
2019 __ jmp(dispatch);
2020 }
2021 __ bind(no_mdo);
2022 // Increment backedge counter in MethodCounters*
2023 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2024 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2025 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2026 rax, false, Assembler::zero, &backedge_counter_overflow);
2027 } else { // not TieredCompilation
2028 // increment counter
2029 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2030 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2031 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2032 __ movl(Address(rcx, be_offset), rax); // store counter
2033
2034 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2035
2036 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2037 __ addl(rax, Address(rcx, be_offset)); // add both counters
2038
2039 if (ProfileInterpreter) {
2040 // Test to see if we should create a method data oop
2041 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2042 __ jcc(Assembler::less, dispatch);
2043
2044 // if no method data exists, go to profile method
2045 __ test_method_data_pointer(rax, profile_method);
2046
2047 if (UseOnStackReplacement) {
2048 // check for overflow against rbx which is the MDO taken count
2049 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2050 __ jcc(Assembler::below, dispatch);
2051
2052 // When ProfileInterpreter is on, the backedge_count comes
2053 // from the MethodData*, which value does not get reset on
2054 // the call to frequency_counter_overflow(). To avoid
2055 // excessive calls to the overflow routine while the method is
2056 // being compiled, add a second test to make sure the overflow
2057 // function is called only once every overflow_frequency.
2058 const int overflow_frequency = 1024;
2059 __ andl(rbx, overflow_frequency - 1);
2060 __ jcc(Assembler::zero, backedge_counter_overflow);
2061
2062 }
2063 } else {
2064 if (UseOnStackReplacement) {
2065 // check for overflow against rax, which is the sum of the
2066 // counters
2067 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2068 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2069
2070 }
2071 }
2072 }
2073 __ bind(dispatch);
2074 }
2075
2076 // Pre-load the next target bytecode into rbx
2077 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2078
2079 // continue with the bytecode @ target
2080 // rax: return bci for jsr's, unused otherwise
2081 // rbx: target bytecode
2082 // r13: target bcp
2083 __ dispatch_only(vtos);
2084
2085 if (UseLoopCounter) {
2086 if (ProfileInterpreter) {
2087 // Out-of-line code to allocate method data oop.
2088 __ bind(profile_method);
2089 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2090 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2091 __ set_method_data_pointer_for_bcp();
2092 __ jmp(dispatch);
2093 }
2094
2095 if (UseOnStackReplacement) {
2096 // invocation counter overflow
2097 __ bind(backedge_counter_overflow);
2098 __ negptr(rdx);
2099 __ addptr(rdx, rbcp); // branch bcp
2100 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2101 __ call_VM(noreg,
2102 CAST_FROM_FN_PTR(address,
2103 InterpreterRuntime::frequency_counter_overflow),
2104 rdx);
2105 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2106
2107 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2108 // rbx: target bytecode
2109 // rdx: scratch
2110 // r14: locals pointer
2111 // r13: bcp
2112 __ testptr(rax, rax); // test result
2113 __ jcc(Assembler::zero, dispatch); // no osr if null
2114 // nmethod may have been invalidated (VM may block upon call_VM return)
2115 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2116 __ jcc(Assembler::notEqual, dispatch);
2117
2118 // We have the address of an on stack replacement routine in rax
2119 // We need to prepare to execute the OSR method. First we must
2120 // migrate the locals and monitors off of the stack.
2121
2122 LP64_ONLY(__ mov(r13, rax)); // save the nmethod
2123 NOT_LP64(__ mov(rbx, rax)); // save the nmethod
2124 NOT_LP64(__ get_thread(rcx));
2125
2126 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2127
2128 // rax is OSR buffer, move it to expected parameter location
2129 LP64_ONLY(__ mov(j_rarg0, rax));
2130 NOT_LP64(__ mov(rcx, rax));
2131 // We use j_rarg definitions here so that registers don't conflict as parameter
2132 // registers change across platforms as we are in the midst of a calling
2133 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2134
2135 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2136 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2137
2138
2139 // pop the interpreter frame
2140 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2141 __ leave(); // remove frame anchor
2142 __ pop(retaddr); // get return address
2143 __ mov(rsp, sender_sp); // set sp to sender sp
2144 // Ensure compiled code always sees stack at proper alignment
2145 __ andptr(rsp, -(StackAlignmentInBytes));
2146
2147 // unlike x86 we need no specialized return from compiled code
2148 // to the interpreter or the call stub.
2149
2150 // push the return address
2151 __ push(retaddr);
2152
2153 // and begin the OSR nmethod
2154 LP64_ONLY(__ jmp(Address(r13, nmethod::osr_entry_point_offset())));
2155 NOT_LP64(__ jmp(Address(rbx, nmethod::osr_entry_point_offset())));
2156 }
2157 }
2158 }
2159
2160 void TemplateTable::if_0cmp(Condition cc) {
2161 transition(itos, vtos);
2162 // assume branch is more often taken than not (loops use backward branches)
2163 Label not_taken;
2164 __ testl(rax, rax);
2165 __ jcc(j_not(cc), not_taken);
2166 branch(false, false);
2167 __ bind(not_taken);
2168 __ profile_not_taken_branch(rax);
2169 }
2170
2171 void TemplateTable::if_icmp(Condition cc) {
2172 transition(itos, vtos);
2173 // assume branch is more often taken than not (loops use backward branches)
2174 Label not_taken;
2175 __ pop_i(rdx);
2176 __ cmpl(rdx, rax);
2177 __ jcc(j_not(cc), not_taken);
2178 branch(false, false);
2179 __ bind(not_taken);
2180 __ profile_not_taken_branch(rax);
2181 }
2182
2183 void TemplateTable::if_nullcmp(Condition cc) {
2184 transition(atos, vtos);
2185 // assume branch is more often taken than not (loops use backward branches)
2186 Label not_taken;
2187 __ testptr(rax, rax);
2188 __ jcc(j_not(cc), not_taken);
2189 branch(false, false);
2190 __ bind(not_taken);
2191 __ profile_not_taken_branch(rax);
2192 }
2193
2194 void TemplateTable::if_acmp(Condition cc) {
2195 transition(atos, vtos);
2196 // assume branch is more often taken than not (loops use backward branches)
2197 Label not_taken;
2198 __ pop_ptr(rdx);
2199 __ cmpptr(rdx, rax);
2200 __ jcc(j_not(cc), not_taken);
2201 branch(false, false);
2202 __ bind(not_taken);
2203 __ profile_not_taken_branch(rax);
2204 }
2205
2206 void TemplateTable::ret() {
2207 transition(vtos, vtos);
2208 locals_index(rbx);
2209 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2210 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2211 __ profile_ret(rbx, rcx);
2212 __ get_method(rax);
2213 __ movptr(rbcp, Address(rax, Method::const_offset()));
2214 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2215 ConstMethod::codes_offset()));
2216 __ dispatch_next(vtos);
2217 }
2218
2219 void TemplateTable::wide_ret() {
2220 transition(vtos, vtos);
2221 locals_index_wide(rbx);
2222 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2223 __ profile_ret(rbx, rcx);
2224 __ get_method(rax);
2225 __ movptr(rbcp, Address(rax, Method::const_offset()));
2226 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2227 __ dispatch_next(vtos);
2228 }
2229
2230 void TemplateTable::tableswitch() {
2231 Label default_case, continue_execution;
2232 transition(itos, vtos);
2233
2234 // align r13/rsi
2235 __ lea(rbx, at_bcp(BytesPerInt));
2236 __ andptr(rbx, -BytesPerInt);
2237 // load lo & hi
2238 __ movl(rcx, Address(rbx, BytesPerInt));
2239 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2240 __ bswapl(rcx);
2241 __ bswapl(rdx);
2242 // check against lo & hi
2243 __ cmpl(rax, rcx);
2244 __ jcc(Assembler::less, default_case);
2245 __ cmpl(rax, rdx);
2246 __ jcc(Assembler::greater, default_case);
2247 // lookup dispatch offset
2248 __ subl(rax, rcx);
2249 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2250 __ profile_switch_case(rax, rbx, rcx);
2251 // continue execution
2252 __ bind(continue_execution);
2253 __ bswapl(rdx);
2254 LP64_ONLY(__ movl2ptr(rdx, rdx));
2255 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2256 __ addptr(rbcp, rdx);
2257 __ dispatch_only(vtos);
2258 // handle default
2259 __ bind(default_case);
2260 __ profile_switch_default(rax);
2261 __ movl(rdx, Address(rbx, 0));
2262 __ jmp(continue_execution);
2263 }
2264
2265 void TemplateTable::lookupswitch() {
2266 transition(itos, itos);
2267 __ stop("lookupswitch bytecode should have been rewritten");
2268 }
2269
2270 void TemplateTable::fast_linearswitch() {
2271 transition(itos, vtos);
2272 Label loop_entry, loop, found, continue_execution;
2273 // bswap rax so we can avoid bswapping the table entries
2274 __ bswapl(rax);
2275 // align r13
2276 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2277 // this instruction (change offsets
2278 // below)
2279 __ andptr(rbx, -BytesPerInt);
2280 // set counter
2281 __ movl(rcx, Address(rbx, BytesPerInt));
2282 __ bswapl(rcx);
2283 __ jmpb(loop_entry);
2284 // table search
2285 __ bind(loop);
2286 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2287 __ jcc(Assembler::equal, found);
2288 __ bind(loop_entry);
2289 __ decrementl(rcx);
2290 __ jcc(Assembler::greaterEqual, loop);
2291 // default case
2292 __ profile_switch_default(rax);
2293 __ movl(rdx, Address(rbx, 0));
2294 __ jmp(continue_execution);
2295 // entry found -> get offset
2296 __ bind(found);
2297 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2298 __ profile_switch_case(rcx, rax, rbx);
2299 // continue execution
2300 __ bind(continue_execution);
2301 __ bswapl(rdx);
2302 __ movl2ptr(rdx, rdx);
2303 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2304 __ addptr(rbcp, rdx);
2305 __ dispatch_only(vtos);
2306 }
2307
2308 void TemplateTable::fast_binaryswitch() {
2309 transition(itos, vtos);
2310 // Implementation using the following core algorithm:
2311 //
2312 // int binary_search(int key, LookupswitchPair* array, int n) {
2313 // // Binary search according to "Methodik des Programmierens" by
2314 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2315 // int i = 0;
2316 // int j = n;
2317 // while (i+1 < j) {
2318 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2319 // // with Q: for all i: 0 <= i < n: key < a[i]
2320 // // where a stands for the array and assuming that the (inexisting)
2321 // // element a[n] is infinitely big.
2322 // int h = (i + j) >> 1;
2323 // // i < h < j
2324 // if (key < array[h].fast_match()) {
2325 // j = h;
2326 // } else {
2327 // i = h;
2328 // }
2329 // }
2330 // // R: a[i] <= key < a[i+1] or Q
2331 // // (i.e., if key is within array, i is the correct index)
2332 // return i;
2333 // }
2334
2335 // Register allocation
2336 const Register key = rax; // already set (tosca)
2337 const Register array = rbx;
2338 const Register i = rcx;
2339 const Register j = rdx;
2340 const Register h = rdi;
2341 const Register temp = rsi;
2342
2343 // Find array start
2344 NOT_LP64(__ save_bcp());
2345
2346 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2347 // get rid of this
2348 // instruction (change
2349 // offsets below)
2350 __ andptr(array, -BytesPerInt);
2351
2352 // Initialize i & j
2353 __ xorl(i, i); // i = 0;
2354 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2355
2356 // Convert j into native byteordering
2357 __ bswapl(j);
2358
2359 // And start
2360 Label entry;
2361 __ jmp(entry);
2362
2363 // binary search loop
2364 {
2365 Label loop;
2366 __ bind(loop);
2367 // int h = (i + j) >> 1;
2368 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2369 __ sarl(h, 1); // h = (i + j) >> 1;
2370 // if (key < array[h].fast_match()) {
2371 // j = h;
2372 // } else {
2373 // i = h;
2374 // }
2375 // Convert array[h].match to native byte-ordering before compare
2376 __ movl(temp, Address(array, h, Address::times_8));
2377 __ bswapl(temp);
2378 __ cmpl(key, temp);
2379 // j = h if (key < array[h].fast_match())
2380 __ cmov32(Assembler::less, j, h);
2381 // i = h if (key >= array[h].fast_match())
2382 __ cmov32(Assembler::greaterEqual, i, h);
2383 // while (i+1 < j)
2384 __ bind(entry);
2385 __ leal(h, Address(i, 1)); // i+1
2386 __ cmpl(h, j); // i+1 < j
2387 __ jcc(Assembler::less, loop);
2388 }
2389
2390 // end of binary search, result index is i (must check again!)
2391 Label default_case;
2392 // Convert array[i].match to native byte-ordering before compare
2393 __ movl(temp, Address(array, i, Address::times_8));
2394 __ bswapl(temp);
2395 __ cmpl(key, temp);
2396 __ jcc(Assembler::notEqual, default_case);
2397
2398 // entry found -> j = offset
2399 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2400 __ profile_switch_case(i, key, array);
2401 __ bswapl(j);
2402 LP64_ONLY(__ movslq(j, j));
2403
2404 NOT_LP64(__ restore_bcp());
2405 NOT_LP64(__ restore_locals()); // restore rdi
2406
2407 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2408 __ addptr(rbcp, j);
2409 __ dispatch_only(vtos);
2410
2411 // default case -> j = default offset
2412 __ bind(default_case);
2413 __ profile_switch_default(i);
2414 __ movl(j, Address(array, -2 * BytesPerInt));
2415 __ bswapl(j);
2416 LP64_ONLY(__ movslq(j, j));
2417
2418 NOT_LP64(__ restore_bcp());
2419 NOT_LP64(__ restore_locals());
2420
2421 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2422 __ addptr(rbcp, j);
2423 __ dispatch_only(vtos);
2424 }
2425
2426 void TemplateTable::_return(TosState state) {
2427 transition(state, state);
2428
2429 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2430
2431 assert(_desc->calls_vm(),
2432 "inconsistent calls_vm information"); // call in remove_activation
2433
2434 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2435 assert(state == vtos, "only valid state");
2436 __ movptr(robj, aaddress(0));
2437 __ load_klass(rdi, robj);
2438 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2439 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2440 Label skip_register_finalizer;
2441 __ jcc(Assembler::zero, skip_register_finalizer);
2442
2443 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2444
2445 __ bind(skip_register_finalizer);
2446 }
2447
2448 __ remove_activation(state, rbcp);
2449 __ jmp(rbcp);
2450 }
2451
2452 // ----------------------------------------------------------------------------
2453 // Volatile variables demand their effects be made known to all CPU's
2454 // in order. Store buffers on most chips allow reads & writes to
2455 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2456 // without some kind of memory barrier (i.e., it's not sufficient that
2457 // the interpreter does not reorder volatile references, the hardware
2458 // also must not reorder them).
2459 //
2460 // According to the new Java Memory Model (JMM):
2461 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2462 // writes act as aquire & release, so:
2463 // (2) A read cannot let unrelated NON-volatile memory refs that
2464 // happen after the read float up to before the read. It's OK for
2465 // non-volatile memory refs that happen before the volatile read to
2466 // float down below it.
2467 // (3) Similar a volatile write cannot let unrelated NON-volatile
2468 // memory refs that happen BEFORE the write float down to after the
2469 // write. It's OK for non-volatile memory refs that happen after the
2470 // volatile write to float up before it.
2471 //
2472 // We only put in barriers around volatile refs (they are expensive),
2473 // not _between_ memory refs (that would require us to track the
2474 // flavor of the previous memory refs). Requirements (2) and (3)
2475 // require some barriers before volatile stores and after volatile
2476 // loads. These nearly cover requirement (1) but miss the
2477 // volatile-store-volatile-load case. This final case is placed after
2478 // volatile-stores although it could just as well go before
2479 // volatile-loads.
2480
2481 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2482 // Helper function to insert a is-volatile test and memory barrier
2483 if(!os::is_MP()) return; // Not needed on single CPU
2484 __ membar(order_constraint);
2485 }
2486
2487 void TemplateTable::resolve_cache_and_index(int byte_no,
2488 Register Rcache,
2489 Register index,
2490 size_t index_size) {
2491 const Register temp = rbx;
2492 assert_different_registers(Rcache, index, temp);
2493
2494 Label resolved;
2495 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2496 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2497 __ cmpl(temp, (int) bytecode()); // have we resolved this bytecode?
2498 __ jcc(Assembler::equal, resolved);
2499
2500 // resolve first time through
2501 address entry;
2502 switch (bytecode()) {
2503 case Bytecodes::_getstatic : // fall through
2504 case Bytecodes::_putstatic : // fall through
2505 case Bytecodes::_getfield : // fall through
2506 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2507 case Bytecodes::_invokevirtual : // fall through
2508 case Bytecodes::_invokespecial : // fall through
2509 case Bytecodes::_invokestatic : // fall through
2510 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2511 case Bytecodes::_invokehandle : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle); break;
2512 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2513 default:
2514 fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2515 break;
2516 }
2517 __ movl(temp, (int)bytecode());
2518 __ call_VM(noreg, entry, temp);
2519 // Update registers with resolved info
2520 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2521 __ bind(resolved);
2522 }
2523
2524 // The cache and index registers must be set before call
2525 void TemplateTable::load_field_cp_cache_entry(Register obj,
2526 Register cache,
2527 Register index,
2528 Register off,
2529 Register flags,
2530 bool is_static = false) {
2531 assert_different_registers(cache, index, flags, off);
2532
2533 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2534 // Field offset
2535 __ movptr(off, Address(cache, index, Address::times_ptr,
2536 in_bytes(cp_base_offset +
2537 ConstantPoolCacheEntry::f2_offset())));
2538 // Flags
2539 __ movl(flags, Address(cache, index, Address::times_ptr,
2540 in_bytes(cp_base_offset +
2541 ConstantPoolCacheEntry::flags_offset())));
2542
2543 // klass overwrite register
2544 if (is_static) {
2545 __ movptr(obj, Address(cache, index, Address::times_ptr,
2546 in_bytes(cp_base_offset +
2547 ConstantPoolCacheEntry::f1_offset())));
2548 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2549 __ movptr(obj, Address(obj, mirror_offset));
2550 }
2551 }
2552
2553 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2554 Register method,
2555 Register itable_index,
2556 Register flags,
2557 bool is_invokevirtual,
2558 bool is_invokevfinal, /*unused*/
2559 bool is_invokedynamic) {
2560 // setup registers
2561 const Register cache = rcx;
2562 const Register index = rdx;
2563 assert_different_registers(method, flags);
2564 assert_different_registers(method, cache, index);
2565 assert_different_registers(itable_index, flags);
2566 assert_different_registers(itable_index, cache, index);
2567 // determine constant pool cache field offsets
2568 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2569 const int method_offset = in_bytes(
2570 ConstantPoolCache::base_offset() +
2571 ((byte_no == f2_byte)
2572 ? ConstantPoolCacheEntry::f2_offset()
2573 : ConstantPoolCacheEntry::f1_offset()));
2574 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2575 ConstantPoolCacheEntry::flags_offset());
2576 // access constant pool cache fields
2577 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2578 ConstantPoolCacheEntry::f2_offset());
2579
2580 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2581 resolve_cache_and_index(byte_no, cache, index, index_size);
2582 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2583
2584 if (itable_index != noreg) {
2585 // pick up itable or appendix index from f2 also:
2586 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2587 }
2588 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2589 }
2590
2591 // The registers cache and index expected to be set before call.
2592 // Correct values of the cache and index registers are preserved.
2593 void TemplateTable::jvmti_post_field_access(Register cache,
2594 Register index,
2595 bool is_static,
2596 bool has_tos) {
2597 if (JvmtiExport::can_post_field_access()) {
2598 // Check to see if a field access watch has been set before we take
2599 // the time to call into the VM.
2600 Label L1;
2601 assert_different_registers(cache, index, rax);
2602 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2603 __ testl(rax,rax);
2604 __ jcc(Assembler::zero, L1);
2605
2606 // cache entry pointer
2607 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2608 __ shll(index, LogBytesPerWord);
2609 __ addptr(cache, index);
2610 if (is_static) {
2611 __ xorptr(rax, rax); // NULL object reference
2612 } else {
2613 __ pop(atos); // Get the object
2614 __ verify_oop(rax);
2615 __ push(atos); // Restore stack state
2616 }
2617 // rax,: object pointer or NULL
2618 // cache: cache entry pointer
2619 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2620 rax, cache);
2621 __ get_cache_and_index_at_bcp(cache, index, 1);
2622 __ bind(L1);
2623 }
2624 }
2625
2626 void TemplateTable::pop_and_check_object(Register r) {
2627 __ pop_ptr(r);
2628 __ null_check(r); // for field access must check obj.
2629 __ verify_oop(r);
2630 }
2631
2632 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2633 transition(vtos, vtos);
2634
2635 const Register cache = rcx;
2636 const Register index = rdx;
2637 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2638 const Register off = rbx;
2639 const Register flags = rax;
2640 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2641
2642 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2643 jvmti_post_field_access(cache, index, is_static, false);
2644 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2645
2646 if (!is_static) pop_and_check_object(obj);
2647
2648 const Address field(obj, off, Address::times_1, 0*wordSize);
2649 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2650
2651 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2652
2653 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2654 // Make sure we don't need to mask edx after the above shift
2655 assert(btos == 0, "change code, btos != 0");
2656
2657 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2658
2659 __ jcc(Assembler::notZero, notByte);
2660 // btos
2661 __ load_signed_byte(rax, field);
2662 __ push(btos);
2663 // Rewrite bytecode to be faster
2664 if (!is_static) {
2665 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2666 }
2667 __ jmp(Done);
2668
2669 __ bind(notByte);
2670 __ cmpl(flags, atos);
2671 __ jcc(Assembler::notEqual, notObj);
2672 // atos
2673 __ load_heap_oop(rax, field);
2674 __ push(atos);
2675 if (!is_static) {
2676 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2677 }
2678 __ jmp(Done);
2679
2680 __ bind(notObj);
2681 __ cmpl(flags, itos);
2682 __ jcc(Assembler::notEqual, notInt);
2683 // itos
2684 __ movl(rax, field);
2685 __ push(itos);
2686 // Rewrite bytecode to be faster
2687 if (!is_static) {
2688 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2689 }
2690 __ jmp(Done);
2691
2692 __ bind(notInt);
2693 __ cmpl(flags, ctos);
2694 __ jcc(Assembler::notEqual, notChar);
2695 // ctos
2696 __ load_unsigned_short(rax, field);
2697 __ push(ctos);
2698 // Rewrite bytecode to be faster
2699 if (!is_static) {
2700 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2701 }
2702 __ jmp(Done);
2703
2704 __ bind(notChar);
2705 __ cmpl(flags, stos);
2706 __ jcc(Assembler::notEqual, notShort);
2707 // stos
2708 __ load_signed_short(rax, field);
2709 __ push(stos);
2710 // Rewrite bytecode to be faster
2711 if (!is_static) {
2712 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2713 }
2714 __ jmp(Done);
2715
2716 __ bind(notShort);
2717 __ cmpl(flags, ltos);
2718 __ jcc(Assembler::notEqual, notLong);
2719 // ltos
2720
2721 #ifndef _LP64
2722 // Generate code as if volatile. There just aren't enough registers to
2723 // save that information and this code is faster than the test.
2724 __ fild_d(field); // Must load atomically
2725 __ subptr(rsp,2*wordSize); // Make space for store
2726 __ fistp_d(Address(rsp,0));
2727 __ pop(rax);
2728 __ pop(rdx);
2729 #else
2730 __ movq(rax, field);
2731 #endif
2732
2733 __ push(ltos);
2734 // Rewrite bytecode to be faster
2735 LP64_ONLY(if (!is_static) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2736 __ jmp(Done);
2737
2738 __ bind(notLong);
2739 __ cmpl(flags, ftos);
2740 __ jcc(Assembler::notEqual, notFloat);
2741 // ftos
2742
2743 LP64_ONLY(__ movflt(xmm0, field));
2744 NOT_LP64(__ fld_s(field));
2745 __ push(ftos);
2746 // Rewrite bytecode to be faster
2747 if (!is_static) {
2748 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2749 }
2750 __ jmp(Done);
2751
2752 __ bind(notFloat);
2753 #ifdef ASSERT
2754 __ cmpl(flags, dtos);
2755 __ jcc(Assembler::notEqual, notDouble);
2756 #endif
2757 // dtos
2758 LP64_ONLY(__ movdbl(xmm0, field));
2759 NOT_LP64(__ fld_d(field));
2760 __ push(dtos);
2761 // Rewrite bytecode to be faster
2762 if (!is_static) {
2763 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2764 }
2765 #ifdef ASSERT
2766 __ jmp(Done);
2767
2768
2769 __ bind(notDouble);
2770 __ stop("Bad state");
2771 #endif
2772
2773 __ bind(Done);
2774 // [jk] not needed currently
2775 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2776 // Assembler::LoadStore));
2777 }
2778
2779 void TemplateTable::getfield(int byte_no) {
2780 getfield_or_static(byte_no, false);
2781 }
2782
2783 void TemplateTable::getstatic(int byte_no) {
2784 getfield_or_static(byte_no, true);
2785 }
2786
2787
2788 // The registers cache and index expected to be set before call.
2789 // The function may destroy various registers, just not the cache and index registers.
2790 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2791
2792 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
2793 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2794 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2795 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
2796
2797 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2798
2799 if (JvmtiExport::can_post_field_modification()) {
2800 // Check to see if a field modification watch has been set before
2801 // we take the time to call into the VM.
2802 Label L1;
2803 assert_different_registers(cache, index, rax);
2804 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2805 __ testl(rax, rax);
2806 __ jcc(Assembler::zero, L1);
2807
2808 __ get_cache_and_index_at_bcp(robj, RDX, 1);
2809
2810
2811 if (is_static) {
2812 // Life is simple. Null out the object pointer.
2813 __ xorl(RBX, RBX);
2814
2815 } else {
2816 // Life is harder. The stack holds the value on top, followed by
2817 // the object. We don't know the size of the value, though; it
2818 // could be one or two words depending on its type. As a result,
2819 // we must find the type to determine where the object is.
2820 #ifndef _LP64
2821 Label two_word, valsize_known;
2822 #endif
2823 __ movl(RCX, Address(robj, RDX,
2824 Address::times_ptr,
2825 in_bytes(cp_base_offset +
2826 ConstantPoolCacheEntry::flags_offset())));
2827 NOT_LP64(__ mov(rbx, rsp));
2828 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
2829
2830 // Make sure we don't need to mask rcx after the above shift
2831 ConstantPoolCacheEntry::verify_tos_state_shift();
2832 #ifdef _LP64
2833 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2834 __ cmpl(c_rarg3, ltos);
2835 __ cmovptr(Assembler::equal,
2836 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2837 __ cmpl(c_rarg3, dtos);
2838 __ cmovptr(Assembler::equal,
2839 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2840 #else
2841 __ cmpl(rcx, ltos);
2842 __ jccb(Assembler::equal, two_word);
2843 __ cmpl(rcx, dtos);
2844 __ jccb(Assembler::equal, two_word);
2845 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2846 __ jmpb(valsize_known);
2847
2848 __ bind(two_word);
2849 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2850
2851 __ bind(valsize_known);
2852 // setup object pointer
2853 __ movptr(rbx, Address(rbx, 0));
2854 #endif
2855 }
2856 // cache entry pointer
2857 __ addptr(robj, in_bytes(cp_base_offset));
2858 __ shll(RDX, LogBytesPerWord);
2859 __ addptr(robj, RDX);
2860 // object (tos)
2861 __ mov(RCX, rsp);
2862 // c_rarg1: object pointer set up above (NULL if static)
2863 // c_rarg2: cache entry pointer
2864 // c_rarg3: jvalue object on the stack
2865 __ call_VM(noreg,
2866 CAST_FROM_FN_PTR(address,
2867 InterpreterRuntime::post_field_modification),
2868 RBX, robj, RCX);
2869 __ get_cache_and_index_at_bcp(cache, index, 1);
2870 __ bind(L1);
2871 }
2872 }
2873
2874 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2875 transition(vtos, vtos);
2876
2877 const Register cache = rcx;
2878 const Register index = rdx;
2879 const Register obj = rcx;
2880 const Register off = rbx;
2881 const Register flags = rax;
2882 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2883
2884 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2885 jvmti_post_field_mod(cache, index, is_static);
2886 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2887
2888 // [jk] not needed currently
2889 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2890 // Assembler::StoreStore));
2891
2892 Label notVolatile, Done;
2893 __ movl(rdx, flags);
2894 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2895 __ andl(rdx, 0x1);
2896
2897 // field addresses
2898 const Address field(obj, off, Address::times_1, 0*wordSize);
2899 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
2900
2901 Label notByte, notInt, notShort, notChar,
2902 notLong, notFloat, notObj, notDouble;
2903
2904 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2905
2906 assert(btos == 0, "change code, btos != 0");
2907 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2908 __ jcc(Assembler::notZero, notByte);
2909
2910 // btos
2911 {
2912 __ pop(btos);
2913 if (!is_static) pop_and_check_object(obj);
2914 __ movb(field, rax);
2915 if (!is_static) {
2916 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
2917 }
2918 __ jmp(Done);
2919 }
2920
2921 __ bind(notByte);
2922 __ cmpl(flags, atos);
2923 __ jcc(Assembler::notEqual, notObj);
2924
2925 // atos
2926 {
2927 __ pop(atos);
2928 if (!is_static) pop_and_check_object(obj);
2929 // Store into the field
2930 do_oop_store(_masm, field, rax, _bs->kind(), false);
2931 if (!is_static) {
2932 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
2933 }
2934 __ jmp(Done);
2935 }
2936
2937 __ bind(notObj);
2938 __ cmpl(flags, itos);
2939 __ jcc(Assembler::notEqual, notInt);
2940
2941 // itos
2942 {
2943 __ pop(itos);
2944 if (!is_static) pop_and_check_object(obj);
2945 __ movl(field, rax);
2946 if (!is_static) {
2947 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
2948 }
2949 __ jmp(Done);
2950 }
2951
2952 __ bind(notInt);
2953 __ cmpl(flags, ctos);
2954 __ jcc(Assembler::notEqual, notChar);
2955
2956 // ctos
2957 {
2958 __ pop(ctos);
2959 if (!is_static) pop_and_check_object(obj);
2960 __ movw(field, rax);
2961 if (!is_static) {
2962 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
2963 }
2964 __ jmp(Done);
2965 }
2966
2967 __ bind(notChar);
2968 __ cmpl(flags, stos);
2969 __ jcc(Assembler::notEqual, notShort);
2970
2971 // stos
2972 {
2973 __ pop(stos);
2974 if (!is_static) pop_and_check_object(obj);
2975 __ movw(field, rax);
2976 if (!is_static) {
2977 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
2978 }
2979 __ jmp(Done);
2980 }
2981
2982 __ bind(notShort);
2983 __ cmpl(flags, ltos);
2984 __ jcc(Assembler::notEqual, notLong);
2985
2986 // ltos
2987 #ifdef _LP64
2988 {
2989 __ pop(ltos);
2990 if (!is_static) pop_and_check_object(obj);
2991 __ movq(field, rax);
2992 if (!is_static) {
2993 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
2994 }
2995 __ jmp(Done);
2996 }
2997 #else
2998 {
2999 Label notVolatileLong;
3000 __ testl(rdx, rdx);
3001 __ jcc(Assembler::zero, notVolatileLong);
3002
3003 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3004 if (!is_static) pop_and_check_object(obj);
3005
3006 // Replace with real volatile test
3007 __ push(rdx);
3008 __ push(rax); // Must update atomically with FIST
3009 __ fild_d(Address(rsp,0)); // So load into FPU register
3010 __ fistp_d(field); // and put into memory atomically
3011 __ addptr(rsp, 2*wordSize);
3012 // volatile_barrier();
3013 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3014 Assembler::StoreStore));
3015 // Don't rewrite volatile version
3016 __ jmp(notVolatile);
3017
3018 __ bind(notVolatileLong);
3019
3020 __ pop(ltos); // overwrites rdx
3021 if (!is_static) pop_and_check_object(obj);
3022 __ movptr(hi, rdx);
3023 __ movptr(field, rax);
3024 // Don't rewrite to _fast_lputfield for potential volatile case.
3025 __ jmp(notVolatile);
3026 }
3027 #endif // _LP64
3028
3029 __ bind(notLong);
3030 __ cmpl(flags, ftos);
3031 __ jcc(Assembler::notEqual, notFloat);
3032
3033 // ftos
3034 {
3035 __ pop(ftos);
3036 if (!is_static) pop_and_check_object(obj);
3037 NOT_LP64( __ fstp_s(field);)
3038 LP64_ONLY( __ movflt(field, xmm0);)
3039 if (!is_static) {
3040 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3041 }
3042 __ jmp(Done);
3043 }
3044
3045 __ bind(notFloat);
3046 #ifdef ASSERT
3047 __ cmpl(flags, dtos);
3048 __ jcc(Assembler::notEqual, notDouble);
3049 #endif
3050
3051 // dtos
3052 {
3053 __ pop(dtos);
3054 if (!is_static) pop_and_check_object(obj);
3055 NOT_LP64( __ fstp_d(field);)
3056 LP64_ONLY( __ movdbl(field, xmm0);)
3057 if (!is_static) {
3058 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3059 }
3060 }
3061
3062 #ifdef ASSERT
3063 __ jmp(Done);
3064
3065 __ bind(notDouble);
3066 __ stop("Bad state");
3067 #endif
3068
3069 __ bind(Done);
3070
3071 // Check for volatile store
3072 __ testl(rdx, rdx);
3073 __ jcc(Assembler::zero, notVolatile);
3074 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3075 Assembler::StoreStore));
3076 __ bind(notVolatile);
3077 }
3078
3079 void TemplateTable::putfield(int byte_no) {
3080 putfield_or_static(byte_no, false);
3081 }
3082
3083 void TemplateTable::putstatic(int byte_no) {
3084 putfield_or_static(byte_no, true);
3085 }
3086
3087 void TemplateTable::jvmti_post_fast_field_mod() {
3088
3089 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3090
3091 if (JvmtiExport::can_post_field_modification()) {
3092 // Check to see if a field modification watch has been set before
3093 // we take the time to call into the VM.
3094 Label L2;
3095 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3096 __ testl(scratch, scratch);
3097 __ jcc(Assembler::zero, L2);
3098 __ pop_ptr(rbx); // copy the object pointer from tos
3099 __ verify_oop(rbx);
3100 __ push_ptr(rbx); // put the object pointer back on tos
3101 // Save tos values before call_VM() clobbers them. Since we have
3102 // to do it for every data type, we use the saved values as the
3103 // jvalue object.
3104 switch (bytecode()) { // load values into the jvalue object
3105 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3106 case Bytecodes::_fast_bputfield: // fall through
3107 case Bytecodes::_fast_sputfield: // fall through
3108 case Bytecodes::_fast_cputfield: // fall through
3109 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3110 case Bytecodes::_fast_dputfield: __ push_d(); break;
3111 case Bytecodes::_fast_fputfield: __ push_f(); break;
3112 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3113
3114 default:
3115 ShouldNotReachHere();
3116 }
3117 __ mov(scratch, rsp); // points to jvalue on the stack
3118 // access constant pool cache entry
3119 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3120 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3121 __ verify_oop(rbx);
3122 // rbx: object pointer copied above
3123 // c_rarg2: cache entry pointer
3124 // c_rarg3: jvalue object on the stack
3125 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3126 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3127
3128 switch (bytecode()) { // restore tos values
3129 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3130 case Bytecodes::_fast_bputfield: // fall through
3131 case Bytecodes::_fast_sputfield: // fall through
3132 case Bytecodes::_fast_cputfield: // fall through
3133 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3134 case Bytecodes::_fast_dputfield: __ pop_d(); break;
3135 case Bytecodes::_fast_fputfield: __ pop_f(); break;
3136 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3137 }
3138 __ bind(L2);
3139 }
3140 }
3141
3142 void TemplateTable::fast_storefield(TosState state) {
3143 transition(state, vtos);
3144
3145 ByteSize base = ConstantPoolCache::base_offset();
3146
3147 jvmti_post_fast_field_mod();
3148
3149 // access constant pool cache
3150 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3151
3152 // test for volatile with rdx but rdx is tos register for lputfield.
3153 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3154 in_bytes(base +
3155 ConstantPoolCacheEntry::flags_offset())));
3156
3157 // replace index with field offset from cache entry
3158 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3159 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3160
3161 // [jk] not needed currently
3162 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3163 // Assembler::StoreStore));
3164
3165 Label notVolatile;
3166 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3167 __ andl(rdx, 0x1);
3168
3169 // Get object from stack
3170 pop_and_check_object(rcx);
3171
3172 // field address
3173 const Address field(rcx, rbx, Address::times_1);
3174
3175 // access field
3176 switch (bytecode()) {
3177 case Bytecodes::_fast_aputfield:
3178 do_oop_store(_masm, field, rax, _bs->kind(), false);
3179 break;
3180 case Bytecodes::_fast_lputfield:
3181 #ifdef _LP64
3182 __ movq(field, rax);
3183 #else
3184 __ stop("should not be rewritten");
3185 #endif
3186 break;
3187 case Bytecodes::_fast_iputfield:
3188 __ movl(field, rax);
3189 break;
3190 case Bytecodes::_fast_bputfield:
3191 __ movb(field, rax);
3192 break;
3193 case Bytecodes::_fast_sputfield:
3194 // fall through
3195 case Bytecodes::_fast_cputfield:
3196 __ movw(field, rax);
3197 break;
3198 case Bytecodes::_fast_fputfield:
3199 NOT_LP64( __ fstp_s(field); )
3200 LP64_ONLY( __ movflt(field, xmm0);)
3201 break;
3202 case Bytecodes::_fast_dputfield:
3203 NOT_LP64( __ fstp_d(field); )
3204 LP64_ONLY( __ movdbl(field, xmm0);)
3205 break;
3206 default:
3207 ShouldNotReachHere();
3208 }
3209
3210 // Check for volatile store
3211 __ testl(rdx, rdx);
3212 __ jcc(Assembler::zero, notVolatile);
3213 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3214 Assembler::StoreStore));
3215 __ bind(notVolatile);
3216 }
3217
3218 void TemplateTable::fast_accessfield(TosState state) {
3219 transition(atos, state);
3220
3221 // Do the JVMTI work here to avoid disturbing the register state below
3222 if (JvmtiExport::can_post_field_access()) {
3223 // Check to see if a field access watch has been set before we
3224 // take the time to call into the VM.
3225 Label L1;
3226 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3227 __ testl(rcx, rcx);
3228 __ jcc(Assembler::zero, L1);
3229 // access constant pool cache entry
3230 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3231 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3232 __ verify_oop(rax);
3233 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3234 LP64_ONLY(__ mov(c_rarg1, rax));
3235 // c_rarg1: object pointer copied above
3236 // c_rarg2: cache entry pointer
3237 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3238 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3239 __ pop_ptr(rax); // restore object pointer
3240 __ bind(L1);
3241 }
3242
3243 // access constant pool cache
3244 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3245 // replace index with field offset from cache entry
3246 // [jk] not needed currently
3247 // if (os::is_MP()) {
3248 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3249 // in_bytes(ConstantPoolCache::base_offset() +
3250 // ConstantPoolCacheEntry::flags_offset())));
3251 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3252 // __ andl(rdx, 0x1);
3253 // }
3254 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3255 in_bytes(ConstantPoolCache::base_offset() +
3256 ConstantPoolCacheEntry::f2_offset())));
3257
3258 // rax: object
3259 __ verify_oop(rax);
3260 __ null_check(rax);
3261 Address field(rax, rbx, Address::times_1);
3262
3263 // access field
3264 switch (bytecode()) {
3265 case Bytecodes::_fast_agetfield:
3266 __ load_heap_oop(rax, field);
3267 __ verify_oop(rax);
3268 break;
3269 case Bytecodes::_fast_lgetfield:
3270 #ifdef _LP64
3271 __ movq(rax, field);
3272 #else
3273 __ stop("should not be rewritten");
3274 #endif
3275 break;
3276 case Bytecodes::_fast_igetfield:
3277 __ movl(rax, field);
3278 break;
3279 case Bytecodes::_fast_bgetfield:
3280 __ movsbl(rax, field);
3281 break;
3282 case Bytecodes::_fast_sgetfield:
3283 __ load_signed_short(rax, field);
3284 break;
3285 case Bytecodes::_fast_cgetfield:
3286 __ load_unsigned_short(rax, field);
3287 break;
3288 case Bytecodes::_fast_fgetfield:
3289 LP64_ONLY(__ movflt(xmm0, field));
3290 NOT_LP64(__ fld_s(field));
3291 break;
3292 case Bytecodes::_fast_dgetfield:
3293 LP64_ONLY(__ movdbl(xmm0, field));
3294 NOT_LP64(__ fld_d(field));
3295 break;
3296 default:
3297 ShouldNotReachHere();
3298 }
3299 // [jk] not needed currently
3300 // if (os::is_MP()) {
3301 // Label notVolatile;
3302 // __ testl(rdx, rdx);
3303 // __ jcc(Assembler::zero, notVolatile);
3304 // __ membar(Assembler::LoadLoad);
3305 // __ bind(notVolatile);
3306 //};
3307 }
3308
3309 void TemplateTable::fast_xaccess(TosState state) {
3310 transition(vtos, state);
3311
3312 // get receiver
3313 __ movptr(rax, aaddress(0));
3314 // access constant pool cache
3315 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3316 __ movptr(rbx,
3317 Address(rcx, rdx, Address::times_ptr,
3318 in_bytes(ConstantPoolCache::base_offset() +
3319 ConstantPoolCacheEntry::f2_offset())));
3320 // make sure exception is reported in correct bcp range (getfield is
3321 // next instruction)
3322 __ increment(rbcp);
3323 __ null_check(rax);
3324 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3325 switch (state) {
3326 case itos:
3327 __ movl(rax, field);
3328 break;
3329 case atos:
3330 __ load_heap_oop(rax, field);
3331 __ verify_oop(rax);
3332 break;
3333 case ftos:
3334 LP64_ONLY(__ movflt(xmm0, field));
3335 NOT_LP64(__ fld_s(field));
3336 break;
3337 default:
3338 ShouldNotReachHere();
3339 }
3340
3341 // [jk] not needed currently
3342 // if (os::is_MP()) {
3343 // Label notVolatile;
3344 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3345 // in_bytes(ConstantPoolCache::base_offset() +
3346 // ConstantPoolCacheEntry::flags_offset())));
3347 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3348 // __ testl(rdx, 0x1);
3349 // __ jcc(Assembler::zero, notVolatile);
3350 // __ membar(Assembler::LoadLoad);
3351 // __ bind(notVolatile);
3352 // }
3353
3354 __ decrement(rbcp);
3355 }
3356
3357 //-----------------------------------------------------------------------------
3358 // Calls
3359
3360 void TemplateTable::count_calls(Register method, Register temp) {
3361 // implemented elsewhere
3362 ShouldNotReachHere();
3363 }
3364
3365 void TemplateTable::prepare_invoke(int byte_no,
3366 Register method, // linked method (or i-klass)
3367 Register index, // itable index, MethodType, etc.
3368 Register recv, // if caller wants to see it
3369 Register flags // if caller wants to test it
3370 ) {
3371 // determine flags
3372 const Bytecodes::Code code = bytecode();
3373 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3374 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3375 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3376 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3377 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3378 const bool load_receiver = (recv != noreg);
3379 const bool save_flags = (flags != noreg);
3380 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3381 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3382 assert(flags == noreg || flags == rdx, "");
3383 assert(recv == noreg || recv == rcx, "");
3384
3385 // setup registers & access constant pool cache
3386 if (recv == noreg) recv = rcx;
3387 if (flags == noreg) flags = rdx;
3388 assert_different_registers(method, index, recv, flags);
3389
3390 // save 'interpreter return address'
3391 __ save_bcp();
3392
3393 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3394
3395 // maybe push appendix to arguments (just before return address)
3396 if (is_invokedynamic || is_invokehandle) {
3397 Label L_no_push;
3398 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3399 __ jcc(Assembler::zero, L_no_push);
3400 // Push the appendix as a trailing parameter.
3401 // This must be done before we get the receiver,
3402 // since the parameter_size includes it.
3403 __ push(rbx);
3404 __ mov(rbx, index);
3405 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3406 __ load_resolved_reference_at_index(index, rbx);
3407 __ pop(rbx);
3408 __ push(index); // push appendix (MethodType, CallSite, etc.)
3409 __ bind(L_no_push);
3410 }
3411
3412 // load receiver if needed (after appendix is pushed so parameter size is correct)
3413 // Note: no return address pushed yet
3414 if (load_receiver) {
3415 __ movl(recv, flags);
3416 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3417 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3418 const int receiver_is_at_end = -1; // back off one slot to get receiver
3419 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3420 __ movptr(recv, recv_addr);
3421 __ verify_oop(recv);
3422 }
3423
3424 if (save_flags) {
3425 __ movl(rbcp, flags);
3426 }
3427
3428 // compute return type
3429 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3430 // Make sure we don't need to mask flags after the above shift
3431 ConstantPoolCacheEntry::verify_tos_state_shift();
3432 // load return address
3433 {
3434 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3435 ExternalAddress table(table_addr);
3436 LP64_ONLY(__ lea(rscratch1, table));
3437 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3438 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3439 }
3440
3441 // push return address
3442 __ push(flags);
3443
3444 // Restore flags value from the constant pool cache, and restore rsi
3445 // for later null checks. r13 is the bytecode pointer
3446 if (save_flags) {
3447 __ movl(flags, rbcp);
3448 __ restore_bcp();
3449 }
3450 }
3451
3452 void TemplateTable::invokevirtual_helper(Register index,
3453 Register recv,
3454 Register flags) {
3455 // Uses temporary registers rax, rdx
3456 assert_different_registers(index, recv, rax, rdx);
3457 assert(index == rbx, "");
3458 assert(recv == rcx, "");
3459
3460 // Test for an invoke of a final method
3461 Label notFinal;
3462 __ movl(rax, flags);
3463 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3464 __ jcc(Assembler::zero, notFinal);
3465
3466 const Register method = index; // method must be rbx
3467 assert(method == rbx,
3468 "Method* must be rbx for interpreter calling convention");
3469
3470 // do the call - the index is actually the method to call
3471 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3472
3473 // It's final, need a null check here!
3474 __ null_check(recv);
3475
3476 // profile this call
3477 __ profile_final_call(rax);
3478 __ profile_arguments_type(rax, method, rbcp, true);
3479
3480 __ jump_from_interpreted(method, rax);
3481
3482 __ bind(notFinal);
3483
3484 // get receiver klass
3485 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3486 __ load_klass(rax, recv);
3487
3488 // profile this call
3489 __ profile_virtual_call(rax, rlocals, rdx);
3490 // get target Method* & entry point
3491 __ lookup_virtual_method(rax, index, method);
3492 __ profile_arguments_type(rdx, method, rbcp, true);
3493 __ jump_from_interpreted(method, rdx);
3494 }
3495
3496 void TemplateTable::invokevirtual(int byte_no) {
3497 transition(vtos, vtos);
3498 assert(byte_no == f2_byte, "use this argument");
3499 prepare_invoke(byte_no,
3500 rbx, // method or vtable index
3501 noreg, // unused itable index
3502 rcx, rdx); // recv, flags
3503
3504 // rbx: index
3505 // rcx: receiver
3506 // rdx: flags
3507
3508 invokevirtual_helper(rbx, rcx, rdx);
3509 }
3510
3511 void TemplateTable::invokespecial(int byte_no) {
3512 transition(vtos, vtos);
3513 assert(byte_no == f1_byte, "use this argument");
3514 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3515 rcx); // get receiver also for null check
3516 __ verify_oop(rcx);
3517 __ null_check(rcx);
3518 // do the call
3519 __ profile_call(rax);
3520 __ profile_arguments_type(rax, rbx, rbcp, false);
3521 __ jump_from_interpreted(rbx, rax);
3522 }
3523
3524 void TemplateTable::invokestatic(int byte_no) {
3525 transition(vtos, vtos);
3526 assert(byte_no == f1_byte, "use this argument");
3527 prepare_invoke(byte_no, rbx); // get f1 Method*
3528 // do the call
3529 __ profile_call(rax);
3530 __ profile_arguments_type(rax, rbx, rbcp, false);
3531 __ jump_from_interpreted(rbx, rax);
3532 }
3533
3534
3535 void TemplateTable::fast_invokevfinal(int byte_no) {
3536 transition(vtos, vtos);
3537 assert(byte_no == f2_byte, "use this argument");
3538 __ stop("fast_invokevfinal not used on x86");
3539 }
3540
3541
3542 void TemplateTable::invokeinterface(int byte_no) {
3543 transition(vtos, vtos);
3544 assert(byte_no == f1_byte, "use this argument");
3545 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3546 rcx, rdx); // recv, flags
3547
3548 // rax: interface klass (from f1)
3549 // rbx: itable index (from f2)
3550 // rcx: receiver
3551 // rdx: flags
3552
3553 // Special case of invokeinterface called for virtual method of
3554 // java.lang.Object. See cpCacheOop.cpp for details.
3555 // This code isn't produced by javac, but could be produced by
3556 // another compliant java compiler.
3557 Label notMethod;
3558 __ movl(rlocals, rdx);
3559 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3560
3561 __ jcc(Assembler::zero, notMethod);
3562
3563 invokevirtual_helper(rbx, rcx, rdx);
3564 __ bind(notMethod);
3565
3566 // Get receiver klass into rdx - also a null check
3567 __ restore_locals(); // restore r14
3568 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3569 __ load_klass(rdx, rcx);
3570
3571 // profile this call
3572 __ profile_virtual_call(rdx, rbcp, rlocals);
3573
3574 Label no_such_interface, no_such_method;
3575
3576 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3577 rdx, rax, rbx,
3578 // outputs: method, scan temp. reg
3579 rbx, rbcp,
3580 no_such_interface);
3581
3582 // rbx: Method* to call
3583 // rcx: receiver
3584 // Check for abstract method error
3585 // Note: This should be done more efficiently via a throw_abstract_method_error
3586 // interpreter entry point and a conditional jump to it in case of a null
3587 // method.
3588 __ testptr(rbx, rbx);
3589 __ jcc(Assembler::zero, no_such_method);
3590
3591 __ profile_arguments_type(rdx, rbx, rbcp, true);
3592
3593 // do the call
3594 // rcx: receiver
3595 // rbx,: Method*
3596 __ jump_from_interpreted(rbx, rdx);
3597 __ should_not_reach_here();
3598
3599 // exception handling code follows...
3600 // note: must restore interpreter registers to canonical
3601 // state for exception handling to work correctly!
3602
3603 __ bind(no_such_method);
3604 // throw exception
3605 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3606 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3607 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3608 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3609 // the call_VM checks for exception, so we should never return here.
3610 __ should_not_reach_here();
3611
3612 __ bind(no_such_interface);
3613 // throw exception
3614 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3615 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3616 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3617 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3618 InterpreterRuntime::throw_IncompatibleClassChangeError));
3619 // the call_VM checks for exception, so we should never return here.
3620 __ should_not_reach_here();
3621 }
3622
3623 void TemplateTable::invokehandle(int byte_no) {
3624 transition(vtos, vtos);
3625 assert(byte_no == f1_byte, "use this argument");
3626 const Register rbx_method = rbx;
3627 const Register rax_mtype = rax;
3628 const Register rcx_recv = rcx;
3629 const Register rdx_flags = rdx;
3630
3631 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3632 __ verify_method_ptr(rbx_method);
3633 __ verify_oop(rcx_recv);
3634 __ null_check(rcx_recv);
3635
3636 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3637 // rbx: MH.invokeExact_MT method (from f2)
3638
3639 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3640
3641 // FIXME: profile the LambdaForm also
3642 __ profile_final_call(rax);
3643 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3644
3645 __ jump_from_interpreted(rbx_method, rdx);
3646 }
3647
3648 void TemplateTable::invokedynamic(int byte_no) {
3649 transition(vtos, vtos);
3650 assert(byte_no == f1_byte, "use this argument");
3651
3652 const Register rbx_method = rbx;
3653 const Register rax_callsite = rax;
3654
3655 prepare_invoke(byte_no, rbx_method, rax_callsite);
3656
3657 // rax: CallSite object (from cpool->resolved_references[f1])
3658 // rbx: MH.linkToCallSite method (from f2)
3659
3660 // Note: rax_callsite is already pushed by prepare_invoke
3661
3662 // %%% should make a type profile for any invokedynamic that takes a ref argument
3663 // profile this call
3664 __ profile_call(rbcp);
3665 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3666
3667 __ verify_oop(rax_callsite);
3668
3669 __ jump_from_interpreted(rbx_method, rdx);
3670 }
3671
3672 //-----------------------------------------------------------------------------
3673 // Allocation
3674
3675 void TemplateTable::_new() {
3676 transition(vtos, atos);
3677 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3678 Label slow_case;
3679 Label slow_case_no_pop;
3680 Label done;
3681 Label initialize_header;
3682 Label initialize_object; // including clearing the fields
3683 Label allocate_shared;
3684
3685 __ get_cpool_and_tags(rcx, rax);
3686
3687 // Make sure the class we're about to instantiate has been resolved.
3688 // This is done before loading InstanceKlass to be consistent with the order
3689 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3690 const int tags_offset = Array<u1>::base_offset_in_bytes();
3691 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3692 __ jcc(Assembler::notEqual, slow_case_no_pop);
3693
3694 // get InstanceKlass
3695 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool)));
3696 __ push(rcx); // save the contexts of klass for initializing the header
3697
3698 // make sure klass is initialized & doesn't have finalizer
3699 // make sure klass is fully initialized
3700 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3701 __ jcc(Assembler::notEqual, slow_case);
3702
3703 // get instance_size in InstanceKlass (scaled to a count of bytes)
3704 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3705 // test to see if it has a finalizer or is malformed in some way
3706 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3707 __ jcc(Assembler::notZero, slow_case);
3708
3709 //
3710 // Allocate the instance
3711 // 1) Try to allocate in the TLAB
3712 // 2) if fail and the object is large allocate in the shared Eden
3713 // 3) if the above fails (or is not applicable), go to a slow case
3714 // (creates a new TLAB, etc.)
3715
3716 const bool allow_shared_alloc =
3717 Universe::heap()->supports_inline_contig_alloc();
3718
3719 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
3720 #ifndef _LP64
3721 if (UseTLAB || allow_shared_alloc) {
3722 __ get_thread(thread);
3723 }
3724 #endif // _LP64
3725
3726 if (UseTLAB) {
3727 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3728 __ lea(rbx, Address(rax, rdx, Address::times_1));
3729 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3730 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3731 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3732 if (ZeroTLAB) {
3733 // the fields have been already cleared
3734 __ jmp(initialize_header);
3735 } else {
3736 // initialize both the header and fields
3737 __ jmp(initialize_object);
3738 }
3739 }
3740
3741 // Allocation in the shared Eden, if allowed.
3742 //
3743 // rdx: instance size in bytes
3744 if (allow_shared_alloc) {
3745 __ bind(allocate_shared);
3746
3747 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3748 ExternalAddress heap_end((address)Universe::heap()->end_addr());
3749
3750 Label retry;
3751 __ bind(retry);
3752 __ movptr(rax, heap_top);
3753 __ lea(rbx, Address(rax, rdx, Address::times_1));
3754 __ cmpptr(rbx, heap_end);
3755 __ jcc(Assembler::above, slow_case);
3756
3757 // Compare rax, with the top addr, and if still equal, store the new
3758 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3759 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3760 //
3761 // rax,: object begin
3762 // rbx,: object end
3763 // rdx: instance size in bytes
3764 __ locked_cmpxchgptr(rbx, heap_top);
3765
3766 // if someone beat us on the allocation, try again, otherwise continue
3767 __ jcc(Assembler::notEqual, retry);
3768
3769 __ incr_allocated_bytes(thread, rdx, 0);
3770 }
3771
3772 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3773 // The object is initialized before the header. If the object size is
3774 // zero, go directly to the header initialization.
3775 __ bind(initialize_object);
3776 __ decrement(rdx, sizeof(oopDesc));
3777 __ jcc(Assembler::zero, initialize_header);
3778
3779 // Initialize topmost object field, divide rdx by 8, check if odd and
3780 // test if zero.
3781 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3782 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3783
3784 // rdx must have been multiple of 8
3785 #ifdef ASSERT
3786 // make sure rdx was multiple of 8
3787 Label L;
3788 // Ignore partial flag stall after shrl() since it is debug VM
3789 __ jccb(Assembler::carryClear, L);
3790 __ stop("object size is not multiple of 2 - adjust this code");
3791 __ bind(L);
3792 // rdx must be > 0, no extra check needed here
3793 #endif
3794
3795 // initialize remaining object fields: rdx was a multiple of 8
3796 { Label loop;
3797 __ bind(loop);
3798 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3799 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3800 __ decrement(rdx);
3801 __ jcc(Assembler::notZero, loop);
3802 }
3803
3804 // initialize object header only.
3805 __ bind(initialize_header);
3806 if (UseBiasedLocking) {
3807 __ pop(rcx); // get saved klass back in the register.
3808 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
3809 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3810 } else {
3811 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3812 (intptr_t)markOopDesc::prototype()); // header
3813 __ pop(rcx); // get saved klass back in the register.
3814 }
3815 #ifdef _LP64
3816 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
3817 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
3818 #endif
3819 __ store_klass(rax, rcx); // klass
3820
3821 {
3822 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3823 // Trigger dtrace event for fastpath
3824 __ push(atos);
3825 __ call_VM_leaf(
3826 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3827 __ pop(atos);
3828 }
3829
3830 __ jmp(done);
3831 }
3832
3833 // slow case
3834 __ bind(slow_case);
3835 __ pop(rcx); // restore stack pointer to what it was when we came in.
3836 __ bind(slow_case_no_pop);
3837
3838 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
3839 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3840
3841 __ get_constant_pool(rarg1);
3842 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
3843 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
3844 __ verify_oop(rax);
3845
3846 // continue
3847 __ bind(done);
3848 }
3849
3850 void TemplateTable::newarray() {
3851 transition(itos, atos);
3852 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
3853 __ load_unsigned_byte(rarg1, at_bcp(1));
3854 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3855 rarg1, rax);
3856 }
3857
3858 void TemplateTable::anewarray() {
3859 transition(itos, atos);
3860
3861 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
3862 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3863
3864 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
3865 __ get_constant_pool(rarg1);
3866 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3867 rarg1, rarg2, rax);
3868 }
3869
3870 void TemplateTable::arraylength() {
3871 transition(atos, itos);
3872 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3873 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3874 }
3875
3876 void TemplateTable::checkcast() {
3877 transition(atos, atos);
3878 Label done, is_null, ok_is_subtype, quicked, resolved;
3879 __ testptr(rax, rax); // object is in rax
3880 __ jcc(Assembler::zero, is_null);
3881
3882 // Get cpool & tags index
3883 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3884 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3885 // See if bytecode has already been quicked
3886 __ cmpb(Address(rdx, rbx,
3887 Address::times_1,
3888 Array<u1>::base_offset_in_bytes()),
3889 JVM_CONSTANT_Class);
3890 __ jcc(Assembler::equal, quicked);
3891 __ push(atos); // save receiver for result, and for GC
3892 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3893
3894 // vm_result_2 has metadata result
3895 #ifndef _LP64
3896 // borrow rdi from locals
3897 __ get_thread(rdi);
3898 __ get_vm_result_2(rax, rdi);
3899 __ restore_locals();
3900 #else
3901 __ get_vm_result_2(rax, r15_thread);
3902 #endif
3903
3904 __ pop_ptr(rdx); // restore receiver
3905 __ jmpb(resolved);
3906
3907 // Get superklass in rax and subklass in rbx
3908 __ bind(quicked);
3909 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
3910 __ movptr(rax, Address(rcx, rbx,
3911 Address::times_ptr, sizeof(ConstantPool)));
3912
3913 __ bind(resolved);
3914 __ load_klass(rbx, rdx);
3915
3916 // Generate subtype check. Blows rcx, rdi. Object in rdx.
3917 // Superklass in rax. Subklass in rbx.
3918 __ gen_subtype_check(rbx, ok_is_subtype);
3919
3920 // Come here on failure
3921 __ push_ptr(rdx);
3922 // object is at TOS
3923 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3924
3925 // Come here on success
3926 __ bind(ok_is_subtype);
3927 __ mov(rax, rdx); // Restore object in rdx
3928
3929 // Collect counts on whether this check-cast sees NULLs a lot or not.
3930 if (ProfileInterpreter) {
3931 __ jmp(done);
3932 __ bind(is_null);
3933 __ profile_null_seen(rcx);
3934 } else {
3935 __ bind(is_null); // same as 'done'
3936 }
3937 __ bind(done);
3938 }
3939
3940 void TemplateTable::instanceof() {
3941 transition(atos, itos);
3942 Label done, is_null, ok_is_subtype, quicked, resolved;
3943 __ testptr(rax, rax);
3944 __ jcc(Assembler::zero, is_null);
3945
3946 // Get cpool & tags index
3947 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3948 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3949 // See if bytecode has already been quicked
3950 __ cmpb(Address(rdx, rbx,
3951 Address::times_1,
3952 Array<u1>::base_offset_in_bytes()),
3953 JVM_CONSTANT_Class);
3954 __ jcc(Assembler::equal, quicked);
3955
3956 __ push(atos); // save receiver for result, and for GC
3957 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3958 // vm_result_2 has metadata result
3959
3960 #ifndef _LP64
3961 // borrow rdi from locals
3962 __ get_thread(rdi);
3963 __ get_vm_result_2(rax, rdi);
3964 __ restore_locals();
3965 #else
3966 __ get_vm_result_2(rax, r15_thread);
3967 #endif
3968
3969 __ pop_ptr(rdx); // restore receiver
3970 __ verify_oop(rdx);
3971 __ load_klass(rdx, rdx);
3972 __ jmpb(resolved);
3973
3974 // Get superklass in rax and subklass in rdx
3975 __ bind(quicked);
3976 __ load_klass(rdx, rax);
3977 __ movptr(rax, Address(rcx, rbx,
3978 Address::times_ptr, sizeof(ConstantPool)));
3979
3980 __ bind(resolved);
3981
3982 // Generate subtype check. Blows rcx, rdi
3983 // Superklass in rax. Subklass in rdx.
3984 __ gen_subtype_check(rdx, ok_is_subtype);
3985
3986 // Come here on failure
3987 __ xorl(rax, rax);
3988 __ jmpb(done);
3989 // Come here on success
3990 __ bind(ok_is_subtype);
3991 __ movl(rax, 1);
3992
3993 // Collect counts on whether this test sees NULLs a lot or not.
3994 if (ProfileInterpreter) {
3995 __ jmp(done);
3996 __ bind(is_null);
3997 __ profile_null_seen(rcx);
3998 } else {
3999 __ bind(is_null); // same as 'done'
4000 }
4001 __ bind(done);
4002 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4003 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4004 }
4005
4006
4007 //----------------------------------------------------------------------------------------------------
4008 // Breakpoints
4009 void TemplateTable::_breakpoint() {
4010 // Note: We get here even if we are single stepping..
4011 // jbug insists on setting breakpoints at every bytecode
4012 // even if we are in single step mode.
4013
4014 transition(vtos, vtos);
4015
4016 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4017
4018 // get the unpatched byte code
4019 __ get_method(rarg);
4020 __ call_VM(noreg,
4021 CAST_FROM_FN_PTR(address,
4022 InterpreterRuntime::get_original_bytecode_at),
4023 rarg, rbcp);
4024 __ mov(rbx, rax); // why?
4025
4026 // post the breakpoint event
4027 __ get_method(rarg);
4028 __ call_VM(noreg,
4029 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4030 rarg, rbcp);
4031
4032 // complete the execution of original bytecode
4033 __ dispatch_only_normal(vtos);
4034 }
4035
4036 //-----------------------------------------------------------------------------
4037 // Exceptions
4038
4039 void TemplateTable::athrow() {
4040 transition(atos, vtos);
4041 __ null_check(rax);
4042 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4043 }
4044
4045 //-----------------------------------------------------------------------------
4046 // Synchronization
4047 //
4048 // Note: monitorenter & exit are symmetric routines; which is reflected
4049 // in the assembly code structure as well
4050 //
4051 // Stack layout:
4052 //
4053 // [expressions ] <--- rsp = expression stack top
4054 // ..
4055 // [expressions ]
4056 // [monitor entry] <--- monitor block top = expression stack bot
4057 // ..
4058 // [monitor entry]
4059 // [frame data ] <--- monitor block bot
4060 // ...
4061 // [saved rbp ] <--- rbp
4062 void TemplateTable::monitorenter() {
4063 transition(atos, vtos);
4064
4065 // check for NULL object
4066 __ null_check(rax);
4067
4068 const Address monitor_block_top(
4069 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4070 const Address monitor_block_bot(
4071 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4072 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4073
4074 Label allocated;
4075
4076 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4077 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4078 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4079
4080 // initialize entry pointer
4081 __ xorl(rmon, rmon); // points to free slot or NULL
4082
4083 // find a free slot in the monitor block (result in rmon)
4084 {
4085 Label entry, loop, exit;
4086 __ movptr(rtop, monitor_block_top); // points to current entry,
4087 // starting with top-most entry
4088 __ lea(rbot, monitor_block_bot); // points to word before bottom
4089 // of monitor block
4090 __ jmpb(entry);
4091
4092 __ bind(loop);
4093 // check if current entry is used
4094 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4095 // if not used then remember entry in rmon
4096 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4097 // check if current entry is for same object
4098 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4099 // if same object then stop searching
4100 __ jccb(Assembler::equal, exit);
4101 // otherwise advance to next entry
4102 __ addptr(rtop, entry_size);
4103 __ bind(entry);
4104 // check if bottom reached
4105 __ cmpptr(rtop, rbot);
4106 // if not at bottom then check this entry
4107 __ jcc(Assembler::notEqual, loop);
4108 __ bind(exit);
4109 }
4110
4111 __ testptr(rmon, rmon); // check if a slot has been found
4112 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4113
4114 // allocate one if there's no free slot
4115 {
4116 Label entry, loop;
4117 // 1. compute new pointers // rsp: old expression stack top
4118 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4119 __ subptr(rsp, entry_size); // move expression stack top
4120 __ subptr(rmon, entry_size); // move expression stack bottom
4121 __ mov(rtop, rsp); // set start value for copy loop
4122 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4123 __ jmp(entry);
4124 // 2. move expression stack contents
4125 __ bind(loop);
4126 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4127 // word from old location
4128 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4129 __ addptr(rtop, wordSize); // advance to next word
4130 __ bind(entry);
4131 __ cmpptr(rtop, rmon); // check if bottom reached
4132 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4133 // copy next word
4134 }
4135
4136 // call run-time routine
4137 // rmon: points to monitor entry
4138 __ bind(allocated);
4139
4140 // Increment bcp to point to the next bytecode, so exception
4141 // handling for async. exceptions work correctly.
4142 // The object has already been poped from the stack, so the
4143 // expression stack looks correct.
4144 __ increment(rbcp);
4145
4146 // store object
4147 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4148 __ lock_object(rmon);
4149
4150 // check to make sure this monitor doesn't cause stack overflow after locking
4151 __ save_bcp(); // in case of exception
4152 __ generate_stack_overflow_check(0);
4153
4154 // The bcp has already been incremented. Just need to dispatch to
4155 // next instruction.
4156 __ dispatch_next(vtos);
4157 }
4158
4159 void TemplateTable::monitorexit() {
4160 transition(atos, vtos);
4161
4162 // check for NULL object
4163 __ null_check(rax);
4164
4165 const Address monitor_block_top(
4166 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4167 const Address monitor_block_bot(
4168 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4169 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4170
4171 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4172 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4173
4174 Label found;
4175
4176 // find matching slot
4177 {
4178 Label entry, loop;
4179 __ movptr(rtop, monitor_block_top); // points to current entry,
4180 // starting with top-most entry
4181 __ lea(rbot, monitor_block_bot); // points to word before bottom
4182 // of monitor block
4183 __ jmpb(entry);
4184
4185 __ bind(loop);
4186 // check if current entry is for same object
4187 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4188 // if same object then stop searching
4189 __ jcc(Assembler::equal, found);
4190 // otherwise advance to next entry
4191 __ addptr(rtop, entry_size);
4192 __ bind(entry);
4193 // check if bottom reached
4194 __ cmpptr(rtop, rbot);
4195 // if not at bottom then check this entry
4196 __ jcc(Assembler::notEqual, loop);
4197 }
4198
4199 // error handling. Unlocking was not block-structured
4200 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4201 InterpreterRuntime::throw_illegal_monitor_state_exception));
4202 __ should_not_reach_here();
4203
4204 // call run-time routine
4205 __ bind(found);
4206 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4207 __ unlock_object(rtop);
4208 __ pop_ptr(rax); // discard object
4209 }
4210
4211 // Wide instructions
4212 void TemplateTable::wide() {
4213 transition(vtos, vtos);
4214 __ load_unsigned_byte(rbx, at_bcp(1));
4215 ExternalAddress wtable((address)Interpreter::_wentry_point);
4216 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4217 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4218 }
4219
4220 // Multi arrays
4221 void TemplateTable::multianewarray() {
4222 transition(vtos, atos);
4223
4224 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4225 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4226 // last dim is on top of stack; we want address of first one:
4227 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4228 // the latter wordSize to point to the beginning of the array.
4229 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4230 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4231 __ load_unsigned_byte(rbx, at_bcp(3));
4232 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4233 }
4234 #endif /* !CC_INTERP */
4235