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 #define __ _masm->
42
43 // Global Register Names
44 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
45 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
46
47 // Platform-dependent initialization
48 void TemplateTable::pd_initialize() {
49 // No x86 specific initialization
50 }
51
52 // Address Computation: local variables
53 static inline Address iaddress(int n) {
54 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
55 }
56
57 static inline Address laddress(int n) {
58 return iaddress(n + 1);
59 }
60
61 #ifndef _LP64
62 static inline Address haddress(int n) {
63 return iaddress(n + 0);
64 }
65 #endif
66
67 static inline Address faddress(int n) {
68 return iaddress(n);
69 }
70
71 static inline Address daddress(int n) {
72 return laddress(n);
73 }
74
75 static inline Address aaddress(int n) {
76 return iaddress(n);
77 }
78
79 static inline Address iaddress(Register r) {
80 return Address(rlocals, r, Address::times_ptr);
81 }
82
83 static inline Address laddress(Register r) {
84 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
85 }
86
87 #ifndef _LP64
88 static inline Address haddress(Register r) {
89 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
90 }
91 #endif
92
93 static inline Address faddress(Register r) {
94 return iaddress(r);
95 }
96
97 static inline Address daddress(Register r) {
98 return laddress(r);
99 }
100
101 static inline Address aaddress(Register r) {
102 return iaddress(r);
103 }
104
105
106 // expression stack
107 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
108 // data beyond the rsp which is potentially unsafe in an MT environment;
109 // an interrupt may overwrite that data.)
110 static inline Address at_rsp () {
111 return Address(rsp, 0);
112 }
113
114 // At top of Java expression stack which may be different than esp(). It
115 // isn't for category 1 objects.
116 static inline Address at_tos () {
117 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
118 }
119
120 static inline Address at_tos_p1() {
121 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
122 }
123
124 static inline Address at_tos_p2() {
125 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
126 }
127
128 // Condition conversion
129 static Assembler::Condition j_not(TemplateTable::Condition cc) {
130 switch (cc) {
131 case TemplateTable::equal : return Assembler::notEqual;
132 case TemplateTable::not_equal : return Assembler::equal;
133 case TemplateTable::less : return Assembler::greaterEqual;
134 case TemplateTable::less_equal : return Assembler::greater;
135 case TemplateTable::greater : return Assembler::lessEqual;
136 case TemplateTable::greater_equal: return Assembler::less;
137 }
138 ShouldNotReachHere();
139 return Assembler::zero;
140 }
141
142
143
144 // Miscelaneous helper routines
145 // Store an oop (or NULL) at the address described by obj.
146 // If val == noreg this means store a NULL
147
148
149 static void do_oop_store(InterpreterMacroAssembler* _masm,
150 Address obj,
151 Register val,
152 BarrierSet::Name barrier,
153 bool precise) {
154 assert(val == noreg || val == rax, "parameter is just for looks");
155 switch (barrier) {
156 #if INCLUDE_ALL_GCS
157 case BarrierSet::G1SATBCTLogging:
158 {
159 // flatten object address if needed
160 // We do it regardless of precise because we need the registers
161 if (obj.index() == noreg && obj.disp() == 0) {
162 if (obj.base() != rdx) {
163 __ movptr(rdx, obj.base());
164 }
165 } else {
166 __ lea(rdx, obj);
167 }
168
169 Register rtmp = LP64_ONLY(r8) NOT_LP64(rsi);
170 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
171
172 NOT_LP64(__ get_thread(rcx));
173 NOT_LP64(__ save_bcp());
174
175 __ g1_write_barrier_pre(rdx /* obj */,
176 rbx /* pre_val */,
177 rthread /* thread */,
178 rtmp /* tmp */,
179 val != noreg /* tosca_live */,
180 false /* expand_call */);
181 if (val == noreg) {
182 __ store_heap_oop_null(Address(rdx, 0));
183 } else {
184 // G1 barrier needs uncompressed oop for region cross check.
185 Register new_val = val;
186 if (UseCompressedOops) {
187 new_val = rbx;
188 __ movptr(new_val, val);
189 }
190 __ store_heap_oop(Address(rdx, 0), val);
191 __ g1_write_barrier_post(rdx /* store_adr */,
192 new_val /* new_val */,
193 rthread /* thread */,
194 rtmp /* tmp */,
195 rbx /* tmp2 */);
196 }
197 NOT_LP64( __ restore_bcp());
198 }
199 break;
200 #endif // INCLUDE_ALL_GCS
201 case BarrierSet::CardTableForRS:
202 case BarrierSet::CardTableExtension:
203 {
204 if (val == noreg) {
205 __ store_heap_oop_null(obj);
206 } else {
207 __ store_heap_oop(obj, val);
208 // flatten object address if needed
209 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
210 __ store_check(obj.base());
211 } else {
212 __ lea(rdx, obj);
213 __ store_check(rdx);
214 }
215 }
216 }
217 break;
218 case BarrierSet::ModRef:
219 if (val == noreg) {
220 __ store_heap_oop_null(obj);
221 } else {
222 __ store_heap_oop(obj, val);
223 }
224 break;
225 default :
226 ShouldNotReachHere();
227
228 }
229 }
230
231 Address TemplateTable::at_bcp(int offset) {
232 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
233 return Address(rbcp, offset);
234 }
235
236
237 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
238 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
239 int byte_no) {
240 if (!RewriteBytecodes) return;
241 Label L_patch_done;
242
243 switch (bc) {
244 case Bytecodes::_fast_aputfield:
245 case Bytecodes::_fast_bputfield:
246 case Bytecodes::_fast_zputfield:
247 case Bytecodes::_fast_cputfield:
248 case Bytecodes::_fast_dputfield:
249 case Bytecodes::_fast_fputfield:
250 case Bytecodes::_fast_iputfield:
251 case Bytecodes::_fast_lputfield:
252 case Bytecodes::_fast_sputfield:
253 {
254 // We skip bytecode quickening for putfield instructions when
255 // the put_code written to the constant pool cache is zero.
256 // This is required so that every execution of this instruction
257 // calls out to InterpreterRuntime::resolve_get_put to do
258 // additional, required work.
259 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
260 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
261 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
262 __ movl(bc_reg, bc);
263 __ cmpl(temp_reg, (int) 0);
264 __ jcc(Assembler::zero, L_patch_done); // don't patch
265 }
266 break;
267 default:
268 assert(byte_no == -1, "sanity");
269 // the pair bytecodes have already done the load.
270 if (load_bc_into_bc_reg) {
271 __ movl(bc_reg, bc);
272 }
273 }
274
275 if (JvmtiExport::can_post_breakpoint()) {
276 Label L_fast_patch;
277 // if a breakpoint is present we can't rewrite the stream directly
278 __ movzbl(temp_reg, at_bcp(0));
279 __ cmpl(temp_reg, Bytecodes::_breakpoint);
280 __ jcc(Assembler::notEqual, L_fast_patch);
281 __ get_method(temp_reg);
282 // Let breakpoint table handling rewrite to quicker bytecode
283 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
284 #ifndef ASSERT
285 __ jmpb(L_patch_done);
286 #else
287 __ jmp(L_patch_done);
288 #endif
289 __ bind(L_fast_patch);
290 }
291
292 #ifdef ASSERT
293 Label L_okay;
294 __ load_unsigned_byte(temp_reg, at_bcp(0));
295 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
296 __ jcc(Assembler::equal, L_okay);
297 __ cmpl(temp_reg, bc_reg);
298 __ jcc(Assembler::equal, L_okay);
299 __ stop("patching the wrong bytecode");
300 __ bind(L_okay);
301 #endif
302
303 // patch bytecode
304 __ movb(at_bcp(0), bc_reg);
305 __ bind(L_patch_done);
306 }
307 // Individual instructions
308
309
310 void TemplateTable::nop() {
311 transition(vtos, vtos);
312 // nothing to do
313 }
314
315 void TemplateTable::shouldnotreachhere() {
316 transition(vtos, vtos);
317 __ stop("shouldnotreachhere bytecode");
318 }
319
320 void TemplateTable::aconst_null() {
321 transition(vtos, atos);
322 __ xorl(rax, rax);
323 }
324
325 void TemplateTable::iconst(int value) {
326 transition(vtos, itos);
327 if (value == 0) {
328 __ xorl(rax, rax);
329 } else {
330 __ movl(rax, value);
331 }
332 }
333
334 void TemplateTable::lconst(int value) {
335 transition(vtos, ltos);
336 if (value == 0) {
337 __ xorl(rax, rax);
338 } else {
339 __ movl(rax, value);
340 }
341 #ifndef _LP64
342 assert(value >= 0, "check this code");
343 __ xorptr(rdx, rdx);
344 #endif
345 }
346
347
348
349 void TemplateTable::fconst(int value) {
350 transition(vtos, ftos);
351 if (UseSSE >= 1) {
352 static float one = 1.0f, two = 2.0f;
353 switch (value) {
354 case 0:
355 __ xorps(xmm0, xmm0);
356 break;
357 case 1:
358 __ movflt(xmm0, ExternalAddress((address) &one));
359 break;
360 case 2:
361 __ movflt(xmm0, ExternalAddress((address) &two));
362 break;
363 default:
364 ShouldNotReachHere();
365 break;
366 }
367 } else {
368 #ifdef _LP64
369 ShouldNotReachHere();
370 #else
371 if (value == 0) { __ fldz();
372 } else if (value == 1) { __ fld1();
373 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
374 } else { ShouldNotReachHere();
375 }
376 #endif // _LP64
377 }
378 }
379
380 void TemplateTable::dconst(int value) {
381 transition(vtos, dtos);
382 if (UseSSE >= 2) {
383 static double one = 1.0;
384 switch (value) {
385 case 0:
386 __ xorpd(xmm0, xmm0);
387 break;
388 case 1:
389 __ movdbl(xmm0, ExternalAddress((address) &one));
390 break;
391 default:
392 ShouldNotReachHere();
393 break;
394 }
395 } else {
396 #ifdef _LP64
397 ShouldNotReachHere();
398 #else
399 if (value == 0) { __ fldz();
400 } else if (value == 1) { __ fld1();
401 } else { ShouldNotReachHere();
402 }
403 #endif
404 }
405 }
406
407 void TemplateTable::bipush() {
408 transition(vtos, itos);
409 __ load_signed_byte(rax, at_bcp(1));
410 }
411
412 void TemplateTable::sipush() {
413 transition(vtos, itos);
414 __ load_unsigned_short(rax, at_bcp(1));
415 __ bswapl(rax);
416 __ sarl(rax, 16);
417 }
418
419 void TemplateTable::ldc(bool wide) {
420 transition(vtos, vtos);
421 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
422 Label call_ldc, notFloat, notClass, Done;
423
424 if (wide) {
425 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
426 } else {
427 __ load_unsigned_byte(rbx, at_bcp(1));
428 }
429
430 __ get_cpool_and_tags(rcx, rax);
431 const int base_offset = ConstantPool::header_size() * wordSize;
432 const int tags_offset = Array<u1>::base_offset_in_bytes();
433
434 // get type
435 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
436
437 // unresolved class - get the resolved class
438 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
439 __ jccb(Assembler::equal, call_ldc);
440
441 // unresolved class in error state - call into runtime to throw the error
442 // from the first resolution attempt
443 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
444 __ jccb(Assembler::equal, call_ldc);
445
446 // resolved class - need to call vm to get java mirror of the class
447 __ cmpl(rdx, JVM_CONSTANT_Class);
448 __ jcc(Assembler::notEqual, notClass);
449
450 __ bind(call_ldc);
451
452 __ movl(rarg, wide);
453 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
454
455 __ push(atos);
456 __ jmp(Done);
457
458 __ bind(notClass);
459 __ cmpl(rdx, JVM_CONSTANT_Float);
460 __ jccb(Assembler::notEqual, notFloat);
461
462 // ftos
463 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
464 __ push(ftos);
465 __ jmp(Done);
466
467 __ bind(notFloat);
468 #ifdef ASSERT
469 {
470 Label L;
471 __ cmpl(rdx, JVM_CONSTANT_Integer);
472 __ jcc(Assembler::equal, L);
473 // String and Object are rewritten to fast_aldc
474 __ stop("unexpected tag type in ldc");
475 __ bind(L);
476 }
477 #endif
478 // itos JVM_CONSTANT_Integer only
479 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
480 __ push(itos);
481 __ bind(Done);
482 }
483
484 // Fast path for caching oop constants.
485 void TemplateTable::fast_aldc(bool wide) {
486 transition(vtos, atos);
487
488 Register result = rax;
489 Register tmp = rdx;
490 int index_size = wide ? sizeof(u2) : sizeof(u1);
491
492 Label resolved;
493
494 // We are resolved if the resolved reference cache entry contains a
495 // non-null object (String, MethodType, etc.)
496 assert_different_registers(result, tmp);
497 __ get_cache_index_at_bcp(tmp, 1, index_size);
498 __ load_resolved_reference_at_index(result, tmp);
499 __ testl(result, result);
500 __ jcc(Assembler::notZero, resolved);
501
502 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
503
504 // first time invocation - must resolve first
505 __ movl(tmp, (int)bytecode());
506 __ call_VM(result, entry, tmp);
507
508 __ bind(resolved);
509
510 if (VerifyOops) {
511 __ verify_oop(result);
512 }
513 }
514
515 void TemplateTable::ldc2_w() {
516 transition(vtos, vtos);
517 Label Long, Done;
518 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
519
520 __ get_cpool_and_tags(rcx, rax);
521 const int base_offset = ConstantPool::header_size() * wordSize;
522 const int tags_offset = Array<u1>::base_offset_in_bytes();
523
524 // get type
525 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
526 JVM_CONSTANT_Double);
527 __ jccb(Assembler::notEqual, Long);
528
529 // dtos
530 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
531 __ push(dtos);
532
533 __ jmpb(Done);
534 __ bind(Long);
535
536 // ltos
537 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
538 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
539 __ push(ltos);
540
541 __ bind(Done);
542 }
543
544 void TemplateTable::locals_index(Register reg, int offset) {
545 __ load_unsigned_byte(reg, at_bcp(offset));
546 __ negptr(reg);
547 }
548
549 void TemplateTable::iload() {
550 iload_internal();
551 }
552
553 void TemplateTable::nofast_iload() {
554 iload_internal(may_not_rewrite);
555 }
556
557 void TemplateTable::iload_internal(RewriteControl rc) {
558 transition(vtos, itos);
559 if (RewriteFrequentPairs && rc == may_rewrite) {
560 Label rewrite, done;
561 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
562 LP64_ONLY(assert(rbx != bc, "register damaged"));
563
564 // get next byte
565 __ load_unsigned_byte(rbx,
566 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
567 // if _iload, wait to rewrite to iload2. We only want to rewrite the
568 // last two iloads in a pair. Comparing against fast_iload means that
569 // the next bytecode is neither an iload or a caload, and therefore
570 // an iload pair.
571 __ cmpl(rbx, Bytecodes::_iload);
572 __ jcc(Assembler::equal, done);
573
574 __ cmpl(rbx, Bytecodes::_fast_iload);
575 __ movl(bc, Bytecodes::_fast_iload2);
576
577 __ jccb(Assembler::equal, rewrite);
578
579 // if _caload, rewrite to fast_icaload
580 __ cmpl(rbx, Bytecodes::_caload);
581 __ movl(bc, Bytecodes::_fast_icaload);
582 __ jccb(Assembler::equal, rewrite);
583
584 // rewrite so iload doesn't check again.
585 __ movl(bc, Bytecodes::_fast_iload);
586
587 // rewrite
588 // bc: fast bytecode
589 __ bind(rewrite);
590 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
591 __ bind(done);
592 }
593
594 // Get the local value into tos
595 locals_index(rbx);
596 __ movl(rax, iaddress(rbx));
597 }
598
599 void TemplateTable::fast_iload2() {
600 transition(vtos, itos);
601 locals_index(rbx);
602 __ movl(rax, iaddress(rbx));
603 __ push(itos);
604 locals_index(rbx, 3);
605 __ movl(rax, iaddress(rbx));
606 }
607
608 void TemplateTable::fast_iload() {
609 transition(vtos, itos);
610 locals_index(rbx);
611 __ movl(rax, iaddress(rbx));
612 }
613
614 void TemplateTable::lload() {
615 transition(vtos, ltos);
616 locals_index(rbx);
617 __ movptr(rax, laddress(rbx));
618 NOT_LP64(__ movl(rdx, haddress(rbx)));
619 }
620
621 void TemplateTable::fload() {
622 transition(vtos, ftos);
623 locals_index(rbx);
624 __ load_float(faddress(rbx));
625 }
626
627 void TemplateTable::dload() {
628 transition(vtos, dtos);
629 locals_index(rbx);
630 __ load_double(daddress(rbx));
631 }
632
633 void TemplateTable::aload() {
634 transition(vtos, atos);
635 locals_index(rbx);
636 __ movptr(rax, aaddress(rbx));
637 }
638
639 void TemplateTable::locals_index_wide(Register reg) {
640 __ load_unsigned_short(reg, at_bcp(2));
641 __ bswapl(reg);
642 __ shrl(reg, 16);
643 __ negptr(reg);
644 }
645
646 void TemplateTable::wide_iload() {
647 transition(vtos, itos);
648 locals_index_wide(rbx);
649 __ movl(rax, iaddress(rbx));
650 }
651
652 void TemplateTable::wide_lload() {
653 transition(vtos, ltos);
654 locals_index_wide(rbx);
655 __ movptr(rax, laddress(rbx));
656 NOT_LP64(__ movl(rdx, haddress(rbx)));
657 }
658
659 void TemplateTable::wide_fload() {
660 transition(vtos, ftos);
661 locals_index_wide(rbx);
662 __ load_float(faddress(rbx));
663 }
664
665 void TemplateTable::wide_dload() {
666 transition(vtos, dtos);
667 locals_index_wide(rbx);
668 __ load_double(daddress(rbx));
669 }
670
671 void TemplateTable::wide_aload() {
672 transition(vtos, atos);
673 locals_index_wide(rbx);
674 __ movptr(rax, aaddress(rbx));
675 }
676
677 void TemplateTable::index_check(Register array, Register index) {
678 // Pop ptr into array
679 __ pop_ptr(array);
680 index_check_without_pop(array, index);
681 }
682
683 void TemplateTable::index_check_without_pop(Register array, Register index) {
684 // destroys rbx
685 // check array
686 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
687 // sign extend index for use by indexed load
688 __ movl2ptr(index, index);
689 // check index
690 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
691 if (index != rbx) {
692 // ??? convention: move aberrant index into rbx for exception message
693 assert(rbx != array, "different registers");
694 __ movl(rbx, index);
695 }
696 __ jump_cc(Assembler::aboveEqual,
697 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
698 }
699
700
701 void TemplateTable::iaload() {
702 transition(itos, itos);
703 // rax: index
704 // rdx: array
705 index_check(rdx, rax); // kills rbx
706 __ movl(rax, Address(rdx, rax,
707 Address::times_4,
708 arrayOopDesc::base_offset_in_bytes(T_INT)));
709 }
710
711 void TemplateTable::laload() {
712 transition(itos, ltos);
713 // rax: index
714 // rdx: array
715 index_check(rdx, rax); // kills rbx
716 NOT_LP64(__ mov(rbx, rax));
717 // rbx,: index
718 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
719 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
720 }
721
722
723
724 void TemplateTable::faload() {
725 transition(itos, ftos);
726 // rax: index
727 // rdx: array
728 index_check(rdx, rax); // kills rbx
729 __ load_float(Address(rdx, rax,
730 Address::times_4,
731 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
732 }
733
734 void TemplateTable::daload() {
735 transition(itos, dtos);
736 // rax: index
737 // rdx: array
738 index_check(rdx, rax); // kills rbx
739 __ load_double(Address(rdx, rax,
740 Address::times_8,
741 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
742 }
743
744 void TemplateTable::aaload() {
745 transition(itos, atos);
746 // rax: index
747 // rdx: array
748 index_check(rdx, rax); // kills rbx
749 __ load_heap_oop(rax, Address(rdx, rax,
750 UseCompressedOops ? Address::times_4 : Address::times_ptr,
751 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
752 }
753
754 void TemplateTable::baload() {
755 transition(itos, itos);
756 // rax: index
757 // rdx: array
758 index_check(rdx, rax); // kills rbx
759 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
760 }
761
762 void TemplateTable::caload() {
763 transition(itos, itos);
764 // rax: index
765 // rdx: array
766 index_check(rdx, rax); // kills rbx
767 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
768 }
769
770 // iload followed by caload frequent pair
771 void TemplateTable::fast_icaload() {
772 transition(vtos, itos);
773 // load index out of locals
774 locals_index(rbx);
775 __ movl(rax, iaddress(rbx));
776
777 // rax: index
778 // rdx: array
779 index_check(rdx, rax); // kills rbx
780 __ load_unsigned_short(rax,
781 Address(rdx, rax,
782 Address::times_2,
783 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
784 }
785
786
787 void TemplateTable::saload() {
788 transition(itos, itos);
789 // rax: index
790 // rdx: array
791 index_check(rdx, rax); // kills rbx
792 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
793 }
794
795 void TemplateTable::iload(int n) {
796 transition(vtos, itos);
797 __ movl(rax, iaddress(n));
798 }
799
800 void TemplateTable::lload(int n) {
801 transition(vtos, ltos);
802 __ movptr(rax, laddress(n));
803 NOT_LP64(__ movptr(rdx, haddress(n)));
804 }
805
806 void TemplateTable::fload(int n) {
807 transition(vtos, ftos);
808 __ load_float(faddress(n));
809 }
810
811 void TemplateTable::dload(int n) {
812 transition(vtos, dtos);
813 __ load_double(daddress(n));
814 }
815
816 void TemplateTable::aload(int n) {
817 transition(vtos, atos);
818 __ movptr(rax, aaddress(n));
819 }
820
821 void TemplateTable::aload_0() {
822 aload_0_internal();
823 }
824
825 void TemplateTable::nofast_aload_0() {
826 aload_0_internal(may_not_rewrite);
827 }
828
829 void TemplateTable::aload_0_internal(RewriteControl rc) {
830 transition(vtos, atos);
831 // According to bytecode histograms, the pairs:
832 //
833 // _aload_0, _fast_igetfield
834 // _aload_0, _fast_agetfield
835 // _aload_0, _fast_fgetfield
836 //
837 // occur frequently. If RewriteFrequentPairs is set, the (slow)
838 // _aload_0 bytecode checks if the next bytecode is either
839 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
840 // rewrites the current bytecode into a pair bytecode; otherwise it
841 // rewrites the current bytecode into _fast_aload_0 that doesn't do
842 // the pair check anymore.
843 //
844 // Note: If the next bytecode is _getfield, the rewrite must be
845 // delayed, otherwise we may miss an opportunity for a pair.
846 //
847 // Also rewrite frequent pairs
848 // aload_0, aload_1
849 // aload_0, iload_1
850 // These bytecodes with a small amount of code are most profitable
851 // to rewrite
852 if (RewriteFrequentPairs && rc == may_rewrite) {
853 Label rewrite, done;
854
855 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
856 LP64_ONLY(assert(rbx != bc, "register damaged"));
857
858 // get next byte
859 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
860
861 // do actual aload_0
862 aload(0);
863
864 // if _getfield then wait with rewrite
865 __ cmpl(rbx, Bytecodes::_getfield);
866 __ jcc(Assembler::equal, done);
867
868 // if _igetfield then reqrite to _fast_iaccess_0
869 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
870 __ cmpl(rbx, Bytecodes::_fast_igetfield);
871 __ movl(bc, Bytecodes::_fast_iaccess_0);
872 __ jccb(Assembler::equal, rewrite);
873
874 // if _agetfield then reqrite to _fast_aaccess_0
875 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
876 __ cmpl(rbx, Bytecodes::_fast_agetfield);
877 __ movl(bc, Bytecodes::_fast_aaccess_0);
878 __ jccb(Assembler::equal, rewrite);
879
880 // if _fgetfield then reqrite to _fast_faccess_0
881 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
882 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
883 __ movl(bc, Bytecodes::_fast_faccess_0);
884 __ jccb(Assembler::equal, rewrite);
885
886 // else rewrite to _fast_aload0
887 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
888 __ movl(bc, Bytecodes::_fast_aload_0);
889
890 // rewrite
891 // bc: fast bytecode
892 __ bind(rewrite);
893 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
894
895 __ bind(done);
896 } else {
897 aload(0);
898 }
899 }
900
901 void TemplateTable::istore() {
902 transition(itos, vtos);
903 locals_index(rbx);
904 __ movl(iaddress(rbx), rax);
905 }
906
907
908 void TemplateTable::lstore() {
909 transition(ltos, vtos);
910 locals_index(rbx);
911 __ movptr(laddress(rbx), rax);
912 NOT_LP64(__ movptr(haddress(rbx), rdx));
913 }
914
915 void TemplateTable::fstore() {
916 transition(ftos, vtos);
917 locals_index(rbx);
918 __ store_float(faddress(rbx));
919 }
920
921 void TemplateTable::dstore() {
922 transition(dtos, vtos);
923 locals_index(rbx);
924 __ store_double(daddress(rbx));
925 }
926
927 void TemplateTable::astore() {
928 transition(vtos, vtos);
929 __ pop_ptr(rax);
930 locals_index(rbx);
931 __ movptr(aaddress(rbx), rax);
932 }
933
934 void TemplateTable::wide_istore() {
935 transition(vtos, vtos);
936 __ pop_i();
937 locals_index_wide(rbx);
938 __ movl(iaddress(rbx), rax);
939 }
940
941 void TemplateTable::wide_lstore() {
942 transition(vtos, vtos);
943 NOT_LP64(__ pop_l(rax, rdx));
944 LP64_ONLY(__ pop_l());
945 locals_index_wide(rbx);
946 __ movptr(laddress(rbx), rax);
947 NOT_LP64(__ movl(haddress(rbx), rdx));
948 }
949
950 void TemplateTable::wide_fstore() {
951 #ifdef _LP64
952 transition(vtos, vtos);
953 __ pop_f(xmm0);
954 locals_index_wide(rbx);
955 __ movflt(faddress(rbx), xmm0);
956 #else
957 wide_istore();
958 #endif
959 }
960
961 void TemplateTable::wide_dstore() {
962 #ifdef _LP64
963 transition(vtos, vtos);
964 __ pop_d(xmm0);
965 locals_index_wide(rbx);
966 __ movdbl(daddress(rbx), xmm0);
967 #else
968 wide_lstore();
969 #endif
970 }
971
972 void TemplateTable::wide_astore() {
973 transition(vtos, vtos);
974 __ pop_ptr(rax);
975 locals_index_wide(rbx);
976 __ movptr(aaddress(rbx), rax);
977 }
978
979 void TemplateTable::iastore() {
980 transition(itos, vtos);
981 __ pop_i(rbx);
982 // rax: value
983 // rbx: index
984 // rdx: array
985 index_check(rdx, rbx); // prefer index in rbx
986 __ movl(Address(rdx, rbx,
987 Address::times_4,
988 arrayOopDesc::base_offset_in_bytes(T_INT)),
989 rax);
990 }
991
992 void TemplateTable::lastore() {
993 transition(ltos, vtos);
994 __ pop_i(rbx);
995 // rax,: low(value)
996 // rcx: array
997 // rdx: high(value)
998 index_check(rcx, rbx); // prefer index in rbx,
999 // rbx,: index
1000 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1001 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1002 }
1003
1004
1005 void TemplateTable::fastore() {
1006 transition(ftos, vtos);
1007 __ pop_i(rbx);
1008 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1009 // rbx: index
1010 // rdx: array
1011 index_check(rdx, rbx); // prefer index in rbx
1012 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1013 }
1014
1015 void TemplateTable::dastore() {
1016 transition(dtos, vtos);
1017 __ pop_i(rbx);
1018 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1019 // rbx: index
1020 // rdx: array
1021 index_check(rdx, rbx); // prefer index in rbx
1022 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1023 }
1024
1025 void TemplateTable::aastore() {
1026 Label is_null, ok_is_subtype, done;
1027 transition(vtos, vtos);
1028 // stack: ..., array, index, value
1029 __ movptr(rax, at_tos()); // value
1030 __ movl(rcx, at_tos_p1()); // index
1031 __ movptr(rdx, at_tos_p2()); // array
1032
1033 Address element_address(rdx, rcx,
1034 UseCompressedOops? Address::times_4 : Address::times_ptr,
1035 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1036
1037 index_check_without_pop(rdx, rcx); // kills rbx
1038 __ testptr(rax, rax);
1039 __ jcc(Assembler::zero, is_null);
1040
1041 // Move subklass into rbx
1042 __ load_klass(rbx, rax);
1043 // Move superklass into rax
1044 __ load_klass(rax, rdx);
1045 __ movptr(rax, Address(rax,
1046 ObjArrayKlass::element_klass_offset()));
1047 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1048 __ lea(rdx, element_address);
1049
1050 // Generate subtype check. Blows rcx, rdi
1051 // Superklass in rax. Subklass in rbx.
1052 __ gen_subtype_check(rbx, ok_is_subtype);
1053
1054 // Come here on failure
1055 // object is at TOS
1056 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1057
1058 // Come here on success
1059 __ bind(ok_is_subtype);
1060
1061 // Get the value we will store
1062 __ movptr(rax, at_tos());
1063 // Now store using the appropriate barrier
1064 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1065 __ jmp(done);
1066
1067 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1068 __ bind(is_null);
1069 __ profile_null_seen(rbx);
1070
1071 // Store a NULL
1072 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1073
1074 // Pop stack arguments
1075 __ bind(done);
1076 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1077 }
1078
1079 void TemplateTable::bastore() {
1080 transition(itos, vtos);
1081 __ pop_i(rbx);
1082 // rax: value
1083 // rbx: index
1084 // rdx: array
1085 index_check(rdx, rbx); // prefer index in rbx
1086 // Need to check whether array is boolean or byte
1087 // since both types share the bastore bytecode.
1088 __ load_klass(rcx, rdx);
1089 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1090 int diffbit = Klass::layout_helper_boolean_diffbit();
1091 __ testl(rcx, diffbit);
1092 Label L_skip;
1093 __ jccb(Assembler::zero, L_skip);
1094 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1095 __ bind(L_skip);
1096 __ movb(Address(rdx, rbx,
1097 Address::times_1,
1098 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1099 rax);
1100 }
1101
1102 void TemplateTable::castore() {
1103 transition(itos, vtos);
1104 __ pop_i(rbx);
1105 // rax: value
1106 // rbx: index
1107 // rdx: array
1108 index_check(rdx, rbx); // prefer index in rbx
1109 __ movw(Address(rdx, rbx,
1110 Address::times_2,
1111 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1112 rax);
1113 }
1114
1115
1116 void TemplateTable::sastore() {
1117 castore();
1118 }
1119
1120 void TemplateTable::istore(int n) {
1121 transition(itos, vtos);
1122 __ movl(iaddress(n), rax);
1123 }
1124
1125 void TemplateTable::lstore(int n) {
1126 transition(ltos, vtos);
1127 __ movptr(laddress(n), rax);
1128 NOT_LP64(__ movptr(haddress(n), rdx));
1129 }
1130
1131 void TemplateTable::fstore(int n) {
1132 transition(ftos, vtos);
1133 __ store_float(faddress(n));
1134 }
1135
1136 void TemplateTable::dstore(int n) {
1137 transition(dtos, vtos);
1138 __ store_double(daddress(n));
1139 }
1140
1141
1142 void TemplateTable::astore(int n) {
1143 transition(vtos, vtos);
1144 __ pop_ptr(rax);
1145 __ movptr(aaddress(n), rax);
1146 }
1147
1148 void TemplateTable::pop() {
1149 transition(vtos, vtos);
1150 __ addptr(rsp, Interpreter::stackElementSize);
1151 }
1152
1153 void TemplateTable::pop2() {
1154 transition(vtos, vtos);
1155 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1156 }
1157
1158
1159 void TemplateTable::dup() {
1160 transition(vtos, vtos);
1161 __ load_ptr(0, rax);
1162 __ push_ptr(rax);
1163 // stack: ..., a, a
1164 }
1165
1166 void TemplateTable::dup_x1() {
1167 transition(vtos, vtos);
1168 // stack: ..., a, b
1169 __ load_ptr( 0, rax); // load b
1170 __ load_ptr( 1, rcx); // load a
1171 __ store_ptr(1, rax); // store b
1172 __ store_ptr(0, rcx); // store a
1173 __ push_ptr(rax); // push b
1174 // stack: ..., b, a, b
1175 }
1176
1177 void TemplateTable::dup_x2() {
1178 transition(vtos, vtos);
1179 // stack: ..., a, b, c
1180 __ load_ptr( 0, rax); // load c
1181 __ load_ptr( 2, rcx); // load a
1182 __ store_ptr(2, rax); // store c in a
1183 __ push_ptr(rax); // push c
1184 // stack: ..., c, b, c, c
1185 __ load_ptr( 2, rax); // load b
1186 __ store_ptr(2, rcx); // store a in b
1187 // stack: ..., c, a, c, c
1188 __ store_ptr(1, rax); // store b in c
1189 // stack: ..., c, a, b, c
1190 }
1191
1192 void TemplateTable::dup2() {
1193 transition(vtos, vtos);
1194 // stack: ..., a, b
1195 __ load_ptr(1, rax); // load a
1196 __ push_ptr(rax); // push a
1197 __ load_ptr(1, rax); // load b
1198 __ push_ptr(rax); // push b
1199 // stack: ..., a, b, a, b
1200 }
1201
1202
1203 void TemplateTable::dup2_x1() {
1204 transition(vtos, vtos);
1205 // stack: ..., a, b, c
1206 __ load_ptr( 0, rcx); // load c
1207 __ load_ptr( 1, rax); // load b
1208 __ push_ptr(rax); // push b
1209 __ push_ptr(rcx); // push c
1210 // stack: ..., a, b, c, b, c
1211 __ store_ptr(3, rcx); // store c in b
1212 // stack: ..., a, c, c, b, c
1213 __ load_ptr( 4, rcx); // load a
1214 __ store_ptr(2, rcx); // store a in 2nd c
1215 // stack: ..., a, c, a, b, c
1216 __ store_ptr(4, rax); // store b in a
1217 // stack: ..., b, c, a, b, c
1218 }
1219
1220 void TemplateTable::dup2_x2() {
1221 transition(vtos, vtos);
1222 // stack: ..., a, b, c, d
1223 __ load_ptr( 0, rcx); // load d
1224 __ load_ptr( 1, rax); // load c
1225 __ push_ptr(rax); // push c
1226 __ push_ptr(rcx); // push d
1227 // stack: ..., a, b, c, d, c, d
1228 __ load_ptr( 4, rax); // load b
1229 __ store_ptr(2, rax); // store b in d
1230 __ store_ptr(4, rcx); // store d in b
1231 // stack: ..., a, d, c, b, c, d
1232 __ load_ptr( 5, rcx); // load a
1233 __ load_ptr( 3, rax); // load c
1234 __ store_ptr(3, rcx); // store a in c
1235 __ store_ptr(5, rax); // store c in a
1236 // stack: ..., c, d, a, b, c, d
1237 }
1238
1239 void TemplateTable::swap() {
1240 transition(vtos, vtos);
1241 // stack: ..., a, b
1242 __ load_ptr( 1, rcx); // load a
1243 __ load_ptr( 0, rax); // load b
1244 __ store_ptr(0, rcx); // store a in b
1245 __ store_ptr(1, rax); // store b in a
1246 // stack: ..., b, a
1247 }
1248
1249 void TemplateTable::iop2(Operation op) {
1250 transition(itos, itos);
1251 switch (op) {
1252 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1253 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1254 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1255 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1256 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1257 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1258 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1259 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1260 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1261 default : ShouldNotReachHere();
1262 }
1263 }
1264
1265 void TemplateTable::lop2(Operation op) {
1266 transition(ltos, ltos);
1267 #ifdef _LP64
1268 switch (op) {
1269 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1270 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1271 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1272 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1273 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1274 default : ShouldNotReachHere();
1275 }
1276 #else
1277 __ pop_l(rbx, rcx);
1278 switch (op) {
1279 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1280 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1281 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1282 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1283 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1284 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1285 default : ShouldNotReachHere();
1286 }
1287 #endif
1288 }
1289
1290 void TemplateTable::idiv() {
1291 transition(itos, itos);
1292 __ movl(rcx, rax);
1293 __ pop_i(rax);
1294 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1295 // they are not equal, one could do a normal division (no correction
1296 // needed), which may speed up this implementation for the common case.
1297 // (see also JVM spec., p.243 & p.271)
1298 __ corrected_idivl(rcx);
1299 }
1300
1301 void TemplateTable::irem() {
1302 transition(itos, itos);
1303 __ movl(rcx, rax);
1304 __ pop_i(rax);
1305 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1306 // they are not equal, one could do a normal division (no correction
1307 // needed), which may speed up this implementation for the common case.
1308 // (see also JVM spec., p.243 & p.271)
1309 __ corrected_idivl(rcx);
1310 __ movl(rax, rdx);
1311 }
1312
1313 void TemplateTable::lmul() {
1314 transition(ltos, ltos);
1315 #ifdef _LP64
1316 __ pop_l(rdx);
1317 __ imulq(rax, rdx);
1318 #else
1319 __ pop_l(rbx, rcx);
1320 __ push(rcx); __ push(rbx);
1321 __ push(rdx); __ push(rax);
1322 __ lmul(2 * wordSize, 0);
1323 __ addptr(rsp, 4 * wordSize); // take off temporaries
1324 #endif
1325 }
1326
1327 void TemplateTable::ldiv() {
1328 transition(ltos, ltos);
1329 #ifdef _LP64
1330 __ mov(rcx, rax);
1331 __ pop_l(rax);
1332 // generate explicit div0 check
1333 __ testq(rcx, rcx);
1334 __ jump_cc(Assembler::zero,
1335 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1336 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1337 // they are not equal, one could do a normal division (no correction
1338 // needed), which may speed up this implementation for the common case.
1339 // (see also JVM spec., p.243 & p.271)
1340 __ corrected_idivq(rcx); // kills rbx
1341 #else
1342 __ pop_l(rbx, rcx);
1343 __ push(rcx); __ push(rbx);
1344 __ push(rdx); __ push(rax);
1345 // check if y = 0
1346 __ orl(rax, rdx);
1347 __ jump_cc(Assembler::zero,
1348 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1349 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1350 __ addptr(rsp, 4 * wordSize); // take off temporaries
1351 #endif
1352 }
1353
1354 void TemplateTable::lrem() {
1355 transition(ltos, ltos);
1356 #ifdef _LP64
1357 __ mov(rcx, rax);
1358 __ pop_l(rax);
1359 __ testq(rcx, rcx);
1360 __ jump_cc(Assembler::zero,
1361 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1362 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1363 // they are not equal, one could do a normal division (no correction
1364 // needed), which may speed up this implementation for the common case.
1365 // (see also JVM spec., p.243 & p.271)
1366 __ corrected_idivq(rcx); // kills rbx
1367 __ mov(rax, rdx);
1368 #else
1369 __ pop_l(rbx, rcx);
1370 __ push(rcx); __ push(rbx);
1371 __ push(rdx); __ push(rax);
1372 // check if y = 0
1373 __ orl(rax, rdx);
1374 __ jump_cc(Assembler::zero,
1375 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1376 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1377 __ addptr(rsp, 4 * wordSize);
1378 #endif
1379 }
1380
1381 void TemplateTable::lshl() {
1382 transition(itos, ltos);
1383 __ movl(rcx, rax); // get shift count
1384 #ifdef _LP64
1385 __ pop_l(rax); // get shift value
1386 __ shlq(rax);
1387 #else
1388 __ pop_l(rax, rdx); // get shift value
1389 __ lshl(rdx, rax);
1390 #endif
1391 }
1392
1393 void TemplateTable::lshr() {
1394 #ifdef _LP64
1395 transition(itos, ltos);
1396 __ movl(rcx, rax); // get shift count
1397 __ pop_l(rax); // get shift value
1398 __ sarq(rax);
1399 #else
1400 transition(itos, ltos);
1401 __ mov(rcx, rax); // get shift count
1402 __ pop_l(rax, rdx); // get shift value
1403 __ lshr(rdx, rax, true);
1404 #endif
1405 }
1406
1407 void TemplateTable::lushr() {
1408 transition(itos, ltos);
1409 #ifdef _LP64
1410 __ movl(rcx, rax); // get shift count
1411 __ pop_l(rax); // get shift value
1412 __ shrq(rax);
1413 #else
1414 __ mov(rcx, rax); // get shift count
1415 __ pop_l(rax, rdx); // get shift value
1416 __ lshr(rdx, rax);
1417 #endif
1418 }
1419
1420 void TemplateTable::fop2(Operation op) {
1421 transition(ftos, ftos);
1422
1423 if (UseSSE >= 1) {
1424 switch (op) {
1425 case add:
1426 __ addss(xmm0, at_rsp());
1427 __ addptr(rsp, Interpreter::stackElementSize);
1428 break;
1429 case sub:
1430 __ movflt(xmm1, xmm0);
1431 __ pop_f(xmm0);
1432 __ subss(xmm0, xmm1);
1433 break;
1434 case mul:
1435 __ mulss(xmm0, at_rsp());
1436 __ addptr(rsp, Interpreter::stackElementSize);
1437 break;
1438 case div:
1439 __ movflt(xmm1, xmm0);
1440 __ pop_f(xmm0);
1441 __ divss(xmm0, xmm1);
1442 break;
1443 case rem:
1444 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1445 // modulo operation. The frem method calls the function
1446 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1447 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1448 // (signalling or quiet) is returned.
1449 //
1450 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1451 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1452 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1453 // The fprem instruction used on x86_32 is functionally equivalent to
1454 // SharedRuntime::frem in that it returns a NaN.
1455 #ifdef _LP64
1456 __ movflt(xmm1, xmm0);
1457 __ pop_f(xmm0);
1458 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1459 #else
1460 __ push_f(xmm0);
1461 __ pop_f();
1462 __ fld_s(at_rsp());
1463 __ fremr(rax);
1464 __ f2ieee();
1465 __ pop(rax); // pop second operand off the stack
1466 __ push_f();
1467 __ pop_f(xmm0);
1468 #endif
1469 break;
1470 default:
1471 ShouldNotReachHere();
1472 break;
1473 }
1474 } else {
1475 #ifdef _LP64
1476 ShouldNotReachHere();
1477 #else
1478 switch (op) {
1479 case add: __ fadd_s (at_rsp()); break;
1480 case sub: __ fsubr_s(at_rsp()); break;
1481 case mul: __ fmul_s (at_rsp()); break;
1482 case div: __ fdivr_s(at_rsp()); break;
1483 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1484 default : ShouldNotReachHere();
1485 }
1486 __ f2ieee();
1487 __ pop(rax); // pop second operand off the stack
1488 #endif // _LP64
1489 }
1490 }
1491
1492 void TemplateTable::dop2(Operation op) {
1493 transition(dtos, dtos);
1494 if (UseSSE >= 2) {
1495 switch (op) {
1496 case add:
1497 __ addsd(xmm0, at_rsp());
1498 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1499 break;
1500 case sub:
1501 __ movdbl(xmm1, xmm0);
1502 __ pop_d(xmm0);
1503 __ subsd(xmm0, xmm1);
1504 break;
1505 case mul:
1506 __ mulsd(xmm0, at_rsp());
1507 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1508 break;
1509 case div:
1510 __ movdbl(xmm1, xmm0);
1511 __ pop_d(xmm0);
1512 __ divsd(xmm0, xmm1);
1513 break;
1514 case rem:
1515 // Similar to fop2(), the modulo operation is performed using the
1516 // SharedRuntime::drem method (on x86_64 platforms) or using the
1517 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1518 #ifdef _LP64
1519 __ movdbl(xmm1, xmm0);
1520 __ pop_d(xmm0);
1521 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1522 #else
1523 __ push_d(xmm0);
1524 __ pop_d();
1525 __ fld_d(at_rsp());
1526 __ fremr(rax);
1527 __ d2ieee();
1528 __ pop(rax);
1529 __ pop(rdx);
1530 __ push_d();
1531 __ pop_d(xmm0);
1532 #endif
1533 break;
1534 default:
1535 ShouldNotReachHere();
1536 break;
1537 }
1538 } else {
1539 #ifdef _LP64
1540 ShouldNotReachHere();
1541 #else
1542 switch (op) {
1543 case add: __ fadd_d (at_rsp()); break;
1544 case sub: __ fsubr_d(at_rsp()); break;
1545 case mul: {
1546 Label L_strict;
1547 Label L_join;
1548 const Address access_flags (rcx, Method::access_flags_offset());
1549 __ get_method(rcx);
1550 __ movl(rcx, access_flags);
1551 __ testl(rcx, JVM_ACC_STRICT);
1552 __ jccb(Assembler::notZero, L_strict);
1553 __ fmul_d (at_rsp());
1554 __ jmpb(L_join);
1555 __ bind(L_strict);
1556 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1557 __ fmulp();
1558 __ fmul_d (at_rsp());
1559 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1560 __ fmulp();
1561 __ bind(L_join);
1562 break;
1563 }
1564 case div: {
1565 Label L_strict;
1566 Label L_join;
1567 const Address access_flags (rcx, Method::access_flags_offset());
1568 __ get_method(rcx);
1569 __ movl(rcx, access_flags);
1570 __ testl(rcx, JVM_ACC_STRICT);
1571 __ jccb(Assembler::notZero, L_strict);
1572 __ fdivr_d(at_rsp());
1573 __ jmp(L_join);
1574 __ bind(L_strict);
1575 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1576 __ fmul_d (at_rsp());
1577 __ fdivrp();
1578 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1579 __ fmulp();
1580 __ bind(L_join);
1581 break;
1582 }
1583 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1584 default : ShouldNotReachHere();
1585 }
1586 __ d2ieee();
1587 // Pop double precision number from rsp.
1588 __ pop(rax);
1589 __ pop(rdx);
1590 #endif
1591 }
1592 }
1593
1594 void TemplateTable::ineg() {
1595 transition(itos, itos);
1596 __ negl(rax);
1597 }
1598
1599 void TemplateTable::lneg() {
1600 transition(ltos, ltos);
1601 LP64_ONLY(__ negq(rax));
1602 NOT_LP64(__ lneg(rdx, rax));
1603 }
1604
1605 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1606 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1607 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1608 // of 128-bits operands for SSE instructions.
1609 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1610 // Store the value to a 128-bits operand.
1611 operand[0] = lo;
1612 operand[1] = hi;
1613 return operand;
1614 }
1615
1616 // Buffer for 128-bits masks used by SSE instructions.
1617 static jlong float_signflip_pool[2*2];
1618 static jlong double_signflip_pool[2*2];
1619
1620 void TemplateTable::fneg() {
1621 transition(ftos, ftos);
1622 if (UseSSE >= 1) {
1623 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1624 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1625 } else {
1626 LP64_ONLY(ShouldNotReachHere());
1627 NOT_LP64(__ fchs());
1628 }
1629 }
1630
1631 void TemplateTable::dneg() {
1632 transition(dtos, dtos);
1633 if (UseSSE >= 2) {
1634 static jlong *double_signflip =
1635 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1636 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1637 } else {
1638 #ifdef _LP64
1639 ShouldNotReachHere();
1640 #else
1641 __ fchs();
1642 #endif
1643 }
1644 }
1645
1646 void TemplateTable::iinc() {
1647 transition(vtos, vtos);
1648 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1649 locals_index(rbx);
1650 __ addl(iaddress(rbx), rdx);
1651 }
1652
1653 void TemplateTable::wide_iinc() {
1654 transition(vtos, vtos);
1655 __ movl(rdx, at_bcp(4)); // get constant
1656 locals_index_wide(rbx);
1657 __ bswapl(rdx); // swap bytes & sign-extend constant
1658 __ sarl(rdx, 16);
1659 __ addl(iaddress(rbx), rdx);
1660 // Note: should probably use only one movl to get both
1661 // the index and the constant -> fix this
1662 }
1663
1664 void TemplateTable::convert() {
1665 #ifdef _LP64
1666 // Checking
1667 #ifdef ASSERT
1668 {
1669 TosState tos_in = ilgl;
1670 TosState tos_out = ilgl;
1671 switch (bytecode()) {
1672 case Bytecodes::_i2l: // fall through
1673 case Bytecodes::_i2f: // fall through
1674 case Bytecodes::_i2d: // fall through
1675 case Bytecodes::_i2b: // fall through
1676 case Bytecodes::_i2c: // fall through
1677 case Bytecodes::_i2s: tos_in = itos; break;
1678 case Bytecodes::_l2i: // fall through
1679 case Bytecodes::_l2f: // fall through
1680 case Bytecodes::_l2d: tos_in = ltos; break;
1681 case Bytecodes::_f2i: // fall through
1682 case Bytecodes::_f2l: // fall through
1683 case Bytecodes::_f2d: tos_in = ftos; break;
1684 case Bytecodes::_d2i: // fall through
1685 case Bytecodes::_d2l: // fall through
1686 case Bytecodes::_d2f: tos_in = dtos; break;
1687 default : ShouldNotReachHere();
1688 }
1689 switch (bytecode()) {
1690 case Bytecodes::_l2i: // fall through
1691 case Bytecodes::_f2i: // fall through
1692 case Bytecodes::_d2i: // fall through
1693 case Bytecodes::_i2b: // fall through
1694 case Bytecodes::_i2c: // fall through
1695 case Bytecodes::_i2s: tos_out = itos; break;
1696 case Bytecodes::_i2l: // fall through
1697 case Bytecodes::_f2l: // fall through
1698 case Bytecodes::_d2l: tos_out = ltos; break;
1699 case Bytecodes::_i2f: // fall through
1700 case Bytecodes::_l2f: // fall through
1701 case Bytecodes::_d2f: tos_out = ftos; break;
1702 case Bytecodes::_i2d: // fall through
1703 case Bytecodes::_l2d: // fall through
1704 case Bytecodes::_f2d: tos_out = dtos; break;
1705 default : ShouldNotReachHere();
1706 }
1707 transition(tos_in, tos_out);
1708 }
1709 #endif // ASSERT
1710
1711 static const int64_t is_nan = 0x8000000000000000L;
1712
1713 // Conversion
1714 switch (bytecode()) {
1715 case Bytecodes::_i2l:
1716 __ movslq(rax, rax);
1717 break;
1718 case Bytecodes::_i2f:
1719 __ cvtsi2ssl(xmm0, rax);
1720 break;
1721 case Bytecodes::_i2d:
1722 __ cvtsi2sdl(xmm0, rax);
1723 break;
1724 case Bytecodes::_i2b:
1725 __ movsbl(rax, rax);
1726 break;
1727 case Bytecodes::_i2c:
1728 __ movzwl(rax, rax);
1729 break;
1730 case Bytecodes::_i2s:
1731 __ movswl(rax, rax);
1732 break;
1733 case Bytecodes::_l2i:
1734 __ movl(rax, rax);
1735 break;
1736 case Bytecodes::_l2f:
1737 __ cvtsi2ssq(xmm0, rax);
1738 break;
1739 case Bytecodes::_l2d:
1740 __ cvtsi2sdq(xmm0, rax);
1741 break;
1742 case Bytecodes::_f2i:
1743 {
1744 Label L;
1745 __ cvttss2sil(rax, xmm0);
1746 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1747 __ jcc(Assembler::notEqual, L);
1748 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1749 __ bind(L);
1750 }
1751 break;
1752 case Bytecodes::_f2l:
1753 {
1754 Label L;
1755 __ cvttss2siq(rax, xmm0);
1756 // NaN or overflow/underflow?
1757 __ cmp64(rax, ExternalAddress((address) &is_nan));
1758 __ jcc(Assembler::notEqual, L);
1759 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1760 __ bind(L);
1761 }
1762 break;
1763 case Bytecodes::_f2d:
1764 __ cvtss2sd(xmm0, xmm0);
1765 break;
1766 case Bytecodes::_d2i:
1767 {
1768 Label L;
1769 __ cvttsd2sil(rax, xmm0);
1770 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1771 __ jcc(Assembler::notEqual, L);
1772 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1773 __ bind(L);
1774 }
1775 break;
1776 case Bytecodes::_d2l:
1777 {
1778 Label L;
1779 __ cvttsd2siq(rax, xmm0);
1780 // NaN or overflow/underflow?
1781 __ cmp64(rax, ExternalAddress((address) &is_nan));
1782 __ jcc(Assembler::notEqual, L);
1783 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1784 __ bind(L);
1785 }
1786 break;
1787 case Bytecodes::_d2f:
1788 __ cvtsd2ss(xmm0, xmm0);
1789 break;
1790 default:
1791 ShouldNotReachHere();
1792 }
1793 #else
1794 // Checking
1795 #ifdef ASSERT
1796 { TosState tos_in = ilgl;
1797 TosState tos_out = ilgl;
1798 switch (bytecode()) {
1799 case Bytecodes::_i2l: // fall through
1800 case Bytecodes::_i2f: // fall through
1801 case Bytecodes::_i2d: // fall through
1802 case Bytecodes::_i2b: // fall through
1803 case Bytecodes::_i2c: // fall through
1804 case Bytecodes::_i2s: tos_in = itos; break;
1805 case Bytecodes::_l2i: // fall through
1806 case Bytecodes::_l2f: // fall through
1807 case Bytecodes::_l2d: tos_in = ltos; break;
1808 case Bytecodes::_f2i: // fall through
1809 case Bytecodes::_f2l: // fall through
1810 case Bytecodes::_f2d: tos_in = ftos; break;
1811 case Bytecodes::_d2i: // fall through
1812 case Bytecodes::_d2l: // fall through
1813 case Bytecodes::_d2f: tos_in = dtos; break;
1814 default : ShouldNotReachHere();
1815 }
1816 switch (bytecode()) {
1817 case Bytecodes::_l2i: // fall through
1818 case Bytecodes::_f2i: // fall through
1819 case Bytecodes::_d2i: // fall through
1820 case Bytecodes::_i2b: // fall through
1821 case Bytecodes::_i2c: // fall through
1822 case Bytecodes::_i2s: tos_out = itos; break;
1823 case Bytecodes::_i2l: // fall through
1824 case Bytecodes::_f2l: // fall through
1825 case Bytecodes::_d2l: tos_out = ltos; break;
1826 case Bytecodes::_i2f: // fall through
1827 case Bytecodes::_l2f: // fall through
1828 case Bytecodes::_d2f: tos_out = ftos; break;
1829 case Bytecodes::_i2d: // fall through
1830 case Bytecodes::_l2d: // fall through
1831 case Bytecodes::_f2d: tos_out = dtos; break;
1832 default : ShouldNotReachHere();
1833 }
1834 transition(tos_in, tos_out);
1835 }
1836 #endif // ASSERT
1837
1838 // Conversion
1839 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1840 switch (bytecode()) {
1841 case Bytecodes::_i2l:
1842 __ extend_sign(rdx, rax);
1843 break;
1844 case Bytecodes::_i2f:
1845 if (UseSSE >= 1) {
1846 __ cvtsi2ssl(xmm0, rax);
1847 } else {
1848 __ push(rax); // store int on tos
1849 __ fild_s(at_rsp()); // load int to ST0
1850 __ f2ieee(); // truncate to float size
1851 __ pop(rcx); // adjust rsp
1852 }
1853 break;
1854 case Bytecodes::_i2d:
1855 if (UseSSE >= 2) {
1856 __ cvtsi2sdl(xmm0, rax);
1857 } else {
1858 __ push(rax); // add one slot for d2ieee()
1859 __ push(rax); // store int on tos
1860 __ fild_s(at_rsp()); // load int to ST0
1861 __ d2ieee(); // truncate to double size
1862 __ pop(rcx); // adjust rsp
1863 __ pop(rcx);
1864 }
1865 break;
1866 case Bytecodes::_i2b:
1867 __ shll(rax, 24); // truncate upper 24 bits
1868 __ sarl(rax, 24); // and sign-extend byte
1869 LP64_ONLY(__ movsbl(rax, rax));
1870 break;
1871 case Bytecodes::_i2c:
1872 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1873 LP64_ONLY(__ movzwl(rax, rax));
1874 break;
1875 case Bytecodes::_i2s:
1876 __ shll(rax, 16); // truncate upper 16 bits
1877 __ sarl(rax, 16); // and sign-extend short
1878 LP64_ONLY(__ movswl(rax, rax));
1879 break;
1880 case Bytecodes::_l2i:
1881 /* nothing to do */
1882 break;
1883 case Bytecodes::_l2f:
1884 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1885 // 64-bit long values to floats. On 32-bit platforms it is not possible
1886 // to use that instruction with 64-bit operands, therefore the FPU is
1887 // used to perform the conversion.
1888 __ push(rdx); // store long on tos
1889 __ push(rax);
1890 __ fild_d(at_rsp()); // load long to ST0
1891 __ f2ieee(); // truncate to float size
1892 __ pop(rcx); // adjust rsp
1893 __ pop(rcx);
1894 if (UseSSE >= 1) {
1895 __ push_f();
1896 __ pop_f(xmm0);
1897 }
1898 break;
1899 case Bytecodes::_l2d:
1900 // On 32-bit platforms the FPU is used for conversion because on
1901 // 32-bit platforms it is not not possible to use the cvtsi2sdq
1902 // instruction with 64-bit operands.
1903 __ push(rdx); // store long on tos
1904 __ push(rax);
1905 __ fild_d(at_rsp()); // load long to ST0
1906 __ d2ieee(); // truncate to double size
1907 __ pop(rcx); // adjust rsp
1908 __ pop(rcx);
1909 if (UseSSE >= 2) {
1910 __ push_d();
1911 __ pop_d(xmm0);
1912 }
1913 break;
1914 case Bytecodes::_f2i:
1915 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1916 // as it returns 0 for any NaN.
1917 if (UseSSE >= 1) {
1918 __ push_f(xmm0);
1919 } else {
1920 __ push(rcx); // reserve space for argument
1921 __ fstp_s(at_rsp()); // pass float argument on stack
1922 }
1923 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1924 break;
1925 case Bytecodes::_f2l:
1926 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
1927 // as it returns 0 for any NaN.
1928 if (UseSSE >= 1) {
1929 __ push_f(xmm0);
1930 } else {
1931 __ push(rcx); // reserve space for argument
1932 __ fstp_s(at_rsp()); // pass float argument on stack
1933 }
1934 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1935 break;
1936 case Bytecodes::_f2d:
1937 if (UseSSE < 1) {
1938 /* nothing to do */
1939 } else if (UseSSE == 1) {
1940 __ push_f(xmm0);
1941 __ pop_f();
1942 } else { // UseSSE >= 2
1943 __ cvtss2sd(xmm0, xmm0);
1944 }
1945 break;
1946 case Bytecodes::_d2i:
1947 if (UseSSE >= 2) {
1948 __ push_d(xmm0);
1949 } else {
1950 __ push(rcx); // reserve space for argument
1951 __ push(rcx);
1952 __ fstp_d(at_rsp()); // pass double argument on stack
1953 }
1954 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1955 break;
1956 case Bytecodes::_d2l:
1957 if (UseSSE >= 2) {
1958 __ push_d(xmm0);
1959 } else {
1960 __ push(rcx); // reserve space for argument
1961 __ push(rcx);
1962 __ fstp_d(at_rsp()); // pass double argument on stack
1963 }
1964 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1965 break;
1966 case Bytecodes::_d2f:
1967 if (UseSSE <= 1) {
1968 __ push(rcx); // reserve space for f2ieee()
1969 __ f2ieee(); // truncate to float size
1970 __ pop(rcx); // adjust rsp
1971 if (UseSSE == 1) {
1972 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
1973 // the conversion is performed using the FPU in this case.
1974 __ push_f();
1975 __ pop_f(xmm0);
1976 }
1977 } else { // UseSSE >= 2
1978 __ cvtsd2ss(xmm0, xmm0);
1979 }
1980 break;
1981 default :
1982 ShouldNotReachHere();
1983 }
1984 #endif
1985 }
1986
1987 void TemplateTable::lcmp() {
1988 transition(ltos, itos);
1989 #ifdef _LP64
1990 Label done;
1991 __ pop_l(rdx);
1992 __ cmpq(rdx, rax);
1993 __ movl(rax, -1);
1994 __ jccb(Assembler::less, done);
1995 __ setb(Assembler::notEqual, rax);
1996 __ movzbl(rax, rax);
1997 __ bind(done);
1998 #else
1999
2000 // y = rdx:rax
2001 __ pop_l(rbx, rcx); // get x = rcx:rbx
2002 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2003 __ mov(rax, rcx);
2004 #endif
2005 }
2006
2007 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2008 if ((is_float && UseSSE >= 1) ||
2009 (!is_float && UseSSE >= 2)) {
2010 Label done;
2011 if (is_float) {
2012 // XXX get rid of pop here, use ... reg, mem32
2013 __ pop_f(xmm1);
2014 __ ucomiss(xmm1, xmm0);
2015 } else {
2016 // XXX get rid of pop here, use ... reg, mem64
2017 __ pop_d(xmm1);
2018 __ ucomisd(xmm1, xmm0);
2019 }
2020 if (unordered_result < 0) {
2021 __ movl(rax, -1);
2022 __ jccb(Assembler::parity, done);
2023 __ jccb(Assembler::below, done);
2024 __ setb(Assembler::notEqual, rdx);
2025 __ movzbl(rax, rdx);
2026 } else {
2027 __ movl(rax, 1);
2028 __ jccb(Assembler::parity, done);
2029 __ jccb(Assembler::above, done);
2030 __ movl(rax, 0);
2031 __ jccb(Assembler::equal, done);
2032 __ decrementl(rax);
2033 }
2034 __ bind(done);
2035 } else {
2036 #ifdef _LP64
2037 ShouldNotReachHere();
2038 #else
2039 if (is_float) {
2040 __ fld_s(at_rsp());
2041 } else {
2042 __ fld_d(at_rsp());
2043 __ pop(rdx);
2044 }
2045 __ pop(rcx);
2046 __ fcmp2int(rax, unordered_result < 0);
2047 #endif // _LP64
2048 }
2049 }
2050
2051 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2052 __ get_method(rcx); // rcx holds method
2053 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2054 // holds bumped taken count
2055
2056 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2057 InvocationCounter::counter_offset();
2058 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2059 InvocationCounter::counter_offset();
2060
2061 // Load up edx with the branch displacement
2062 if (is_wide) {
2063 __ movl(rdx, at_bcp(1));
2064 } else {
2065 __ load_signed_short(rdx, at_bcp(1));
2066 }
2067 __ bswapl(rdx);
2068
2069 if (!is_wide) {
2070 __ sarl(rdx, 16);
2071 }
2072 LP64_ONLY(__ movl2ptr(rdx, rdx));
2073
2074 // Handle all the JSR stuff here, then exit.
2075 // It's much shorter and cleaner than intermingling with the non-JSR
2076 // normal-branch stuff occurring below.
2077 if (is_jsr) {
2078 // Pre-load the next target bytecode into rbx
2079 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2080
2081 // compute return address as bci in rax
2082 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2083 in_bytes(ConstMethod::codes_offset())));
2084 __ subptr(rax, Address(rcx, Method::const_offset()));
2085 // Adjust the bcp in r13 by the displacement in rdx
2086 __ addptr(rbcp, rdx);
2087 // jsr returns atos that is not an oop
2088 __ push_i(rax);
2089 __ dispatch_only(vtos);
2090 return;
2091 }
2092
2093 // Normal (non-jsr) branch handling
2094
2095 // Adjust the bcp in r13 by the displacement in rdx
2096 __ addptr(rbcp, rdx);
2097
2098 assert(UseLoopCounter || !UseOnStackReplacement,
2099 "on-stack-replacement requires loop counters");
2100 Label backedge_counter_overflow;
2101 Label profile_method;
2102 Label dispatch;
2103 if (UseLoopCounter) {
2104 // increment backedge counter for backward branches
2105 // rax: MDO
2106 // rbx: MDO bumped taken-count
2107 // rcx: method
2108 // rdx: target offset
2109 // r13: target bcp
2110 // r14: locals pointer
2111 __ testl(rdx, rdx); // check if forward or backward branch
2112 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2113
2114 // check if MethodCounters exists
2115 Label has_counters;
2116 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2117 __ testptr(rax, rax);
2118 __ jcc(Assembler::notZero, has_counters);
2119 __ push(rdx);
2120 __ push(rcx);
2121 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2122 rcx);
2123 __ pop(rcx);
2124 __ pop(rdx);
2125 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2126 __ testptr(rax, rax);
2127 __ jcc(Assembler::zero, dispatch);
2128 __ bind(has_counters);
2129
2130 if (TieredCompilation) {
2131 Label no_mdo;
2132 int increment = InvocationCounter::count_increment;
2133 if (ProfileInterpreter) {
2134 // Are we profiling?
2135 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2136 __ testptr(rbx, rbx);
2137 __ jccb(Assembler::zero, no_mdo);
2138 // Increment the MDO backedge counter
2139 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2140 in_bytes(InvocationCounter::counter_offset()));
2141 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2142 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2143 rax, false, Assembler::zero, &backedge_counter_overflow);
2144 __ jmp(dispatch);
2145 }
2146 __ bind(no_mdo);
2147 // Increment backedge counter in MethodCounters*
2148 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2149 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2150 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2151 rax, false, Assembler::zero, &backedge_counter_overflow);
2152 } else { // not TieredCompilation
2153 // increment counter
2154 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2155 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2156 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2157 __ movl(Address(rcx, be_offset), rax); // store counter
2158
2159 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2160
2161 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2162 __ addl(rax, Address(rcx, be_offset)); // add both counters
2163
2164 if (ProfileInterpreter) {
2165 // Test to see if we should create a method data oop
2166 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2167 __ jcc(Assembler::less, dispatch);
2168
2169 // if no method data exists, go to profile method
2170 __ test_method_data_pointer(rax, profile_method);
2171
2172 if (UseOnStackReplacement) {
2173 // check for overflow against rbx which is the MDO taken count
2174 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2175 __ jcc(Assembler::below, dispatch);
2176
2177 // When ProfileInterpreter is on, the backedge_count comes
2178 // from the MethodData*, which value does not get reset on
2179 // the call to frequency_counter_overflow(). To avoid
2180 // excessive calls to the overflow routine while the method is
2181 // being compiled, add a second test to make sure the overflow
2182 // function is called only once every overflow_frequency.
2183 const int overflow_frequency = 1024;
2184 __ andl(rbx, overflow_frequency - 1);
2185 __ jcc(Assembler::zero, backedge_counter_overflow);
2186
2187 }
2188 } else {
2189 if (UseOnStackReplacement) {
2190 // check for overflow against rax, which is the sum of the
2191 // counters
2192 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2193 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2194
2195 }
2196 }
2197 }
2198 __ bind(dispatch);
2199 }
2200
2201 // Pre-load the next target bytecode into rbx
2202 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2203
2204 // continue with the bytecode @ target
2205 // rax: return bci for jsr's, unused otherwise
2206 // rbx: target bytecode
2207 // r13: target bcp
2208 __ dispatch_only(vtos);
2209
2210 if (UseLoopCounter) {
2211 if (ProfileInterpreter) {
2212 // Out-of-line code to allocate method data oop.
2213 __ bind(profile_method);
2214 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2215 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2216 __ set_method_data_pointer_for_bcp();
2217 __ jmp(dispatch);
2218 }
2219
2220 if (UseOnStackReplacement) {
2221 // invocation counter overflow
2222 __ bind(backedge_counter_overflow);
2223 __ negptr(rdx);
2224 __ addptr(rdx, rbcp); // branch bcp
2225 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2226 __ call_VM(noreg,
2227 CAST_FROM_FN_PTR(address,
2228 InterpreterRuntime::frequency_counter_overflow),
2229 rdx);
2230 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2231
2232 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2233 // rbx: target bytecode
2234 // rdx: scratch
2235 // r14: locals pointer
2236 // r13: bcp
2237 __ testptr(rax, rax); // test result
2238 __ jcc(Assembler::zero, dispatch); // no osr if null
2239 // nmethod may have been invalidated (VM may block upon call_VM return)
2240 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2241 __ jcc(Assembler::notEqual, dispatch);
2242
2243 // We have the address of an on stack replacement routine in rax
2244 // We need to prepare to execute the OSR method. First we must
2245 // migrate the locals and monitors off of the stack.
2246
2247 LP64_ONLY(__ mov(r13, rax)); // save the nmethod
2248 NOT_LP64(__ mov(rbx, rax)); // save the nmethod
2249 NOT_LP64(__ get_thread(rcx));
2250
2251 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2252
2253 // rax is OSR buffer, move it to expected parameter location
2254 LP64_ONLY(__ mov(j_rarg0, rax));
2255 NOT_LP64(__ mov(rcx, rax));
2256 // We use j_rarg definitions here so that registers don't conflict as parameter
2257 // registers change across platforms as we are in the midst of a calling
2258 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2259
2260 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2261 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2262
2263
2264 // pop the interpreter frame
2265 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2266 __ leave(); // remove frame anchor
2267 __ pop(retaddr); // get return address
2268 __ mov(rsp, sender_sp); // set sp to sender sp
2269 // Ensure compiled code always sees stack at proper alignment
2270 __ andptr(rsp, -(StackAlignmentInBytes));
2271
2272 // unlike x86 we need no specialized return from compiled code
2273 // to the interpreter or the call stub.
2274
2275 // push the return address
2276 __ push(retaddr);
2277
2278 // and begin the OSR nmethod
2279 LP64_ONLY(__ jmp(Address(r13, nmethod::osr_entry_point_offset())));
2280 NOT_LP64(__ jmp(Address(rbx, nmethod::osr_entry_point_offset())));
2281 }
2282 }
2283 }
2284
2285 void TemplateTable::if_0cmp(Condition cc) {
2286 transition(itos, vtos);
2287 // assume branch is more often taken than not (loops use backward branches)
2288 Label not_taken;
2289 __ testl(rax, rax);
2290 __ jcc(j_not(cc), not_taken);
2291 branch(false, false);
2292 __ bind(not_taken);
2293 __ profile_not_taken_branch(rax);
2294 }
2295
2296 void TemplateTable::if_icmp(Condition cc) {
2297 transition(itos, vtos);
2298 // assume branch is more often taken than not (loops use backward branches)
2299 Label not_taken;
2300 __ pop_i(rdx);
2301 __ cmpl(rdx, rax);
2302 __ jcc(j_not(cc), not_taken);
2303 branch(false, false);
2304 __ bind(not_taken);
2305 __ profile_not_taken_branch(rax);
2306 }
2307
2308 void TemplateTable::if_nullcmp(Condition cc) {
2309 transition(atos, vtos);
2310 // assume branch is more often taken than not (loops use backward branches)
2311 Label not_taken;
2312 __ testptr(rax, rax);
2313 __ jcc(j_not(cc), not_taken);
2314 branch(false, false);
2315 __ bind(not_taken);
2316 __ profile_not_taken_branch(rax);
2317 }
2318
2319 void TemplateTable::if_acmp(Condition cc) {
2320 transition(atos, vtos);
2321 // assume branch is more often taken than not (loops use backward branches)
2322 Label not_taken;
2323 __ pop_ptr(rdx);
2324 __ cmpptr(rdx, rax);
2325 __ jcc(j_not(cc), not_taken);
2326 branch(false, false);
2327 __ bind(not_taken);
2328 __ profile_not_taken_branch(rax);
2329 }
2330
2331 void TemplateTable::ret() {
2332 transition(vtos, vtos);
2333 locals_index(rbx);
2334 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2335 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2336 __ profile_ret(rbx, rcx);
2337 __ get_method(rax);
2338 __ movptr(rbcp, Address(rax, Method::const_offset()));
2339 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2340 ConstMethod::codes_offset()));
2341 __ dispatch_next(vtos);
2342 }
2343
2344 void TemplateTable::wide_ret() {
2345 transition(vtos, vtos);
2346 locals_index_wide(rbx);
2347 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2348 __ profile_ret(rbx, rcx);
2349 __ get_method(rax);
2350 __ movptr(rbcp, Address(rax, Method::const_offset()));
2351 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2352 __ dispatch_next(vtos);
2353 }
2354
2355 void TemplateTable::tableswitch() {
2356 Label default_case, continue_execution;
2357 transition(itos, vtos);
2358
2359 // align r13/rsi
2360 __ lea(rbx, at_bcp(BytesPerInt));
2361 __ andptr(rbx, -BytesPerInt);
2362 // load lo & hi
2363 __ movl(rcx, Address(rbx, BytesPerInt));
2364 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2365 __ bswapl(rcx);
2366 __ bswapl(rdx);
2367 // check against lo & hi
2368 __ cmpl(rax, rcx);
2369 __ jcc(Assembler::less, default_case);
2370 __ cmpl(rax, rdx);
2371 __ jcc(Assembler::greater, default_case);
2372 // lookup dispatch offset
2373 __ subl(rax, rcx);
2374 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2375 __ profile_switch_case(rax, rbx, rcx);
2376 // continue execution
2377 __ bind(continue_execution);
2378 __ bswapl(rdx);
2379 LP64_ONLY(__ movl2ptr(rdx, rdx));
2380 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2381 __ addptr(rbcp, rdx);
2382 __ dispatch_only(vtos);
2383 // handle default
2384 __ bind(default_case);
2385 __ profile_switch_default(rax);
2386 __ movl(rdx, Address(rbx, 0));
2387 __ jmp(continue_execution);
2388 }
2389
2390 void TemplateTable::lookupswitch() {
2391 transition(itos, itos);
2392 __ stop("lookupswitch bytecode should have been rewritten");
2393 }
2394
2395 void TemplateTable::fast_linearswitch() {
2396 transition(itos, vtos);
2397 Label loop_entry, loop, found, continue_execution;
2398 // bswap rax so we can avoid bswapping the table entries
2399 __ bswapl(rax);
2400 // align r13
2401 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2402 // this instruction (change offsets
2403 // below)
2404 __ andptr(rbx, -BytesPerInt);
2405 // set counter
2406 __ movl(rcx, Address(rbx, BytesPerInt));
2407 __ bswapl(rcx);
2408 __ jmpb(loop_entry);
2409 // table search
2410 __ bind(loop);
2411 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2412 __ jcc(Assembler::equal, found);
2413 __ bind(loop_entry);
2414 __ decrementl(rcx);
2415 __ jcc(Assembler::greaterEqual, loop);
2416 // default case
2417 __ profile_switch_default(rax);
2418 __ movl(rdx, Address(rbx, 0));
2419 __ jmp(continue_execution);
2420 // entry found -> get offset
2421 __ bind(found);
2422 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2423 __ profile_switch_case(rcx, rax, rbx);
2424 // continue execution
2425 __ bind(continue_execution);
2426 __ bswapl(rdx);
2427 __ movl2ptr(rdx, rdx);
2428 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2429 __ addptr(rbcp, rdx);
2430 __ dispatch_only(vtos);
2431 }
2432
2433 void TemplateTable::fast_binaryswitch() {
2434 transition(itos, vtos);
2435 // Implementation using the following core algorithm:
2436 //
2437 // int binary_search(int key, LookupswitchPair* array, int n) {
2438 // // Binary search according to "Methodik des Programmierens" by
2439 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2440 // int i = 0;
2441 // int j = n;
2442 // while (i+1 < j) {
2443 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2444 // // with Q: for all i: 0 <= i < n: key < a[i]
2445 // // where a stands for the array and assuming that the (inexisting)
2446 // // element a[n] is infinitely big.
2447 // int h = (i + j) >> 1;
2448 // // i < h < j
2449 // if (key < array[h].fast_match()) {
2450 // j = h;
2451 // } else {
2452 // i = h;
2453 // }
2454 // }
2455 // // R: a[i] <= key < a[i+1] or Q
2456 // // (i.e., if key is within array, i is the correct index)
2457 // return i;
2458 // }
2459
2460 // Register allocation
2461 const Register key = rax; // already set (tosca)
2462 const Register array = rbx;
2463 const Register i = rcx;
2464 const Register j = rdx;
2465 const Register h = rdi;
2466 const Register temp = rsi;
2467
2468 // Find array start
2469 NOT_LP64(__ save_bcp());
2470
2471 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2472 // get rid of this
2473 // instruction (change
2474 // offsets below)
2475 __ andptr(array, -BytesPerInt);
2476
2477 // Initialize i & j
2478 __ xorl(i, i); // i = 0;
2479 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2480
2481 // Convert j into native byteordering
2482 __ bswapl(j);
2483
2484 // And start
2485 Label entry;
2486 __ jmp(entry);
2487
2488 // binary search loop
2489 {
2490 Label loop;
2491 __ bind(loop);
2492 // int h = (i + j) >> 1;
2493 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2494 __ sarl(h, 1); // h = (i + j) >> 1;
2495 // if (key < array[h].fast_match()) {
2496 // j = h;
2497 // } else {
2498 // i = h;
2499 // }
2500 // Convert array[h].match to native byte-ordering before compare
2501 __ movl(temp, Address(array, h, Address::times_8));
2502 __ bswapl(temp);
2503 __ cmpl(key, temp);
2504 // j = h if (key < array[h].fast_match())
2505 __ cmov32(Assembler::less, j, h);
2506 // i = h if (key >= array[h].fast_match())
2507 __ cmov32(Assembler::greaterEqual, i, h);
2508 // while (i+1 < j)
2509 __ bind(entry);
2510 __ leal(h, Address(i, 1)); // i+1
2511 __ cmpl(h, j); // i+1 < j
2512 __ jcc(Assembler::less, loop);
2513 }
2514
2515 // end of binary search, result index is i (must check again!)
2516 Label default_case;
2517 // Convert array[i].match to native byte-ordering before compare
2518 __ movl(temp, Address(array, i, Address::times_8));
2519 __ bswapl(temp);
2520 __ cmpl(key, temp);
2521 __ jcc(Assembler::notEqual, default_case);
2522
2523 // entry found -> j = offset
2524 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2525 __ profile_switch_case(i, key, array);
2526 __ bswapl(j);
2527 LP64_ONLY(__ movslq(j, j));
2528
2529 NOT_LP64(__ restore_bcp());
2530 NOT_LP64(__ restore_locals()); // restore rdi
2531
2532 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2533 __ addptr(rbcp, j);
2534 __ dispatch_only(vtos);
2535
2536 // default case -> j = default offset
2537 __ bind(default_case);
2538 __ profile_switch_default(i);
2539 __ movl(j, Address(array, -2 * BytesPerInt));
2540 __ bswapl(j);
2541 LP64_ONLY(__ movslq(j, j));
2542
2543 NOT_LP64(__ restore_bcp());
2544 NOT_LP64(__ restore_locals());
2545
2546 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2547 __ addptr(rbcp, j);
2548 __ dispatch_only(vtos);
2549 }
2550
2551 void TemplateTable::_return(TosState state) {
2552 transition(state, state);
2553
2554 assert(_desc->calls_vm(),
2555 "inconsistent calls_vm information"); // call in remove_activation
2556
2557 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2558 assert(state == vtos, "only valid state");
2559 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2560 __ movptr(robj, aaddress(0));
2561 __ load_klass(rdi, robj);
2562 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2563 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2564 Label skip_register_finalizer;
2565 __ jcc(Assembler::zero, skip_register_finalizer);
2566
2567 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2568
2569 __ bind(skip_register_finalizer);
2570 }
2571
2572 // Narrow result if state is itos but result type is smaller.
2573 // Need to narrow in the return bytecode rather than in generate_return_entry
2574 // since compiled code callers expect the result to already be narrowed.
2575 if (state == itos) {
2576 __ narrow(rax);
2577 }
2578 __ remove_activation(state, rbcp);
2579
2580 __ jmp(rbcp);
2581 }
2582
2583 // ----------------------------------------------------------------------------
2584 // Volatile variables demand their effects be made known to all CPU's
2585 // in order. Store buffers on most chips allow reads & writes to
2586 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2587 // without some kind of memory barrier (i.e., it's not sufficient that
2588 // the interpreter does not reorder volatile references, the hardware
2589 // also must not reorder them).
2590 //
2591 // According to the new Java Memory Model (JMM):
2592 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2593 // writes act as aquire & release, so:
2594 // (2) A read cannot let unrelated NON-volatile memory refs that
2595 // happen after the read float up to before the read. It's OK for
2596 // non-volatile memory refs that happen before the volatile read to
2597 // float down below it.
2598 // (3) Similar a volatile write cannot let unrelated NON-volatile
2599 // memory refs that happen BEFORE the write float down to after the
2600 // write. It's OK for non-volatile memory refs that happen after the
2601 // volatile write to float up before it.
2602 //
2603 // We only put in barriers around volatile refs (they are expensive),
2604 // not _between_ memory refs (that would require us to track the
2605 // flavor of the previous memory refs). Requirements (2) and (3)
2606 // require some barriers before volatile stores and after volatile
2607 // loads. These nearly cover requirement (1) but miss the
2608 // volatile-store-volatile-load case. This final case is placed after
2609 // volatile-stores although it could just as well go before
2610 // volatile-loads.
2611
2612 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2613 // Helper function to insert a is-volatile test and memory barrier
2614 if(!os::is_MP()) return; // Not needed on single CPU
2615 __ membar(order_constraint);
2616 }
2617
2618 void TemplateTable::resolve_cache_and_index(int byte_no,
2619 Register Rcache,
2620 Register index,
2621 size_t index_size) {
2622 const Register temp = rbx;
2623 assert_different_registers(Rcache, index, temp);
2624
2625 Label resolved;
2626
2627 Bytecodes::Code code = bytecode();
2628 switch (code) {
2629 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2630 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2631 }
2632
2633 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2634 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2635 __ cmpl(temp, code); // have we resolved this bytecode?
2636 __ jcc(Assembler::equal, resolved);
2637
2638 // resolve first time through
2639 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2640 __ movl(temp, code);
2641 __ call_VM(noreg, entry, temp);
2642 // Update registers with resolved info
2643 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2644 __ bind(resolved);
2645 }
2646
2647 // The cache and index registers must be set before call
2648 void TemplateTable::load_field_cp_cache_entry(Register obj,
2649 Register cache,
2650 Register index,
2651 Register off,
2652 Register flags,
2653 bool is_static = false) {
2654 assert_different_registers(cache, index, flags, off);
2655
2656 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2657 // Field offset
2658 __ movptr(off, Address(cache, index, Address::times_ptr,
2659 in_bytes(cp_base_offset +
2660 ConstantPoolCacheEntry::f2_offset())));
2661 // Flags
2662 __ movl(flags, Address(cache, index, Address::times_ptr,
2663 in_bytes(cp_base_offset +
2664 ConstantPoolCacheEntry::flags_offset())));
2665
2666 // klass overwrite register
2667 if (is_static) {
2668 __ movptr(obj, Address(cache, index, Address::times_ptr,
2669 in_bytes(cp_base_offset +
2670 ConstantPoolCacheEntry::f1_offset())));
2671 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2672 __ movptr(obj, Address(obj, mirror_offset));
2673 }
2674 }
2675
2676 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2677 Register method,
2678 Register itable_index,
2679 Register flags,
2680 bool is_invokevirtual,
2681 bool is_invokevfinal, /*unused*/
2682 bool is_invokedynamic) {
2683 // setup registers
2684 const Register cache = rcx;
2685 const Register index = rdx;
2686 assert_different_registers(method, flags);
2687 assert_different_registers(method, cache, index);
2688 assert_different_registers(itable_index, flags);
2689 assert_different_registers(itable_index, cache, index);
2690 // determine constant pool cache field offsets
2691 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2692 const int method_offset = in_bytes(
2693 ConstantPoolCache::base_offset() +
2694 ((byte_no == f2_byte)
2695 ? ConstantPoolCacheEntry::f2_offset()
2696 : ConstantPoolCacheEntry::f1_offset()));
2697 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2698 ConstantPoolCacheEntry::flags_offset());
2699 // access constant pool cache fields
2700 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2701 ConstantPoolCacheEntry::f2_offset());
2702
2703 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2704 resolve_cache_and_index(byte_no, cache, index, index_size);
2705 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2706
2707 if (itable_index != noreg) {
2708 // pick up itable or appendix index from f2 also:
2709 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2710 }
2711 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2712 }
2713
2714 // The registers cache and index expected to be set before call.
2715 // Correct values of the cache and index registers are preserved.
2716 void TemplateTable::jvmti_post_field_access(Register cache,
2717 Register index,
2718 bool is_static,
2719 bool has_tos) {
2720 if (JvmtiExport::can_post_field_access()) {
2721 // Check to see if a field access watch has been set before we take
2722 // the time to call into the VM.
2723 Label L1;
2724 assert_different_registers(cache, index, rax);
2725 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2726 __ testl(rax,rax);
2727 __ jcc(Assembler::zero, L1);
2728
2729 // cache entry pointer
2730 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2731 __ shll(index, LogBytesPerWord);
2732 __ addptr(cache, index);
2733 if (is_static) {
2734 __ xorptr(rax, rax); // NULL object reference
2735 } else {
2736 __ pop(atos); // Get the object
2737 __ verify_oop(rax);
2738 __ push(atos); // Restore stack state
2739 }
2740 // rax,: object pointer or NULL
2741 // cache: cache entry pointer
2742 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2743 rax, cache);
2744 __ get_cache_and_index_at_bcp(cache, index, 1);
2745 __ bind(L1);
2746 }
2747 }
2748
2749 void TemplateTable::pop_and_check_object(Register r) {
2750 __ pop_ptr(r);
2751 __ null_check(r); // for field access must check obj.
2752 __ verify_oop(r);
2753 }
2754
2755 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2756 transition(vtos, vtos);
2757
2758 const Register cache = rcx;
2759 const Register index = rdx;
2760 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2761 const Register off = rbx;
2762 const Register flags = rax;
2763 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2764
2765 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2766 jvmti_post_field_access(cache, index, is_static, false);
2767 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2768
2769 if (!is_static) pop_and_check_object(obj);
2770
2771 const Address field(obj, off, Address::times_1, 0*wordSize);
2772 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2773
2774 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2775
2776 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2777 // Make sure we don't need to mask edx after the above shift
2778 assert(btos == 0, "change code, btos != 0");
2779
2780 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2781
2782 __ jcc(Assembler::notZero, notByte);
2783 // btos
2784 __ load_signed_byte(rax, field);
2785 __ push(btos);
2786 // Rewrite bytecode to be faster
2787 if (!is_static && rc == may_rewrite) {
2788 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2789 }
2790 __ jmp(Done);
2791
2792 __ bind(notByte);
2793 __ cmpl(flags, ztos);
2794 __ jcc(Assembler::notEqual, notBool);
2795
2796 // ztos (same code as btos)
2797 __ load_signed_byte(rax, field);
2798 __ push(ztos);
2799 // Rewrite bytecode to be faster
2800 if (!is_static && rc == may_rewrite) {
2801 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2802 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2803 }
2804 __ jmp(Done);
2805
2806 __ bind(notBool);
2807 __ cmpl(flags, atos);
2808 __ jcc(Assembler::notEqual, notObj);
2809 // atos
2810 __ load_heap_oop(rax, field);
2811 __ push(atos);
2812 if (!is_static && rc == may_rewrite) {
2813 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2814 }
2815 __ jmp(Done);
2816
2817 __ bind(notObj);
2818 __ cmpl(flags, itos);
2819 __ jcc(Assembler::notEqual, notInt);
2820 // itos
2821 __ movl(rax, field);
2822 __ push(itos);
2823 // Rewrite bytecode to be faster
2824 if (!is_static && rc == may_rewrite) {
2825 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2826 }
2827 __ jmp(Done);
2828
2829 __ bind(notInt);
2830 __ cmpl(flags, ctos);
2831 __ jcc(Assembler::notEqual, notChar);
2832 // ctos
2833 __ load_unsigned_short(rax, field);
2834 __ push(ctos);
2835 // Rewrite bytecode to be faster
2836 if (!is_static && rc == may_rewrite) {
2837 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2838 }
2839 __ jmp(Done);
2840
2841 __ bind(notChar);
2842 __ cmpl(flags, stos);
2843 __ jcc(Assembler::notEqual, notShort);
2844 // stos
2845 __ load_signed_short(rax, field);
2846 __ push(stos);
2847 // Rewrite bytecode to be faster
2848 if (!is_static && rc == may_rewrite) {
2849 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2850 }
2851 __ jmp(Done);
2852
2853 __ bind(notShort);
2854 __ cmpl(flags, ltos);
2855 __ jcc(Assembler::notEqual, notLong);
2856 // ltos
2857
2858 #ifndef _LP64
2859 // Generate code as if volatile. There just aren't enough registers to
2860 // save that information and this code is faster than the test.
2861 __ fild_d(field); // Must load atomically
2862 __ subptr(rsp,2*wordSize); // Make space for store
2863 __ fistp_d(Address(rsp,0));
2864 __ pop(rax);
2865 __ pop(rdx);
2866 #else
2867 __ movq(rax, field);
2868 #endif
2869
2870 __ push(ltos);
2871 // Rewrite bytecode to be faster
2872 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2873 __ jmp(Done);
2874
2875 __ bind(notLong);
2876 __ cmpl(flags, ftos);
2877 __ jcc(Assembler::notEqual, notFloat);
2878 // ftos
2879
2880 __ load_float(field);
2881 __ push(ftos);
2882 // Rewrite bytecode to be faster
2883 if (!is_static && rc == may_rewrite) {
2884 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2885 }
2886 __ jmp(Done);
2887
2888 __ bind(notFloat);
2889 #ifdef ASSERT
2890 __ cmpl(flags, dtos);
2891 __ jcc(Assembler::notEqual, notDouble);
2892 #endif
2893 // dtos
2894 __ load_double(field);
2895 __ push(dtos);
2896 // Rewrite bytecode to be faster
2897 if (!is_static && rc == may_rewrite) {
2898 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2899 }
2900 #ifdef ASSERT
2901 __ jmp(Done);
2902
2903
2904 __ bind(notDouble);
2905 __ stop("Bad state");
2906 #endif
2907
2908 __ bind(Done);
2909 // [jk] not needed currently
2910 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2911 // Assembler::LoadStore));
2912 }
2913
2914 void TemplateTable::getfield(int byte_no) {
2915 getfield_or_static(byte_no, false);
2916 }
2917
2918 void TemplateTable::nofast_getfield(int byte_no) {
2919 getfield_or_static(byte_no, false, may_not_rewrite);
2920 }
2921
2922 void TemplateTable::getstatic(int byte_no) {
2923 getfield_or_static(byte_no, true);
2924 }
2925
2926
2927 // The registers cache and index expected to be set before call.
2928 // The function may destroy various registers, just not the cache and index registers.
2929 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2930
2931 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
2932 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2933 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2934 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
2935
2936 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2937
2938 if (JvmtiExport::can_post_field_modification()) {
2939 // Check to see if a field modification watch has been set before
2940 // we take the time to call into the VM.
2941 Label L1;
2942 assert_different_registers(cache, index, rax);
2943 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2944 __ testl(rax, rax);
2945 __ jcc(Assembler::zero, L1);
2946
2947 __ get_cache_and_index_at_bcp(robj, RDX, 1);
2948
2949
2950 if (is_static) {
2951 // Life is simple. Null out the object pointer.
2952 __ xorl(RBX, RBX);
2953
2954 } else {
2955 // Life is harder. The stack holds the value on top, followed by
2956 // the object. We don't know the size of the value, though; it
2957 // could be one or two words depending on its type. As a result,
2958 // we must find the type to determine where the object is.
2959 #ifndef _LP64
2960 Label two_word, valsize_known;
2961 #endif
2962 __ movl(RCX, Address(robj, RDX,
2963 Address::times_ptr,
2964 in_bytes(cp_base_offset +
2965 ConstantPoolCacheEntry::flags_offset())));
2966 NOT_LP64(__ mov(rbx, rsp));
2967 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
2968
2969 // Make sure we don't need to mask rcx after the above shift
2970 ConstantPoolCacheEntry::verify_tos_state_shift();
2971 #ifdef _LP64
2972 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2973 __ cmpl(c_rarg3, ltos);
2974 __ cmovptr(Assembler::equal,
2975 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2976 __ cmpl(c_rarg3, dtos);
2977 __ cmovptr(Assembler::equal,
2978 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2979 #else
2980 __ cmpl(rcx, ltos);
2981 __ jccb(Assembler::equal, two_word);
2982 __ cmpl(rcx, dtos);
2983 __ jccb(Assembler::equal, two_word);
2984 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2985 __ jmpb(valsize_known);
2986
2987 __ bind(two_word);
2988 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2989
2990 __ bind(valsize_known);
2991 // setup object pointer
2992 __ movptr(rbx, Address(rbx, 0));
2993 #endif
2994 }
2995 // cache entry pointer
2996 __ addptr(robj, in_bytes(cp_base_offset));
2997 __ shll(RDX, LogBytesPerWord);
2998 __ addptr(robj, RDX);
2999 // object (tos)
3000 __ mov(RCX, rsp);
3001 // c_rarg1: object pointer set up above (NULL if static)
3002 // c_rarg2: cache entry pointer
3003 // c_rarg3: jvalue object on the stack
3004 __ call_VM(noreg,
3005 CAST_FROM_FN_PTR(address,
3006 InterpreterRuntime::post_field_modification),
3007 RBX, robj, RCX);
3008 __ get_cache_and_index_at_bcp(cache, index, 1);
3009 __ bind(L1);
3010 }
3011 }
3012
3013 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3014 transition(vtos, vtos);
3015
3016 const Register cache = rcx;
3017 const Register index = rdx;
3018 const Register obj = rcx;
3019 const Register off = rbx;
3020 const Register flags = rax;
3021 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3022
3023 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3024 jvmti_post_field_mod(cache, index, is_static);
3025 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3026
3027 // [jk] not needed currently
3028 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3029 // Assembler::StoreStore));
3030
3031 Label notVolatile, Done;
3032 __ movl(rdx, flags);
3033 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3034 __ andl(rdx, 0x1);
3035
3036 // field addresses
3037 const Address field(obj, off, Address::times_1, 0*wordSize);
3038 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3039
3040 Label notByte, notBool, notInt, notShort, notChar,
3041 notLong, notFloat, notObj, notDouble;
3042
3043 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3044
3045 assert(btos == 0, "change code, btos != 0");
3046 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3047 __ jcc(Assembler::notZero, notByte);
3048
3049 // btos
3050 {
3051 __ pop(btos);
3052 if (!is_static) pop_and_check_object(obj);
3053 __ movb(field, rax);
3054 if (!is_static && rc == may_rewrite) {
3055 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3056 }
3057 __ jmp(Done);
3058 }
3059
3060 __ bind(notByte);
3061 __ cmpl(flags, ztos);
3062 __ jcc(Assembler::notEqual, notBool);
3063
3064 // ztos
3065 {
3066 __ pop(ztos);
3067 if (!is_static) pop_and_check_object(obj);
3068 __ andl(rax, 0x1);
3069 __ movb(field, rax);
3070 if (!is_static && rc == may_rewrite) {
3071 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3072 }
3073 __ jmp(Done);
3074 }
3075
3076 __ bind(notBool);
3077 __ cmpl(flags, atos);
3078 __ jcc(Assembler::notEqual, notObj);
3079
3080 // atos
3081 {
3082 __ pop(atos);
3083 if (!is_static) pop_and_check_object(obj);
3084 // Store into the field
3085 do_oop_store(_masm, field, rax, _bs->kind(), false);
3086 if (!is_static && rc == may_rewrite) {
3087 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3088 }
3089 __ jmp(Done);
3090 }
3091
3092 __ bind(notObj);
3093 __ cmpl(flags, itos);
3094 __ jcc(Assembler::notEqual, notInt);
3095
3096 // itos
3097 {
3098 __ pop(itos);
3099 if (!is_static) pop_and_check_object(obj);
3100 __ movl(field, rax);
3101 if (!is_static && rc == may_rewrite) {
3102 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3103 }
3104 __ jmp(Done);
3105 }
3106
3107 __ bind(notInt);
3108 __ cmpl(flags, ctos);
3109 __ jcc(Assembler::notEqual, notChar);
3110
3111 // ctos
3112 {
3113 __ pop(ctos);
3114 if (!is_static) pop_and_check_object(obj);
3115 __ movw(field, rax);
3116 if (!is_static && rc == may_rewrite) {
3117 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3118 }
3119 __ jmp(Done);
3120 }
3121
3122 __ bind(notChar);
3123 __ cmpl(flags, stos);
3124 __ jcc(Assembler::notEqual, notShort);
3125
3126 // stos
3127 {
3128 __ pop(stos);
3129 if (!is_static) pop_and_check_object(obj);
3130 __ movw(field, rax);
3131 if (!is_static && rc == may_rewrite) {
3132 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3133 }
3134 __ jmp(Done);
3135 }
3136
3137 __ bind(notShort);
3138 __ cmpl(flags, ltos);
3139 __ jcc(Assembler::notEqual, notLong);
3140
3141 // ltos
3142 #ifdef _LP64
3143 {
3144 __ pop(ltos);
3145 if (!is_static) pop_and_check_object(obj);
3146 __ movq(field, rax);
3147 if (!is_static && rc == may_rewrite) {
3148 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3149 }
3150 __ jmp(Done);
3151 }
3152 #else
3153 {
3154 Label notVolatileLong;
3155 __ testl(rdx, rdx);
3156 __ jcc(Assembler::zero, notVolatileLong);
3157
3158 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3159 if (!is_static) pop_and_check_object(obj);
3160
3161 // Replace with real volatile test
3162 __ push(rdx);
3163 __ push(rax); // Must update atomically with FIST
3164 __ fild_d(Address(rsp,0)); // So load into FPU register
3165 __ fistp_d(field); // and put into memory atomically
3166 __ addptr(rsp, 2*wordSize);
3167 // volatile_barrier();
3168 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3169 Assembler::StoreStore));
3170 // Don't rewrite volatile version
3171 __ jmp(notVolatile);
3172
3173 __ bind(notVolatileLong);
3174
3175 __ pop(ltos); // overwrites rdx
3176 if (!is_static) pop_and_check_object(obj);
3177 __ movptr(hi, rdx);
3178 __ movptr(field, rax);
3179 // Don't rewrite to _fast_lputfield for potential volatile case.
3180 __ jmp(notVolatile);
3181 }
3182 #endif // _LP64
3183
3184 __ bind(notLong);
3185 __ cmpl(flags, ftos);
3186 __ jcc(Assembler::notEqual, notFloat);
3187
3188 // ftos
3189 {
3190 __ pop(ftos);
3191 if (!is_static) pop_and_check_object(obj);
3192 __ store_float(field);
3193 if (!is_static && rc == may_rewrite) {
3194 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3195 }
3196 __ jmp(Done);
3197 }
3198
3199 __ bind(notFloat);
3200 #ifdef ASSERT
3201 __ cmpl(flags, dtos);
3202 __ jcc(Assembler::notEqual, notDouble);
3203 #endif
3204
3205 // dtos
3206 {
3207 __ pop(dtos);
3208 if (!is_static) pop_and_check_object(obj);
3209 __ store_double(field);
3210 if (!is_static && rc == may_rewrite) {
3211 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3212 }
3213 }
3214
3215 #ifdef ASSERT
3216 __ jmp(Done);
3217
3218 __ bind(notDouble);
3219 __ stop("Bad state");
3220 #endif
3221
3222 __ bind(Done);
3223
3224 // Check for volatile store
3225 __ testl(rdx, rdx);
3226 __ jcc(Assembler::zero, notVolatile);
3227 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3228 Assembler::StoreStore));
3229 __ bind(notVolatile);
3230 }
3231
3232 void TemplateTable::putfield(int byte_no) {
3233 putfield_or_static(byte_no, false);
3234 }
3235
3236 void TemplateTable::nofast_putfield(int byte_no) {
3237 putfield_or_static(byte_no, false, may_not_rewrite);
3238 }
3239
3240 void TemplateTable::putstatic(int byte_no) {
3241 putfield_or_static(byte_no, true);
3242 }
3243
3244 void TemplateTable::jvmti_post_fast_field_mod() {
3245
3246 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3247
3248 if (JvmtiExport::can_post_field_modification()) {
3249 // Check to see if a field modification watch has been set before
3250 // we take the time to call into the VM.
3251 Label L2;
3252 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3253 __ testl(scratch, scratch);
3254 __ jcc(Assembler::zero, L2);
3255 __ pop_ptr(rbx); // copy the object pointer from tos
3256 __ verify_oop(rbx);
3257 __ push_ptr(rbx); // put the object pointer back on tos
3258 // Save tos values before call_VM() clobbers them. Since we have
3259 // to do it for every data type, we use the saved values as the
3260 // jvalue object.
3261 switch (bytecode()) { // load values into the jvalue object
3262 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3263 case Bytecodes::_fast_bputfield: // fall through
3264 case Bytecodes::_fast_zputfield: // fall through
3265 case Bytecodes::_fast_sputfield: // fall through
3266 case Bytecodes::_fast_cputfield: // fall through
3267 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3268 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3269 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3270 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3271
3272 default:
3273 ShouldNotReachHere();
3274 }
3275 __ mov(scratch, rsp); // points to jvalue on the stack
3276 // access constant pool cache entry
3277 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3278 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3279 __ verify_oop(rbx);
3280 // rbx: object pointer copied above
3281 // c_rarg2: cache entry pointer
3282 // c_rarg3: jvalue object on the stack
3283 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3284 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3285
3286 switch (bytecode()) { // restore tos values
3287 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3288 case Bytecodes::_fast_bputfield: // fall through
3289 case Bytecodes::_fast_zputfield: // fall through
3290 case Bytecodes::_fast_sputfield: // fall through
3291 case Bytecodes::_fast_cputfield: // fall through
3292 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3293 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3294 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3295 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3296 }
3297 __ bind(L2);
3298 }
3299 }
3300
3301 void TemplateTable::fast_storefield(TosState state) {
3302 transition(state, vtos);
3303
3304 ByteSize base = ConstantPoolCache::base_offset();
3305
3306 jvmti_post_fast_field_mod();
3307
3308 // access constant pool cache
3309 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3310
3311 // test for volatile with rdx but rdx is tos register for lputfield.
3312 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3313 in_bytes(base +
3314 ConstantPoolCacheEntry::flags_offset())));
3315
3316 // replace index with field offset from cache entry
3317 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3318 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3319
3320 // [jk] not needed currently
3321 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3322 // Assembler::StoreStore));
3323
3324 Label notVolatile;
3325 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3326 __ andl(rdx, 0x1);
3327
3328 // Get object from stack
3329 pop_and_check_object(rcx);
3330
3331 // field address
3332 const Address field(rcx, rbx, Address::times_1);
3333
3334 // access field
3335 switch (bytecode()) {
3336 case Bytecodes::_fast_aputfield:
3337 do_oop_store(_masm, field, rax, _bs->kind(), false);
3338 break;
3339 case Bytecodes::_fast_lputfield:
3340 #ifdef _LP64
3341 __ movq(field, rax);
3342 #else
3343 __ stop("should not be rewritten");
3344 #endif
3345 break;
3346 case Bytecodes::_fast_iputfield:
3347 __ movl(field, rax);
3348 break;
3349 case Bytecodes::_fast_zputfield:
3350 __ andl(rax, 0x1); // boolean is true if LSB is 1
3351 // fall through to bputfield
3352 case Bytecodes::_fast_bputfield:
3353 __ movb(field, rax);
3354 break;
3355 case Bytecodes::_fast_sputfield:
3356 // fall through
3357 case Bytecodes::_fast_cputfield:
3358 __ movw(field, rax);
3359 break;
3360 case Bytecodes::_fast_fputfield:
3361 __ store_float(field);
3362 break;
3363 case Bytecodes::_fast_dputfield:
3364 __ store_double(field);
3365 break;
3366 default:
3367 ShouldNotReachHere();
3368 }
3369
3370 // Check for volatile store
3371 __ testl(rdx, rdx);
3372 __ jcc(Assembler::zero, notVolatile);
3373 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3374 Assembler::StoreStore));
3375 __ bind(notVolatile);
3376 }
3377
3378 void TemplateTable::fast_accessfield(TosState state) {
3379 transition(atos, state);
3380
3381 // Do the JVMTI work here to avoid disturbing the register state below
3382 if (JvmtiExport::can_post_field_access()) {
3383 // Check to see if a field access watch has been set before we
3384 // take the time to call into the VM.
3385 Label L1;
3386 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3387 __ testl(rcx, rcx);
3388 __ jcc(Assembler::zero, L1);
3389 // access constant pool cache entry
3390 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3391 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3392 __ verify_oop(rax);
3393 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3394 LP64_ONLY(__ mov(c_rarg1, rax));
3395 // c_rarg1: object pointer copied above
3396 // c_rarg2: cache entry pointer
3397 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3398 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3399 __ pop_ptr(rax); // restore object pointer
3400 __ bind(L1);
3401 }
3402
3403 // access constant pool cache
3404 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3405 // replace index with field offset from cache entry
3406 // [jk] not needed currently
3407 // if (os::is_MP()) {
3408 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3409 // in_bytes(ConstantPoolCache::base_offset() +
3410 // ConstantPoolCacheEntry::flags_offset())));
3411 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3412 // __ andl(rdx, 0x1);
3413 // }
3414 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3415 in_bytes(ConstantPoolCache::base_offset() +
3416 ConstantPoolCacheEntry::f2_offset())));
3417
3418 // rax: object
3419 __ verify_oop(rax);
3420 __ null_check(rax);
3421 Address field(rax, rbx, Address::times_1);
3422
3423 // access field
3424 switch (bytecode()) {
3425 case Bytecodes::_fast_agetfield:
3426 __ load_heap_oop(rax, field);
3427 __ verify_oop(rax);
3428 break;
3429 case Bytecodes::_fast_lgetfield:
3430 #ifdef _LP64
3431 __ movq(rax, field);
3432 #else
3433 __ stop("should not be rewritten");
3434 #endif
3435 break;
3436 case Bytecodes::_fast_igetfield:
3437 __ movl(rax, field);
3438 break;
3439 case Bytecodes::_fast_bgetfield:
3440 __ movsbl(rax, field);
3441 break;
3442 case Bytecodes::_fast_sgetfield:
3443 __ load_signed_short(rax, field);
3444 break;
3445 case Bytecodes::_fast_cgetfield:
3446 __ load_unsigned_short(rax, field);
3447 break;
3448 case Bytecodes::_fast_fgetfield:
3449 __ load_float(field);
3450 break;
3451 case Bytecodes::_fast_dgetfield:
3452 __ load_double(field);
3453 break;
3454 default:
3455 ShouldNotReachHere();
3456 }
3457 // [jk] not needed currently
3458 // if (os::is_MP()) {
3459 // Label notVolatile;
3460 // __ testl(rdx, rdx);
3461 // __ jcc(Assembler::zero, notVolatile);
3462 // __ membar(Assembler::LoadLoad);
3463 // __ bind(notVolatile);
3464 //};
3465 }
3466
3467 void TemplateTable::fast_xaccess(TosState state) {
3468 transition(vtos, state);
3469
3470 // get receiver
3471 __ movptr(rax, aaddress(0));
3472 // access constant pool cache
3473 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3474 __ movptr(rbx,
3475 Address(rcx, rdx, Address::times_ptr,
3476 in_bytes(ConstantPoolCache::base_offset() +
3477 ConstantPoolCacheEntry::f2_offset())));
3478 // make sure exception is reported in correct bcp range (getfield is
3479 // next instruction)
3480 __ increment(rbcp);
3481 __ null_check(rax);
3482 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3483 switch (state) {
3484 case itos:
3485 __ movl(rax, field);
3486 break;
3487 case atos:
3488 __ load_heap_oop(rax, field);
3489 __ verify_oop(rax);
3490 break;
3491 case ftos:
3492 __ load_float(field);
3493 break;
3494 default:
3495 ShouldNotReachHere();
3496 }
3497
3498 // [jk] not needed currently
3499 // if (os::is_MP()) {
3500 // Label notVolatile;
3501 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3502 // in_bytes(ConstantPoolCache::base_offset() +
3503 // ConstantPoolCacheEntry::flags_offset())));
3504 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3505 // __ testl(rdx, 0x1);
3506 // __ jcc(Assembler::zero, notVolatile);
3507 // __ membar(Assembler::LoadLoad);
3508 // __ bind(notVolatile);
3509 // }
3510
3511 __ decrement(rbcp);
3512 }
3513
3514 //-----------------------------------------------------------------------------
3515 // Calls
3516
3517 void TemplateTable::count_calls(Register method, Register temp) {
3518 // implemented elsewhere
3519 ShouldNotReachHere();
3520 }
3521
3522 void TemplateTable::prepare_invoke(int byte_no,
3523 Register method, // linked method (or i-klass)
3524 Register index, // itable index, MethodType, etc.
3525 Register recv, // if caller wants to see it
3526 Register flags // if caller wants to test it
3527 ) {
3528 // determine flags
3529 const Bytecodes::Code code = bytecode();
3530 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3531 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3532 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3533 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3534 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3535 const bool load_receiver = (recv != noreg);
3536 const bool save_flags = (flags != noreg);
3537 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3538 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3539 assert(flags == noreg || flags == rdx, "");
3540 assert(recv == noreg || recv == rcx, "");
3541
3542 // setup registers & access constant pool cache
3543 if (recv == noreg) recv = rcx;
3544 if (flags == noreg) flags = rdx;
3545 assert_different_registers(method, index, recv, flags);
3546
3547 // save 'interpreter return address'
3548 __ save_bcp();
3549
3550 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3551
3552 // maybe push appendix to arguments (just before return address)
3553 if (is_invokedynamic || is_invokehandle) {
3554 Label L_no_push;
3555 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3556 __ jcc(Assembler::zero, L_no_push);
3557 // Push the appendix as a trailing parameter.
3558 // This must be done before we get the receiver,
3559 // since the parameter_size includes it.
3560 __ push(rbx);
3561 __ mov(rbx, index);
3562 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3563 __ load_resolved_reference_at_index(index, rbx);
3564 __ pop(rbx);
3565 __ push(index); // push appendix (MethodType, CallSite, etc.)
3566 __ bind(L_no_push);
3567 }
3568
3569 // load receiver if needed (after appendix is pushed so parameter size is correct)
3570 // Note: no return address pushed yet
3571 if (load_receiver) {
3572 __ movl(recv, flags);
3573 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3574 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3575 const int receiver_is_at_end = -1; // back off one slot to get receiver
3576 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3577 __ movptr(recv, recv_addr);
3578 __ verify_oop(recv);
3579 }
3580
3581 if (save_flags) {
3582 __ movl(rbcp, flags);
3583 }
3584
3585 // compute return type
3586 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3587 // Make sure we don't need to mask flags after the above shift
3588 ConstantPoolCacheEntry::verify_tos_state_shift();
3589 // load return address
3590 {
3591 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3592 ExternalAddress table(table_addr);
3593 LP64_ONLY(__ lea(rscratch1, table));
3594 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3595 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3596 }
3597
3598 // push return address
3599 __ push(flags);
3600
3601 // Restore flags value from the constant pool cache, and restore rsi
3602 // for later null checks. r13 is the bytecode pointer
3603 if (save_flags) {
3604 __ movl(flags, rbcp);
3605 __ restore_bcp();
3606 }
3607 }
3608
3609 void TemplateTable::invokevirtual_helper(Register index,
3610 Register recv,
3611 Register flags) {
3612 // Uses temporary registers rax, rdx
3613 assert_different_registers(index, recv, rax, rdx);
3614 assert(index == rbx, "");
3615 assert(recv == rcx, "");
3616
3617 // Test for an invoke of a final method
3618 Label notFinal;
3619 __ movl(rax, flags);
3620 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3621 __ jcc(Assembler::zero, notFinal);
3622
3623 const Register method = index; // method must be rbx
3624 assert(method == rbx,
3625 "Method* must be rbx for interpreter calling convention");
3626
3627 // do the call - the index is actually the method to call
3628 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3629
3630 // It's final, need a null check here!
3631 __ null_check(recv);
3632
3633 // profile this call
3634 __ profile_final_call(rax);
3635 __ profile_arguments_type(rax, method, rbcp, true);
3636
3637 __ jump_from_interpreted(method, rax);
3638
3639 __ bind(notFinal);
3640
3641 // get receiver klass
3642 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3643 __ load_klass(rax, recv);
3644
3645 // profile this call
3646 __ profile_virtual_call(rax, rlocals, rdx);
3647 // get target Method* & entry point
3648 __ lookup_virtual_method(rax, index, method);
3649 __ profile_called_method(method, rdx, rbcp);
3650
3651 __ profile_arguments_type(rdx, method, rbcp, true);
3652 __ jump_from_interpreted(method, rdx);
3653 }
3654
3655 void TemplateTable::invokevirtual(int byte_no) {
3656 transition(vtos, vtos);
3657 assert(byte_no == f2_byte, "use this argument");
3658 prepare_invoke(byte_no,
3659 rbx, // method or vtable index
3660 noreg, // unused itable index
3661 rcx, rdx); // recv, flags
3662
3663 // rbx: index
3664 // rcx: receiver
3665 // rdx: flags
3666
3667 invokevirtual_helper(rbx, rcx, rdx);
3668 }
3669
3670 void TemplateTable::invokespecial(int byte_no) {
3671 transition(vtos, vtos);
3672 assert(byte_no == f1_byte, "use this argument");
3673 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3674 rcx); // get receiver also for null check
3675 __ verify_oop(rcx);
3676 __ null_check(rcx);
3677 // do the call
3678 __ profile_call(rax);
3679 __ profile_arguments_type(rax, rbx, rbcp, false);
3680 __ jump_from_interpreted(rbx, rax);
3681 }
3682
3683 void TemplateTable::invokestatic(int byte_no) {
3684 transition(vtos, vtos);
3685 assert(byte_no == f1_byte, "use this argument");
3686 prepare_invoke(byte_no, rbx); // get f1 Method*
3687 // do the call
3688 __ profile_call(rax);
3689 __ profile_arguments_type(rax, rbx, rbcp, false);
3690 __ jump_from_interpreted(rbx, rax);
3691 }
3692
3693
3694 void TemplateTable::fast_invokevfinal(int byte_no) {
3695 transition(vtos, vtos);
3696 assert(byte_no == f2_byte, "use this argument");
3697 __ stop("fast_invokevfinal not used on x86");
3698 }
3699
3700
3701 void TemplateTable::invokeinterface(int byte_no) {
3702 transition(vtos, vtos);
3703 assert(byte_no == f1_byte, "use this argument");
3704 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3705 rcx, rdx); // recv, flags
3706
3707 // rax: interface klass (from f1)
3708 // rbx: itable index (from f2)
3709 // rcx: receiver
3710 // rdx: flags
3711
3712 // Special case of invokeinterface called for virtual method of
3713 // java.lang.Object. See cpCacheOop.cpp for details.
3714 // This code isn't produced by javac, but could be produced by
3715 // another compliant java compiler.
3716 Label notMethod;
3717 __ movl(rlocals, rdx);
3718 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3719
3720 __ jcc(Assembler::zero, notMethod);
3721
3722 invokevirtual_helper(rbx, rcx, rdx);
3723 __ bind(notMethod);
3724
3725 // Get receiver klass into rdx - also a null check
3726 __ restore_locals(); // restore r14
3727 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3728 __ load_klass(rdx, rcx);
3729
3730 // profile this call
3731 __ profile_virtual_call(rdx, rbcp, rlocals);
3732
3733 Label no_such_interface, no_such_method;
3734
3735 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3736 rdx, rax, rbx,
3737 // outputs: method, scan temp. reg
3738 rbx, rbcp,
3739 no_such_interface);
3740
3741 // rbx: Method* to call
3742 // rcx: receiver
3743 // Check for abstract method error
3744 // Note: This should be done more efficiently via a throw_abstract_method_error
3745 // interpreter entry point and a conditional jump to it in case of a null
3746 // method.
3747 __ testptr(rbx, rbx);
3748 __ jcc(Assembler::zero, no_such_method);
3749
3750 __ profile_called_method(rbx, rbcp, rdx);
3751 __ profile_arguments_type(rdx, rbx, rbcp, true);
3752
3753 // do the call
3754 // rcx: receiver
3755 // rbx,: Method*
3756 __ jump_from_interpreted(rbx, rdx);
3757 __ should_not_reach_here();
3758
3759 // exception handling code follows...
3760 // note: must restore interpreter registers to canonical
3761 // state for exception handling to work correctly!
3762
3763 __ bind(no_such_method);
3764 // throw exception
3765 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3766 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3767 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3768 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3769 // the call_VM checks for exception, so we should never return here.
3770 __ should_not_reach_here();
3771
3772 __ bind(no_such_interface);
3773 // throw exception
3774 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3775 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3776 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3777 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3778 InterpreterRuntime::throw_IncompatibleClassChangeError));
3779 // the call_VM checks for exception, so we should never return here.
3780 __ should_not_reach_here();
3781 }
3782
3783 void TemplateTable::invokehandle(int byte_no) {
3784 transition(vtos, vtos);
3785 assert(byte_no == f1_byte, "use this argument");
3786 const Register rbx_method = rbx;
3787 const Register rax_mtype = rax;
3788 const Register rcx_recv = rcx;
3789 const Register rdx_flags = rdx;
3790
3791 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3792 __ verify_method_ptr(rbx_method);
3793 __ verify_oop(rcx_recv);
3794 __ null_check(rcx_recv);
3795
3796 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3797 // rbx: MH.invokeExact_MT method (from f2)
3798
3799 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3800
3801 // FIXME: profile the LambdaForm also
3802 __ profile_final_call(rax);
3803 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3804
3805 __ jump_from_interpreted(rbx_method, rdx);
3806 }
3807
3808 void TemplateTable::invokedynamic(int byte_no) {
3809 transition(vtos, vtos);
3810 assert(byte_no == f1_byte, "use this argument");
3811
3812 const Register rbx_method = rbx;
3813 const Register rax_callsite = rax;
3814
3815 prepare_invoke(byte_no, rbx_method, rax_callsite);
3816
3817 // rax: CallSite object (from cpool->resolved_references[f1])
3818 // rbx: MH.linkToCallSite method (from f2)
3819
3820 // Note: rax_callsite is already pushed by prepare_invoke
3821
3822 // %%% should make a type profile for any invokedynamic that takes a ref argument
3823 // profile this call
3824 __ profile_call(rbcp);
3825 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3826
3827 __ verify_oop(rax_callsite);
3828
3829 __ jump_from_interpreted(rbx_method, rdx);
3830 }
3831
3832 //-----------------------------------------------------------------------------
3833 // Allocation
3834
3835 void TemplateTable::_new() {
3836 transition(vtos, atos);
3837 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3838 Label slow_case;
3839 Label slow_case_no_pop;
3840 Label done;
3841 Label initialize_header;
3842 Label initialize_object; // including clearing the fields
3843 Label allocate_shared;
3844
3845 __ get_cpool_and_tags(rcx, rax);
3846
3847 // Make sure the class we're about to instantiate has been resolved.
3848 // This is done before loading InstanceKlass to be consistent with the order
3849 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3850 const int tags_offset = Array<u1>::base_offset_in_bytes();
3851 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3852 __ jcc(Assembler::notEqual, slow_case_no_pop);
3853
3854 // get InstanceKlass
3855 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool)));
3856 __ push(rcx); // save the contexts of klass for initializing the header
3857
3858 // make sure klass is initialized & doesn't have finalizer
3859 // make sure klass is fully initialized
3860 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3861 __ jcc(Assembler::notEqual, slow_case);
3862
3863 // get instance_size in InstanceKlass (scaled to a count of bytes)
3864 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3865 // test to see if it has a finalizer or is malformed in some way
3866 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3867 __ jcc(Assembler::notZero, slow_case);
3868
3869 //
3870 // Allocate the instance
3871 // 1) Try to allocate in the TLAB
3872 // 2) if fail and the object is large allocate in the shared Eden
3873 // 3) if the above fails (or is not applicable), go to a slow case
3874 // (creates a new TLAB, etc.)
3875
3876 const bool allow_shared_alloc =
3877 Universe::heap()->supports_inline_contig_alloc();
3878
3879 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
3880 #ifndef _LP64
3881 if (UseTLAB || allow_shared_alloc) {
3882 __ get_thread(thread);
3883 }
3884 #endif // _LP64
3885
3886 if (UseTLAB) {
3887 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3888 __ lea(rbx, Address(rax, rdx, Address::times_1));
3889 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3890 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3891 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3892 if (ZeroTLAB) {
3893 // the fields have been already cleared
3894 __ jmp(initialize_header);
3895 } else {
3896 // initialize both the header and fields
3897 __ jmp(initialize_object);
3898 }
3899 }
3900
3901 // Allocation in the shared Eden, if allowed.
3902 //
3903 // rdx: instance size in bytes
3904 if (allow_shared_alloc) {
3905 __ bind(allocate_shared);
3906
3907 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3908 ExternalAddress heap_end((address)Universe::heap()->end_addr());
3909
3910 Label retry;
3911 __ bind(retry);
3912 __ movptr(rax, heap_top);
3913 __ lea(rbx, Address(rax, rdx, Address::times_1));
3914 __ cmpptr(rbx, heap_end);
3915 __ jcc(Assembler::above, slow_case);
3916
3917 // Compare rax, with the top addr, and if still equal, store the new
3918 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3919 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3920 //
3921 // rax,: object begin
3922 // rbx,: object end
3923 // rdx: instance size in bytes
3924 __ locked_cmpxchgptr(rbx, heap_top);
3925
3926 // if someone beat us on the allocation, try again, otherwise continue
3927 __ jcc(Assembler::notEqual, retry);
3928
3929 __ incr_allocated_bytes(thread, rdx, 0);
3930 }
3931
3932 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3933 // The object is initialized before the header. If the object size is
3934 // zero, go directly to the header initialization.
3935 __ bind(initialize_object);
3936 __ decrement(rdx, sizeof(oopDesc));
3937 __ jcc(Assembler::zero, initialize_header);
3938
3939 // Initialize topmost object field, divide rdx by 8, check if odd and
3940 // test if zero.
3941 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3942 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3943
3944 // rdx must have been multiple of 8
3945 #ifdef ASSERT
3946 // make sure rdx was multiple of 8
3947 Label L;
3948 // Ignore partial flag stall after shrl() since it is debug VM
3949 __ jccb(Assembler::carryClear, L);
3950 __ stop("object size is not multiple of 2 - adjust this code");
3951 __ bind(L);
3952 // rdx must be > 0, no extra check needed here
3953 #endif
3954
3955 // initialize remaining object fields: rdx was a multiple of 8
3956 { Label loop;
3957 __ bind(loop);
3958 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3959 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3960 __ decrement(rdx);
3961 __ jcc(Assembler::notZero, loop);
3962 }
3963
3964 // initialize object header only.
3965 __ bind(initialize_header);
3966 if (UseBiasedLocking) {
3967 __ pop(rcx); // get saved klass back in the register.
3968 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
3969 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3970 } else {
3971 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3972 (intptr_t)markOopDesc::prototype()); // header
3973 __ pop(rcx); // get saved klass back in the register.
3974 }
3975 #ifdef _LP64
3976 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
3977 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
3978 #endif
3979 __ store_klass(rax, rcx); // klass
3980
3981 {
3982 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3983 // Trigger dtrace event for fastpath
3984 __ push(atos);
3985 __ call_VM_leaf(
3986 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3987 __ pop(atos);
3988 }
3989
3990 __ jmp(done);
3991 }
3992
3993 // slow case
3994 __ bind(slow_case);
3995 __ pop(rcx); // restore stack pointer to what it was when we came in.
3996 __ bind(slow_case_no_pop);
3997
3998 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
3999 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4000
4001 __ get_constant_pool(rarg1);
4002 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4003 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4004 __ verify_oop(rax);
4005
4006 // continue
4007 __ bind(done);
4008 }
4009
4010 void TemplateTable::newarray() {
4011 transition(itos, atos);
4012 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4013 __ load_unsigned_byte(rarg1, at_bcp(1));
4014 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4015 rarg1, rax);
4016 }
4017
4018 void TemplateTable::anewarray() {
4019 transition(itos, atos);
4020
4021 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4022 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4023
4024 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4025 __ get_constant_pool(rarg1);
4026 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4027 rarg1, rarg2, rax);
4028 }
4029
4030 void TemplateTable::arraylength() {
4031 transition(atos, itos);
4032 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4033 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4034 }
4035
4036 void TemplateTable::checkcast() {
4037 transition(atos, atos);
4038 Label done, is_null, ok_is_subtype, quicked, resolved;
4039 __ testptr(rax, rax); // object is in rax
4040 __ jcc(Assembler::zero, is_null);
4041
4042 // Get cpool & tags index
4043 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4044 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4045 // See if bytecode has already been quicked
4046 __ cmpb(Address(rdx, rbx,
4047 Address::times_1,
4048 Array<u1>::base_offset_in_bytes()),
4049 JVM_CONSTANT_Class);
4050 __ jcc(Assembler::equal, quicked);
4051 __ push(atos); // save receiver for result, and for GC
4052 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4053
4054 // vm_result_2 has metadata result
4055 #ifndef _LP64
4056 // borrow rdi from locals
4057 __ get_thread(rdi);
4058 __ get_vm_result_2(rax, rdi);
4059 __ restore_locals();
4060 #else
4061 __ get_vm_result_2(rax, r15_thread);
4062 #endif
4063
4064 __ pop_ptr(rdx); // restore receiver
4065 __ jmpb(resolved);
4066
4067 // Get superklass in rax and subklass in rbx
4068 __ bind(quicked);
4069 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4070 __ movptr(rax, Address(rcx, rbx,
4071 Address::times_ptr, sizeof(ConstantPool)));
4072
4073 __ bind(resolved);
4074 __ load_klass(rbx, rdx);
4075
4076 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4077 // Superklass in rax. Subklass in rbx.
4078 __ gen_subtype_check(rbx, ok_is_subtype);
4079
4080 // Come here on failure
4081 __ push_ptr(rdx);
4082 // object is at TOS
4083 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4084
4085 // Come here on success
4086 __ bind(ok_is_subtype);
4087 __ mov(rax, rdx); // Restore object in rdx
4088
4089 // Collect counts on whether this check-cast sees NULLs a lot or not.
4090 if (ProfileInterpreter) {
4091 __ jmp(done);
4092 __ bind(is_null);
4093 __ profile_null_seen(rcx);
4094 } else {
4095 __ bind(is_null); // same as 'done'
4096 }
4097 __ bind(done);
4098 }
4099
4100 void TemplateTable::instanceof() {
4101 transition(atos, itos);
4102 Label done, is_null, ok_is_subtype, quicked, resolved;
4103 __ testptr(rax, rax);
4104 __ jcc(Assembler::zero, is_null);
4105
4106 // Get cpool & tags index
4107 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4108 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4109 // See if bytecode has already been quicked
4110 __ cmpb(Address(rdx, rbx,
4111 Address::times_1,
4112 Array<u1>::base_offset_in_bytes()),
4113 JVM_CONSTANT_Class);
4114 __ jcc(Assembler::equal, quicked);
4115
4116 __ push(atos); // save receiver for result, and for GC
4117 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4118 // vm_result_2 has metadata result
4119
4120 #ifndef _LP64
4121 // borrow rdi from locals
4122 __ get_thread(rdi);
4123 __ get_vm_result_2(rax, rdi);
4124 __ restore_locals();
4125 #else
4126 __ get_vm_result_2(rax, r15_thread);
4127 #endif
4128
4129 __ pop_ptr(rdx); // restore receiver
4130 __ verify_oop(rdx);
4131 __ load_klass(rdx, rdx);
4132 __ jmpb(resolved);
4133
4134 // Get superklass in rax and subklass in rdx
4135 __ bind(quicked);
4136 __ load_klass(rdx, rax);
4137 __ movptr(rax, Address(rcx, rbx,
4138 Address::times_ptr, sizeof(ConstantPool)));
4139
4140 __ bind(resolved);
4141
4142 // Generate subtype check. Blows rcx, rdi
4143 // Superklass in rax. Subklass in rdx.
4144 __ gen_subtype_check(rdx, ok_is_subtype);
4145
4146 // Come here on failure
4147 __ xorl(rax, rax);
4148 __ jmpb(done);
4149 // Come here on success
4150 __ bind(ok_is_subtype);
4151 __ movl(rax, 1);
4152
4153 // Collect counts on whether this test sees NULLs a lot or not.
4154 if (ProfileInterpreter) {
4155 __ jmp(done);
4156 __ bind(is_null);
4157 __ profile_null_seen(rcx);
4158 } else {
4159 __ bind(is_null); // same as 'done'
4160 }
4161 __ bind(done);
4162 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4163 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4164 }
4165
4166
4167 //----------------------------------------------------------------------------------------------------
4168 // Breakpoints
4169 void TemplateTable::_breakpoint() {
4170 // Note: We get here even if we are single stepping..
4171 // jbug insists on setting breakpoints at every bytecode
4172 // even if we are in single step mode.
4173
4174 transition(vtos, vtos);
4175
4176 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4177
4178 // get the unpatched byte code
4179 __ get_method(rarg);
4180 __ call_VM(noreg,
4181 CAST_FROM_FN_PTR(address,
4182 InterpreterRuntime::get_original_bytecode_at),
4183 rarg, rbcp);
4184 __ mov(rbx, rax); // why?
4185
4186 // post the breakpoint event
4187 __ get_method(rarg);
4188 __ call_VM(noreg,
4189 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4190 rarg, rbcp);
4191
4192 // complete the execution of original bytecode
4193 __ dispatch_only_normal(vtos);
4194 }
4195
4196 //-----------------------------------------------------------------------------
4197 // Exceptions
4198
4199 void TemplateTable::athrow() {
4200 transition(atos, vtos);
4201 __ null_check(rax);
4202 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4203 }
4204
4205 //-----------------------------------------------------------------------------
4206 // Synchronization
4207 //
4208 // Note: monitorenter & exit are symmetric routines; which is reflected
4209 // in the assembly code structure as well
4210 //
4211 // Stack layout:
4212 //
4213 // [expressions ] <--- rsp = expression stack top
4214 // ..
4215 // [expressions ]
4216 // [monitor entry] <--- monitor block top = expression stack bot
4217 // ..
4218 // [monitor entry]
4219 // [frame data ] <--- monitor block bot
4220 // ...
4221 // [saved rbp ] <--- rbp
4222 void TemplateTable::monitorenter() {
4223 transition(atos, vtos);
4224
4225 // check for NULL object
4226 __ null_check(rax);
4227
4228 const Address monitor_block_top(
4229 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4230 const Address monitor_block_bot(
4231 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4232 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4233
4234 Label allocated;
4235
4236 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4237 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4238 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4239
4240 // initialize entry pointer
4241 __ xorl(rmon, rmon); // points to free slot or NULL
4242
4243 // find a free slot in the monitor block (result in rmon)
4244 {
4245 Label entry, loop, exit;
4246 __ movptr(rtop, monitor_block_top); // points to current entry,
4247 // starting with top-most entry
4248 __ lea(rbot, monitor_block_bot); // points to word before bottom
4249 // of monitor block
4250 __ jmpb(entry);
4251
4252 __ bind(loop);
4253 // check if current entry is used
4254 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4255 // if not used then remember entry in rmon
4256 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4257 // check if current entry is for same object
4258 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4259 // if same object then stop searching
4260 __ jccb(Assembler::equal, exit);
4261 // otherwise advance to next entry
4262 __ addptr(rtop, entry_size);
4263 __ bind(entry);
4264 // check if bottom reached
4265 __ cmpptr(rtop, rbot);
4266 // if not at bottom then check this entry
4267 __ jcc(Assembler::notEqual, loop);
4268 __ bind(exit);
4269 }
4270
4271 __ testptr(rmon, rmon); // check if a slot has been found
4272 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4273
4274 // allocate one if there's no free slot
4275 {
4276 Label entry, loop;
4277 // 1. compute new pointers // rsp: old expression stack top
4278 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4279 __ subptr(rsp, entry_size); // move expression stack top
4280 __ subptr(rmon, entry_size); // move expression stack bottom
4281 __ mov(rtop, rsp); // set start value for copy loop
4282 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4283 __ jmp(entry);
4284 // 2. move expression stack contents
4285 __ bind(loop);
4286 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4287 // word from old location
4288 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4289 __ addptr(rtop, wordSize); // advance to next word
4290 __ bind(entry);
4291 __ cmpptr(rtop, rmon); // check if bottom reached
4292 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4293 // copy next word
4294 }
4295
4296 // call run-time routine
4297 // rmon: points to monitor entry
4298 __ bind(allocated);
4299
4300 // Increment bcp to point to the next bytecode, so exception
4301 // handling for async. exceptions work correctly.
4302 // The object has already been poped from the stack, so the
4303 // expression stack looks correct.
4304 __ increment(rbcp);
4305
4306 // store object
4307 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4308 __ lock_object(rmon);
4309
4310 // check to make sure this monitor doesn't cause stack overflow after locking
4311 __ save_bcp(); // in case of exception
4312 __ generate_stack_overflow_check(0);
4313
4314 // The bcp has already been incremented. Just need to dispatch to
4315 // next instruction.
4316 __ dispatch_next(vtos);
4317 }
4318
4319 void TemplateTable::monitorexit() {
4320 transition(atos, vtos);
4321
4322 // check for NULL object
4323 __ null_check(rax);
4324
4325 const Address monitor_block_top(
4326 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4327 const Address monitor_block_bot(
4328 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4329 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4330
4331 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4332 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4333
4334 Label found;
4335
4336 // find matching slot
4337 {
4338 Label entry, loop;
4339 __ movptr(rtop, monitor_block_top); // points to current entry,
4340 // starting with top-most entry
4341 __ lea(rbot, monitor_block_bot); // points to word before bottom
4342 // of monitor block
4343 __ jmpb(entry);
4344
4345 __ bind(loop);
4346 // check if current entry is for same object
4347 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4348 // if same object then stop searching
4349 __ jcc(Assembler::equal, found);
4350 // otherwise advance to next entry
4351 __ addptr(rtop, entry_size);
4352 __ bind(entry);
4353 // check if bottom reached
4354 __ cmpptr(rtop, rbot);
4355 // if not at bottom then check this entry
4356 __ jcc(Assembler::notEqual, loop);
4357 }
4358
4359 // error handling. Unlocking was not block-structured
4360 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4361 InterpreterRuntime::throw_illegal_monitor_state_exception));
4362 __ should_not_reach_here();
4363
4364 // call run-time routine
4365 __ bind(found);
4366 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4367 __ unlock_object(rtop);
4368 __ pop_ptr(rax); // discard object
4369 }
4370
4371 // Wide instructions
4372 void TemplateTable::wide() {
4373 transition(vtos, vtos);
4374 __ load_unsigned_byte(rbx, at_bcp(1));
4375 ExternalAddress wtable((address)Interpreter::_wentry_point);
4376 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4377 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4378 }
4379
4380 // Multi arrays
4381 void TemplateTable::multianewarray() {
4382 transition(vtos, atos);
4383
4384 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4385 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4386 // last dim is on top of stack; we want address of first one:
4387 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4388 // the latter wordSize to point to the beginning of the array.
4389 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4390 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4391 __ load_unsigned_byte(rbx, at_bcp(3));
4392 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4393 }