1 /*
2 * Copyright (c) 2013, Red Hat Inc.
3 * Copyright (c) 2003, 2011, Oracle and/or its affiliates.
4 * All rights reserved.
5 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 *
7 * This code is free software; you can redistribute it and/or modify it
8 * under the terms of the GNU General Public License version 2 only, as
9 * published by the Free Software Foundation.
10 *
11 * This code is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 * version 2 for more details (a copy is included in the LICENSE file that
15 * accompanied this code).
16 *
17 * You should have received a copy of the GNU General Public License version
18 * 2 along with this work; if not, write to the Free Software Foundation,
19 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
20 *
21 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
22 * or visit www.oracle.com if you need additional information or have any
23 * questions.
24 *
25 */
26
27 #include "precompiled.hpp"
28 #include "asm/macroAssembler.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "memory/universe.inline.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/methodHandles.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/stubRoutines.hpp"
40 #include "runtime/synchronizer.hpp"
41 #if INCLUDE_ALL_GCS
42 #include "shenandoahBarrierSetAssembler_aarch64.hpp"
43 #endif
44
45 #ifndef CC_INTERP
46
47 #define __ _masm->
48
49 // Platform-dependent initialization
50
51 void TemplateTable::pd_initialize() {
52 // No amd64 specific initialization
53 }
54
55 // Address computation: local variables
56
57 static inline Address iaddress(int n) {
58 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
59 }
60
61 static inline Address laddress(int n) {
62 return iaddress(n + 1);
63 }
64
65 static inline Address faddress(int n) {
66 return iaddress(n);
67 }
68
69 static inline Address daddress(int n) {
70 return laddress(n);
71 }
72
73 static inline Address aaddress(int n) {
74 return iaddress(n);
75 }
76
77 static inline Address iaddress(Register r) {
78 return Address(rlocals, r, Address::lsl(3));
79 }
80
81 static inline Address laddress(Register r, Register scratch,
82 InterpreterMacroAssembler* _masm) {
83 __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
84 return Address(scratch, Interpreter::local_offset_in_bytes(1));
85 }
86
87 static inline Address faddress(Register r) {
88 return iaddress(r);
89 }
90
91 static inline Address daddress(Register r, Register scratch,
92 InterpreterMacroAssembler* _masm) {
93 return laddress(r, scratch, _masm);
94 }
95
96 static inline Address aaddress(Register r) {
97 return iaddress(r);
98 }
99
100 static inline Address at_rsp() {
101 return Address(esp, 0);
102 }
103
104 // At top of Java expression stack which may be different than esp(). It
105 // isn't for category 1 objects.
106 static inline Address at_tos () {
107 return Address(esp, Interpreter::expr_offset_in_bytes(0));
108 }
109
110 static inline Address at_tos_p1() {
111 return Address(esp, Interpreter::expr_offset_in_bytes(1));
112 }
113
114 static inline Address at_tos_p2() {
115 return Address(esp, Interpreter::expr_offset_in_bytes(2));
116 }
117
118 static inline Address at_tos_p3() {
119 return Address(esp, Interpreter::expr_offset_in_bytes(3));
120 }
121
122 static inline Address at_tos_p4() {
123 return Address(esp, Interpreter::expr_offset_in_bytes(4));
124 }
125
126 static inline Address at_tos_p5() {
127 return Address(esp, Interpreter::expr_offset_in_bytes(5));
128 }
129
130 // Condition conversion
131 static Assembler::Condition j_not(TemplateTable::Condition cc) {
132 switch (cc) {
133 case TemplateTable::equal : return Assembler::NE;
134 case TemplateTable::not_equal : return Assembler::EQ;
135 case TemplateTable::less : return Assembler::GE;
136 case TemplateTable::less_equal : return Assembler::GT;
137 case TemplateTable::greater : return Assembler::LE;
138 case TemplateTable::greater_equal: return Assembler::LT;
139 }
140 ShouldNotReachHere();
141 return Assembler::EQ;
142 }
143
144
145 // Miscelaneous helper routines
146 // Store an oop (or NULL) at the Address described by obj.
147 // If val == noreg this means store a NULL
148 static void do_oop_store(InterpreterMacroAssembler* _masm,
149 Address obj,
150 Register val,
151 BarrierSet::Name barrier,
152 bool precise) {
153 assert(val == noreg || val == r0, "parameter is just for looks");
154 switch (barrier) {
155 #if INCLUDE_ALL_GCS
156 case BarrierSet::G1SATBCT:
157 case BarrierSet::G1SATBCTLogging:
158 {
159 // flatten object address if needed
160 if (obj.index() == noreg && obj.offset() == 0) {
161 if (obj.base() != r3) {
162 __ mov(r3, obj.base());
163 }
164 } else {
165 __ lea(r3, obj);
166 }
167 __ g1_write_barrier_pre(r3 /* obj */,
168 r1 /* pre_val */,
169 rthread /* thread */,
170 r10 /* tmp */,
171 val != noreg /* tosca_live */,
172 false /* expand_call */);
173 if (val == noreg) {
174 __ store_heap_oop_null(Address(r3, 0));
175 } else {
176 // G1 barrier needs uncompressed oop for region cross check.
177 Register new_val = val;
178 if (UseCompressedOops) {
179 new_val = rscratch2;
180 __ mov(new_val, val);
181 }
182 __ store_heap_oop(Address(r3, 0), val);
183 __ g1_write_barrier_post(r3 /* store_adr */,
184 new_val /* new_val */,
185 rthread /* thread */,
186 r10 /* tmp */,
187 r1 /* tmp2 */);
188 }
189
190 }
191 break;
192 case BarrierSet::ShenandoahBarrierSet:
193 {
194 // flatten object address if needed
195 if (obj.index() == noreg && obj.offset() == 0) {
196 if (obj.base() != r3) {
197 __ mov(r3, obj.base());
198 }
199 } else {
200 __ lea(r3, obj);
201 }
202 if (ShenandoahSATBBarrier) {
203 __ g1_write_barrier_pre(r3 /* obj */,
204 r1 /* pre_val */,
205 rthread /* thread */,
206 r10 /* tmp */,
207 val != noreg /* tosca_live */,
208 false /* expand_call */);
209 }
210 if (val == noreg) {
211 __ store_heap_oop_null(Address(r3, 0));
212 } else {
213 if (ShenandoahStoreValEnqueueBarrier) {
214 ShenandoahBarrierSetAssembler::bsasm()->storeval_barrier(_masm, val, r10);
215 }
216 __ store_heap_oop(Address(r3, 0), val);
217 }
218
219 }
220 break;
221 #endif // INCLUDE_ALL_GCS
222 case BarrierSet::CardTableModRef:
223 case BarrierSet::CardTableExtension:
224 {
225 if (val == noreg) {
226 __ store_heap_oop_null(obj);
227 } else {
228 __ store_heap_oop(obj, val);
229 // flatten object address if needed
230 if (!precise || (obj.index() == noreg && obj.offset() == 0)) {
231 __ store_check(obj.base());
232 } else {
233 __ lea(r3, obj);
234 __ store_check(r3);
235 }
236 }
237 }
238 break;
239 case BarrierSet::ModRef:
240 case BarrierSet::Other:
241 if (val == noreg) {
242 __ store_heap_oop_null(obj);
243 } else {
244 __ store_heap_oop(obj, val);
245 }
246 break;
247 default :
248 ShouldNotReachHere();
249
250 }
251 }
252
253 Address TemplateTable::at_bcp(int offset) {
254 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
255 return Address(rbcp, offset);
256 }
257
258 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
259 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
260 int byte_no)
261 {
262 if (!RewriteBytecodes) return;
263 Label L_patch_done;
264
265 switch (bc) {
266 case Bytecodes::_fast_aputfield:
267 case Bytecodes::_fast_bputfield:
268 case Bytecodes::_fast_zputfield:
269 case Bytecodes::_fast_cputfield:
270 case Bytecodes::_fast_dputfield:
271 case Bytecodes::_fast_fputfield:
272 case Bytecodes::_fast_iputfield:
273 case Bytecodes::_fast_lputfield:
274 case Bytecodes::_fast_sputfield:
275 {
276 // We skip bytecode quickening for putfield instructions when
277 // the put_code written to the constant pool cache is zero.
278 // This is required so that every execution of this instruction
279 // calls out to InterpreterRuntime::resolve_get_put to do
280 // additional, required work.
281 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
282 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
283 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
284 __ movw(bc_reg, bc);
285 __ cmpw(temp_reg, (unsigned) 0);
286 __ br(Assembler::EQ, L_patch_done); // don't patch
287 }
288 break;
289 default:
290 assert(byte_no == -1, "sanity");
291 // the pair bytecodes have already done the load.
292 if (load_bc_into_bc_reg) {
293 __ movw(bc_reg, bc);
294 }
295 }
296
297 if (JvmtiExport::can_post_breakpoint()) {
298 Label L_fast_patch;
299 // if a breakpoint is present we can't rewrite the stream directly
300 __ load_unsigned_byte(temp_reg, at_bcp(0));
301 __ cmpw(temp_reg, Bytecodes::_breakpoint);
302 __ br(Assembler::NE, L_fast_patch);
303 // Let breakpoint table handling rewrite to quicker bytecode
304 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
305 __ b(L_patch_done);
306 __ bind(L_fast_patch);
307 }
308
309 #ifdef ASSERT
310 Label L_okay;
311 __ load_unsigned_byte(temp_reg, at_bcp(0));
312 __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
313 __ br(Assembler::EQ, L_okay);
314 __ cmpw(temp_reg, bc_reg);
315 __ br(Assembler::EQ, L_okay);
316 __ stop("patching the wrong bytecode");
317 __ bind(L_okay);
318 #endif
319
320 // patch bytecode
321 __ strb(bc_reg, at_bcp(0));
322 __ bind(L_patch_done);
323 }
324
325
326 // Individual instructions
327
328 void TemplateTable::nop() {
329 transition(vtos, vtos);
330 // nothing to do
331 }
332
333 void TemplateTable::shouldnotreachhere() {
334 transition(vtos, vtos);
335 __ stop("shouldnotreachhere bytecode");
336 }
337
338 void TemplateTable::aconst_null()
339 {
340 transition(vtos, atos);
341 __ mov(r0, 0);
342 }
343
344 void TemplateTable::iconst(int value)
345 {
346 transition(vtos, itos);
347 __ mov(r0, value);
348 }
349
350 void TemplateTable::lconst(int value)
351 {
352 __ mov(r0, value);
353 }
354
355 void TemplateTable::fconst(int value)
356 {
357 transition(vtos, ftos);
358 switch (value) {
359 case 0:
360 __ fmovs(v0, zr);
361 break;
362 case 1:
363 __ fmovs(v0, 1.0);
364 break;
365 case 2:
366 __ fmovs(v0, 2.0);
367 break;
368 default:
369 ShouldNotReachHere();
370 break;
371 }
372 }
373
374 void TemplateTable::dconst(int value)
375 {
376 transition(vtos, dtos);
377 switch (value) {
378 case 0:
379 __ fmovd(v0, zr);
380 break;
381 case 1:
382 __ fmovd(v0, 1.0);
383 break;
384 case 2:
385 __ fmovd(v0, 2.0);
386 break;
387 default:
388 ShouldNotReachHere();
389 break;
390 }
391 }
392
393 void TemplateTable::bipush()
394 {
395 transition(vtos, itos);
396 __ load_signed_byte32(r0, at_bcp(1));
397 }
398
399 void TemplateTable::sipush()
400 {
401 transition(vtos, itos);
402 __ load_unsigned_short(r0, at_bcp(1));
403 __ revw(r0, r0);
404 __ asrw(r0, r0, 16);
405 }
406
407 void TemplateTable::ldc(bool wide)
408 {
409 transition(vtos, vtos);
410 Label call_ldc, notFloat, notClass, Done;
411
412 if (wide) {
413 __ get_unsigned_2_byte_index_at_bcp(r1, 1);
414 } else {
415 __ load_unsigned_byte(r1, at_bcp(1));
416 }
417 __ get_cpool_and_tags(r2, r0);
418
419 const int base_offset = ConstantPool::header_size() * wordSize;
420 const int tags_offset = Array<u1>::base_offset_in_bytes();
421
422 // get type
423 __ add(r3, r1, tags_offset);
424 __ lea(r3, Address(r0, r3));
425 __ ldarb(r3, r3);
426
427 // unresolved class - get the resolved class
428 __ cmp(r3, JVM_CONSTANT_UnresolvedClass);
429 __ br(Assembler::EQ, call_ldc);
430
431 // unresolved class in error state - call into runtime to throw the error
432 // from the first resolution attempt
433 __ cmp(r3, JVM_CONSTANT_UnresolvedClassInError);
434 __ br(Assembler::EQ, call_ldc);
435
436 // resolved class - need to call vm to get java mirror of the class
437 __ cmp(r3, JVM_CONSTANT_Class);
438 __ br(Assembler::NE, notClass);
439
440 __ bind(call_ldc);
441 __ mov(c_rarg1, wide);
442 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
443 __ push_ptr(r0);
444 __ verify_oop(r0);
445 __ b(Done);
446
447 __ bind(notClass);
448 __ cmp(r3, JVM_CONSTANT_Float);
449 __ br(Assembler::NE, notFloat);
450 // ftos
451 __ adds(r1, r2, r1, Assembler::LSL, 3);
452 __ ldrs(v0, Address(r1, base_offset));
453 __ push_f();
454 __ b(Done);
455
456 __ bind(notFloat);
457 #ifdef ASSERT
458 {
459 Label L;
460 __ cmp(r3, JVM_CONSTANT_Integer);
461 __ br(Assembler::EQ, L);
462 // String and Object are rewritten to fast_aldc
463 __ stop("unexpected tag type in ldc");
464 __ bind(L);
465 }
466 #endif
467 // itos JVM_CONSTANT_Integer only
468 __ adds(r1, r2, r1, Assembler::LSL, 3);
469 __ ldrw(r0, Address(r1, base_offset));
470 __ push_i(r0);
471 __ bind(Done);
472 }
473
474 // Fast path for caching oop constants.
475 void TemplateTable::fast_aldc(bool wide)
476 {
477 transition(vtos, atos);
478
479 Register result = r0;
480 Register tmp = r1;
481 int index_size = wide ? sizeof(u2) : sizeof(u1);
482
483 Label resolved;
484
485 // We are resolved if the resolved reference cache entry contains a
486 // non-null object (String, MethodType, etc.)
487 assert_different_registers(result, tmp);
488 __ get_cache_index_at_bcp(tmp, 1, index_size);
489 __ load_resolved_reference_at_index(result, tmp);
490 __ cbnz(result, resolved);
491
492 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
493
494 // first time invocation - must resolve first
495 __ mov(tmp, (int)bytecode());
496 __ call_VM(result, entry, tmp);
497
498 __ bind(resolved);
499
500 if (VerifyOops) {
501 __ verify_oop(result);
502 }
503 }
504
505 void TemplateTable::ldc2_w()
506 {
507 transition(vtos, vtos);
508 Label Long, Done;
509 __ get_unsigned_2_byte_index_at_bcp(r0, 1);
510
511 __ get_cpool_and_tags(r1, r2);
512 const int base_offset = ConstantPool::header_size() * wordSize;
513 const int tags_offset = Array<u1>::base_offset_in_bytes();
514
515 // get type
516 __ lea(r2, Address(r2, r0, Address::lsl(0)));
517 __ load_unsigned_byte(r2, Address(r2, tags_offset));
518 __ cmpw(r2, (int)JVM_CONSTANT_Double);
519 __ br(Assembler::NE, Long);
520 // dtos
521 __ lea (r2, Address(r1, r0, Address::lsl(3)));
522 __ ldrd(v0, Address(r2, base_offset));
523 __ push_d();
524 __ b(Done);
525
526 __ bind(Long);
527 // ltos
528 __ lea(r0, Address(r1, r0, Address::lsl(3)));
529 __ ldr(r0, Address(r0, base_offset));
530 __ push_l();
531
532 __ bind(Done);
533 }
534
535 void TemplateTable::locals_index(Register reg, int offset)
536 {
537 __ ldrb(reg, at_bcp(offset));
538 __ neg(reg, reg);
539 }
540
541 void TemplateTable::iload()
542 {
543 transition(vtos, itos);
544 if (RewriteFrequentPairs) {
545 Label rewrite, done;
546 Register bc = r4;
547
548 // get next bytecode
549 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
550
551 // if _iload, wait to rewrite to iload2. We only want to rewrite the
552 // last two iloads in a pair. Comparing against fast_iload means that
553 // the next bytecode is neither an iload or a caload, and therefore
554 // an iload pair.
555 __ cmpw(r1, Bytecodes::_iload);
556 __ br(Assembler::EQ, done);
557
558 // if _fast_iload rewrite to _fast_iload2
559 __ cmpw(r1, Bytecodes::_fast_iload);
560 __ movw(bc, Bytecodes::_fast_iload2);
561 __ br(Assembler::EQ, rewrite);
562
563 // if _caload rewrite to _fast_icaload
564 __ cmpw(r1, Bytecodes::_caload);
565 __ movw(bc, Bytecodes::_fast_icaload);
566 __ br(Assembler::EQ, rewrite);
567
568 // else rewrite to _fast_iload
569 __ movw(bc, Bytecodes::_fast_iload);
570
571 // rewrite
572 // bc: new bytecode
573 __ bind(rewrite);
574 patch_bytecode(Bytecodes::_iload, bc, r1, false);
575 __ bind(done);
576
577 }
578
579 // do iload, get the local value into tos
580 locals_index(r1);
581 __ ldr(r0, iaddress(r1));
582
583 }
584
585 void TemplateTable::fast_iload2()
586 {
587 transition(vtos, itos);
588 locals_index(r1);
589 __ ldr(r0, iaddress(r1));
590 __ push(itos);
591 locals_index(r1, 3);
592 __ ldr(r0, iaddress(r1));
593 }
594
595 void TemplateTable::fast_iload()
596 {
597 transition(vtos, itos);
598 locals_index(r1);
599 __ ldr(r0, iaddress(r1));
600 }
601
602 void TemplateTable::lload()
603 {
604 transition(vtos, ltos);
605 __ ldrb(r1, at_bcp(1));
606 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
607 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
608 }
609
610 void TemplateTable::fload()
611 {
612 transition(vtos, ftos);
613 locals_index(r1);
614 // n.b. we use ldrd here because this is a 64 bit slot
615 // this is comparable to the iload case
616 __ ldrd(v0, faddress(r1));
617 }
618
619 void TemplateTable::dload()
620 {
621 transition(vtos, dtos);
622 __ ldrb(r1, at_bcp(1));
623 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
624 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
625 }
626
627 void TemplateTable::aload()
628 {
629 transition(vtos, atos);
630 locals_index(r1);
631 __ ldr(r0, iaddress(r1));
632 }
633
634 void TemplateTable::locals_index_wide(Register reg) {
635 __ ldrh(reg, at_bcp(2));
636 __ rev16w(reg, reg);
637 __ neg(reg, reg);
638 }
639
640 void TemplateTable::wide_iload() {
641 transition(vtos, itos);
642 locals_index_wide(r1);
643 __ ldr(r0, iaddress(r1));
644 }
645
646 void TemplateTable::wide_lload()
647 {
648 transition(vtos, ltos);
649 __ ldrh(r1, at_bcp(2));
650 __ rev16w(r1, r1);
651 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
652 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
653 }
654
655 void TemplateTable::wide_fload()
656 {
657 transition(vtos, ftos);
658 locals_index_wide(r1);
659 // n.b. we use ldrd here because this is a 64 bit slot
660 // this is comparable to the iload case
661 __ ldrd(v0, faddress(r1));
662 }
663
664 void TemplateTable::wide_dload()
665 {
666 transition(vtos, dtos);
667 __ ldrh(r1, at_bcp(2));
668 __ rev16w(r1, r1);
669 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
670 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
671 }
672
673 void TemplateTable::wide_aload()
674 {
675 transition(vtos, atos);
676 locals_index_wide(r1);
677 __ ldr(r0, aaddress(r1));
678 }
679
680 void TemplateTable::index_check(Register array, Register index)
681 {
682 // destroys r1, rscratch1
683 // check array
684 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
685 // sign extend index for use by indexed load
686 // __ movl2ptr(index, index);
687 // check index
688 Register length = rscratch1;
689 __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
690 __ cmpw(index, length);
691 if (index != r1) {
692 // ??? convention: move aberrant index into r1 for exception message
693 assert(r1 != array, "different registers");
694 __ mov(r1, index);
695 }
696 Label ok;
697 __ br(Assembler::LO, ok);
698 __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
699 __ br(rscratch1);
700 __ bind(ok);
701 }
702
703 void TemplateTable::iaload()
704 {
705 transition(itos, itos);
706 __ mov(r1, r0);
707 __ pop_ptr(r0);
708 // r0: array
709 // r1: index
710 index_check(r0, r1); // leaves index in r1, kills rscratch1
711 __ lea(r1, Address(r0, r1, Address::uxtw(2)));
712 __ ldrw(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_INT)));
713 }
714
715 void TemplateTable::laload()
716 {
717 transition(itos, ltos);
718 __ mov(r1, r0);
719 __ pop_ptr(r0);
720 // r0: array
721 // r1: index
722 index_check(r0, r1); // leaves index in r1, kills rscratch1
723 __ lea(r1, Address(r0, r1, Address::uxtw(3)));
724 __ ldr(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_LONG)));
725 }
726
727 void TemplateTable::faload()
728 {
729 transition(itos, ftos);
730 __ mov(r1, r0);
731 __ pop_ptr(r0);
732 // r0: array
733 // r1: index
734 index_check(r0, r1); // leaves index in r1, kills rscratch1
735 __ lea(r1, Address(r0, r1, Address::uxtw(2)));
736 __ ldrs(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
737 }
738
739 void TemplateTable::daload()
740 {
741 transition(itos, dtos);
742 __ mov(r1, r0);
743 __ pop_ptr(r0);
744 // r0: array
745 // r1: index
746 index_check(r0, r1); // leaves index in r1, kills rscratch1
747 __ lea(r1, Address(r0, r1, Address::uxtw(3)));
748 __ ldrd(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
749 }
750
751 void TemplateTable::aaload()
752 {
753 transition(itos, atos);
754 __ mov(r1, r0);
755 __ pop_ptr(r0);
756 // r0: array
757 // r1: index
758 index_check(r0, r1); // leaves index in r1, kills rscratch1
759 int s = (UseCompressedOops ? 2 : 3);
760 __ lea(r1, Address(r0, r1, Address::uxtw(s)));
761 __ load_heap_oop(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
762 }
763
764 void TemplateTable::baload()
765 {
766 transition(itos, itos);
767 __ mov(r1, r0);
768 __ pop_ptr(r0);
769 // r0: array
770 // r1: index
771 index_check(r0, r1); // leaves index in r1, kills rscratch1
772 __ lea(r1, Address(r0, r1, Address::uxtw(0)));
773 __ load_signed_byte(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
774 }
775
776 void TemplateTable::caload()
777 {
778 transition(itos, itos);
779 __ mov(r1, r0);
780 __ pop_ptr(r0);
781 // r0: array
782 // r1: index
783 index_check(r0, r1); // leaves index in r1, kills rscratch1
784 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
785 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
786 }
787
788 // iload followed by caload frequent pair
789 void TemplateTable::fast_icaload()
790 {
791 transition(vtos, itos);
792 // load index out of locals
793 locals_index(r2);
794 __ ldr(r1, iaddress(r2));
795
796 __ pop_ptr(r0);
797
798 // r0: array
799 // r1: index
800 index_check(r0, r1); // leaves index in r1, kills rscratch1
801 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
802 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
803 }
804
805 void TemplateTable::saload()
806 {
807 transition(itos, itos);
808 __ mov(r1, r0);
809 __ pop_ptr(r0);
810 // r0: array
811 // r1: index
812 index_check(r0, r1); // leaves index in r1, kills rscratch1
813 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
814 __ load_signed_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
815 }
816
817 void TemplateTable::iload(int n)
818 {
819 transition(vtos, itos);
820 __ ldr(r0, iaddress(n));
821 }
822
823 void TemplateTable::lload(int n)
824 {
825 transition(vtos, ltos);
826 __ ldr(r0, laddress(n));
827 }
828
829 void TemplateTable::fload(int n)
830 {
831 transition(vtos, ftos);
832 __ ldrs(v0, faddress(n));
833 }
834
835 void TemplateTable::dload(int n)
836 {
837 transition(vtos, dtos);
838 __ ldrd(v0, daddress(n));
839 }
840
841 void TemplateTable::aload(int n)
842 {
843 transition(vtos, atos);
844 __ ldr(r0, iaddress(n));
845 }
846
847 void TemplateTable::aload_0()
848 {
849 // According to bytecode histograms, the pairs:
850 //
851 // _aload_0, _fast_igetfield
852 // _aload_0, _fast_agetfield
853 // _aload_0, _fast_fgetfield
854 //
855 // occur frequently. If RewriteFrequentPairs is set, the (slow)
856 // _aload_0 bytecode checks if the next bytecode is either
857 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
858 // rewrites the current bytecode into a pair bytecode; otherwise it
859 // rewrites the current bytecode into _fast_aload_0 that doesn't do
860 // the pair check anymore.
861 //
862 // Note: If the next bytecode is _getfield, the rewrite must be
863 // delayed, otherwise we may miss an opportunity for a pair.
864 //
865 // Also rewrite frequent pairs
866 // aload_0, aload_1
867 // aload_0, iload_1
868 // These bytecodes with a small amount of code are most profitable
869 // to rewrite
870 if (RewriteFrequentPairs) {
871 Label rewrite, done;
872 const Register bc = r4;
873
874 // get next bytecode
875 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
876
877 // do actual aload_0
878 aload(0);
879
880 // if _getfield then wait with rewrite
881 __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
882 __ br(Assembler::EQ, done);
883
884 // if _igetfield then reqrite to _fast_iaccess_0
885 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
886 __ cmpw(r1, Bytecodes::_fast_igetfield);
887 __ movw(bc, Bytecodes::_fast_iaccess_0);
888 __ br(Assembler::EQ, rewrite);
889
890 // if _agetfield then reqrite to _fast_aaccess_0
891 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
892 __ cmpw(r1, Bytecodes::_fast_agetfield);
893 __ movw(bc, Bytecodes::_fast_aaccess_0);
894 __ br(Assembler::EQ, rewrite);
895
896 // if _fgetfield then reqrite to _fast_faccess_0
897 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
898 __ cmpw(r1, Bytecodes::_fast_fgetfield);
899 __ movw(bc, Bytecodes::_fast_faccess_0);
900 __ br(Assembler::EQ, rewrite);
901
902 // else rewrite to _fast_aload0
903 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
904 __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
905
906 // rewrite
907 // bc: new bytecode
908 __ bind(rewrite);
909 patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
910
911 __ bind(done);
912 } else {
913 aload(0);
914 }
915 }
916
917 void TemplateTable::istore()
918 {
919 transition(itos, vtos);
920 locals_index(r1);
921 // FIXME: We're being very pernickerty here storing a jint in a
922 // local with strw, which costs an extra instruction over what we'd
923 // be able to do with a simple str. We should just store the whole
924 // word.
925 __ lea(rscratch1, iaddress(r1));
926 __ strw(r0, Address(rscratch1));
927 }
928
929 void TemplateTable::lstore()
930 {
931 transition(ltos, vtos);
932 locals_index(r1);
933 __ str(r0, laddress(r1, rscratch1, _masm));
934 }
935
936 void TemplateTable::fstore() {
937 transition(ftos, vtos);
938 locals_index(r1);
939 __ lea(rscratch1, iaddress(r1));
940 __ strs(v0, Address(rscratch1));
941 }
942
943 void TemplateTable::dstore() {
944 transition(dtos, vtos);
945 locals_index(r1);
946 __ strd(v0, daddress(r1, rscratch1, _masm));
947 }
948
949 void TemplateTable::astore()
950 {
951 transition(vtos, vtos);
952 __ pop_ptr(r0);
953 locals_index(r1);
954 __ str(r0, aaddress(r1));
955 }
956
957 void TemplateTable::wide_istore() {
958 transition(vtos, vtos);
959 __ pop_i();
960 locals_index_wide(r1);
961 __ lea(rscratch1, iaddress(r1));
962 __ strw(r0, Address(rscratch1));
963 }
964
965 void TemplateTable::wide_lstore() {
966 transition(vtos, vtos);
967 __ pop_l();
968 locals_index_wide(r1);
969 __ str(r0, laddress(r1, rscratch1, _masm));
970 }
971
972 void TemplateTable::wide_fstore() {
973 transition(vtos, vtos);
974 __ pop_f();
975 locals_index_wide(r1);
976 __ lea(rscratch1, faddress(r1));
977 __ strs(v0, rscratch1);
978 }
979
980 void TemplateTable::wide_dstore() {
981 transition(vtos, vtos);
982 __ pop_d();
983 locals_index_wide(r1);
984 __ strd(v0, daddress(r1, rscratch1, _masm));
985 }
986
987 void TemplateTable::wide_astore() {
988 transition(vtos, vtos);
989 __ pop_ptr(r0);
990 locals_index_wide(r1);
991 __ str(r0, aaddress(r1));
992 }
993
994 void TemplateTable::iastore() {
995 transition(itos, vtos);
996 __ pop_i(r1);
997 __ pop_ptr(r3);
998 // r0: value
999 // r1: index
1000 // r3: array
1001 index_check(r3, r1); // prefer index in r1
1002 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
1003 __ strw(r0, Address(rscratch1,
1004 arrayOopDesc::base_offset_in_bytes(T_INT)));
1005 }
1006
1007 void TemplateTable::lastore() {
1008 transition(ltos, vtos);
1009 __ pop_i(r1);
1010 __ pop_ptr(r3);
1011 // r0: value
1012 // r1: index
1013 // r3: array
1014 index_check(r3, r1); // prefer index in r1
1015 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
1016 __ str(r0, Address(rscratch1,
1017 arrayOopDesc::base_offset_in_bytes(T_LONG)));
1018 }
1019
1020 void TemplateTable::fastore() {
1021 transition(ftos, vtos);
1022 __ pop_i(r1);
1023 __ pop_ptr(r3);
1024 // v0: value
1025 // r1: index
1026 // r3: array
1027 index_check(r3, r1); // prefer index in r1
1028 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
1029 __ strs(v0, Address(rscratch1,
1030 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1031 }
1032
1033 void TemplateTable::dastore() {
1034 transition(dtos, vtos);
1035 __ pop_i(r1);
1036 __ pop_ptr(r3);
1037 // v0: value
1038 // r1: index
1039 // r3: array
1040 index_check(r3, r1); // prefer index in r1
1041 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
1042 __ strd(v0, Address(rscratch1,
1043 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1044 }
1045
1046 void TemplateTable::aastore() {
1047 Label is_null, ok_is_subtype, done;
1048 transition(vtos, vtos);
1049 // stack: ..., array, index, value
1050 __ ldr(r0, at_tos()); // value
1051 __ ldr(r2, at_tos_p1()); // index
1052 __ ldr(r3, at_tos_p2()); // array
1053
1054 Address element_address(r4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1055
1056 index_check(r3, r2); // kills r1
1057 __ lea(r4, Address(r3, r2, Address::uxtw(UseCompressedOops? 2 : 3)));
1058
1059 // do array store check - check for NULL value first
1060 __ cbz(r0, is_null);
1061
1062 // Move subklass into r1
1063 __ load_klass(r1, r0);
1064 // Move superklass into r0
1065 __ load_klass(r0, r3);
1066 __ ldr(r0, Address(r0,
1067 ObjArrayKlass::element_klass_offset()));
1068 // Compress array + index*oopSize + 12 into a single register. Frees r2.
1069
1070 // Generate subtype check. Blows r2, r5
1071 // Superklass in r0. Subklass in r1.
1072 __ gen_subtype_check(r1, ok_is_subtype);
1073
1074 // Come here on failure
1075 // object is at TOS
1076 __ b(Interpreter::_throw_ArrayStoreException_entry);
1077
1078 // Come here on success
1079 __ bind(ok_is_subtype);
1080
1081 // Get the value we will store
1082 __ ldr(r0, at_tos());
1083 // Now store using the appropriate barrier
1084 do_oop_store(_masm, element_address, r0, _bs->kind(), true);
1085 __ b(done);
1086
1087 // Have a NULL in r0, r3=array, r2=index. Store NULL at ary[idx]
1088 __ bind(is_null);
1089 __ profile_null_seen(r2);
1090
1091 // Store a NULL
1092 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1093
1094 // Pop stack arguments
1095 __ bind(done);
1096 __ add(esp, esp, 3 * Interpreter::stackElementSize);
1097 }
1098
1099 void TemplateTable::bastore()
1100 {
1101 transition(itos, vtos);
1102 __ pop_i(r1);
1103 __ pop_ptr(r3);
1104 // r0: value
1105 // r1: index
1106 // r3: array
1107 index_check(r3, r1); // prefer index in r1
1108
1109 // Need to check whether array is boolean or byte
1110 // since both types share the bastore bytecode.
1111 __ load_klass(r2, r3);
1112 __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1113 int diffbit = Klass::layout_helper_boolean_diffbit();
1114 __ andw(rscratch1, r2, diffbit);
1115 Label L_skip;
1116 __ cbzw(rscratch1, L_skip);
1117 __ andw(r0, r0, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1118 __ bind(L_skip);
1119
1120 __ lea(rscratch1, Address(r3, r1, Address::uxtw(0)));
1121 __ strb(r0, Address(rscratch1,
1122 arrayOopDesc::base_offset_in_bytes(T_BYTE)));
1123 }
1124
1125 void TemplateTable::castore()
1126 {
1127 transition(itos, vtos);
1128 __ pop_i(r1);
1129 __ pop_ptr(r3);
1130 // r0: value
1131 // r1: index
1132 // r3: array
1133 index_check(r3, r1); // prefer index in r1
1134 __ lea(rscratch1, Address(r3, r1, Address::uxtw(1)));
1135 __ strh(r0, Address(rscratch1,
1136 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1137 }
1138
1139 void TemplateTable::sastore()
1140 {
1141 castore();
1142 }
1143
1144 void TemplateTable::istore(int n)
1145 {
1146 transition(itos, vtos);
1147 __ str(r0, iaddress(n));
1148 }
1149
1150 void TemplateTable::lstore(int n)
1151 {
1152 transition(ltos, vtos);
1153 __ str(r0, laddress(n));
1154 }
1155
1156 void TemplateTable::fstore(int n)
1157 {
1158 transition(ftos, vtos);
1159 __ strs(v0, faddress(n));
1160 }
1161
1162 void TemplateTable::dstore(int n)
1163 {
1164 transition(dtos, vtos);
1165 __ strd(v0, daddress(n));
1166 }
1167
1168 void TemplateTable::astore(int n)
1169 {
1170 transition(vtos, vtos);
1171 __ pop_ptr(r0);
1172 __ str(r0, iaddress(n));
1173 }
1174
1175 void TemplateTable::pop()
1176 {
1177 transition(vtos, vtos);
1178 __ add(esp, esp, Interpreter::stackElementSize);
1179 }
1180
1181 void TemplateTable::pop2()
1182 {
1183 transition(vtos, vtos);
1184 __ add(esp, esp, 2 * Interpreter::stackElementSize);
1185 }
1186
1187 void TemplateTable::dup()
1188 {
1189 transition(vtos, vtos);
1190 __ ldr(r0, Address(esp, 0));
1191 __ push(r0);
1192 // stack: ..., a, a
1193 }
1194
1195 void TemplateTable::dup_x1()
1196 {
1197 transition(vtos, vtos);
1198 // stack: ..., a, b
1199 __ ldr(r0, at_tos()); // load b
1200 __ ldr(r2, at_tos_p1()); // load a
1201 __ str(r0, at_tos_p1()); // store b
1202 __ str(r2, at_tos()); // store a
1203 __ push(r0); // push b
1204 // stack: ..., b, a, b
1205 }
1206
1207 void TemplateTable::dup_x2()
1208 {
1209 transition(vtos, vtos);
1210 // stack: ..., a, b, c
1211 __ ldr(r0, at_tos()); // load c
1212 __ ldr(r2, at_tos_p2()); // load a
1213 __ str(r0, at_tos_p2()); // store c in a
1214 __ push(r0); // push c
1215 // stack: ..., c, b, c, c
1216 __ ldr(r0, at_tos_p2()); // load b
1217 __ str(r2, at_tos_p2()); // store a in b
1218 // stack: ..., c, a, c, c
1219 __ str(r0, at_tos_p1()); // store b in c
1220 // stack: ..., c, a, b, c
1221 }
1222
1223 void TemplateTable::dup2()
1224 {
1225 transition(vtos, vtos);
1226 // stack: ..., a, b
1227 __ ldr(r0, at_tos_p1()); // load a
1228 __ push(r0); // push a
1229 __ ldr(r0, at_tos_p1()); // load b
1230 __ push(r0); // push b
1231 // stack: ..., a, b, a, b
1232 }
1233
1234 void TemplateTable::dup2_x1()
1235 {
1236 transition(vtos, vtos);
1237 // stack: ..., a, b, c
1238 __ ldr(r2, at_tos()); // load c
1239 __ ldr(r0, at_tos_p1()); // load b
1240 __ push(r0); // push b
1241 __ push(r2); // push c
1242 // stack: ..., a, b, c, b, c
1243 __ str(r2, at_tos_p3()); // store c in b
1244 // stack: ..., a, c, c, b, c
1245 __ ldr(r2, at_tos_p4()); // load a
1246 __ str(r2, at_tos_p2()); // store a in 2nd c
1247 // stack: ..., a, c, a, b, c
1248 __ str(r0, at_tos_p4()); // store b in a
1249 // stack: ..., b, c, a, b, c
1250 }
1251
1252 void TemplateTable::dup2_x2()
1253 {
1254 transition(vtos, vtos);
1255 // stack: ..., a, b, c, d
1256 __ ldr(r2, at_tos()); // load d
1257 __ ldr(r0, at_tos_p1()); // load c
1258 __ push(r0) ; // push c
1259 __ push(r2); // push d
1260 // stack: ..., a, b, c, d, c, d
1261 __ ldr(r0, at_tos_p4()); // load b
1262 __ str(r0, at_tos_p2()); // store b in d
1263 __ str(r2, at_tos_p4()); // store d in b
1264 // stack: ..., a, d, c, b, c, d
1265 __ ldr(r2, at_tos_p5()); // load a
1266 __ ldr(r0, at_tos_p3()); // load c
1267 __ str(r2, at_tos_p3()); // store a in c
1268 __ str(r0, at_tos_p5()); // store c in a
1269 // stack: ..., c, d, a, b, c, d
1270 }
1271
1272 void TemplateTable::swap()
1273 {
1274 transition(vtos, vtos);
1275 // stack: ..., a, b
1276 __ ldr(r2, at_tos_p1()); // load a
1277 __ ldr(r0, at_tos()); // load b
1278 __ str(r2, at_tos()); // store a in b
1279 __ str(r0, at_tos_p1()); // store b in a
1280 // stack: ..., b, a
1281 }
1282
1283 void TemplateTable::iop2(Operation op)
1284 {
1285 transition(itos, itos);
1286 // r0 <== r1 op r0
1287 __ pop_i(r1);
1288 switch (op) {
1289 case add : __ addw(r0, r1, r0); break;
1290 case sub : __ subw(r0, r1, r0); break;
1291 case mul : __ mulw(r0, r1, r0); break;
1292 case _and : __ andw(r0, r1, r0); break;
1293 case _or : __ orrw(r0, r1, r0); break;
1294 case _xor : __ eorw(r0, r1, r0); break;
1295 case shl : __ lslvw(r0, r1, r0); break;
1296 case shr : __ asrvw(r0, r1, r0); break;
1297 case ushr : __ lsrvw(r0, r1, r0);break;
1298 default : ShouldNotReachHere();
1299 }
1300 }
1301
1302 void TemplateTable::lop2(Operation op)
1303 {
1304 transition(ltos, ltos);
1305 // r0 <== r1 op r0
1306 __ pop_l(r1);
1307 switch (op) {
1308 case add : __ add(r0, r1, r0); break;
1309 case sub : __ sub(r0, r1, r0); break;
1310 case mul : __ mul(r0, r1, r0); break;
1311 case _and : __ andr(r0, r1, r0); break;
1312 case _or : __ orr(r0, r1, r0); break;
1313 case _xor : __ eor(r0, r1, r0); break;
1314 default : ShouldNotReachHere();
1315 }
1316 }
1317
1318 void TemplateTable::idiv()
1319 {
1320 transition(itos, itos);
1321 // explicitly check for div0
1322 Label no_div0;
1323 __ cbnzw(r0, no_div0);
1324 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1325 __ br(rscratch1);
1326 __ bind(no_div0);
1327 __ pop_i(r1);
1328 // r0 <== r1 idiv r0
1329 __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1330 }
1331
1332 void TemplateTable::irem()
1333 {
1334 transition(itos, itos);
1335 // explicitly check for div0
1336 Label no_div0;
1337 __ cbnzw(r0, no_div0);
1338 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1339 __ br(rscratch1);
1340 __ bind(no_div0);
1341 __ pop_i(r1);
1342 // r0 <== r1 irem r0
1343 __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1344 }
1345
1346 void TemplateTable::lmul()
1347 {
1348 transition(ltos, ltos);
1349 __ pop_l(r1);
1350 __ mul(r0, r0, r1);
1351 }
1352
1353 void TemplateTable::ldiv()
1354 {
1355 transition(ltos, ltos);
1356 // explicitly check for div0
1357 Label no_div0;
1358 __ cbnz(r0, no_div0);
1359 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1360 __ br(rscratch1);
1361 __ bind(no_div0);
1362 __ pop_l(r1);
1363 // r0 <== r1 ldiv r0
1364 __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1365 }
1366
1367 void TemplateTable::lrem()
1368 {
1369 transition(ltos, ltos);
1370 // explicitly check for div0
1371 Label no_div0;
1372 __ cbnz(r0, no_div0);
1373 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1374 __ br(rscratch1);
1375 __ bind(no_div0);
1376 __ pop_l(r1);
1377 // r0 <== r1 lrem r0
1378 __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1379 }
1380
1381 void TemplateTable::lshl()
1382 {
1383 transition(itos, ltos);
1384 // shift count is in r0
1385 __ pop_l(r1);
1386 __ lslv(r0, r1, r0);
1387 }
1388
1389 void TemplateTable::lshr()
1390 {
1391 transition(itos, ltos);
1392 // shift count is in r0
1393 __ pop_l(r1);
1394 __ asrv(r0, r1, r0);
1395 }
1396
1397 void TemplateTable::lushr()
1398 {
1399 transition(itos, ltos);
1400 // shift count is in r0
1401 __ pop_l(r1);
1402 __ lsrv(r0, r1, r0);
1403 }
1404
1405 void TemplateTable::fop2(Operation op)
1406 {
1407 transition(ftos, ftos);
1408 switch (op) {
1409 case add:
1410 // n.b. use ldrd because this is a 64 bit slot
1411 __ pop_f(v1);
1412 __ fadds(v0, v1, v0);
1413 break;
1414 case sub:
1415 __ pop_f(v1);
1416 __ fsubs(v0, v1, v0);
1417 break;
1418 case mul:
1419 __ pop_f(v1);
1420 __ fmuls(v0, v1, v0);
1421 break;
1422 case div:
1423 __ pop_f(v1);
1424 __ fdivs(v0, v1, v0);
1425 break;
1426 case rem:
1427 __ fmovs(v1, v0);
1428 __ pop_f(v0);
1429 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem));
1430 break;
1431 default:
1432 ShouldNotReachHere();
1433 break;
1434 }
1435 }
1436
1437 void TemplateTable::dop2(Operation op)
1438 {
1439 transition(dtos, dtos);
1440 switch (op) {
1441 case add:
1442 // n.b. use ldrd because this is a 64 bit slot
1443 __ pop_d(v1);
1444 __ faddd(v0, v1, v0);
1445 break;
1446 case sub:
1447 __ pop_d(v1);
1448 __ fsubd(v0, v1, v0);
1449 break;
1450 case mul:
1451 __ pop_d(v1);
1452 __ fmuld(v0, v1, v0);
1453 break;
1454 case div:
1455 __ pop_d(v1);
1456 __ fdivd(v0, v1, v0);
1457 break;
1458 case rem:
1459 __ fmovd(v1, v0);
1460 __ pop_d(v0);
1461 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem));
1462 break;
1463 default:
1464 ShouldNotReachHere();
1465 break;
1466 }
1467 }
1468
1469 void TemplateTable::ineg()
1470 {
1471 transition(itos, itos);
1472 __ negw(r0, r0);
1473
1474 }
1475
1476 void TemplateTable::lneg()
1477 {
1478 transition(ltos, ltos);
1479 __ neg(r0, r0);
1480 }
1481
1482 void TemplateTable::fneg()
1483 {
1484 transition(ftos, ftos);
1485 __ fnegs(v0, v0);
1486 }
1487
1488 void TemplateTable::dneg()
1489 {
1490 transition(dtos, dtos);
1491 __ fnegd(v0, v0);
1492 }
1493
1494 void TemplateTable::iinc()
1495 {
1496 transition(vtos, vtos);
1497 __ load_signed_byte(r1, at_bcp(2)); // get constant
1498 locals_index(r2);
1499 __ ldr(r0, iaddress(r2));
1500 __ addw(r0, r0, r1);
1501 __ str(r0, iaddress(r2));
1502 }
1503
1504 void TemplateTable::wide_iinc()
1505 {
1506 transition(vtos, vtos);
1507 // __ mov(r1, zr);
1508 __ ldrw(r1, at_bcp(2)); // get constant and index
1509 __ rev16(r1, r1);
1510 __ ubfx(r2, r1, 0, 16);
1511 __ neg(r2, r2);
1512 __ sbfx(r1, r1, 16, 16);
1513 __ ldr(r0, iaddress(r2));
1514 __ addw(r0, r0, r1);
1515 __ str(r0, iaddress(r2));
1516 }
1517
1518 void TemplateTable::convert()
1519 {
1520 // Checking
1521 #ifdef ASSERT
1522 {
1523 TosState tos_in = ilgl;
1524 TosState tos_out = ilgl;
1525 switch (bytecode()) {
1526 case Bytecodes::_i2l: // fall through
1527 case Bytecodes::_i2f: // fall through
1528 case Bytecodes::_i2d: // fall through
1529 case Bytecodes::_i2b: // fall through
1530 case Bytecodes::_i2c: // fall through
1531 case Bytecodes::_i2s: tos_in = itos; break;
1532 case Bytecodes::_l2i: // fall through
1533 case Bytecodes::_l2f: // fall through
1534 case Bytecodes::_l2d: tos_in = ltos; break;
1535 case Bytecodes::_f2i: // fall through
1536 case Bytecodes::_f2l: // fall through
1537 case Bytecodes::_f2d: tos_in = ftos; break;
1538 case Bytecodes::_d2i: // fall through
1539 case Bytecodes::_d2l: // fall through
1540 case Bytecodes::_d2f: tos_in = dtos; break;
1541 default : ShouldNotReachHere();
1542 }
1543 switch (bytecode()) {
1544 case Bytecodes::_l2i: // fall through
1545 case Bytecodes::_f2i: // fall through
1546 case Bytecodes::_d2i: // fall through
1547 case Bytecodes::_i2b: // fall through
1548 case Bytecodes::_i2c: // fall through
1549 case Bytecodes::_i2s: tos_out = itos; break;
1550 case Bytecodes::_i2l: // fall through
1551 case Bytecodes::_f2l: // fall through
1552 case Bytecodes::_d2l: tos_out = ltos; break;
1553 case Bytecodes::_i2f: // fall through
1554 case Bytecodes::_l2f: // fall through
1555 case Bytecodes::_d2f: tos_out = ftos; break;
1556 case Bytecodes::_i2d: // fall through
1557 case Bytecodes::_l2d: // fall through
1558 case Bytecodes::_f2d: tos_out = dtos; break;
1559 default : ShouldNotReachHere();
1560 }
1561 transition(tos_in, tos_out);
1562 }
1563 #endif // ASSERT
1564 // static const int64_t is_nan = 0x8000000000000000L;
1565
1566 // Conversion
1567 switch (bytecode()) {
1568 case Bytecodes::_i2l:
1569 __ sxtw(r0, r0);
1570 break;
1571 case Bytecodes::_i2f:
1572 __ scvtfws(v0, r0);
1573 break;
1574 case Bytecodes::_i2d:
1575 __ scvtfwd(v0, r0);
1576 break;
1577 case Bytecodes::_i2b:
1578 __ sxtbw(r0, r0);
1579 break;
1580 case Bytecodes::_i2c:
1581 __ uxthw(r0, r0);
1582 break;
1583 case Bytecodes::_i2s:
1584 __ sxthw(r0, r0);
1585 break;
1586 case Bytecodes::_l2i:
1587 __ uxtw(r0, r0);
1588 break;
1589 case Bytecodes::_l2f:
1590 __ scvtfs(v0, r0);
1591 break;
1592 case Bytecodes::_l2d:
1593 __ scvtfd(v0, r0);
1594 break;
1595 case Bytecodes::_f2i:
1596 {
1597 Label L_Okay;
1598 __ clear_fpsr();
1599 __ fcvtzsw(r0, v0);
1600 __ get_fpsr(r1);
1601 __ cbzw(r1, L_Okay);
1602 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i));
1603 __ bind(L_Okay);
1604 }
1605 break;
1606 case Bytecodes::_f2l:
1607 {
1608 Label L_Okay;
1609 __ clear_fpsr();
1610 __ fcvtzs(r0, v0);
1611 __ get_fpsr(r1);
1612 __ cbzw(r1, L_Okay);
1613 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l));
1614 __ bind(L_Okay);
1615 }
1616 break;
1617 case Bytecodes::_f2d:
1618 __ fcvts(v0, v0);
1619 break;
1620 case Bytecodes::_d2i:
1621 {
1622 Label L_Okay;
1623 __ clear_fpsr();
1624 __ fcvtzdw(r0, v0);
1625 __ get_fpsr(r1);
1626 __ cbzw(r1, L_Okay);
1627 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i));
1628 __ bind(L_Okay);
1629 }
1630 break;
1631 case Bytecodes::_d2l:
1632 {
1633 Label L_Okay;
1634 __ clear_fpsr();
1635 __ fcvtzd(r0, v0);
1636 __ get_fpsr(r1);
1637 __ cbzw(r1, L_Okay);
1638 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l));
1639 __ bind(L_Okay);
1640 }
1641 break;
1642 case Bytecodes::_d2f:
1643 __ fcvtd(v0, v0);
1644 break;
1645 default:
1646 ShouldNotReachHere();
1647 }
1648 }
1649
1650 void TemplateTable::lcmp()
1651 {
1652 transition(ltos, itos);
1653 Label done;
1654 __ pop_l(r1);
1655 __ cmp(r1, r0);
1656 __ mov(r0, (u_int64_t)-1L);
1657 __ br(Assembler::LT, done);
1658 // __ mov(r0, 1UL);
1659 // __ csel(r0, r0, zr, Assembler::NE);
1660 // and here is a faster way
1661 __ csinc(r0, zr, zr, Assembler::EQ);
1662 __ bind(done);
1663 }
1664
1665 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1666 {
1667 Label done;
1668 if (is_float) {
1669 // XXX get rid of pop here, use ... reg, mem32
1670 __ pop_f(v1);
1671 __ fcmps(v1, v0);
1672 } else {
1673 // XXX get rid of pop here, use ... reg, mem64
1674 __ pop_d(v1);
1675 __ fcmpd(v1, v0);
1676 }
1677 if (unordered_result < 0) {
1678 // we want -1 for unordered or less than, 0 for equal and 1 for
1679 // greater than.
1680 __ mov(r0, (u_int64_t)-1L);
1681 // for FP LT tests less than or unordered
1682 __ br(Assembler::LT, done);
1683 // install 0 for EQ otherwise 1
1684 __ csinc(r0, zr, zr, Assembler::EQ);
1685 } else {
1686 // we want -1 for less than, 0 for equal and 1 for unordered or
1687 // greater than.
1688 __ mov(r0, 1L);
1689 // for FP HI tests greater than or unordered
1690 __ br(Assembler::HI, done);
1691 // install 0 for EQ otherwise ~0
1692 __ csinv(r0, zr, zr, Assembler::EQ);
1693
1694 }
1695 __ bind(done);
1696 }
1697
1698 void TemplateTable::branch(bool is_jsr, bool is_wide)
1699 {
1700 // We might be moving to a safepoint. The thread which calls
1701 // Interpreter::notice_safepoints() will effectively flush its cache
1702 // when it makes a system call, but we need to do something to
1703 // ensure that we see the changed dispatch table.
1704 __ membar(MacroAssembler::LoadLoad);
1705
1706 __ profile_taken_branch(r0, r1);
1707 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1708 InvocationCounter::counter_offset();
1709 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1710 InvocationCounter::counter_offset();
1711
1712 // load branch displacement
1713 if (!is_wide) {
1714 __ ldrh(r2, at_bcp(1));
1715 __ rev16(r2, r2);
1716 // sign extend the 16 bit value in r2
1717 __ sbfm(r2, r2, 0, 15);
1718 } else {
1719 __ ldrw(r2, at_bcp(1));
1720 __ revw(r2, r2);
1721 // sign extend the 32 bit value in r2
1722 __ sbfm(r2, r2, 0, 31);
1723 }
1724
1725 // Handle all the JSR stuff here, then exit.
1726 // It's much shorter and cleaner than intermingling with the non-JSR
1727 // normal-branch stuff occurring below.
1728
1729 if (is_jsr) {
1730 // Pre-load the next target bytecode into rscratch1
1731 __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1732 // compute return address as bci
1733 __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1734 __ add(rscratch2, rscratch2,
1735 in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1736 __ sub(r1, rbcp, rscratch2);
1737 __ push_i(r1);
1738 // Adjust the bcp by the 16-bit displacement in r2
1739 __ add(rbcp, rbcp, r2);
1740 __ dispatch_only(vtos);
1741 return;
1742 }
1743
1744 // Normal (non-jsr) branch handling
1745
1746 // Adjust the bcp by the displacement in r2
1747 __ add(rbcp, rbcp, r2);
1748
1749 assert(UseLoopCounter || !UseOnStackReplacement,
1750 "on-stack-replacement requires loop counters");
1751 Label backedge_counter_overflow;
1752 Label profile_method;
1753 Label dispatch;
1754 if (UseLoopCounter) {
1755 // increment backedge counter for backward branches
1756 // r0: MDO
1757 // w1: MDO bumped taken-count
1758 // r2: target offset
1759 __ cmp(r2, zr);
1760 __ br(Assembler::GT, dispatch); // count only if backward branch
1761
1762 // ECN: FIXME: This code smells
1763 // check if MethodCounters exists
1764 Label has_counters;
1765 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1766 __ cbnz(rscratch1, has_counters);
1767 __ push(r0);
1768 __ push(r1);
1769 __ push(r2);
1770 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1771 InterpreterRuntime::build_method_counters), rmethod);
1772 __ pop(r2);
1773 __ pop(r1);
1774 __ pop(r0);
1775 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1776 __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1777 __ bind(has_counters);
1778
1779 if (TieredCompilation) {
1780 Label no_mdo;
1781 int increment = InvocationCounter::count_increment;
1782 int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1783 if (ProfileInterpreter) {
1784 // Are we profiling?
1785 __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1786 __ cbz(r1, no_mdo);
1787 // Increment the MDO backedge counter
1788 const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1789 in_bytes(InvocationCounter::counter_offset()));
1790 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1791 r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow);
1792 __ b(dispatch);
1793 }
1794 __ bind(no_mdo);
1795 // Increment backedge counter in MethodCounters*
1796 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1797 __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1798 r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow);
1799 } else {
1800 // increment counter
1801 __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1802 __ ldrw(r0, Address(rscratch2, be_offset)); // load backedge counter
1803 __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1804 __ strw(rscratch1, Address(rscratch2, be_offset)); // store counter
1805
1806 __ ldrw(r0, Address(rscratch2, inv_offset)); // load invocation counter
1807 __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1808 __ addw(r0, r0, rscratch1); // add both counters
1809
1810 if (ProfileInterpreter) {
1811 // Test to see if we should create a method data oop
1812 __ lea(rscratch1, ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1813 __ ldrw(rscratch1, rscratch1);
1814 __ cmpw(r0, rscratch1);
1815 __ br(Assembler::LT, dispatch);
1816
1817 // if no method data exists, go to profile method
1818 __ test_method_data_pointer(r0, profile_method);
1819
1820 if (UseOnStackReplacement) {
1821 // check for overflow against w1 which is the MDO taken count
1822 __ lea(rscratch1, ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1823 __ ldrw(rscratch1, rscratch1);
1824 __ cmpw(r1, rscratch1);
1825 __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1826
1827 // When ProfileInterpreter is on, the backedge_count comes
1828 // from the MethodData*, which value does not get reset on
1829 // the call to frequency_counter_overflow(). To avoid
1830 // excessive calls to the overflow routine while the method is
1831 // being compiled, add a second test to make sure the overflow
1832 // function is called only once every overflow_frequency.
1833 const int overflow_frequency = 1024;
1834 __ andsw(r1, r1, overflow_frequency - 1);
1835 __ br(Assembler::EQ, backedge_counter_overflow);
1836
1837 }
1838 } else {
1839 if (UseOnStackReplacement) {
1840 // check for overflow against w0, which is the sum of the
1841 // counters
1842 __ lea(rscratch1, ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1843 __ ldrw(rscratch1, rscratch1);
1844 __ cmpw(r0, rscratch1);
1845 __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1846 }
1847 }
1848 }
1849 }
1850 __ bind(dispatch);
1851
1852 // Pre-load the next target bytecode into rscratch1
1853 __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1854
1855 // continue with the bytecode @ target
1856 // rscratch1: target bytecode
1857 // rbcp: target bcp
1858 __ dispatch_only(vtos);
1859
1860 if (UseLoopCounter) {
1861 if (ProfileInterpreter) {
1862 // Out-of-line code to allocate method data oop.
1863 __ bind(profile_method);
1864 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1865 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
1866 __ set_method_data_pointer_for_bcp();
1867 __ b(dispatch);
1868 }
1869
1870 if (TieredCompilation || UseOnStackReplacement) {
1871 // invocation counter overflow
1872 __ bind(backedge_counter_overflow);
1873 __ neg(r2, r2);
1874 __ add(r2, r2, rbcp); // branch bcp
1875 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1876 __ call_VM(noreg,
1877 CAST_FROM_FN_PTR(address,
1878 InterpreterRuntime::frequency_counter_overflow),
1879 r2);
1880 if (!UseOnStackReplacement)
1881 __ b(dispatch);
1882 }
1883
1884 if (UseOnStackReplacement) {
1885 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
1886
1887 // r0: osr nmethod (osr ok) or NULL (osr not possible)
1888 // w1: target bytecode
1889 // r2: scratch
1890 __ cbz(r0, dispatch); // test result -- no osr if null
1891 // nmethod may have been invalidated (VM may block upon call_VM return)
1892 __ ldrw(r2, Address(r0, nmethod::entry_bci_offset()));
1893 // InvalidOSREntryBci == -2 which overflows cmpw as unsigned
1894 // use cmnw against -InvalidOSREntryBci which does the same thing
1895 __ cmn(r2, -InvalidOSREntryBci);
1896 __ br(Assembler::EQ, dispatch);
1897
1898 // We have the address of an on stack replacement routine in r0
1899 // We need to prepare to execute the OSR method. First we must
1900 // migrate the locals and monitors off of the stack.
1901
1902 __ mov(r19, r0); // save the nmethod
1903
1904 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1905
1906 // r0 is OSR buffer, move it to expected parameter location
1907 __ mov(j_rarg0, r0);
1908
1909 // remove activation
1910 // get sender esp
1911 __ ldr(esp,
1912 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1913 // remove frame anchor
1914 __ leave();
1915 // Ensure compiled code always sees stack at proper alignment
1916 __ andr(sp, esp, -16);
1917
1918 // and begin the OSR nmethod
1919 __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1920 __ br(rscratch1);
1921 }
1922 }
1923 }
1924
1925
1926 void TemplateTable::if_0cmp(Condition cc)
1927 {
1928 transition(itos, vtos);
1929 // assume branch is more often taken than not (loops use backward branches)
1930 Label not_taken;
1931 if (cc == equal)
1932 __ cbnzw(r0, not_taken);
1933 else if (cc == not_equal)
1934 __ cbzw(r0, not_taken);
1935 else {
1936 __ andsw(zr, r0, r0);
1937 __ br(j_not(cc), not_taken);
1938 }
1939
1940 branch(false, false);
1941 __ bind(not_taken);
1942 __ profile_not_taken_branch(r0);
1943 }
1944
1945 void TemplateTable::if_icmp(Condition cc)
1946 {
1947 transition(itos, vtos);
1948 // assume branch is more often taken than not (loops use backward branches)
1949 Label not_taken;
1950 __ pop_i(r1);
1951 __ cmpw(r1, r0, Assembler::LSL);
1952 __ br(j_not(cc), not_taken);
1953 branch(false, false);
1954 __ bind(not_taken);
1955 __ profile_not_taken_branch(r0);
1956 }
1957
1958 void TemplateTable::if_nullcmp(Condition cc)
1959 {
1960 transition(atos, vtos);
1961 // assume branch is more often taken than not (loops use backward branches)
1962 Label not_taken;
1963 if (cc == equal)
1964 __ cbnz(r0, not_taken);
1965 else
1966 __ cbz(r0, not_taken);
1967 branch(false, false);
1968 __ bind(not_taken);
1969 __ profile_not_taken_branch(r0);
1970 }
1971
1972 void TemplateTable::if_acmp(Condition cc)
1973 {
1974 transition(atos, vtos);
1975 // assume branch is more often taken than not (loops use backward branches)
1976 Label not_taken;
1977 __ pop_ptr(r1);
1978 __ cmp(r1, r0);
1979 __ br(j_not(cc), not_taken);
1980 branch(false, false);
1981 __ bind(not_taken);
1982 __ profile_not_taken_branch(r0);
1983 }
1984
1985 void TemplateTable::ret() {
1986 transition(vtos, vtos);
1987 // We might be moving to a safepoint. The thread which calls
1988 // Interpreter::notice_safepoints() will effectively flush its cache
1989 // when it makes a system call, but we need to do something to
1990 // ensure that we see the changed dispatch table.
1991 __ membar(MacroAssembler::LoadLoad);
1992
1993 locals_index(r1);
1994 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1995 __ profile_ret(r1, r2);
1996 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1997 __ lea(rbcp, Address(rbcp, r1));
1998 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1999 __ dispatch_next(vtos);
2000 }
2001
2002 void TemplateTable::wide_ret() {
2003 transition(vtos, vtos);
2004 locals_index_wide(r1);
2005 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2006 __ profile_ret(r1, r2);
2007 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2008 __ lea(rbcp, Address(rbcp, r1));
2009 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2010 __ dispatch_next(vtos);
2011 }
2012
2013
2014 void TemplateTable::tableswitch() {
2015 Label default_case, continue_execution;
2016 transition(itos, vtos);
2017 // align rbcp
2018 __ lea(r1, at_bcp(BytesPerInt));
2019 __ andr(r1, r1, -BytesPerInt);
2020 // load lo & hi
2021 __ ldrw(r2, Address(r1, BytesPerInt));
2022 __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2023 __ rev32(r2, r2);
2024 __ rev32(r3, r3);
2025 // check against lo & hi
2026 __ cmpw(r0, r2);
2027 __ br(Assembler::LT, default_case);
2028 __ cmpw(r0, r3);
2029 __ br(Assembler::GT, default_case);
2030 // lookup dispatch offset
2031 __ subw(r0, r0, r2);
2032 __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2033 __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2034 __ profile_switch_case(r0, r1, r2);
2035 // continue execution
2036 __ bind(continue_execution);
2037 __ rev32(r3, r3);
2038 __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2039 __ add(rbcp, rbcp, r3, ext::sxtw);
2040 __ dispatch_only(vtos);
2041 // handle default
2042 __ bind(default_case);
2043 __ profile_switch_default(r0);
2044 __ ldrw(r3, Address(r1, 0));
2045 __ b(continue_execution);
2046 }
2047
2048 void TemplateTable::lookupswitch() {
2049 transition(itos, itos);
2050 __ stop("lookupswitch bytecode should have been rewritten");
2051 }
2052
2053 void TemplateTable::fast_linearswitch() {
2054 transition(itos, vtos);
2055 Label loop_entry, loop, found, continue_execution;
2056 // bswap r0 so we can avoid bswapping the table entries
2057 __ rev32(r0, r0);
2058 // align rbcp
2059 __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2060 // this instruction (change offsets
2061 // below)
2062 __ andr(r19, r19, -BytesPerInt);
2063 // set counter
2064 __ ldrw(r1, Address(r19, BytesPerInt));
2065 __ rev32(r1, r1);
2066 __ b(loop_entry);
2067 // table search
2068 __ bind(loop);
2069 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2070 __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2071 __ cmpw(r0, rscratch1);
2072 __ br(Assembler::EQ, found);
2073 __ bind(loop_entry);
2074 __ subs(r1, r1, 1);
2075 __ br(Assembler::PL, loop);
2076 // default case
2077 __ profile_switch_default(r0);
2078 __ ldrw(r3, Address(r19, 0));
2079 __ b(continue_execution);
2080 // entry found -> get offset
2081 __ bind(found);
2082 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2083 __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2084 __ profile_switch_case(r1, r0, r19);
2085 // continue execution
2086 __ bind(continue_execution);
2087 __ rev32(r3, r3);
2088 __ add(rbcp, rbcp, r3, ext::sxtw);
2089 __ ldrb(rscratch1, Address(rbcp, 0));
2090 __ dispatch_only(vtos);
2091 }
2092
2093 void TemplateTable::fast_binaryswitch() {
2094 transition(itos, vtos);
2095 // Implementation using the following core algorithm:
2096 //
2097 // int binary_search(int key, LookupswitchPair* array, int n) {
2098 // // Binary search according to "Methodik des Programmierens" by
2099 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2100 // int i = 0;
2101 // int j = n;
2102 // while (i+1 < j) {
2103 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2104 // // with Q: for all i: 0 <= i < n: key < a[i]
2105 // // where a stands for the array and assuming that the (inexisting)
2106 // // element a[n] is infinitely big.
2107 // int h = (i + j) >> 1;
2108 // // i < h < j
2109 // if (key < array[h].fast_match()) {
2110 // j = h;
2111 // } else {
2112 // i = h;
2113 // }
2114 // }
2115 // // R: a[i] <= key < a[i+1] or Q
2116 // // (i.e., if key is within array, i is the correct index)
2117 // return i;
2118 // }
2119
2120 // Register allocation
2121 const Register key = r0; // already set (tosca)
2122 const Register array = r1;
2123 const Register i = r2;
2124 const Register j = r3;
2125 const Register h = rscratch1;
2126 const Register temp = rscratch2;
2127
2128 // Find array start
2129 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2130 // get rid of this
2131 // instruction (change
2132 // offsets below)
2133 __ andr(array, array, -BytesPerInt);
2134
2135 // Initialize i & j
2136 __ mov(i, 0); // i = 0;
2137 __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2138
2139 // Convert j into native byteordering
2140 __ rev32(j, j);
2141
2142 // And start
2143 Label entry;
2144 __ b(entry);
2145
2146 // binary search loop
2147 {
2148 Label loop;
2149 __ bind(loop);
2150 // int h = (i + j) >> 1;
2151 __ addw(h, i, j); // h = i + j;
2152 __ lsrw(h, h, 1); // h = (i + j) >> 1;
2153 // if (key < array[h].fast_match()) {
2154 // j = h;
2155 // } else {
2156 // i = h;
2157 // }
2158 // Convert array[h].match to native byte-ordering before compare
2159 __ ldr(temp, Address(array, h, Address::lsl(3)));
2160 __ rev32(temp, temp);
2161 __ cmpw(key, temp);
2162 // j = h if (key < array[h].fast_match())
2163 __ csel(j, h, j, Assembler::LT);
2164 // i = h if (key >= array[h].fast_match())
2165 __ csel(i, h, i, Assembler::GE);
2166 // while (i+1 < j)
2167 __ bind(entry);
2168 __ addw(h, i, 1); // i+1
2169 __ cmpw(h, j); // i+1 < j
2170 __ br(Assembler::LT, loop);
2171 }
2172
2173 // end of binary search, result index is i (must check again!)
2174 Label default_case;
2175 // Convert array[i].match to native byte-ordering before compare
2176 __ ldr(temp, Address(array, i, Address::lsl(3)));
2177 __ rev32(temp, temp);
2178 __ cmpw(key, temp);
2179 __ br(Assembler::NE, default_case);
2180
2181 // entry found -> j = offset
2182 __ add(j, array, i, ext::uxtx, 3);
2183 __ ldrw(j, Address(j, BytesPerInt));
2184 __ profile_switch_case(i, key, array);
2185 __ rev32(j, j);
2186 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2187 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2188 __ dispatch_only(vtos);
2189
2190 // default case -> j = default offset
2191 __ bind(default_case);
2192 __ profile_switch_default(i);
2193 __ ldrw(j, Address(array, -2 * BytesPerInt));
2194 __ rev32(j, j);
2195 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2196 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2197 __ dispatch_only(vtos);
2198 }
2199
2200
2201 void TemplateTable::_return(TosState state)
2202 {
2203 transition(state, state);
2204 assert(_desc->calls_vm(),
2205 "inconsistent calls_vm information"); // call in remove_activation
2206
2207 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2208 assert(state == vtos, "only valid state");
2209
2210 __ ldr(c_rarg1, aaddress(0));
2211 __ load_klass(r3, c_rarg1);
2212 __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2213 __ tst(r3, JVM_ACC_HAS_FINALIZER);
2214 Label skip_register_finalizer;
2215 __ br(Assembler::EQ, skip_register_finalizer);
2216
2217 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2218
2219 __ bind(skip_register_finalizer);
2220 }
2221
2222 // Issue a StoreStore barrier after all stores but before return
2223 // from any constructor for any class with a final field. We don't
2224 // know if this is a finalizer, so we always do so.
2225 if (_desc->bytecode() == Bytecodes::_return)
2226 __ membar(MacroAssembler::StoreStore);
2227
2228 // Narrow result if state is itos but result type is smaller.
2229 // Need to narrow in the return bytecode rather than in generate_return_entry
2230 // since compiled code callers expect the result to already be narrowed.
2231 if (state == itos) {
2232 __ narrow(r0);
2233 }
2234
2235 __ remove_activation(state);
2236 __ ret(lr);
2237 }
2238
2239 // ----------------------------------------------------------------------------
2240 // Volatile variables demand their effects be made known to all CPU's
2241 // in order. Store buffers on most chips allow reads & writes to
2242 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2243 // without some kind of memory barrier (i.e., it's not sufficient that
2244 // the interpreter does not reorder volatile references, the hardware
2245 // also must not reorder them).
2246 //
2247 // According to the new Java Memory Model (JMM):
2248 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2249 // writes act as aquire & release, so:
2250 // (2) A read cannot let unrelated NON-volatile memory refs that
2251 // happen after the read float up to before the read. It's OK for
2252 // non-volatile memory refs that happen before the volatile read to
2253 // float down below it.
2254 // (3) Similar a volatile write cannot let unrelated NON-volatile
2255 // memory refs that happen BEFORE the write float down to after the
2256 // write. It's OK for non-volatile memory refs that happen after the
2257 // volatile write to float up before it.
2258 //
2259 // We only put in barriers around volatile refs (they are expensive),
2260 // not _between_ memory refs (that would require us to track the
2261 // flavor of the previous memory refs). Requirements (2) and (3)
2262 // require some barriers before volatile stores and after volatile
2263 // loads. These nearly cover requirement (1) but miss the
2264 // volatile-store-volatile-load case. This final case is placed after
2265 // volatile-stores although it could just as well go before
2266 // volatile-loads.
2267
2268 void TemplateTable::resolve_cache_and_index(int byte_no,
2269 Register Rcache,
2270 Register index,
2271 size_t index_size) {
2272 const Register temp = r19;
2273 assert_different_registers(Rcache, index, temp);
2274
2275 Label resolved;
2276 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2277 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2278 __ cmp(temp, (int) bytecode()); // have we resolved this bytecode?
2279 __ br(Assembler::EQ, resolved);
2280
2281 // resolve first time through
2282 address entry;
2283 switch (bytecode()) {
2284 case Bytecodes::_getstatic:
2285 case Bytecodes::_putstatic:
2286 case Bytecodes::_getfield:
2287 case Bytecodes::_putfield:
2288 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put);
2289 break;
2290 case Bytecodes::_invokevirtual:
2291 case Bytecodes::_invokespecial:
2292 case Bytecodes::_invokestatic:
2293 case Bytecodes::_invokeinterface:
2294 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke);
2295 break;
2296 case Bytecodes::_invokehandle:
2297 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle);
2298 break;
2299 case Bytecodes::_invokedynamic:
2300 entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic);
2301 break;
2302 default:
2303 fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2304 break;
2305 }
2306 __ mov(temp, (int) bytecode());
2307 __ call_VM(noreg, entry, temp);
2308
2309 // Update registers with resolved info
2310 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2311 // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2312 // so all clients ofthis method must be modified accordingly
2313 __ bind(resolved);
2314 }
2315
2316 // The Rcache and index registers must be set before call
2317 // n.b unlike x86 cache already includes the index offset
2318 void TemplateTable::load_field_cp_cache_entry(Register obj,
2319 Register cache,
2320 Register index,
2321 Register off,
2322 Register flags,
2323 bool is_static = false) {
2324 assert_different_registers(cache, index, flags, off);
2325
2326 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2327 // Field offset
2328 __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2329 ConstantPoolCacheEntry::f2_offset())));
2330 // Flags
2331 __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2332 ConstantPoolCacheEntry::flags_offset())));
2333
2334 // klass overwrite register
2335 if (is_static) {
2336 __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2337 ConstantPoolCacheEntry::f1_offset())));
2338 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2339 __ ldr(obj, Address(obj, mirror_offset));
2340 }
2341 }
2342
2343 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2344 Register method,
2345 Register itable_index,
2346 Register flags,
2347 bool is_invokevirtual,
2348 bool is_invokevfinal, /*unused*/
2349 bool is_invokedynamic) {
2350 // setup registers
2351 const Register cache = rscratch2;
2352 const Register index = r4;
2353 assert_different_registers(method, flags);
2354 assert_different_registers(method, cache, index);
2355 assert_different_registers(itable_index, flags);
2356 assert_different_registers(itable_index, cache, index);
2357 // determine constant pool cache field offsets
2358 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2359 const int method_offset = in_bytes(
2360 ConstantPoolCache::base_offset() +
2361 (is_invokevirtual
2362 ? ConstantPoolCacheEntry::f2_offset()
2363 : ConstantPoolCacheEntry::f1_offset()));
2364 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2365 ConstantPoolCacheEntry::flags_offset());
2366 // access constant pool cache fields
2367 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2368 ConstantPoolCacheEntry::f2_offset());
2369
2370 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2371 resolve_cache_and_index(byte_no, cache, index, index_size);
2372 __ ldr(method, Address(cache, method_offset));
2373
2374 if (itable_index != noreg) {
2375 __ ldr(itable_index, Address(cache, index_offset));
2376 }
2377 __ ldrw(flags, Address(cache, flags_offset));
2378 }
2379
2380
2381 // The registers cache and index expected to be set before call.
2382 // Correct values of the cache and index registers are preserved.
2383 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2384 bool is_static, bool has_tos) {
2385 // do the JVMTI work here to avoid disturbing the register state below
2386 // We use c_rarg registers here because we want to use the register used in
2387 // the call to the VM
2388 if (JvmtiExport::can_post_field_access()) {
2389 // Check to see if a field access watch has been set before we
2390 // take the time to call into the VM.
2391 Label L1;
2392 assert_different_registers(cache, index, r0);
2393 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2394 __ ldrw(r0, Address(rscratch1));
2395 __ cbzw(r0, L1);
2396
2397 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2398 __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2399
2400 if (is_static) {
2401 __ mov(c_rarg1, zr); // NULL object reference
2402 } else {
2403 __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2404 __ verify_oop(c_rarg1);
2405 }
2406 // c_rarg1: object pointer or NULL
2407 // c_rarg2: cache entry pointer
2408 // c_rarg3: jvalue object on the stack
2409 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2410 InterpreterRuntime::post_field_access),
2411 c_rarg1, c_rarg2, c_rarg3);
2412 __ get_cache_and_index_at_bcp(cache, index, 1);
2413 __ bind(L1);
2414 }
2415 }
2416
2417 void TemplateTable::pop_and_check_object(Register r)
2418 {
2419 __ pop_ptr(r);
2420 __ null_check(r); // for field access must check obj.
2421 __ verify_oop(r);
2422 }
2423
2424 void TemplateTable::getfield_or_static(int byte_no, bool is_static)
2425 {
2426 const Register cache = r2;
2427 const Register index = r3;
2428 const Register obj = r4;
2429 const Register off = r19;
2430 const Register flags = r0;
2431 const Register raw_flags = r6;
2432 const Register bc = r4; // uses same reg as obj, so don't mix them
2433
2434 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2435 jvmti_post_field_access(cache, index, is_static, false);
2436 load_field_cp_cache_entry(obj, cache, index, off, raw_flags, is_static);
2437
2438 if (!is_static) {
2439 // obj is on the stack
2440 pop_and_check_object(obj);
2441 }
2442
2443 // 8179954: We need to make sure that the code generated for
2444 // volatile accesses forms a sequentially-consistent set of
2445 // operations when combined with STLR and LDAR. Without a leading
2446 // membar it's possible for a simple Dekker test to fail if loads
2447 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
2448 // the stores in one method and we interpret the loads in another.
2449 if (! UseBarriersForVolatile) {
2450 Label notVolatile;
2451 __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2452 __ membar(MacroAssembler::AnyAny);
2453 __ bind(notVolatile);
2454 }
2455
2456 const Address field(obj, off);
2457
2458 Label Done, notByte, notBool, notInt, notShort, notChar,
2459 notLong, notFloat, notObj, notDouble;
2460
2461 // x86 uses a shift and mask or wings it with a shift plus assert
2462 // the mask is not needed. aarch64 just uses bitfield extract
2463 __ ubfxw(flags, raw_flags, ConstantPoolCacheEntry::tos_state_shift,
2464 ConstantPoolCacheEntry::tos_state_bits);
2465
2466 assert(btos == 0, "change code, btos != 0");
2467 __ cbnz(flags, notByte);
2468
2469 // btos
2470 __ load_signed_byte(r0, field);
2471 __ push(btos);
2472 // Rewrite bytecode to be faster
2473 if (!is_static) {
2474 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2475 }
2476 __ b(Done);
2477
2478 __ bind(notByte);
2479 __ cmp(flags, ztos);
2480 __ br(Assembler::NE, notBool);
2481
2482 // ztos (same code as btos)
2483 __ ldrsb(r0, field);
2484 __ push(ztos);
2485 // Rewrite bytecode to be faster
2486 if (!is_static) {
2487 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2488 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2489 }
2490 __ b(Done);
2491
2492 __ bind(notBool);
2493 __ cmp(flags, atos);
2494 __ br(Assembler::NE, notObj);
2495 // atos
2496 __ load_heap_oop(r0, field);
2497 __ push(atos);
2498 if (!is_static) {
2499 patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2500 }
2501 __ b(Done);
2502
2503 __ bind(notObj);
2504 __ cmp(flags, itos);
2505 __ br(Assembler::NE, notInt);
2506 // itos
2507 __ ldrw(r0, field);
2508 __ push(itos);
2509 // Rewrite bytecode to be faster
2510 if (!is_static) {
2511 patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2512 }
2513 __ b(Done);
2514
2515 __ bind(notInt);
2516 __ cmp(flags, ctos);
2517 __ br(Assembler::NE, notChar);
2518 // ctos
2519 __ load_unsigned_short(r0, field);
2520 __ push(ctos);
2521 // Rewrite bytecode to be faster
2522 if (!is_static) {
2523 patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2524 }
2525 __ b(Done);
2526
2527 __ bind(notChar);
2528 __ cmp(flags, stos);
2529 __ br(Assembler::NE, notShort);
2530 // stos
2531 __ load_signed_short(r0, field);
2532 __ push(stos);
2533 // Rewrite bytecode to be faster
2534 if (!is_static) {
2535 patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2536 }
2537 __ b(Done);
2538
2539 __ bind(notShort);
2540 __ cmp(flags, ltos);
2541 __ br(Assembler::NE, notLong);
2542 // ltos
2543 __ ldr(r0, field);
2544 __ push(ltos);
2545 // Rewrite bytecode to be faster
2546 if (!is_static) {
2547 patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2548 }
2549 __ b(Done);
2550
2551 __ bind(notLong);
2552 __ cmp(flags, ftos);
2553 __ br(Assembler::NE, notFloat);
2554 // ftos
2555 __ ldrs(v0, field);
2556 __ push(ftos);
2557 // Rewrite bytecode to be faster
2558 if (!is_static) {
2559 patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2560 }
2561 __ b(Done);
2562
2563 __ bind(notFloat);
2564 #ifdef ASSERT
2565 __ cmp(flags, dtos);
2566 __ br(Assembler::NE, notDouble);
2567 #endif
2568 // dtos
2569 __ ldrd(v0, field);
2570 __ push(dtos);
2571 // Rewrite bytecode to be faster
2572 if (!is_static) {
2573 patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2574 }
2575 #ifdef ASSERT
2576 __ b(Done);
2577
2578 __ bind(notDouble);
2579 __ stop("Bad state");
2580 #endif
2581
2582 __ bind(Done);
2583
2584 Label notVolatile;
2585 __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2586 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2587 __ bind(notVolatile);
2588 }
2589
2590
2591 void TemplateTable::getfield(int byte_no)
2592 {
2593 getfield_or_static(byte_no, false);
2594 }
2595
2596 void TemplateTable::getstatic(int byte_no)
2597 {
2598 getfield_or_static(byte_no, true);
2599 }
2600
2601 // The registers cache and index expected to be set before call.
2602 // The function may destroy various registers, just not the cache and index registers.
2603 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2604 transition(vtos, vtos);
2605
2606 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2607
2608 if (JvmtiExport::can_post_field_modification()) {
2609 // Check to see if a field modification watch has been set before
2610 // we take the time to call into the VM.
2611 Label L1;
2612 assert_different_registers(cache, index, r0);
2613 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2614 __ ldrw(r0, Address(rscratch1));
2615 __ cbz(r0, L1);
2616
2617 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2618
2619 if (is_static) {
2620 // Life is simple. Null out the object pointer.
2621 __ mov(c_rarg1, zr);
2622 } else {
2623 // Life is harder. The stack holds the value on top, followed by
2624 // the object. We don't know the size of the value, though; it
2625 // could be one or two words depending on its type. As a result,
2626 // we must find the type to determine where the object is.
2627 __ ldrw(c_rarg3, Address(c_rarg2,
2628 in_bytes(cp_base_offset +
2629 ConstantPoolCacheEntry::flags_offset())));
2630 __ lsr(c_rarg3, c_rarg3,
2631 ConstantPoolCacheEntry::tos_state_shift);
2632 ConstantPoolCacheEntry::verify_tos_state_shift();
2633 Label nope2, done, ok;
2634 __ ldr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2635 __ cmpw(c_rarg3, ltos);
2636 __ br(Assembler::EQ, ok);
2637 __ cmpw(c_rarg3, dtos);
2638 __ br(Assembler::NE, nope2);
2639 __ bind(ok);
2640 __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2641 __ bind(nope2);
2642 }
2643 // cache entry pointer
2644 __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2645 // object (tos)
2646 __ mov(c_rarg3, esp);
2647 // c_rarg1: object pointer set up above (NULL if static)
2648 // c_rarg2: cache entry pointer
2649 // c_rarg3: jvalue object on the stack
2650 __ call_VM(noreg,
2651 CAST_FROM_FN_PTR(address,
2652 InterpreterRuntime::post_field_modification),
2653 c_rarg1, c_rarg2, c_rarg3);
2654 __ get_cache_and_index_at_bcp(cache, index, 1);
2655 __ bind(L1);
2656 }
2657 }
2658
2659 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2660 transition(vtos, vtos);
2661
2662 const Register cache = r2;
2663 const Register index = r3;
2664 const Register obj = r2;
2665 const Register off = r19;
2666 const Register flags = r0;
2667 const Register bc = r4;
2668
2669 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2670 jvmti_post_field_mod(cache, index, is_static);
2671 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2672
2673 Label Done;
2674 __ mov(r5, flags);
2675
2676 {
2677 Label notVolatile;
2678 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2679 __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2680 __ bind(notVolatile);
2681 }
2682
2683 // field address
2684 const Address field(obj, off);
2685
2686 Label notByte, notBool, notInt, notShort, notChar,
2687 notLong, notFloat, notObj, notDouble;
2688
2689 // x86 uses a shift and mask or wings it with a shift plus assert
2690 // the mask is not needed. aarch64 just uses bitfield extract
2691 __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
2692
2693 assert(btos == 0, "change code, btos != 0");
2694 __ cbnz(flags, notByte);
2695
2696 // btos
2697 {
2698 __ pop(btos);
2699 if (!is_static) pop_and_check_object(obj);
2700 __ strb(r0, field);
2701 if (!is_static) {
2702 patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2703 }
2704 __ b(Done);
2705 }
2706
2707 __ bind(notByte);
2708 __ cmp(flags, ztos);
2709 __ br(Assembler::NE, notBool);
2710
2711 // ztos
2712 {
2713 __ pop(ztos);
2714 if (!is_static) pop_and_check_object(obj);
2715 __ andw(r0, r0, 0x1);
2716 __ strb(r0, field);
2717 if (!is_static) {
2718 patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2719 }
2720 __ b(Done);
2721 }
2722
2723 __ bind(notBool);
2724 __ cmp(flags, atos);
2725 __ br(Assembler::NE, notObj);
2726
2727 // atos
2728 {
2729 __ pop(atos);
2730 if (!is_static) pop_and_check_object(obj);
2731 // Store into the field
2732 do_oop_store(_masm, field, r0, _bs->kind(), false);
2733 if (!is_static) {
2734 patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2735 }
2736 __ b(Done);
2737 }
2738
2739 __ bind(notObj);
2740 __ cmp(flags, itos);
2741 __ br(Assembler::NE, notInt);
2742
2743 // itos
2744 {
2745 __ pop(itos);
2746 if (!is_static) pop_and_check_object(obj);
2747 __ strw(r0, field);
2748 if (!is_static) {
2749 patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2750 }
2751 __ b(Done);
2752 }
2753
2754 __ bind(notInt);
2755 __ cmp(flags, ctos);
2756 __ br(Assembler::NE, notChar);
2757
2758 // ctos
2759 {
2760 __ pop(ctos);
2761 if (!is_static) pop_and_check_object(obj);
2762 __ strh(r0, field);
2763 if (!is_static) {
2764 patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2765 }
2766 __ b(Done);
2767 }
2768
2769 __ bind(notChar);
2770 __ cmp(flags, stos);
2771 __ br(Assembler::NE, notShort);
2772
2773 // stos
2774 {
2775 __ pop(stos);
2776 if (!is_static) pop_and_check_object(obj);
2777 __ strh(r0, field);
2778 if (!is_static) {
2779 patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2780 }
2781 __ b(Done);
2782 }
2783
2784 __ bind(notShort);
2785 __ cmp(flags, ltos);
2786 __ br(Assembler::NE, notLong);
2787
2788 // ltos
2789 {
2790 __ pop(ltos);
2791 if (!is_static) pop_and_check_object(obj);
2792 __ str(r0, field);
2793 if (!is_static) {
2794 patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2795 }
2796 __ b(Done);
2797 }
2798
2799 __ bind(notLong);
2800 __ cmp(flags, ftos);
2801 __ br(Assembler::NE, notFloat);
2802
2803 // ftos
2804 {
2805 __ pop(ftos);
2806 if (!is_static) pop_and_check_object(obj);
2807 __ strs(v0, field);
2808 if (!is_static) {
2809 patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2810 }
2811 __ b(Done);
2812 }
2813
2814 __ bind(notFloat);
2815 #ifdef ASSERT
2816 __ cmp(flags, dtos);
2817 __ br(Assembler::NE, notDouble);
2818 #endif
2819
2820 // dtos
2821 {
2822 __ pop(dtos);
2823 if (!is_static) pop_and_check_object(obj);
2824 __ strd(v0, field);
2825 if (!is_static) {
2826 patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2827 }
2828 }
2829
2830 #ifdef ASSERT
2831 __ b(Done);
2832
2833 __ bind(notDouble);
2834 __ stop("Bad state");
2835 #endif
2836
2837 __ bind(Done);
2838
2839 {
2840 Label notVolatile;
2841 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2842 __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
2843 __ bind(notVolatile);
2844 }
2845 }
2846
2847 void TemplateTable::putfield(int byte_no)
2848 {
2849 putfield_or_static(byte_no, false);
2850 }
2851
2852 void TemplateTable::putstatic(int byte_no) {
2853 putfield_or_static(byte_no, true);
2854 }
2855
2856 void TemplateTable::jvmti_post_fast_field_mod()
2857 {
2858 if (JvmtiExport::can_post_field_modification()) {
2859 // Check to see if a field modification watch has been set before
2860 // we take the time to call into the VM.
2861 Label L2;
2862 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2863 __ ldrw(c_rarg3, Address(rscratch1));
2864 __ cbzw(c_rarg3, L2);
2865 __ pop_ptr(r19); // copy the object pointer from tos
2866 __ verify_oop(r19);
2867 __ push_ptr(r19); // put the object pointer back on tos
2868 // Save tos values before call_VM() clobbers them. Since we have
2869 // to do it for every data type, we use the saved values as the
2870 // jvalue object.
2871 switch (bytecode()) { // load values into the jvalue object
2872 case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2873 case Bytecodes::_fast_bputfield: // fall through
2874 case Bytecodes::_fast_zputfield: // fall through
2875 case Bytecodes::_fast_sputfield: // fall through
2876 case Bytecodes::_fast_cputfield: // fall through
2877 case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2878 case Bytecodes::_fast_dputfield: __ push_d(); break;
2879 case Bytecodes::_fast_fputfield: __ push_f(); break;
2880 case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2881
2882 default:
2883 ShouldNotReachHere();
2884 }
2885 __ mov(c_rarg3, esp); // points to jvalue on the stack
2886 // access constant pool cache entry
2887 __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2888 __ verify_oop(r19);
2889 // r19: object pointer copied above
2890 // c_rarg2: cache entry pointer
2891 // c_rarg3: jvalue object on the stack
2892 __ call_VM(noreg,
2893 CAST_FROM_FN_PTR(address,
2894 InterpreterRuntime::post_field_modification),
2895 r19, c_rarg2, c_rarg3);
2896
2897 switch (bytecode()) { // restore tos values
2898 case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2899 case Bytecodes::_fast_bputfield: // fall through
2900 case Bytecodes::_fast_zputfield: // fall through
2901 case Bytecodes::_fast_sputfield: // fall through
2902 case Bytecodes::_fast_cputfield: // fall through
2903 case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2904 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2905 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2906 case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2907 }
2908 __ bind(L2);
2909 }
2910 }
2911
2912 void TemplateTable::fast_storefield(TosState state)
2913 {
2914 transition(state, vtos);
2915
2916 ByteSize base = ConstantPoolCache::base_offset();
2917
2918 jvmti_post_fast_field_mod();
2919
2920 // access constant pool cache
2921 __ get_cache_and_index_at_bcp(r2, r1, 1);
2922
2923 // Must prevent reordering of the following cp cache loads with bytecode load
2924 __ membar(MacroAssembler::LoadLoad);
2925
2926 // test for volatile with r3
2927 __ ldrw(r3, Address(r2, in_bytes(base +
2928 ConstantPoolCacheEntry::flags_offset())));
2929
2930 // replace index with field offset from cache entry
2931 __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2932
2933 {
2934 Label notVolatile;
2935 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2936 __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2937 __ bind(notVolatile);
2938 }
2939
2940 Label notVolatile;
2941
2942 // Get object from stack
2943 pop_and_check_object(r2);
2944
2945 // field address
2946 const Address field(r2, r1);
2947
2948 // access field
2949 switch (bytecode()) {
2950 case Bytecodes::_fast_aputfield:
2951 do_oop_store(_masm, field, r0, _bs->kind(), false);
2952 break;
2953 case Bytecodes::_fast_lputfield:
2954 __ str(r0, field);
2955 break;
2956 case Bytecodes::_fast_iputfield:
2957 __ strw(r0, field);
2958 break;
2959 case Bytecodes::_fast_zputfield:
2960 __ andw(r0, r0, 0x1); // boolean is true if LSB is 1
2961 // fall through to bputfield
2962 case Bytecodes::_fast_bputfield:
2963 __ strb(r0, field);
2964 break;
2965 case Bytecodes::_fast_sputfield:
2966 // fall through
2967 case Bytecodes::_fast_cputfield:
2968 __ strh(r0, field);
2969 break;
2970 case Bytecodes::_fast_fputfield:
2971 __ strs(v0, field);
2972 break;
2973 case Bytecodes::_fast_dputfield:
2974 __ strd(v0, field);
2975 break;
2976 default:
2977 ShouldNotReachHere();
2978 }
2979
2980 {
2981 Label notVolatile;
2982 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2983 __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
2984 __ bind(notVolatile);
2985 }
2986 }
2987
2988
2989 void TemplateTable::fast_accessfield(TosState state)
2990 {
2991 transition(atos, state);
2992 // Do the JVMTI work here to avoid disturbing the register state below
2993 if (JvmtiExport::can_post_field_access()) {
2994 // Check to see if a field access watch has been set before we
2995 // take the time to call into the VM.
2996 Label L1;
2997 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2998 __ ldrw(r2, Address(rscratch1));
2999 __ cbzw(r2, L1);
3000 // access constant pool cache entry
3001 __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
3002 __ verify_oop(r0);
3003 __ push_ptr(r0); // save object pointer before call_VM() clobbers it
3004 __ mov(c_rarg1, r0);
3005 // c_rarg1: object pointer copied above
3006 // c_rarg2: cache entry pointer
3007 __ call_VM(noreg,
3008 CAST_FROM_FN_PTR(address,
3009 InterpreterRuntime::post_field_access),
3010 c_rarg1, c_rarg2);
3011 __ pop_ptr(r0); // restore object pointer
3012 __ bind(L1);
3013 }
3014
3015 // access constant pool cache
3016 __ get_cache_and_index_at_bcp(r2, r1, 1);
3017
3018 // Must prevent reordering of the following cp cache loads with bytecode load
3019 __ membar(MacroAssembler::LoadLoad);
3020
3021 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3022 ConstantPoolCacheEntry::f2_offset())));
3023 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3024 ConstantPoolCacheEntry::flags_offset())));
3025
3026 // r0: object
3027 __ verify_oop(r0);
3028 __ null_check(r0);
3029 const Address field(r0, r1);
3030
3031 // 8179954: We need to make sure that the code generated for
3032 // volatile accesses forms a sequentially-consistent set of
3033 // operations when combined with STLR and LDAR. Without a leading
3034 // membar it's possible for a simple Dekker test to fail if loads
3035 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
3036 // the stores in one method and we interpret the loads in another.
3037 if (! UseBarriersForVolatile) {
3038 Label notVolatile;
3039 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3040 __ membar(MacroAssembler::AnyAny);
3041 __ bind(notVolatile);
3042 }
3043
3044 // access field
3045 switch (bytecode()) {
3046 case Bytecodes::_fast_agetfield:
3047 __ load_heap_oop(r0, field);
3048 __ verify_oop(r0);
3049 break;
3050 case Bytecodes::_fast_lgetfield:
3051 __ ldr(r0, field);
3052 break;
3053 case Bytecodes::_fast_igetfield:
3054 __ ldrw(r0, field);
3055 break;
3056 case Bytecodes::_fast_bgetfield:
3057 __ load_signed_byte(r0, field);
3058 break;
3059 case Bytecodes::_fast_sgetfield:
3060 __ load_signed_short(r0, field);
3061 break;
3062 case Bytecodes::_fast_cgetfield:
3063 __ load_unsigned_short(r0, field);
3064 break;
3065 case Bytecodes::_fast_fgetfield:
3066 __ ldrs(v0, field);
3067 break;
3068 case Bytecodes::_fast_dgetfield:
3069 __ ldrd(v0, field);
3070 break;
3071 default:
3072 ShouldNotReachHere();
3073 }
3074 {
3075 Label notVolatile;
3076 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3077 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3078 __ bind(notVolatile);
3079 }
3080 }
3081
3082 void TemplateTable::fast_xaccess(TosState state)
3083 {
3084 transition(vtos, state);
3085
3086 // get receiver
3087 __ ldr(r0, aaddress(0));
3088 // access constant pool cache
3089 __ get_cache_and_index_at_bcp(r2, r3, 2);
3090 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3091 ConstantPoolCacheEntry::f2_offset())));
3092
3093 // 8179954: We need to make sure that the code generated for
3094 // volatile accesses forms a sequentially-consistent set of
3095 // operations when combined with STLR and LDAR. Without a leading
3096 // membar it's possible for a simple Dekker test to fail if loads
3097 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
3098 // the stores in one method and we interpret the loads in another.
3099 if (! UseBarriersForVolatile) {
3100 Label notVolatile;
3101 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3102 ConstantPoolCacheEntry::flags_offset())));
3103 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3104 __ membar(MacroAssembler::AnyAny);
3105 __ bind(notVolatile);
3106 }
3107
3108 // make sure exception is reported in correct bcp range (getfield is
3109 // next instruction)
3110 __ increment(rbcp);
3111 __ null_check(r0);
3112 switch (state) {
3113 case itos:
3114 __ ldrw(r0, Address(r0, r1, Address::lsl(0)));
3115 break;
3116 case atos:
3117 __ load_heap_oop(r0, Address(r0, r1, Address::lsl(0)));
3118 __ verify_oop(r0);
3119 break;
3120 case ftos:
3121 __ ldrs(v0, Address(r0, r1, Address::lsl(0)));
3122 break;
3123 default:
3124 ShouldNotReachHere();
3125 }
3126
3127 {
3128 Label notVolatile;
3129 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3130 ConstantPoolCacheEntry::flags_offset())));
3131 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3132 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3133 __ bind(notVolatile);
3134 }
3135
3136 __ decrement(rbcp);
3137 }
3138
3139
3140
3141 //-----------------------------------------------------------------------------
3142 // Calls
3143
3144 void TemplateTable::count_calls(Register method, Register temp)
3145 {
3146 __ call_Unimplemented();
3147 }
3148
3149 void TemplateTable::prepare_invoke(int byte_no,
3150 Register method, // linked method (or i-klass)
3151 Register index, // itable index, MethodType, etc.
3152 Register recv, // if caller wants to see it
3153 Register flags // if caller wants to test it
3154 ) {
3155 // determine flags
3156 Bytecodes::Code code = bytecode();
3157 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3158 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3159 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3160 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3161 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3162 const bool load_receiver = (recv != noreg);
3163 const bool save_flags = (flags != noreg);
3164 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3165 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3166 assert(flags == noreg || flags == r3, "");
3167 assert(recv == noreg || recv == r2, "");
3168
3169 // setup registers & access constant pool cache
3170 if (recv == noreg) recv = r2;
3171 if (flags == noreg) flags = r3;
3172 assert_different_registers(method, index, recv, flags);
3173
3174 // save 'interpreter return address'
3175 __ save_bcp();
3176
3177 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3178
3179 // maybe push appendix to arguments (just before return address)
3180 if (is_invokedynamic || is_invokehandle) {
3181 Label L_no_push;
3182 __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3183 // Push the appendix as a trailing parameter.
3184 // This must be done before we get the receiver,
3185 // since the parameter_size includes it.
3186 __ push(r19);
3187 __ mov(r19, index);
3188 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3189 __ load_resolved_reference_at_index(index, r19);
3190 __ pop(r19);
3191 __ push(index); // push appendix (MethodType, CallSite, etc.)
3192 __ bind(L_no_push);
3193 }
3194
3195 // load receiver if needed (note: no return address pushed yet)
3196 if (load_receiver) {
3197 __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3198 // FIXME -- is this actually correct? looks like it should be 2
3199 // const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3200 // const int receiver_is_at_end = -1; // back off one slot to get receiver
3201 // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3202 // __ movptr(recv, recv_addr);
3203 __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3204 __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3205 __ verify_oop(recv);
3206 }
3207
3208 // compute return type
3209 // x86 uses a shift and mask or wings it with a shift plus assert
3210 // the mask is not needed. aarch64 just uses bitfield extract
3211 __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
3212 // load return address
3213 {
3214 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3215 __ mov(rscratch1, table_addr);
3216 __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3217 }
3218 }
3219
3220
3221 void TemplateTable::invokevirtual_helper(Register index,
3222 Register recv,
3223 Register flags)
3224 {
3225 // Uses temporary registers r0, r3
3226 assert_different_registers(index, recv, r0, r3);
3227 // Test for an invoke of a final method
3228 Label notFinal;
3229 __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3230
3231 const Register method = index; // method must be rmethod
3232 assert(method == rmethod,
3233 "methodOop must be rmethod for interpreter calling convention");
3234
3235 // do the call - the index is actually the method to call
3236 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3237
3238 // It's final, need a null check here!
3239 __ null_check(recv);
3240
3241 // profile this call
3242 __ profile_final_call(r0);
3243 __ profile_arguments_type(r0, method, r4, true);
3244
3245 __ jump_from_interpreted(method, r0);
3246
3247 __ bind(notFinal);
3248
3249 // get receiver klass
3250 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3251 __ load_klass(r0, recv);
3252
3253 // profile this call
3254 __ profile_virtual_call(r0, rlocals, r3);
3255
3256 // get target methodOop & entry point
3257 __ lookup_virtual_method(r0, index, method);
3258 __ profile_arguments_type(r3, method, r4, true);
3259 // FIXME -- this looks completely redundant. is it?
3260 // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3261 __ jump_from_interpreted(method, r3);
3262 }
3263
3264 void TemplateTable::invokevirtual(int byte_no)
3265 {
3266 transition(vtos, vtos);
3267 assert(byte_no == f2_byte, "use this argument");
3268
3269 prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3270
3271 // rmethod: index (actually a Method*)
3272 // r2: receiver
3273 // r3: flags
3274
3275 invokevirtual_helper(rmethod, r2, r3);
3276 }
3277
3278 void TemplateTable::invokespecial(int byte_no)
3279 {
3280 transition(vtos, vtos);
3281 assert(byte_no == f1_byte, "use this argument");
3282
3283 prepare_invoke(byte_no, rmethod, noreg, // get f1 Method*
3284 r2); // get receiver also for null check
3285 __ verify_oop(r2);
3286 __ null_check(r2);
3287 // do the call
3288 __ profile_call(r0);
3289 __ profile_arguments_type(r0, rmethod, rbcp, false);
3290 __ jump_from_interpreted(rmethod, r0);
3291 }
3292
3293 void TemplateTable::invokestatic(int byte_no)
3294 {
3295 transition(vtos, vtos);
3296 assert(byte_no == f1_byte, "use this argument");
3297
3298 prepare_invoke(byte_no, rmethod); // get f1 Method*
3299 // do the call
3300 __ profile_call(r0);
3301 __ profile_arguments_type(r0, rmethod, r4, false);
3302 __ jump_from_interpreted(rmethod, r0);
3303 }
3304
3305 void TemplateTable::fast_invokevfinal(int byte_no)
3306 {
3307 __ call_Unimplemented();
3308 }
3309
3310 void TemplateTable::invokeinterface(int byte_no) {
3311 transition(vtos, vtos);
3312 assert(byte_no == f1_byte, "use this argument");
3313
3314 prepare_invoke(byte_no, r0, rmethod, // get f1 Klass*, f2 Method*
3315 r2, r3); // recv, flags
3316
3317 // r0: interface klass (from f1)
3318 // rmethod: method (from f2)
3319 // r2: receiver
3320 // r3: flags
3321
3322 // Special case of invokeinterface called for virtual method of
3323 // java.lang.Object. See cpCacheOop.cpp for details.
3324 // This code isn't produced by javac, but could be produced by
3325 // another compliant java compiler.
3326 Label notMethod;
3327 __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notMethod);
3328
3329 invokevirtual_helper(rmethod, r2, r3);
3330 __ bind(notMethod);
3331
3332 // Get receiver klass into r3 - also a null check
3333 __ restore_locals();
3334 __ null_check(r2, oopDesc::klass_offset_in_bytes());
3335 __ load_klass(r3, r2);
3336
3337 Label no_such_interface, no_such_method;
3338
3339 // Receiver subtype check against REFC.
3340 // Superklass in r0. Subklass in r3. Blows rscratch2, r13.
3341 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3342 r3, r0, noreg,
3343 // outputs: scan temp. reg, scan temp. reg
3344 rscratch2, r13,
3345 no_such_interface,
3346 /*return_method=*/false);
3347
3348 // profile this call
3349 __ profile_virtual_call(r3, r13, r19);
3350
3351 // Get declaring interface class from method, and itable index
3352 __ ldr(r0, Address(rmethod, Method::const_offset()));
3353 __ ldr(r0, Address(r0, ConstMethod::constants_offset()));
3354 __ ldr(r0, Address(r0, ConstantPool::pool_holder_offset_in_bytes()));
3355 __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
3356 __ subw(rmethod, rmethod, Method::itable_index_max);
3357 __ negw(rmethod, rmethod);
3358
3359 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3360 r3, r0, rmethod,
3361 // outputs: method, scan temp. reg
3362 rmethod, r13,
3363 no_such_interface);
3364
3365 // rmethod,: methodOop to call
3366 // r2: receiver
3367 // Check for abstract method error
3368 // Note: This should be done more efficiently via a throw_abstract_method_error
3369 // interpreter entry point and a conditional jump to it in case of a null
3370 // method.
3371 __ cbz(rmethod, no_such_method);
3372
3373 __ profile_arguments_type(r3, rmethod, r13, true);
3374
3375 // do the call
3376 // r2: receiver
3377 // rmethod,: methodOop
3378 __ jump_from_interpreted(rmethod, r3);
3379 __ should_not_reach_here();
3380
3381 // exception handling code follows...
3382 // note: must restore interpreter registers to canonical
3383 // state for exception handling to work correctly!
3384
3385 __ bind(no_such_method);
3386 // throw exception
3387 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3388 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3389 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3390 // the call_VM checks for exception, so we should never return here.
3391 __ should_not_reach_here();
3392
3393 __ bind(no_such_interface);
3394 // throw exception
3395 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3396 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3397 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3398 InterpreterRuntime::throw_IncompatibleClassChangeError));
3399 // the call_VM checks for exception, so we should never return here.
3400 __ should_not_reach_here();
3401 return;
3402 }
3403
3404 void TemplateTable::invokehandle(int byte_no) {
3405 transition(vtos, vtos);
3406 assert(byte_no == f1_byte, "use this argument");
3407
3408 if (!EnableInvokeDynamic) {
3409 // rewriter does not generate this bytecode
3410 __ should_not_reach_here();
3411 return;
3412 }
3413
3414 prepare_invoke(byte_no, rmethod, r0, r2);
3415 __ verify_method_ptr(r2);
3416 __ verify_oop(r2);
3417 __ null_check(r2);
3418
3419 // FIXME: profile the LambdaForm also
3420
3421 // r13 is safe to use here as a scratch reg because it is about to
3422 // be clobbered by jump_from_interpreted().
3423 __ profile_final_call(r13);
3424 __ profile_arguments_type(r13, rmethod, r4, true);
3425
3426 __ jump_from_interpreted(rmethod, r0);
3427 }
3428
3429 void TemplateTable::invokedynamic(int byte_no) {
3430 transition(vtos, vtos);
3431 assert(byte_no == f1_byte, "use this argument");
3432
3433 if (!EnableInvokeDynamic) {
3434 // We should not encounter this bytecode if !EnableInvokeDynamic.
3435 // The verifier will stop it. However, if we get past the verifier,
3436 // this will stop the thread in a reasonable way, without crashing the JVM.
3437 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3438 InterpreterRuntime::throw_IncompatibleClassChangeError));
3439 // the call_VM checks for exception, so we should never return here.
3440 __ should_not_reach_here();
3441 return;
3442 }
3443
3444 prepare_invoke(byte_no, rmethod, r0);
3445
3446 // r0: CallSite object (from cpool->resolved_references[])
3447 // rmethod: MH.linkToCallSite method (from f2)
3448
3449 // Note: r0_callsite is already pushed by prepare_invoke
3450
3451 // %%% should make a type profile for any invokedynamic that takes a ref argument
3452 // profile this call
3453 __ profile_call(rbcp);
3454 __ profile_arguments_type(r3, rmethod, r13, false);
3455
3456 __ verify_oop(r0);
3457
3458 __ jump_from_interpreted(rmethod, r0);
3459 }
3460
3461
3462 //-----------------------------------------------------------------------------
3463 // Allocation
3464
3465 void TemplateTable::_new() {
3466 transition(vtos, atos);
3467
3468 __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3469 Label slow_case;
3470 Label done;
3471 Label initialize_header;
3472 Label initialize_object; // including clearing the fields
3473 Label allocate_shared;
3474
3475 __ get_cpool_and_tags(r4, r0);
3476 // Make sure the class we're about to instantiate has been resolved.
3477 // This is done before loading InstanceKlass to be consistent with the order
3478 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3479 const int tags_offset = Array<u1>::base_offset_in_bytes();
3480 __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3481 __ lea(rscratch1, Address(rscratch1, tags_offset));
3482 __ ldarb(rscratch1, rscratch1);
3483 __ cmp(rscratch1, JVM_CONSTANT_Class);
3484 __ br(Assembler::NE, slow_case);
3485
3486 // get InstanceKlass
3487 __ lea(r4, Address(r4, r3, Address::lsl(3)));
3488 __ ldr(r4, Address(r4, sizeof(ConstantPool)));
3489
3490 // make sure klass is initialized & doesn't have finalizer
3491 // make sure klass is fully initialized
3492 __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3493 __ cmp(rscratch1, InstanceKlass::fully_initialized);
3494 __ br(Assembler::NE, slow_case);
3495
3496 // get instance_size in InstanceKlass (scaled to a count of bytes)
3497 __ ldrw(r3,
3498 Address(r4,
3499 Klass::layout_helper_offset()));
3500 // test to see if it has a finalizer or is malformed in some way
3501 __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3502
3503 // Allocate the instance
3504 // 1) Try to allocate in the TLAB
3505 // 2) if fail and the object is large allocate in the shared Eden
3506 // 3) if the above fails (or is not applicable), go to a slow case
3507 // (creates a new TLAB, etc.)
3508
3509 const bool allow_shared_alloc =
3510 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3511
3512 if (UseTLAB) {
3513 __ tlab_allocate(r0, r3, 0, noreg, r1,
3514 allow_shared_alloc ? allocate_shared : slow_case);
3515
3516 if (ZeroTLAB) {
3517 // the fields have been already cleared
3518 __ b(initialize_header);
3519 } else {
3520 // initialize both the header and fields
3521 __ b(initialize_object);
3522 }
3523 }
3524
3525 // Allocation in the shared Eden, if allowed.
3526 //
3527 // r3: instance size in bytes
3528 if (allow_shared_alloc) {
3529 __ bind(allocate_shared);
3530
3531 __ eden_allocate(r0, r3, 0, r10, slow_case);
3532 __ incr_allocated_bytes(rthread, r3, 0, rscratch1);
3533 }
3534
3535 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3536 // The object is initialized before the header. If the object size is
3537 // zero, go directly to the header initialization.
3538 __ bind(initialize_object);
3539 __ sub(r3, r3, sizeof(oopDesc));
3540 __ cbz(r3, initialize_header);
3541
3542 // Initialize object fields
3543 {
3544 __ add(r2, r0, sizeof(oopDesc));
3545 Label loop;
3546 __ bind(loop);
3547 __ str(zr, Address(__ post(r2, BytesPerLong)));
3548 __ sub(r3, r3, BytesPerLong);
3549 __ cbnz(r3, loop);
3550 }
3551
3552 // initialize object header only.
3553 __ bind(initialize_header);
3554 if (UseBiasedLocking) {
3555 __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3556 } else {
3557 __ mov(rscratch1, (intptr_t)markOopDesc::prototype());
3558 }
3559 __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3560 __ store_klass_gap(r0, zr); // zero klass gap for compressed oops
3561 __ store_klass(r0, r4); // store klass last
3562
3563 {
3564 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3565 // Trigger dtrace event for fastpath
3566 __ push(atos); // save the return value
3567 __ call_VM_leaf(
3568 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3569 __ pop(atos); // restore the return value
3570
3571 }
3572 __ b(done);
3573 }
3574
3575 // slow case
3576 __ bind(slow_case);
3577 __ get_constant_pool(c_rarg1);
3578 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3579 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3580 __ verify_oop(r0);
3581
3582 // continue
3583 __ bind(done);
3584 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3585 __ membar(Assembler::StoreStore);
3586 }
3587
3588 void TemplateTable::newarray() {
3589 transition(itos, atos);
3590 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3591 __ mov(c_rarg2, r0);
3592 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3593 c_rarg1, c_rarg2);
3594 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3595 __ membar(Assembler::StoreStore);
3596 }
3597
3598 void TemplateTable::anewarray() {
3599 transition(itos, atos);
3600 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3601 __ get_constant_pool(c_rarg1);
3602 __ mov(c_rarg3, r0);
3603 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3604 c_rarg1, c_rarg2, c_rarg3);
3605 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3606 __ membar(Assembler::StoreStore);
3607 }
3608
3609 void TemplateTable::arraylength() {
3610 transition(atos, itos);
3611 __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3612 __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3613 }
3614
3615 void TemplateTable::checkcast()
3616 {
3617 transition(atos, atos);
3618 Label done, is_null, ok_is_subtype, quicked, resolved;
3619 __ cbz(r0, is_null);
3620
3621 // Get cpool & tags index
3622 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3623 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3624 // See if bytecode has already been quicked
3625 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3626 __ lea(r1, Address(rscratch1, r19));
3627 __ ldarb(r1, r1);
3628 __ cmp(r1, JVM_CONSTANT_Class);
3629 __ br(Assembler::EQ, quicked);
3630
3631 __ push(atos); // save receiver for result, and for GC
3632 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3633 // vm_result_2 has metadata result
3634 __ get_vm_result_2(r0, rthread);
3635 __ pop(r3); // restore receiver
3636 __ b(resolved);
3637
3638 // Get superklass in r0 and subklass in r3
3639 __ bind(quicked);
3640 __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3641 __ lea(r0, Address(r2, r19, Address::lsl(3)));
3642 __ ldr(r0, Address(r0, sizeof(ConstantPool)));
3643
3644 __ bind(resolved);
3645 __ load_klass(r19, r3);
3646
3647 // Generate subtype check. Blows r2, r5. Object in r3.
3648 // Superklass in r0. Subklass in r19.
3649 __ gen_subtype_check(r19, ok_is_subtype);
3650
3651 // Come here on failure
3652 __ push(r3);
3653 // object is at TOS
3654 __ b(Interpreter::_throw_ClassCastException_entry);
3655
3656 // Come here on success
3657 __ bind(ok_is_subtype);
3658 __ mov(r0, r3); // Restore object in r3
3659
3660 // Collect counts on whether this test sees NULLs a lot or not.
3661 if (ProfileInterpreter) {
3662 __ b(done);
3663 __ bind(is_null);
3664 __ profile_null_seen(r2);
3665 } else {
3666 __ bind(is_null); // same as 'done'
3667 }
3668 __ bind(done);
3669 }
3670
3671 void TemplateTable::instanceof() {
3672 transition(atos, itos);
3673 Label done, is_null, ok_is_subtype, quicked, resolved;
3674 __ cbz(r0, is_null);
3675
3676 // Get cpool & tags index
3677 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3678 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3679 // See if bytecode has already been quicked
3680 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3681 __ lea(r1, Address(rscratch1, r19));
3682 __ ldarb(r1, r1);
3683 __ cmp(r1, JVM_CONSTANT_Class);
3684 __ br(Assembler::EQ, quicked);
3685
3686 __ push(atos); // save receiver for result, and for GC
3687 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3688 // vm_result_2 has metadata result
3689 __ get_vm_result_2(r0, rthread);
3690 __ pop(r3); // restore receiver
3691 __ verify_oop(r3);
3692 __ load_klass(r3, r3);
3693 __ b(resolved);
3694
3695 // Get superklass in r0 and subklass in r3
3696 __ bind(quicked);
3697 __ load_klass(r3, r0);
3698 __ lea(r0, Address(r2, r19, Address::lsl(3)));
3699 __ ldr(r0, Address(r0, sizeof(ConstantPool)));
3700
3701 __ bind(resolved);
3702
3703 // Generate subtype check. Blows r2, r5
3704 // Superklass in r0. Subklass in r3.
3705 __ gen_subtype_check(r3, ok_is_subtype);
3706
3707 // Come here on failure
3708 __ mov(r0, 0);
3709 __ b(done);
3710 // Come here on success
3711 __ bind(ok_is_subtype);
3712 __ mov(r0, 1);
3713
3714 // Collect counts on whether this test sees NULLs a lot or not.
3715 if (ProfileInterpreter) {
3716 __ b(done);
3717 __ bind(is_null);
3718 __ profile_null_seen(r2);
3719 } else {
3720 __ bind(is_null); // same as 'done'
3721 }
3722 __ bind(done);
3723 // r0 = 0: obj == NULL or obj is not an instanceof the specified klass
3724 // r0 = 1: obj != NULL and obj is an instanceof the specified klass
3725 }
3726
3727 //-----------------------------------------------------------------------------
3728 // Breakpoints
3729 void TemplateTable::_breakpoint() {
3730 // Note: We get here even if we are single stepping..
3731 // jbug inists on setting breakpoints at every bytecode
3732 // even if we are in single step mode.
3733
3734 transition(vtos, vtos);
3735
3736 // get the unpatched byte code
3737 __ get_method(c_rarg1);
3738 __ call_VM(noreg,
3739 CAST_FROM_FN_PTR(address,
3740 InterpreterRuntime::get_original_bytecode_at),
3741 c_rarg1, rbcp);
3742 __ mov(r19, r0);
3743
3744 // post the breakpoint event
3745 __ call_VM(noreg,
3746 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3747 rmethod, rbcp);
3748
3749 // complete the execution of original bytecode
3750 __ mov(rscratch1, r19);
3751 __ dispatch_only_normal(vtos);
3752 }
3753
3754 //-----------------------------------------------------------------------------
3755 // Exceptions
3756
3757 void TemplateTable::athrow() {
3758 transition(atos, vtos);
3759 __ null_check(r0);
3760 __ b(Interpreter::throw_exception_entry());
3761 }
3762
3763 //-----------------------------------------------------------------------------
3764 // Synchronization
3765 //
3766 // Note: monitorenter & exit are symmetric routines; which is reflected
3767 // in the assembly code structure as well
3768 //
3769 // Stack layout:
3770 //
3771 // [expressions ] <--- esp = expression stack top
3772 // ..
3773 // [expressions ]
3774 // [monitor entry] <--- monitor block top = expression stack bot
3775 // ..
3776 // [monitor entry]
3777 // [frame data ] <--- monitor block bot
3778 // ...
3779 // [saved rbp ] <--- rbp
3780 void TemplateTable::monitorenter()
3781 {
3782 transition(atos, vtos);
3783
3784 // check for NULL object
3785 __ null_check(r0);
3786
3787 const Address monitor_block_top(
3788 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3789 const Address monitor_block_bot(
3790 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3791 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3792
3793 Label allocated;
3794
3795 // initialize entry pointer
3796 __ mov(c_rarg1, zr); // points to free slot or NULL
3797
3798 // find a free slot in the monitor block (result in c_rarg1)
3799 {
3800 Label entry, loop, exit;
3801 __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3802 // starting with top-most entry
3803 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3804
3805 __ b(entry);
3806
3807 __ bind(loop);
3808 // check if current entry is used
3809 // if not used then remember entry in c_rarg1
3810 __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3811 __ cmp(zr, rscratch1);
3812 __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3813 // check if current entry is for same object
3814 __ cmp(r0, rscratch1);
3815 // if same object then stop searching
3816 __ br(Assembler::EQ, exit);
3817 // otherwise advance to next entry
3818 __ add(c_rarg3, c_rarg3, entry_size);
3819 __ bind(entry);
3820 // check if bottom reached
3821 __ cmp(c_rarg3, c_rarg2);
3822 // if not at bottom then check this entry
3823 __ br(Assembler::NE, loop);
3824 __ bind(exit);
3825 }
3826
3827 __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3828 // if found, continue with that on
3829
3830 // allocate one if there's no free slot
3831 {
3832 Label entry, loop;
3833 // 1. compute new pointers // rsp: old expression stack top
3834 __ ldr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
3835 __ sub(esp, esp, entry_size); // move expression stack top
3836 __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3837 __ mov(c_rarg3, esp); // set start value for copy loop
3838 __ str(c_rarg1, monitor_block_bot); // set new monitor block bottom
3839
3840 __ sub(sp, sp, entry_size); // make room for the monitor
3841
3842 __ b(entry);
3843 // 2. move expression stack contents
3844 __ bind(loop);
3845 __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3846 // word from old location
3847 __ str(c_rarg2, Address(c_rarg3, 0)); // and store it at new location
3848 __ add(c_rarg3, c_rarg3, wordSize); // advance to next word
3849 __ bind(entry);
3850 __ cmp(c_rarg3, c_rarg1); // check if bottom reached
3851 __ br(Assembler::NE, loop); // if not at bottom then
3852 // copy next word
3853 }
3854
3855 // call run-time routine
3856 // c_rarg1: points to monitor entry
3857 __ bind(allocated);
3858
3859 // Increment bcp to point to the next bytecode, so exception
3860 // handling for async. exceptions work correctly.
3861 // The object has already been poped from the stack, so the
3862 // expression stack looks correct.
3863 __ increment(rbcp);
3864
3865 // store object
3866 __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3867 __ lock_object(c_rarg1);
3868
3869 // check to make sure this monitor doesn't cause stack overflow after locking
3870 __ save_bcp(); // in case of exception
3871 __ generate_stack_overflow_check(0);
3872
3873 // The bcp has already been incremented. Just need to dispatch to
3874 // next instruction.
3875 __ dispatch_next(vtos);
3876 }
3877
3878
3879 void TemplateTable::monitorexit()
3880 {
3881 transition(atos, vtos);
3882
3883 // check for NULL object
3884 __ null_check(r0);
3885
3886 const Address monitor_block_top(
3887 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3888 const Address monitor_block_bot(
3889 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3890 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3891
3892 Label found;
3893
3894 // find matching slot
3895 {
3896 Label entry, loop;
3897 __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3898 // starting with top-most entry
3899 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3900 // of monitor block
3901 __ b(entry);
3902
3903 __ bind(loop);
3904 // check if current entry is for same object
3905 __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3906 __ cmp(r0, rscratch1);
3907 // if same object then stop searching
3908 __ br(Assembler::EQ, found);
3909 // otherwise advance to next entry
3910 __ add(c_rarg1, c_rarg1, entry_size);
3911 __ bind(entry);
3912 // check if bottom reached
3913 __ cmp(c_rarg1, c_rarg2);
3914 // if not at bottom then check this entry
3915 __ br(Assembler::NE, loop);
3916 }
3917
3918 // error handling. Unlocking was not block-structured
3919 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3920 InterpreterRuntime::throw_illegal_monitor_state_exception));
3921 __ should_not_reach_here();
3922
3923 // call run-time routine
3924 __ bind(found);
3925 __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3926 __ unlock_object(c_rarg1);
3927 __ pop_ptr(r0); // discard object
3928 }
3929
3930
3931 // Wide instructions
3932 void TemplateTable::wide()
3933 {
3934 __ load_unsigned_byte(r19, at_bcp(1));
3935 __ mov(rscratch1, (address)Interpreter::_wentry_point);
3936 __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
3937 __ br(rscratch1);
3938 }
3939
3940
3941 // Multi arrays
3942 void TemplateTable::multianewarray() {
3943 transition(vtos, atos);
3944 __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
3945 // last dim is on top of stack; we want address of first one:
3946 // first_addr = last_addr + (ndims - 1) * wordSize
3947 __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
3948 __ sub(c_rarg1, c_rarg1, wordSize);
3949 call_VM(r0,
3950 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3951 c_rarg1);
3952 __ load_unsigned_byte(r1, at_bcp(3));
3953 __ lea(esp, Address(esp, r1, Address::uxtw(3)));
3954 }
3955 #endif // !CC_INTERP