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