1 /*
2 * Copyright (c) 2003, 2019, Oracle and/or its affiliates. All rights reserved.
3 * Copyright (c) 2014, Red Hat Inc. All rights reserved.
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #include "precompiled.hpp"
27 #include "asm/macroAssembler.inline.hpp"
28 #include "gc/shared/barrierSetAssembler.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/interp_masm.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "memory/universe.hpp"
34 #include "oops/methodData.hpp"
35 #include "oops/method.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "prims/methodHandles.hpp"
39 #include "runtime/frame.inline.hpp"
40 #include "runtime/sharedRuntime.hpp"
41 #include "runtime/stubRoutines.hpp"
42 #include "runtime/synchronizer.hpp"
43
44 #define __ _masm->
45
46 // Platform-dependent initialization
47
48 void TemplateTable::pd_initialize() {
49 // No aarch64 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 dst,
147 Register val,
148 DecoratorSet decorators) {
149 assert(val == noreg || val == r0, "parameter is just for looks");
150 __ store_heap_oop(dst, val, r10, r1, noreg, decorators);
151 }
152
153 static void do_oop_load(InterpreterMacroAssembler* _masm,
154 Address src,
155 Register dst,
156 DecoratorSet decorators) {
157 __ load_heap_oop(dst, src, r10, r1, decorators);
158 }
159
160 Address TemplateTable::at_bcp(int offset) {
161 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
162 return Address(rbcp, offset);
163 }
164
165 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
166 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
167 int byte_no)
168 {
169 if (!RewriteBytecodes) return;
170 Label L_patch_done;
171
172 switch (bc) {
173 case Bytecodes::_fast_qputfield:
174 case Bytecodes::_fast_aputfield:
175 case Bytecodes::_fast_bputfield:
176 case Bytecodes::_fast_zputfield:
177 case Bytecodes::_fast_cputfield:
178 case Bytecodes::_fast_dputfield:
179 case Bytecodes::_fast_fputfield:
180 case Bytecodes::_fast_iputfield:
181 case Bytecodes::_fast_lputfield:
182 case Bytecodes::_fast_sputfield:
183 {
184 // We skip bytecode quickening for putfield instructions when
185 // the put_code written to the constant pool cache is zero.
186 // This is required so that every execution of this instruction
187 // calls out to InterpreterRuntime::resolve_get_put to do
188 // additional, required work.
189 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
190 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
191 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
192 __ movw(bc_reg, bc);
193 __ cbzw(temp_reg, L_patch_done); // don't patch
194 }
195 break;
196 default:
197 assert(byte_no == -1, "sanity");
198 // the pair bytecodes have already done the load.
199 if (load_bc_into_bc_reg) {
200 __ movw(bc_reg, bc);
201 }
202 }
203
204 if (JvmtiExport::can_post_breakpoint()) {
205 Label L_fast_patch;
206 // if a breakpoint is present we can't rewrite the stream directly
207 __ load_unsigned_byte(temp_reg, at_bcp(0));
208 __ cmpw(temp_reg, Bytecodes::_breakpoint);
209 __ br(Assembler::NE, L_fast_patch);
210 // Let breakpoint table handling rewrite to quicker bytecode
211 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
212 __ b(L_patch_done);
213 __ bind(L_fast_patch);
214 }
215
216 #ifdef ASSERT
217 Label L_okay;
218 __ load_unsigned_byte(temp_reg, at_bcp(0));
219 __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
220 __ br(Assembler::EQ, L_okay);
221 __ cmpw(temp_reg, bc_reg);
222 __ br(Assembler::EQ, L_okay);
223 __ stop("patching the wrong bytecode");
224 __ bind(L_okay);
225 #endif
226
227 // patch bytecode
228 __ strb(bc_reg, at_bcp(0));
229 __ bind(L_patch_done);
230 }
231
232
233 // Individual instructions
234
235 void TemplateTable::nop() {
236 transition(vtos, vtos);
237 // nothing to do
238 }
239
240 void TemplateTable::shouldnotreachhere() {
241 transition(vtos, vtos);
242 __ stop("shouldnotreachhere bytecode");
243 }
244
245 void TemplateTable::aconst_null()
246 {
247 transition(vtos, atos);
248 __ mov(r0, 0);
249 }
250
251 void TemplateTable::iconst(int value)
252 {
253 transition(vtos, itos);
254 __ mov(r0, value);
255 }
256
257 void TemplateTable::lconst(int value)
258 {
259 __ mov(r0, value);
260 }
261
262 void TemplateTable::fconst(int value)
263 {
264 transition(vtos, ftos);
265 switch (value) {
266 case 0:
267 __ fmovs(v0, zr);
268 break;
269 case 1:
270 __ fmovs(v0, 1.0);
271 break;
272 case 2:
273 __ fmovs(v0, 2.0);
274 break;
275 default:
276 ShouldNotReachHere();
277 break;
278 }
279 }
280
281 void TemplateTable::dconst(int value)
282 {
283 transition(vtos, dtos);
284 switch (value) {
285 case 0:
286 __ fmovd(v0, zr);
287 break;
288 case 1:
289 __ fmovd(v0, 1.0);
290 break;
291 case 2:
292 __ fmovd(v0, 2.0);
293 break;
294 default:
295 ShouldNotReachHere();
296 break;
297 }
298 }
299
300 void TemplateTable::bipush()
301 {
302 transition(vtos, itos);
303 __ load_signed_byte32(r0, at_bcp(1));
304 }
305
306 void TemplateTable::sipush()
307 {
308 transition(vtos, itos);
309 __ load_unsigned_short(r0, at_bcp(1));
310 __ revw(r0, r0);
311 __ asrw(r0, r0, 16);
312 }
313
314 void TemplateTable::ldc(bool wide)
315 {
316 transition(vtos, vtos);
317 Label call_ldc, notFloat, notClass, notInt, Done;
318
319 if (wide) {
320 __ get_unsigned_2_byte_index_at_bcp(r1, 1);
321 } else {
322 __ load_unsigned_byte(r1, at_bcp(1));
323 }
324 __ get_cpool_and_tags(r2, r0);
325
326 const int base_offset = ConstantPool::header_size() * wordSize;
327 const int tags_offset = Array<u1>::base_offset_in_bytes();
328
329 // get type
330 __ add(r3, r1, tags_offset);
331 __ lea(r3, Address(r0, r3));
332 __ ldarb(r3, r3);
333
334 // unresolved class - get the resolved class
335 __ cmp(r3, (u1)JVM_CONSTANT_UnresolvedClass);
336 __ br(Assembler::EQ, call_ldc);
337
338 // unresolved class in error state - call into runtime to throw the error
339 // from the first resolution attempt
340 __ cmp(r3, (u1)JVM_CONSTANT_UnresolvedClassInError);
341 __ br(Assembler::EQ, call_ldc);
342
343 // resolved class - need to call vm to get java mirror of the class
344 __ cmp(r3, (u1)JVM_CONSTANT_Class);
345 __ br(Assembler::NE, notClass);
346
347 __ bind(call_ldc);
348 __ mov(c_rarg1, wide);
349 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
350 __ push_ptr(r0);
351 __ verify_oop(r0);
352 __ b(Done);
353
354 __ bind(notClass);
355 __ cmp(r3, (u1)JVM_CONSTANT_Float);
356 __ br(Assembler::NE, notFloat);
357 // ftos
358 __ adds(r1, r2, r1, Assembler::LSL, 3);
359 __ ldrs(v0, Address(r1, base_offset));
360 __ push_f();
361 __ b(Done);
362
363 __ bind(notFloat);
364
365 __ cmp(r3, (u1)JVM_CONSTANT_Integer);
366 __ br(Assembler::NE, notInt);
367
368 // itos
369 __ adds(r1, r2, r1, Assembler::LSL, 3);
370 __ ldrw(r0, Address(r1, base_offset));
371 __ push_i(r0);
372 __ b(Done);
373
374 __ bind(notInt);
375 condy_helper(Done);
376
377 __ bind(Done);
378 }
379
380 // Fast path for caching oop constants.
381 void TemplateTable::fast_aldc(bool wide)
382 {
383 transition(vtos, atos);
384
385 Register result = r0;
386 Register tmp = r1;
387 Register rarg = r2;
388
389 int index_size = wide ? sizeof(u2) : sizeof(u1);
390
391 Label resolved;
392
393 // We are resolved if the resolved reference cache entry contains a
394 // non-null object (String, MethodType, etc.)
395 assert_different_registers(result, tmp);
396 __ get_cache_index_at_bcp(tmp, 1, index_size);
397 __ load_resolved_reference_at_index(result, tmp);
398 __ cbnz(result, resolved);
399
400 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
401
402 // first time invocation - must resolve first
403 __ mov(rarg, (int)bytecode());
404 __ call_VM(result, entry, rarg);
405
406 __ bind(resolved);
407
408 { // Check for the null sentinel.
409 // If we just called the VM, it already did the mapping for us,
410 // but it's harmless to retry.
411 Label notNull;
412
413 // Stash null_sentinel address to get its value later
414 __ movptr(rarg, (uintptr_t)Universe::the_null_sentinel_addr());
415 __ ldr(tmp, Address(rarg));
416 __ cmpoop(result, tmp);
417 __ br(Assembler::NE, notNull);
418 __ mov(result, 0); // NULL object reference
419 __ bind(notNull);
420 }
421
422 if (VerifyOops) {
423 // Safe to call with 0 result
424 __ verify_oop(result);
425 }
426 }
427
428 void TemplateTable::ldc2_w()
429 {
430 transition(vtos, vtos);
431 Label notDouble, notLong, Done;
432 __ get_unsigned_2_byte_index_at_bcp(r0, 1);
433
434 __ get_cpool_and_tags(r1, r2);
435 const int base_offset = ConstantPool::header_size() * wordSize;
436 const int tags_offset = Array<u1>::base_offset_in_bytes();
437
438 // get type
439 __ lea(r2, Address(r2, r0, Address::lsl(0)));
440 __ load_unsigned_byte(r2, Address(r2, tags_offset));
441 __ cmpw(r2, (int)JVM_CONSTANT_Double);
442 __ br(Assembler::NE, notDouble);
443
444 // dtos
445 __ lea (r2, Address(r1, r0, Address::lsl(3)));
446 __ ldrd(v0, Address(r2, base_offset));
447 __ push_d();
448 __ b(Done);
449
450 __ bind(notDouble);
451 __ cmpw(r2, (int)JVM_CONSTANT_Long);
452 __ br(Assembler::NE, notLong);
453
454 // ltos
455 __ lea(r0, Address(r1, r0, Address::lsl(3)));
456 __ ldr(r0, Address(r0, base_offset));
457 __ push_l();
458 __ b(Done);
459
460 __ bind(notLong);
461 condy_helper(Done);
462
463 __ bind(Done);
464 }
465
466 void TemplateTable::condy_helper(Label& Done)
467 {
468 Register obj = r0;
469 Register rarg = r1;
470 Register flags = r2;
471 Register off = r3;
472
473 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
474
475 __ mov(rarg, (int) bytecode());
476 __ call_VM(obj, entry, rarg);
477
478 __ get_vm_result_2(flags, rthread);
479
480 // VMr = obj = base address to find primitive value to push
481 // VMr2 = flags = (tos, off) using format of CPCE::_flags
482 __ mov(off, flags);
483 __ andw(off, off, ConstantPoolCacheEntry::field_index_mask);
484
485 const Address field(obj, off);
486
487 // What sort of thing are we loading?
488 // x86 uses a shift and mask or wings it with a shift plus assert
489 // the mask is not needed. aarch64 just uses bitfield extract
490 __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift,
491 ConstantPoolCacheEntry::tos_state_bits);
492
493 switch (bytecode()) {
494 case Bytecodes::_ldc:
495 case Bytecodes::_ldc_w:
496 {
497 // tos in (itos, ftos, stos, btos, ctos, ztos)
498 Label notInt, notFloat, notShort, notByte, notChar, notBool;
499 __ cmpw(flags, itos);
500 __ br(Assembler::NE, notInt);
501 // itos
502 __ ldrw(r0, field);
503 __ push(itos);
504 __ b(Done);
505
506 __ bind(notInt);
507 __ cmpw(flags, ftos);
508 __ br(Assembler::NE, notFloat);
509 // ftos
510 __ load_float(field);
511 __ push(ftos);
512 __ b(Done);
513
514 __ bind(notFloat);
515 __ cmpw(flags, stos);
516 __ br(Assembler::NE, notShort);
517 // stos
518 __ load_signed_short(r0, field);
519 __ push(stos);
520 __ b(Done);
521
522 __ bind(notShort);
523 __ cmpw(flags, btos);
524 __ br(Assembler::NE, notByte);
525 // btos
526 __ load_signed_byte(r0, field);
527 __ push(btos);
528 __ b(Done);
529
530 __ bind(notByte);
531 __ cmpw(flags, ctos);
532 __ br(Assembler::NE, notChar);
533 // ctos
534 __ load_unsigned_short(r0, field);
535 __ push(ctos);
536 __ b(Done);
537
538 __ bind(notChar);
539 __ cmpw(flags, ztos);
540 __ br(Assembler::NE, notBool);
541 // ztos
542 __ load_signed_byte(r0, field);
543 __ push(ztos);
544 __ b(Done);
545
546 __ bind(notBool);
547 break;
548 }
549
550 case Bytecodes::_ldc2_w:
551 {
552 Label notLong, notDouble;
553 __ cmpw(flags, ltos);
554 __ br(Assembler::NE, notLong);
555 // ltos
556 __ ldr(r0, field);
557 __ push(ltos);
558 __ b(Done);
559
560 __ bind(notLong);
561 __ cmpw(flags, dtos);
562 __ br(Assembler::NE, notDouble);
563 // dtos
564 __ load_double(field);
565 __ push(dtos);
566 __ b(Done);
567
568 __ bind(notDouble);
569 break;
570 }
571
572 default:
573 ShouldNotReachHere();
574 }
575
576 __ stop("bad ldc/condy");
577 }
578
579 void TemplateTable::locals_index(Register reg, int offset)
580 {
581 __ ldrb(reg, at_bcp(offset));
582 __ neg(reg, reg);
583 }
584
585 void TemplateTable::iload() {
586 iload_internal();
587 }
588
589 void TemplateTable::nofast_iload() {
590 iload_internal(may_not_rewrite);
591 }
592
593 void TemplateTable::iload_internal(RewriteControl rc) {
594 transition(vtos, itos);
595 if (RewriteFrequentPairs && rc == may_rewrite) {
596 Label rewrite, done;
597 Register bc = r4;
598
599 // get next bytecode
600 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
601
602 // if _iload, wait to rewrite to iload2. We only want to rewrite the
603 // last two iloads in a pair. Comparing against fast_iload means that
604 // the next bytecode is neither an iload or a caload, and therefore
605 // an iload pair.
606 __ cmpw(r1, Bytecodes::_iload);
607 __ br(Assembler::EQ, done);
608
609 // if _fast_iload rewrite to _fast_iload2
610 __ cmpw(r1, Bytecodes::_fast_iload);
611 __ movw(bc, Bytecodes::_fast_iload2);
612 __ br(Assembler::EQ, rewrite);
613
614 // if _caload rewrite to _fast_icaload
615 __ cmpw(r1, Bytecodes::_caload);
616 __ movw(bc, Bytecodes::_fast_icaload);
617 __ br(Assembler::EQ, rewrite);
618
619 // else rewrite to _fast_iload
620 __ movw(bc, Bytecodes::_fast_iload);
621
622 // rewrite
623 // bc: new bytecode
624 __ bind(rewrite);
625 patch_bytecode(Bytecodes::_iload, bc, r1, false);
626 __ bind(done);
627
628 }
629
630 // do iload, get the local value into tos
631 locals_index(r1);
632 __ ldr(r0, iaddress(r1));
633
634 }
635
636 void TemplateTable::fast_iload2()
637 {
638 transition(vtos, itos);
639 locals_index(r1);
640 __ ldr(r0, iaddress(r1));
641 __ push(itos);
642 locals_index(r1, 3);
643 __ ldr(r0, iaddress(r1));
644 }
645
646 void TemplateTable::fast_iload()
647 {
648 transition(vtos, itos);
649 locals_index(r1);
650 __ ldr(r0, iaddress(r1));
651 }
652
653 void TemplateTable::lload()
654 {
655 transition(vtos, ltos);
656 __ ldrb(r1, at_bcp(1));
657 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
658 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
659 }
660
661 void TemplateTable::fload()
662 {
663 transition(vtos, ftos);
664 locals_index(r1);
665 // n.b. we use ldrd here because this is a 64 bit slot
666 // this is comparable to the iload case
667 __ ldrd(v0, faddress(r1));
668 }
669
670 void TemplateTable::dload()
671 {
672 transition(vtos, dtos);
673 __ ldrb(r1, at_bcp(1));
674 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
675 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
676 }
677
678 void TemplateTable::aload()
679 {
680 transition(vtos, atos);
681 locals_index(r1);
682 __ ldr(r0, iaddress(r1));
683 }
684
685 void TemplateTable::locals_index_wide(Register reg) {
686 __ ldrh(reg, at_bcp(2));
687 __ rev16w(reg, reg);
688 __ neg(reg, reg);
689 }
690
691 void TemplateTable::wide_iload() {
692 transition(vtos, itos);
693 locals_index_wide(r1);
694 __ ldr(r0, iaddress(r1));
695 }
696
697 void TemplateTable::wide_lload()
698 {
699 transition(vtos, ltos);
700 __ ldrh(r1, at_bcp(2));
701 __ rev16w(r1, r1);
702 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
703 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
704 }
705
706 void TemplateTable::wide_fload()
707 {
708 transition(vtos, ftos);
709 locals_index_wide(r1);
710 // n.b. we use ldrd here because this is a 64 bit slot
711 // this is comparable to the iload case
712 __ ldrd(v0, faddress(r1));
713 }
714
715 void TemplateTable::wide_dload()
716 {
717 transition(vtos, dtos);
718 __ ldrh(r1, at_bcp(2));
719 __ rev16w(r1, r1);
720 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
721 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
722 }
723
724 void TemplateTable::wide_aload()
725 {
726 transition(vtos, atos);
727 locals_index_wide(r1);
728 __ ldr(r0, aaddress(r1));
729 }
730
731 void TemplateTable::index_check(Register array, Register index)
732 {
733 // destroys r1, rscratch1
734 // check array
735 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
736 // sign extend index for use by indexed load
737 // __ movl2ptr(index, index);
738 // check index
739 Register length = rscratch1;
740 __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
741 __ cmpw(index, length);
742 if (index != r1) {
743 // ??? convention: move aberrant index into r1 for exception message
744 assert(r1 != array, "different registers");
745 __ mov(r1, index);
746 }
747 Label ok;
748 __ br(Assembler::LO, ok);
749 // ??? convention: move array into r3 for exception message
750 __ mov(r3, array);
751 __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
752 __ br(rscratch1);
753 __ bind(ok);
754 }
755
756 void TemplateTable::iaload()
757 {
758 transition(itos, itos);
759 __ mov(r1, r0);
760 __ pop_ptr(r0);
761 // r0: array
762 // r1: index
763 index_check(r0, r1); // leaves index in r1, kills rscratch1
764 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_INT) >> 2);
765 __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(2)), noreg, noreg);
766 }
767
768 void TemplateTable::laload()
769 {
770 transition(itos, ltos);
771 __ mov(r1, r0);
772 __ pop_ptr(r0);
773 // r0: array
774 // r1: index
775 index_check(r0, r1); // leaves index in r1, kills rscratch1
776 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_LONG) >> 3);
777 __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(3)), noreg, noreg);
778 }
779
780 void TemplateTable::faload()
781 {
782 transition(itos, ftos);
783 __ mov(r1, r0);
784 __ pop_ptr(r0);
785 // r0: array
786 // r1: index
787 index_check(r0, r1); // leaves index in r1, kills rscratch1
788 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT) >> 2);
789 __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(2)), noreg, noreg);
790 }
791
792 void TemplateTable::daload()
793 {
794 transition(itos, dtos);
795 __ mov(r1, r0);
796 __ pop_ptr(r0);
797 // r0: array
798 // r1: index
799 index_check(r0, r1); // leaves index in r1, kills rscratch1
800 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) >> 3);
801 __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(3)), noreg, noreg);
802 }
803
804 void TemplateTable::aaload()
805 {
806 transition(itos, atos);
807 __ mov(r1, r0);
808 __ pop_ptr(r0);
809 // r0: array
810 // r1: index
811 index_check(r0, r1); // leaves index in r1, kills rscratch1
812 if (ValueArrayFlatten) {
813 Label is_flat_array, done;
814
815 __ test_flattened_array_oop(r0, r8 /*temp*/, is_flat_array);
816 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
817 do_oop_load(_masm, Address(r0, r1, Address::uxtw(LogBytesPerHeapOop)), r0, IS_ARRAY);
818
819 __ b(done);
820 __ bind(is_flat_array);
821 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_load), r0, r1);
822 __ bind(done);
823 } else {
824 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
825 do_oop_load(_masm, Address(r0, r1, Address::uxtw(LogBytesPerHeapOop)), r0, IS_ARRAY);
826 }
827 }
828
829 void TemplateTable::baload()
830 {
831 transition(itos, itos);
832 __ mov(r1, r0);
833 __ pop_ptr(r0);
834 // r0: array
835 // r1: index
836 index_check(r0, r1); // leaves index in r1, kills rscratch1
837 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_BYTE) >> 0);
838 __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(0)), noreg, noreg);
839 }
840
841 void TemplateTable::caload()
842 {
843 transition(itos, itos);
844 __ mov(r1, r0);
845 __ pop_ptr(r0);
846 // r0: array
847 // r1: index
848 index_check(r0, r1); // leaves index in r1, kills rscratch1
849 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
850 __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
851 }
852
853 // iload followed by caload frequent pair
854 void TemplateTable::fast_icaload()
855 {
856 transition(vtos, itos);
857 // load index out of locals
858 locals_index(r2);
859 __ ldr(r1, iaddress(r2));
860
861 __ pop_ptr(r0);
862
863 // r0: array
864 // r1: index
865 index_check(r0, r1); // leaves index in r1, kills rscratch1
866 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
867 __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
868 }
869
870 void TemplateTable::saload()
871 {
872 transition(itos, itos);
873 __ mov(r1, r0);
874 __ pop_ptr(r0);
875 // r0: array
876 // r1: index
877 index_check(r0, r1); // leaves index in r1, kills rscratch1
878 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_SHORT) >> 1);
879 __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, r0, Address(r0, r1, Address::uxtw(1)), noreg, noreg);
880 }
881
882 void TemplateTable::iload(int n)
883 {
884 transition(vtos, itos);
885 __ ldr(r0, iaddress(n));
886 }
887
888 void TemplateTable::lload(int n)
889 {
890 transition(vtos, ltos);
891 __ ldr(r0, laddress(n));
892 }
893
894 void TemplateTable::fload(int n)
895 {
896 transition(vtos, ftos);
897 __ ldrs(v0, faddress(n));
898 }
899
900 void TemplateTable::dload(int n)
901 {
902 transition(vtos, dtos);
903 __ ldrd(v0, daddress(n));
904 }
905
906 void TemplateTable::aload(int n)
907 {
908 transition(vtos, atos);
909 __ ldr(r0, iaddress(n));
910 }
911
912 void TemplateTable::aload_0() {
913 aload_0_internal();
914 }
915
916 void TemplateTable::nofast_aload_0() {
917 aload_0_internal(may_not_rewrite);
918 }
919
920 void TemplateTable::aload_0_internal(RewriteControl rc) {
921 // According to bytecode histograms, the pairs:
922 //
923 // _aload_0, _fast_igetfield
924 // _aload_0, _fast_agetfield
925 // _aload_0, _fast_fgetfield
926 //
927 // occur frequently. If RewriteFrequentPairs is set, the (slow)
928 // _aload_0 bytecode checks if the next bytecode is either
929 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
930 // rewrites the current bytecode into a pair bytecode; otherwise it
931 // rewrites the current bytecode into _fast_aload_0 that doesn't do
932 // the pair check anymore.
933 //
934 // Note: If the next bytecode is _getfield, the rewrite must be
935 // delayed, otherwise we may miss an opportunity for a pair.
936 //
937 // Also rewrite frequent pairs
938 // aload_0, aload_1
939 // aload_0, iload_1
940 // These bytecodes with a small amount of code are most profitable
941 // to rewrite
942 if (RewriteFrequentPairs && rc == may_rewrite) {
943 Label rewrite, done;
944 const Register bc = r4;
945
946 // get next bytecode
947 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
948
949 // if _getfield then wait with rewrite
950 __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
951 __ br(Assembler::EQ, done);
952
953 // if _igetfield then rewrite to _fast_iaccess_0
954 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
955 __ cmpw(r1, Bytecodes::_fast_igetfield);
956 __ movw(bc, Bytecodes::_fast_iaccess_0);
957 __ br(Assembler::EQ, rewrite);
958
959 // if _agetfield then rewrite to _fast_aaccess_0
960 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
961 __ cmpw(r1, Bytecodes::_fast_agetfield);
962 __ movw(bc, Bytecodes::_fast_aaccess_0);
963 __ br(Assembler::EQ, rewrite);
964
965 // if _fgetfield then rewrite to _fast_faccess_0
966 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
967 __ cmpw(r1, Bytecodes::_fast_fgetfield);
968 __ movw(bc, Bytecodes::_fast_faccess_0);
969 __ br(Assembler::EQ, rewrite);
970
971 // else rewrite to _fast_aload0
972 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
973 __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
974
975 // rewrite
976 // bc: new bytecode
977 __ bind(rewrite);
978 patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
979
980 __ bind(done);
981 }
982
983 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
984 aload(0);
985 }
986
987 void TemplateTable::istore()
988 {
989 transition(itos, vtos);
990 locals_index(r1);
991 // FIXME: We're being very pernickerty here storing a jint in a
992 // local with strw, which costs an extra instruction over what we'd
993 // be able to do with a simple str. We should just store the whole
994 // word.
995 __ lea(rscratch1, iaddress(r1));
996 __ strw(r0, Address(rscratch1));
997 }
998
999 void TemplateTable::lstore()
1000 {
1001 transition(ltos, vtos);
1002 locals_index(r1);
1003 __ str(r0, laddress(r1, rscratch1, _masm));
1004 }
1005
1006 void TemplateTable::fstore() {
1007 transition(ftos, vtos);
1008 locals_index(r1);
1009 __ lea(rscratch1, iaddress(r1));
1010 __ strs(v0, Address(rscratch1));
1011 }
1012
1013 void TemplateTable::dstore() {
1014 transition(dtos, vtos);
1015 locals_index(r1);
1016 __ strd(v0, daddress(r1, rscratch1, _masm));
1017 }
1018
1019 void TemplateTable::astore()
1020 {
1021 transition(vtos, vtos);
1022 __ pop_ptr(r0);
1023 locals_index(r1);
1024 __ str(r0, aaddress(r1));
1025 }
1026
1027 void TemplateTable::wide_istore() {
1028 transition(vtos, vtos);
1029 __ pop_i();
1030 locals_index_wide(r1);
1031 __ lea(rscratch1, iaddress(r1));
1032 __ strw(r0, Address(rscratch1));
1033 }
1034
1035 void TemplateTable::wide_lstore() {
1036 transition(vtos, vtos);
1037 __ pop_l();
1038 locals_index_wide(r1);
1039 __ str(r0, laddress(r1, rscratch1, _masm));
1040 }
1041
1042 void TemplateTable::wide_fstore() {
1043 transition(vtos, vtos);
1044 __ pop_f();
1045 locals_index_wide(r1);
1046 __ lea(rscratch1, faddress(r1));
1047 __ strs(v0, rscratch1);
1048 }
1049
1050 void TemplateTable::wide_dstore() {
1051 transition(vtos, vtos);
1052 __ pop_d();
1053 locals_index_wide(r1);
1054 __ strd(v0, daddress(r1, rscratch1, _masm));
1055 }
1056
1057 void TemplateTable::wide_astore() {
1058 transition(vtos, vtos);
1059 __ pop_ptr(r0);
1060 locals_index_wide(r1);
1061 __ str(r0, aaddress(r1));
1062 }
1063
1064 void TemplateTable::iastore() {
1065 transition(itos, vtos);
1066 __ pop_i(r1);
1067 __ pop_ptr(r3);
1068 // r0: value
1069 // r1: index
1070 // r3: array
1071 index_check(r3, r1); // prefer index in r1
1072 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_INT) >> 2);
1073 __ access_store_at(T_INT, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(2)), r0, noreg, noreg);
1074 }
1075
1076 void TemplateTable::lastore() {
1077 transition(ltos, vtos);
1078 __ pop_i(r1);
1079 __ pop_ptr(r3);
1080 // r0: value
1081 // r1: index
1082 // r3: array
1083 index_check(r3, r1); // prefer index in r1
1084 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_LONG) >> 3);
1085 __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(3)), r0, noreg, noreg);
1086 }
1087
1088 void TemplateTable::fastore() {
1089 transition(ftos, vtos);
1090 __ pop_i(r1);
1091 __ pop_ptr(r3);
1092 // v0: value
1093 // r1: index
1094 // r3: array
1095 index_check(r3, r1); // prefer index in r1
1096 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT) >> 2);
1097 __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(2)), noreg /* ftos */, noreg, noreg);
1098 }
1099
1100 void TemplateTable::dastore() {
1101 transition(dtos, vtos);
1102 __ pop_i(r1);
1103 __ pop_ptr(r3);
1104 // v0: value
1105 // r1: index
1106 // r3: array
1107 index_check(r3, r1); // prefer index in r1
1108 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE) >> 3);
1109 __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(3)), noreg /* dtos */, noreg, noreg);
1110 }
1111
1112 void TemplateTable::aastore() {
1113 Label is_null, ok_is_subtype, done;
1114 transition(vtos, vtos);
1115 // stack: ..., array, index, value
1116 __ ldr(r0, at_tos()); // value
1117 __ ldr(r2, at_tos_p1()); // index
1118 __ ldr(r3, at_tos_p2()); // array
1119
1120 Address element_address(r3, r4, Address::uxtw(LogBytesPerHeapOop));
1121
1122 index_check(r3, r2); // kills r1
1123
1124 // FIXME: Could we remove the line below?
1125 __ add(r4, r2, arrayOopDesc::base_offset_in_bytes(T_OBJECT) >> LogBytesPerHeapOop);
1126
1127 // do array store check - check for NULL value first
1128 __ cbz(r0, is_null);
1129
1130 Label is_flat_array;
1131 if (ValueArrayFlatten) {
1132 __ test_flattened_array_oop(r3, r8 /*temp*/, is_flat_array);
1133 }
1134
1135 // Move subklass into r1
1136 __ load_klass(r1, r0);
1137
1138 // Move superklass into r0
1139 __ load_klass(r0, r3);
1140 __ ldr(r0, Address(r0, ObjArrayKlass::element_klass_offset()));
1141 // Compress array + index*oopSize + 12 into a single register. Frees r2.
1142
1143 // Generate subtype check. Blows r2, r5
1144 // Superklass in r0. Subklass in r1.
1145
1146 __ gen_subtype_check(r1, ok_is_subtype);
1147
1148 // Come here on failure
1149 // object is at TOS
1150 __ b(Interpreter::_throw_ArrayStoreException_entry);
1151
1152
1153 // Come here on success
1154 __ bind(ok_is_subtype);
1155
1156
1157 // Get the value we will store
1158 __ ldr(r0, at_tos());
1159 // Now store using the appropriate barrier
1160 do_oop_store(_masm, element_address, r0, IS_ARRAY);
1161 __ b(done);
1162
1163 // Have a NULL in r0, r3=array, r2=index. Store NULL at ary[idx]
1164 __ bind(is_null);
1165 __ profile_null_seen(r2);
1166
1167 if (EnableValhalla) {
1168 Label is_null_into_value_array_npe, store_null;
1169
1170 // No way to store null in flat array
1171 __ test_null_free_array_oop(r3, r8, is_null_into_value_array_npe);
1172 __ b(store_null);
1173
1174 __ bind(is_null_into_value_array_npe);
1175 __ b(ExternalAddress(Interpreter::_throw_NullPointerException_entry));
1176
1177 __ bind(store_null);
1178 }
1179
1180 // Store a NULL
1181 do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1182 __ b(done);
1183
1184 if (EnableValhalla) {
1185 Label is_type_ok;
1186
1187 // store non-null value
1188 __ bind(is_flat_array);
1189
1190 // Simplistic type check...
1191 // r0 - value, r2 - index, r3 - array.
1192
1193 // Profile the not-null value's klass.
1194 // Load value class
1195 __ load_klass(r1, r0);
1196 __ profile_typecheck(r2, r1, r0); // blows r2, and r0
1197
1198 // flat value array needs exact type match
1199 // is "r8 == r0" (value subclass == array element superclass)
1200
1201 // Move element klass into r0
1202
1203 __ load_klass(r0, r3);
1204
1205 __ ldr(r0, Address(r0, ArrayKlass::element_klass_offset()));
1206 __ cmp(r0, r1);
1207 __ br(Assembler::EQ, is_type_ok);
1208
1209 __ profile_typecheck_failed(r2);
1210 __ b(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1211
1212 __ bind(is_type_ok);
1213
1214 // Reload from TOS to be safe, because of profile_typecheck that blows r2 and r0.
1215 // FIXME: Should we really do it?
1216 __ ldr(r1, at_tos()); // value
1217 __ mov(r2, r3); // array, ldr(r2, at_tos_p2());
1218 __ ldr(r3, at_tos_p1()); // index
1219 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_store), r1, r2, r3);
1220 }
1221
1222
1223 // Pop stack arguments
1224 __ bind(done);
1225 __ add(esp, esp, 3 * Interpreter::stackElementSize);
1226 }
1227
1228 void TemplateTable::bastore()
1229 {
1230 transition(itos, vtos);
1231 __ pop_i(r1);
1232 __ pop_ptr(r3);
1233 // r0: value
1234 // r1: index
1235 // r3: array
1236 index_check(r3, r1); // prefer index in r1
1237
1238 // Need to check whether array is boolean or byte
1239 // since both types share the bastore bytecode.
1240 __ load_klass(r2, r3);
1241 __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1242 int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit());
1243 Label L_skip;
1244 __ tbz(r2, diffbit_index, L_skip);
1245 __ andw(r0, r0, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1246 __ bind(L_skip);
1247
1248 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_BYTE) >> 0);
1249 __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(0)), r0, noreg, noreg);
1250 }
1251
1252 void TemplateTable::castore()
1253 {
1254 transition(itos, vtos);
1255 __ pop_i(r1);
1256 __ pop_ptr(r3);
1257 // r0: value
1258 // r1: index
1259 // r3: array
1260 index_check(r3, r1); // prefer index in r1
1261 __ add(r1, r1, arrayOopDesc::base_offset_in_bytes(T_CHAR) >> 1);
1262 __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY, Address(r3, r1, Address::uxtw(1)), r0, noreg, noreg);
1263 }
1264
1265 void TemplateTable::sastore()
1266 {
1267 castore();
1268 }
1269
1270 void TemplateTable::istore(int n)
1271 {
1272 transition(itos, vtos);
1273 __ str(r0, iaddress(n));
1274 }
1275
1276 void TemplateTable::lstore(int n)
1277 {
1278 transition(ltos, vtos);
1279 __ str(r0, laddress(n));
1280 }
1281
1282 void TemplateTable::fstore(int n)
1283 {
1284 transition(ftos, vtos);
1285 __ strs(v0, faddress(n));
1286 }
1287
1288 void TemplateTable::dstore(int n)
1289 {
1290 transition(dtos, vtos);
1291 __ strd(v0, daddress(n));
1292 }
1293
1294 void TemplateTable::astore(int n)
1295 {
1296 transition(vtos, vtos);
1297 __ pop_ptr(r0);
1298 __ str(r0, iaddress(n));
1299 }
1300
1301 void TemplateTable::pop()
1302 {
1303 transition(vtos, vtos);
1304 __ add(esp, esp, Interpreter::stackElementSize);
1305 }
1306
1307 void TemplateTable::pop2()
1308 {
1309 transition(vtos, vtos);
1310 __ add(esp, esp, 2 * Interpreter::stackElementSize);
1311 }
1312
1313 void TemplateTable::dup()
1314 {
1315 transition(vtos, vtos);
1316 __ ldr(r0, Address(esp, 0));
1317 __ push(r0);
1318 // stack: ..., a, a
1319 }
1320
1321 void TemplateTable::dup_x1()
1322 {
1323 transition(vtos, vtos);
1324 // stack: ..., a, b
1325 __ ldr(r0, at_tos()); // load b
1326 __ ldr(r2, at_tos_p1()); // load a
1327 __ str(r0, at_tos_p1()); // store b
1328 __ str(r2, at_tos()); // store a
1329 __ push(r0); // push b
1330 // stack: ..., b, a, b
1331 }
1332
1333 void TemplateTable::dup_x2()
1334 {
1335 transition(vtos, vtos);
1336 // stack: ..., a, b, c
1337 __ ldr(r0, at_tos()); // load c
1338 __ ldr(r2, at_tos_p2()); // load a
1339 __ str(r0, at_tos_p2()); // store c in a
1340 __ push(r0); // push c
1341 // stack: ..., c, b, c, c
1342 __ ldr(r0, at_tos_p2()); // load b
1343 __ str(r2, at_tos_p2()); // store a in b
1344 // stack: ..., c, a, c, c
1345 __ str(r0, at_tos_p1()); // store b in c
1346 // stack: ..., c, a, b, c
1347 }
1348
1349 void TemplateTable::dup2()
1350 {
1351 transition(vtos, vtos);
1352 // stack: ..., a, b
1353 __ ldr(r0, at_tos_p1()); // load a
1354 __ push(r0); // push a
1355 __ ldr(r0, at_tos_p1()); // load b
1356 __ push(r0); // push b
1357 // stack: ..., a, b, a, b
1358 }
1359
1360 void TemplateTable::dup2_x1()
1361 {
1362 transition(vtos, vtos);
1363 // stack: ..., a, b, c
1364 __ ldr(r2, at_tos()); // load c
1365 __ ldr(r0, at_tos_p1()); // load b
1366 __ push(r0); // push b
1367 __ push(r2); // push c
1368 // stack: ..., a, b, c, b, c
1369 __ str(r2, at_tos_p3()); // store c in b
1370 // stack: ..., a, c, c, b, c
1371 __ ldr(r2, at_tos_p4()); // load a
1372 __ str(r2, at_tos_p2()); // store a in 2nd c
1373 // stack: ..., a, c, a, b, c
1374 __ str(r0, at_tos_p4()); // store b in a
1375 // stack: ..., b, c, a, b, c
1376 }
1377
1378 void TemplateTable::dup2_x2()
1379 {
1380 transition(vtos, vtos);
1381 // stack: ..., a, b, c, d
1382 __ ldr(r2, at_tos()); // load d
1383 __ ldr(r0, at_tos_p1()); // load c
1384 __ push(r0) ; // push c
1385 __ push(r2); // push d
1386 // stack: ..., a, b, c, d, c, d
1387 __ ldr(r0, at_tos_p4()); // load b
1388 __ str(r0, at_tos_p2()); // store b in d
1389 __ str(r2, at_tos_p4()); // store d in b
1390 // stack: ..., a, d, c, b, c, d
1391 __ ldr(r2, at_tos_p5()); // load a
1392 __ ldr(r0, at_tos_p3()); // load c
1393 __ str(r2, at_tos_p3()); // store a in c
1394 __ str(r0, at_tos_p5()); // store c in a
1395 // stack: ..., c, d, a, b, c, d
1396 }
1397
1398 void TemplateTable::swap()
1399 {
1400 transition(vtos, vtos);
1401 // stack: ..., a, b
1402 __ ldr(r2, at_tos_p1()); // load a
1403 __ ldr(r0, at_tos()); // load b
1404 __ str(r2, at_tos()); // store a in b
1405 __ str(r0, at_tos_p1()); // store b in a
1406 // stack: ..., b, a
1407 }
1408
1409 void TemplateTable::iop2(Operation op)
1410 {
1411 transition(itos, itos);
1412 // r0 <== r1 op r0
1413 __ pop_i(r1);
1414 switch (op) {
1415 case add : __ addw(r0, r1, r0); break;
1416 case sub : __ subw(r0, r1, r0); break;
1417 case mul : __ mulw(r0, r1, r0); break;
1418 case _and : __ andw(r0, r1, r0); break;
1419 case _or : __ orrw(r0, r1, r0); break;
1420 case _xor : __ eorw(r0, r1, r0); break;
1421 case shl : __ lslvw(r0, r1, r0); break;
1422 case shr : __ asrvw(r0, r1, r0); break;
1423 case ushr : __ lsrvw(r0, r1, r0);break;
1424 default : ShouldNotReachHere();
1425 }
1426 }
1427
1428 void TemplateTable::lop2(Operation op)
1429 {
1430 transition(ltos, ltos);
1431 // r0 <== r1 op r0
1432 __ pop_l(r1);
1433 switch (op) {
1434 case add : __ add(r0, r1, r0); break;
1435 case sub : __ sub(r0, r1, r0); break;
1436 case mul : __ mul(r0, r1, r0); break;
1437 case _and : __ andr(r0, r1, r0); break;
1438 case _or : __ orr(r0, r1, r0); break;
1439 case _xor : __ eor(r0, r1, r0); break;
1440 default : ShouldNotReachHere();
1441 }
1442 }
1443
1444 void TemplateTable::idiv()
1445 {
1446 transition(itos, itos);
1447 // explicitly check for div0
1448 Label no_div0;
1449 __ cbnzw(r0, no_div0);
1450 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1451 __ br(rscratch1);
1452 __ bind(no_div0);
1453 __ pop_i(r1);
1454 // r0 <== r1 idiv r0
1455 __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1456 }
1457
1458 void TemplateTable::irem()
1459 {
1460 transition(itos, itos);
1461 // explicitly check for div0
1462 Label no_div0;
1463 __ cbnzw(r0, no_div0);
1464 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1465 __ br(rscratch1);
1466 __ bind(no_div0);
1467 __ pop_i(r1);
1468 // r0 <== r1 irem r0
1469 __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1470 }
1471
1472 void TemplateTable::lmul()
1473 {
1474 transition(ltos, ltos);
1475 __ pop_l(r1);
1476 __ mul(r0, r0, r1);
1477 }
1478
1479 void TemplateTable::ldiv()
1480 {
1481 transition(ltos, ltos);
1482 // explicitly check for div0
1483 Label no_div0;
1484 __ cbnz(r0, no_div0);
1485 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1486 __ br(rscratch1);
1487 __ bind(no_div0);
1488 __ pop_l(r1);
1489 // r0 <== r1 ldiv r0
1490 __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1491 }
1492
1493 void TemplateTable::lrem()
1494 {
1495 transition(ltos, ltos);
1496 // explicitly check for div0
1497 Label no_div0;
1498 __ cbnz(r0, no_div0);
1499 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1500 __ br(rscratch1);
1501 __ bind(no_div0);
1502 __ pop_l(r1);
1503 // r0 <== r1 lrem r0
1504 __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1505 }
1506
1507 void TemplateTable::lshl()
1508 {
1509 transition(itos, ltos);
1510 // shift count is in r0
1511 __ pop_l(r1);
1512 __ lslv(r0, r1, r0);
1513 }
1514
1515 void TemplateTable::lshr()
1516 {
1517 transition(itos, ltos);
1518 // shift count is in r0
1519 __ pop_l(r1);
1520 __ asrv(r0, r1, r0);
1521 }
1522
1523 void TemplateTable::lushr()
1524 {
1525 transition(itos, ltos);
1526 // shift count is in r0
1527 __ pop_l(r1);
1528 __ lsrv(r0, r1, r0);
1529 }
1530
1531 void TemplateTable::fop2(Operation op)
1532 {
1533 transition(ftos, ftos);
1534 switch (op) {
1535 case add:
1536 // n.b. use ldrd because this is a 64 bit slot
1537 __ pop_f(v1);
1538 __ fadds(v0, v1, v0);
1539 break;
1540 case sub:
1541 __ pop_f(v1);
1542 __ fsubs(v0, v1, v0);
1543 break;
1544 case mul:
1545 __ pop_f(v1);
1546 __ fmuls(v0, v1, v0);
1547 break;
1548 case div:
1549 __ pop_f(v1);
1550 __ fdivs(v0, v1, v0);
1551 break;
1552 case rem:
1553 __ fmovs(v1, v0);
1554 __ pop_f(v0);
1555 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::frem),
1556 0, 2, MacroAssembler::ret_type_float);
1557 break;
1558 default:
1559 ShouldNotReachHere();
1560 break;
1561 }
1562 }
1563
1564 void TemplateTable::dop2(Operation op)
1565 {
1566 transition(dtos, dtos);
1567 switch (op) {
1568 case add:
1569 // n.b. use ldrd because this is a 64 bit slot
1570 __ pop_d(v1);
1571 __ faddd(v0, v1, v0);
1572 break;
1573 case sub:
1574 __ pop_d(v1);
1575 __ fsubd(v0, v1, v0);
1576 break;
1577 case mul:
1578 __ pop_d(v1);
1579 __ fmuld(v0, v1, v0);
1580 break;
1581 case div:
1582 __ pop_d(v1);
1583 __ fdivd(v0, v1, v0);
1584 break;
1585 case rem:
1586 __ fmovd(v1, v0);
1587 __ pop_d(v0);
1588 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::drem),
1589 0, 2, MacroAssembler::ret_type_double);
1590 break;
1591 default:
1592 ShouldNotReachHere();
1593 break;
1594 }
1595 }
1596
1597 void TemplateTable::ineg()
1598 {
1599 transition(itos, itos);
1600 __ negw(r0, r0);
1601
1602 }
1603
1604 void TemplateTable::lneg()
1605 {
1606 transition(ltos, ltos);
1607 __ neg(r0, r0);
1608 }
1609
1610 void TemplateTable::fneg()
1611 {
1612 transition(ftos, ftos);
1613 __ fnegs(v0, v0);
1614 }
1615
1616 void TemplateTable::dneg()
1617 {
1618 transition(dtos, dtos);
1619 __ fnegd(v0, v0);
1620 }
1621
1622 void TemplateTable::iinc()
1623 {
1624 transition(vtos, vtos);
1625 __ load_signed_byte(r1, at_bcp(2)); // get constant
1626 locals_index(r2);
1627 __ ldr(r0, iaddress(r2));
1628 __ addw(r0, r0, r1);
1629 __ str(r0, iaddress(r2));
1630 }
1631
1632 void TemplateTable::wide_iinc()
1633 {
1634 transition(vtos, vtos);
1635 // __ mov(r1, zr);
1636 __ ldrw(r1, at_bcp(2)); // get constant and index
1637 __ rev16(r1, r1);
1638 __ ubfx(r2, r1, 0, 16);
1639 __ neg(r2, r2);
1640 __ sbfx(r1, r1, 16, 16);
1641 __ ldr(r0, iaddress(r2));
1642 __ addw(r0, r0, r1);
1643 __ str(r0, iaddress(r2));
1644 }
1645
1646 void TemplateTable::convert()
1647 {
1648 // Checking
1649 #ifdef ASSERT
1650 {
1651 TosState tos_in = ilgl;
1652 TosState tos_out = ilgl;
1653 switch (bytecode()) {
1654 case Bytecodes::_i2l: // fall through
1655 case Bytecodes::_i2f: // fall through
1656 case Bytecodes::_i2d: // fall through
1657 case Bytecodes::_i2b: // fall through
1658 case Bytecodes::_i2c: // fall through
1659 case Bytecodes::_i2s: tos_in = itos; break;
1660 case Bytecodes::_l2i: // fall through
1661 case Bytecodes::_l2f: // fall through
1662 case Bytecodes::_l2d: tos_in = ltos; break;
1663 case Bytecodes::_f2i: // fall through
1664 case Bytecodes::_f2l: // fall through
1665 case Bytecodes::_f2d: tos_in = ftos; break;
1666 case Bytecodes::_d2i: // fall through
1667 case Bytecodes::_d2l: // fall through
1668 case Bytecodes::_d2f: tos_in = dtos; break;
1669 default : ShouldNotReachHere();
1670 }
1671 switch (bytecode()) {
1672 case Bytecodes::_l2i: // fall through
1673 case Bytecodes::_f2i: // fall through
1674 case Bytecodes::_d2i: // fall through
1675 case Bytecodes::_i2b: // fall through
1676 case Bytecodes::_i2c: // fall through
1677 case Bytecodes::_i2s: tos_out = itos; break;
1678 case Bytecodes::_i2l: // fall through
1679 case Bytecodes::_f2l: // fall through
1680 case Bytecodes::_d2l: tos_out = ltos; break;
1681 case Bytecodes::_i2f: // fall through
1682 case Bytecodes::_l2f: // fall through
1683 case Bytecodes::_d2f: tos_out = ftos; break;
1684 case Bytecodes::_i2d: // fall through
1685 case Bytecodes::_l2d: // fall through
1686 case Bytecodes::_f2d: tos_out = dtos; break;
1687 default : ShouldNotReachHere();
1688 }
1689 transition(tos_in, tos_out);
1690 }
1691 #endif // ASSERT
1692 // static const int64_t is_nan = 0x8000000000000000L;
1693
1694 // Conversion
1695 switch (bytecode()) {
1696 case Bytecodes::_i2l:
1697 __ sxtw(r0, r0);
1698 break;
1699 case Bytecodes::_i2f:
1700 __ scvtfws(v0, r0);
1701 break;
1702 case Bytecodes::_i2d:
1703 __ scvtfwd(v0, r0);
1704 break;
1705 case Bytecodes::_i2b:
1706 __ sxtbw(r0, r0);
1707 break;
1708 case Bytecodes::_i2c:
1709 __ uxthw(r0, r0);
1710 break;
1711 case Bytecodes::_i2s:
1712 __ sxthw(r0, r0);
1713 break;
1714 case Bytecodes::_l2i:
1715 __ uxtw(r0, r0);
1716 break;
1717 case Bytecodes::_l2f:
1718 __ scvtfs(v0, r0);
1719 break;
1720 case Bytecodes::_l2d:
1721 __ scvtfd(v0, r0);
1722 break;
1723 case Bytecodes::_f2i:
1724 {
1725 Label L_Okay;
1726 __ clear_fpsr();
1727 __ fcvtzsw(r0, v0);
1728 __ get_fpsr(r1);
1729 __ cbzw(r1, L_Okay);
1730 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2i),
1731 0, 1, MacroAssembler::ret_type_integral);
1732 __ bind(L_Okay);
1733 }
1734 break;
1735 case Bytecodes::_f2l:
1736 {
1737 Label L_Okay;
1738 __ clear_fpsr();
1739 __ fcvtzs(r0, v0);
1740 __ get_fpsr(r1);
1741 __ cbzw(r1, L_Okay);
1742 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2l),
1743 0, 1, MacroAssembler::ret_type_integral);
1744 __ bind(L_Okay);
1745 }
1746 break;
1747 case Bytecodes::_f2d:
1748 __ fcvts(v0, v0);
1749 break;
1750 case Bytecodes::_d2i:
1751 {
1752 Label L_Okay;
1753 __ clear_fpsr();
1754 __ fcvtzdw(r0, v0);
1755 __ get_fpsr(r1);
1756 __ cbzw(r1, L_Okay);
1757 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2i),
1758 0, 1, MacroAssembler::ret_type_integral);
1759 __ bind(L_Okay);
1760 }
1761 break;
1762 case Bytecodes::_d2l:
1763 {
1764 Label L_Okay;
1765 __ clear_fpsr();
1766 __ fcvtzd(r0, v0);
1767 __ get_fpsr(r1);
1768 __ cbzw(r1, L_Okay);
1769 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2l),
1770 0, 1, MacroAssembler::ret_type_integral);
1771 __ bind(L_Okay);
1772 }
1773 break;
1774 case Bytecodes::_d2f:
1775 __ fcvtd(v0, v0);
1776 break;
1777 default:
1778 ShouldNotReachHere();
1779 }
1780 }
1781
1782 void TemplateTable::lcmp()
1783 {
1784 transition(ltos, itos);
1785 Label done;
1786 __ pop_l(r1);
1787 __ cmp(r1, r0);
1788 __ mov(r0, (u_int64_t)-1L);
1789 __ br(Assembler::LT, done);
1790 // __ mov(r0, 1UL);
1791 // __ csel(r0, r0, zr, Assembler::NE);
1792 // and here is a faster way
1793 __ csinc(r0, zr, zr, Assembler::EQ);
1794 __ bind(done);
1795 }
1796
1797 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1798 {
1799 Label done;
1800 if (is_float) {
1801 // XXX get rid of pop here, use ... reg, mem32
1802 __ pop_f(v1);
1803 __ fcmps(v1, v0);
1804 } else {
1805 // XXX get rid of pop here, use ... reg, mem64
1806 __ pop_d(v1);
1807 __ fcmpd(v1, v0);
1808 }
1809 if (unordered_result < 0) {
1810 // we want -1 for unordered or less than, 0 for equal and 1 for
1811 // greater than.
1812 __ mov(r0, (u_int64_t)-1L);
1813 // for FP LT tests less than or unordered
1814 __ br(Assembler::LT, done);
1815 // install 0 for EQ otherwise 1
1816 __ csinc(r0, zr, zr, Assembler::EQ);
1817 } else {
1818 // we want -1 for less than, 0 for equal and 1 for unordered or
1819 // greater than.
1820 __ mov(r0, 1L);
1821 // for FP HI tests greater than or unordered
1822 __ br(Assembler::HI, done);
1823 // install 0 for EQ otherwise ~0
1824 __ csinv(r0, zr, zr, Assembler::EQ);
1825
1826 }
1827 __ bind(done);
1828 }
1829
1830 void TemplateTable::branch(bool is_jsr, bool is_wide)
1831 {
1832 // We might be moving to a safepoint. The thread which calls
1833 // Interpreter::notice_safepoints() will effectively flush its cache
1834 // when it makes a system call, but we need to do something to
1835 // ensure that we see the changed dispatch table.
1836 __ membar(MacroAssembler::LoadLoad);
1837
1838 __ profile_taken_branch(r0, r1);
1839 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1840 InvocationCounter::counter_offset();
1841 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1842 InvocationCounter::counter_offset();
1843
1844 // load branch displacement
1845 if (!is_wide) {
1846 __ ldrh(r2, at_bcp(1));
1847 __ rev16(r2, r2);
1848 // sign extend the 16 bit value in r2
1849 __ sbfm(r2, r2, 0, 15);
1850 } else {
1851 __ ldrw(r2, at_bcp(1));
1852 __ revw(r2, r2);
1853 // sign extend the 32 bit value in r2
1854 __ sbfm(r2, r2, 0, 31);
1855 }
1856
1857 // Handle all the JSR stuff here, then exit.
1858 // It's much shorter and cleaner than intermingling with the non-JSR
1859 // normal-branch stuff occurring below.
1860
1861 if (is_jsr) {
1862 // Pre-load the next target bytecode into rscratch1
1863 __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1864 // compute return address as bci
1865 __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1866 __ add(rscratch2, rscratch2,
1867 in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1868 __ sub(r1, rbcp, rscratch2);
1869 __ push_i(r1);
1870 // Adjust the bcp by the 16-bit displacement in r2
1871 __ add(rbcp, rbcp, r2);
1872 __ dispatch_only(vtos, /*generate_poll*/true);
1873 return;
1874 }
1875
1876 // Normal (non-jsr) branch handling
1877
1878 // Adjust the bcp by the displacement in r2
1879 __ add(rbcp, rbcp, r2);
1880
1881 assert(UseLoopCounter || !UseOnStackReplacement,
1882 "on-stack-replacement requires loop counters");
1883 Label backedge_counter_overflow;
1884 Label profile_method;
1885 Label dispatch;
1886 if (UseLoopCounter) {
1887 // increment backedge counter for backward branches
1888 // r0: MDO
1889 // w1: MDO bumped taken-count
1890 // r2: target offset
1891 __ cmp(r2, zr);
1892 __ br(Assembler::GT, dispatch); // count only if backward branch
1893
1894 // ECN: FIXME: This code smells
1895 // check if MethodCounters exists
1896 Label has_counters;
1897 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1898 __ cbnz(rscratch1, has_counters);
1899 __ push(r0);
1900 __ push(r1);
1901 __ push(r2);
1902 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1903 InterpreterRuntime::build_method_counters), rmethod);
1904 __ pop(r2);
1905 __ pop(r1);
1906 __ pop(r0);
1907 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1908 __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1909 __ bind(has_counters);
1910
1911 if (TieredCompilation) {
1912 Label no_mdo;
1913 int increment = InvocationCounter::count_increment;
1914 if (ProfileInterpreter) {
1915 // Are we profiling?
1916 __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1917 __ cbz(r1, no_mdo);
1918 // Increment the MDO backedge counter
1919 const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1920 in_bytes(InvocationCounter::counter_offset()));
1921 const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1922 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1923 r0, rscratch1, false, Assembler::EQ,
1924 UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1925 __ b(dispatch);
1926 }
1927 __ bind(no_mdo);
1928 // Increment backedge counter in MethodCounters*
1929 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1930 const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1931 __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1932 r0, rscratch2, false, Assembler::EQ,
1933 UseOnStackReplacement ? &backedge_counter_overflow : &dispatch);
1934 } else { // not TieredCompilation
1935 // increment counter
1936 __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1937 __ ldrw(r0, Address(rscratch2, be_offset)); // load backedge counter
1938 __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1939 __ strw(rscratch1, Address(rscratch2, be_offset)); // store counter
1940
1941 __ ldrw(r0, Address(rscratch2, inv_offset)); // load invocation counter
1942 __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1943 __ addw(r0, r0, rscratch1); // add both counters
1944
1945 if (ProfileInterpreter) {
1946 // Test to see if we should create a method data oop
1947 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
1948 __ cmpw(r0, rscratch1);
1949 __ br(Assembler::LT, dispatch);
1950
1951 // if no method data exists, go to profile method
1952 __ test_method_data_pointer(r0, profile_method);
1953
1954 if (UseOnStackReplacement) {
1955 // check for overflow against w1 which is the MDO taken count
1956 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1957 __ cmpw(r1, rscratch1);
1958 __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1959
1960 // When ProfileInterpreter is on, the backedge_count comes
1961 // from the MethodData*, which value does not get reset on
1962 // the call to frequency_counter_overflow(). To avoid
1963 // excessive calls to the overflow routine while the method is
1964 // being compiled, add a second test to make sure the overflow
1965 // function is called only once every overflow_frequency.
1966 const int overflow_frequency = 1024;
1967 __ andsw(r1, r1, overflow_frequency - 1);
1968 __ br(Assembler::EQ, backedge_counter_overflow);
1969
1970 }
1971 } else {
1972 if (UseOnStackReplacement) {
1973 // check for overflow against w0, which is the sum of the
1974 // counters
1975 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1976 __ cmpw(r0, rscratch1);
1977 __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1978 }
1979 }
1980 }
1981 __ bind(dispatch);
1982 }
1983
1984 // Pre-load the next target bytecode into rscratch1
1985 __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1986
1987 // continue with the bytecode @ target
1988 // rscratch1: target bytecode
1989 // rbcp: target bcp
1990 __ dispatch_only(vtos, /*generate_poll*/true);
1991
1992 if (UseLoopCounter) {
1993 if (ProfileInterpreter) {
1994 // Out-of-line code to allocate method data oop.
1995 __ bind(profile_method);
1996 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1997 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
1998 __ set_method_data_pointer_for_bcp();
1999 __ b(dispatch);
2000 }
2001
2002 if (UseOnStackReplacement) {
2003 // invocation counter overflow
2004 __ bind(backedge_counter_overflow);
2005 __ neg(r2, r2);
2006 __ add(r2, r2, rbcp); // branch bcp
2007 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2008 __ call_VM(noreg,
2009 CAST_FROM_FN_PTR(address,
2010 InterpreterRuntime::frequency_counter_overflow),
2011 r2);
2012 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
2013
2014 // r0: osr nmethod (osr ok) or NULL (osr not possible)
2015 // w1: target bytecode
2016 // r2: scratch
2017 __ cbz(r0, dispatch); // test result -- no osr if null
2018 // nmethod may have been invalidated (VM may block upon call_VM return)
2019 __ ldrb(r2, Address(r0, nmethod::state_offset()));
2020 if (nmethod::in_use != 0)
2021 __ sub(r2, r2, nmethod::in_use);
2022 __ cbnz(r2, dispatch);
2023
2024 // We have the address of an on stack replacement routine in r0
2025 // We need to prepare to execute the OSR method. First we must
2026 // migrate the locals and monitors off of the stack.
2027
2028 __ mov(r19, r0); // save the nmethod
2029
2030 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2031
2032 // r0 is OSR buffer, move it to expected parameter location
2033 __ mov(j_rarg0, r0);
2034
2035 // remove activation
2036 // get sender esp
2037 __ ldr(esp,
2038 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
2039 // remove frame anchor
2040 __ leave();
2041 // Ensure compiled code always sees stack at proper alignment
2042 __ andr(sp, esp, -16);
2043
2044 // and begin the OSR nmethod
2045 __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
2046 __ br(rscratch1);
2047 }
2048 }
2049 }
2050
2051
2052 void TemplateTable::if_0cmp(Condition cc)
2053 {
2054 transition(itos, vtos);
2055 // assume branch is more often taken than not (loops use backward branches)
2056 Label not_taken;
2057 if (cc == equal)
2058 __ cbnzw(r0, not_taken);
2059 else if (cc == not_equal)
2060 __ cbzw(r0, not_taken);
2061 else {
2062 __ andsw(zr, r0, r0);
2063 __ br(j_not(cc), not_taken);
2064 }
2065
2066 branch(false, false);
2067 __ bind(not_taken);
2068 __ profile_not_taken_branch(r0);
2069 }
2070
2071 void TemplateTable::if_icmp(Condition cc)
2072 {
2073 transition(itos, vtos);
2074 // assume branch is more often taken than not (loops use backward branches)
2075 Label not_taken;
2076 __ pop_i(r1);
2077 __ cmpw(r1, r0, Assembler::LSL);
2078 __ br(j_not(cc), not_taken);
2079 branch(false, false);
2080 __ bind(not_taken);
2081 __ profile_not_taken_branch(r0);
2082 }
2083
2084 void TemplateTable::if_nullcmp(Condition cc)
2085 {
2086 transition(atos, vtos);
2087 // assume branch is more often taken than not (loops use backward branches)
2088 Label not_taken;
2089 if (cc == equal)
2090 __ cbnz(r0, not_taken);
2091 else
2092 __ cbz(r0, not_taken);
2093 branch(false, false);
2094 __ bind(not_taken);
2095 __ profile_not_taken_branch(r0);
2096 }
2097
2098 void TemplateTable::if_acmp(Condition cc) {
2099 transition(atos, vtos);
2100 // assume branch is more often taken than not (loops use backward branches)
2101 Label taken, not_taken;
2102 __ pop_ptr(r1);
2103
2104 Register is_value_mask = rscratch1;
2105 __ mov(is_value_mask, markOopDesc::always_locked_pattern);
2106
2107 if (EnableValhalla) {
2108 __ cmp(r1, r0);
2109 __ br(Assembler::EQ, (cc == equal) ? taken : not_taken);
2110
2111 // might be substitutable, test if either r0 or r1 is null
2112 __ andr(r2, r0, r1);
2113 __ cbz(r2, (cc == equal) ? not_taken : taken);
2114
2115 // and both are values ?
2116 __ ldr(r2, Address(r1, oopDesc::mark_offset_in_bytes()));
2117 __ andr(r2, r2, is_value_mask);
2118 __ ldr(r4, Address(r0, oopDesc::mark_offset_in_bytes()));
2119 __ andr(r4, r4, is_value_mask);
2120 __ andr(r2, r2, r4);
2121 __ cmp(r2, is_value_mask);
2122 __ br(Assembler::NE, (cc == equal) ? not_taken : taken);
2123
2124 // same value klass ?
2125 __ load_metadata(r2, r1);
2126 __ load_metadata(r4, r0);
2127 __ cmp(r2, r4);
2128 __ br(Assembler::NE, (cc == equal) ? not_taken : taken);
2129
2130 // Know both are the same type, let's test for substitutability...
2131 if (cc == equal) {
2132 invoke_is_substitutable(r0, r1, taken, not_taken);
2133 } else {
2134 invoke_is_substitutable(r0, r1, not_taken, taken);
2135 }
2136 __ stop("Not reachable");
2137 }
2138
2139 __ cmpoop(r1, r0);
2140 __ br(j_not(cc), not_taken);
2141 __ bind(taken);
2142 branch(false, false);
2143 __ bind(not_taken);
2144 __ profile_not_taken_branch(r0);
2145 }
2146
2147 void TemplateTable::invoke_is_substitutable(Register aobj, Register bobj,
2148 Label& is_subst, Label& not_subst) {
2149
2150 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::is_substitutable), aobj, bobj);
2151 // Restored... r0 answer, jmp to outcome...
2152 __ cbz(r0, not_subst);
2153 __ b(is_subst);
2154 }
2155
2156
2157 void TemplateTable::ret() {
2158 transition(vtos, vtos);
2159 // We might be moving to a safepoint. The thread which calls
2160 // Interpreter::notice_safepoints() will effectively flush its cache
2161 // when it makes a system call, but we need to do something to
2162 // ensure that we see the changed dispatch table.
2163 __ membar(MacroAssembler::LoadLoad);
2164
2165 locals_index(r1);
2166 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2167 __ profile_ret(r1, r2);
2168 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2169 __ lea(rbcp, Address(rbcp, r1));
2170 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2171 __ dispatch_next(vtos, 0, /*generate_poll*/true);
2172 }
2173
2174 void TemplateTable::wide_ret() {
2175 transition(vtos, vtos);
2176 locals_index_wide(r1);
2177 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
2178 __ profile_ret(r1, r2);
2179 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
2180 __ lea(rbcp, Address(rbcp, r1));
2181 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
2182 __ dispatch_next(vtos, 0, /*generate_poll*/true);
2183 }
2184
2185
2186 void TemplateTable::tableswitch() {
2187 Label default_case, continue_execution;
2188 transition(itos, vtos);
2189 // align rbcp
2190 __ lea(r1, at_bcp(BytesPerInt));
2191 __ andr(r1, r1, -BytesPerInt);
2192 // load lo & hi
2193 __ ldrw(r2, Address(r1, BytesPerInt));
2194 __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2195 __ rev32(r2, r2);
2196 __ rev32(r3, r3);
2197 // check against lo & hi
2198 __ cmpw(r0, r2);
2199 __ br(Assembler::LT, default_case);
2200 __ cmpw(r0, r3);
2201 __ br(Assembler::GT, default_case);
2202 // lookup dispatch offset
2203 __ subw(r0, r0, r2);
2204 __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2205 __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2206 __ profile_switch_case(r0, r1, r2);
2207 // continue execution
2208 __ bind(continue_execution);
2209 __ rev32(r3, r3);
2210 __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2211 __ add(rbcp, rbcp, r3, ext::sxtw);
2212 __ dispatch_only(vtos, /*generate_poll*/true);
2213 // handle default
2214 __ bind(default_case);
2215 __ profile_switch_default(r0);
2216 __ ldrw(r3, Address(r1, 0));
2217 __ b(continue_execution);
2218 }
2219
2220 void TemplateTable::lookupswitch() {
2221 transition(itos, itos);
2222 __ stop("lookupswitch bytecode should have been rewritten");
2223 }
2224
2225 void TemplateTable::fast_linearswitch() {
2226 transition(itos, vtos);
2227 Label loop_entry, loop, found, continue_execution;
2228 // bswap r0 so we can avoid bswapping the table entries
2229 __ rev32(r0, r0);
2230 // align rbcp
2231 __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2232 // this instruction (change offsets
2233 // below)
2234 __ andr(r19, r19, -BytesPerInt);
2235 // set counter
2236 __ ldrw(r1, Address(r19, BytesPerInt));
2237 __ rev32(r1, r1);
2238 __ b(loop_entry);
2239 // table search
2240 __ bind(loop);
2241 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2242 __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2243 __ cmpw(r0, rscratch1);
2244 __ br(Assembler::EQ, found);
2245 __ bind(loop_entry);
2246 __ subs(r1, r1, 1);
2247 __ br(Assembler::PL, loop);
2248 // default case
2249 __ profile_switch_default(r0);
2250 __ ldrw(r3, Address(r19, 0));
2251 __ b(continue_execution);
2252 // entry found -> get offset
2253 __ bind(found);
2254 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2255 __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2256 __ profile_switch_case(r1, r0, r19);
2257 // continue execution
2258 __ bind(continue_execution);
2259 __ rev32(r3, r3);
2260 __ add(rbcp, rbcp, r3, ext::sxtw);
2261 __ ldrb(rscratch1, Address(rbcp, 0));
2262 __ dispatch_only(vtos, /*generate_poll*/true);
2263 }
2264
2265 void TemplateTable::fast_binaryswitch() {
2266 transition(itos, vtos);
2267 // Implementation using the following core algorithm:
2268 //
2269 // int binary_search(int key, LookupswitchPair* array, int n) {
2270 // // Binary search according to "Methodik des Programmierens" by
2271 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2272 // int i = 0;
2273 // int j = n;
2274 // while (i+1 < j) {
2275 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2276 // // with Q: for all i: 0 <= i < n: key < a[i]
2277 // // where a stands for the array and assuming that the (inexisting)
2278 // // element a[n] is infinitely big.
2279 // int h = (i + j) >> 1;
2280 // // i < h < j
2281 // if (key < array[h].fast_match()) {
2282 // j = h;
2283 // } else {
2284 // i = h;
2285 // }
2286 // }
2287 // // R: a[i] <= key < a[i+1] or Q
2288 // // (i.e., if key is within array, i is the correct index)
2289 // return i;
2290 // }
2291
2292 // Register allocation
2293 const Register key = r0; // already set (tosca)
2294 const Register array = r1;
2295 const Register i = r2;
2296 const Register j = r3;
2297 const Register h = rscratch1;
2298 const Register temp = rscratch2;
2299
2300 // Find array start
2301 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2302 // get rid of this
2303 // instruction (change
2304 // offsets below)
2305 __ andr(array, array, -BytesPerInt);
2306
2307 // Initialize i & j
2308 __ mov(i, 0); // i = 0;
2309 __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2310
2311 // Convert j into native byteordering
2312 __ rev32(j, j);
2313
2314 // And start
2315 Label entry;
2316 __ b(entry);
2317
2318 // binary search loop
2319 {
2320 Label loop;
2321 __ bind(loop);
2322 // int h = (i + j) >> 1;
2323 __ addw(h, i, j); // h = i + j;
2324 __ lsrw(h, h, 1); // h = (i + j) >> 1;
2325 // if (key < array[h].fast_match()) {
2326 // j = h;
2327 // } else {
2328 // i = h;
2329 // }
2330 // Convert array[h].match to native byte-ordering before compare
2331 __ ldr(temp, Address(array, h, Address::lsl(3)));
2332 __ rev32(temp, temp);
2333 __ cmpw(key, temp);
2334 // j = h if (key < array[h].fast_match())
2335 __ csel(j, h, j, Assembler::LT);
2336 // i = h if (key >= array[h].fast_match())
2337 __ csel(i, h, i, Assembler::GE);
2338 // while (i+1 < j)
2339 __ bind(entry);
2340 __ addw(h, i, 1); // i+1
2341 __ cmpw(h, j); // i+1 < j
2342 __ br(Assembler::LT, loop);
2343 }
2344
2345 // end of binary search, result index is i (must check again!)
2346 Label default_case;
2347 // Convert array[i].match to native byte-ordering before compare
2348 __ ldr(temp, Address(array, i, Address::lsl(3)));
2349 __ rev32(temp, temp);
2350 __ cmpw(key, temp);
2351 __ br(Assembler::NE, default_case);
2352
2353 // entry found -> j = offset
2354 __ add(j, array, i, ext::uxtx, 3);
2355 __ ldrw(j, Address(j, BytesPerInt));
2356 __ profile_switch_case(i, key, array);
2357 __ rev32(j, j);
2358 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2359 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2360 __ dispatch_only(vtos, /*generate_poll*/true);
2361
2362 // default case -> j = default offset
2363 __ bind(default_case);
2364 __ profile_switch_default(i);
2365 __ ldrw(j, Address(array, -2 * BytesPerInt));
2366 __ rev32(j, j);
2367 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2368 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2369 __ dispatch_only(vtos, /*generate_poll*/true);
2370 }
2371
2372
2373 void TemplateTable::_return(TosState state)
2374 {
2375 transition(state, state);
2376 assert(_desc->calls_vm(),
2377 "inconsistent calls_vm information"); // call in remove_activation
2378
2379 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2380 assert(state == vtos, "only valid state");
2381
2382 __ ldr(c_rarg1, aaddress(0));
2383 __ load_klass(r3, c_rarg1);
2384 __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2385 Label skip_register_finalizer;
2386 __ tbz(r3, exact_log2(JVM_ACC_HAS_FINALIZER), skip_register_finalizer);
2387
2388 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2389
2390 __ bind(skip_register_finalizer);
2391 }
2392
2393 // Issue a StoreStore barrier after all stores but before return
2394 // from any constructor for any class with a final field. We don't
2395 // know if this is a finalizer, so we always do so.
2396 if (_desc->bytecode() == Bytecodes::_return)
2397 __ membar(MacroAssembler::StoreStore);
2398
2399 // Narrow result if state is itos but result type is smaller.
2400 // Need to narrow in the return bytecode rather than in generate_return_entry
2401 // since compiled code callers expect the result to already be narrowed.
2402 if (state == itos) {
2403 __ narrow(r0);
2404 }
2405
2406 __ remove_activation(state);
2407 __ ret(lr);
2408 }
2409
2410 // ----------------------------------------------------------------------------
2411 // Volatile variables demand their effects be made known to all CPU's
2412 // in order. Store buffers on most chips allow reads & writes to
2413 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2414 // without some kind of memory barrier (i.e., it's not sufficient that
2415 // the interpreter does not reorder volatile references, the hardware
2416 // also must not reorder them).
2417 //
2418 // According to the new Java Memory Model (JMM):
2419 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2420 // writes act as aquire & release, so:
2421 // (2) A read cannot let unrelated NON-volatile memory refs that
2422 // happen after the read float up to before the read. It's OK for
2423 // non-volatile memory refs that happen before the volatile read to
2424 // float down below it.
2425 // (3) Similar a volatile write cannot let unrelated NON-volatile
2426 // memory refs that happen BEFORE the write float down to after the
2427 // write. It's OK for non-volatile memory refs that happen after the
2428 // volatile write to float up before it.
2429 //
2430 // We only put in barriers around volatile refs (they are expensive),
2431 // not _between_ memory refs (that would require us to track the
2432 // flavor of the previous memory refs). Requirements (2) and (3)
2433 // require some barriers before volatile stores and after volatile
2434 // loads. These nearly cover requirement (1) but miss the
2435 // volatile-store-volatile-load case. This final case is placed after
2436 // volatile-stores although it could just as well go before
2437 // volatile-loads.
2438
2439 void TemplateTable::resolve_cache_and_index(int byte_no,
2440 Register Rcache,
2441 Register index,
2442 size_t index_size) {
2443 const Register temp = r19;
2444 assert_different_registers(Rcache, index, temp);
2445
2446 Label resolved;
2447
2448 Bytecodes::Code code = bytecode();
2449 switch (code) {
2450 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2451 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2452 default: break;
2453 }
2454
2455 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2456 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2457 __ subs(zr, temp, (int) code); // have we resolved this bytecode?
2458 __ br(Assembler::EQ, resolved);
2459
2460 // resolve first time through
2461 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2462 __ mov(temp, (int) code);
2463 __ call_VM(noreg, entry, temp);
2464
2465 // Update registers with resolved info
2466 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2467 // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2468 // so all clients ofthis method must be modified accordingly
2469 __ bind(resolved);
2470 }
2471
2472 // The Rcache and index registers must be set before call
2473 // n.b unlike x86 cache already includes the index offset
2474 void TemplateTable::load_field_cp_cache_entry(Register obj,
2475 Register cache,
2476 Register index,
2477 Register off,
2478 Register flags,
2479 bool is_static = false) {
2480 assert_different_registers(cache, index, flags, off);
2481
2482 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2483 // Field offset
2484 __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2485 ConstantPoolCacheEntry::f2_offset())));
2486 // Flags
2487 __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2488 ConstantPoolCacheEntry::flags_offset())));
2489
2490 // klass overwrite register
2491 if (is_static) {
2492 __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2493 ConstantPoolCacheEntry::f1_offset())));
2494 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2495 __ ldr(obj, Address(obj, mirror_offset));
2496 __ resolve_oop_handle(obj);
2497 }
2498 }
2499
2500 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2501 Register method,
2502 Register itable_index,
2503 Register flags,
2504 bool is_invokevirtual,
2505 bool is_invokevfinal, /*unused*/
2506 bool is_invokedynamic) {
2507 // setup registers
2508 const Register cache = rscratch2;
2509 const Register index = r4;
2510 assert_different_registers(method, flags);
2511 assert_different_registers(method, cache, index);
2512 assert_different_registers(itable_index, flags);
2513 assert_different_registers(itable_index, cache, index);
2514 // determine constant pool cache field offsets
2515 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2516 const int method_offset = in_bytes(
2517 ConstantPoolCache::base_offset() +
2518 (is_invokevirtual
2519 ? ConstantPoolCacheEntry::f2_offset()
2520 : ConstantPoolCacheEntry::f1_offset()));
2521 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2522 ConstantPoolCacheEntry::flags_offset());
2523 // access constant pool cache fields
2524 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2525 ConstantPoolCacheEntry::f2_offset());
2526
2527 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2528 resolve_cache_and_index(byte_no, cache, index, index_size);
2529 __ ldr(method, Address(cache, method_offset));
2530
2531 if (itable_index != noreg) {
2532 __ ldr(itable_index, Address(cache, index_offset));
2533 }
2534 __ ldrw(flags, Address(cache, flags_offset));
2535 }
2536
2537
2538 // The registers cache and index expected to be set before call.
2539 // Correct values of the cache and index registers are preserved.
2540 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2541 bool is_static, bool has_tos) {
2542 // do the JVMTI work here to avoid disturbing the register state below
2543 // We use c_rarg registers here because we want to use the register used in
2544 // the call to the VM
2545 if (JvmtiExport::can_post_field_access()) {
2546 // Check to see if a field access watch has been set before we
2547 // take the time to call into the VM.
2548 Label L1;
2549 assert_different_registers(cache, index, r0);
2550 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2551 __ ldrw(r0, Address(rscratch1));
2552 __ cbzw(r0, L1);
2553
2554 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2555 __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2556
2557 if (is_static) {
2558 __ mov(c_rarg1, zr); // NULL object reference
2559 } else {
2560 __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2561 __ verify_oop(c_rarg1);
2562 }
2563 // c_rarg1: object pointer or NULL
2564 // c_rarg2: cache entry pointer
2565 // c_rarg3: jvalue object on the stack
2566 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2567 InterpreterRuntime::post_field_access),
2568 c_rarg1, c_rarg2, c_rarg3);
2569 __ get_cache_and_index_at_bcp(cache, index, 1);
2570 __ bind(L1);
2571 }
2572 }
2573
2574 void TemplateTable::pop_and_check_object(Register r)
2575 {
2576 __ pop_ptr(r);
2577 __ null_check(r); // for field access must check obj.
2578 __ verify_oop(r);
2579 }
2580
2581 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2582 {
2583 const Register cache = r2;
2584 const Register index = r3;
2585 const Register obj = r4;
2586 const Register off = r19;
2587 const Register flags = r0;
2588 const Register raw_flags = r6;
2589 const Register bc = r4; // uses same reg as obj, so don't mix them
2590
2591 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2592 jvmti_post_field_access(cache, index, is_static, false);
2593 load_field_cp_cache_entry(obj, cache, index, off, raw_flags, is_static);
2594
2595 if (!is_static) {
2596 // obj is on the stack
2597 pop_and_check_object(obj);
2598 }
2599
2600 // 8179954: We need to make sure that the code generated for
2601 // volatile accesses forms a sequentially-consistent set of
2602 // operations when combined with STLR and LDAR. Without a leading
2603 // membar it's possible for a simple Dekker test to fail if loads
2604 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
2605 // the stores in one method and we interpret the loads in another.
2606 if (! UseBarriersForVolatile) {
2607 Label notVolatile;
2608 __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2609 __ membar(MacroAssembler::AnyAny);
2610 __ bind(notVolatile);
2611 }
2612
2613 const Address field(obj, off);
2614
2615 Label Done, notByte, notBool, notInt, notShort, notChar,
2616 notLong, notFloat, notObj, notDouble;
2617
2618 // x86 uses a shift and mask or wings it with a shift plus assert
2619 // the mask is not needed. aarch64 just uses bitfield extract
2620 __ ubfxw(flags, raw_flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
2621
2622 assert(btos == 0, "change code, btos != 0");
2623 __ cbnz(flags, notByte);
2624
2625 // Don't rewrite getstatic, only getfield
2626 if (is_static) rc = may_not_rewrite;
2627
2628 // btos
2629 __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
2630 __ push(btos);
2631 // Rewrite bytecode to be faster
2632 if (rc == may_rewrite) {
2633 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2634 }
2635 __ b(Done);
2636
2637 __ bind(notByte);
2638 __ cmp(flags, (u1)ztos);
2639 __ br(Assembler::NE, notBool);
2640
2641 // ztos (same code as btos)
2642 __ access_load_at(T_BOOLEAN, IN_HEAP, r0, field, noreg, noreg);
2643 __ push(ztos);
2644 // Rewrite bytecode to be faster
2645 if (rc == may_rewrite) {
2646 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2647 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2648 }
2649 __ b(Done);
2650
2651 __ bind(notBool);
2652 __ cmp(flags, (u1)atos);
2653 __ br(Assembler::NE, notObj);
2654 // atos
2655 if (!EnableValhalla) {
2656 do_oop_load(_masm, field, r0, IN_HEAP);
2657 __ push(atos);
2658 if (rc == may_rewrite) {
2659 patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2660 }
2661 __ b(Done);
2662 } else { // Valhalla
2663
2664 if (is_static) {
2665 __ load_heap_oop(r0, field);
2666 Label isFlattenable, isUninitialized;
2667 // Issue below if the static field has not been initialized yet
2668 __ test_field_is_flattenable(raw_flags, r8 /*temp*/, isFlattenable);
2669 // Not flattenable case
2670 __ push(atos);
2671 __ b(Done);
2672 // Flattenable case, must not return null even if uninitialized
2673 __ bind(isFlattenable);
2674 __ cbz(r0, isUninitialized);
2675 __ push(atos);
2676 __ b(Done);
2677 __ bind(isUninitialized);
2678 __ andw(raw_flags, raw_flags, ConstantPoolCacheEntry::field_index_mask);
2679 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_static_value_field), obj, raw_flags);
2680 __ verify_oop(r0);
2681 __ push(atos);
2682 __ b(Done);
2683 } else {
2684 Label isFlattened, isInitialized, isFlattenable, rewriteFlattenable;
2685 __ test_field_is_flattenable(raw_flags, r8 /*temp*/, isFlattenable);
2686 // Non-flattenable field case, also covers the object case
2687 __ load_heap_oop(r0, field);
2688 __ push(atos);
2689 if (rc == may_rewrite) {
2690 patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2691 }
2692 __ b(Done);
2693 __ bind(isFlattenable);
2694 __ test_field_is_flattened(raw_flags, r8 /* temp */, isFlattened);
2695 // Non-flattened field case
2696 __ load_heap_oop(r0, field);
2697 __ cbnz(r0, isInitialized);
2698 __ andw(raw_flags, raw_flags, ConstantPoolCacheEntry::field_index_mask);
2699 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_instance_value_field), obj, raw_flags);
2700 __ bind(isInitialized);
2701 __ verify_oop(r0);
2702 __ push(atos);
2703 __ b(rewriteFlattenable);
2704 __ bind(isFlattened);
2705 __ ldr(r10, Address(cache, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f1_offset())));
2706 __ andw(raw_flags, raw_flags, ConstantPoolCacheEntry::field_index_mask);
2707 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flattened_field), obj, raw_flags, r10);
2708 __ verify_oop(r0);
2709 __ push(atos);
2710 __ bind(rewriteFlattenable);
2711 if (rc == may_rewrite) {
2712 patch_bytecode(Bytecodes::_fast_qgetfield, bc, r1);
2713 }
2714 __ b(Done);
2715 }
2716 }
2717
2718 __ bind(notObj);
2719 __ cmp(flags, (u1)itos);
2720 __ br(Assembler::NE, notInt);
2721 // itos
2722 __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
2723 __ push(itos);
2724 // Rewrite bytecode to be faster
2725 if (rc == may_rewrite) {
2726 patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2727 }
2728 __ b(Done);
2729
2730 __ bind(notInt);
2731 __ cmp(flags, (u1)ctos);
2732 __ br(Assembler::NE, notChar);
2733 // ctos
2734 __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
2735 __ push(ctos);
2736 // Rewrite bytecode to be faster
2737 if (rc == may_rewrite) {
2738 patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2739 }
2740 __ b(Done);
2741
2742 __ bind(notChar);
2743 __ cmp(flags, (u1)stos);
2744 __ br(Assembler::NE, notShort);
2745 // stos
2746 __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
2747 __ push(stos);
2748 // Rewrite bytecode to be faster
2749 if (rc == may_rewrite) {
2750 patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2751 }
2752 __ b(Done);
2753
2754 __ bind(notShort);
2755 __ cmp(flags, (u1)ltos);
2756 __ br(Assembler::NE, notLong);
2757 // ltos
2758 __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
2759 __ push(ltos);
2760 // Rewrite bytecode to be faster
2761 if (rc == may_rewrite) {
2762 patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2763 }
2764 __ b(Done);
2765
2766 __ bind(notLong);
2767 __ cmp(flags, (u1)ftos);
2768 __ br(Assembler::NE, notFloat);
2769 // ftos
2770 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2771 __ push(ftos);
2772 // Rewrite bytecode to be faster
2773 if (rc == may_rewrite) {
2774 patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2775 }
2776 __ b(Done);
2777
2778 __ bind(notFloat);
2779 #ifdef ASSERT
2780 __ cmp(flags, (u1)dtos);
2781 __ br(Assembler::NE, notDouble);
2782 #endif
2783 // dtos
2784 __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2785 __ push(dtos);
2786 // Rewrite bytecode to be faster
2787 if (rc == may_rewrite) {
2788 patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2789 }
2790 #ifdef ASSERT
2791 __ b(Done);
2792
2793 __ bind(notDouble);
2794 __ stop("Bad state");
2795 #endif
2796
2797 __ bind(Done);
2798
2799 Label notVolatile;
2800 __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2801 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2802 __ bind(notVolatile);
2803 }
2804
2805
2806 void TemplateTable::getfield(int byte_no)
2807 {
2808 getfield_or_static(byte_no, false);
2809 }
2810
2811 void TemplateTable::nofast_getfield(int byte_no) {
2812 getfield_or_static(byte_no, false, may_not_rewrite);
2813 }
2814
2815 void TemplateTable::getstatic(int byte_no)
2816 {
2817 getfield_or_static(byte_no, true);
2818 }
2819
2820 // The registers cache and index expected to be set before call.
2821 // The function may destroy various registers, just not the cache and index registers.
2822 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2823 transition(vtos, vtos);
2824
2825 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2826
2827 if (JvmtiExport::can_post_field_modification()) {
2828 // Check to see if a field modification watch has been set before
2829 // we take the time to call into the VM.
2830 Label L1;
2831 assert_different_registers(cache, index, r0);
2832 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2833 __ ldrw(r0, Address(rscratch1));
2834 __ cbz(r0, L1);
2835
2836 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2837
2838 if (is_static) {
2839 // Life is simple. Null out the object pointer.
2840 __ mov(c_rarg1, zr);
2841 } else {
2842 // Life is harder. The stack holds the value on top, followed by
2843 // the object. We don't know the size of the value, though; it
2844 // could be one or two words depending on its type. As a result,
2845 // we must find the type to determine where the object is.
2846 __ ldrw(c_rarg3, Address(c_rarg2,
2847 in_bytes(cp_base_offset +
2848 ConstantPoolCacheEntry::flags_offset())));
2849 __ lsr(c_rarg3, c_rarg3,
2850 ConstantPoolCacheEntry::tos_state_shift);
2851 ConstantPoolCacheEntry::verify_tos_state_shift();
2852 Label nope2, done, ok;
2853 __ ldr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2854 __ cmpw(c_rarg3, ltos);
2855 __ br(Assembler::EQ, ok);
2856 __ cmpw(c_rarg3, dtos);
2857 __ br(Assembler::NE, nope2);
2858 __ bind(ok);
2859 __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2860 __ bind(nope2);
2861 }
2862 // cache entry pointer
2863 __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2864 // object (tos)
2865 __ mov(c_rarg3, esp);
2866 // c_rarg1: object pointer set up above (NULL if static)
2867 // c_rarg2: cache entry pointer
2868 // c_rarg3: jvalue object on the stack
2869 __ call_VM(noreg,
2870 CAST_FROM_FN_PTR(address,
2871 InterpreterRuntime::post_field_modification),
2872 c_rarg1, c_rarg2, c_rarg3);
2873 __ get_cache_and_index_at_bcp(cache, index, 1);
2874 __ bind(L1);
2875 }
2876 }
2877
2878 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2879 transition(vtos, vtos);
2880
2881 const Register cache = r2;
2882 const Register index = r3;
2883 const Register obj = r2;
2884 const Register off = r19;
2885 const Register flags = r0;
2886 const Register flags2 = r6;
2887 const Register bc = r4;
2888
2889 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2890 jvmti_post_field_mod(cache, index, is_static);
2891 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2892
2893 Label Done;
2894 __ mov(r5, flags);
2895
2896 {
2897 Label notVolatile;
2898 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2899 __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
2900 __ bind(notVolatile);
2901 }
2902
2903 // field address
2904 const Address field(obj, off);
2905
2906 Label notByte, notBool, notInt, notShort, notChar,
2907 notLong, notFloat, notObj, notDouble;
2908
2909 __ mov(flags2, flags);
2910
2911 // x86 uses a shift and mask or wings it with a shift plus assert
2912 // the mask is not needed. aarch64 just uses bitfield extract
2913 __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
2914
2915 assert(btos == 0, "change code, btos != 0");
2916 __ cbnz(flags, notByte);
2917
2918 // Don't rewrite putstatic, only putfield
2919 if (is_static) rc = may_not_rewrite;
2920
2921 // btos
2922 {
2923 __ pop(btos);
2924 if (!is_static) pop_and_check_object(obj);
2925 __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg);
2926 if (rc == may_rewrite) {
2927 patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2928 }
2929 __ b(Done);
2930 }
2931
2932 __ bind(notByte);
2933 __ cmp(flags, (u1)ztos);
2934 __ br(Assembler::NE, notBool);
2935
2936 // ztos
2937 {
2938 __ pop(ztos);
2939 if (!is_static) pop_and_check_object(obj);
2940 __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg);
2941 if (rc == may_rewrite) {
2942 patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2943 }
2944 __ b(Done);
2945 }
2946
2947 __ bind(notBool);
2948 __ cmp(flags, (u1)atos);
2949 __ br(Assembler::NE, notObj);
2950
2951 // atos
2952 {
2953 if (!EnableValhalla) {
2954 __ pop(atos);
2955 if (!is_static) pop_and_check_object(obj);
2956 // Store into the field
2957 do_oop_store(_masm, field, r0, IN_HEAP);
2958 if (rc == may_rewrite) {
2959 patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2960 }
2961 __ b(Done);
2962 } else { // Valhalla
2963
2964 __ pop(atos);
2965 if (is_static) {
2966 Label notFlattenable;
2967 __ test_field_is_not_flattenable(flags2, r8 /* temp */, notFlattenable);
2968 __ null_check(r0);
2969 __ bind(notFlattenable);
2970 do_oop_store(_masm, field, r0, IN_HEAP);
2971 __ b(Done);
2972 } else {
2973 Label isFlattenable, isFlattened, notBuffered, notBuffered2, rewriteNotFlattenable, rewriteFlattenable;
2974 __ test_field_is_flattenable(flags2, r8 /*temp*/, isFlattenable);
2975 // Not flattenable case, covers not flattenable values and objects
2976 pop_and_check_object(obj);
2977 // Store into the field
2978 do_oop_store(_masm, field, r0, IN_HEAP);
2979 __ bind(rewriteNotFlattenable);
2980 if (rc == may_rewrite) {
2981 patch_bytecode(Bytecodes::_fast_aputfield, bc, r19, true, byte_no);
2982 }
2983 __ b(Done);
2984 // Implementation of the flattenable semantic
2985 __ bind(isFlattenable);
2986 __ null_check(r0);
2987 __ test_field_is_flattened(flags2, r8 /*temp*/, isFlattened);
2988 // Not flattened case
2989 pop_and_check_object(obj);
2990 // Store into the field
2991 do_oop_store(_masm, field, r0, IN_HEAP);
2992 __ b(rewriteFlattenable);
2993 __ bind(isFlattened);
2994 pop_and_check_object(obj);
2995 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_flattened_value), r0, off, obj);
2996 __ bind(rewriteFlattenable);
2997 if (rc == may_rewrite) {
2998 patch_bytecode(Bytecodes::_fast_qputfield, bc, r19, true, byte_no);
2999 }
3000 __ b(Done);
3001 }
3002 } // Valhalla
3003 }
3004
3005 __ bind(notObj);
3006 __ cmp(flags, (u1)itos);
3007 __ br(Assembler::NE, notInt);
3008
3009 // itos
3010 {
3011 __ pop(itos);
3012 if (!is_static) pop_and_check_object(obj);
3013 __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg);
3014 if (rc == may_rewrite) {
3015 patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
3016 }
3017 __ b(Done);
3018 }
3019
3020 __ bind(notInt);
3021 __ cmp(flags, (u1)ctos);
3022 __ br(Assembler::NE, notChar);
3023
3024 // ctos
3025 {
3026 __ pop(ctos);
3027 if (!is_static) pop_and_check_object(obj);
3028 __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg);
3029 if (rc == may_rewrite) {
3030 patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
3031 }
3032 __ b(Done);
3033 }
3034
3035 __ bind(notChar);
3036 __ cmp(flags, (u1)stos);
3037 __ br(Assembler::NE, notShort);
3038
3039 // stos
3040 {
3041 __ pop(stos);
3042 if (!is_static) pop_and_check_object(obj);
3043 __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg);
3044 if (rc == may_rewrite) {
3045 patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
3046 }
3047 __ b(Done);
3048 }
3049
3050 __ bind(notShort);
3051 __ cmp(flags, (u1)ltos);
3052 __ br(Assembler::NE, notLong);
3053
3054 // ltos
3055 {
3056 __ pop(ltos);
3057 if (!is_static) pop_and_check_object(obj);
3058 __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg);
3059 if (rc == may_rewrite) {
3060 patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
3061 }
3062 __ b(Done);
3063 }
3064
3065 __ bind(notLong);
3066 __ cmp(flags, (u1)ftos);
3067 __ br(Assembler::NE, notFloat);
3068
3069 // ftos
3070 {
3071 __ pop(ftos);
3072 if (!is_static) pop_and_check_object(obj);
3073 __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg);
3074 if (rc == may_rewrite) {
3075 patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
3076 }
3077 __ b(Done);
3078 }
3079
3080 __ bind(notFloat);
3081 #ifdef ASSERT
3082 __ cmp(flags, (u1)dtos);
3083 __ br(Assembler::NE, notDouble);
3084 #endif
3085
3086 // dtos
3087 {
3088 __ pop(dtos);
3089 if (!is_static) pop_and_check_object(obj);
3090 __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg);
3091 if (rc == may_rewrite) {
3092 patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
3093 }
3094 }
3095
3096 #ifdef ASSERT
3097 __ b(Done);
3098
3099 __ bind(notDouble);
3100 __ stop("Bad state");
3101 #endif
3102
3103 __ bind(Done);
3104
3105 {
3106 Label notVolatile;
3107 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3108 __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
3109 __ bind(notVolatile);
3110 }
3111 }
3112
3113 void TemplateTable::putfield(int byte_no)
3114 {
3115 putfield_or_static(byte_no, false);
3116 }
3117
3118 void TemplateTable::nofast_putfield(int byte_no) {
3119 putfield_or_static(byte_no, false, may_not_rewrite);
3120 }
3121
3122 void TemplateTable::putstatic(int byte_no) {
3123 putfield_or_static(byte_no, true);
3124 }
3125
3126 void TemplateTable::jvmti_post_fast_field_mod()
3127 {
3128 if (JvmtiExport::can_post_field_modification()) {
3129 // Check to see if a field modification watch has been set before
3130 // we take the time to call into the VM.
3131 Label L2;
3132 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3133 __ ldrw(c_rarg3, Address(rscratch1));
3134 __ cbzw(c_rarg3, L2);
3135 __ pop_ptr(r19); // copy the object pointer from tos
3136 __ verify_oop(r19);
3137 __ push_ptr(r19); // put the object pointer back on tos
3138 // Save tos values before call_VM() clobbers them. Since we have
3139 // to do it for every data type, we use the saved values as the
3140 // jvalue object.
3141 switch (bytecode()) { // load values into the jvalue object
3142 case Bytecodes::_fast_qputfield: //fall through
3143 case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
3144 case Bytecodes::_fast_bputfield: // fall through
3145 case Bytecodes::_fast_zputfield: // fall through
3146 case Bytecodes::_fast_sputfield: // fall through
3147 case Bytecodes::_fast_cputfield: // fall through
3148 case Bytecodes::_fast_iputfield: __ push_i(r0); break;
3149 case Bytecodes::_fast_dputfield: __ push_d(); break;
3150 case Bytecodes::_fast_fputfield: __ push_f(); break;
3151 case Bytecodes::_fast_lputfield: __ push_l(r0); break;
3152
3153 default:
3154 ShouldNotReachHere();
3155 }
3156 __ mov(c_rarg3, esp); // points to jvalue on the stack
3157 // access constant pool cache entry
3158 __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
3159 __ verify_oop(r19);
3160 // r19: object pointer copied above
3161 // c_rarg2: cache entry pointer
3162 // c_rarg3: jvalue object on the stack
3163 __ call_VM(noreg,
3164 CAST_FROM_FN_PTR(address,
3165 InterpreterRuntime::post_field_modification),
3166 r19, c_rarg2, c_rarg3);
3167
3168 switch (bytecode()) { // restore tos values
3169 case Bytecodes::_fast_qputfield: //fall through
3170 case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
3171 case Bytecodes::_fast_bputfield: // fall through
3172 case Bytecodes::_fast_zputfield: // fall through
3173 case Bytecodes::_fast_sputfield: // fall through
3174 case Bytecodes::_fast_cputfield: // fall through
3175 case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
3176 case Bytecodes::_fast_dputfield: __ pop_d(); break;
3177 case Bytecodes::_fast_fputfield: __ pop_f(); break;
3178 case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
3179 default: break;
3180 }
3181 __ bind(L2);
3182 }
3183 }
3184
3185 void TemplateTable::fast_storefield(TosState state)
3186 {
3187 transition(state, vtos);
3188
3189 ByteSize base = ConstantPoolCache::base_offset();
3190
3191 jvmti_post_fast_field_mod();
3192
3193 // access constant pool cache
3194 __ get_cache_and_index_at_bcp(r2, r1, 1);
3195
3196 // test for volatile with r3
3197 __ ldrw(r3, Address(r2, in_bytes(base +
3198 ConstantPoolCacheEntry::flags_offset())));
3199
3200 // replace index with field offset from cache entry
3201 __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3202
3203 {
3204 Label notVolatile;
3205 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3206 __ membar(MacroAssembler::StoreStore | MacroAssembler::LoadStore);
3207 __ bind(notVolatile);
3208 }
3209
3210 Label notVolatile;
3211
3212 // Get object from stack
3213 pop_and_check_object(r2);
3214
3215 // field address
3216 const Address field(r2, r1);
3217
3218 // access field
3219 switch (bytecode()) {
3220 case Bytecodes::_fast_qputfield: //fall through
3221 {
3222 Label isFlattened, done;
3223 __ null_check(r0);
3224 __ test_field_is_flattened(r3, r8 /* temp */, isFlattened);
3225 // No Flattened case
3226 do_oop_store(_masm, field, r0, IN_HEAP);
3227 __ b(done);
3228 __ bind(isFlattened);
3229 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_flattened_value), r0, r1, r2);
3230 __ bind(done);
3231 }
3232 break;
3233 case Bytecodes::_fast_aputfield:
3234 do_oop_store(_masm, field, r0, IN_HEAP);
3235 break;
3236 case Bytecodes::_fast_lputfield:
3237 __ access_store_at(T_LONG, IN_HEAP, field, r0, noreg, noreg);
3238 break;
3239 case Bytecodes::_fast_iputfield:
3240 __ access_store_at(T_INT, IN_HEAP, field, r0, noreg, noreg);
3241 break;
3242 case Bytecodes::_fast_zputfield:
3243 __ access_store_at(T_BOOLEAN, IN_HEAP, field, r0, noreg, noreg);
3244 break;
3245 case Bytecodes::_fast_bputfield:
3246 __ access_store_at(T_BYTE, IN_HEAP, field, r0, noreg, noreg);
3247 break;
3248 case Bytecodes::_fast_sputfield:
3249 __ access_store_at(T_SHORT, IN_HEAP, field, r0, noreg, noreg);
3250 break;
3251 case Bytecodes::_fast_cputfield:
3252 __ access_store_at(T_CHAR, IN_HEAP, field, r0, noreg, noreg);
3253 break;
3254 case Bytecodes::_fast_fputfield:
3255 __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg);
3256 break;
3257 case Bytecodes::_fast_dputfield:
3258 __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg);
3259 break;
3260 default:
3261 ShouldNotReachHere();
3262 }
3263
3264 {
3265 Label notVolatile;
3266 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3267 __ membar(MacroAssembler::StoreLoad | MacroAssembler::StoreStore);
3268 __ bind(notVolatile);
3269 }
3270 }
3271
3272
3273 void TemplateTable::fast_accessfield(TosState state)
3274 {
3275 transition(atos, state);
3276 // Do the JVMTI work here to avoid disturbing the register state below
3277 if (JvmtiExport::can_post_field_access()) {
3278 // Check to see if a field access watch has been set before we
3279 // take the time to call into the VM.
3280 Label L1;
3281 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3282 __ ldrw(r2, Address(rscratch1));
3283 __ cbzw(r2, L1);
3284 // access constant pool cache entry
3285 __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
3286 __ verify_oop(r0);
3287 __ push_ptr(r0); // save object pointer before call_VM() clobbers it
3288 __ mov(c_rarg1, r0);
3289 // c_rarg1: object pointer copied above
3290 // c_rarg2: cache entry pointer
3291 __ call_VM(noreg,
3292 CAST_FROM_FN_PTR(address,
3293 InterpreterRuntime::post_field_access),
3294 c_rarg1, c_rarg2);
3295 __ pop_ptr(r0); // restore object pointer
3296 __ bind(L1);
3297 }
3298
3299 // access constant pool cache
3300 __ get_cache_and_index_at_bcp(r2, r1, 1);
3301 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3302 ConstantPoolCacheEntry::f2_offset())));
3303 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3304 ConstantPoolCacheEntry::flags_offset())));
3305
3306 // r0: object
3307 __ verify_oop(r0);
3308 __ null_check(r0);
3309 const Address field(r0, r1);
3310
3311 // 8179954: We need to make sure that the code generated for
3312 // volatile accesses forms a sequentially-consistent set of
3313 // operations when combined with STLR and LDAR. Without a leading
3314 // membar it's possible for a simple Dekker test to fail if loads
3315 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
3316 // the stores in one method and we interpret the loads in another.
3317 if (! UseBarriersForVolatile) {
3318 Label notVolatile;
3319 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3320 __ membar(MacroAssembler::AnyAny);
3321 __ bind(notVolatile);
3322 }
3323
3324 // access field
3325 switch (bytecode()) {
3326 case Bytecodes::_fast_qgetfield:
3327 {
3328 Label isFlattened, isInitialized, Done;
3329 // FIXME: We don't need to reload registers multiple times, but stay close to x86 code
3330 __ ldrw(r9, Address(r2, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset())));
3331 __ test_field_is_flattened(r9, r8 /* temp */, isFlattened);
3332 // Non-flattened field case
3333 __ mov(r9, r0);
3334 __ load_heap_oop(r0, field);
3335 __ cbnz(r0, isInitialized);
3336 __ mov(r0, r9);
3337 __ ldrw(r9, Address(r2, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset())));
3338 __ andw(r9, r9, ConstantPoolCacheEntry::field_index_mask);
3339 __ call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_instance_value_field), r0, r9);
3340 __ bind(isInitialized);
3341 __ verify_oop(r0);
3342 __ b(Done);
3343 __ bind(isFlattened);
3344 __ ldrw(r9, Address(r2, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset())));
3345 __ andw(r9, r9, ConstantPoolCacheEntry::field_index_mask);
3346 __ ldr(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f1_offset())));
3347 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flattened_field), r0, r9, r3);
3348 __ verify_oop(r0);
3349 __ bind(Done);
3350 }
3351 break;
3352 case Bytecodes::_fast_agetfield:
3353 do_oop_load(_masm, field, r0, IN_HEAP);
3354 __ verify_oop(r0);
3355 break;
3356 case Bytecodes::_fast_lgetfield:
3357 __ access_load_at(T_LONG, IN_HEAP, r0, field, noreg, noreg);
3358 break;
3359 case Bytecodes::_fast_igetfield:
3360 __ access_load_at(T_INT, IN_HEAP, r0, field, noreg, noreg);
3361 break;
3362 case Bytecodes::_fast_bgetfield:
3363 __ access_load_at(T_BYTE, IN_HEAP, r0, field, noreg, noreg);
3364 break;
3365 case Bytecodes::_fast_sgetfield:
3366 __ access_load_at(T_SHORT, IN_HEAP, r0, field, noreg, noreg);
3367 break;
3368 case Bytecodes::_fast_cgetfield:
3369 __ access_load_at(T_CHAR, IN_HEAP, r0, field, noreg, noreg);
3370 break;
3371 case Bytecodes::_fast_fgetfield:
3372 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3373 break;
3374 case Bytecodes::_fast_dgetfield:
3375 __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3376 break;
3377 default:
3378 ShouldNotReachHere();
3379 }
3380 {
3381 Label notVolatile;
3382 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3383 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3384 __ bind(notVolatile);
3385 }
3386 }
3387
3388 void TemplateTable::fast_xaccess(TosState state)
3389 {
3390 transition(vtos, state);
3391
3392 // get receiver
3393 __ ldr(r0, aaddress(0));
3394 // access constant pool cache
3395 __ get_cache_and_index_at_bcp(r2, r3, 2);
3396 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3397 ConstantPoolCacheEntry::f2_offset())));
3398
3399 // 8179954: We need to make sure that the code generated for
3400 // volatile accesses forms a sequentially-consistent set of
3401 // operations when combined with STLR and LDAR. Without a leading
3402 // membar it's possible for a simple Dekker test to fail if loads
3403 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
3404 // the stores in one method and we interpret the loads in another.
3405 if (! UseBarriersForVolatile) {
3406 Label notVolatile;
3407 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3408 ConstantPoolCacheEntry::flags_offset())));
3409 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3410 __ membar(MacroAssembler::AnyAny);
3411 __ bind(notVolatile);
3412 }
3413
3414 // make sure exception is reported in correct bcp range (getfield is
3415 // next instruction)
3416 __ increment(rbcp);
3417 __ null_check(r0);
3418 switch (state) {
3419 case itos:
3420 __ access_load_at(T_INT, IN_HEAP, r0, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3421 break;
3422 case atos:
3423 do_oop_load(_masm, Address(r0, r1, Address::lsl(0)), r0, IN_HEAP);
3424 __ verify_oop(r0);
3425 break;
3426 case ftos:
3427 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, Address(r0, r1, Address::lsl(0)), noreg, noreg);
3428 break;
3429 default:
3430 ShouldNotReachHere();
3431 }
3432
3433 {
3434 Label notVolatile;
3435 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3436 ConstantPoolCacheEntry::flags_offset())));
3437 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3438 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3439 __ bind(notVolatile);
3440 }
3441
3442 __ decrement(rbcp);
3443 }
3444
3445
3446
3447 //-----------------------------------------------------------------------------
3448 // Calls
3449
3450 void TemplateTable::count_calls(Register method, Register temp)
3451 {
3452 __ call_Unimplemented();
3453 }
3454
3455 void TemplateTable::prepare_invoke(int byte_no,
3456 Register method, // linked method (or i-klass)
3457 Register index, // itable index, MethodType, etc.
3458 Register recv, // if caller wants to see it
3459 Register flags // if caller wants to test it
3460 ) {
3461 // determine flags
3462 Bytecodes::Code code = bytecode();
3463 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3464 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3465 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3466 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3467 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3468 const bool load_receiver = (recv != noreg);
3469 const bool save_flags = (flags != noreg);
3470 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3471 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3472 assert(flags == noreg || flags == r3, "");
3473 assert(recv == noreg || recv == r2, "");
3474
3475 // setup registers & access constant pool cache
3476 if (recv == noreg) recv = r2;
3477 if (flags == noreg) flags = r3;
3478 assert_different_registers(method, index, recv, flags);
3479
3480 // save 'interpreter return address'
3481 __ save_bcp();
3482
3483 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3484
3485 // maybe push appendix to arguments (just before return address)
3486 if (is_invokedynamic || is_invokehandle) {
3487 Label L_no_push;
3488 __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3489 // Push the appendix as a trailing parameter.
3490 // This must be done before we get the receiver,
3491 // since the parameter_size includes it.
3492 __ push(r19);
3493 __ mov(r19, index);
3494 __ load_resolved_reference_at_index(index, r19);
3495 __ pop(r19);
3496 __ push(index); // push appendix (MethodType, CallSite, etc.)
3497 __ bind(L_no_push);
3498 }
3499
3500 // load receiver if needed (note: no return address pushed yet)
3501 if (load_receiver) {
3502 __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3503 // FIXME -- is this actually correct? looks like it should be 2
3504 // const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3505 // const int receiver_is_at_end = -1; // back off one slot to get receiver
3506 // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3507 // __ movptr(recv, recv_addr);
3508 __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3509 __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3510 __ verify_oop(recv);
3511 }
3512
3513 // compute return type
3514 // x86 uses a shift and mask or wings it with a shift plus assert
3515 // the mask is not needed. aarch64 just uses bitfield extract
3516 __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
3517 // load return address
3518 {
3519 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3520 __ mov(rscratch1, table_addr);
3521 __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3522 }
3523 }
3524
3525
3526 void TemplateTable::invokevirtual_helper(Register index,
3527 Register recv,
3528 Register flags)
3529 {
3530 // Uses temporary registers r0, r3
3531 assert_different_registers(index, recv, r0, r3);
3532 // Test for an invoke of a final method
3533 Label notFinal;
3534 __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3535
3536 const Register method = index; // method must be rmethod
3537 assert(method == rmethod,
3538 "methodOop must be rmethod for interpreter calling convention");
3539
3540 // do the call - the index is actually the method to call
3541 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3542
3543 // It's final, need a null check here!
3544 __ null_check(recv);
3545
3546 // profile this call
3547 __ profile_final_call(r0);
3548 __ profile_arguments_type(r0, method, r4, true);
3549
3550 __ jump_from_interpreted(method, r0);
3551
3552 __ bind(notFinal);
3553
3554 // get receiver klass
3555 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3556 __ load_klass(r0, recv);
3557
3558 // profile this call
3559 __ profile_virtual_call(r0, rlocals, r3);
3560
3561 // get target methodOop & entry point
3562 __ lookup_virtual_method(r0, index, method);
3563 __ profile_arguments_type(r3, method, r4, true);
3564 // FIXME -- this looks completely redundant. is it?
3565 // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3566 __ jump_from_interpreted(method, r3);
3567 }
3568
3569 void TemplateTable::invokevirtual(int byte_no)
3570 {
3571 transition(vtos, vtos);
3572 assert(byte_no == f2_byte, "use this argument");
3573
3574 prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3575
3576 // rmethod: index (actually a Method*)
3577 // r2: receiver
3578 // r3: flags
3579
3580 invokevirtual_helper(rmethod, r2, r3);
3581 }
3582
3583 void TemplateTable::invokespecial(int byte_no)
3584 {
3585 transition(vtos, vtos);
3586 assert(byte_no == f1_byte, "use this argument");
3587
3588 prepare_invoke(byte_no, rmethod, noreg, // get f1 Method*
3589 r2); // get receiver also for null check
3590 __ verify_oop(r2);
3591 __ null_check(r2);
3592 // do the call
3593 __ profile_call(r0);
3594 __ profile_arguments_type(r0, rmethod, rbcp, false);
3595 __ jump_from_interpreted(rmethod, r0);
3596 }
3597
3598 void TemplateTable::invokestatic(int byte_no)
3599 {
3600 transition(vtos, vtos);
3601 assert(byte_no == f1_byte, "use this argument");
3602
3603 prepare_invoke(byte_no, rmethod); // get f1 Method*
3604 // do the call
3605 __ profile_call(r0);
3606 __ profile_arguments_type(r0, rmethod, r4, false);
3607 __ jump_from_interpreted(rmethod, r0);
3608 }
3609
3610 void TemplateTable::fast_invokevfinal(int byte_no)
3611 {
3612 __ call_Unimplemented();
3613 }
3614
3615 void TemplateTable::invokeinterface(int byte_no) {
3616 transition(vtos, vtos);
3617 assert(byte_no == f1_byte, "use this argument");
3618
3619 prepare_invoke(byte_no, r0, rmethod, // get f1 Klass*, f2 Method*
3620 r2, r3); // recv, flags
3621
3622 // r0: interface klass (from f1)
3623 // rmethod: method (from f2)
3624 // r2: receiver
3625 // r3: flags
3626
3627 // First check for Object case, then private interface method,
3628 // then regular interface method.
3629
3630 // Special case of invokeinterface called for virtual method of
3631 // java.lang.Object. See cpCache.cpp for details.
3632 Label notObjectMethod;
3633 __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notObjectMethod);
3634
3635 invokevirtual_helper(rmethod, r2, r3);
3636 __ bind(notObjectMethod);
3637
3638 Label no_such_interface;
3639
3640 // Check for private method invocation - indicated by vfinal
3641 Label notVFinal;
3642 __ tbz(r3, ConstantPoolCacheEntry::is_vfinal_shift, notVFinal);
3643
3644 // Get receiver klass into r3 - also a null check
3645 __ null_check(r2, oopDesc::klass_offset_in_bytes());
3646 __ load_klass(r3, r2);
3647
3648 Label subtype;
3649 __ check_klass_subtype(r3, r0, r4, subtype);
3650 // If we get here the typecheck failed
3651 __ b(no_such_interface);
3652 __ bind(subtype);
3653
3654 __ profile_final_call(r0);
3655 __ profile_arguments_type(r0, rmethod, r4, true);
3656 __ jump_from_interpreted(rmethod, r0);
3657
3658 __ bind(notVFinal);
3659
3660 // Get receiver klass into r3 - also a null check
3661 __ restore_locals();
3662 __ null_check(r2, oopDesc::klass_offset_in_bytes());
3663 __ load_klass(r3, r2);
3664
3665 Label no_such_method;
3666
3667 // Preserve method for throw_AbstractMethodErrorVerbose.
3668 __ mov(r16, rmethod);
3669 // Receiver subtype check against REFC.
3670 // Superklass in r0. Subklass in r3. Blows rscratch2, r13
3671 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3672 r3, r0, noreg,
3673 // outputs: scan temp. reg, scan temp. reg
3674 rscratch2, r13,
3675 no_such_interface,
3676 /*return_method=*/false);
3677
3678 // profile this call
3679 __ profile_virtual_call(r3, r13, r19);
3680
3681 // Get declaring interface class from method, and itable index
3682 __ ldr(r0, Address(rmethod, Method::const_offset()));
3683 __ ldr(r0, Address(r0, ConstMethod::constants_offset()));
3684 __ ldr(r0, Address(r0, ConstantPool::pool_holder_offset_in_bytes()));
3685 __ ldrw(rmethod, Address(rmethod, Method::itable_index_offset()));
3686 __ subw(rmethod, rmethod, Method::itable_index_max);
3687 __ negw(rmethod, rmethod);
3688
3689 // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3690 __ mov(rlocals, r3);
3691 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3692 rlocals, r0, rmethod,
3693 // outputs: method, scan temp. reg
3694 rmethod, r13,
3695 no_such_interface);
3696
3697 // rmethod,: methodOop to call
3698 // r2: receiver
3699 // Check for abstract method error
3700 // Note: This should be done more efficiently via a throw_abstract_method_error
3701 // interpreter entry point and a conditional jump to it in case of a null
3702 // method.
3703 __ cbz(rmethod, no_such_method);
3704
3705 __ profile_arguments_type(r3, rmethod, r13, true);
3706
3707 // do the call
3708 // r2: receiver
3709 // rmethod,: methodOop
3710 __ jump_from_interpreted(rmethod, r3);
3711 __ should_not_reach_here();
3712
3713 // exception handling code follows...
3714 // note: must restore interpreter registers to canonical
3715 // state for exception handling to work correctly!
3716
3717 __ bind(no_such_method);
3718 // throw exception
3719 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3720 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3721 // Pass arguments for generating a verbose error message.
3722 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose), r3, r16);
3723 // the call_VM checks for exception, so we should never return here.
3724 __ should_not_reach_here();
3725
3726 __ bind(no_such_interface);
3727 // throw exception
3728 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3729 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3730 // Pass arguments for generating a verbose error message.
3731 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3732 InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose), r3, r0);
3733 // the call_VM checks for exception, so we should never return here.
3734 __ should_not_reach_here();
3735 return;
3736 }
3737
3738 void TemplateTable::invokehandle(int byte_no) {
3739 transition(vtos, vtos);
3740 assert(byte_no == f1_byte, "use this argument");
3741
3742 prepare_invoke(byte_no, rmethod, r0, r2);
3743 __ verify_method_ptr(r2);
3744 __ verify_oop(r2);
3745 __ null_check(r2);
3746
3747 // FIXME: profile the LambdaForm also
3748
3749 // r13 is safe to use here as a scratch reg because it is about to
3750 // be clobbered by jump_from_interpreted().
3751 __ profile_final_call(r13);
3752 __ profile_arguments_type(r13, rmethod, r4, true);
3753
3754 __ jump_from_interpreted(rmethod, r0);
3755 }
3756
3757 void TemplateTable::invokedynamic(int byte_no) {
3758 transition(vtos, vtos);
3759 assert(byte_no == f1_byte, "use this argument");
3760
3761 prepare_invoke(byte_no, rmethod, r0);
3762
3763 // r0: CallSite object (from cpool->resolved_references[])
3764 // rmethod: MH.linkToCallSite method (from f2)
3765
3766 // Note: r0_callsite is already pushed by prepare_invoke
3767
3768 // %%% should make a type profile for any invokedynamic that takes a ref argument
3769 // profile this call
3770 __ profile_call(rbcp);
3771 __ profile_arguments_type(r3, rmethod, r13, false);
3772
3773 __ verify_oop(r0);
3774
3775 __ jump_from_interpreted(rmethod, r0);
3776 }
3777
3778
3779 //-----------------------------------------------------------------------------
3780 // Allocation
3781
3782 void TemplateTable::_new() {
3783 transition(vtos, atos);
3784
3785 __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3786 Label slow_case;
3787 Label done;
3788 Label initialize_header;
3789 Label initialize_object; // including clearing the fields
3790
3791 __ get_cpool_and_tags(r4, r0);
3792 // Make sure the class we're about to instantiate has been resolved.
3793 // This is done before loading InstanceKlass to be consistent with the order
3794 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3795 const int tags_offset = Array<u1>::base_offset_in_bytes();
3796 __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3797 __ lea(rscratch1, Address(rscratch1, tags_offset));
3798 __ ldarb(rscratch1, rscratch1);
3799 __ cmp(rscratch1, (u1)JVM_CONSTANT_Class);
3800 __ br(Assembler::NE, slow_case);
3801
3802 // get InstanceKlass
3803 __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
3804
3805 // make sure klass is initialized & doesn't have finalizer
3806 // make sure klass is fully initialized
3807 __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3808 __ cmp(rscratch1, (u1)InstanceKlass::fully_initialized);
3809 __ br(Assembler::NE, slow_case);
3810
3811 // get instance_size in InstanceKlass (scaled to a count of bytes)
3812 __ ldrw(r3,
3813 Address(r4,
3814 Klass::layout_helper_offset()));
3815 // test to see if it has a finalizer or is malformed in some way
3816 __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3817
3818 // Allocate the instance:
3819 // If TLAB is enabled:
3820 // Try to allocate in the TLAB.
3821 // If fails, go to the slow path.
3822 // Else If inline contiguous allocations are enabled:
3823 // Try to allocate in eden.
3824 // If fails due to heap end, go to slow path.
3825 //
3826 // If TLAB is enabled OR inline contiguous is enabled:
3827 // Initialize the allocation.
3828 // Exit.
3829 //
3830 // Go to slow path.
3831 const bool allow_shared_alloc =
3832 Universe::heap()->supports_inline_contig_alloc();
3833
3834 if (UseTLAB) {
3835 __ tlab_allocate(r0, r3, 0, noreg, r1, slow_case);
3836
3837 if (ZeroTLAB) {
3838 // the fields have been already cleared
3839 __ b(initialize_header);
3840 } else {
3841 // initialize both the header and fields
3842 __ b(initialize_object);
3843 }
3844 } else {
3845 // Allocation in the shared Eden, if allowed.
3846 //
3847 // r3: instance size in bytes
3848 if (allow_shared_alloc) {
3849 __ eden_allocate(r0, r3, 0, r10, slow_case);
3850 }
3851 }
3852
3853 // If UseTLAB or allow_shared_alloc are true, the object is created above and
3854 // there is an initialize need. Otherwise, skip and go to the slow path.
3855 if (UseTLAB || allow_shared_alloc) {
3856 // The object is initialized before the header. If the object size is
3857 // zero, go directly to the header initialization.
3858 __ bind(initialize_object);
3859 __ sub(r3, r3, sizeof(oopDesc));
3860 __ cbz(r3, initialize_header);
3861
3862 // Initialize object fields
3863 {
3864 __ add(r2, r0, sizeof(oopDesc));
3865 Label loop;
3866 __ bind(loop);
3867 __ str(zr, Address(__ post(r2, BytesPerLong)));
3868 __ sub(r3, r3, BytesPerLong);
3869 __ cbnz(r3, loop);
3870 }
3871
3872 // initialize object header only.
3873 __ bind(initialize_header);
3874 if (UseBiasedLocking) {
3875 __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3876 } else {
3877 __ mov(rscratch1, (intptr_t)markOopDesc::prototype());
3878 }
3879 __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3880 __ store_klass_gap(r0, zr); // zero klass gap for compressed oops
3881 __ store_klass(r0, r4); // store klass last
3882
3883 {
3884 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3885 // Trigger dtrace event for fastpath
3886 __ push(atos); // save the return value
3887 __ call_VM_leaf(
3888 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3889 __ pop(atos); // restore the return value
3890
3891 }
3892 __ b(done);
3893 }
3894
3895 // slow case
3896 __ bind(slow_case);
3897 __ get_constant_pool(c_rarg1);
3898 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3899 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3900 __ verify_oop(r0);
3901
3902 // continue
3903 __ bind(done);
3904 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3905 __ membar(Assembler::StoreStore);
3906 }
3907
3908 void TemplateTable::defaultvalue() {
3909 transition(vtos, atos);
3910 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3911 __ get_constant_pool(c_rarg1);
3912 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::defaultvalue),
3913 c_rarg1, c_rarg2);
3914 __ verify_oop(r0);
3915 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3916 __ membar(Assembler::StoreStore);
3917 }
3918
3919 void TemplateTable::withfield() {
3920 transition(vtos, atos);
3921 resolve_cache_and_index(f2_byte, c_rarg1 /*cache*/, c_rarg2 /*index*/, sizeof(u2));
3922
3923 // n.b. unlike x86 cache is now rcpool plus the indexed offset
3924 // so using rcpool to meet shared code expectations
3925
3926 call_VM(r1, CAST_FROM_FN_PTR(address, InterpreterRuntime::withfield), rcpool);
3927 __ verify_oop(r1);
3928 __ add(esp, esp, r0);
3929 __ mov(r0, r1);
3930 }
3931
3932 void TemplateTable::newarray() {
3933 transition(itos, atos);
3934 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3935 __ mov(c_rarg2, r0);
3936 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3937 c_rarg1, c_rarg2);
3938 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3939 __ membar(Assembler::StoreStore);
3940 }
3941
3942 void TemplateTable::anewarray() {
3943 transition(itos, atos);
3944 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3945 __ get_constant_pool(c_rarg1);
3946 __ mov(c_rarg3, r0);
3947 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3948 c_rarg1, c_rarg2, c_rarg3);
3949 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3950 __ membar(Assembler::StoreStore);
3951 }
3952
3953 void TemplateTable::arraylength() {
3954 transition(atos, itos);
3955 __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3956 __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3957 }
3958
3959 void TemplateTable::checkcast()
3960 {
3961 transition(atos, atos);
3962 Label done, is_null, ok_is_subtype, quicked, resolved;
3963 __ cbz(r0, is_null);
3964
3965 // Get cpool & tags index
3966 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3967 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3968 // See if bytecode has already been quicked
3969 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3970 __ lea(r1, Address(rscratch1, r19));
3971 __ ldarb(r1, r1);
3972 __ cmp(r1, (u1)JVM_CONSTANT_Class);
3973 __ br(Assembler::EQ, quicked);
3974
3975 __ push(atos); // save receiver for result, and for GC
3976 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3977 // vm_result_2 has metadata result
3978 __ get_vm_result_2(r0, rthread);
3979 __ pop(r3); // restore receiver
3980 __ b(resolved);
3981
3982 // Get superklass in r0 and subklass in r3
3983 __ bind(quicked);
3984 __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3985 __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass
3986
3987 __ bind(resolved);
3988 __ load_klass(r19, r3);
3989
3990 // Generate subtype check. Blows r2, r5. Object in r3.
3991 // Superklass in r0. Subklass in r19.
3992 __ gen_subtype_check(r19, ok_is_subtype);
3993
3994 // Come here on failure
3995 __ push(r3);
3996 // object is at TOS
3997 __ b(Interpreter::_throw_ClassCastException_entry);
3998
3999 // Come here on success
4000 __ bind(ok_is_subtype);
4001 __ mov(r0, r3); // Restore object in r3
4002
4003 __ b(done);
4004 __ bind(is_null);
4005
4006 // Collect counts on whether this test sees NULLs a lot or not.
4007 if (ProfileInterpreter) {
4008 __ profile_null_seen(r2);
4009 }
4010
4011 if (EnableValhalla) {
4012 // Get cpool & tags index
4013 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
4014 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
4015 // See if bytecode has already been quicked
4016 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
4017 __ lea(r1, Address(rscratch1, r19));
4018 __ ldarb(r1, r1);
4019 // See if CP entry is a Q-descriptor
4020 __ andr (r1, r1, JVM_CONSTANT_QDescBit);
4021 __ cmp(r1, (u1) JVM_CONSTANT_QDescBit);
4022 __ br(Assembler::NE, done);
4023 __ b(ExternalAddress(Interpreter::_throw_NullPointerException_entry));
4024 }
4025
4026 __ bind(done);
4027 }
4028
4029 void TemplateTable::instanceof() {
4030 transition(atos, itos);
4031 Label done, is_null, ok_is_subtype, quicked, resolved;
4032 __ cbz(r0, is_null);
4033
4034 // Get cpool & tags index
4035 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
4036 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
4037 // See if bytecode has already been quicked
4038 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
4039 __ lea(r1, Address(rscratch1, r19));
4040 __ ldarb(r1, r1);
4041 __ cmp(r1, (u1)JVM_CONSTANT_Class);
4042 __ br(Assembler::EQ, quicked);
4043
4044 __ push(atos); // save receiver for result, and for GC
4045 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4046 // vm_result_2 has metadata result
4047 __ get_vm_result_2(r0, rthread);
4048 __ pop(r3); // restore receiver
4049 __ verify_oop(r3);
4050 __ load_klass(r3, r3);
4051 __ b(resolved);
4052
4053 // Get superklass in r0 and subklass in r3
4054 __ bind(quicked);
4055 __ load_klass(r3, r0);
4056 __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1);
4057
4058 __ bind(resolved);
4059
4060 // Generate subtype check. Blows r2, r5
4061 // Superklass in r0. Subklass in r3.
4062 __ gen_subtype_check(r3, ok_is_subtype);
4063
4064 // Come here on failure
4065 __ mov(r0, 0);
4066 __ b(done);
4067 // Come here on success
4068 __ bind(ok_is_subtype);
4069 __ mov(r0, 1);
4070
4071 // Collect counts on whether this test sees NULLs a lot or not.
4072 if (ProfileInterpreter) {
4073 __ b(done);
4074 __ bind(is_null);
4075 __ profile_null_seen(r2);
4076 } else {
4077 __ bind(is_null); // same as 'done'
4078 }
4079 __ bind(done);
4080 // r0 = 0: obj == NULL or obj is not an instanceof the specified klass
4081 // r0 = 1: obj != NULL and obj is an instanceof the specified klass
4082 }
4083
4084 //-----------------------------------------------------------------------------
4085 // Breakpoints
4086 void TemplateTable::_breakpoint() {
4087 // Note: We get here even if we are single stepping..
4088 // jbug inists on setting breakpoints at every bytecode
4089 // even if we are in single step mode.
4090
4091 transition(vtos, vtos);
4092
4093 // get the unpatched byte code
4094 __ get_method(c_rarg1);
4095 __ call_VM(noreg,
4096 CAST_FROM_FN_PTR(address,
4097 InterpreterRuntime::get_original_bytecode_at),
4098 c_rarg1, rbcp);
4099 __ mov(r19, r0);
4100
4101 // post the breakpoint event
4102 __ call_VM(noreg,
4103 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4104 rmethod, rbcp);
4105
4106 // complete the execution of original bytecode
4107 __ mov(rscratch1, r19);
4108 __ dispatch_only_normal(vtos);
4109 }
4110
4111 //-----------------------------------------------------------------------------
4112 // Exceptions
4113
4114 void TemplateTable::athrow() {
4115 transition(atos, vtos);
4116 __ null_check(r0);
4117 __ b(Interpreter::throw_exception_entry());
4118 }
4119
4120 //-----------------------------------------------------------------------------
4121 // Synchronization
4122 //
4123 // Note: monitorenter & exit are symmetric routines; which is reflected
4124 // in the assembly code structure as well
4125 //
4126 // Stack layout:
4127 //
4128 // [expressions ] <--- esp = expression stack top
4129 // ..
4130 // [expressions ]
4131 // [monitor entry] <--- monitor block top = expression stack bot
4132 // ..
4133 // [monitor entry]
4134 // [frame data ] <--- monitor block bot
4135 // ...
4136 // [saved rbp ] <--- rbp
4137 void TemplateTable::monitorenter()
4138 {
4139 transition(atos, vtos);
4140
4141 // check for NULL object
4142 __ null_check(r0);
4143
4144 __ resolve(IS_NOT_NULL, r0);
4145
4146 const Address monitor_block_top(
4147 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4148 const Address monitor_block_bot(
4149 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
4150 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4151
4152 Label allocated;
4153
4154 // initialize entry pointer
4155 __ mov(c_rarg1, zr); // points to free slot or NULL
4156
4157 // find a free slot in the monitor block (result in c_rarg1)
4158 {
4159 Label entry, loop, exit;
4160 __ ldr(c_rarg3, monitor_block_top); // points to current entry,
4161 // starting with top-most entry
4162 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
4163
4164 __ b(entry);
4165
4166 __ bind(loop);
4167 // check if current entry is used
4168 // if not used then remember entry in c_rarg1
4169 __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
4170 __ cmp(zr, rscratch1);
4171 __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
4172 // check if current entry is for same object
4173 __ cmp(r0, rscratch1);
4174 // if same object then stop searching
4175 __ br(Assembler::EQ, exit);
4176 // otherwise advance to next entry
4177 __ add(c_rarg3, c_rarg3, entry_size);
4178 __ bind(entry);
4179 // check if bottom reached
4180 __ cmp(c_rarg3, c_rarg2);
4181 // if not at bottom then check this entry
4182 __ br(Assembler::NE, loop);
4183 __ bind(exit);
4184 }
4185
4186 __ cbnz(c_rarg1, allocated); // check if a slot has been found and
4187 // if found, continue with that on
4188
4189 // allocate one if there's no free slot
4190 {
4191 Label entry, loop;
4192 // 1. compute new pointers // rsp: old expression stack top
4193 __ ldr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
4194 __ sub(esp, esp, entry_size); // move expression stack top
4195 __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
4196 __ mov(c_rarg3, esp); // set start value for copy loop
4197 __ str(c_rarg1, monitor_block_bot); // set new monitor block bottom
4198
4199 __ sub(sp, sp, entry_size); // make room for the monitor
4200
4201 __ b(entry);
4202 // 2. move expression stack contents
4203 __ bind(loop);
4204 __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
4205 // word from old location
4206 __ str(c_rarg2, Address(c_rarg3, 0)); // and store it at new location
4207 __ add(c_rarg3, c_rarg3, wordSize); // advance to next word
4208 __ bind(entry);
4209 __ cmp(c_rarg3, c_rarg1); // check if bottom reached
4210 __ br(Assembler::NE, loop); // if not at bottom then
4211 // copy next word
4212 }
4213
4214 // call run-time routine
4215 // c_rarg1: points to monitor entry
4216 __ bind(allocated);
4217
4218 // Increment bcp to point to the next bytecode, so exception
4219 // handling for async. exceptions work correctly.
4220 // The object has already been poped from the stack, so the
4221 // expression stack looks correct.
4222 __ increment(rbcp);
4223
4224 // store object
4225 __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
4226 __ lock_object(c_rarg1);
4227
4228 // check to make sure this monitor doesn't cause stack overflow after locking
4229 __ save_bcp(); // in case of exception
4230 __ generate_stack_overflow_check(0);
4231
4232 // The bcp has already been incremented. Just need to dispatch to
4233 // next instruction.
4234 __ dispatch_next(vtos);
4235 }
4236
4237
4238 void TemplateTable::monitorexit()
4239 {
4240 transition(atos, vtos);
4241
4242 // check for NULL object
4243 __ null_check(r0);
4244
4245 __ resolve(IS_NOT_NULL, r0);
4246
4247 const Address monitor_block_top(
4248 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4249 const Address monitor_block_bot(
4250 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
4251 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4252
4253 Label found;
4254
4255 // find matching slot
4256 {
4257 Label entry, loop;
4258 __ ldr(c_rarg1, monitor_block_top); // points to current entry,
4259 // starting with top-most entry
4260 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
4261 // of monitor block
4262 __ b(entry);
4263
4264 __ bind(loop);
4265 // check if current entry is for same object
4266 __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
4267 __ cmp(r0, rscratch1);
4268 // if same object then stop searching
4269 __ br(Assembler::EQ, found);
4270 // otherwise advance to next entry
4271 __ add(c_rarg1, c_rarg1, entry_size);
4272 __ bind(entry);
4273 // check if bottom reached
4274 __ cmp(c_rarg1, c_rarg2);
4275 // if not at bottom then check this entry
4276 __ br(Assembler::NE, loop);
4277 }
4278
4279 // error handling. Unlocking was not block-structured
4280 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4281 InterpreterRuntime::throw_illegal_monitor_state_exception));
4282 __ should_not_reach_here();
4283
4284 // call run-time routine
4285 __ bind(found);
4286 __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
4287 __ unlock_object(c_rarg1);
4288 __ pop_ptr(r0); // discard object
4289 }
4290
4291
4292 // Wide instructions
4293 void TemplateTable::wide()
4294 {
4295 __ load_unsigned_byte(r19, at_bcp(1));
4296 __ mov(rscratch1, (address)Interpreter::_wentry_point);
4297 __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
4298 __ br(rscratch1);
4299 }
4300
4301
4302 // Multi arrays
4303 void TemplateTable::multianewarray() {
4304 transition(vtos, atos);
4305 __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
4306 // last dim is on top of stack; we want address of first one:
4307 // first_addr = last_addr + (ndims - 1) * wordSize
4308 __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
4309 __ sub(c_rarg1, c_rarg1, wordSize);
4310 call_VM(r0,
4311 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
4312 c_rarg1);
4313 __ load_unsigned_byte(r1, at_bcp(3));
4314 __ lea(esp, Address(esp, r1, Address::uxtw(3)));
4315 }