1 /*
2 * Copyright (c) 2003, 2015, 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.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "interpreter/interp_masm.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "memory/universe.inline.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/methodHandles.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/stubRoutines.hpp"
40 #include "runtime/synchronizer.hpp"
41
42 #ifndef CC_INTERP
43
44 #define __ _masm->
45
46 // Platform-dependent initialization
47
48 void TemplateTable::pd_initialize() {
49 // No 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 obj,
147 Register val,
148 BarrierSet::Name barrier,
149 bool precise) {
150 assert(val == noreg || val == r0, "parameter is just for looks");
151 switch (barrier) {
152 #if INCLUDE_ALL_GCS
153 case BarrierSet::G1SATBCTLogging:
154 {
155 // flatten object address if needed
156 if (obj.index() == noreg && obj.offset() == 0) {
157 if (obj.base() != r3) {
158 __ mov(r3, obj.base());
159 }
160 } else {
161 __ lea(r3, obj);
162 }
163 __ g1_write_barrier_pre(r3 /* obj */,
164 r1 /* pre_val */,
165 rthread /* thread */,
166 r10 /* tmp */,
167 val != noreg /* tosca_live */,
168 false /* expand_call */);
169 if (val == noreg) {
170 __ store_heap_oop_null(Address(r3, 0));
171 } else {
172 // G1 barrier needs uncompressed oop for region cross check.
173 Register new_val = val;
174 if (UseCompressedOops) {
175 new_val = rscratch1;
176 __ mov(new_val, val);
177 }
178 __ store_heap_oop(Address(r3, 0), val);
179 __ g1_write_barrier_post(r3 /* store_adr */,
180 new_val /* new_val */,
181 rthread /* thread */,
182 r10 /* tmp */,
183 r1 /* tmp2 */);
184 }
185
186 }
187 break;
188 #endif // INCLUDE_ALL_GCS
189 case BarrierSet::CardTableForRS:
190 case BarrierSet::CardTableExtension:
191 {
192 if (val == noreg) {
193 __ store_heap_oop_null(obj);
194 } else {
195 __ store_heap_oop(obj, val);
196 // flatten object address if needed
197 if (!precise || (obj.index() == noreg && obj.offset() == 0)) {
198 __ store_check(obj.base());
199 } else {
200 __ lea(r3, obj);
201 __ store_check(r3);
202 }
203 }
204 }
205 break;
206 case BarrierSet::ModRef:
207 if (val == noreg) {
208 __ store_heap_oop_null(obj);
209 } else {
210 __ store_heap_oop(obj, val);
211 }
212 break;
213 default :
214 ShouldNotReachHere();
215
216 }
217 }
218
219 Address TemplateTable::at_bcp(int offset) {
220 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
221 return Address(rbcp, offset);
222 }
223
224 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
225 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
226 int byte_no)
227 {
228 if (!RewriteBytecodes) return;
229 Label L_patch_done;
230
231 switch (bc) {
232 case Bytecodes::_fast_aputfield:
233 case Bytecodes::_fast_bputfield:
234 case Bytecodes::_fast_cputfield:
235 case Bytecodes::_fast_dputfield:
236 case Bytecodes::_fast_fputfield:
237 case Bytecodes::_fast_iputfield:
238 case Bytecodes::_fast_lputfield:
239 case Bytecodes::_fast_sputfield:
240 {
241 // We skip bytecode quickening for putfield instructions when
242 // the put_code written to the constant pool cache is zero.
243 // This is required so that every execution of this instruction
244 // calls out to InterpreterRuntime::resolve_get_put to do
245 // additional, required work.
246 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
247 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
248 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
249 __ movw(bc_reg, bc);
250 __ cmpw(temp_reg, (unsigned) 0);
251 __ br(Assembler::EQ, L_patch_done); // don't patch
252 }
253 break;
254 default:
255 assert(byte_no == -1, "sanity");
256 // the pair bytecodes have already done the load.
257 if (load_bc_into_bc_reg) {
258 __ movw(bc_reg, bc);
259 }
260 }
261
262 if (JvmtiExport::can_post_breakpoint()) {
263 Label L_fast_patch;
264 // if a breakpoint is present we can't rewrite the stream directly
265 __ load_unsigned_byte(temp_reg, at_bcp(0));
266 __ cmpw(temp_reg, Bytecodes::_breakpoint);
267 __ br(Assembler::NE, L_fast_patch);
268 // Let breakpoint table handling rewrite to quicker bytecode
269 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
270 __ b(L_patch_done);
271 __ bind(L_fast_patch);
272 }
273
274 #ifdef ASSERT
275 Label L_okay;
276 __ load_unsigned_byte(temp_reg, at_bcp(0));
277 __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
278 __ br(Assembler::EQ, L_okay);
279 __ cmpw(temp_reg, bc_reg);
280 __ br(Assembler::EQ, L_okay);
281 __ stop("patching the wrong bytecode");
282 __ bind(L_okay);
283 #endif
284
285 // patch bytecode
286 __ strb(bc_reg, at_bcp(0));
287 __ bind(L_patch_done);
288 }
289
290
291 // Individual instructions
292
293 void TemplateTable::nop() {
294 transition(vtos, vtos);
295 // nothing to do
296 }
297
298 void TemplateTable::shouldnotreachhere() {
299 transition(vtos, vtos);
300 __ stop("shouldnotreachhere bytecode");
301 }
302
303 void TemplateTable::aconst_null()
304 {
305 transition(vtos, atos);
306 __ mov(r0, 0);
307 }
308
309 void TemplateTable::iconst(int value)
310 {
311 transition(vtos, itos);
312 __ mov(r0, value);
313 }
314
315 void TemplateTable::lconst(int value)
316 {
317 __ mov(r0, value);
318 }
319
320 void TemplateTable::fconst(int value)
321 {
322 transition(vtos, ftos);
323 switch (value) {
324 case 0:
325 __ fmovs(v0, zr);
326 break;
327 case 1:
328 __ fmovs(v0, 1.0);
329 break;
330 case 2:
331 __ fmovs(v0, 2.0);
332 break;
333 default:
334 ShouldNotReachHere();
335 break;
336 }
337 }
338
339 void TemplateTable::dconst(int value)
340 {
341 transition(vtos, dtos);
342 switch (value) {
343 case 0:
344 __ fmovd(v0, zr);
345 break;
346 case 1:
347 __ fmovd(v0, 1.0);
348 break;
349 case 2:
350 __ fmovd(v0, 2.0);
351 break;
352 default:
353 ShouldNotReachHere();
354 break;
355 }
356 }
357
358 void TemplateTable::bipush()
359 {
360 transition(vtos, itos);
361 __ load_signed_byte32(r0, at_bcp(1));
362 }
363
364 void TemplateTable::sipush()
365 {
366 transition(vtos, itos);
367 __ load_unsigned_short(r0, at_bcp(1));
368 __ revw(r0, r0);
369 __ asrw(r0, r0, 16);
370 }
371
372 void TemplateTable::ldc(bool wide)
373 {
374 transition(vtos, vtos);
375 Label call_ldc, notFloat, notClass, Done;
376
377 if (wide) {
378 __ get_unsigned_2_byte_index_at_bcp(r1, 1);
379 } else {
380 __ load_unsigned_byte(r1, at_bcp(1));
381 }
382 __ get_cpool_and_tags(r2, r0);
383
384 const int base_offset = ConstantPool::header_size() * wordSize;
385 const int tags_offset = Array<u1>::base_offset_in_bytes();
386
387 // get type
388 __ add(r3, r1, tags_offset);
389 __ lea(r3, Address(r0, r3));
390 __ ldarb(r3, r3);
391
392 // unresolved class - get the resolved class
393 __ cmp(r3, JVM_CONSTANT_UnresolvedClass);
394 __ br(Assembler::EQ, call_ldc);
395
396 // unresolved class in error state - call into runtime to throw the error
397 // from the first resolution attempt
398 __ cmp(r3, JVM_CONSTANT_UnresolvedClassInError);
399 __ br(Assembler::EQ, call_ldc);
400
401 // resolved class - need to call vm to get java mirror of the class
402 __ cmp(r3, JVM_CONSTANT_Class);
403 __ br(Assembler::NE, notClass);
404
405 __ bind(call_ldc);
406 __ mov(c_rarg1, wide);
407 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
408 __ push_ptr(r0);
409 __ verify_oop(r0);
410 __ b(Done);
411
412 __ bind(notClass);
413 __ cmp(r3, JVM_CONSTANT_Float);
414 __ br(Assembler::NE, notFloat);
415 // ftos
416 __ adds(r1, r2, r1, Assembler::LSL, 3);
417 __ ldrs(v0, Address(r1, base_offset));
418 __ push_f();
419 __ b(Done);
420
421 __ bind(notFloat);
422 #ifdef ASSERT
423 {
424 Label L;
425 __ cmp(r3, JVM_CONSTANT_Integer);
426 __ br(Assembler::EQ, L);
427 // String and Object are rewritten to fast_aldc
428 __ stop("unexpected tag type in ldc");
429 __ bind(L);
430 }
431 #endif
432 // itos JVM_CONSTANT_Integer only
433 __ adds(r1, r2, r1, Assembler::LSL, 3);
434 __ ldrw(r0, Address(r1, base_offset));
435 __ push_i(r0);
436 __ bind(Done);
437 }
438
439 // Fast path for caching oop constants.
440 void TemplateTable::fast_aldc(bool wide)
441 {
442 transition(vtos, atos);
443
444 Register result = r0;
445 Register tmp = r1;
446 int index_size = wide ? sizeof(u2) : sizeof(u1);
447
448 Label resolved;
449
450 // We are resolved if the resolved reference cache entry contains a
451 // non-null object (String, MethodType, etc.)
452 assert_different_registers(result, tmp);
453 __ get_cache_index_at_bcp(tmp, 1, index_size);
454 __ load_resolved_reference_at_index(result, tmp);
455 __ cbnz(result, resolved);
456
457 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
458
459 // first time invocation - must resolve first
460 __ mov(tmp, (int)bytecode());
461 __ call_VM(result, entry, tmp);
462
463 __ bind(resolved);
464
465 if (VerifyOops) {
466 __ verify_oop(result);
467 }
468 }
469
470 void TemplateTable::ldc2_w()
471 {
472 transition(vtos, vtos);
473 Label Long, Done;
474 __ get_unsigned_2_byte_index_at_bcp(r0, 1);
475
476 __ get_cpool_and_tags(r1, r2);
477 const int base_offset = ConstantPool::header_size() * wordSize;
478 const int tags_offset = Array<u1>::base_offset_in_bytes();
479
480 // get type
481 __ lea(r2, Address(r2, r0, Address::lsl(0)));
482 __ load_unsigned_byte(r2, Address(r2, tags_offset));
483 __ cmpw(r2, (int)JVM_CONSTANT_Double);
484 __ br(Assembler::NE, Long);
485 // dtos
486 __ lea (r2, Address(r1, r0, Address::lsl(3)));
487 __ ldrd(v0, Address(r2, base_offset));
488 __ push_d();
489 __ b(Done);
490
491 __ bind(Long);
492 // ltos
493 __ lea(r0, Address(r1, r0, Address::lsl(3)));
494 __ ldr(r0, Address(r0, base_offset));
495 __ push_l();
496
497 __ bind(Done);
498 }
499
500 void TemplateTable::locals_index(Register reg, int offset)
501 {
502 __ ldrb(reg, at_bcp(offset));
503 __ neg(reg, reg);
504 }
505
506 void TemplateTable::iload() {
507 iload_internal();
508 }
509
510 void TemplateTable::nofast_iload() {
511 iload_internal(may_not_rewrite);
512 }
513
514 void TemplateTable::iload_internal(RewriteControl rc) {
515 transition(vtos, itos);
516 if (RewriteFrequentPairs && rc == may_rewrite) {
517 Label rewrite, done;
518 Register bc = r4;
519
520 // get next bytecode
521 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
522
523 // if _iload, wait to rewrite to iload2. We only want to rewrite the
524 // last two iloads in a pair. Comparing against fast_iload means that
525 // the next bytecode is neither an iload or a caload, and therefore
526 // an iload pair.
527 __ cmpw(r1, Bytecodes::_iload);
528 __ br(Assembler::EQ, done);
529
530 // if _fast_iload rewrite to _fast_iload2
531 __ cmpw(r1, Bytecodes::_fast_iload);
532 __ movw(bc, Bytecodes::_fast_iload2);
533 __ br(Assembler::EQ, rewrite);
534
535 // if _caload rewrite to _fast_icaload
536 __ cmpw(r1, Bytecodes::_caload);
537 __ movw(bc, Bytecodes::_fast_icaload);
538 __ br(Assembler::EQ, rewrite);
539
540 // else rewrite to _fast_iload
541 __ movw(bc, Bytecodes::_fast_iload);
542
543 // rewrite
544 // bc: new bytecode
545 __ bind(rewrite);
546 patch_bytecode(Bytecodes::_iload, bc, r1, false);
547 __ bind(done);
548
549 }
550
551 // do iload, get the local value into tos
552 locals_index(r1);
553 __ ldr(r0, iaddress(r1));
554
555 }
556
557 void TemplateTable::fast_iload2()
558 {
559 transition(vtos, itos);
560 locals_index(r1);
561 __ ldr(r0, iaddress(r1));
562 __ push(itos);
563 locals_index(r1, 3);
564 __ ldr(r0, iaddress(r1));
565 }
566
567 void TemplateTable::fast_iload()
568 {
569 transition(vtos, itos);
570 locals_index(r1);
571 __ ldr(r0, iaddress(r1));
572 }
573
574 void TemplateTable::lload()
575 {
576 transition(vtos, ltos);
577 __ ldrb(r1, at_bcp(1));
578 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
579 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
580 }
581
582 void TemplateTable::fload()
583 {
584 transition(vtos, ftos);
585 locals_index(r1);
586 // n.b. we use ldrd here because this is a 64 bit slot
587 // this is comparable to the iload case
588 __ ldrd(v0, faddress(r1));
589 }
590
591 void TemplateTable::dload()
592 {
593 transition(vtos, dtos);
594 __ ldrb(r1, at_bcp(1));
595 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
596 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
597 }
598
599 void TemplateTable::aload()
600 {
601 transition(vtos, atos);
602 locals_index(r1);
603 __ ldr(r0, iaddress(r1));
604 }
605
606 void TemplateTable::locals_index_wide(Register reg) {
607 __ ldrh(reg, at_bcp(2));
608 __ rev16w(reg, reg);
609 __ neg(reg, reg);
610 }
611
612 void TemplateTable::wide_iload() {
613 transition(vtos, itos);
614 locals_index_wide(r1);
615 __ ldr(r0, iaddress(r1));
616 }
617
618 void TemplateTable::wide_lload()
619 {
620 transition(vtos, ltos);
621 __ ldrh(r1, at_bcp(2));
622 __ rev16w(r1, r1);
623 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
624 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
625 }
626
627 void TemplateTable::wide_fload()
628 {
629 transition(vtos, ftos);
630 locals_index_wide(r1);
631 // n.b. we use ldrd here because this is a 64 bit slot
632 // this is comparable to the iload case
633 __ ldrd(v0, faddress(r1));
634 }
635
636 void TemplateTable::wide_dload()
637 {
638 transition(vtos, dtos);
639 __ ldrh(r1, at_bcp(2));
640 __ rev16w(r1, r1);
641 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
642 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
643 }
644
645 void TemplateTable::wide_aload()
646 {
647 transition(vtos, atos);
648 locals_index_wide(r1);
649 __ ldr(r0, aaddress(r1));
650 }
651
652 void TemplateTable::index_check(Register array, Register index)
653 {
654 // destroys r1, rscratch1
655 // check array
656 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
657 // sign extend index for use by indexed load
658 // __ movl2ptr(index, index);
659 // check index
660 Register length = rscratch1;
661 __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
662 __ cmpw(index, length);
663 if (index != r1) {
664 // ??? convention: move aberrant index into r1 for exception message
665 assert(r1 != array, "different registers");
666 __ mov(r1, index);
667 }
668 Label ok;
669 __ br(Assembler::LO, ok);
670 __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
671 __ br(rscratch1);
672 __ bind(ok);
673 }
674
675 void TemplateTable::iaload()
676 {
677 transition(itos, itos);
678 __ mov(r1, r0);
679 __ pop_ptr(r0);
680 // r0: array
681 // r1: index
682 index_check(r0, r1); // leaves index in r1, kills rscratch1
683 __ lea(r1, Address(r0, r1, Address::uxtw(2)));
684 __ ldrw(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_INT)));
685 }
686
687 void TemplateTable::laload()
688 {
689 transition(itos, ltos);
690 __ mov(r1, r0);
691 __ pop_ptr(r0);
692 // r0: array
693 // r1: index
694 index_check(r0, r1); // leaves index in r1, kills rscratch1
695 __ lea(r1, Address(r0, r1, Address::uxtw(3)));
696 __ ldr(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_LONG)));
697 }
698
699 void TemplateTable::faload()
700 {
701 transition(itos, ftos);
702 __ mov(r1, r0);
703 __ pop_ptr(r0);
704 // r0: array
705 // r1: index
706 index_check(r0, r1); // leaves index in r1, kills rscratch1
707 __ lea(r1, Address(r0, r1, Address::uxtw(2)));
708 __ ldrs(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
709 }
710
711 void TemplateTable::daload()
712 {
713 transition(itos, dtos);
714 __ mov(r1, r0);
715 __ pop_ptr(r0);
716 // r0: array
717 // r1: index
718 index_check(r0, r1); // leaves index in r1, kills rscratch1
719 __ lea(r1, Address(r0, r1, Address::uxtw(3)));
720 __ ldrd(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
721 }
722
723 void TemplateTable::aaload()
724 {
725 transition(itos, atos);
726 __ mov(r1, r0);
727 __ pop_ptr(r0);
728 // r0: array
729 // r1: index
730 index_check(r0, r1); // leaves index in r1, kills rscratch1
731 int s = (UseCompressedOops ? 2 : 3);
732 __ lea(r1, Address(r0, r1, Address::uxtw(s)));
733 __ load_heap_oop(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
734 }
735
736 void TemplateTable::baload()
737 {
738 transition(itos, itos);
739 __ mov(r1, r0);
740 __ pop_ptr(r0);
741 // r0: array
742 // r1: index
743 index_check(r0, r1); // leaves index in r1, kills rscratch1
744 __ lea(r1, Address(r0, r1, Address::uxtw(0)));
745 __ load_signed_byte(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
746 }
747
748 void TemplateTable::caload()
749 {
750 transition(itos, itos);
751 __ mov(r1, r0);
752 __ pop_ptr(r0);
753 // r0: array
754 // r1: index
755 index_check(r0, r1); // leaves index in r1, kills rscratch1
756 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
757 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
758 }
759
760 // iload followed by caload frequent pair
761 void TemplateTable::fast_icaload()
762 {
763 transition(vtos, itos);
764 // load index out of locals
765 locals_index(r2);
766 __ ldr(r1, iaddress(r2));
767
768 __ pop_ptr(r0);
769
770 // r0: array
771 // r1: index
772 index_check(r0, r1); // leaves index in r1, kills rscratch1
773 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
774 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
775 }
776
777 void TemplateTable::saload()
778 {
779 transition(itos, itos);
780 __ mov(r1, r0);
781 __ pop_ptr(r0);
782 // r0: array
783 // r1: index
784 index_check(r0, r1); // leaves index in r1, kills rscratch1
785 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
786 __ load_signed_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
787 }
788
789 void TemplateTable::iload(int n)
790 {
791 transition(vtos, itos);
792 __ ldr(r0, iaddress(n));
793 }
794
795 void TemplateTable::lload(int n)
796 {
797 transition(vtos, ltos);
798 __ ldr(r0, laddress(n));
799 }
800
801 void TemplateTable::fload(int n)
802 {
803 transition(vtos, ftos);
804 __ ldrs(v0, faddress(n));
805 }
806
807 void TemplateTable::dload(int n)
808 {
809 transition(vtos, dtos);
810 __ ldrd(v0, daddress(n));
811 }
812
813 void TemplateTable::aload(int n)
814 {
815 transition(vtos, atos);
816 __ ldr(r0, iaddress(n));
817 }
818
819 void TemplateTable::aload_0() {
820 aload_0_internal();
821 }
822
823 void TemplateTable::nofast_aload_0() {
824 aload_0_internal(may_not_rewrite);
825 }
826
827 void TemplateTable::aload_0_internal(RewriteControl rc) {
828 // According to bytecode histograms, the pairs:
829 //
830 // _aload_0, _fast_igetfield
831 // _aload_0, _fast_agetfield
832 // _aload_0, _fast_fgetfield
833 //
834 // occur frequently. If RewriteFrequentPairs is set, the (slow)
835 // _aload_0 bytecode checks if the next bytecode is either
836 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
837 // rewrites the current bytecode into a pair bytecode; otherwise it
838 // rewrites the current bytecode into _fast_aload_0 that doesn't do
839 // the pair check anymore.
840 //
841 // Note: If the next bytecode is _getfield, the rewrite must be
842 // delayed, otherwise we may miss an opportunity for a pair.
843 //
844 // Also rewrite frequent pairs
845 // aload_0, aload_1
846 // aload_0, iload_1
847 // These bytecodes with a small amount of code are most profitable
848 // to rewrite
849 if (RewriteFrequentPairs && rc == may_rewrite) {
850 Label rewrite, done;
851 const Register bc = r4;
852
853 // get next bytecode
854 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
855
856 // do actual aload_0
857 aload(0);
858
859 // if _getfield then wait with rewrite
860 __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
861 __ br(Assembler::EQ, done);
862
863 // if _igetfield then reqrite to _fast_iaccess_0
864 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
865 __ cmpw(r1, Bytecodes::_fast_igetfield);
866 __ movw(bc, Bytecodes::_fast_iaccess_0);
867 __ br(Assembler::EQ, rewrite);
868
869 // if _agetfield then reqrite to _fast_aaccess_0
870 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
871 __ cmpw(r1, Bytecodes::_fast_agetfield);
872 __ movw(bc, Bytecodes::_fast_aaccess_0);
873 __ br(Assembler::EQ, rewrite);
874
875 // if _fgetfield then reqrite to _fast_faccess_0
876 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
877 __ cmpw(r1, Bytecodes::_fast_fgetfield);
878 __ movw(bc, Bytecodes::_fast_faccess_0);
879 __ br(Assembler::EQ, rewrite);
880
881 // else rewrite to _fast_aload0
882 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
883 __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
884
885 // rewrite
886 // bc: new bytecode
887 __ bind(rewrite);
888 patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
889
890 __ bind(done);
891 } else {
892 aload(0);
893 }
894 }
895
896 void TemplateTable::istore()
897 {
898 transition(itos, vtos);
899 locals_index(r1);
900 // FIXME: We're being very pernickerty here storing a jint in a
901 // local with strw, which costs an extra instruction over what we'd
902 // be able to do with a simple str. We should just store the whole
903 // word.
904 __ lea(rscratch1, iaddress(r1));
905 __ strw(r0, Address(rscratch1));
906 }
907
908 void TemplateTable::lstore()
909 {
910 transition(ltos, vtos);
911 locals_index(r1);
912 __ str(r0, laddress(r1, rscratch1, _masm));
913 }
914
915 void TemplateTable::fstore() {
916 transition(ftos, vtos);
917 locals_index(r1);
918 __ lea(rscratch1, iaddress(r1));
919 __ strs(v0, Address(rscratch1));
920 }
921
922 void TemplateTable::dstore() {
923 transition(dtos, vtos);
924 locals_index(r1);
925 __ strd(v0, daddress(r1, rscratch1, _masm));
926 }
927
928 void TemplateTable::astore()
929 {
930 transition(vtos, vtos);
931 __ pop_ptr(r0);
932 locals_index(r1);
933 __ str(r0, aaddress(r1));
934 }
935
936 void TemplateTable::wide_istore() {
937 transition(vtos, vtos);
938 __ pop_i();
939 locals_index_wide(r1);
940 __ lea(rscratch1, iaddress(r1));
941 __ strw(r0, Address(rscratch1));
942 }
943
944 void TemplateTable::wide_lstore() {
945 transition(vtos, vtos);
946 __ pop_l();
947 locals_index_wide(r1);
948 __ str(r0, laddress(r1, rscratch1, _masm));
949 }
950
951 void TemplateTable::wide_fstore() {
952 transition(vtos, vtos);
953 __ pop_f();
954 locals_index_wide(r1);
955 __ lea(rscratch1, faddress(r1));
956 __ strs(v0, rscratch1);
957 }
958
959 void TemplateTable::wide_dstore() {
960 transition(vtos, vtos);
961 __ pop_d();
962 locals_index_wide(r1);
963 __ strd(v0, daddress(r1, rscratch1, _masm));
964 }
965
966 void TemplateTable::wide_astore() {
967 transition(vtos, vtos);
968 __ pop_ptr(r0);
969 locals_index_wide(r1);
970 __ str(r0, aaddress(r1));
971 }
972
973 void TemplateTable::iastore() {
974 transition(itos, vtos);
975 __ pop_i(r1);
976 __ pop_ptr(r3);
977 // r0: value
978 // r1: index
979 // r3: array
980 index_check(r3, r1); // prefer index in r1
981 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
982 __ strw(r0, Address(rscratch1,
983 arrayOopDesc::base_offset_in_bytes(T_INT)));
984 }
985
986 void TemplateTable::lastore() {
987 transition(ltos, vtos);
988 __ pop_i(r1);
989 __ pop_ptr(r3);
990 // r0: value
991 // r1: index
992 // r3: array
993 index_check(r3, r1); // prefer index in r1
994 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
995 __ str(r0, Address(rscratch1,
996 arrayOopDesc::base_offset_in_bytes(T_LONG)));
997 }
998
999 void TemplateTable::fastore() {
1000 transition(ftos, vtos);
1001 __ pop_i(r1);
1002 __ pop_ptr(r3);
1003 // v0: value
1004 // r1: index
1005 // r3: array
1006 index_check(r3, r1); // prefer index in r1
1007 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
1008 __ strs(v0, Address(rscratch1,
1009 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1010 }
1011
1012 void TemplateTable::dastore() {
1013 transition(dtos, vtos);
1014 __ pop_i(r1);
1015 __ pop_ptr(r3);
1016 // v0: value
1017 // r1: index
1018 // r3: array
1019 index_check(r3, r1); // prefer index in r1
1020 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
1021 __ strd(v0, Address(rscratch1,
1022 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1023 }
1024
1025 void TemplateTable::aastore() {
1026 Label is_null, ok_is_subtype, done;
1027 transition(vtos, vtos);
1028 // stack: ..., array, index, value
1029 __ ldr(r0, at_tos()); // value
1030 __ ldr(r2, at_tos_p1()); // index
1031 __ ldr(r3, at_tos_p2()); // array
1032
1033 Address element_address(r4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1034
1035 index_check(r3, r2); // kills r1
1036 __ lea(r4, Address(r3, r2, Address::uxtw(UseCompressedOops? 2 : 3)));
1037
1038 // do array store check - check for NULL value first
1039 __ cbz(r0, is_null);
1040
1041 // Move subklass into r1
1042 __ load_klass(r1, r0);
1043 // Move superklass into r0
1044 __ load_klass(r0, r3);
1045 __ ldr(r0, Address(r0,
1046 ObjArrayKlass::element_klass_offset()));
1047 // Compress array + index*oopSize + 12 into a single register. Frees r2.
1048
1049 // Generate subtype check. Blows r2, r5
1050 // Superklass in r0. Subklass in r1.
1051 __ gen_subtype_check(r1, ok_is_subtype);
1052
1053 // Come here on failure
1054 // object is at TOS
1055 __ b(Interpreter::_throw_ArrayStoreException_entry);
1056
1057 // Come here on success
1058 __ bind(ok_is_subtype);
1059
1060 // Get the value we will store
1061 __ ldr(r0, at_tos());
1062 // Now store using the appropriate barrier
1063 do_oop_store(_masm, element_address, r0, _bs->kind(), true);
1064 __ b(done);
1065
1066 // Have a NULL in r0, r3=array, r2=index. Store NULL at ary[idx]
1067 __ bind(is_null);
1068 __ profile_null_seen(r2);
1069
1070 // Store a NULL
1071 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1072
1073 // Pop stack arguments
1074 __ bind(done);
1075 __ add(esp, esp, 3 * Interpreter::stackElementSize);
1076 }
1077
1078 void TemplateTable::bastore()
1079 {
1080 transition(itos, vtos);
1081 __ pop_i(r1);
1082 __ pop_ptr(r3);
1083 // r0: value
1084 // r1: index
1085 // r3: array
1086 index_check(r3, r1); // prefer index in r1
1087 __ lea(rscratch1, Address(r3, r1, Address::uxtw(0)));
1088 __ strb(r0, Address(rscratch1,
1089 arrayOopDesc::base_offset_in_bytes(T_BYTE)));
1090 }
1091
1092 void TemplateTable::castore()
1093 {
1094 transition(itos, vtos);
1095 __ pop_i(r1);
1096 __ pop_ptr(r3);
1097 // r0: value
1098 // r1: index
1099 // r3: array
1100 index_check(r3, r1); // prefer index in r1
1101 __ lea(rscratch1, Address(r3, r1, Address::uxtw(1)));
1102 __ strh(r0, Address(rscratch1,
1103 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1104 }
1105
1106 void TemplateTable::sastore()
1107 {
1108 castore();
1109 }
1110
1111 void TemplateTable::istore(int n)
1112 {
1113 transition(itos, vtos);
1114 __ str(r0, iaddress(n));
1115 }
1116
1117 void TemplateTable::lstore(int n)
1118 {
1119 transition(ltos, vtos);
1120 __ str(r0, laddress(n));
1121 }
1122
1123 void TemplateTable::fstore(int n)
1124 {
1125 transition(ftos, vtos);
1126 __ strs(v0, faddress(n));
1127 }
1128
1129 void TemplateTable::dstore(int n)
1130 {
1131 transition(dtos, vtos);
1132 __ strd(v0, daddress(n));
1133 }
1134
1135 void TemplateTable::astore(int n)
1136 {
1137 transition(vtos, vtos);
1138 __ pop_ptr(r0);
1139 __ str(r0, iaddress(n));
1140 }
1141
1142 void TemplateTable::pop()
1143 {
1144 transition(vtos, vtos);
1145 __ add(esp, esp, Interpreter::stackElementSize);
1146 }
1147
1148 void TemplateTable::pop2()
1149 {
1150 transition(vtos, vtos);
1151 __ add(esp, esp, 2 * Interpreter::stackElementSize);
1152 }
1153
1154 void TemplateTable::dup()
1155 {
1156 transition(vtos, vtos);
1157 __ ldr(r0, Address(esp, 0));
1158 __ push(r0);
1159 // stack: ..., a, a
1160 }
1161
1162 void TemplateTable::dup_x1()
1163 {
1164 transition(vtos, vtos);
1165 // stack: ..., a, b
1166 __ ldr(r0, at_tos()); // load b
1167 __ ldr(r2, at_tos_p1()); // load a
1168 __ str(r0, at_tos_p1()); // store b
1169 __ str(r2, at_tos()); // store a
1170 __ push(r0); // push b
1171 // stack: ..., b, a, b
1172 }
1173
1174 void TemplateTable::dup_x2()
1175 {
1176 transition(vtos, vtos);
1177 // stack: ..., a, b, c
1178 __ ldr(r0, at_tos()); // load c
1179 __ ldr(r2, at_tos_p2()); // load a
1180 __ str(r0, at_tos_p2()); // store c in a
1181 __ push(r0); // push c
1182 // stack: ..., c, b, c, c
1183 __ ldr(r0, at_tos_p2()); // load b
1184 __ str(r2, at_tos_p2()); // store a in b
1185 // stack: ..., c, a, c, c
1186 __ str(r0, at_tos_p1()); // store b in c
1187 // stack: ..., c, a, b, c
1188 }
1189
1190 void TemplateTable::dup2()
1191 {
1192 transition(vtos, vtos);
1193 // stack: ..., a, b
1194 __ ldr(r0, at_tos_p1()); // load a
1195 __ push(r0); // push a
1196 __ ldr(r0, at_tos_p1()); // load b
1197 __ push(r0); // push b
1198 // stack: ..., a, b, a, b
1199 }
1200
1201 void TemplateTable::dup2_x1()
1202 {
1203 transition(vtos, vtos);
1204 // stack: ..., a, b, c
1205 __ ldr(r2, at_tos()); // load c
1206 __ ldr(r0, at_tos_p1()); // load b
1207 __ push(r0); // push b
1208 __ push(r2); // push c
1209 // stack: ..., a, b, c, b, c
1210 __ str(r2, at_tos_p3()); // store c in b
1211 // stack: ..., a, c, c, b, c
1212 __ ldr(r2, at_tos_p4()); // load a
1213 __ str(r2, at_tos_p2()); // store a in 2nd c
1214 // stack: ..., a, c, a, b, c
1215 __ str(r0, at_tos_p4()); // store b in a
1216 // stack: ..., b, c, a, b, c
1217 }
1218
1219 void TemplateTable::dup2_x2()
1220 {
1221 transition(vtos, vtos);
1222 // stack: ..., a, b, c, d
1223 __ ldr(r2, at_tos()); // load d
1224 __ ldr(r0, at_tos_p1()); // load c
1225 __ push(r0) ; // push c
1226 __ push(r2); // push d
1227 // stack: ..., a, b, c, d, c, d
1228 __ ldr(r0, at_tos_p4()); // load b
1229 __ str(r0, at_tos_p2()); // store b in d
1230 __ str(r2, at_tos_p4()); // store d in b
1231 // stack: ..., a, d, c, b, c, d
1232 __ ldr(r2, at_tos_p5()); // load a
1233 __ ldr(r0, at_tos_p3()); // load c
1234 __ str(r2, at_tos_p3()); // store a in c
1235 __ str(r0, at_tos_p5()); // store c in a
1236 // stack: ..., c, d, a, b, c, d
1237 }
1238
1239 void TemplateTable::swap()
1240 {
1241 transition(vtos, vtos);
1242 // stack: ..., a, b
1243 __ ldr(r2, at_tos_p1()); // load a
1244 __ ldr(r0, at_tos()); // load b
1245 __ str(r2, at_tos()); // store a in b
1246 __ str(r0, at_tos_p1()); // store b in a
1247 // stack: ..., b, a
1248 }
1249
1250 void TemplateTable::iop2(Operation op)
1251 {
1252 transition(itos, itos);
1253 // r0 <== r1 op r0
1254 __ pop_i(r1);
1255 switch (op) {
1256 case add : __ addw(r0, r1, r0); break;
1257 case sub : __ subw(r0, r1, r0); break;
1258 case mul : __ mulw(r0, r1, r0); break;
1259 case _and : __ andw(r0, r1, r0); break;
1260 case _or : __ orrw(r0, r1, r0); break;
1261 case _xor : __ eorw(r0, r1, r0); break;
1262 case shl : __ lslvw(r0, r1, r0); break;
1263 case shr : __ asrvw(r0, r1, r0); break;
1264 case ushr : __ lsrvw(r0, r1, r0);break;
1265 default : ShouldNotReachHere();
1266 }
1267 }
1268
1269 void TemplateTable::lop2(Operation op)
1270 {
1271 transition(ltos, ltos);
1272 // r0 <== r1 op r0
1273 __ pop_l(r1);
1274 switch (op) {
1275 case add : __ add(r0, r1, r0); break;
1276 case sub : __ sub(r0, r1, r0); break;
1277 case mul : __ mul(r0, r1, r0); break;
1278 case _and : __ andr(r0, r1, r0); break;
1279 case _or : __ orr(r0, r1, r0); break;
1280 case _xor : __ eor(r0, r1, r0); break;
1281 default : ShouldNotReachHere();
1282 }
1283 }
1284
1285 void TemplateTable::idiv()
1286 {
1287 transition(itos, itos);
1288 // explicitly check for div0
1289 Label no_div0;
1290 __ cbnzw(r0, no_div0);
1291 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1292 __ br(rscratch1);
1293 __ bind(no_div0);
1294 __ pop_i(r1);
1295 // r0 <== r1 idiv r0
1296 __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1297 }
1298
1299 void TemplateTable::irem()
1300 {
1301 transition(itos, itos);
1302 // explicitly check for div0
1303 Label no_div0;
1304 __ cbnzw(r0, no_div0);
1305 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1306 __ br(rscratch1);
1307 __ bind(no_div0);
1308 __ pop_i(r1);
1309 // r0 <== r1 irem r0
1310 __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1311 }
1312
1313 void TemplateTable::lmul()
1314 {
1315 transition(ltos, ltos);
1316 __ pop_l(r1);
1317 __ mul(r0, r0, r1);
1318 }
1319
1320 void TemplateTable::ldiv()
1321 {
1322 transition(ltos, ltos);
1323 // explicitly check for div0
1324 Label no_div0;
1325 __ cbnz(r0, no_div0);
1326 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1327 __ br(rscratch1);
1328 __ bind(no_div0);
1329 __ pop_l(r1);
1330 // r0 <== r1 ldiv r0
1331 __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1332 }
1333
1334 void TemplateTable::lrem()
1335 {
1336 transition(ltos, ltos);
1337 // explicitly check for div0
1338 Label no_div0;
1339 __ cbnz(r0, no_div0);
1340 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1341 __ br(rscratch1);
1342 __ bind(no_div0);
1343 __ pop_l(r1);
1344 // r0 <== r1 lrem r0
1345 __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1346 }
1347
1348 void TemplateTable::lshl()
1349 {
1350 transition(itos, ltos);
1351 // shift count is in r0
1352 __ pop_l(r1);
1353 __ lslv(r0, r1, r0);
1354 }
1355
1356 void TemplateTable::lshr()
1357 {
1358 transition(itos, ltos);
1359 // shift count is in r0
1360 __ pop_l(r1);
1361 __ asrv(r0, r1, r0);
1362 }
1363
1364 void TemplateTable::lushr()
1365 {
1366 transition(itos, ltos);
1367 // shift count is in r0
1368 __ pop_l(r1);
1369 __ lsrv(r0, r1, r0);
1370 }
1371
1372 void TemplateTable::fop2(Operation op)
1373 {
1374 transition(ftos, ftos);
1375 switch (op) {
1376 case add:
1377 // n.b. use ldrd because this is a 64 bit slot
1378 __ pop_f(v1);
1379 __ fadds(v0, v1, v0);
1380 break;
1381 case sub:
1382 __ pop_f(v1);
1383 __ fsubs(v0, v1, v0);
1384 break;
1385 case mul:
1386 __ pop_f(v1);
1387 __ fmuls(v0, v1, v0);
1388 break;
1389 case div:
1390 __ pop_f(v1);
1391 __ fdivs(v0, v1, v0);
1392 break;
1393 case rem:
1394 __ fmovs(v1, v0);
1395 __ pop_f(v0);
1396 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::frem),
1397 0, 2, MacroAssembler::ret_type_float);
1398 break;
1399 default:
1400 ShouldNotReachHere();
1401 break;
1402 }
1403 }
1404
1405 void TemplateTable::dop2(Operation op)
1406 {
1407 transition(dtos, dtos);
1408 switch (op) {
1409 case add:
1410 // n.b. use ldrd because this is a 64 bit slot
1411 __ pop_d(v1);
1412 __ faddd(v0, v1, v0);
1413 break;
1414 case sub:
1415 __ pop_d(v1);
1416 __ fsubd(v0, v1, v0);
1417 break;
1418 case mul:
1419 __ pop_d(v1);
1420 __ fmuld(v0, v1, v0);
1421 break;
1422 case div:
1423 __ pop_d(v1);
1424 __ fdivd(v0, v1, v0);
1425 break;
1426 case rem:
1427 __ fmovd(v1, v0);
1428 __ pop_d(v0);
1429 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::drem),
1430 0, 2, MacroAssembler::ret_type_double);
1431 break;
1432 default:
1433 ShouldNotReachHere();
1434 break;
1435 }
1436 }
1437
1438 void TemplateTable::ineg()
1439 {
1440 transition(itos, itos);
1441 __ negw(r0, r0);
1442
1443 }
1444
1445 void TemplateTable::lneg()
1446 {
1447 transition(ltos, ltos);
1448 __ neg(r0, r0);
1449 }
1450
1451 void TemplateTable::fneg()
1452 {
1453 transition(ftos, ftos);
1454 __ fnegs(v0, v0);
1455 }
1456
1457 void TemplateTable::dneg()
1458 {
1459 transition(dtos, dtos);
1460 __ fnegd(v0, v0);
1461 }
1462
1463 void TemplateTable::iinc()
1464 {
1465 transition(vtos, vtos);
1466 __ load_signed_byte(r1, at_bcp(2)); // get constant
1467 locals_index(r2);
1468 __ ldr(r0, iaddress(r2));
1469 __ addw(r0, r0, r1);
1470 __ str(r0, iaddress(r2));
1471 }
1472
1473 void TemplateTable::wide_iinc()
1474 {
1475 transition(vtos, vtos);
1476 // __ mov(r1, zr);
1477 __ ldrw(r1, at_bcp(2)); // get constant and index
1478 __ rev16(r1, r1);
1479 __ ubfx(r2, r1, 0, 16);
1480 __ neg(r2, r2);
1481 __ sbfx(r1, r1, 16, 16);
1482 __ ldr(r0, iaddress(r2));
1483 __ addw(r0, r0, r1);
1484 __ str(r0, iaddress(r2));
1485 }
1486
1487 void TemplateTable::convert()
1488 {
1489 // Checking
1490 #ifdef ASSERT
1491 {
1492 TosState tos_in = ilgl;
1493 TosState tos_out = ilgl;
1494 switch (bytecode()) {
1495 case Bytecodes::_i2l: // fall through
1496 case Bytecodes::_i2f: // fall through
1497 case Bytecodes::_i2d: // fall through
1498 case Bytecodes::_i2b: // fall through
1499 case Bytecodes::_i2c: // fall through
1500 case Bytecodes::_i2s: tos_in = itos; break;
1501 case Bytecodes::_l2i: // fall through
1502 case Bytecodes::_l2f: // fall through
1503 case Bytecodes::_l2d: tos_in = ltos; break;
1504 case Bytecodes::_f2i: // fall through
1505 case Bytecodes::_f2l: // fall through
1506 case Bytecodes::_f2d: tos_in = ftos; break;
1507 case Bytecodes::_d2i: // fall through
1508 case Bytecodes::_d2l: // fall through
1509 case Bytecodes::_d2f: tos_in = dtos; break;
1510 default : ShouldNotReachHere();
1511 }
1512 switch (bytecode()) {
1513 case Bytecodes::_l2i: // fall through
1514 case Bytecodes::_f2i: // fall through
1515 case Bytecodes::_d2i: // fall through
1516 case Bytecodes::_i2b: // fall through
1517 case Bytecodes::_i2c: // fall through
1518 case Bytecodes::_i2s: tos_out = itos; break;
1519 case Bytecodes::_i2l: // fall through
1520 case Bytecodes::_f2l: // fall through
1521 case Bytecodes::_d2l: tos_out = ltos; break;
1522 case Bytecodes::_i2f: // fall through
1523 case Bytecodes::_l2f: // fall through
1524 case Bytecodes::_d2f: tos_out = ftos; break;
1525 case Bytecodes::_i2d: // fall through
1526 case Bytecodes::_l2d: // fall through
1527 case Bytecodes::_f2d: tos_out = dtos; break;
1528 default : ShouldNotReachHere();
1529 }
1530 transition(tos_in, tos_out);
1531 }
1532 #endif // ASSERT
1533 // static const int64_t is_nan = 0x8000000000000000L;
1534
1535 // Conversion
1536 switch (bytecode()) {
1537 case Bytecodes::_i2l:
1538 __ sxtw(r0, r0);
1539 break;
1540 case Bytecodes::_i2f:
1541 __ scvtfws(v0, r0);
1542 break;
1543 case Bytecodes::_i2d:
1544 __ scvtfwd(v0, r0);
1545 break;
1546 case Bytecodes::_i2b:
1547 __ sxtbw(r0, r0);
1548 break;
1549 case Bytecodes::_i2c:
1550 __ uxthw(r0, r0);
1551 break;
1552 case Bytecodes::_i2s:
1553 __ sxthw(r0, r0);
1554 break;
1555 case Bytecodes::_l2i:
1556 __ uxtw(r0, r0);
1557 break;
1558 case Bytecodes::_l2f:
1559 __ scvtfs(v0, r0);
1560 break;
1561 case Bytecodes::_l2d:
1562 __ scvtfd(v0, r0);
1563 break;
1564 case Bytecodes::_f2i:
1565 {
1566 Label L_Okay;
1567 __ clear_fpsr();
1568 __ fcvtzsw(r0, v0);
1569 __ get_fpsr(r1);
1570 __ cbzw(r1, L_Okay);
1571 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2i),
1572 0, 1, MacroAssembler::ret_type_integral);
1573 __ bind(L_Okay);
1574 }
1575 break;
1576 case Bytecodes::_f2l:
1577 {
1578 Label L_Okay;
1579 __ clear_fpsr();
1580 __ fcvtzs(r0, v0);
1581 __ get_fpsr(r1);
1582 __ cbzw(r1, L_Okay);
1583 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2l),
1584 0, 1, MacroAssembler::ret_type_integral);
1585 __ bind(L_Okay);
1586 }
1587 break;
1588 case Bytecodes::_f2d:
1589 __ fcvts(v0, v0);
1590 break;
1591 case Bytecodes::_d2i:
1592 {
1593 Label L_Okay;
1594 __ clear_fpsr();
1595 __ fcvtzdw(r0, v0);
1596 __ get_fpsr(r1);
1597 __ cbzw(r1, L_Okay);
1598 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2i),
1599 0, 1, MacroAssembler::ret_type_integral);
1600 __ bind(L_Okay);
1601 }
1602 break;
1603 case Bytecodes::_d2l:
1604 {
1605 Label L_Okay;
1606 __ clear_fpsr();
1607 __ fcvtzd(r0, v0);
1608 __ get_fpsr(r1);
1609 __ cbzw(r1, L_Okay);
1610 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2l),
1611 0, 1, MacroAssembler::ret_type_integral);
1612 __ bind(L_Okay);
1613 }
1614 break;
1615 case Bytecodes::_d2f:
1616 __ fcvtd(v0, v0);
1617 break;
1618 default:
1619 ShouldNotReachHere();
1620 }
1621 }
1622
1623 void TemplateTable::lcmp()
1624 {
1625 transition(ltos, itos);
1626 Label done;
1627 __ pop_l(r1);
1628 __ cmp(r1, r0);
1629 __ mov(r0, (u_int64_t)-1L);
1630 __ br(Assembler::LT, done);
1631 // __ mov(r0, 1UL);
1632 // __ csel(r0, r0, zr, Assembler::NE);
1633 // and here is a faster way
1634 __ csinc(r0, zr, zr, Assembler::EQ);
1635 __ bind(done);
1636 }
1637
1638 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1639 {
1640 Label done;
1641 if (is_float) {
1642 // XXX get rid of pop here, use ... reg, mem32
1643 __ pop_f(v1);
1644 __ fcmps(v1, v0);
1645 } else {
1646 // XXX get rid of pop here, use ... reg, mem64
1647 __ pop_d(v1);
1648 __ fcmpd(v1, v0);
1649 }
1650 if (unordered_result < 0) {
1651 // we want -1 for unordered or less than, 0 for equal and 1 for
1652 // greater than.
1653 __ mov(r0, (u_int64_t)-1L);
1654 // for FP LT tests less than or unordered
1655 __ br(Assembler::LT, done);
1656 // install 0 for EQ otherwise 1
1657 __ csinc(r0, zr, zr, Assembler::EQ);
1658 } else {
1659 // we want -1 for less than, 0 for equal and 1 for unordered or
1660 // greater than.
1661 __ mov(r0, 1L);
1662 // for FP HI tests greater than or unordered
1663 __ br(Assembler::HI, done);
1664 // install 0 for EQ otherwise ~0
1665 __ csinv(r0, zr, zr, Assembler::EQ);
1666
1667 }
1668 __ bind(done);
1669 }
1670
1671 void TemplateTable::branch(bool is_jsr, bool is_wide)
1672 {
1673 // We might be moving to a safepoint. The thread which calls
1674 // Interpreter::notice_safepoints() will effectively flush its cache
1675 // when it makes a system call, but we need to do something to
1676 // ensure that we see the changed dispatch table.
1677 __ membar(MacroAssembler::LoadLoad);
1678
1679 __ profile_taken_branch(r0, r1);
1680 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1681 InvocationCounter::counter_offset();
1682 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1683 InvocationCounter::counter_offset();
1684
1685 // load branch displacement
1686 if (!is_wide) {
1687 __ ldrh(r2, at_bcp(1));
1688 __ rev16(r2, r2);
1689 // sign extend the 16 bit value in r2
1690 __ sbfm(r2, r2, 0, 15);
1691 } else {
1692 __ ldrw(r2, at_bcp(1));
1693 __ revw(r2, r2);
1694 // sign extend the 32 bit value in r2
1695 __ sbfm(r2, r2, 0, 31);
1696 }
1697
1698 // Handle all the JSR stuff here, then exit.
1699 // It's much shorter and cleaner than intermingling with the non-JSR
1700 // normal-branch stuff occurring below.
1701
1702 if (is_jsr) {
1703 // Pre-load the next target bytecode into rscratch1
1704 __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1705 // compute return address as bci
1706 __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1707 __ add(rscratch2, rscratch2,
1708 in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1709 __ sub(r1, rbcp, rscratch2);
1710 __ push_i(r1);
1711 // Adjust the bcp by the 16-bit displacement in r2
1712 __ add(rbcp, rbcp, r2);
1713 __ dispatch_only(vtos);
1714 return;
1715 }
1716
1717 // Normal (non-jsr) branch handling
1718
1719 // Adjust the bcp by the displacement in r2
1720 __ add(rbcp, rbcp, r2);
1721
1722 assert(UseLoopCounter || !UseOnStackReplacement,
1723 "on-stack-replacement requires loop counters");
1724 Label backedge_counter_overflow;
1725 Label profile_method;
1726 Label dispatch;
1727 if (UseLoopCounter) {
1728 // increment backedge counter for backward branches
1729 // r0: MDO
1730 // w1: MDO bumped taken-count
1731 // r2: target offset
1732 __ cmp(r2, zr);
1733 __ br(Assembler::GT, dispatch); // count only if backward branch
1734
1735 // ECN: FIXME: This code smells
1736 // check if MethodCounters exists
1737 Label has_counters;
1738 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1739 __ cbnz(rscratch1, has_counters);
1740 __ push(r0);
1741 __ push(r1);
1742 __ push(r2);
1743 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1744 InterpreterRuntime::build_method_counters), rmethod);
1745 __ pop(r2);
1746 __ pop(r1);
1747 __ pop(r0);
1748 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1749 __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1750 __ bind(has_counters);
1751
1752 if (TieredCompilation) {
1753 Label no_mdo;
1754 int increment = InvocationCounter::count_increment;
1755 if (ProfileInterpreter) {
1756 // Are we profiling?
1757 __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1758 __ cbz(r1, no_mdo);
1759 // Increment the MDO backedge counter
1760 const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1761 in_bytes(InvocationCounter::counter_offset()));
1762 const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1763 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1764 r0, rscratch1, false, Assembler::EQ, &backedge_counter_overflow);
1765 __ b(dispatch);
1766 }
1767 __ bind(no_mdo);
1768 // Increment backedge counter in MethodCounters*
1769 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1770 const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1771 __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1772 r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow);
1773 } else { // not TieredCompilation
1774 // increment counter
1775 __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1776 __ ldrw(r0, Address(rscratch2, be_offset)); // load backedge counter
1777 __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1778 __ strw(rscratch1, Address(rscratch2, be_offset)); // store counter
1779
1780 __ ldrw(r0, Address(rscratch2, inv_offset)); // load invocation counter
1781 __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1782 __ addw(r0, r0, rscratch1); // add both counters
1783
1784 if (ProfileInterpreter) {
1785 // Test to see if we should create a method data oop
1786 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
1787 __ cmpw(r0, rscratch1);
1788 __ br(Assembler::LT, dispatch);
1789
1790 // if no method data exists, go to profile method
1791 __ test_method_data_pointer(r0, profile_method);
1792
1793 if (UseOnStackReplacement) {
1794 // check for overflow against w1 which is the MDO taken count
1795 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1796 __ cmpw(r1, rscratch1);
1797 __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1798
1799 // When ProfileInterpreter is on, the backedge_count comes
1800 // from the MethodData*, which value does not get reset on
1801 // the call to frequency_counter_overflow(). To avoid
1802 // excessive calls to the overflow routine while the method is
1803 // being compiled, add a second test to make sure the overflow
1804 // function is called only once every overflow_frequency.
1805 const int overflow_frequency = 1024;
1806 __ andsw(r1, r1, overflow_frequency - 1);
1807 __ br(Assembler::EQ, backedge_counter_overflow);
1808
1809 }
1810 } else {
1811 if (UseOnStackReplacement) {
1812 // check for overflow against w0, which is the sum of the
1813 // counters
1814 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1815 __ cmpw(r0, rscratch1);
1816 __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1817 }
1818 }
1819 }
1820 }
1821 __ bind(dispatch);
1822
1823 // Pre-load the next target bytecode into rscratch1
1824 __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1825
1826 // continue with the bytecode @ target
1827 // rscratch1: target bytecode
1828 // rbcp: target bcp
1829 __ dispatch_only(vtos);
1830
1831 if (UseLoopCounter) {
1832 if (ProfileInterpreter) {
1833 // Out-of-line code to allocate method data oop.
1834 __ bind(profile_method);
1835 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1836 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
1837 __ set_method_data_pointer_for_bcp();
1838 __ b(dispatch);
1839 }
1840
1841 if (TieredCompilation || UseOnStackReplacement) {
1842 // invocation counter overflow
1843 __ bind(backedge_counter_overflow);
1844 __ neg(r2, r2);
1845 __ add(r2, r2, rbcp); // branch bcp
1846 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1847 __ call_VM(noreg,
1848 CAST_FROM_FN_PTR(address,
1849 InterpreterRuntime::frequency_counter_overflow),
1850 r2);
1851 if (!UseOnStackReplacement)
1852 __ b(dispatch);
1853 }
1854
1855 if (UseOnStackReplacement) {
1856 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
1857
1858 // r0: osr nmethod (osr ok) or NULL (osr not possible)
1859 // w1: target bytecode
1860 // r2: scratch
1861 __ cbz(r0, dispatch); // test result -- no osr if null
1862 // nmethod may have been invalidated (VM may block upon call_VM return)
1863 __ ldrb(r2, Address(r0, nmethod::state_offset()));
1864 if (nmethod::in_use != 0)
1865 __ sub(r2, r2, nmethod::in_use);
1866 __ cbnz(r2, dispatch);
1867
1868 // We have the address of an on stack replacement routine in r0
1869 // We need to prepare to execute the OSR method. First we must
1870 // migrate the locals and monitors off of the stack.
1871
1872 __ mov(r19, r0); // save the nmethod
1873
1874 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1875
1876 // r0 is OSR buffer, move it to expected parameter location
1877 __ mov(j_rarg0, r0);
1878
1879 // remove activation
1880 // get sender esp
1881 __ ldr(esp,
1882 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1883 // remove frame anchor
1884 __ leave();
1885 // Ensure compiled code always sees stack at proper alignment
1886 __ andr(sp, esp, -16);
1887
1888 // and begin the OSR nmethod
1889 __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1890 __ br(rscratch1);
1891 }
1892 }
1893 }
1894
1895
1896 void TemplateTable::if_0cmp(Condition cc)
1897 {
1898 transition(itos, vtos);
1899 // assume branch is more often taken than not (loops use backward branches)
1900 Label not_taken;
1901 if (cc == equal)
1902 __ cbnzw(r0, not_taken);
1903 else if (cc == not_equal)
1904 __ cbzw(r0, not_taken);
1905 else {
1906 __ andsw(zr, r0, r0);
1907 __ br(j_not(cc), not_taken);
1908 }
1909
1910 branch(false, false);
1911 __ bind(not_taken);
1912 __ profile_not_taken_branch(r0);
1913 }
1914
1915 void TemplateTable::if_icmp(Condition cc)
1916 {
1917 transition(itos, vtos);
1918 // assume branch is more often taken than not (loops use backward branches)
1919 Label not_taken;
1920 __ pop_i(r1);
1921 __ cmpw(r1, r0, Assembler::LSL);
1922 __ br(j_not(cc), not_taken);
1923 branch(false, false);
1924 __ bind(not_taken);
1925 __ profile_not_taken_branch(r0);
1926 }
1927
1928 void TemplateTable::if_nullcmp(Condition cc)
1929 {
1930 transition(atos, vtos);
1931 // assume branch is more often taken than not (loops use backward branches)
1932 Label not_taken;
1933 if (cc == equal)
1934 __ cbnz(r0, not_taken);
1935 else
1936 __ cbz(r0, not_taken);
1937 branch(false, false);
1938 __ bind(not_taken);
1939 __ profile_not_taken_branch(r0);
1940 }
1941
1942 void TemplateTable::if_acmp(Condition cc)
1943 {
1944 transition(atos, vtos);
1945 // assume branch is more often taken than not (loops use backward branches)
1946 Label not_taken;
1947 __ pop_ptr(r1);
1948 __ cmp(r1, r0);
1949 __ br(j_not(cc), not_taken);
1950 branch(false, false);
1951 __ bind(not_taken);
1952 __ profile_not_taken_branch(r0);
1953 }
1954
1955 void TemplateTable::ret() {
1956 transition(vtos, vtos);
1957 // We might be moving to a safepoint. The thread which calls
1958 // Interpreter::notice_safepoints() will effectively flush its cache
1959 // when it makes a system call, but we need to do something to
1960 // ensure that we see the changed dispatch table.
1961 __ membar(MacroAssembler::LoadLoad);
1962
1963 locals_index(r1);
1964 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1965 __ profile_ret(r1, r2);
1966 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1967 __ lea(rbcp, Address(rbcp, r1));
1968 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1969 __ dispatch_next(vtos);
1970 }
1971
1972 void TemplateTable::wide_ret() {
1973 transition(vtos, vtos);
1974 locals_index_wide(r1);
1975 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1976 __ profile_ret(r1, r2);
1977 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1978 __ lea(rbcp, Address(rbcp, r1));
1979 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1980 __ dispatch_next(vtos);
1981 }
1982
1983
1984 void TemplateTable::tableswitch() {
1985 Label default_case, continue_execution;
1986 transition(itos, vtos);
1987 // align rbcp
1988 __ lea(r1, at_bcp(BytesPerInt));
1989 __ andr(r1, r1, -BytesPerInt);
1990 // load lo & hi
1991 __ ldrw(r2, Address(r1, BytesPerInt));
1992 __ ldrw(r3, Address(r1, 2 * BytesPerInt));
1993 __ rev32(r2, r2);
1994 __ rev32(r3, r3);
1995 // check against lo & hi
1996 __ cmpw(r0, r2);
1997 __ br(Assembler::LT, default_case);
1998 __ cmpw(r0, r3);
1999 __ br(Assembler::GT, default_case);
2000 // lookup dispatch offset
2001 __ subw(r0, r0, r2);
2002 __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2003 __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2004 __ profile_switch_case(r0, r1, r2);
2005 // continue execution
2006 __ bind(continue_execution);
2007 __ rev32(r3, r3);
2008 __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2009 __ add(rbcp, rbcp, r3, ext::sxtw);
2010 __ dispatch_only(vtos);
2011 // handle default
2012 __ bind(default_case);
2013 __ profile_switch_default(r0);
2014 __ ldrw(r3, Address(r1, 0));
2015 __ b(continue_execution);
2016 }
2017
2018 void TemplateTable::lookupswitch() {
2019 transition(itos, itos);
2020 __ stop("lookupswitch bytecode should have been rewritten");
2021 }
2022
2023 void TemplateTable::fast_linearswitch() {
2024 transition(itos, vtos);
2025 Label loop_entry, loop, found, continue_execution;
2026 // bswap r0 so we can avoid bswapping the table entries
2027 __ rev32(r0, r0);
2028 // align rbcp
2029 __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2030 // this instruction (change offsets
2031 // below)
2032 __ andr(r19, r19, -BytesPerInt);
2033 // set counter
2034 __ ldrw(r1, Address(r19, BytesPerInt));
2035 __ rev32(r1, r1);
2036 __ b(loop_entry);
2037 // table search
2038 __ bind(loop);
2039 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2040 __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2041 __ cmpw(r0, rscratch1);
2042 __ br(Assembler::EQ, found);
2043 __ bind(loop_entry);
2044 __ subs(r1, r1, 1);
2045 __ br(Assembler::PL, loop);
2046 // default case
2047 __ profile_switch_default(r0);
2048 __ ldrw(r3, Address(r19, 0));
2049 __ b(continue_execution);
2050 // entry found -> get offset
2051 __ bind(found);
2052 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2053 __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2054 __ profile_switch_case(r1, r0, r19);
2055 // continue execution
2056 __ bind(continue_execution);
2057 __ rev32(r3, r3);
2058 __ add(rbcp, rbcp, r3, ext::sxtw);
2059 __ ldrb(rscratch1, Address(rbcp, 0));
2060 __ dispatch_only(vtos);
2061 }
2062
2063 void TemplateTable::fast_binaryswitch() {
2064 transition(itos, vtos);
2065 // Implementation using the following core algorithm:
2066 //
2067 // int binary_search(int key, LookupswitchPair* array, int n) {
2068 // // Binary search according to "Methodik des Programmierens" by
2069 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2070 // int i = 0;
2071 // int j = n;
2072 // while (i+1 < j) {
2073 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2074 // // with Q: for all i: 0 <= i < n: key < a[i]
2075 // // where a stands for the array and assuming that the (inexisting)
2076 // // element a[n] is infinitely big.
2077 // int h = (i + j) >> 1;
2078 // // i < h < j
2079 // if (key < array[h].fast_match()) {
2080 // j = h;
2081 // } else {
2082 // i = h;
2083 // }
2084 // }
2085 // // R: a[i] <= key < a[i+1] or Q
2086 // // (i.e., if key is within array, i is the correct index)
2087 // return i;
2088 // }
2089
2090 // Register allocation
2091 const Register key = r0; // already set (tosca)
2092 const Register array = r1;
2093 const Register i = r2;
2094 const Register j = r3;
2095 const Register h = rscratch1;
2096 const Register temp = rscratch2;
2097
2098 // Find array start
2099 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2100 // get rid of this
2101 // instruction (change
2102 // offsets below)
2103 __ andr(array, array, -BytesPerInt);
2104
2105 // Initialize i & j
2106 __ mov(i, 0); // i = 0;
2107 __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2108
2109 // Convert j into native byteordering
2110 __ rev32(j, j);
2111
2112 // And start
2113 Label entry;
2114 __ b(entry);
2115
2116 // binary search loop
2117 {
2118 Label loop;
2119 __ bind(loop);
2120 // int h = (i + j) >> 1;
2121 __ addw(h, i, j); // h = i + j;
2122 __ lsrw(h, h, 1); // h = (i + j) >> 1;
2123 // if (key < array[h].fast_match()) {
2124 // j = h;
2125 // } else {
2126 // i = h;
2127 // }
2128 // Convert array[h].match to native byte-ordering before compare
2129 __ ldr(temp, Address(array, h, Address::lsl(3)));
2130 __ rev32(temp, temp);
2131 __ cmpw(key, temp);
2132 // j = h if (key < array[h].fast_match())
2133 __ csel(j, h, j, Assembler::LT);
2134 // i = h if (key >= array[h].fast_match())
2135 __ csel(i, h, i, Assembler::GE);
2136 // while (i+1 < j)
2137 __ bind(entry);
2138 __ addw(h, i, 1); // i+1
2139 __ cmpw(h, j); // i+1 < j
2140 __ br(Assembler::LT, loop);
2141 }
2142
2143 // end of binary search, result index is i (must check again!)
2144 Label default_case;
2145 // Convert array[i].match to native byte-ordering before compare
2146 __ ldr(temp, Address(array, i, Address::lsl(3)));
2147 __ rev32(temp, temp);
2148 __ cmpw(key, temp);
2149 __ br(Assembler::NE, default_case);
2150
2151 // entry found -> j = offset
2152 __ add(j, array, i, ext::uxtx, 3);
2153 __ ldrw(j, Address(j, BytesPerInt));
2154 __ profile_switch_case(i, key, array);
2155 __ rev32(j, j);
2156 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2157 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2158 __ dispatch_only(vtos);
2159
2160 // default case -> j = default offset
2161 __ bind(default_case);
2162 __ profile_switch_default(i);
2163 __ ldrw(j, Address(array, -2 * BytesPerInt));
2164 __ rev32(j, j);
2165 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2166 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2167 __ dispatch_only(vtos);
2168 }
2169
2170
2171 void TemplateTable::_return(TosState state)
2172 {
2173 transition(state, state);
2174 assert(_desc->calls_vm(),
2175 "inconsistent calls_vm information"); // call in remove_activation
2176
2177 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2178 assert(state == vtos, "only valid state");
2179
2180 __ ldr(c_rarg1, aaddress(0));
2181 __ load_klass(r3, c_rarg1);
2182 __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2183 __ tst(r3, JVM_ACC_HAS_FINALIZER);
2184 Label skip_register_finalizer;
2185 __ br(Assembler::EQ, skip_register_finalizer);
2186
2187 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2188
2189 __ bind(skip_register_finalizer);
2190 }
2191
2192 // Issue a StoreStore barrier after all stores but before return
2193 // from any constructor for any class with a final field. We don't
2194 // know if this is a finalizer, so we always do so.
2195 if (_desc->bytecode() == Bytecodes::_return)
2196 __ membar(MacroAssembler::StoreStore);
2197
2198 __ remove_activation(state);
2199 __ ret(lr);
2200 }
2201
2202 // ----------------------------------------------------------------------------
2203 // Volatile variables demand their effects be made known to all CPU's
2204 // in order. Store buffers on most chips allow reads & writes to
2205 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2206 // without some kind of memory barrier (i.e., it's not sufficient that
2207 // the interpreter does not reorder volatile references, the hardware
2208 // also must not reorder them).
2209 //
2210 // According to the new Java Memory Model (JMM):
2211 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2212 // writes act as aquire & release, so:
2213 // (2) A read cannot let unrelated NON-volatile memory refs that
2214 // happen after the read float up to before the read. It's OK for
2215 // non-volatile memory refs that happen before the volatile read to
2216 // float down below it.
2217 // (3) Similar a volatile write cannot let unrelated NON-volatile
2218 // memory refs that happen BEFORE the write float down to after the
2219 // write. It's OK for non-volatile memory refs that happen after the
2220 // volatile write to float up before it.
2221 //
2222 // We only put in barriers around volatile refs (they are expensive),
2223 // not _between_ memory refs (that would require us to track the
2224 // flavor of the previous memory refs). Requirements (2) and (3)
2225 // require some barriers before volatile stores and after volatile
2226 // loads. These nearly cover requirement (1) but miss the
2227 // volatile-store-volatile-load case. This final case is placed after
2228 // volatile-stores although it could just as well go before
2229 // volatile-loads.
2230
2231 void TemplateTable::resolve_cache_and_index(int byte_no,
2232 Register Rcache,
2233 Register index,
2234 size_t index_size) {
2235 const Register temp = r19;
2236 assert_different_registers(Rcache, index, temp);
2237
2238 Label resolved;
2239
2240 Bytecodes::Code code = bytecode();
2241 switch (code) {
2242 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2243 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2244 }
2245
2246 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2247 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2248 __ cmp(temp, (int) code); // have we resolved this bytecode?
2249 __ br(Assembler::EQ, resolved);
2250
2251 // resolve first time through
2252 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2253 __ mov(temp, (int) code);
2254 __ call_VM(noreg, entry, temp);
2255
2256 // Update registers with resolved info
2257 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2258 // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2259 // so all clients ofthis method must be modified accordingly
2260 __ bind(resolved);
2261 }
2262
2263 // The Rcache and index registers must be set before call
2264 // n.b unlike x86 cache already includes the index offset
2265 void TemplateTable::load_field_cp_cache_entry(Register obj,
2266 Register cache,
2267 Register index,
2268 Register off,
2269 Register flags,
2270 bool is_static = false) {
2271 assert_different_registers(cache, index, flags, off);
2272
2273 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2274 // Field offset
2275 __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2276 ConstantPoolCacheEntry::f2_offset())));
2277 // Flags
2278 __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2279 ConstantPoolCacheEntry::flags_offset())));
2280
2281 // klass overwrite register
2282 if (is_static) {
2283 __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2284 ConstantPoolCacheEntry::f1_offset())));
2285 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2286 __ ldr(obj, Address(obj, mirror_offset));
2287 }
2288 }
2289
2290 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2291 Register method,
2292 Register itable_index,
2293 Register flags,
2294 bool is_invokevirtual,
2295 bool is_invokevfinal, /*unused*/
2296 bool is_invokedynamic) {
2297 // setup registers
2298 const Register cache = rscratch2;
2299 const Register index = r4;
2300 assert_different_registers(method, flags);
2301 assert_different_registers(method, cache, index);
2302 assert_different_registers(itable_index, flags);
2303 assert_different_registers(itable_index, cache, index);
2304 // determine constant pool cache field offsets
2305 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2306 const int method_offset = in_bytes(
2307 ConstantPoolCache::base_offset() +
2308 (is_invokevirtual
2309 ? ConstantPoolCacheEntry::f2_offset()
2310 : ConstantPoolCacheEntry::f1_offset()));
2311 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2312 ConstantPoolCacheEntry::flags_offset());
2313 // access constant pool cache fields
2314 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2315 ConstantPoolCacheEntry::f2_offset());
2316
2317 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2318 resolve_cache_and_index(byte_no, cache, index, index_size);
2319 __ ldr(method, Address(cache, method_offset));
2320
2321 if (itable_index != noreg) {
2322 __ ldr(itable_index, Address(cache, index_offset));
2323 }
2324 __ ldrw(flags, Address(cache, flags_offset));
2325 }
2326
2327
2328 // The registers cache and index expected to be set before call.
2329 // Correct values of the cache and index registers are preserved.
2330 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2331 bool is_static, bool has_tos) {
2332 // do the JVMTI work here to avoid disturbing the register state below
2333 // We use c_rarg registers here because we want to use the register used in
2334 // the call to the VM
2335 if (JvmtiExport::can_post_field_access()) {
2336 // Check to see if a field access watch has been set before we
2337 // take the time to call into the VM.
2338 Label L1;
2339 assert_different_registers(cache, index, r0);
2340 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2341 __ ldrw(r0, Address(rscratch1));
2342 __ cbzw(r0, L1);
2343
2344 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2345 __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2346
2347 if (is_static) {
2348 __ mov(c_rarg1, zr); // NULL object reference
2349 } else {
2350 __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2351 __ verify_oop(c_rarg1);
2352 }
2353 // c_rarg1: object pointer or NULL
2354 // c_rarg2: cache entry pointer
2355 // c_rarg3: jvalue object on the stack
2356 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2357 InterpreterRuntime::post_field_access),
2358 c_rarg1, c_rarg2, c_rarg3);
2359 __ get_cache_and_index_at_bcp(cache, index, 1);
2360 __ bind(L1);
2361 }
2362 }
2363
2364 void TemplateTable::pop_and_check_object(Register r)
2365 {
2366 __ pop_ptr(r);
2367 __ null_check(r); // for field access must check obj.
2368 __ verify_oop(r);
2369 }
2370
2371 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2372 {
2373 const Register cache = r2;
2374 const Register index = r3;
2375 const Register obj = r4;
2376 const Register off = r19;
2377 const Register flags = r0;
2378 const Register bc = r4; // uses same reg as obj, so don't mix them
2379
2380 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2381 jvmti_post_field_access(cache, index, is_static, false);
2382 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2383
2384 if (!is_static) {
2385 // obj is on the stack
2386 pop_and_check_object(obj);
2387 }
2388
2389 const Address field(obj, off);
2390
2391 Label Done, notByte, notInt, notShort, notChar,
2392 notLong, notFloat, notObj, notDouble;
2393
2394 // x86 uses a shift and mask or wings it with a shift plus assert
2395 // the mask is not needed. aarch64 just uses bitfield extract
2396 __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
2397
2398 assert(btos == 0, "change code, btos != 0");
2399 __ cbnz(flags, notByte);
2400
2401 // Don't rewrite getstatic, only getfield
2402 if (is_static) rc = may_not_rewrite;
2403
2404 // btos
2405 __ load_signed_byte(r0, field);
2406 __ push(btos);
2407 // Rewrite bytecode to be faster
2408 if (rc == may_rewrite) {
2409 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2410 }
2411 __ b(Done);
2412
2413 __ bind(notByte);
2414 __ cmp(flags, atos);
2415 __ br(Assembler::NE, notObj);
2416 // atos
2417 __ load_heap_oop(r0, field);
2418 __ push(atos);
2419 if (rc == may_rewrite) {
2420 patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2421 }
2422 __ b(Done);
2423
2424 __ bind(notObj);
2425 __ cmp(flags, itos);
2426 __ br(Assembler::NE, notInt);
2427 // itos
2428 __ ldrw(r0, field);
2429 __ push(itos);
2430 // Rewrite bytecode to be faster
2431 if (rc == may_rewrite) {
2432 patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2433 }
2434 __ b(Done);
2435
2436 __ bind(notInt);
2437 __ cmp(flags, ctos);
2438 __ br(Assembler::NE, notChar);
2439 // ctos
2440 __ load_unsigned_short(r0, field);
2441 __ push(ctos);
2442 // Rewrite bytecode to be faster
2443 if (rc == may_rewrite) {
2444 patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2445 }
2446 __ b(Done);
2447
2448 __ bind(notChar);
2449 __ cmp(flags, stos);
2450 __ br(Assembler::NE, notShort);
2451 // stos
2452 __ load_signed_short(r0, field);
2453 __ push(stos);
2454 // Rewrite bytecode to be faster
2455 if (rc == may_rewrite) {
2456 patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2457 }
2458 __ b(Done);
2459
2460 __ bind(notShort);
2461 __ cmp(flags, ltos);
2462 __ br(Assembler::NE, notLong);
2463 // ltos
2464 __ ldr(r0, field);
2465 __ push(ltos);
2466 // Rewrite bytecode to be faster
2467 if (rc == may_rewrite) {
2468 patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2469 }
2470 __ b(Done);
2471
2472 __ bind(notLong);
2473 __ cmp(flags, ftos);
2474 __ br(Assembler::NE, notFloat);
2475 // ftos
2476 __ ldrs(v0, field);
2477 __ push(ftos);
2478 // Rewrite bytecode to be faster
2479 if (rc == may_rewrite) {
2480 patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2481 }
2482 __ b(Done);
2483
2484 __ bind(notFloat);
2485 #ifdef ASSERT
2486 __ cmp(flags, dtos);
2487 __ br(Assembler::NE, notDouble);
2488 #endif
2489 // dtos
2490 __ ldrd(v0, field);
2491 __ push(dtos);
2492 // Rewrite bytecode to be faster
2493 if (rc == may_rewrite) {
2494 patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2495 }
2496 #ifdef ASSERT
2497 __ b(Done);
2498
2499 __ bind(notDouble);
2500 __ stop("Bad state");
2501 #endif
2502
2503 __ bind(Done);
2504 // It's really not worth bothering to check whether this field
2505 // really is volatile in the slow case.
2506 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2507 }
2508
2509
2510 void TemplateTable::getfield(int byte_no)
2511 {
2512 getfield_or_static(byte_no, false);
2513 }
2514
2515 void TemplateTable::nofast_getfield(int byte_no) {
2516 getfield_or_static(byte_no, false, may_not_rewrite);
2517 }
2518
2519 void TemplateTable::getstatic(int byte_no)
2520 {
2521 getfield_or_static(byte_no, true);
2522 }
2523
2524 // The registers cache and index expected to be set before call.
2525 // The function may destroy various registers, just not the cache and index registers.
2526 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2527 transition(vtos, vtos);
2528
2529 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2530
2531 if (JvmtiExport::can_post_field_modification()) {
2532 // Check to see if a field modification watch has been set before
2533 // we take the time to call into the VM.
2534 Label L1;
2535 assert_different_registers(cache, index, r0);
2536 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2537 __ ldrw(r0, Address(rscratch1));
2538 __ cbz(r0, L1);
2539
2540 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2541
2542 if (is_static) {
2543 // Life is simple. Null out the object pointer.
2544 __ mov(c_rarg1, zr);
2545 } else {
2546 // Life is harder. The stack holds the value on top, followed by
2547 // the object. We don't know the size of the value, though; it
2548 // could be one or two words depending on its type. As a result,
2549 // we must find the type to determine where the object is.
2550 __ ldrw(c_rarg3, Address(c_rarg2,
2551 in_bytes(cp_base_offset +
2552 ConstantPoolCacheEntry::flags_offset())));
2553 __ lsr(c_rarg3, c_rarg3,
2554 ConstantPoolCacheEntry::tos_state_shift);
2555 ConstantPoolCacheEntry::verify_tos_state_shift();
2556 Label nope2, done, ok;
2557 __ ldr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2558 __ cmpw(c_rarg3, ltos);
2559 __ br(Assembler::EQ, ok);
2560 __ cmpw(c_rarg3, dtos);
2561 __ br(Assembler::NE, nope2);
2562 __ bind(ok);
2563 __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2564 __ bind(nope2);
2565 }
2566 // cache entry pointer
2567 __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2568 // object (tos)
2569 __ mov(c_rarg3, esp);
2570 // c_rarg1: object pointer set up above (NULL if static)
2571 // c_rarg2: cache entry pointer
2572 // c_rarg3: jvalue object on the stack
2573 __ call_VM(noreg,
2574 CAST_FROM_FN_PTR(address,
2575 InterpreterRuntime::post_field_modification),
2576 c_rarg1, c_rarg2, c_rarg3);
2577 __ get_cache_and_index_at_bcp(cache, index, 1);
2578 __ bind(L1);
2579 }
2580 }
2581
2582 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2583 transition(vtos, vtos);
2584
2585 const Register cache = r2;
2586 const Register index = r3;
2587 const Register obj = r2;
2588 const Register off = r19;
2589 const Register flags = r0;
2590 const Register bc = r4;
2591
2592 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2593 jvmti_post_field_mod(cache, index, is_static);
2594 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2595
2596 Label Done;
2597 __ mov(r5, flags);
2598
2599 {
2600 Label notVolatile;
2601 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2602 __ membar(MacroAssembler::StoreStore);
2603 __ bind(notVolatile);
2604 }
2605
2606 // field address
2607 const Address field(obj, off);
2608
2609 Label notByte, notInt, notShort, notChar,
2610 notLong, notFloat, notObj, notDouble;
2611
2612 // x86 uses a shift and mask or wings it with a shift plus assert
2613 // the mask is not needed. aarch64 just uses bitfield extract
2614 __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
2615
2616 assert(btos == 0, "change code, btos != 0");
2617 __ cbnz(flags, notByte);
2618
2619 // Don't rewrite putstatic, only putfield
2620 if (is_static) rc = may_not_rewrite;
2621
2622 // btos
2623 {
2624 __ pop(btos);
2625 if (!is_static) pop_and_check_object(obj);
2626 __ strb(r0, field);
2627 if (rc == may_rewrite) {
2628 patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2629 }
2630 __ b(Done);
2631 }
2632
2633 __ bind(notByte);
2634 __ cmp(flags, atos);
2635 __ br(Assembler::NE, notObj);
2636
2637 // atos
2638 {
2639 __ pop(atos);
2640 if (!is_static) pop_and_check_object(obj);
2641 // Store into the field
2642 do_oop_store(_masm, field, r0, _bs->kind(), false);
2643 if (rc == may_rewrite) {
2644 patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2645 }
2646 __ b(Done);
2647 }
2648
2649 __ bind(notObj);
2650 __ cmp(flags, itos);
2651 __ br(Assembler::NE, notInt);
2652
2653 // itos
2654 {
2655 __ pop(itos);
2656 if (!is_static) pop_and_check_object(obj);
2657 __ strw(r0, field);
2658 if (rc == may_rewrite) {
2659 patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2660 }
2661 __ b(Done);
2662 }
2663
2664 __ bind(notInt);
2665 __ cmp(flags, ctos);
2666 __ br(Assembler::NE, notChar);
2667
2668 // ctos
2669 {
2670 __ pop(ctos);
2671 if (!is_static) pop_and_check_object(obj);
2672 __ strh(r0, field);
2673 if (rc == may_rewrite) {
2674 patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2675 }
2676 __ b(Done);
2677 }
2678
2679 __ bind(notChar);
2680 __ cmp(flags, stos);
2681 __ br(Assembler::NE, notShort);
2682
2683 // stos
2684 {
2685 __ pop(stos);
2686 if (!is_static) pop_and_check_object(obj);
2687 __ strh(r0, field);
2688 if (rc == may_rewrite) {
2689 patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2690 }
2691 __ b(Done);
2692 }
2693
2694 __ bind(notShort);
2695 __ cmp(flags, ltos);
2696 __ br(Assembler::NE, notLong);
2697
2698 // ltos
2699 {
2700 __ pop(ltos);
2701 if (!is_static) pop_and_check_object(obj);
2702 __ str(r0, field);
2703 if (rc == may_rewrite) {
2704 patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2705 }
2706 __ b(Done);
2707 }
2708
2709 __ bind(notLong);
2710 __ cmp(flags, ftos);
2711 __ br(Assembler::NE, notFloat);
2712
2713 // ftos
2714 {
2715 __ pop(ftos);
2716 if (!is_static) pop_and_check_object(obj);
2717 __ strs(v0, field);
2718 if (rc == may_rewrite) {
2719 patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2720 }
2721 __ b(Done);
2722 }
2723
2724 __ bind(notFloat);
2725 #ifdef ASSERT
2726 __ cmp(flags, dtos);
2727 __ br(Assembler::NE, notDouble);
2728 #endif
2729
2730 // dtos
2731 {
2732 __ pop(dtos);
2733 if (!is_static) pop_and_check_object(obj);
2734 __ strd(v0, field);
2735 if (rc == may_rewrite) {
2736 patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2737 }
2738 }
2739
2740 #ifdef ASSERT
2741 __ b(Done);
2742
2743 __ bind(notDouble);
2744 __ stop("Bad state");
2745 #endif
2746
2747 __ bind(Done);
2748
2749 {
2750 Label notVolatile;
2751 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2752 __ membar(MacroAssembler::StoreLoad);
2753 __ bind(notVolatile);
2754 }
2755 }
2756
2757 void TemplateTable::putfield(int byte_no)
2758 {
2759 putfield_or_static(byte_no, false);
2760 }
2761
2762 void TemplateTable::nofast_putfield(int byte_no) {
2763 putfield_or_static(byte_no, false, may_not_rewrite);
2764 }
2765
2766 void TemplateTable::putstatic(int byte_no) {
2767 putfield_or_static(byte_no, true);
2768 }
2769
2770 void TemplateTable::jvmti_post_fast_field_mod()
2771 {
2772 if (JvmtiExport::can_post_field_modification()) {
2773 // Check to see if a field modification watch has been set before
2774 // we take the time to call into the VM.
2775 Label L2;
2776 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2777 __ ldrw(c_rarg3, Address(rscratch1));
2778 __ cbzw(c_rarg3, L2);
2779 __ pop_ptr(r19); // copy the object pointer from tos
2780 __ verify_oop(r19);
2781 __ push_ptr(r19); // put the object pointer back on tos
2782 // Save tos values before call_VM() clobbers them. Since we have
2783 // to do it for every data type, we use the saved values as the
2784 // jvalue object.
2785 switch (bytecode()) { // load values into the jvalue object
2786 case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2787 case Bytecodes::_fast_bputfield: // fall through
2788 case Bytecodes::_fast_sputfield: // fall through
2789 case Bytecodes::_fast_cputfield: // fall through
2790 case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2791 case Bytecodes::_fast_dputfield: __ push_d(); break;
2792 case Bytecodes::_fast_fputfield: __ push_f(); break;
2793 case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2794
2795 default:
2796 ShouldNotReachHere();
2797 }
2798 __ mov(c_rarg3, esp); // points to jvalue on the stack
2799 // access constant pool cache entry
2800 __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2801 __ verify_oop(r19);
2802 // r19: object pointer copied above
2803 // c_rarg2: cache entry pointer
2804 // c_rarg3: jvalue object on the stack
2805 __ call_VM(noreg,
2806 CAST_FROM_FN_PTR(address,
2807 InterpreterRuntime::post_field_modification),
2808 r19, c_rarg2, c_rarg3);
2809
2810 switch (bytecode()) { // restore tos values
2811 case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2812 case Bytecodes::_fast_bputfield: // fall through
2813 case Bytecodes::_fast_sputfield: // fall through
2814 case Bytecodes::_fast_cputfield: // fall through
2815 case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2816 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2817 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2818 case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2819 }
2820 __ bind(L2);
2821 }
2822 }
2823
2824 void TemplateTable::fast_storefield(TosState state)
2825 {
2826 transition(state, vtos);
2827
2828 ByteSize base = ConstantPoolCache::base_offset();
2829
2830 jvmti_post_fast_field_mod();
2831
2832 // access constant pool cache
2833 __ get_cache_and_index_at_bcp(r2, r1, 1);
2834
2835 // test for volatile with r3
2836 __ ldrw(r3, Address(r2, in_bytes(base +
2837 ConstantPoolCacheEntry::flags_offset())));
2838
2839 // replace index with field offset from cache entry
2840 __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2841
2842 {
2843 Label notVolatile;
2844 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2845 __ membar(MacroAssembler::StoreStore);
2846 __ bind(notVolatile);
2847 }
2848
2849 Label notVolatile;
2850
2851 // Get object from stack
2852 pop_and_check_object(r2);
2853
2854 // field address
2855 const Address field(r2, r1);
2856
2857 // access field
2858 switch (bytecode()) {
2859 case Bytecodes::_fast_aputfield:
2860 do_oop_store(_masm, field, r0, _bs->kind(), false);
2861 break;
2862 case Bytecodes::_fast_lputfield:
2863 __ str(r0, field);
2864 break;
2865 case Bytecodes::_fast_iputfield:
2866 __ strw(r0, field);
2867 break;
2868 case Bytecodes::_fast_bputfield:
2869 __ strb(r0, field);
2870 break;
2871 case Bytecodes::_fast_sputfield:
2872 // fall through
2873 case Bytecodes::_fast_cputfield:
2874 __ strh(r0, field);
2875 break;
2876 case Bytecodes::_fast_fputfield:
2877 __ strs(v0, field);
2878 break;
2879 case Bytecodes::_fast_dputfield:
2880 __ strd(v0, field);
2881 break;
2882 default:
2883 ShouldNotReachHere();
2884 }
2885
2886 {
2887 Label notVolatile;
2888 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2889 __ membar(MacroAssembler::StoreLoad);
2890 __ bind(notVolatile);
2891 }
2892 }
2893
2894
2895 void TemplateTable::fast_accessfield(TosState state)
2896 {
2897 transition(atos, state);
2898 // Do the JVMTI work here to avoid disturbing the register state below
2899 if (JvmtiExport::can_post_field_access()) {
2900 // Check to see if a field access watch has been set before we
2901 // take the time to call into the VM.
2902 Label L1;
2903 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2904 __ ldrw(r2, Address(rscratch1));
2905 __ cbzw(r2, L1);
2906 // access constant pool cache entry
2907 __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
2908 __ verify_oop(r0);
2909 __ push_ptr(r0); // save object pointer before call_VM() clobbers it
2910 __ mov(c_rarg1, r0);
2911 // c_rarg1: object pointer copied above
2912 // c_rarg2: cache entry pointer
2913 __ call_VM(noreg,
2914 CAST_FROM_FN_PTR(address,
2915 InterpreterRuntime::post_field_access),
2916 c_rarg1, c_rarg2);
2917 __ pop_ptr(r0); // restore object pointer
2918 __ bind(L1);
2919 }
2920
2921 // access constant pool cache
2922 __ get_cache_and_index_at_bcp(r2, r1, 1);
2923 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2924 ConstantPoolCacheEntry::f2_offset())));
2925 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2926 ConstantPoolCacheEntry::flags_offset())));
2927
2928 // r0: object
2929 __ verify_oop(r0);
2930 __ null_check(r0);
2931 const Address field(r0, r1);
2932
2933 // access field
2934 switch (bytecode()) {
2935 case Bytecodes::_fast_agetfield:
2936 __ load_heap_oop(r0, field);
2937 __ verify_oop(r0);
2938 break;
2939 case Bytecodes::_fast_lgetfield:
2940 __ ldr(r0, field);
2941 break;
2942 case Bytecodes::_fast_igetfield:
2943 __ ldrw(r0, field);
2944 break;
2945 case Bytecodes::_fast_bgetfield:
2946 __ load_signed_byte(r0, field);
2947 break;
2948 case Bytecodes::_fast_sgetfield:
2949 __ load_signed_short(r0, field);
2950 break;
2951 case Bytecodes::_fast_cgetfield:
2952 __ load_unsigned_short(r0, field);
2953 break;
2954 case Bytecodes::_fast_fgetfield:
2955 __ ldrs(v0, field);
2956 break;
2957 case Bytecodes::_fast_dgetfield:
2958 __ ldrd(v0, field);
2959 break;
2960 default:
2961 ShouldNotReachHere();
2962 }
2963 {
2964 Label notVolatile;
2965 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2966 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2967 __ bind(notVolatile);
2968 }
2969 }
2970
2971 void TemplateTable::fast_xaccess(TosState state)
2972 {
2973 transition(vtos, state);
2974
2975 // get receiver
2976 __ ldr(r0, aaddress(0));
2977 // access constant pool cache
2978 __ get_cache_and_index_at_bcp(r2, r3, 2);
2979 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2980 ConstantPoolCacheEntry::f2_offset())));
2981 // make sure exception is reported in correct bcp range (getfield is
2982 // next instruction)
2983 __ increment(rbcp);
2984 __ null_check(r0);
2985 switch (state) {
2986 case itos:
2987 __ ldr(r0, Address(r0, r1, Address::lsl(0)));
2988 break;
2989 case atos:
2990 __ load_heap_oop(r0, Address(r0, r1, Address::lsl(0)));
2991 __ verify_oop(r0);
2992 break;
2993 case ftos:
2994 __ ldrs(v0, Address(r0, r1, Address::lsl(0)));
2995 break;
2996 default:
2997 ShouldNotReachHere();
2998 }
2999
3000 {
3001 Label notVolatile;
3002 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3003 ConstantPoolCacheEntry::flags_offset())));
3004 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3005 __ membar(MacroAssembler::LoadLoad);
3006 __ bind(notVolatile);
3007 }
3008
3009 __ decrement(rbcp);
3010 }
3011
3012
3013
3014 //-----------------------------------------------------------------------------
3015 // Calls
3016
3017 void TemplateTable::count_calls(Register method, Register temp)
3018 {
3019 __ call_Unimplemented();
3020 }
3021
3022 void TemplateTable::prepare_invoke(int byte_no,
3023 Register method, // linked method (or i-klass)
3024 Register index, // itable index, MethodType, etc.
3025 Register recv, // if caller wants to see it
3026 Register flags // if caller wants to test it
3027 ) {
3028 // determine flags
3029 Bytecodes::Code code = bytecode();
3030 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3031 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3032 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3033 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3034 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3035 const bool load_receiver = (recv != noreg);
3036 const bool save_flags = (flags != noreg);
3037 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3038 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3039 assert(flags == noreg || flags == r3, "");
3040 assert(recv == noreg || recv == r2, "");
3041
3042 // setup registers & access constant pool cache
3043 if (recv == noreg) recv = r2;
3044 if (flags == noreg) flags = r3;
3045 assert_different_registers(method, index, recv, flags);
3046
3047 // save 'interpreter return address'
3048 __ save_bcp();
3049
3050 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3051
3052 // maybe push appendix to arguments (just before return address)
3053 if (is_invokedynamic || is_invokehandle) {
3054 Label L_no_push;
3055 __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3056 // Push the appendix as a trailing parameter.
3057 // This must be done before we get the receiver,
3058 // since the parameter_size includes it.
3059 __ push(r19);
3060 __ mov(r19, index);
3061 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3062 __ load_resolved_reference_at_index(index, r19);
3063 __ pop(r19);
3064 __ push(index); // push appendix (MethodType, CallSite, etc.)
3065 __ bind(L_no_push);
3066 }
3067
3068 // load receiver if needed (note: no return address pushed yet)
3069 if (load_receiver) {
3070 __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3071 // FIXME -- is this actually correct? looks like it should be 2
3072 // const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3073 // const int receiver_is_at_end = -1; // back off one slot to get receiver
3074 // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3075 // __ movptr(recv, recv_addr);
3076 __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3077 __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3078 __ verify_oop(recv);
3079 }
3080
3081 // compute return type
3082 // x86 uses a shift and mask or wings it with a shift plus assert
3083 // the mask is not needed. aarch64 just uses bitfield extract
3084 __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
3085 // load return address
3086 {
3087 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3088 __ mov(rscratch1, table_addr);
3089 __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3090 }
3091 }
3092
3093
3094 void TemplateTable::invokevirtual_helper(Register index,
3095 Register recv,
3096 Register flags)
3097 {
3098 // Uses temporary registers r0, r3
3099 assert_different_registers(index, recv, r0, r3);
3100 // Test for an invoke of a final method
3101 Label notFinal;
3102 __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3103
3104 const Register method = index; // method must be rmethod
3105 assert(method == rmethod,
3106 "methodOop must be rmethod for interpreter calling convention");
3107
3108 // do the call - the index is actually the method to call
3109 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3110
3111 // It's final, need a null check here!
3112 __ null_check(recv);
3113
3114 // profile this call
3115 __ profile_final_call(r0);
3116 __ profile_arguments_type(r0, method, r4, true);
3117
3118 __ jump_from_interpreted(method, r0);
3119
3120 __ bind(notFinal);
3121
3122 // get receiver klass
3123 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3124 __ load_klass(r0, recv);
3125
3126 // profile this call
3127 __ profile_virtual_call(r0, rlocals, r3);
3128
3129 // get target methodOop & entry point
3130 __ lookup_virtual_method(r0, index, method);
3131 __ profile_arguments_type(r3, method, r4, true);
3132 // FIXME -- this looks completely redundant. is it?
3133 // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3134 __ jump_from_interpreted(method, r3);
3135 }
3136
3137 void TemplateTable::invokevirtual(int byte_no)
3138 {
3139 transition(vtos, vtos);
3140 assert(byte_no == f2_byte, "use this argument");
3141
3142 prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3143
3144 // rmethod: index (actually a Method*)
3145 // r2: receiver
3146 // r3: flags
3147
3148 invokevirtual_helper(rmethod, r2, r3);
3149 }
3150
3151 void TemplateTable::invokespecial(int byte_no)
3152 {
3153 transition(vtos, vtos);
3154 assert(byte_no == f1_byte, "use this argument");
3155
3156 prepare_invoke(byte_no, rmethod, noreg, // get f1 Method*
3157 r2); // get receiver also for null check
3158 __ verify_oop(r2);
3159 __ null_check(r2);
3160 // do the call
3161 __ profile_call(r0);
3162 __ profile_arguments_type(r0, rmethod, rbcp, false);
3163 __ jump_from_interpreted(rmethod, r0);
3164 }
3165
3166 void TemplateTable::invokestatic(int byte_no)
3167 {
3168 transition(vtos, vtos);
3169 assert(byte_no == f1_byte, "use this argument");
3170
3171 prepare_invoke(byte_no, rmethod); // get f1 Method*
3172 // do the call
3173 __ profile_call(r0);
3174 __ profile_arguments_type(r0, rmethod, r4, false);
3175 __ jump_from_interpreted(rmethod, r0);
3176 }
3177
3178 void TemplateTable::fast_invokevfinal(int byte_no)
3179 {
3180 __ call_Unimplemented();
3181 }
3182
3183 void TemplateTable::invokeinterface(int byte_no) {
3184 transition(vtos, vtos);
3185 assert(byte_no == f1_byte, "use this argument");
3186
3187 prepare_invoke(byte_no, r0, rmethod, // get f1 Klass*, f2 itable index
3188 r2, r3); // recv, flags
3189
3190 // r0: interface klass (from f1)
3191 // rmethod: itable index (from f2)
3192 // r2: receiver
3193 // r3: flags
3194
3195 // Special case of invokeinterface called for virtual method of
3196 // java.lang.Object. See cpCacheOop.cpp for details.
3197 // This code isn't produced by javac, but could be produced by
3198 // another compliant java compiler.
3199 Label notMethod;
3200 __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notMethod);
3201
3202 invokevirtual_helper(rmethod, r2, r3);
3203 __ bind(notMethod);
3204
3205 // Get receiver klass into r3 - also a null check
3206 __ restore_locals();
3207 __ null_check(r2, oopDesc::klass_offset_in_bytes());
3208 __ load_klass(r3, r2);
3209
3210 // profile this call
3211 __ profile_virtual_call(r3, r13, r19);
3212
3213 Label no_such_interface, no_such_method;
3214
3215 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3216 r3, r0, rmethod,
3217 // outputs: method, scan temp. reg
3218 rmethod, r13,
3219 no_such_interface);
3220
3221 // rmethod,: methodOop to call
3222 // r2: receiver
3223 // Check for abstract method error
3224 // Note: This should be done more efficiently via a throw_abstract_method_error
3225 // interpreter entry point and a conditional jump to it in case of a null
3226 // method.
3227 __ cbz(rmethod, no_such_method);
3228
3229 __ profile_arguments_type(r3, rmethod, r13, true);
3230
3231 // do the call
3232 // r2: receiver
3233 // rmethod,: methodOop
3234 __ jump_from_interpreted(rmethod, r3);
3235 __ should_not_reach_here();
3236
3237 // exception handling code follows...
3238 // note: must restore interpreter registers to canonical
3239 // state for exception handling to work correctly!
3240
3241 __ bind(no_such_method);
3242 // throw exception
3243 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3244 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3245 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3246 // the call_VM checks for exception, so we should never return here.
3247 __ should_not_reach_here();
3248
3249 __ bind(no_such_interface);
3250 // throw exception
3251 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3252 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3253 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3254 InterpreterRuntime::throw_IncompatibleClassChangeError));
3255 // the call_VM checks for exception, so we should never return here.
3256 __ should_not_reach_here();
3257 return;
3258 }
3259
3260 void TemplateTable::invokehandle(int byte_no) {
3261 transition(vtos, vtos);
3262 assert(byte_no == f1_byte, "use this argument");
3263
3264 prepare_invoke(byte_no, rmethod, r0, r2);
3265 __ verify_method_ptr(r2);
3266 __ verify_oop(r2);
3267 __ null_check(r2);
3268
3269 // FIXME: profile the LambdaForm also
3270
3271 // r13 is safe to use here as a scratch reg because it is about to
3272 // be clobbered by jump_from_interpreted().
3273 __ profile_final_call(r13);
3274 __ profile_arguments_type(r13, rmethod, r4, true);
3275
3276 __ jump_from_interpreted(rmethod, r0);
3277 }
3278
3279 void TemplateTable::invokedynamic(int byte_no) {
3280 transition(vtos, vtos);
3281 assert(byte_no == f1_byte, "use this argument");
3282
3283 prepare_invoke(byte_no, rmethod, r0);
3284
3285 // r0: CallSite object (from cpool->resolved_references[])
3286 // rmethod: MH.linkToCallSite method (from f2)
3287
3288 // Note: r0_callsite is already pushed by prepare_invoke
3289
3290 // %%% should make a type profile for any invokedynamic that takes a ref argument
3291 // profile this call
3292 __ profile_call(rbcp);
3293 __ profile_arguments_type(r3, rmethod, r13, false);
3294
3295 __ verify_oop(r0);
3296
3297 __ jump_from_interpreted(rmethod, r0);
3298 }
3299
3300
3301 //-----------------------------------------------------------------------------
3302 // Allocation
3303
3304 void TemplateTable::_new() {
3305 transition(vtos, atos);
3306
3307 __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3308 Label slow_case;
3309 Label done;
3310 Label initialize_header;
3311 Label initialize_object; // including clearing the fields
3312 Label allocate_shared;
3313
3314 __ get_cpool_and_tags(r4, r0);
3315 // Make sure the class we're about to instantiate has been resolved.
3316 // This is done before loading InstanceKlass to be consistent with the order
3317 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3318 const int tags_offset = Array<u1>::base_offset_in_bytes();
3319 __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3320 __ lea(rscratch1, Address(rscratch1, tags_offset));
3321 __ ldarb(rscratch1, rscratch1);
3322 __ cmp(rscratch1, JVM_CONSTANT_Class);
3323 __ br(Assembler::NE, slow_case);
3324
3325 // get InstanceKlass
3326 __ lea(r4, Address(r4, r3, Address::lsl(3)));
3327 __ ldr(r4, Address(r4, sizeof(ConstantPool)));
3328
3329 // make sure klass is initialized & doesn't have finalizer
3330 // make sure klass is fully initialized
3331 __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3332 __ cmp(rscratch1, InstanceKlass::fully_initialized);
3333 __ br(Assembler::NE, slow_case);
3334
3335 // get instance_size in InstanceKlass (scaled to a count of bytes)
3336 __ ldrw(r3,
3337 Address(r4,
3338 Klass::layout_helper_offset()));
3339 // test to see if it has a finalizer or is malformed in some way
3340 __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3341
3342 // Allocate the instance
3343 // 1) Try to allocate in the TLAB
3344 // 2) if fail and the object is large allocate in the shared Eden
3345 // 3) if the above fails (or is not applicable), go to a slow case
3346 // (creates a new TLAB, etc.)
3347
3348 const bool allow_shared_alloc =
3349 Universe::heap()->supports_inline_contig_alloc();
3350
3351 if (UseTLAB) {
3352 __ tlab_allocate(r0, r3, 0, noreg, r1,
3353 allow_shared_alloc ? allocate_shared : slow_case);
3354
3355 if (ZeroTLAB) {
3356 // the fields have been already cleared
3357 __ b(initialize_header);
3358 } else {
3359 // initialize both the header and fields
3360 __ b(initialize_object);
3361 }
3362 }
3363
3364 // Allocation in the shared Eden, if allowed.
3365 //
3366 // r3: instance size in bytes
3367 if (allow_shared_alloc) {
3368 __ bind(allocate_shared);
3369
3370 __ eden_allocate(r0, r3, 0, r10, slow_case);
3371 __ incr_allocated_bytes(rthread, r3, 0, rscratch1);
3372 }
3373
3374 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3375 // The object is initialized before the header. If the object size is
3376 // zero, go directly to the header initialization.
3377 __ bind(initialize_object);
3378 __ sub(r3, r3, sizeof(oopDesc));
3379 __ cbz(r3, initialize_header);
3380
3381 // Initialize object fields
3382 {
3383 __ add(r2, r0, sizeof(oopDesc));
3384 Label loop;
3385 __ bind(loop);
3386 __ str(zr, Address(__ post(r2, BytesPerLong)));
3387 __ sub(r3, r3, BytesPerLong);
3388 __ cbnz(r3, loop);
3389 }
3390
3391 // initialize object header only.
3392 __ bind(initialize_header);
3393 if (UseBiasedLocking) {
3394 __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3395 } else {
3396 __ mov(rscratch1, (intptr_t)markOopDesc::prototype());
3397 }
3398 __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3399 __ store_klass_gap(r0, zr); // zero klass gap for compressed oops
3400 __ store_klass(r0, r4); // store klass last
3401
3402 {
3403 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3404 // Trigger dtrace event for fastpath
3405 __ push(atos); // save the return value
3406 __ call_VM_leaf(
3407 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3408 __ pop(atos); // restore the return value
3409
3410 }
3411 __ b(done);
3412 }
3413
3414 // slow case
3415 __ bind(slow_case);
3416 __ get_constant_pool(c_rarg1);
3417 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3418 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3419 __ verify_oop(r0);
3420
3421 // continue
3422 __ bind(done);
3423 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3424 __ membar(Assembler::StoreStore);
3425 }
3426
3427 void TemplateTable::newarray() {
3428 transition(itos, atos);
3429 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3430 __ mov(c_rarg2, r0);
3431 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3432 c_rarg1, c_rarg2);
3433 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3434 __ membar(Assembler::StoreStore);
3435 }
3436
3437 void TemplateTable::anewarray() {
3438 transition(itos, atos);
3439 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3440 __ get_constant_pool(c_rarg1);
3441 __ mov(c_rarg3, r0);
3442 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3443 c_rarg1, c_rarg2, c_rarg3);
3444 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3445 __ membar(Assembler::StoreStore);
3446 }
3447
3448 void TemplateTable::arraylength() {
3449 transition(atos, itos);
3450 __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3451 __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3452 }
3453
3454 void TemplateTable::checkcast()
3455 {
3456 transition(atos, atos);
3457 Label done, is_null, ok_is_subtype, quicked, resolved;
3458 __ cbz(r0, is_null);
3459
3460 // Get cpool & tags index
3461 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3462 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3463 // See if bytecode has already been quicked
3464 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3465 __ lea(r1, Address(rscratch1, r19));
3466 __ ldarb(r1, r1);
3467 __ cmp(r1, JVM_CONSTANT_Class);
3468 __ br(Assembler::EQ, quicked);
3469
3470 __ push(atos); // save receiver for result, and for GC
3471 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3472 // vm_result_2 has metadata result
3473 __ get_vm_result_2(r0, rthread);
3474 __ pop(r3); // restore receiver
3475 __ b(resolved);
3476
3477 // Get superklass in r0 and subklass in r3
3478 __ bind(quicked);
3479 __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3480 __ lea(r0, Address(r2, r19, Address::lsl(3)));
3481 __ ldr(r0, Address(r0, sizeof(ConstantPool)));
3482
3483 __ bind(resolved);
3484 __ load_klass(r19, r3);
3485
3486 // Generate subtype check. Blows r2, r5. Object in r3.
3487 // Superklass in r0. Subklass in r19.
3488 __ gen_subtype_check(r19, ok_is_subtype);
3489
3490 // Come here on failure
3491 __ push(r3);
3492 // object is at TOS
3493 __ b(Interpreter::_throw_ClassCastException_entry);
3494
3495 // Come here on success
3496 __ bind(ok_is_subtype);
3497 __ mov(r0, r3); // Restore object in r3
3498
3499 // Collect counts on whether this test sees NULLs a lot or not.
3500 if (ProfileInterpreter) {
3501 __ b(done);
3502 __ bind(is_null);
3503 __ profile_null_seen(r2);
3504 } else {
3505 __ bind(is_null); // same as 'done'
3506 }
3507 __ bind(done);
3508 }
3509
3510 void TemplateTable::instanceof() {
3511 transition(atos, itos);
3512 Label done, is_null, ok_is_subtype, quicked, resolved;
3513 __ cbz(r0, is_null);
3514
3515 // Get cpool & tags index
3516 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3517 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3518 // See if bytecode has already been quicked
3519 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3520 __ lea(r1, Address(rscratch1, r19));
3521 __ ldarb(r1, r1);
3522 __ cmp(r1, JVM_CONSTANT_Class);
3523 __ br(Assembler::EQ, quicked);
3524
3525 __ push(atos); // save receiver for result, and for GC
3526 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3527 // vm_result_2 has metadata result
3528 __ get_vm_result_2(r0, rthread);
3529 __ pop(r3); // restore receiver
3530 __ verify_oop(r3);
3531 __ load_klass(r3, r3);
3532 __ b(resolved);
3533
3534 // Get superklass in r0 and subklass in r3
3535 __ bind(quicked);
3536 __ load_klass(r3, r0);
3537 __ lea(r0, Address(r2, r19, Address::lsl(3)));
3538 __ ldr(r0, Address(r0, sizeof(ConstantPool)));
3539
3540 __ bind(resolved);
3541
3542 // Generate subtype check. Blows r2, r5
3543 // Superklass in r0. Subklass in r3.
3544 __ gen_subtype_check(r3, ok_is_subtype);
3545
3546 // Come here on failure
3547 __ mov(r0, 0);
3548 __ b(done);
3549 // Come here on success
3550 __ bind(ok_is_subtype);
3551 __ mov(r0, 1);
3552
3553 // Collect counts on whether this test sees NULLs a lot or not.
3554 if (ProfileInterpreter) {
3555 __ b(done);
3556 __ bind(is_null);
3557 __ profile_null_seen(r2);
3558 } else {
3559 __ bind(is_null); // same as 'done'
3560 }
3561 __ bind(done);
3562 // r0 = 0: obj == NULL or obj is not an instanceof the specified klass
3563 // r0 = 1: obj != NULL and obj is an instanceof the specified klass
3564 }
3565
3566 //-----------------------------------------------------------------------------
3567 // Breakpoints
3568 void TemplateTable::_breakpoint() {
3569 // Note: We get here even if we are single stepping..
3570 // jbug inists on setting breakpoints at every bytecode
3571 // even if we are in single step mode.
3572
3573 transition(vtos, vtos);
3574
3575 // get the unpatched byte code
3576 __ get_method(c_rarg1);
3577 __ call_VM(noreg,
3578 CAST_FROM_FN_PTR(address,
3579 InterpreterRuntime::get_original_bytecode_at),
3580 c_rarg1, rbcp);
3581 __ mov(r19, r0);
3582
3583 // post the breakpoint event
3584 __ call_VM(noreg,
3585 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3586 rmethod, rbcp);
3587
3588 // complete the execution of original bytecode
3589 __ mov(rscratch1, r19);
3590 __ dispatch_only_normal(vtos);
3591 }
3592
3593 //-----------------------------------------------------------------------------
3594 // Exceptions
3595
3596 void TemplateTable::athrow() {
3597 transition(atos, vtos);
3598 __ null_check(r0);
3599 __ b(Interpreter::throw_exception_entry());
3600 }
3601
3602 //-----------------------------------------------------------------------------
3603 // Synchronization
3604 //
3605 // Note: monitorenter & exit are symmetric routines; which is reflected
3606 // in the assembly code structure as well
3607 //
3608 // Stack layout:
3609 //
3610 // [expressions ] <--- esp = expression stack top
3611 // ..
3612 // [expressions ]
3613 // [monitor entry] <--- monitor block top = expression stack bot
3614 // ..
3615 // [monitor entry]
3616 // [frame data ] <--- monitor block bot
3617 // ...
3618 // [saved rbp ] <--- rbp
3619 void TemplateTable::monitorenter()
3620 {
3621 transition(atos, vtos);
3622
3623 // check for NULL object
3624 __ null_check(r0);
3625
3626 const Address monitor_block_top(
3627 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3628 const Address monitor_block_bot(
3629 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3630 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3631
3632 Label allocated;
3633
3634 // initialize entry pointer
3635 __ mov(c_rarg1, zr); // points to free slot or NULL
3636
3637 // find a free slot in the monitor block (result in c_rarg1)
3638 {
3639 Label entry, loop, exit;
3640 __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3641 // starting with top-most entry
3642 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3643
3644 __ b(entry);
3645
3646 __ bind(loop);
3647 // check if current entry is used
3648 // if not used then remember entry in c_rarg1
3649 __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3650 __ cmp(zr, rscratch1);
3651 __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3652 // check if current entry is for same object
3653 __ cmp(r0, rscratch1);
3654 // if same object then stop searching
3655 __ br(Assembler::EQ, exit);
3656 // otherwise advance to next entry
3657 __ add(c_rarg3, c_rarg3, entry_size);
3658 __ bind(entry);
3659 // check if bottom reached
3660 __ cmp(c_rarg3, c_rarg2);
3661 // if not at bottom then check this entry
3662 __ br(Assembler::NE, loop);
3663 __ bind(exit);
3664 }
3665
3666 __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3667 // if found, continue with that on
3668
3669 // allocate one if there's no free slot
3670 {
3671 Label entry, loop, no_adjust;
3672 // 1. compute new pointers // rsp: old expression stack top
3673 __ ldr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
3674 __ sub(esp, esp, entry_size); // move expression stack top
3675 __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3676 __ mov(c_rarg3, esp); // set start value for copy loop
3677 __ str(c_rarg1, monitor_block_bot); // set new monitor block bottom
3678
3679 __ cmp(sp, c_rarg3); // Check if we need to move sp
3680 __ br(Assembler::LO, no_adjust); // to allow more stack space
3681 // for our new esp
3682 __ sub(sp, sp, 2 * wordSize);
3683 __ bind(no_adjust);
3684
3685 __ b(entry);
3686 // 2. move expression stack contents
3687 __ bind(loop);
3688 __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3689 // word from old location
3690 __ str(c_rarg2, Address(c_rarg3, 0)); // and store it at new location
3691 __ add(c_rarg3, c_rarg3, wordSize); // advance to next word
3692 __ bind(entry);
3693 __ cmp(c_rarg3, c_rarg1); // check if bottom reached
3694 __ br(Assembler::NE, loop); // if not at bottom then
3695 // copy next word
3696 }
3697
3698 // call run-time routine
3699 // c_rarg1: points to monitor entry
3700 __ bind(allocated);
3701
3702 // Increment bcp to point to the next bytecode, so exception
3703 // handling for async. exceptions work correctly.
3704 // The object has already been poped from the stack, so the
3705 // expression stack looks correct.
3706 __ increment(rbcp);
3707
3708 // store object
3709 __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3710 __ lock_object(c_rarg1);
3711
3712 // check to make sure this monitor doesn't cause stack overflow after locking
3713 __ save_bcp(); // in case of exception
3714 __ generate_stack_overflow_check(0);
3715
3716 // The bcp has already been incremented. Just need to dispatch to
3717 // next instruction.
3718 __ dispatch_next(vtos);
3719 }
3720
3721
3722 void TemplateTable::monitorexit()
3723 {
3724 transition(atos, vtos);
3725
3726 // check for NULL object
3727 __ null_check(r0);
3728
3729 const Address monitor_block_top(
3730 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3731 const Address monitor_block_bot(
3732 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3733 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3734
3735 Label found;
3736
3737 // find matching slot
3738 {
3739 Label entry, loop;
3740 __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3741 // starting with top-most entry
3742 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3743 // of monitor block
3744 __ b(entry);
3745
3746 __ bind(loop);
3747 // check if current entry is for same object
3748 __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3749 __ cmp(r0, rscratch1);
3750 // if same object then stop searching
3751 __ br(Assembler::EQ, found);
3752 // otherwise advance to next entry
3753 __ add(c_rarg1, c_rarg1, entry_size);
3754 __ bind(entry);
3755 // check if bottom reached
3756 __ cmp(c_rarg1, c_rarg2);
3757 // if not at bottom then check this entry
3758 __ br(Assembler::NE, loop);
3759 }
3760
3761 // error handling. Unlocking was not block-structured
3762 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3763 InterpreterRuntime::throw_illegal_monitor_state_exception));
3764 __ should_not_reach_here();
3765
3766 // call run-time routine
3767 __ bind(found);
3768 __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3769 __ unlock_object(c_rarg1);
3770 __ pop_ptr(r0); // discard object
3771 }
3772
3773
3774 // Wide instructions
3775 void TemplateTable::wide()
3776 {
3777 __ load_unsigned_byte(r19, at_bcp(1));
3778 __ mov(rscratch1, (address)Interpreter::_wentry_point);
3779 __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
3780 __ br(rscratch1);
3781 }
3782
3783
3784 // Multi arrays
3785 void TemplateTable::multianewarray() {
3786 transition(vtos, atos);
3787 __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
3788 // last dim is on top of stack; we want address of first one:
3789 // first_addr = last_addr + (ndims - 1) * wordSize
3790 __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
3791 __ sub(c_rarg1, c_rarg1, wordSize);
3792 call_VM(r0,
3793 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3794 c_rarg1);
3795 __ load_unsigned_byte(r1, at_bcp(3));
3796 __ lea(esp, Address(esp, r1, Address::uxtw(3)));
3797 }
3798 #endif // !CC_INTERP