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 #define __ _masm->
43
44 // Platform-dependent initialization
45
46 void TemplateTable::pd_initialize() {
47 // No aarch64 specific initialization
48 }
49
50 // Address computation: local variables
51
52 static inline Address iaddress(int n) {
53 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
54 }
55
56 static inline Address laddress(int n) {
57 return iaddress(n + 1);
58 }
59
60 static inline Address faddress(int n) {
61 return iaddress(n);
62 }
63
64 static inline Address daddress(int n) {
65 return laddress(n);
66 }
67
68 static inline Address aaddress(int n) {
69 return iaddress(n);
70 }
71
72 static inline Address iaddress(Register r) {
73 return Address(rlocals, r, Address::lsl(3));
74 }
75
76 static inline Address laddress(Register r, Register scratch,
77 InterpreterMacroAssembler* _masm) {
78 __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
79 return Address(scratch, Interpreter::local_offset_in_bytes(1));
80 }
81
82 static inline Address faddress(Register r) {
83 return iaddress(r);
84 }
85
86 static inline Address daddress(Register r, Register scratch,
87 InterpreterMacroAssembler* _masm) {
88 return laddress(r, scratch, _masm);
89 }
90
91 static inline Address aaddress(Register r) {
92 return iaddress(r);
93 }
94
95 static inline Address at_rsp() {
96 return Address(esp, 0);
97 }
98
99 // At top of Java expression stack which may be different than esp(). It
100 // isn't for category 1 objects.
101 static inline Address at_tos () {
102 return Address(esp, Interpreter::expr_offset_in_bytes(0));
103 }
104
105 static inline Address at_tos_p1() {
106 return Address(esp, Interpreter::expr_offset_in_bytes(1));
107 }
108
109 static inline Address at_tos_p2() {
110 return Address(esp, Interpreter::expr_offset_in_bytes(2));
111 }
112
113 static inline Address at_tos_p3() {
114 return Address(esp, Interpreter::expr_offset_in_bytes(3));
115 }
116
117 static inline Address at_tos_p4() {
118 return Address(esp, Interpreter::expr_offset_in_bytes(4));
119 }
120
121 static inline Address at_tos_p5() {
122 return Address(esp, Interpreter::expr_offset_in_bytes(5));
123 }
124
125 // Condition conversion
126 static Assembler::Condition j_not(TemplateTable::Condition cc) {
127 switch (cc) {
128 case TemplateTable::equal : return Assembler::NE;
129 case TemplateTable::not_equal : return Assembler::EQ;
130 case TemplateTable::less : return Assembler::GE;
131 case TemplateTable::less_equal : return Assembler::GT;
132 case TemplateTable::greater : return Assembler::LE;
133 case TemplateTable::greater_equal: return Assembler::LT;
134 }
135 ShouldNotReachHere();
136 return Assembler::EQ;
137 }
138
139
140 // Miscelaneous helper routines
141 // Store an oop (or NULL) at the Address described by obj.
142 // If val == noreg this means store a NULL
143 static void do_oop_store(InterpreterMacroAssembler* _masm,
144 Address obj,
145 Register val,
146 BarrierSet::Name barrier,
147 bool precise) {
148 assert(val == noreg || val == r0, "parameter is just for looks");
149 switch (barrier) {
150 #if INCLUDE_ALL_GCS
151 case BarrierSet::G1SATBCTLogging:
152 {
153 // flatten object address if needed
154 if (obj.index() == noreg && obj.offset() == 0) {
155 if (obj.base() != r3) {
156 __ mov(r3, obj.base());
157 }
158 } else {
159 __ lea(r3, obj);
160 }
161 __ g1_write_barrier_pre(r3 /* obj */,
162 r1 /* pre_val */,
163 rthread /* thread */,
164 r10 /* tmp */,
165 val != noreg /* tosca_live */,
166 false /* expand_call */);
167 if (val == noreg) {
168 __ store_heap_oop_null(Address(r3, 0));
169 } else {
170 // G1 barrier needs uncompressed oop for region cross check.
171 Register new_val = val;
172 if (UseCompressedOops) {
173 new_val = rscratch1;
174 __ mov(new_val, val);
175 }
176 __ store_heap_oop(Address(r3, 0), val);
177 __ g1_write_barrier_post(r3 /* store_adr */,
178 new_val /* new_val */,
179 rthread /* thread */,
180 r10 /* tmp */,
181 r1 /* tmp2 */);
182 }
183
184 }
185 break;
186 #endif // INCLUDE_ALL_GCS
187 case BarrierSet::CardTableForRS:
188 case BarrierSet::CardTableExtension:
189 {
190 if (val == noreg) {
191 __ store_heap_oop_null(obj);
192 } else {
193 __ store_heap_oop(obj, val);
194 // flatten object address if needed
195 if (!precise || (obj.index() == noreg && obj.offset() == 0)) {
196 __ store_check(obj.base());
197 } else {
198 __ lea(r3, obj);
199 __ store_check(r3);
200 }
201 }
202 }
203 break;
204 case BarrierSet::ModRef:
205 if (val == noreg) {
206 __ store_heap_oop_null(obj);
207 } else {
208 __ store_heap_oop(obj, val);
209 }
210 break;
211 default :
212 ShouldNotReachHere();
213
214 }
215 }
216
217 Address TemplateTable::at_bcp(int offset) {
218 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
219 return Address(rbcp, offset);
220 }
221
222 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
223 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
224 int byte_no)
225 {
226 if (!RewriteBytecodes) return;
227 Label L_patch_done;
228
229 switch (bc) {
230 case Bytecodes::_fast_aputfield:
231 case Bytecodes::_fast_bputfield:
232 case Bytecodes::_fast_zputfield:
233 case Bytecodes::_fast_cputfield:
234 case Bytecodes::_fast_dputfield:
235 case Bytecodes::_fast_fputfield:
236 case Bytecodes::_fast_iputfield:
237 case Bytecodes::_fast_lputfield:
238 case Bytecodes::_fast_sputfield:
239 {
240 // We skip bytecode quickening for putfield instructions when
241 // the put_code written to the constant pool cache is zero.
242 // This is required so that every execution of this instruction
243 // calls out to InterpreterRuntime::resolve_get_put to do
244 // additional, required work.
245 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
246 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
247 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
248 __ movw(bc_reg, bc);
249 __ cmpw(temp_reg, (unsigned) 0);
250 __ br(Assembler::EQ, L_patch_done); // don't patch
251 }
252 break;
253 default:
254 assert(byte_no == -1, "sanity");
255 // the pair bytecodes have already done the load.
256 if (load_bc_into_bc_reg) {
257 __ movw(bc_reg, bc);
258 }
259 }
260
261 if (JvmtiExport::can_post_breakpoint()) {
262 Label L_fast_patch;
263 // if a breakpoint is present we can't rewrite the stream directly
264 __ load_unsigned_byte(temp_reg, at_bcp(0));
265 __ cmpw(temp_reg, Bytecodes::_breakpoint);
266 __ br(Assembler::NE, L_fast_patch);
267 // Let breakpoint table handling rewrite to quicker bytecode
268 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
269 __ b(L_patch_done);
270 __ bind(L_fast_patch);
271 }
272
273 #ifdef ASSERT
274 Label L_okay;
275 __ load_unsigned_byte(temp_reg, at_bcp(0));
276 __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
277 __ br(Assembler::EQ, L_okay);
278 __ cmpw(temp_reg, bc_reg);
279 __ br(Assembler::EQ, L_okay);
280 __ stop("patching the wrong bytecode");
281 __ bind(L_okay);
282 #endif
283
284 // patch bytecode
285 __ strb(bc_reg, at_bcp(0));
286 __ bind(L_patch_done);
287 }
288
289
290 // Individual instructions
291
292 void TemplateTable::nop() {
293 transition(vtos, vtos);
294 // nothing to do
295 }
296
297 void TemplateTable::shouldnotreachhere() {
298 transition(vtos, vtos);
299 __ stop("shouldnotreachhere bytecode");
300 }
301
302 void TemplateTable::aconst_null()
303 {
304 transition(vtos, atos);
305 __ mov(r0, 0);
306 }
307
308 void TemplateTable::iconst(int value)
309 {
310 transition(vtos, itos);
311 __ mov(r0, value);
312 }
313
314 void TemplateTable::lconst(int value)
315 {
316 __ mov(r0, value);
317 }
318
319 void TemplateTable::fconst(int value)
320 {
321 transition(vtos, ftos);
322 switch (value) {
323 case 0:
324 __ fmovs(v0, zr);
325 break;
326 case 1:
327 __ fmovs(v0, 1.0);
328 break;
329 case 2:
330 __ fmovs(v0, 2.0);
331 break;
332 default:
333 ShouldNotReachHere();
334 break;
335 }
336 }
337
338 void TemplateTable::dconst(int value)
339 {
340 transition(vtos, dtos);
341 switch (value) {
342 case 0:
343 __ fmovd(v0, zr);
344 break;
345 case 1:
346 __ fmovd(v0, 1.0);
347 break;
348 case 2:
349 __ fmovd(v0, 2.0);
350 break;
351 default:
352 ShouldNotReachHere();
353 break;
354 }
355 }
356
357 void TemplateTable::bipush()
358 {
359 transition(vtos, itos);
360 __ load_signed_byte32(r0, at_bcp(1));
361 }
362
363 void TemplateTable::sipush()
364 {
365 transition(vtos, itos);
366 __ load_unsigned_short(r0, at_bcp(1));
367 __ revw(r0, r0);
368 __ asrw(r0, r0, 16);
369 }
370
371 void TemplateTable::ldc(bool wide)
372 {
373 transition(vtos, vtos);
374 Label call_ldc, notFloat, notClass, Done;
375
376 if (wide) {
377 __ get_unsigned_2_byte_index_at_bcp(r1, 1);
378 } else {
379 __ load_unsigned_byte(r1, at_bcp(1));
380 }
381 __ get_cpool_and_tags(r2, r0);
382
383 const int base_offset = ConstantPool::header_size() * wordSize;
384 const int tags_offset = Array<u1>::base_offset_in_bytes();
385
386 // get type
387 __ add(r3, r1, tags_offset);
388 __ lea(r3, Address(r0, r3));
389 __ ldarb(r3, r3);
390
391 // unresolved class - get the resolved class
392 __ cmp(r3, JVM_CONSTANT_UnresolvedClass);
393 __ br(Assembler::EQ, call_ldc);
394
395 // unresolved class in error state - call into runtime to throw the error
396 // from the first resolution attempt
397 __ cmp(r3, JVM_CONSTANT_UnresolvedClassInError);
398 __ br(Assembler::EQ, call_ldc);
399
400 // resolved class - need to call vm to get java mirror of the class
401 __ cmp(r3, JVM_CONSTANT_Class);
402 __ br(Assembler::NE, notClass);
403
404 __ bind(call_ldc);
405 __ mov(c_rarg1, wide);
406 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
407 __ push_ptr(r0);
408 __ verify_oop(r0);
409 __ b(Done);
410
411 __ bind(notClass);
412 __ cmp(r3, JVM_CONSTANT_Float);
413 __ br(Assembler::NE, notFloat);
414 // ftos
415 __ adds(r1, r2, r1, Assembler::LSL, 3);
416 __ ldrs(v0, Address(r1, base_offset));
417 __ push_f();
418 __ b(Done);
419
420 __ bind(notFloat);
421 #ifdef ASSERT
422 {
423 Label L;
424 __ cmp(r3, JVM_CONSTANT_Integer);
425 __ br(Assembler::EQ, L);
426 // String and Object are rewritten to fast_aldc
427 __ stop("unexpected tag type in ldc");
428 __ bind(L);
429 }
430 #endif
431 // itos JVM_CONSTANT_Integer only
432 __ adds(r1, r2, r1, Assembler::LSL, 3);
433 __ ldrw(r0, Address(r1, base_offset));
434 __ push_i(r0);
435 __ bind(Done);
436 }
437
438 // Fast path for caching oop constants.
439 void TemplateTable::fast_aldc(bool wide)
440 {
441 transition(vtos, atos);
442
443 Register result = r0;
444 Register tmp = r1;
445 int index_size = wide ? sizeof(u2) : sizeof(u1);
446
447 Label resolved;
448
449 // We are resolved if the resolved reference cache entry contains a
450 // non-null object (String, MethodType, etc.)
451 assert_different_registers(result, tmp);
452 __ get_cache_index_at_bcp(tmp, 1, index_size);
453 __ load_resolved_reference_at_index(result, tmp);
454 __ cbnz(result, resolved);
455
456 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
457
458 // first time invocation - must resolve first
459 __ mov(tmp, (int)bytecode());
460 __ call_VM(result, entry, tmp);
461
462 __ bind(resolved);
463
464 if (VerifyOops) {
465 __ verify_oop(result);
466 }
467 }
468
469 void TemplateTable::ldc2_w()
470 {
471 transition(vtos, vtos);
472 Label Long, Done;
473 __ get_unsigned_2_byte_index_at_bcp(r0, 1);
474
475 __ get_cpool_and_tags(r1, r2);
476 const int base_offset = ConstantPool::header_size() * wordSize;
477 const int tags_offset = Array<u1>::base_offset_in_bytes();
478
479 // get type
480 __ lea(r2, Address(r2, r0, Address::lsl(0)));
481 __ load_unsigned_byte(r2, Address(r2, tags_offset));
482 __ cmpw(r2, (int)JVM_CONSTANT_Double);
483 __ br(Assembler::NE, Long);
484 // dtos
485 __ lea (r2, Address(r1, r0, Address::lsl(3)));
486 __ ldrd(v0, Address(r2, base_offset));
487 __ push_d();
488 __ b(Done);
489
490 __ bind(Long);
491 // ltos
492 __ lea(r0, Address(r1, r0, Address::lsl(3)));
493 __ ldr(r0, Address(r0, base_offset));
494 __ push_l();
495
496 __ bind(Done);
497 }
498
499 void TemplateTable::locals_index(Register reg, int offset)
500 {
501 __ ldrb(reg, at_bcp(offset));
502 __ neg(reg, reg);
503 }
504
505 void TemplateTable::iload() {
506 iload_internal();
507 }
508
509 void TemplateTable::nofast_iload() {
510 iload_internal(may_not_rewrite);
511 }
512
513 void TemplateTable::iload_internal(RewriteControl rc) {
514 transition(vtos, itos);
515 if (RewriteFrequentPairs && rc == may_rewrite) {
516 Label rewrite, done;
517 Register bc = r4;
518
519 // get next bytecode
520 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
521
522 // if _iload, wait to rewrite to iload2. We only want to rewrite the
523 // last two iloads in a pair. Comparing against fast_iload means that
524 // the next bytecode is neither an iload or a caload, and therefore
525 // an iload pair.
526 __ cmpw(r1, Bytecodes::_iload);
527 __ br(Assembler::EQ, done);
528
529 // if _fast_iload rewrite to _fast_iload2
530 __ cmpw(r1, Bytecodes::_fast_iload);
531 __ movw(bc, Bytecodes::_fast_iload2);
532 __ br(Assembler::EQ, rewrite);
533
534 // if _caload rewrite to _fast_icaload
535 __ cmpw(r1, Bytecodes::_caload);
536 __ movw(bc, Bytecodes::_fast_icaload);
537 __ br(Assembler::EQ, rewrite);
538
539 // else rewrite to _fast_iload
540 __ movw(bc, Bytecodes::_fast_iload);
541
542 // rewrite
543 // bc: new bytecode
544 __ bind(rewrite);
545 patch_bytecode(Bytecodes::_iload, bc, r1, false);
546 __ bind(done);
547
548 }
549
550 // do iload, get the local value into tos
551 locals_index(r1);
552 __ ldr(r0, iaddress(r1));
553
554 }
555
556 void TemplateTable::fast_iload2()
557 {
558 transition(vtos, itos);
559 locals_index(r1);
560 __ ldr(r0, iaddress(r1));
561 __ push(itos);
562 locals_index(r1, 3);
563 __ ldr(r0, iaddress(r1));
564 }
565
566 void TemplateTable::fast_iload()
567 {
568 transition(vtos, itos);
569 locals_index(r1);
570 __ ldr(r0, iaddress(r1));
571 }
572
573 void TemplateTable::lload()
574 {
575 transition(vtos, ltos);
576 __ ldrb(r1, at_bcp(1));
577 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
578 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
579 }
580
581 void TemplateTable::fload()
582 {
583 transition(vtos, ftos);
584 locals_index(r1);
585 // n.b. we use ldrd here because this is a 64 bit slot
586 // this is comparable to the iload case
587 __ ldrd(v0, faddress(r1));
588 }
589
590 void TemplateTable::dload()
591 {
592 transition(vtos, dtos);
593 __ ldrb(r1, at_bcp(1));
594 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
595 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
596 }
597
598 void TemplateTable::aload()
599 {
600 transition(vtos, atos);
601 locals_index(r1);
602 __ ldr(r0, iaddress(r1));
603 }
604
605 void TemplateTable::locals_index_wide(Register reg) {
606 __ ldrh(reg, at_bcp(2));
607 __ rev16w(reg, reg);
608 __ neg(reg, reg);
609 }
610
611 void TemplateTable::wide_iload() {
612 transition(vtos, itos);
613 locals_index_wide(r1);
614 __ ldr(r0, iaddress(r1));
615 }
616
617 void TemplateTable::wide_lload()
618 {
619 transition(vtos, ltos);
620 __ ldrh(r1, at_bcp(2));
621 __ rev16w(r1, r1);
622 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
623 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
624 }
625
626 void TemplateTable::wide_fload()
627 {
628 transition(vtos, ftos);
629 locals_index_wide(r1);
630 // n.b. we use ldrd here because this is a 64 bit slot
631 // this is comparable to the iload case
632 __ ldrd(v0, faddress(r1));
633 }
634
635 void TemplateTable::wide_dload()
636 {
637 transition(vtos, dtos);
638 __ ldrh(r1, at_bcp(2));
639 __ rev16w(r1, r1);
640 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
641 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
642 }
643
644 void TemplateTable::wide_aload()
645 {
646 transition(vtos, atos);
647 locals_index_wide(r1);
648 __ ldr(r0, aaddress(r1));
649 }
650
651 void TemplateTable::index_check(Register array, Register index)
652 {
653 // destroys r1, rscratch1
654 // check array
655 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
656 // sign extend index for use by indexed load
657 // __ movl2ptr(index, index);
658 // check index
659 Register length = rscratch1;
660 __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
661 __ cmpw(index, length);
662 if (index != r1) {
663 // ??? convention: move aberrant index into r1 for exception message
664 assert(r1 != array, "different registers");
665 __ mov(r1, index);
666 }
667 Label ok;
668 __ br(Assembler::LO, ok);
669 __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
670 __ br(rscratch1);
671 __ bind(ok);
672 }
673
674 void TemplateTable::iaload()
675 {
676 transition(itos, itos);
677 __ mov(r1, r0);
678 __ pop_ptr(r0);
679 // r0: array
680 // r1: index
681 index_check(r0, r1); // leaves index in r1, kills rscratch1
682 __ lea(r1, Address(r0, r1, Address::uxtw(2)));
683 __ ldrw(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_INT)));
684 }
685
686 void TemplateTable::laload()
687 {
688 transition(itos, ltos);
689 __ mov(r1, r0);
690 __ pop_ptr(r0);
691 // r0: array
692 // r1: index
693 index_check(r0, r1); // leaves index in r1, kills rscratch1
694 __ lea(r1, Address(r0, r1, Address::uxtw(3)));
695 __ ldr(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_LONG)));
696 }
697
698 void TemplateTable::faload()
699 {
700 transition(itos, ftos);
701 __ mov(r1, r0);
702 __ pop_ptr(r0);
703 // r0: array
704 // r1: index
705 index_check(r0, r1); // leaves index in r1, kills rscratch1
706 __ lea(r1, Address(r0, r1, Address::uxtw(2)));
707 __ ldrs(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
708 }
709
710 void TemplateTable::daload()
711 {
712 transition(itos, dtos);
713 __ mov(r1, r0);
714 __ pop_ptr(r0);
715 // r0: array
716 // r1: index
717 index_check(r0, r1); // leaves index in r1, kills rscratch1
718 __ lea(r1, Address(r0, r1, Address::uxtw(3)));
719 __ ldrd(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
720 }
721
722 void TemplateTable::aaload()
723 {
724 transition(itos, atos);
725 __ mov(r1, r0);
726 __ pop_ptr(r0);
727 // r0: array
728 // r1: index
729 index_check(r0, r1); // leaves index in r1, kills rscratch1
730 int s = (UseCompressedOops ? 2 : 3);
731 __ lea(r1, Address(r0, r1, Address::uxtw(s)));
732 __ load_heap_oop(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
733 }
734
735 void TemplateTable::baload()
736 {
737 transition(itos, itos);
738 __ mov(r1, r0);
739 __ pop_ptr(r0);
740 // r0: array
741 // r1: index
742 index_check(r0, r1); // leaves index in r1, kills rscratch1
743 __ lea(r1, Address(r0, r1, Address::uxtw(0)));
744 __ load_signed_byte(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
745 }
746
747 void TemplateTable::caload()
748 {
749 transition(itos, itos);
750 __ mov(r1, r0);
751 __ pop_ptr(r0);
752 // r0: array
753 // r1: index
754 index_check(r0, r1); // leaves index in r1, kills rscratch1
755 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
756 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
757 }
758
759 // iload followed by caload frequent pair
760 void TemplateTable::fast_icaload()
761 {
762 transition(vtos, itos);
763 // load index out of locals
764 locals_index(r2);
765 __ ldr(r1, iaddress(r2));
766
767 __ pop_ptr(r0);
768
769 // r0: array
770 // r1: index
771 index_check(r0, r1); // leaves index in r1, kills rscratch1
772 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
773 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
774 }
775
776 void TemplateTable::saload()
777 {
778 transition(itos, itos);
779 __ mov(r1, r0);
780 __ pop_ptr(r0);
781 // r0: array
782 // r1: index
783 index_check(r0, r1); // leaves index in r1, kills rscratch1
784 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
785 __ load_signed_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
786 }
787
788 void TemplateTable::iload(int n)
789 {
790 transition(vtos, itos);
791 __ ldr(r0, iaddress(n));
792 }
793
794 void TemplateTable::lload(int n)
795 {
796 transition(vtos, ltos);
797 __ ldr(r0, laddress(n));
798 }
799
800 void TemplateTable::fload(int n)
801 {
802 transition(vtos, ftos);
803 __ ldrs(v0, faddress(n));
804 }
805
806 void TemplateTable::dload(int n)
807 {
808 transition(vtos, dtos);
809 __ ldrd(v0, daddress(n));
810 }
811
812 void TemplateTable::aload(int n)
813 {
814 transition(vtos, atos);
815 __ ldr(r0, iaddress(n));
816 }
817
818 void TemplateTable::aload_0() {
819 aload_0_internal();
820 }
821
822 void TemplateTable::nofast_aload_0() {
823 aload_0_internal(may_not_rewrite);
824 }
825
826 void TemplateTable::aload_0_internal(RewriteControl rc) {
827 // According to bytecode histograms, the pairs:
828 //
829 // _aload_0, _fast_igetfield
830 // _aload_0, _fast_agetfield
831 // _aload_0, _fast_fgetfield
832 //
833 // occur frequently. If RewriteFrequentPairs is set, the (slow)
834 // _aload_0 bytecode checks if the next bytecode is either
835 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
836 // rewrites the current bytecode into a pair bytecode; otherwise it
837 // rewrites the current bytecode into _fast_aload_0 that doesn't do
838 // the pair check anymore.
839 //
840 // Note: If the next bytecode is _getfield, the rewrite must be
841 // delayed, otherwise we may miss an opportunity for a pair.
842 //
843 // Also rewrite frequent pairs
844 // aload_0, aload_1
845 // aload_0, iload_1
846 // These bytecodes with a small amount of code are most profitable
847 // to rewrite
848 if (RewriteFrequentPairs && rc == may_rewrite) {
849 Label rewrite, done;
850 const Register bc = r4;
851
852 // get next bytecode
853 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
854
855 // do actual aload_0
856 aload(0);
857
858 // if _getfield then wait with rewrite
859 __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
860 __ br(Assembler::EQ, done);
861
862 // if _igetfield then reqrite to _fast_iaccess_0
863 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
864 __ cmpw(r1, Bytecodes::_fast_igetfield);
865 __ movw(bc, Bytecodes::_fast_iaccess_0);
866 __ br(Assembler::EQ, rewrite);
867
868 // if _agetfield then reqrite to _fast_aaccess_0
869 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
870 __ cmpw(r1, Bytecodes::_fast_agetfield);
871 __ movw(bc, Bytecodes::_fast_aaccess_0);
872 __ br(Assembler::EQ, rewrite);
873
874 // if _fgetfield then reqrite to _fast_faccess_0
875 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
876 __ cmpw(r1, Bytecodes::_fast_fgetfield);
877 __ movw(bc, Bytecodes::_fast_faccess_0);
878 __ br(Assembler::EQ, rewrite);
879
880 // else rewrite to _fast_aload0
881 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
882 __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
883
884 // rewrite
885 // bc: new bytecode
886 __ bind(rewrite);
887 patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
888
889 __ bind(done);
890 } else {
891 aload(0);
892 }
893 }
894
895 void TemplateTable::istore()
896 {
897 transition(itos, vtos);
898 locals_index(r1);
899 // FIXME: We're being very pernickerty here storing a jint in a
900 // local with strw, which costs an extra instruction over what we'd
901 // be able to do with a simple str. We should just store the whole
902 // word.
903 __ lea(rscratch1, iaddress(r1));
904 __ strw(r0, Address(rscratch1));
905 }
906
907 void TemplateTable::lstore()
908 {
909 transition(ltos, vtos);
910 locals_index(r1);
911 __ str(r0, laddress(r1, rscratch1, _masm));
912 }
913
914 void TemplateTable::fstore() {
915 transition(ftos, vtos);
916 locals_index(r1);
917 __ lea(rscratch1, iaddress(r1));
918 __ strs(v0, Address(rscratch1));
919 }
920
921 void TemplateTable::dstore() {
922 transition(dtos, vtos);
923 locals_index(r1);
924 __ strd(v0, daddress(r1, rscratch1, _masm));
925 }
926
927 void TemplateTable::astore()
928 {
929 transition(vtos, vtos);
930 __ pop_ptr(r0);
931 locals_index(r1);
932 __ str(r0, aaddress(r1));
933 }
934
935 void TemplateTable::wide_istore() {
936 transition(vtos, vtos);
937 __ pop_i();
938 locals_index_wide(r1);
939 __ lea(rscratch1, iaddress(r1));
940 __ strw(r0, Address(rscratch1));
941 }
942
943 void TemplateTable::wide_lstore() {
944 transition(vtos, vtos);
945 __ pop_l();
946 locals_index_wide(r1);
947 __ str(r0, laddress(r1, rscratch1, _masm));
948 }
949
950 void TemplateTable::wide_fstore() {
951 transition(vtos, vtos);
952 __ pop_f();
953 locals_index_wide(r1);
954 __ lea(rscratch1, faddress(r1));
955 __ strs(v0, rscratch1);
956 }
957
958 void TemplateTable::wide_dstore() {
959 transition(vtos, vtos);
960 __ pop_d();
961 locals_index_wide(r1);
962 __ strd(v0, daddress(r1, rscratch1, _masm));
963 }
964
965 void TemplateTable::wide_astore() {
966 transition(vtos, vtos);
967 __ pop_ptr(r0);
968 locals_index_wide(r1);
969 __ str(r0, aaddress(r1));
970 }
971
972 void TemplateTable::iastore() {
973 transition(itos, vtos);
974 __ pop_i(r1);
975 __ pop_ptr(r3);
976 // r0: value
977 // r1: index
978 // r3: array
979 index_check(r3, r1); // prefer index in r1
980 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
981 __ strw(r0, Address(rscratch1,
982 arrayOopDesc::base_offset_in_bytes(T_INT)));
983 }
984
985 void TemplateTable::lastore() {
986 transition(ltos, vtos);
987 __ pop_i(r1);
988 __ pop_ptr(r3);
989 // r0: value
990 // r1: index
991 // r3: array
992 index_check(r3, r1); // prefer index in r1
993 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
994 __ str(r0, Address(rscratch1,
995 arrayOopDesc::base_offset_in_bytes(T_LONG)));
996 }
997
998 void TemplateTable::fastore() {
999 transition(ftos, vtos);
1000 __ pop_i(r1);
1001 __ pop_ptr(r3);
1002 // v0: value
1003 // r1: index
1004 // r3: array
1005 index_check(r3, r1); // prefer index in r1
1006 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
1007 __ strs(v0, Address(rscratch1,
1008 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1009 }
1010
1011 void TemplateTable::dastore() {
1012 transition(dtos, vtos);
1013 __ pop_i(r1);
1014 __ pop_ptr(r3);
1015 // v0: value
1016 // r1: index
1017 // r3: array
1018 index_check(r3, r1); // prefer index in r1
1019 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
1020 __ strd(v0, Address(rscratch1,
1021 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1022 }
1023
1024 void TemplateTable::aastore() {
1025 Label is_null, ok_is_subtype, done;
1026 transition(vtos, vtos);
1027 // stack: ..., array, index, value
1028 __ ldr(r0, at_tos()); // value
1029 __ ldr(r2, at_tos_p1()); // index
1030 __ ldr(r3, at_tos_p2()); // array
1031
1032 Address element_address(r4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1033
1034 index_check(r3, r2); // kills r1
1035 __ lea(r4, Address(r3, r2, Address::uxtw(UseCompressedOops? 2 : 3)));
1036
1037 // do array store check - check for NULL value first
1038 __ cbz(r0, is_null);
1039
1040 // Move subklass into r1
1041 __ load_klass(r1, r0);
1042 // Move superklass into r0
1043 __ load_klass(r0, r3);
1044 __ ldr(r0, Address(r0,
1045 ObjArrayKlass::element_klass_offset()));
1046 // Compress array + index*oopSize + 12 into a single register. Frees r2.
1047
1048 // Generate subtype check. Blows r2, r5
1049 // Superklass in r0. Subklass in r1.
1050 __ gen_subtype_check(r1, ok_is_subtype);
1051
1052 // Come here on failure
1053 // object is at TOS
1054 __ b(Interpreter::_throw_ArrayStoreException_entry);
1055
1056 // Come here on success
1057 __ bind(ok_is_subtype);
1058
1059 // Get the value we will store
1060 __ ldr(r0, at_tos());
1061 // Now store using the appropriate barrier
1062 do_oop_store(_masm, element_address, r0, _bs->kind(), true);
1063 __ b(done);
1064
1065 // Have a NULL in r0, r3=array, r2=index. Store NULL at ary[idx]
1066 __ bind(is_null);
1067 __ profile_null_seen(r2);
1068
1069 // Store a NULL
1070 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1071
1072 // Pop stack arguments
1073 __ bind(done);
1074 __ add(esp, esp, 3 * Interpreter::stackElementSize);
1075 }
1076
1077 void TemplateTable::bastore()
1078 {
1079 transition(itos, vtos);
1080 __ pop_i(r1);
1081 __ pop_ptr(r3);
1082 // r0: value
1083 // r1: index
1084 // r3: array
1085 index_check(r3, r1); // prefer index in r1
1086
1087 // Need to check whether array is boolean or byte
1088 // since both types share the bastore bytecode.
1089 __ load_klass(r2, r3);
1090 __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1091 int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit());
1092 Label L_skip;
1093 __ tbz(r2, diffbit_index, L_skip);
1094 __ andw(r0, r0, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1095 __ bind(L_skip);
1096
1097 __ lea(rscratch1, Address(r3, r1, Address::uxtw(0)));
1098 __ strb(r0, Address(rscratch1,
1099 arrayOopDesc::base_offset_in_bytes(T_BYTE)));
1100 }
1101
1102 void TemplateTable::castore()
1103 {
1104 transition(itos, vtos);
1105 __ pop_i(r1);
1106 __ pop_ptr(r3);
1107 // r0: value
1108 // r1: index
1109 // r3: array
1110 index_check(r3, r1); // prefer index in r1
1111 __ lea(rscratch1, Address(r3, r1, Address::uxtw(1)));
1112 __ strh(r0, Address(rscratch1,
1113 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1114 }
1115
1116 void TemplateTable::sastore()
1117 {
1118 castore();
1119 }
1120
1121 void TemplateTable::istore(int n)
1122 {
1123 transition(itos, vtos);
1124 __ str(r0, iaddress(n));
1125 }
1126
1127 void TemplateTable::lstore(int n)
1128 {
1129 transition(ltos, vtos);
1130 __ str(r0, laddress(n));
1131 }
1132
1133 void TemplateTable::fstore(int n)
1134 {
1135 transition(ftos, vtos);
1136 __ strs(v0, faddress(n));
1137 }
1138
1139 void TemplateTable::dstore(int n)
1140 {
1141 transition(dtos, vtos);
1142 __ strd(v0, daddress(n));
1143 }
1144
1145 void TemplateTable::astore(int n)
1146 {
1147 transition(vtos, vtos);
1148 __ pop_ptr(r0);
1149 __ str(r0, iaddress(n));
1150 }
1151
1152 void TemplateTable::pop()
1153 {
1154 transition(vtos, vtos);
1155 __ add(esp, esp, Interpreter::stackElementSize);
1156 }
1157
1158 void TemplateTable::pop2()
1159 {
1160 transition(vtos, vtos);
1161 __ add(esp, esp, 2 * Interpreter::stackElementSize);
1162 }
1163
1164 void TemplateTable::dup()
1165 {
1166 transition(vtos, vtos);
1167 __ ldr(r0, Address(esp, 0));
1168 __ push(r0);
1169 // stack: ..., a, a
1170 }
1171
1172 void TemplateTable::dup_x1()
1173 {
1174 transition(vtos, vtos);
1175 // stack: ..., a, b
1176 __ ldr(r0, at_tos()); // load b
1177 __ ldr(r2, at_tos_p1()); // load a
1178 __ str(r0, at_tos_p1()); // store b
1179 __ str(r2, at_tos()); // store a
1180 __ push(r0); // push b
1181 // stack: ..., b, a, b
1182 }
1183
1184 void TemplateTable::dup_x2()
1185 {
1186 transition(vtos, vtos);
1187 // stack: ..., a, b, c
1188 __ ldr(r0, at_tos()); // load c
1189 __ ldr(r2, at_tos_p2()); // load a
1190 __ str(r0, at_tos_p2()); // store c in a
1191 __ push(r0); // push c
1192 // stack: ..., c, b, c, c
1193 __ ldr(r0, at_tos_p2()); // load b
1194 __ str(r2, at_tos_p2()); // store a in b
1195 // stack: ..., c, a, c, c
1196 __ str(r0, at_tos_p1()); // store b in c
1197 // stack: ..., c, a, b, c
1198 }
1199
1200 void TemplateTable::dup2()
1201 {
1202 transition(vtos, vtos);
1203 // stack: ..., a, b
1204 __ ldr(r0, at_tos_p1()); // load a
1205 __ push(r0); // push a
1206 __ ldr(r0, at_tos_p1()); // load b
1207 __ push(r0); // push b
1208 // stack: ..., a, b, a, b
1209 }
1210
1211 void TemplateTable::dup2_x1()
1212 {
1213 transition(vtos, vtos);
1214 // stack: ..., a, b, c
1215 __ ldr(r2, at_tos()); // load c
1216 __ ldr(r0, at_tos_p1()); // load b
1217 __ push(r0); // push b
1218 __ push(r2); // push c
1219 // stack: ..., a, b, c, b, c
1220 __ str(r2, at_tos_p3()); // store c in b
1221 // stack: ..., a, c, c, b, c
1222 __ ldr(r2, at_tos_p4()); // load a
1223 __ str(r2, at_tos_p2()); // store a in 2nd c
1224 // stack: ..., a, c, a, b, c
1225 __ str(r0, at_tos_p4()); // store b in a
1226 // stack: ..., b, c, a, b, c
1227 }
1228
1229 void TemplateTable::dup2_x2()
1230 {
1231 transition(vtos, vtos);
1232 // stack: ..., a, b, c, d
1233 __ ldr(r2, at_tos()); // load d
1234 __ ldr(r0, at_tos_p1()); // load c
1235 __ push(r0) ; // push c
1236 __ push(r2); // push d
1237 // stack: ..., a, b, c, d, c, d
1238 __ ldr(r0, at_tos_p4()); // load b
1239 __ str(r0, at_tos_p2()); // store b in d
1240 __ str(r2, at_tos_p4()); // store d in b
1241 // stack: ..., a, d, c, b, c, d
1242 __ ldr(r2, at_tos_p5()); // load a
1243 __ ldr(r0, at_tos_p3()); // load c
1244 __ str(r2, at_tos_p3()); // store a in c
1245 __ str(r0, at_tos_p5()); // store c in a
1246 // stack: ..., c, d, a, b, c, d
1247 }
1248
1249 void TemplateTable::swap()
1250 {
1251 transition(vtos, vtos);
1252 // stack: ..., a, b
1253 __ ldr(r2, at_tos_p1()); // load a
1254 __ ldr(r0, at_tos()); // load b
1255 __ str(r2, at_tos()); // store a in b
1256 __ str(r0, at_tos_p1()); // store b in a
1257 // stack: ..., b, a
1258 }
1259
1260 void TemplateTable::iop2(Operation op)
1261 {
1262 transition(itos, itos);
1263 // r0 <== r1 op r0
1264 __ pop_i(r1);
1265 switch (op) {
1266 case add : __ addw(r0, r1, r0); break;
1267 case sub : __ subw(r0, r1, r0); break;
1268 case mul : __ mulw(r0, r1, r0); break;
1269 case _and : __ andw(r0, r1, r0); break;
1270 case _or : __ orrw(r0, r1, r0); break;
1271 case _xor : __ eorw(r0, r1, r0); break;
1272 case shl : __ lslvw(r0, r1, r0); break;
1273 case shr : __ asrvw(r0, r1, r0); break;
1274 case ushr : __ lsrvw(r0, r1, r0);break;
1275 default : ShouldNotReachHere();
1276 }
1277 }
1278
1279 void TemplateTable::lop2(Operation op)
1280 {
1281 transition(ltos, ltos);
1282 // r0 <== r1 op r0
1283 __ pop_l(r1);
1284 switch (op) {
1285 case add : __ add(r0, r1, r0); break;
1286 case sub : __ sub(r0, r1, r0); break;
1287 case mul : __ mul(r0, r1, r0); break;
1288 case _and : __ andr(r0, r1, r0); break;
1289 case _or : __ orr(r0, r1, r0); break;
1290 case _xor : __ eor(r0, r1, r0); break;
1291 default : ShouldNotReachHere();
1292 }
1293 }
1294
1295 void TemplateTable::idiv()
1296 {
1297 transition(itos, itos);
1298 // explicitly check for div0
1299 Label no_div0;
1300 __ cbnzw(r0, no_div0);
1301 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1302 __ br(rscratch1);
1303 __ bind(no_div0);
1304 __ pop_i(r1);
1305 // r0 <== r1 idiv r0
1306 __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1307 }
1308
1309 void TemplateTable::irem()
1310 {
1311 transition(itos, itos);
1312 // explicitly check for div0
1313 Label no_div0;
1314 __ cbnzw(r0, no_div0);
1315 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1316 __ br(rscratch1);
1317 __ bind(no_div0);
1318 __ pop_i(r1);
1319 // r0 <== r1 irem r0
1320 __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1321 }
1322
1323 void TemplateTable::lmul()
1324 {
1325 transition(ltos, ltos);
1326 __ pop_l(r1);
1327 __ mul(r0, r0, r1);
1328 }
1329
1330 void TemplateTable::ldiv()
1331 {
1332 transition(ltos, ltos);
1333 // explicitly check for div0
1334 Label no_div0;
1335 __ cbnz(r0, no_div0);
1336 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1337 __ br(rscratch1);
1338 __ bind(no_div0);
1339 __ pop_l(r1);
1340 // r0 <== r1 ldiv r0
1341 __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1342 }
1343
1344 void TemplateTable::lrem()
1345 {
1346 transition(ltos, ltos);
1347 // explicitly check for div0
1348 Label no_div0;
1349 __ cbnz(r0, no_div0);
1350 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1351 __ br(rscratch1);
1352 __ bind(no_div0);
1353 __ pop_l(r1);
1354 // r0 <== r1 lrem r0
1355 __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1356 }
1357
1358 void TemplateTable::lshl()
1359 {
1360 transition(itos, ltos);
1361 // shift count is in r0
1362 __ pop_l(r1);
1363 __ lslv(r0, r1, r0);
1364 }
1365
1366 void TemplateTable::lshr()
1367 {
1368 transition(itos, ltos);
1369 // shift count is in r0
1370 __ pop_l(r1);
1371 __ asrv(r0, r1, r0);
1372 }
1373
1374 void TemplateTable::lushr()
1375 {
1376 transition(itos, ltos);
1377 // shift count is in r0
1378 __ pop_l(r1);
1379 __ lsrv(r0, r1, r0);
1380 }
1381
1382 void TemplateTable::fop2(Operation op)
1383 {
1384 transition(ftos, ftos);
1385 switch (op) {
1386 case add:
1387 // n.b. use ldrd because this is a 64 bit slot
1388 __ pop_f(v1);
1389 __ fadds(v0, v1, v0);
1390 break;
1391 case sub:
1392 __ pop_f(v1);
1393 __ fsubs(v0, v1, v0);
1394 break;
1395 case mul:
1396 __ pop_f(v1);
1397 __ fmuls(v0, v1, v0);
1398 break;
1399 case div:
1400 __ pop_f(v1);
1401 __ fdivs(v0, v1, v0);
1402 break;
1403 case rem:
1404 __ fmovs(v1, v0);
1405 __ pop_f(v0);
1406 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::frem),
1407 0, 2, MacroAssembler::ret_type_float);
1408 break;
1409 default:
1410 ShouldNotReachHere();
1411 break;
1412 }
1413 }
1414
1415 void TemplateTable::dop2(Operation op)
1416 {
1417 transition(dtos, dtos);
1418 switch (op) {
1419 case add:
1420 // n.b. use ldrd because this is a 64 bit slot
1421 __ pop_d(v1);
1422 __ faddd(v0, v1, v0);
1423 break;
1424 case sub:
1425 __ pop_d(v1);
1426 __ fsubd(v0, v1, v0);
1427 break;
1428 case mul:
1429 __ pop_d(v1);
1430 __ fmuld(v0, v1, v0);
1431 break;
1432 case div:
1433 __ pop_d(v1);
1434 __ fdivd(v0, v1, v0);
1435 break;
1436 case rem:
1437 __ fmovd(v1, v0);
1438 __ pop_d(v0);
1439 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::drem),
1440 0, 2, MacroAssembler::ret_type_double);
1441 break;
1442 default:
1443 ShouldNotReachHere();
1444 break;
1445 }
1446 }
1447
1448 void TemplateTable::ineg()
1449 {
1450 transition(itos, itos);
1451 __ negw(r0, r0);
1452
1453 }
1454
1455 void TemplateTable::lneg()
1456 {
1457 transition(ltos, ltos);
1458 __ neg(r0, r0);
1459 }
1460
1461 void TemplateTable::fneg()
1462 {
1463 transition(ftos, ftos);
1464 __ fnegs(v0, v0);
1465 }
1466
1467 void TemplateTable::dneg()
1468 {
1469 transition(dtos, dtos);
1470 __ fnegd(v0, v0);
1471 }
1472
1473 void TemplateTable::iinc()
1474 {
1475 transition(vtos, vtos);
1476 __ load_signed_byte(r1, at_bcp(2)); // get constant
1477 locals_index(r2);
1478 __ ldr(r0, iaddress(r2));
1479 __ addw(r0, r0, r1);
1480 __ str(r0, iaddress(r2));
1481 }
1482
1483 void TemplateTable::wide_iinc()
1484 {
1485 transition(vtos, vtos);
1486 // __ mov(r1, zr);
1487 __ ldrw(r1, at_bcp(2)); // get constant and index
1488 __ rev16(r1, r1);
1489 __ ubfx(r2, r1, 0, 16);
1490 __ neg(r2, r2);
1491 __ sbfx(r1, r1, 16, 16);
1492 __ ldr(r0, iaddress(r2));
1493 __ addw(r0, r0, r1);
1494 __ str(r0, iaddress(r2));
1495 }
1496
1497 void TemplateTable::convert()
1498 {
1499 // Checking
1500 #ifdef ASSERT
1501 {
1502 TosState tos_in = ilgl;
1503 TosState tos_out = ilgl;
1504 switch (bytecode()) {
1505 case Bytecodes::_i2l: // fall through
1506 case Bytecodes::_i2f: // fall through
1507 case Bytecodes::_i2d: // fall through
1508 case Bytecodes::_i2b: // fall through
1509 case Bytecodes::_i2c: // fall through
1510 case Bytecodes::_i2s: tos_in = itos; break;
1511 case Bytecodes::_l2i: // fall through
1512 case Bytecodes::_l2f: // fall through
1513 case Bytecodes::_l2d: tos_in = ltos; break;
1514 case Bytecodes::_f2i: // fall through
1515 case Bytecodes::_f2l: // fall through
1516 case Bytecodes::_f2d: tos_in = ftos; break;
1517 case Bytecodes::_d2i: // fall through
1518 case Bytecodes::_d2l: // fall through
1519 case Bytecodes::_d2f: tos_in = dtos; break;
1520 default : ShouldNotReachHere();
1521 }
1522 switch (bytecode()) {
1523 case Bytecodes::_l2i: // fall through
1524 case Bytecodes::_f2i: // fall through
1525 case Bytecodes::_d2i: // fall through
1526 case Bytecodes::_i2b: // fall through
1527 case Bytecodes::_i2c: // fall through
1528 case Bytecodes::_i2s: tos_out = itos; break;
1529 case Bytecodes::_i2l: // fall through
1530 case Bytecodes::_f2l: // fall through
1531 case Bytecodes::_d2l: tos_out = ltos; break;
1532 case Bytecodes::_i2f: // fall through
1533 case Bytecodes::_l2f: // fall through
1534 case Bytecodes::_d2f: tos_out = ftos; break;
1535 case Bytecodes::_i2d: // fall through
1536 case Bytecodes::_l2d: // fall through
1537 case Bytecodes::_f2d: tos_out = dtos; break;
1538 default : ShouldNotReachHere();
1539 }
1540 transition(tos_in, tos_out);
1541 }
1542 #endif // ASSERT
1543 // static const int64_t is_nan = 0x8000000000000000L;
1544
1545 // Conversion
1546 switch (bytecode()) {
1547 case Bytecodes::_i2l:
1548 __ sxtw(r0, r0);
1549 break;
1550 case Bytecodes::_i2f:
1551 __ scvtfws(v0, r0);
1552 break;
1553 case Bytecodes::_i2d:
1554 __ scvtfwd(v0, r0);
1555 break;
1556 case Bytecodes::_i2b:
1557 __ sxtbw(r0, r0);
1558 break;
1559 case Bytecodes::_i2c:
1560 __ uxthw(r0, r0);
1561 break;
1562 case Bytecodes::_i2s:
1563 __ sxthw(r0, r0);
1564 break;
1565 case Bytecodes::_l2i:
1566 __ uxtw(r0, r0);
1567 break;
1568 case Bytecodes::_l2f:
1569 __ scvtfs(v0, r0);
1570 break;
1571 case Bytecodes::_l2d:
1572 __ scvtfd(v0, r0);
1573 break;
1574 case Bytecodes::_f2i:
1575 {
1576 Label L_Okay;
1577 __ clear_fpsr();
1578 __ fcvtzsw(r0, v0);
1579 __ get_fpsr(r1);
1580 __ cbzw(r1, L_Okay);
1581 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2i),
1582 0, 1, MacroAssembler::ret_type_integral);
1583 __ bind(L_Okay);
1584 }
1585 break;
1586 case Bytecodes::_f2l:
1587 {
1588 Label L_Okay;
1589 __ clear_fpsr();
1590 __ fcvtzs(r0, v0);
1591 __ get_fpsr(r1);
1592 __ cbzw(r1, L_Okay);
1593 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2l),
1594 0, 1, MacroAssembler::ret_type_integral);
1595 __ bind(L_Okay);
1596 }
1597 break;
1598 case Bytecodes::_f2d:
1599 __ fcvts(v0, v0);
1600 break;
1601 case Bytecodes::_d2i:
1602 {
1603 Label L_Okay;
1604 __ clear_fpsr();
1605 __ fcvtzdw(r0, v0);
1606 __ get_fpsr(r1);
1607 __ cbzw(r1, L_Okay);
1608 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2i),
1609 0, 1, MacroAssembler::ret_type_integral);
1610 __ bind(L_Okay);
1611 }
1612 break;
1613 case Bytecodes::_d2l:
1614 {
1615 Label L_Okay;
1616 __ clear_fpsr();
1617 __ fcvtzd(r0, v0);
1618 __ get_fpsr(r1);
1619 __ cbzw(r1, L_Okay);
1620 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2l),
1621 0, 1, MacroAssembler::ret_type_integral);
1622 __ bind(L_Okay);
1623 }
1624 break;
1625 case Bytecodes::_d2f:
1626 __ fcvtd(v0, v0);
1627 break;
1628 default:
1629 ShouldNotReachHere();
1630 }
1631 }
1632
1633 void TemplateTable::lcmp()
1634 {
1635 transition(ltos, itos);
1636 Label done;
1637 __ pop_l(r1);
1638 __ cmp(r1, r0);
1639 __ mov(r0, (u_int64_t)-1L);
1640 __ br(Assembler::LT, done);
1641 // __ mov(r0, 1UL);
1642 // __ csel(r0, r0, zr, Assembler::NE);
1643 // and here is a faster way
1644 __ csinc(r0, zr, zr, Assembler::EQ);
1645 __ bind(done);
1646 }
1647
1648 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1649 {
1650 Label done;
1651 if (is_float) {
1652 // XXX get rid of pop here, use ... reg, mem32
1653 __ pop_f(v1);
1654 __ fcmps(v1, v0);
1655 } else {
1656 // XXX get rid of pop here, use ... reg, mem64
1657 __ pop_d(v1);
1658 __ fcmpd(v1, v0);
1659 }
1660 if (unordered_result < 0) {
1661 // we want -1 for unordered or less than, 0 for equal and 1 for
1662 // greater than.
1663 __ mov(r0, (u_int64_t)-1L);
1664 // for FP LT tests less than or unordered
1665 __ br(Assembler::LT, done);
1666 // install 0 for EQ otherwise 1
1667 __ csinc(r0, zr, zr, Assembler::EQ);
1668 } else {
1669 // we want -1 for less than, 0 for equal and 1 for unordered or
1670 // greater than.
1671 __ mov(r0, 1L);
1672 // for FP HI tests greater than or unordered
1673 __ br(Assembler::HI, done);
1674 // install 0 for EQ otherwise ~0
1675 __ csinv(r0, zr, zr, Assembler::EQ);
1676
1677 }
1678 __ bind(done);
1679 }
1680
1681 void TemplateTable::branch(bool is_jsr, bool is_wide)
1682 {
1683 // We might be moving to a safepoint. The thread which calls
1684 // Interpreter::notice_safepoints() will effectively flush its cache
1685 // when it makes a system call, but we need to do something to
1686 // ensure that we see the changed dispatch table.
1687 __ membar(MacroAssembler::LoadLoad);
1688
1689 __ profile_taken_branch(r0, r1);
1690 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1691 InvocationCounter::counter_offset();
1692 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1693 InvocationCounter::counter_offset();
1694
1695 // load branch displacement
1696 if (!is_wide) {
1697 __ ldrh(r2, at_bcp(1));
1698 __ rev16(r2, r2);
1699 // sign extend the 16 bit value in r2
1700 __ sbfm(r2, r2, 0, 15);
1701 } else {
1702 __ ldrw(r2, at_bcp(1));
1703 __ revw(r2, r2);
1704 // sign extend the 32 bit value in r2
1705 __ sbfm(r2, r2, 0, 31);
1706 }
1707
1708 // Handle all the JSR stuff here, then exit.
1709 // It's much shorter and cleaner than intermingling with the non-JSR
1710 // normal-branch stuff occurring below.
1711
1712 if (is_jsr) {
1713 // Pre-load the next target bytecode into rscratch1
1714 __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1715 // compute return address as bci
1716 __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1717 __ add(rscratch2, rscratch2,
1718 in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1719 __ sub(r1, rbcp, rscratch2);
1720 __ push_i(r1);
1721 // Adjust the bcp by the 16-bit displacement in r2
1722 __ add(rbcp, rbcp, r2);
1723 __ dispatch_only(vtos);
1724 return;
1725 }
1726
1727 // Normal (non-jsr) branch handling
1728
1729 // Adjust the bcp by the displacement in r2
1730 __ add(rbcp, rbcp, r2);
1731
1732 assert(UseLoopCounter || !UseOnStackReplacement,
1733 "on-stack-replacement requires loop counters");
1734 Label backedge_counter_overflow;
1735 Label profile_method;
1736 Label dispatch;
1737 if (UseLoopCounter) {
1738 // increment backedge counter for backward branches
1739 // r0: MDO
1740 // w1: MDO bumped taken-count
1741 // r2: target offset
1742 __ cmp(r2, zr);
1743 __ br(Assembler::GT, dispatch); // count only if backward branch
1744
1745 // ECN: FIXME: This code smells
1746 // check if MethodCounters exists
1747 Label has_counters;
1748 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1749 __ cbnz(rscratch1, has_counters);
1750 __ push(r0);
1751 __ push(r1);
1752 __ push(r2);
1753 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1754 InterpreterRuntime::build_method_counters), rmethod);
1755 __ pop(r2);
1756 __ pop(r1);
1757 __ pop(r0);
1758 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1759 __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1760 __ bind(has_counters);
1761
1762 if (TieredCompilation) {
1763 Label no_mdo;
1764 int increment = InvocationCounter::count_increment;
1765 if (ProfileInterpreter) {
1766 // Are we profiling?
1767 __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1768 __ cbz(r1, no_mdo);
1769 // Increment the MDO backedge counter
1770 const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1771 in_bytes(InvocationCounter::counter_offset()));
1772 const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1773 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1774 r0, rscratch1, false, Assembler::EQ, &backedge_counter_overflow);
1775 __ b(dispatch);
1776 }
1777 __ bind(no_mdo);
1778 // Increment backedge counter in MethodCounters*
1779 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1780 const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1781 __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1782 r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow);
1783 } else { // not TieredCompilation
1784 // increment counter
1785 __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1786 __ ldrw(r0, Address(rscratch2, be_offset)); // load backedge counter
1787 __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1788 __ strw(rscratch1, Address(rscratch2, be_offset)); // store counter
1789
1790 __ ldrw(r0, Address(rscratch2, inv_offset)); // load invocation counter
1791 __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1792 __ addw(r0, r0, rscratch1); // add both counters
1793
1794 if (ProfileInterpreter) {
1795 // Test to see if we should create a method data oop
1796 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
1797 __ cmpw(r0, rscratch1);
1798 __ br(Assembler::LT, dispatch);
1799
1800 // if no method data exists, go to profile method
1801 __ test_method_data_pointer(r0, profile_method);
1802
1803 if (UseOnStackReplacement) {
1804 // check for overflow against w1 which is the MDO taken count
1805 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1806 __ cmpw(r1, rscratch1);
1807 __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1808
1809 // When ProfileInterpreter is on, the backedge_count comes
1810 // from the MethodData*, which value does not get reset on
1811 // the call to frequency_counter_overflow(). To avoid
1812 // excessive calls to the overflow routine while the method is
1813 // being compiled, add a second test to make sure the overflow
1814 // function is called only once every overflow_frequency.
1815 const int overflow_frequency = 1024;
1816 __ andsw(r1, r1, overflow_frequency - 1);
1817 __ br(Assembler::EQ, backedge_counter_overflow);
1818
1819 }
1820 } else {
1821 if (UseOnStackReplacement) {
1822 // check for overflow against w0, which is the sum of the
1823 // counters
1824 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1825 __ cmpw(r0, rscratch1);
1826 __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1827 }
1828 }
1829 }
1830 }
1831 __ bind(dispatch);
1832
1833 // Pre-load the next target bytecode into rscratch1
1834 __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1835
1836 // continue with the bytecode @ target
1837 // rscratch1: target bytecode
1838 // rbcp: target bcp
1839 __ dispatch_only(vtos);
1840
1841 if (UseLoopCounter) {
1842 if (ProfileInterpreter) {
1843 // Out-of-line code to allocate method data oop.
1844 __ bind(profile_method);
1845 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1846 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
1847 __ set_method_data_pointer_for_bcp();
1848 __ b(dispatch);
1849 }
1850
1851 if (TieredCompilation || UseOnStackReplacement) {
1852 // invocation counter overflow
1853 __ bind(backedge_counter_overflow);
1854 __ neg(r2, r2);
1855 __ add(r2, r2, rbcp); // branch bcp
1856 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1857 __ call_VM(noreg,
1858 CAST_FROM_FN_PTR(address,
1859 InterpreterRuntime::frequency_counter_overflow),
1860 r2);
1861 if (!UseOnStackReplacement)
1862 __ b(dispatch);
1863 }
1864
1865 if (UseOnStackReplacement) {
1866 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
1867
1868 // r0: osr nmethod (osr ok) or NULL (osr not possible)
1869 // w1: target bytecode
1870 // r2: scratch
1871 __ cbz(r0, dispatch); // test result -- no osr if null
1872 // nmethod may have been invalidated (VM may block upon call_VM return)
1873 __ ldrb(r2, Address(r0, nmethod::state_offset()));
1874 if (nmethod::in_use != 0)
1875 __ sub(r2, r2, nmethod::in_use);
1876 __ cbnz(r2, dispatch);
1877
1878 // We have the address of an on stack replacement routine in r0
1879 // We need to prepare to execute the OSR method. First we must
1880 // migrate the locals and monitors off of the stack.
1881
1882 __ mov(r19, r0); // save the nmethod
1883
1884 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1885
1886 // r0 is OSR buffer, move it to expected parameter location
1887 __ mov(j_rarg0, r0);
1888
1889 // remove activation
1890 // get sender esp
1891 __ ldr(esp,
1892 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1893 // remove frame anchor
1894 __ leave();
1895 // Ensure compiled code always sees stack at proper alignment
1896 __ andr(sp, esp, -16);
1897
1898 // and begin the OSR nmethod
1899 __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1900 __ br(rscratch1);
1901 }
1902 }
1903 }
1904
1905
1906 void TemplateTable::if_0cmp(Condition cc)
1907 {
1908 transition(itos, vtos);
1909 // assume branch is more often taken than not (loops use backward branches)
1910 Label not_taken;
1911 if (cc == equal)
1912 __ cbnzw(r0, not_taken);
1913 else if (cc == not_equal)
1914 __ cbzw(r0, not_taken);
1915 else {
1916 __ andsw(zr, r0, r0);
1917 __ br(j_not(cc), not_taken);
1918 }
1919
1920 branch(false, false);
1921 __ bind(not_taken);
1922 __ profile_not_taken_branch(r0);
1923 }
1924
1925 void TemplateTable::if_icmp(Condition cc)
1926 {
1927 transition(itos, vtos);
1928 // assume branch is more often taken than not (loops use backward branches)
1929 Label not_taken;
1930 __ pop_i(r1);
1931 __ cmpw(r1, r0, Assembler::LSL);
1932 __ br(j_not(cc), not_taken);
1933 branch(false, false);
1934 __ bind(not_taken);
1935 __ profile_not_taken_branch(r0);
1936 }
1937
1938 void TemplateTable::if_nullcmp(Condition cc)
1939 {
1940 transition(atos, vtos);
1941 // assume branch is more often taken than not (loops use backward branches)
1942 Label not_taken;
1943 if (cc == equal)
1944 __ cbnz(r0, not_taken);
1945 else
1946 __ cbz(r0, not_taken);
1947 branch(false, false);
1948 __ bind(not_taken);
1949 __ profile_not_taken_branch(r0);
1950 }
1951
1952 void TemplateTable::if_acmp(Condition cc)
1953 {
1954 transition(atos, vtos);
1955 // assume branch is more often taken than not (loops use backward branches)
1956 Label not_taken;
1957 __ pop_ptr(r1);
1958 __ cmp(r1, r0);
1959 __ br(j_not(cc), not_taken);
1960 branch(false, false);
1961 __ bind(not_taken);
1962 __ profile_not_taken_branch(r0);
1963 }
1964
1965 void TemplateTable::ret() {
1966 transition(vtos, vtos);
1967 // We might be moving to a safepoint. The thread which calls
1968 // Interpreter::notice_safepoints() will effectively flush its cache
1969 // when it makes a system call, but we need to do something to
1970 // ensure that we see the changed dispatch table.
1971 __ membar(MacroAssembler::LoadLoad);
1972
1973 locals_index(r1);
1974 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1975 __ profile_ret(r1, r2);
1976 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1977 __ lea(rbcp, Address(rbcp, r1));
1978 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1979 __ dispatch_next(vtos);
1980 }
1981
1982 void TemplateTable::wide_ret() {
1983 transition(vtos, vtos);
1984 locals_index_wide(r1);
1985 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1986 __ profile_ret(r1, r2);
1987 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1988 __ lea(rbcp, Address(rbcp, r1));
1989 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1990 __ dispatch_next(vtos);
1991 }
1992
1993
1994 void TemplateTable::tableswitch() {
1995 Label default_case, continue_execution;
1996 transition(itos, vtos);
1997 // align rbcp
1998 __ lea(r1, at_bcp(BytesPerInt));
1999 __ andr(r1, r1, -BytesPerInt);
2000 // load lo & hi
2001 __ ldrw(r2, Address(r1, BytesPerInt));
2002 __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2003 __ rev32(r2, r2);
2004 __ rev32(r3, r3);
2005 // check against lo & hi
2006 __ cmpw(r0, r2);
2007 __ br(Assembler::LT, default_case);
2008 __ cmpw(r0, r3);
2009 __ br(Assembler::GT, default_case);
2010 // lookup dispatch offset
2011 __ subw(r0, r0, r2);
2012 __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2013 __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2014 __ profile_switch_case(r0, r1, r2);
2015 // continue execution
2016 __ bind(continue_execution);
2017 __ rev32(r3, r3);
2018 __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2019 __ add(rbcp, rbcp, r3, ext::sxtw);
2020 __ dispatch_only(vtos);
2021 // handle default
2022 __ bind(default_case);
2023 __ profile_switch_default(r0);
2024 __ ldrw(r3, Address(r1, 0));
2025 __ b(continue_execution);
2026 }
2027
2028 void TemplateTable::lookupswitch() {
2029 transition(itos, itos);
2030 __ stop("lookupswitch bytecode should have been rewritten");
2031 }
2032
2033 void TemplateTable::fast_linearswitch() {
2034 transition(itos, vtos);
2035 Label loop_entry, loop, found, continue_execution;
2036 // bswap r0 so we can avoid bswapping the table entries
2037 __ rev32(r0, r0);
2038 // align rbcp
2039 __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2040 // this instruction (change offsets
2041 // below)
2042 __ andr(r19, r19, -BytesPerInt);
2043 // set counter
2044 __ ldrw(r1, Address(r19, BytesPerInt));
2045 __ rev32(r1, r1);
2046 __ b(loop_entry);
2047 // table search
2048 __ bind(loop);
2049 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2050 __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2051 __ cmpw(r0, rscratch1);
2052 __ br(Assembler::EQ, found);
2053 __ bind(loop_entry);
2054 __ subs(r1, r1, 1);
2055 __ br(Assembler::PL, loop);
2056 // default case
2057 __ profile_switch_default(r0);
2058 __ ldrw(r3, Address(r19, 0));
2059 __ b(continue_execution);
2060 // entry found -> get offset
2061 __ bind(found);
2062 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2063 __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2064 __ profile_switch_case(r1, r0, r19);
2065 // continue execution
2066 __ bind(continue_execution);
2067 __ rev32(r3, r3);
2068 __ add(rbcp, rbcp, r3, ext::sxtw);
2069 __ ldrb(rscratch1, Address(rbcp, 0));
2070 __ dispatch_only(vtos);
2071 }
2072
2073 void TemplateTable::fast_binaryswitch() {
2074 transition(itos, vtos);
2075 // Implementation using the following core algorithm:
2076 //
2077 // int binary_search(int key, LookupswitchPair* array, int n) {
2078 // // Binary search according to "Methodik des Programmierens" by
2079 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2080 // int i = 0;
2081 // int j = n;
2082 // while (i+1 < j) {
2083 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2084 // // with Q: for all i: 0 <= i < n: key < a[i]
2085 // // where a stands for the array and assuming that the (inexisting)
2086 // // element a[n] is infinitely big.
2087 // int h = (i + j) >> 1;
2088 // // i < h < j
2089 // if (key < array[h].fast_match()) {
2090 // j = h;
2091 // } else {
2092 // i = h;
2093 // }
2094 // }
2095 // // R: a[i] <= key < a[i+1] or Q
2096 // // (i.e., if key is within array, i is the correct index)
2097 // return i;
2098 // }
2099
2100 // Register allocation
2101 const Register key = r0; // already set (tosca)
2102 const Register array = r1;
2103 const Register i = r2;
2104 const Register j = r3;
2105 const Register h = rscratch1;
2106 const Register temp = rscratch2;
2107
2108 // Find array start
2109 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2110 // get rid of this
2111 // instruction (change
2112 // offsets below)
2113 __ andr(array, array, -BytesPerInt);
2114
2115 // Initialize i & j
2116 __ mov(i, 0); // i = 0;
2117 __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2118
2119 // Convert j into native byteordering
2120 __ rev32(j, j);
2121
2122 // And start
2123 Label entry;
2124 __ b(entry);
2125
2126 // binary search loop
2127 {
2128 Label loop;
2129 __ bind(loop);
2130 // int h = (i + j) >> 1;
2131 __ addw(h, i, j); // h = i + j;
2132 __ lsrw(h, h, 1); // h = (i + j) >> 1;
2133 // if (key < array[h].fast_match()) {
2134 // j = h;
2135 // } else {
2136 // i = h;
2137 // }
2138 // Convert array[h].match to native byte-ordering before compare
2139 __ ldr(temp, Address(array, h, Address::lsl(3)));
2140 __ rev32(temp, temp);
2141 __ cmpw(key, temp);
2142 // j = h if (key < array[h].fast_match())
2143 __ csel(j, h, j, Assembler::LT);
2144 // i = h if (key >= array[h].fast_match())
2145 __ csel(i, h, i, Assembler::GE);
2146 // while (i+1 < j)
2147 __ bind(entry);
2148 __ addw(h, i, 1); // i+1
2149 __ cmpw(h, j); // i+1 < j
2150 __ br(Assembler::LT, loop);
2151 }
2152
2153 // end of binary search, result index is i (must check again!)
2154 Label default_case;
2155 // Convert array[i].match to native byte-ordering before compare
2156 __ ldr(temp, Address(array, i, Address::lsl(3)));
2157 __ rev32(temp, temp);
2158 __ cmpw(key, temp);
2159 __ br(Assembler::NE, default_case);
2160
2161 // entry found -> j = offset
2162 __ add(j, array, i, ext::uxtx, 3);
2163 __ ldrw(j, Address(j, BytesPerInt));
2164 __ profile_switch_case(i, key, array);
2165 __ rev32(j, j);
2166 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2167 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2168 __ dispatch_only(vtos);
2169
2170 // default case -> j = default offset
2171 __ bind(default_case);
2172 __ profile_switch_default(i);
2173 __ ldrw(j, Address(array, -2 * BytesPerInt));
2174 __ rev32(j, j);
2175 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2176 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2177 __ dispatch_only(vtos);
2178 }
2179
2180
2181 void TemplateTable::_return(TosState state)
2182 {
2183 transition(state, state);
2184 assert(_desc->calls_vm(),
2185 "inconsistent calls_vm information"); // call in remove_activation
2186
2187 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2188 assert(state == vtos, "only valid state");
2189
2190 __ ldr(c_rarg1, aaddress(0));
2191 __ load_klass(r3, c_rarg1);
2192 __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2193 __ tst(r3, JVM_ACC_HAS_FINALIZER);
2194 Label skip_register_finalizer;
2195 __ br(Assembler::EQ, skip_register_finalizer);
2196
2197 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2198
2199 __ bind(skip_register_finalizer);
2200 }
2201
2202 // Issue a StoreStore barrier after all stores but before return
2203 // from any constructor for any class with a final field. We don't
2204 // know if this is a finalizer, so we always do so.
2205 if (_desc->bytecode() == Bytecodes::_return)
2206 __ membar(MacroAssembler::StoreStore);
2207
2208 // Narrow result if state is itos but result type is smaller.
2209 // Need to narrow in the return bytecode rather than in generate_return_entry
2210 // since compiled code callers expect the result to already be narrowed.
2211 if (state == itos) {
2212 __ narrow(r0);
2213 }
2214
2215 __ remove_activation(state);
2216 __ ret(lr);
2217 }
2218
2219 // ----------------------------------------------------------------------------
2220 // Volatile variables demand their effects be made known to all CPU's
2221 // in order. Store buffers on most chips allow reads & writes to
2222 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2223 // without some kind of memory barrier (i.e., it's not sufficient that
2224 // the interpreter does not reorder volatile references, the hardware
2225 // also must not reorder them).
2226 //
2227 // According to the new Java Memory Model (JMM):
2228 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2229 // writes act as aquire & release, so:
2230 // (2) A read cannot let unrelated NON-volatile memory refs that
2231 // happen after the read float up to before the read. It's OK for
2232 // non-volatile memory refs that happen before the volatile read to
2233 // float down below it.
2234 // (3) Similar a volatile write cannot let unrelated NON-volatile
2235 // memory refs that happen BEFORE the write float down to after the
2236 // write. It's OK for non-volatile memory refs that happen after the
2237 // volatile write to float up before it.
2238 //
2239 // We only put in barriers around volatile refs (they are expensive),
2240 // not _between_ memory refs (that would require us to track the
2241 // flavor of the previous memory refs). Requirements (2) and (3)
2242 // require some barriers before volatile stores and after volatile
2243 // loads. These nearly cover requirement (1) but miss the
2244 // volatile-store-volatile-load case. This final case is placed after
2245 // volatile-stores although it could just as well go before
2246 // volatile-loads.
2247
2248 void TemplateTable::resolve_cache_and_index(int byte_no,
2249 Register Rcache,
2250 Register index,
2251 size_t index_size) {
2252 const Register temp = r19;
2253 assert_different_registers(Rcache, index, temp);
2254
2255 Label resolved;
2256
2257 Bytecodes::Code code = bytecode();
2258 switch (code) {
2259 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2260 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2261 }
2262
2263 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2264 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2265 __ cmp(temp, (int) code); // have we resolved this bytecode?
2266 __ br(Assembler::EQ, resolved);
2267
2268 // resolve first time through
2269 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2270 __ mov(temp, (int) code);
2271 __ call_VM(noreg, entry, temp);
2272
2273 // Update registers with resolved info
2274 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2275 // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2276 // so all clients ofthis method must be modified accordingly
2277 __ bind(resolved);
2278 }
2279
2280 // The Rcache and index registers must be set before call
2281 // n.b unlike x86 cache already includes the index offset
2282 void TemplateTable::load_field_cp_cache_entry(Register obj,
2283 Register cache,
2284 Register index,
2285 Register off,
2286 Register flags,
2287 bool is_static = false) {
2288 assert_different_registers(cache, index, flags, off);
2289
2290 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2291 // Field offset
2292 __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2293 ConstantPoolCacheEntry::f2_offset())));
2294 // Flags
2295 __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2296 ConstantPoolCacheEntry::flags_offset())));
2297
2298 // klass overwrite register
2299 if (is_static) {
2300 __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2301 ConstantPoolCacheEntry::f1_offset())));
2302 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2303 __ ldr(obj, Address(obj, mirror_offset));
2304 }
2305 }
2306
2307 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2308 Register method,
2309 Register itable_index,
2310 Register flags,
2311 bool is_invokevirtual,
2312 bool is_invokevfinal, /*unused*/
2313 bool is_invokedynamic) {
2314 // setup registers
2315 const Register cache = rscratch2;
2316 const Register index = r4;
2317 assert_different_registers(method, flags);
2318 assert_different_registers(method, cache, index);
2319 assert_different_registers(itable_index, flags);
2320 assert_different_registers(itable_index, cache, index);
2321 // determine constant pool cache field offsets
2322 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2323 const int method_offset = in_bytes(
2324 ConstantPoolCache::base_offset() +
2325 (is_invokevirtual
2326 ? ConstantPoolCacheEntry::f2_offset()
2327 : ConstantPoolCacheEntry::f1_offset()));
2328 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2329 ConstantPoolCacheEntry::flags_offset());
2330 // access constant pool cache fields
2331 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2332 ConstantPoolCacheEntry::f2_offset());
2333
2334 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2335 resolve_cache_and_index(byte_no, cache, index, index_size);
2336 __ ldr(method, Address(cache, method_offset));
2337
2338 if (itable_index != noreg) {
2339 __ ldr(itable_index, Address(cache, index_offset));
2340 }
2341 __ ldrw(flags, Address(cache, flags_offset));
2342 }
2343
2344
2345 // The registers cache and index expected to be set before call.
2346 // Correct values of the cache and index registers are preserved.
2347 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2348 bool is_static, bool has_tos) {
2349 // do the JVMTI work here to avoid disturbing the register state below
2350 // We use c_rarg registers here because we want to use the register used in
2351 // the call to the VM
2352 if (JvmtiExport::can_post_field_access()) {
2353 // Check to see if a field access watch has been set before we
2354 // take the time to call into the VM.
2355 Label L1;
2356 assert_different_registers(cache, index, r0);
2357 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2358 __ ldrw(r0, Address(rscratch1));
2359 __ cbzw(r0, L1);
2360
2361 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2362 __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2363
2364 if (is_static) {
2365 __ mov(c_rarg1, zr); // NULL object reference
2366 } else {
2367 __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2368 __ verify_oop(c_rarg1);
2369 }
2370 // c_rarg1: object pointer or NULL
2371 // c_rarg2: cache entry pointer
2372 // c_rarg3: jvalue object on the stack
2373 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2374 InterpreterRuntime::post_field_access),
2375 c_rarg1, c_rarg2, c_rarg3);
2376 __ get_cache_and_index_at_bcp(cache, index, 1);
2377 __ bind(L1);
2378 }
2379 }
2380
2381 void TemplateTable::pop_and_check_object(Register r)
2382 {
2383 __ pop_ptr(r);
2384 __ null_check(r); // for field access must check obj.
2385 __ verify_oop(r);
2386 }
2387
2388 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2389 {
2390 const Register cache = r2;
2391 const Register index = r3;
2392 const Register obj = r4;
2393 const Register off = r19;
2394 const Register flags = r0;
2395 const Register bc = r4; // uses same reg as obj, so don't mix them
2396
2397 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2398 jvmti_post_field_access(cache, index, is_static, false);
2399 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2400
2401 if (!is_static) {
2402 // obj is on the stack
2403 pop_and_check_object(obj);
2404 }
2405
2406 const Address field(obj, off);
2407
2408 Label Done, notByte, notBool, notInt, notShort, notChar,
2409 notLong, notFloat, notObj, notDouble;
2410
2411 // x86 uses a shift and mask or wings it with a shift plus assert
2412 // the mask is not needed. aarch64 just uses bitfield extract
2413 __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
2414
2415 assert(btos == 0, "change code, btos != 0");
2416 __ cbnz(flags, notByte);
2417
2418 // Don't rewrite getstatic, only getfield
2419 if (is_static) rc = may_not_rewrite;
2420
2421 // btos
2422 __ load_signed_byte(r0, field);
2423 __ push(btos);
2424 // Rewrite bytecode to be faster
2425 if (rc == may_rewrite) {
2426 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2427 }
2428 __ b(Done);
2429
2430 __ bind(notByte);
2431 __ cmp(flags, ztos);
2432 __ br(Assembler::NE, notBool);
2433
2434 // ztos (same code as btos)
2435 __ ldrsb(r0, field);
2436 __ push(ztos);
2437 // Rewrite bytecode to be faster
2438 if (!is_static) {
2439 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2440 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2441 }
2442 __ b(Done);
2443
2444 __ bind(notBool);
2445 __ cmp(flags, atos);
2446 __ br(Assembler::NE, notObj);
2447 // atos
2448 __ load_heap_oop(r0, field);
2449 __ push(atos);
2450 if (rc == may_rewrite) {
2451 patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2452 }
2453 __ b(Done);
2454
2455 __ bind(notObj);
2456 __ cmp(flags, itos);
2457 __ br(Assembler::NE, notInt);
2458 // itos
2459 __ ldrw(r0, field);
2460 __ push(itos);
2461 // Rewrite bytecode to be faster
2462 if (rc == may_rewrite) {
2463 patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2464 }
2465 __ b(Done);
2466
2467 __ bind(notInt);
2468 __ cmp(flags, ctos);
2469 __ br(Assembler::NE, notChar);
2470 // ctos
2471 __ load_unsigned_short(r0, field);
2472 __ push(ctos);
2473 // Rewrite bytecode to be faster
2474 if (rc == may_rewrite) {
2475 patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2476 }
2477 __ b(Done);
2478
2479 __ bind(notChar);
2480 __ cmp(flags, stos);
2481 __ br(Assembler::NE, notShort);
2482 // stos
2483 __ load_signed_short(r0, field);
2484 __ push(stos);
2485 // Rewrite bytecode to be faster
2486 if (rc == may_rewrite) {
2487 patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2488 }
2489 __ b(Done);
2490
2491 __ bind(notShort);
2492 __ cmp(flags, ltos);
2493 __ br(Assembler::NE, notLong);
2494 // ltos
2495 __ ldr(r0, field);
2496 __ push(ltos);
2497 // Rewrite bytecode to be faster
2498 if (rc == may_rewrite) {
2499 patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2500 }
2501 __ b(Done);
2502
2503 __ bind(notLong);
2504 __ cmp(flags, ftos);
2505 __ br(Assembler::NE, notFloat);
2506 // ftos
2507 __ ldrs(v0, field);
2508 __ push(ftos);
2509 // Rewrite bytecode to be faster
2510 if (rc == may_rewrite) {
2511 patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2512 }
2513 __ b(Done);
2514
2515 __ bind(notFloat);
2516 #ifdef ASSERT
2517 __ cmp(flags, dtos);
2518 __ br(Assembler::NE, notDouble);
2519 #endif
2520 // dtos
2521 __ ldrd(v0, field);
2522 __ push(dtos);
2523 // Rewrite bytecode to be faster
2524 if (rc == may_rewrite) {
2525 patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2526 }
2527 #ifdef ASSERT
2528 __ b(Done);
2529
2530 __ bind(notDouble);
2531 __ stop("Bad state");
2532 #endif
2533
2534 __ bind(Done);
2535 // It's really not worth bothering to check whether this field
2536 // really is volatile in the slow case.
2537 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2538 }
2539
2540
2541 void TemplateTable::getfield(int byte_no)
2542 {
2543 getfield_or_static(byte_no, false);
2544 }
2545
2546 void TemplateTable::nofast_getfield(int byte_no) {
2547 getfield_or_static(byte_no, false, may_not_rewrite);
2548 }
2549
2550 void TemplateTable::getstatic(int byte_no)
2551 {
2552 getfield_or_static(byte_no, true);
2553 }
2554
2555 // The registers cache and index expected to be set before call.
2556 // The function may destroy various registers, just not the cache and index registers.
2557 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2558 transition(vtos, vtos);
2559
2560 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2561
2562 if (JvmtiExport::can_post_field_modification()) {
2563 // Check to see if a field modification watch has been set before
2564 // we take the time to call into the VM.
2565 Label L1;
2566 assert_different_registers(cache, index, r0);
2567 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2568 __ ldrw(r0, Address(rscratch1));
2569 __ cbz(r0, L1);
2570
2571 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2572
2573 if (is_static) {
2574 // Life is simple. Null out the object pointer.
2575 __ mov(c_rarg1, zr);
2576 } else {
2577 // Life is harder. The stack holds the value on top, followed by
2578 // the object. We don't know the size of the value, though; it
2579 // could be one or two words depending on its type. As a result,
2580 // we must find the type to determine where the object is.
2581 __ ldrw(c_rarg3, Address(c_rarg2,
2582 in_bytes(cp_base_offset +
2583 ConstantPoolCacheEntry::flags_offset())));
2584 __ lsr(c_rarg3, c_rarg3,
2585 ConstantPoolCacheEntry::tos_state_shift);
2586 ConstantPoolCacheEntry::verify_tos_state_shift();
2587 Label nope2, done, ok;
2588 __ ldr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2589 __ cmpw(c_rarg3, ltos);
2590 __ br(Assembler::EQ, ok);
2591 __ cmpw(c_rarg3, dtos);
2592 __ br(Assembler::NE, nope2);
2593 __ bind(ok);
2594 __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2595 __ bind(nope2);
2596 }
2597 // cache entry pointer
2598 __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2599 // object (tos)
2600 __ mov(c_rarg3, esp);
2601 // c_rarg1: object pointer set up above (NULL if static)
2602 // c_rarg2: cache entry pointer
2603 // c_rarg3: jvalue object on the stack
2604 __ call_VM(noreg,
2605 CAST_FROM_FN_PTR(address,
2606 InterpreterRuntime::post_field_modification),
2607 c_rarg1, c_rarg2, c_rarg3);
2608 __ get_cache_and_index_at_bcp(cache, index, 1);
2609 __ bind(L1);
2610 }
2611 }
2612
2613 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2614 transition(vtos, vtos);
2615
2616 const Register cache = r2;
2617 const Register index = r3;
2618 const Register obj = r2;
2619 const Register off = r19;
2620 const Register flags = r0;
2621 const Register bc = r4;
2622
2623 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2624 jvmti_post_field_mod(cache, index, is_static);
2625 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2626
2627 Label Done;
2628 __ mov(r5, flags);
2629
2630 {
2631 Label notVolatile;
2632 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2633 __ membar(MacroAssembler::StoreStore);
2634 __ bind(notVolatile);
2635 }
2636
2637 // field address
2638 const Address field(obj, off);
2639
2640 Label notByte, notBool, notInt, notShort, notChar,
2641 notLong, notFloat, notObj, notDouble;
2642
2643 // x86 uses a shift and mask or wings it with a shift plus assert
2644 // the mask is not needed. aarch64 just uses bitfield extract
2645 __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
2646
2647 assert(btos == 0, "change code, btos != 0");
2648 __ cbnz(flags, notByte);
2649
2650 // Don't rewrite putstatic, only putfield
2651 if (is_static) rc = may_not_rewrite;
2652
2653 // btos
2654 {
2655 __ pop(btos);
2656 if (!is_static) pop_and_check_object(obj);
2657 __ strb(r0, field);
2658 if (rc == may_rewrite) {
2659 patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2660 }
2661 __ b(Done);
2662 }
2663
2664 __ bind(notByte);
2665 __ cmp(flags, ztos);
2666 __ br(Assembler::NE, notBool);
2667
2668 // ztos
2669 {
2670 __ pop(ztos);
2671 if (!is_static) pop_and_check_object(obj);
2672 __ andw(r0, r0, 0x1);
2673 __ strb(r0, field);
2674 if (!is_static) {
2675 patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2676 }
2677 __ b(Done);
2678 }
2679
2680 __ bind(notBool);
2681 __ cmp(flags, atos);
2682 __ br(Assembler::NE, notObj);
2683
2684 // atos
2685 {
2686 __ pop(atos);
2687 if (!is_static) pop_and_check_object(obj);
2688 // Store into the field
2689 do_oop_store(_masm, field, r0, _bs->kind(), false);
2690 if (rc == may_rewrite) {
2691 patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2692 }
2693 __ b(Done);
2694 }
2695
2696 __ bind(notObj);
2697 __ cmp(flags, itos);
2698 __ br(Assembler::NE, notInt);
2699
2700 // itos
2701 {
2702 __ pop(itos);
2703 if (!is_static) pop_and_check_object(obj);
2704 __ strw(r0, field);
2705 if (rc == may_rewrite) {
2706 patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2707 }
2708 __ b(Done);
2709 }
2710
2711 __ bind(notInt);
2712 __ cmp(flags, ctos);
2713 __ br(Assembler::NE, notChar);
2714
2715 // ctos
2716 {
2717 __ pop(ctos);
2718 if (!is_static) pop_and_check_object(obj);
2719 __ strh(r0, field);
2720 if (rc == may_rewrite) {
2721 patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2722 }
2723 __ b(Done);
2724 }
2725
2726 __ bind(notChar);
2727 __ cmp(flags, stos);
2728 __ br(Assembler::NE, notShort);
2729
2730 // stos
2731 {
2732 __ pop(stos);
2733 if (!is_static) pop_and_check_object(obj);
2734 __ strh(r0, field);
2735 if (rc == may_rewrite) {
2736 patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2737 }
2738 __ b(Done);
2739 }
2740
2741 __ bind(notShort);
2742 __ cmp(flags, ltos);
2743 __ br(Assembler::NE, notLong);
2744
2745 // ltos
2746 {
2747 __ pop(ltos);
2748 if (!is_static) pop_and_check_object(obj);
2749 __ str(r0, field);
2750 if (rc == may_rewrite) {
2751 patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2752 }
2753 __ b(Done);
2754 }
2755
2756 __ bind(notLong);
2757 __ cmp(flags, ftos);
2758 __ br(Assembler::NE, notFloat);
2759
2760 // ftos
2761 {
2762 __ pop(ftos);
2763 if (!is_static) pop_and_check_object(obj);
2764 __ strs(v0, field);
2765 if (rc == may_rewrite) {
2766 patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2767 }
2768 __ b(Done);
2769 }
2770
2771 __ bind(notFloat);
2772 #ifdef ASSERT
2773 __ cmp(flags, dtos);
2774 __ br(Assembler::NE, notDouble);
2775 #endif
2776
2777 // dtos
2778 {
2779 __ pop(dtos);
2780 if (!is_static) pop_and_check_object(obj);
2781 __ strd(v0, field);
2782 if (rc == may_rewrite) {
2783 patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2784 }
2785 }
2786
2787 #ifdef ASSERT
2788 __ b(Done);
2789
2790 __ bind(notDouble);
2791 __ stop("Bad state");
2792 #endif
2793
2794 __ bind(Done);
2795
2796 {
2797 Label notVolatile;
2798 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2799 __ membar(MacroAssembler::StoreLoad);
2800 __ bind(notVolatile);
2801 }
2802 }
2803
2804 void TemplateTable::putfield(int byte_no)
2805 {
2806 putfield_or_static(byte_no, false);
2807 }
2808
2809 void TemplateTable::nofast_putfield(int byte_no) {
2810 putfield_or_static(byte_no, false, may_not_rewrite);
2811 }
2812
2813 void TemplateTable::putstatic(int byte_no) {
2814 putfield_or_static(byte_no, true);
2815 }
2816
2817 void TemplateTable::jvmti_post_fast_field_mod()
2818 {
2819 if (JvmtiExport::can_post_field_modification()) {
2820 // Check to see if a field modification watch has been set before
2821 // we take the time to call into the VM.
2822 Label L2;
2823 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2824 __ ldrw(c_rarg3, Address(rscratch1));
2825 __ cbzw(c_rarg3, L2);
2826 __ pop_ptr(r19); // copy the object pointer from tos
2827 __ verify_oop(r19);
2828 __ push_ptr(r19); // put the object pointer back on tos
2829 // Save tos values before call_VM() clobbers them. Since we have
2830 // to do it for every data type, we use the saved values as the
2831 // jvalue object.
2832 switch (bytecode()) { // load values into the jvalue object
2833 case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2834 case Bytecodes::_fast_bputfield: // fall through
2835 case Bytecodes::_fast_zputfield: // fall through
2836 case Bytecodes::_fast_sputfield: // fall through
2837 case Bytecodes::_fast_cputfield: // fall through
2838 case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2839 case Bytecodes::_fast_dputfield: __ push_d(); break;
2840 case Bytecodes::_fast_fputfield: __ push_f(); break;
2841 case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2842
2843 default:
2844 ShouldNotReachHere();
2845 }
2846 __ mov(c_rarg3, esp); // points to jvalue on the stack
2847 // access constant pool cache entry
2848 __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2849 __ verify_oop(r19);
2850 // r19: object pointer copied above
2851 // c_rarg2: cache entry pointer
2852 // c_rarg3: jvalue object on the stack
2853 __ call_VM(noreg,
2854 CAST_FROM_FN_PTR(address,
2855 InterpreterRuntime::post_field_modification),
2856 r19, c_rarg2, c_rarg3);
2857
2858 switch (bytecode()) { // restore tos values
2859 case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2860 case Bytecodes::_fast_bputfield: // fall through
2861 case Bytecodes::_fast_zputfield: // fall through
2862 case Bytecodes::_fast_sputfield: // fall through
2863 case Bytecodes::_fast_cputfield: // fall through
2864 case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2865 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2866 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2867 case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2868 }
2869 __ bind(L2);
2870 }
2871 }
2872
2873 void TemplateTable::fast_storefield(TosState state)
2874 {
2875 transition(state, vtos);
2876
2877 ByteSize base = ConstantPoolCache::base_offset();
2878
2879 jvmti_post_fast_field_mod();
2880
2881 // access constant pool cache
2882 __ get_cache_and_index_at_bcp(r2, r1, 1);
2883
2884 // test for volatile with r3
2885 __ ldrw(r3, Address(r2, in_bytes(base +
2886 ConstantPoolCacheEntry::flags_offset())));
2887
2888 // replace index with field offset from cache entry
2889 __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2890
2891 {
2892 Label notVolatile;
2893 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2894 __ membar(MacroAssembler::StoreStore);
2895 __ bind(notVolatile);
2896 }
2897
2898 Label notVolatile;
2899
2900 // Get object from stack
2901 pop_and_check_object(r2);
2902
2903 // field address
2904 const Address field(r2, r1);
2905
2906 // access field
2907 switch (bytecode()) {
2908 case Bytecodes::_fast_aputfield:
2909 do_oop_store(_masm, field, r0, _bs->kind(), false);
2910 break;
2911 case Bytecodes::_fast_lputfield:
2912 __ str(r0, field);
2913 break;
2914 case Bytecodes::_fast_iputfield:
2915 __ strw(r0, field);
2916 break;
2917 case Bytecodes::_fast_zputfield:
2918 __ andw(r0, r0, 0x1); // boolean is true if LSB is 1
2919 // fall through to bputfield
2920 case Bytecodes::_fast_bputfield:
2921 __ strb(r0, field);
2922 break;
2923 case Bytecodes::_fast_sputfield:
2924 // fall through
2925 case Bytecodes::_fast_cputfield:
2926 __ strh(r0, field);
2927 break;
2928 case Bytecodes::_fast_fputfield:
2929 __ strs(v0, field);
2930 break;
2931 case Bytecodes::_fast_dputfield:
2932 __ strd(v0, field);
2933 break;
2934 default:
2935 ShouldNotReachHere();
2936 }
2937
2938 {
2939 Label notVolatile;
2940 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2941 __ membar(MacroAssembler::StoreLoad);
2942 __ bind(notVolatile);
2943 }
2944 }
2945
2946
2947 void TemplateTable::fast_accessfield(TosState state)
2948 {
2949 transition(atos, state);
2950 // Do the JVMTI work here to avoid disturbing the register state below
2951 if (JvmtiExport::can_post_field_access()) {
2952 // Check to see if a field access watch has been set before we
2953 // take the time to call into the VM.
2954 Label L1;
2955 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2956 __ ldrw(r2, Address(rscratch1));
2957 __ cbzw(r2, L1);
2958 // access constant pool cache entry
2959 __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
2960 __ verify_oop(r0);
2961 __ push_ptr(r0); // save object pointer before call_VM() clobbers it
2962 __ mov(c_rarg1, r0);
2963 // c_rarg1: object pointer copied above
2964 // c_rarg2: cache entry pointer
2965 __ call_VM(noreg,
2966 CAST_FROM_FN_PTR(address,
2967 InterpreterRuntime::post_field_access),
2968 c_rarg1, c_rarg2);
2969 __ pop_ptr(r0); // restore object pointer
2970 __ bind(L1);
2971 }
2972
2973 // access constant pool cache
2974 __ get_cache_and_index_at_bcp(r2, r1, 1);
2975 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2976 ConstantPoolCacheEntry::f2_offset())));
2977 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2978 ConstantPoolCacheEntry::flags_offset())));
2979
2980 // r0: object
2981 __ verify_oop(r0);
2982 __ null_check(r0);
2983 const Address field(r0, r1);
2984
2985 // access field
2986 switch (bytecode()) {
2987 case Bytecodes::_fast_agetfield:
2988 __ load_heap_oop(r0, field);
2989 __ verify_oop(r0);
2990 break;
2991 case Bytecodes::_fast_lgetfield:
2992 __ ldr(r0, field);
2993 break;
2994 case Bytecodes::_fast_igetfield:
2995 __ ldrw(r0, field);
2996 break;
2997 case Bytecodes::_fast_bgetfield:
2998 __ load_signed_byte(r0, field);
2999 break;
3000 case Bytecodes::_fast_sgetfield:
3001 __ load_signed_short(r0, field);
3002 break;
3003 case Bytecodes::_fast_cgetfield:
3004 __ load_unsigned_short(r0, field);
3005 break;
3006 case Bytecodes::_fast_fgetfield:
3007 __ ldrs(v0, field);
3008 break;
3009 case Bytecodes::_fast_dgetfield:
3010 __ ldrd(v0, field);
3011 break;
3012 default:
3013 ShouldNotReachHere();
3014 }
3015 {
3016 Label notVolatile;
3017 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3018 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3019 __ bind(notVolatile);
3020 }
3021 }
3022
3023 void TemplateTable::fast_xaccess(TosState state)
3024 {
3025 transition(vtos, state);
3026
3027 // get receiver
3028 __ ldr(r0, aaddress(0));
3029 // access constant pool cache
3030 __ get_cache_and_index_at_bcp(r2, r3, 2);
3031 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3032 ConstantPoolCacheEntry::f2_offset())));
3033 // make sure exception is reported in correct bcp range (getfield is
3034 // next instruction)
3035 __ increment(rbcp);
3036 __ null_check(r0);
3037 switch (state) {
3038 case itos:
3039 __ ldrw(r0, Address(r0, r1, Address::lsl(0)));
3040 break;
3041 case atos:
3042 __ load_heap_oop(r0, Address(r0, r1, Address::lsl(0)));
3043 __ verify_oop(r0);
3044 break;
3045 case ftos:
3046 __ ldrs(v0, Address(r0, r1, Address::lsl(0)));
3047 break;
3048 default:
3049 ShouldNotReachHere();
3050 }
3051
3052 {
3053 Label notVolatile;
3054 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3055 ConstantPoolCacheEntry::flags_offset())));
3056 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3057 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3058 __ bind(notVolatile);
3059 }
3060
3061 __ decrement(rbcp);
3062 }
3063
3064
3065
3066 //-----------------------------------------------------------------------------
3067 // Calls
3068
3069 void TemplateTable::count_calls(Register method, Register temp)
3070 {
3071 __ call_Unimplemented();
3072 }
3073
3074 void TemplateTable::prepare_invoke(int byte_no,
3075 Register method, // linked method (or i-klass)
3076 Register index, // itable index, MethodType, etc.
3077 Register recv, // if caller wants to see it
3078 Register flags // if caller wants to test it
3079 ) {
3080 // determine flags
3081 Bytecodes::Code code = bytecode();
3082 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3083 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3084 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3085 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3086 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3087 const bool load_receiver = (recv != noreg);
3088 const bool save_flags = (flags != noreg);
3089 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3090 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3091 assert(flags == noreg || flags == r3, "");
3092 assert(recv == noreg || recv == r2, "");
3093
3094 // setup registers & access constant pool cache
3095 if (recv == noreg) recv = r2;
3096 if (flags == noreg) flags = r3;
3097 assert_different_registers(method, index, recv, flags);
3098
3099 // save 'interpreter return address'
3100 __ save_bcp();
3101
3102 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3103
3104 // maybe push appendix to arguments (just before return address)
3105 if (is_invokedynamic || is_invokehandle) {
3106 Label L_no_push;
3107 __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3108 // Push the appendix as a trailing parameter.
3109 // This must be done before we get the receiver,
3110 // since the parameter_size includes it.
3111 __ push(r19);
3112 __ mov(r19, index);
3113 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3114 __ load_resolved_reference_at_index(index, r19);
3115 __ pop(r19);
3116 __ push(index); // push appendix (MethodType, CallSite, etc.)
3117 __ bind(L_no_push);
3118 }
3119
3120 // load receiver if needed (note: no return address pushed yet)
3121 if (load_receiver) {
3122 __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3123 // FIXME -- is this actually correct? looks like it should be 2
3124 // const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3125 // const int receiver_is_at_end = -1; // back off one slot to get receiver
3126 // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3127 // __ movptr(recv, recv_addr);
3128 __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3129 __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3130 __ verify_oop(recv);
3131 }
3132
3133 // compute return type
3134 // x86 uses a shift and mask or wings it with a shift plus assert
3135 // the mask is not needed. aarch64 just uses bitfield extract
3136 __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
3137 // load return address
3138 {
3139 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3140 __ mov(rscratch1, table_addr);
3141 __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3142 }
3143 }
3144
3145
3146 void TemplateTable::invokevirtual_helper(Register index,
3147 Register recv,
3148 Register flags)
3149 {
3150 // Uses temporary registers r0, r3
3151 assert_different_registers(index, recv, r0, r3);
3152 // Test for an invoke of a final method
3153 Label notFinal;
3154 __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3155
3156 const Register method = index; // method must be rmethod
3157 assert(method == rmethod,
3158 "methodOop must be rmethod for interpreter calling convention");
3159
3160 // do the call - the index is actually the method to call
3161 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3162
3163 // It's final, need a null check here!
3164 __ null_check(recv);
3165
3166 // profile this call
3167 __ profile_final_call(r0);
3168 __ profile_arguments_type(r0, method, r4, true);
3169
3170 __ jump_from_interpreted(method, r0);
3171
3172 __ bind(notFinal);
3173
3174 // get receiver klass
3175 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3176 __ load_klass(r0, recv);
3177
3178 // profile this call
3179 __ profile_virtual_call(r0, rlocals, r3);
3180
3181 // get target methodOop & entry point
3182 __ lookup_virtual_method(r0, index, method);
3183 __ profile_arguments_type(r3, method, r4, true);
3184 // FIXME -- this looks completely redundant. is it?
3185 // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3186 __ jump_from_interpreted(method, r3);
3187 }
3188
3189 void TemplateTable::invokevirtual(int byte_no)
3190 {
3191 transition(vtos, vtos);
3192 assert(byte_no == f2_byte, "use this argument");
3193
3194 prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3195
3196 // rmethod: index (actually a Method*)
3197 // r2: receiver
3198 // r3: flags
3199
3200 invokevirtual_helper(rmethod, r2, r3);
3201 }
3202
3203 void TemplateTable::invokespecial(int byte_no)
3204 {
3205 transition(vtos, vtos);
3206 assert(byte_no == f1_byte, "use this argument");
3207
3208 prepare_invoke(byte_no, rmethod, noreg, // get f1 Method*
3209 r2); // get receiver also for null check
3210 __ verify_oop(r2);
3211 __ null_check(r2);
3212 // do the call
3213 __ profile_call(r0);
3214 __ profile_arguments_type(r0, rmethod, rbcp, false);
3215 __ jump_from_interpreted(rmethod, r0);
3216 }
3217
3218 void TemplateTable::invokestatic(int byte_no)
3219 {
3220 transition(vtos, vtos);
3221 assert(byte_no == f1_byte, "use this argument");
3222
3223 prepare_invoke(byte_no, rmethod); // get f1 Method*
3224 // do the call
3225 __ profile_call(r0);
3226 __ profile_arguments_type(r0, rmethod, r4, false);
3227 __ jump_from_interpreted(rmethod, r0);
3228 }
3229
3230 void TemplateTable::fast_invokevfinal(int byte_no)
3231 {
3232 __ call_Unimplemented();
3233 }
3234
3235 void TemplateTable::invokeinterface(int byte_no) {
3236 transition(vtos, vtos);
3237 assert(byte_no == f1_byte, "use this argument");
3238
3239 prepare_invoke(byte_no, r0, rmethod, // get f1 Klass*, f2 itable index
3240 r2, r3); // recv, flags
3241
3242 // r0: interface klass (from f1)
3243 // rmethod: itable index (from f2)
3244 // r2: receiver
3245 // r3: flags
3246
3247 // Special case of invokeinterface called for virtual method of
3248 // java.lang.Object. See cpCacheOop.cpp for details.
3249 // This code isn't produced by javac, but could be produced by
3250 // another compliant java compiler.
3251 Label notMethod;
3252 __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notMethod);
3253
3254 invokevirtual_helper(rmethod, r2, r3);
3255 __ bind(notMethod);
3256
3257 // Get receiver klass into r3 - also a null check
3258 __ restore_locals();
3259 __ null_check(r2, oopDesc::klass_offset_in_bytes());
3260 __ load_klass(r3, r2);
3261
3262 // profile this call
3263 __ profile_virtual_call(r3, r13, r19);
3264
3265 Label no_such_interface, no_such_method;
3266
3267 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3268 r3, r0, rmethod,
3269 // outputs: method, scan temp. reg
3270 rmethod, r13,
3271 no_such_interface);
3272
3273 // rmethod,: methodOop to call
3274 // r2: receiver
3275 // Check for abstract method error
3276 // Note: This should be done more efficiently via a throw_abstract_method_error
3277 // interpreter entry point and a conditional jump to it in case of a null
3278 // method.
3279 __ cbz(rmethod, no_such_method);
3280
3281 __ profile_arguments_type(r3, rmethod, r13, true);
3282
3283 // do the call
3284 // r2: receiver
3285 // rmethod,: methodOop
3286 __ jump_from_interpreted(rmethod, r3);
3287 __ should_not_reach_here();
3288
3289 // exception handling code follows...
3290 // note: must restore interpreter registers to canonical
3291 // state for exception handling to work correctly!
3292
3293 __ bind(no_such_method);
3294 // throw exception
3295 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3296 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3297 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3298 // the call_VM checks for exception, so we should never return here.
3299 __ should_not_reach_here();
3300
3301 __ bind(no_such_interface);
3302 // throw exception
3303 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3304 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3305 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3306 InterpreterRuntime::throw_IncompatibleClassChangeError));
3307 // the call_VM checks for exception, so we should never return here.
3308 __ should_not_reach_here();
3309 return;
3310 }
3311
3312 void TemplateTable::invokehandle(int byte_no) {
3313 transition(vtos, vtos);
3314 assert(byte_no == f1_byte, "use this argument");
3315
3316 prepare_invoke(byte_no, rmethod, r0, r2);
3317 __ verify_method_ptr(r2);
3318 __ verify_oop(r2);
3319 __ null_check(r2);
3320
3321 // FIXME: profile the LambdaForm also
3322
3323 // r13 is safe to use here as a scratch reg because it is about to
3324 // be clobbered by jump_from_interpreted().
3325 __ profile_final_call(r13);
3326 __ profile_arguments_type(r13, rmethod, r4, true);
3327
3328 __ jump_from_interpreted(rmethod, r0);
3329 }
3330
3331 void TemplateTable::invokedynamic(int byte_no) {
3332 transition(vtos, vtos);
3333 assert(byte_no == f1_byte, "use this argument");
3334
3335 prepare_invoke(byte_no, rmethod, r0);
3336
3337 // r0: CallSite object (from cpool->resolved_references[])
3338 // rmethod: MH.linkToCallSite method (from f2)
3339
3340 // Note: r0_callsite is already pushed by prepare_invoke
3341
3342 // %%% should make a type profile for any invokedynamic that takes a ref argument
3343 // profile this call
3344 __ profile_call(rbcp);
3345 __ profile_arguments_type(r3, rmethod, r13, false);
3346
3347 __ verify_oop(r0);
3348
3349 __ jump_from_interpreted(rmethod, r0);
3350 }
3351
3352
3353 //-----------------------------------------------------------------------------
3354 // Allocation
3355
3356 void TemplateTable::_new() {
3357 transition(vtos, atos);
3358
3359 __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3360 Label slow_case;
3361 Label done;
3362 Label initialize_header;
3363 Label initialize_object; // including clearing the fields
3364 Label allocate_shared;
3365
3366 __ get_cpool_and_tags(r4, r0);
3367 // Make sure the class we're about to instantiate has been resolved.
3368 // This is done before loading InstanceKlass to be consistent with the order
3369 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3370 const int tags_offset = Array<u1>::base_offset_in_bytes();
3371 __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3372 __ lea(rscratch1, Address(rscratch1, tags_offset));
3373 __ ldarb(rscratch1, rscratch1);
3374 __ cmp(rscratch1, JVM_CONSTANT_Class);
3375 __ br(Assembler::NE, slow_case);
3376
3377 // get InstanceKlass
3378 __ lea(r4, Address(r4, r3, Address::lsl(3)));
3379 __ ldr(r4, Address(r4, sizeof(ConstantPool)));
3380
3381 // make sure klass is initialized & doesn't have finalizer
3382 // make sure klass is fully initialized
3383 __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3384 __ cmp(rscratch1, InstanceKlass::fully_initialized);
3385 __ br(Assembler::NE, slow_case);
3386
3387 // get instance_size in InstanceKlass (scaled to a count of bytes)
3388 __ ldrw(r3,
3389 Address(r4,
3390 Klass::layout_helper_offset()));
3391 // test to see if it has a finalizer or is malformed in some way
3392 __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3393
3394 // Allocate the instance
3395 // 1) Try to allocate in the TLAB
3396 // 2) if fail and the object is large allocate in the shared Eden
3397 // 3) if the above fails (or is not applicable), go to a slow case
3398 // (creates a new TLAB, etc.)
3399
3400 const bool allow_shared_alloc =
3401 Universe::heap()->supports_inline_contig_alloc();
3402
3403 if (UseTLAB) {
3404 __ tlab_allocate(r0, r3, 0, noreg, r1,
3405 allow_shared_alloc ? allocate_shared : slow_case);
3406
3407 if (ZeroTLAB) {
3408 // the fields have been already cleared
3409 __ b(initialize_header);
3410 } else {
3411 // initialize both the header and fields
3412 __ b(initialize_object);
3413 }
3414 }
3415
3416 // Allocation in the shared Eden, if allowed.
3417 //
3418 // r3: instance size in bytes
3419 if (allow_shared_alloc) {
3420 __ bind(allocate_shared);
3421
3422 __ eden_allocate(r0, r3, 0, r10, slow_case);
3423 __ incr_allocated_bytes(rthread, r3, 0, rscratch1);
3424 }
3425
3426 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3427 // The object is initialized before the header. If the object size is
3428 // zero, go directly to the header initialization.
3429 __ bind(initialize_object);
3430 __ sub(r3, r3, sizeof(oopDesc));
3431 __ cbz(r3, initialize_header);
3432
3433 // Initialize object fields
3434 {
3435 __ add(r2, r0, sizeof(oopDesc));
3436 Label loop;
3437 __ bind(loop);
3438 __ str(zr, Address(__ post(r2, BytesPerLong)));
3439 __ sub(r3, r3, BytesPerLong);
3440 __ cbnz(r3, loop);
3441 }
3442
3443 // initialize object header only.
3444 __ bind(initialize_header);
3445 if (UseBiasedLocking) {
3446 __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3447 } else {
3448 __ mov(rscratch1, (intptr_t)markOopDesc::prototype());
3449 }
3450 __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3451 __ store_klass_gap(r0, zr); // zero klass gap for compressed oops
3452 __ store_klass(r0, r4); // store klass last
3453
3454 {
3455 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3456 // Trigger dtrace event for fastpath
3457 __ push(atos); // save the return value
3458 __ call_VM_leaf(
3459 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3460 __ pop(atos); // restore the return value
3461
3462 }
3463 __ b(done);
3464 }
3465
3466 // slow case
3467 __ bind(slow_case);
3468 __ get_constant_pool(c_rarg1);
3469 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3470 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3471 __ verify_oop(r0);
3472
3473 // continue
3474 __ bind(done);
3475 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3476 __ membar(Assembler::StoreStore);
3477 }
3478
3479 void TemplateTable::newarray() {
3480 transition(itos, atos);
3481 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3482 __ mov(c_rarg2, r0);
3483 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3484 c_rarg1, c_rarg2);
3485 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3486 __ membar(Assembler::StoreStore);
3487 }
3488
3489 void TemplateTable::anewarray() {
3490 transition(itos, atos);
3491 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3492 __ get_constant_pool(c_rarg1);
3493 __ mov(c_rarg3, r0);
3494 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3495 c_rarg1, c_rarg2, c_rarg3);
3496 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3497 __ membar(Assembler::StoreStore);
3498 }
3499
3500 void TemplateTable::arraylength() {
3501 transition(atos, itos);
3502 __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3503 __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3504 }
3505
3506 void TemplateTable::checkcast()
3507 {
3508 transition(atos, atos);
3509 Label done, is_null, ok_is_subtype, quicked, resolved;
3510 __ cbz(r0, is_null);
3511
3512 // Get cpool & tags index
3513 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3514 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3515 // See if bytecode has already been quicked
3516 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3517 __ lea(r1, Address(rscratch1, r19));
3518 __ ldarb(r1, r1);
3519 __ cmp(r1, JVM_CONSTANT_Class);
3520 __ br(Assembler::EQ, quicked);
3521
3522 __ push(atos); // save receiver for result, and for GC
3523 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3524 // vm_result_2 has metadata result
3525 __ get_vm_result_2(r0, rthread);
3526 __ pop(r3); // restore receiver
3527 __ b(resolved);
3528
3529 // Get superklass in r0 and subklass in r3
3530 __ bind(quicked);
3531 __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3532 __ lea(r0, Address(r2, r19, Address::lsl(3)));
3533 __ ldr(r0, Address(r0, sizeof(ConstantPool)));
3534
3535 __ bind(resolved);
3536 __ load_klass(r19, r3);
3537
3538 // Generate subtype check. Blows r2, r5. Object in r3.
3539 // Superklass in r0. Subklass in r19.
3540 __ gen_subtype_check(r19, ok_is_subtype);
3541
3542 // Come here on failure
3543 __ push(r3);
3544 // object is at TOS
3545 __ b(Interpreter::_throw_ClassCastException_entry);
3546
3547 // Come here on success
3548 __ bind(ok_is_subtype);
3549 __ mov(r0, r3); // Restore object in r3
3550
3551 // Collect counts on whether this test sees NULLs a lot or not.
3552 if (ProfileInterpreter) {
3553 __ b(done);
3554 __ bind(is_null);
3555 __ profile_null_seen(r2);
3556 } else {
3557 __ bind(is_null); // same as 'done'
3558 }
3559 __ bind(done);
3560 }
3561
3562 void TemplateTable::instanceof() {
3563 transition(atos, itos);
3564 Label done, is_null, ok_is_subtype, quicked, resolved;
3565 __ cbz(r0, is_null);
3566
3567 // Get cpool & tags index
3568 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3569 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3570 // See if bytecode has already been quicked
3571 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3572 __ lea(r1, Address(rscratch1, r19));
3573 __ ldarb(r1, r1);
3574 __ cmp(r1, JVM_CONSTANT_Class);
3575 __ br(Assembler::EQ, quicked);
3576
3577 __ push(atos); // save receiver for result, and for GC
3578 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3579 // vm_result_2 has metadata result
3580 __ get_vm_result_2(r0, rthread);
3581 __ pop(r3); // restore receiver
3582 __ verify_oop(r3);
3583 __ load_klass(r3, r3);
3584 __ b(resolved);
3585
3586 // Get superklass in r0 and subklass in r3
3587 __ bind(quicked);
3588 __ load_klass(r3, r0);
3589 __ lea(r0, Address(r2, r19, Address::lsl(3)));
3590 __ ldr(r0, Address(r0, sizeof(ConstantPool)));
3591
3592 __ bind(resolved);
3593
3594 // Generate subtype check. Blows r2, r5
3595 // Superklass in r0. Subklass in r3.
3596 __ gen_subtype_check(r3, ok_is_subtype);
3597
3598 // Come here on failure
3599 __ mov(r0, 0);
3600 __ b(done);
3601 // Come here on success
3602 __ bind(ok_is_subtype);
3603 __ mov(r0, 1);
3604
3605 // Collect counts on whether this test sees NULLs a lot or not.
3606 if (ProfileInterpreter) {
3607 __ b(done);
3608 __ bind(is_null);
3609 __ profile_null_seen(r2);
3610 } else {
3611 __ bind(is_null); // same as 'done'
3612 }
3613 __ bind(done);
3614 // r0 = 0: obj == NULL or obj is not an instanceof the specified klass
3615 // r0 = 1: obj != NULL and obj is an instanceof the specified klass
3616 }
3617
3618 //-----------------------------------------------------------------------------
3619 // Breakpoints
3620 void TemplateTable::_breakpoint() {
3621 // Note: We get here even if we are single stepping..
3622 // jbug inists on setting breakpoints at every bytecode
3623 // even if we are in single step mode.
3624
3625 transition(vtos, vtos);
3626
3627 // get the unpatched byte code
3628 __ get_method(c_rarg1);
3629 __ call_VM(noreg,
3630 CAST_FROM_FN_PTR(address,
3631 InterpreterRuntime::get_original_bytecode_at),
3632 c_rarg1, rbcp);
3633 __ mov(r19, r0);
3634
3635 // post the breakpoint event
3636 __ call_VM(noreg,
3637 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3638 rmethod, rbcp);
3639
3640 // complete the execution of original bytecode
3641 __ mov(rscratch1, r19);
3642 __ dispatch_only_normal(vtos);
3643 }
3644
3645 //-----------------------------------------------------------------------------
3646 // Exceptions
3647
3648 void TemplateTable::athrow() {
3649 transition(atos, vtos);
3650 __ null_check(r0);
3651 __ b(Interpreter::throw_exception_entry());
3652 }
3653
3654 //-----------------------------------------------------------------------------
3655 // Synchronization
3656 //
3657 // Note: monitorenter & exit are symmetric routines; which is reflected
3658 // in the assembly code structure as well
3659 //
3660 // Stack layout:
3661 //
3662 // [expressions ] <--- esp = expression stack top
3663 // ..
3664 // [expressions ]
3665 // [monitor entry] <--- monitor block top = expression stack bot
3666 // ..
3667 // [monitor entry]
3668 // [frame data ] <--- monitor block bot
3669 // ...
3670 // [saved rbp ] <--- rbp
3671 void TemplateTable::monitorenter()
3672 {
3673 transition(atos, vtos);
3674
3675 // check for NULL object
3676 __ null_check(r0);
3677
3678 const Address monitor_block_top(
3679 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3680 const Address monitor_block_bot(
3681 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3682 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3683
3684 Label allocated;
3685
3686 // initialize entry pointer
3687 __ mov(c_rarg1, zr); // points to free slot or NULL
3688
3689 // find a free slot in the monitor block (result in c_rarg1)
3690 {
3691 Label entry, loop, exit;
3692 __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3693 // starting with top-most entry
3694 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3695
3696 __ b(entry);
3697
3698 __ bind(loop);
3699 // check if current entry is used
3700 // if not used then remember entry in c_rarg1
3701 __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3702 __ cmp(zr, rscratch1);
3703 __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3704 // check if current entry is for same object
3705 __ cmp(r0, rscratch1);
3706 // if same object then stop searching
3707 __ br(Assembler::EQ, exit);
3708 // otherwise advance to next entry
3709 __ add(c_rarg3, c_rarg3, entry_size);
3710 __ bind(entry);
3711 // check if bottom reached
3712 __ cmp(c_rarg3, c_rarg2);
3713 // if not at bottom then check this entry
3714 __ br(Assembler::NE, loop);
3715 __ bind(exit);
3716 }
3717
3718 __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3719 // if found, continue with that on
3720
3721 // allocate one if there's no free slot
3722 {
3723 Label entry, loop, no_adjust;
3724 // 1. compute new pointers // rsp: old expression stack top
3725 __ ldr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
3726 __ sub(esp, esp, entry_size); // move expression stack top
3727 __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3728 __ mov(c_rarg3, esp); // set start value for copy loop
3729 __ str(c_rarg1, monitor_block_bot); // set new monitor block bottom
3730
3731 __ cmp(sp, c_rarg3); // Check if we need to move sp
3732 __ br(Assembler::LO, no_adjust); // to allow more stack space
3733 // for our new esp
3734 __ sub(sp, sp, 2 * wordSize);
3735 __ bind(no_adjust);
3736
3737 __ b(entry);
3738 // 2. move expression stack contents
3739 __ bind(loop);
3740 __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3741 // word from old location
3742 __ str(c_rarg2, Address(c_rarg3, 0)); // and store it at new location
3743 __ add(c_rarg3, c_rarg3, wordSize); // advance to next word
3744 __ bind(entry);
3745 __ cmp(c_rarg3, c_rarg1); // check if bottom reached
3746 __ br(Assembler::NE, loop); // if not at bottom then
3747 // copy next word
3748 }
3749
3750 // call run-time routine
3751 // c_rarg1: points to monitor entry
3752 __ bind(allocated);
3753
3754 // Increment bcp to point to the next bytecode, so exception
3755 // handling for async. exceptions work correctly.
3756 // The object has already been poped from the stack, so the
3757 // expression stack looks correct.
3758 __ increment(rbcp);
3759
3760 // store object
3761 __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3762 __ lock_object(c_rarg1);
3763
3764 // check to make sure this monitor doesn't cause stack overflow after locking
3765 __ save_bcp(); // in case of exception
3766 __ generate_stack_overflow_check(0);
3767
3768 // The bcp has already been incremented. Just need to dispatch to
3769 // next instruction.
3770 __ dispatch_next(vtos);
3771 }
3772
3773
3774 void TemplateTable::monitorexit()
3775 {
3776 transition(atos, vtos);
3777
3778 // check for NULL object
3779 __ null_check(r0);
3780
3781 const Address monitor_block_top(
3782 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3783 const Address monitor_block_bot(
3784 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3785 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3786
3787 Label found;
3788
3789 // find matching slot
3790 {
3791 Label entry, loop;
3792 __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3793 // starting with top-most entry
3794 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3795 // of monitor block
3796 __ b(entry);
3797
3798 __ bind(loop);
3799 // check if current entry is for same object
3800 __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3801 __ cmp(r0, rscratch1);
3802 // if same object then stop searching
3803 __ br(Assembler::EQ, found);
3804 // otherwise advance to next entry
3805 __ add(c_rarg1, c_rarg1, entry_size);
3806 __ bind(entry);
3807 // check if bottom reached
3808 __ cmp(c_rarg1, c_rarg2);
3809 // if not at bottom then check this entry
3810 __ br(Assembler::NE, loop);
3811 }
3812
3813 // error handling. Unlocking was not block-structured
3814 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3815 InterpreterRuntime::throw_illegal_monitor_state_exception));
3816 __ should_not_reach_here();
3817
3818 // call run-time routine
3819 __ bind(found);
3820 __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3821 __ unlock_object(c_rarg1);
3822 __ pop_ptr(r0); // discard object
3823 }
3824
3825
3826 // Wide instructions
3827 void TemplateTable::wide()
3828 {
3829 __ load_unsigned_byte(r19, at_bcp(1));
3830 __ mov(rscratch1, (address)Interpreter::_wentry_point);
3831 __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
3832 __ br(rscratch1);
3833 }
3834
3835
3836 // Multi arrays
3837 void TemplateTable::multianewarray() {
3838 transition(vtos, atos);
3839 __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
3840 // last dim is on top of stack; we want address of first one:
3841 // first_addr = last_addr + (ndims - 1) * wordSize
3842 __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
3843 __ sub(c_rarg1, c_rarg1, wordSize);
3844 call_VM(r0,
3845 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3846 c_rarg1);
3847 __ load_unsigned_byte(r1, at_bcp(3));
3848 __ lea(esp, Address(esp, r1, Address::uxtw(3)));
3849 }