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