1 /*
2 * Copyright (c) 2003, 2017, 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 "gc/shared/barrierSetCodeGen.hpp"
29 #include "interpreter/interpreter.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/interp_masm.hpp"
32 #include "interpreter/templateTable.hpp"
33 #include "memory/universe.inline.hpp"
34 #include "oops/methodData.hpp"
35 #include "oops/method.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "prims/methodHandles.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/stubRoutines.hpp"
41 #include "runtime/synchronizer.hpp"
42
43 #define __ _masm->
44
45 // Platform-dependent initialization
46
47 void TemplateTable::pd_initialize() {
48 // No aarch64 specific initialization
49 }
50
51 // Address computation: local variables
52
53 static inline Address iaddress(int n) {
54 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
55 }
56
57 static inline Address laddress(int n) {
58 return iaddress(n + 1);
59 }
60
61 static inline Address faddress(int n) {
62 return iaddress(n);
63 }
64
65 static inline Address daddress(int n) {
66 return laddress(n);
67 }
68
69 static inline Address aaddress(int n) {
70 return iaddress(n);
71 }
72
73 static inline Address iaddress(Register r) {
74 return Address(rlocals, r, Address::lsl(3));
75 }
76
77 static inline Address laddress(Register r, Register scratch,
78 InterpreterMacroAssembler* _masm) {
79 __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
80 return Address(scratch, Interpreter::local_offset_in_bytes(1));
81 }
82
83 static inline Address faddress(Register r) {
84 return iaddress(r);
85 }
86
87 static inline Address daddress(Register r, Register scratch,
88 InterpreterMacroAssembler* _masm) {
89 return laddress(r, scratch, _masm);
90 }
91
92 static inline Address aaddress(Register r) {
93 return iaddress(r);
94 }
95
96 static inline Address at_rsp() {
97 return Address(esp, 0);
98 }
99
100 // At top of Java expression stack which may be different than esp(). It
101 // isn't for category 1 objects.
102 static inline Address at_tos () {
103 return Address(esp, Interpreter::expr_offset_in_bytes(0));
104 }
105
106 static inline Address at_tos_p1() {
107 return Address(esp, Interpreter::expr_offset_in_bytes(1));
108 }
109
110 static inline Address at_tos_p2() {
111 return Address(esp, Interpreter::expr_offset_in_bytes(2));
112 }
113
114 static inline Address at_tos_p3() {
115 return Address(esp, Interpreter::expr_offset_in_bytes(3));
116 }
117
118 static inline Address at_tos_p4() {
119 return Address(esp, Interpreter::expr_offset_in_bytes(4));
120 }
121
122 static inline Address at_tos_p5() {
123 return Address(esp, Interpreter::expr_offset_in_bytes(5));
124 }
125
126 // Condition conversion
127 static Assembler::Condition j_not(TemplateTable::Condition cc) {
128 switch (cc) {
129 case TemplateTable::equal : return Assembler::NE;
130 case TemplateTable::not_equal : return Assembler::EQ;
131 case TemplateTable::less : return Assembler::GE;
132 case TemplateTable::less_equal : return Assembler::GT;
133 case TemplateTable::greater : return Assembler::LE;
134 case TemplateTable::greater_equal: return Assembler::LT;
135 }
136 ShouldNotReachHere();
137 return Assembler::EQ;
138 }
139
140
141 // Miscelaneous helper routines
142 // Store an oop (or NULL) at the Address described by obj.
143 // If val == noreg this means store a NULL
144 static void do_oop_store(InterpreterMacroAssembler* _masm,
145 Address dst,
146 Register val,
147 DecoratorSet decorators) {
148 assert(val == noreg || val == r0, "parameter is just for looks");
149 BarrierSetCodeGen *code_gen = Universe::heap()->barrier_set()->code_gen();
150 code_gen->store_at(_masm, decorators, T_OBJECT, dst, val, /*tmp1*/ r10, /*tmp2*/ r1);
151 }
152
153 static void do_oop_load(InterpreterMacroAssembler* _masm,
154 Address src,
155 Register dst,
156 DecoratorSet decorators) {
157 BarrierSetCodeGen *code_gen = Universe::heap()->barrier_set()->code_gen();
158 code_gen->load_at(_masm, decorators, T_OBJECT, dst, src, /*tmp1*/ r10, /*tmp_thread*/ r1);
159 }
160
161 Address TemplateTable::at_bcp(int offset) {
162 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
163 return Address(rbcp, offset);
164 }
165
166 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
167 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
168 int byte_no)
169 {
170 if (!RewriteBytecodes) return;
171 Label L_patch_done;
172
173 switch (bc) {
174 case Bytecodes::_fast_aputfield:
175 case Bytecodes::_fast_bputfield:
176 case Bytecodes::_fast_zputfield:
177 case Bytecodes::_fast_cputfield:
178 case Bytecodes::_fast_dputfield:
179 case Bytecodes::_fast_fputfield:
180 case Bytecodes::_fast_iputfield:
181 case Bytecodes::_fast_lputfield:
182 case Bytecodes::_fast_sputfield:
183 {
184 // We skip bytecode quickening for putfield instructions when
185 // the put_code written to the constant pool cache is zero.
186 // This is required so that every execution of this instruction
187 // calls out to InterpreterRuntime::resolve_get_put to do
188 // additional, required work.
189 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
190 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
191 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
192 __ movw(bc_reg, bc);
193 __ cbzw(temp_reg, L_patch_done); // don't patch
194 }
195 break;
196 default:
197 assert(byte_no == -1, "sanity");
198 // the pair bytecodes have already done the load.
199 if (load_bc_into_bc_reg) {
200 __ movw(bc_reg, bc);
201 }
202 }
203
204 if (JvmtiExport::can_post_breakpoint()) {
205 Label L_fast_patch;
206 // if a breakpoint is present we can't rewrite the stream directly
207 __ load_unsigned_byte(temp_reg, at_bcp(0));
208 __ cmpw(temp_reg, Bytecodes::_breakpoint);
209 __ br(Assembler::NE, L_fast_patch);
210 // Let breakpoint table handling rewrite to quicker bytecode
211 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
212 __ b(L_patch_done);
213 __ bind(L_fast_patch);
214 }
215
216 #ifdef ASSERT
217 Label L_okay;
218 __ load_unsigned_byte(temp_reg, at_bcp(0));
219 __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
220 __ br(Assembler::EQ, L_okay);
221 __ cmpw(temp_reg, bc_reg);
222 __ br(Assembler::EQ, L_okay);
223 __ stop("patching the wrong bytecode");
224 __ bind(L_okay);
225 #endif
226
227 // patch bytecode
228 __ strb(bc_reg, at_bcp(0));
229 __ bind(L_patch_done);
230 }
231
232
233 // Individual instructions
234
235 void TemplateTable::nop() {
236 transition(vtos, vtos);
237 // nothing to do
238 }
239
240 void TemplateTable::shouldnotreachhere() {
241 transition(vtos, vtos);
242 __ stop("shouldnotreachhere bytecode");
243 }
244
245 void TemplateTable::aconst_null()
246 {
247 transition(vtos, atos);
248 __ mov(r0, 0);
249 }
250
251 void TemplateTable::iconst(int value)
252 {
253 transition(vtos, itos);
254 __ mov(r0, value);
255 }
256
257 void TemplateTable::lconst(int value)
258 {
259 __ mov(r0, value);
260 }
261
262 void TemplateTable::fconst(int value)
263 {
264 transition(vtos, ftos);
265 switch (value) {
266 case 0:
267 __ fmovs(v0, zr);
268 break;
269 case 1:
270 __ fmovs(v0, 1.0);
271 break;
272 case 2:
273 __ fmovs(v0, 2.0);
274 break;
275 default:
276 ShouldNotReachHere();
277 break;
278 }
279 }
280
281 void TemplateTable::dconst(int value)
282 {
283 transition(vtos, dtos);
284 switch (value) {
285 case 0:
286 __ fmovd(v0, zr);
287 break;
288 case 1:
289 __ fmovd(v0, 1.0);
290 break;
291 case 2:
292 __ fmovd(v0, 2.0);
293 break;
294 default:
295 ShouldNotReachHere();
296 break;
297 }
298 }
299
300 void TemplateTable::bipush()
301 {
302 transition(vtos, itos);
303 __ load_signed_byte32(r0, at_bcp(1));
304 }
305
306 void TemplateTable::sipush()
307 {
308 transition(vtos, itos);
309 __ load_unsigned_short(r0, at_bcp(1));
310 __ revw(r0, r0);
311 __ asrw(r0, r0, 16);
312 }
313
314 void TemplateTable::ldc(bool wide)
315 {
316 transition(vtos, vtos);
317 Label call_ldc, notFloat, notClass, Done;
318
319 if (wide) {
320 __ get_unsigned_2_byte_index_at_bcp(r1, 1);
321 } else {
322 __ load_unsigned_byte(r1, at_bcp(1));
323 }
324 __ get_cpool_and_tags(r2, r0);
325
326 const int base_offset = ConstantPool::header_size() * wordSize;
327 const int tags_offset = Array<u1>::base_offset_in_bytes();
328
329 // get type
330 __ add(r3, r1, tags_offset);
331 __ lea(r3, Address(r0, r3));
332 __ ldarb(r3, r3);
333
334 // unresolved class - get the resolved class
335 __ cmp(r3, JVM_CONSTANT_UnresolvedClass);
336 __ br(Assembler::EQ, call_ldc);
337
338 // unresolved class in error state - call into runtime to throw the error
339 // from the first resolution attempt
340 __ cmp(r3, JVM_CONSTANT_UnresolvedClassInError);
341 __ br(Assembler::EQ, call_ldc);
342
343 // resolved class - need to call vm to get java mirror of the class
344 __ cmp(r3, JVM_CONSTANT_Class);
345 __ br(Assembler::NE, notClass);
346
347 __ bind(call_ldc);
348 __ mov(c_rarg1, wide);
349 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
350 __ push_ptr(r0);
351 __ verify_oop(r0);
352 __ b(Done);
353
354 __ bind(notClass);
355 __ cmp(r3, JVM_CONSTANT_Float);
356 __ br(Assembler::NE, notFloat);
357 // ftos
358 __ adds(r1, r2, r1, Assembler::LSL, 3);
359 __ ldrs(v0, Address(r1, base_offset));
360 __ push_f();
361 __ b(Done);
362
363 __ bind(notFloat);
364 #ifdef ASSERT
365 {
366 Label L;
367 __ cmp(r3, JVM_CONSTANT_Integer);
368 __ br(Assembler::EQ, L);
369 // String and Object are rewritten to fast_aldc
370 __ stop("unexpected tag type in ldc");
371 __ bind(L);
372 }
373 #endif
374 // itos JVM_CONSTANT_Integer only
375 __ adds(r1, r2, r1, Assembler::LSL, 3);
376 __ ldrw(r0, Address(r1, base_offset));
377 __ push_i(r0);
378 __ bind(Done);
379 }
380
381 // Fast path for caching oop constants.
382 void TemplateTable::fast_aldc(bool wide)
383 {
384 transition(vtos, atos);
385
386 Register result = r0;
387 Register tmp = r1;
388 int index_size = wide ? sizeof(u2) : sizeof(u1);
389
390 Label resolved;
391
392 // We are resolved if the resolved reference cache entry contains a
393 // non-null object (String, MethodType, etc.)
394 assert_different_registers(result, tmp);
395 __ get_cache_index_at_bcp(tmp, 1, index_size);
396 __ load_resolved_reference_at_index(result, tmp);
397 __ cbnz(result, resolved);
398
399 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
400
401 // first time invocation - must resolve first
402 __ mov(tmp, (int)bytecode());
403 __ call_VM(result, entry, tmp);
404
405 __ bind(resolved);
406
407 if (VerifyOops) {
408 __ verify_oop(result);
409 }
410 }
411
412 void TemplateTable::ldc2_w()
413 {
414 transition(vtos, vtos);
415 Label Long, Done;
416 __ get_unsigned_2_byte_index_at_bcp(r0, 1);
417
418 __ get_cpool_and_tags(r1, r2);
419 const int base_offset = ConstantPool::header_size() * wordSize;
420 const int tags_offset = Array<u1>::base_offset_in_bytes();
421
422 // get type
423 __ lea(r2, Address(r2, r0, Address::lsl(0)));
424 __ load_unsigned_byte(r2, Address(r2, tags_offset));
425 __ cmpw(r2, (int)JVM_CONSTANT_Double);
426 __ br(Assembler::NE, Long);
427 // dtos
428 __ lea (r2, Address(r1, r0, Address::lsl(3)));
429 __ ldrd(v0, Address(r2, base_offset));
430 __ push_d();
431 __ b(Done);
432
433 __ bind(Long);
434 // ltos
435 __ lea(r0, Address(r1, r0, Address::lsl(3)));
436 __ ldr(r0, Address(r0, base_offset));
437 __ push_l();
438
439 __ bind(Done);
440 }
441
442 void TemplateTable::locals_index(Register reg, int offset)
443 {
444 __ ldrb(reg, at_bcp(offset));
445 __ neg(reg, reg);
446 }
447
448 void TemplateTable::iload() {
449 iload_internal();
450 }
451
452 void TemplateTable::nofast_iload() {
453 iload_internal(may_not_rewrite);
454 }
455
456 void TemplateTable::iload_internal(RewriteControl rc) {
457 transition(vtos, itos);
458 if (RewriteFrequentPairs && rc == may_rewrite) {
459 Label rewrite, done;
460 Register bc = r4;
461
462 // get next bytecode
463 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
464
465 // if _iload, wait to rewrite to iload2. We only want to rewrite the
466 // last two iloads in a pair. Comparing against fast_iload means that
467 // the next bytecode is neither an iload or a caload, and therefore
468 // an iload pair.
469 __ cmpw(r1, Bytecodes::_iload);
470 __ br(Assembler::EQ, done);
471
472 // if _fast_iload rewrite to _fast_iload2
473 __ cmpw(r1, Bytecodes::_fast_iload);
474 __ movw(bc, Bytecodes::_fast_iload2);
475 __ br(Assembler::EQ, rewrite);
476
477 // if _caload rewrite to _fast_icaload
478 __ cmpw(r1, Bytecodes::_caload);
479 __ movw(bc, Bytecodes::_fast_icaload);
480 __ br(Assembler::EQ, rewrite);
481
482 // else rewrite to _fast_iload
483 __ movw(bc, Bytecodes::_fast_iload);
484
485 // rewrite
486 // bc: new bytecode
487 __ bind(rewrite);
488 patch_bytecode(Bytecodes::_iload, bc, r1, false);
489 __ bind(done);
490
491 }
492
493 // do iload, get the local value into tos
494 locals_index(r1);
495 __ ldr(r0, iaddress(r1));
496
497 }
498
499 void TemplateTable::fast_iload2()
500 {
501 transition(vtos, itos);
502 locals_index(r1);
503 __ ldr(r0, iaddress(r1));
504 __ push(itos);
505 locals_index(r1, 3);
506 __ ldr(r0, iaddress(r1));
507 }
508
509 void TemplateTable::fast_iload()
510 {
511 transition(vtos, itos);
512 locals_index(r1);
513 __ ldr(r0, iaddress(r1));
514 }
515
516 void TemplateTable::lload()
517 {
518 transition(vtos, ltos);
519 __ ldrb(r1, at_bcp(1));
520 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
521 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
522 }
523
524 void TemplateTable::fload()
525 {
526 transition(vtos, ftos);
527 locals_index(r1);
528 // n.b. we use ldrd here because this is a 64 bit slot
529 // this is comparable to the iload case
530 __ ldrd(v0, faddress(r1));
531 }
532
533 void TemplateTable::dload()
534 {
535 transition(vtos, dtos);
536 __ ldrb(r1, at_bcp(1));
537 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
538 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
539 }
540
541 void TemplateTable::aload()
542 {
543 transition(vtos, atos);
544 locals_index(r1);
545 __ ldr(r0, iaddress(r1));
546 }
547
548 void TemplateTable::locals_index_wide(Register reg) {
549 __ ldrh(reg, at_bcp(2));
550 __ rev16w(reg, reg);
551 __ neg(reg, reg);
552 }
553
554 void TemplateTable::wide_iload() {
555 transition(vtos, itos);
556 locals_index_wide(r1);
557 __ ldr(r0, iaddress(r1));
558 }
559
560 void TemplateTable::wide_lload()
561 {
562 transition(vtos, ltos);
563 __ ldrh(r1, at_bcp(2));
564 __ rev16w(r1, r1);
565 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
566 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
567 }
568
569 void TemplateTable::wide_fload()
570 {
571 transition(vtos, ftos);
572 locals_index_wide(r1);
573 // n.b. we use ldrd here because this is a 64 bit slot
574 // this is comparable to the iload case
575 __ ldrd(v0, faddress(r1));
576 }
577
578 void TemplateTable::wide_dload()
579 {
580 transition(vtos, dtos);
581 __ ldrh(r1, at_bcp(2));
582 __ rev16w(r1, r1);
583 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
584 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
585 }
586
587 void TemplateTable::wide_aload()
588 {
589 transition(vtos, atos);
590 locals_index_wide(r1);
591 __ ldr(r0, aaddress(r1));
592 }
593
594 void TemplateTable::index_check(Register array, Register index)
595 {
596 // destroys r1, rscratch1
597 // check array
598 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
599 // sign extend index for use by indexed load
600 // __ movl2ptr(index, index);
601 // check index
602 Register length = rscratch1;
603 __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
604 __ cmpw(index, length);
605 if (index != r1) {
606 // ??? convention: move aberrant index into r1 for exception message
607 assert(r1 != array, "different registers");
608 __ mov(r1, index);
609 }
610 Label ok;
611 __ br(Assembler::LO, ok);
612 __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
613 __ br(rscratch1);
614 __ bind(ok);
615 }
616
617 void TemplateTable::iaload()
618 {
619 transition(itos, itos);
620 __ mov(r1, r0);
621 __ pop_ptr(r0);
622 // r0: array
623 // r1: index
624 index_check(r0, r1); // leaves index in r1, kills rscratch1
625 __ lea(r1, Address(r0, r1, Address::uxtw(2)));
626 __ ldrw(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_INT)));
627 }
628
629 void TemplateTable::laload()
630 {
631 transition(itos, ltos);
632 __ mov(r1, r0);
633 __ pop_ptr(r0);
634 // r0: array
635 // r1: index
636 index_check(r0, r1); // leaves index in r1, kills rscratch1
637 __ lea(r1, Address(r0, r1, Address::uxtw(3)));
638 __ ldr(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_LONG)));
639 }
640
641 void TemplateTable::faload()
642 {
643 transition(itos, ftos);
644 __ mov(r1, r0);
645 __ pop_ptr(r0);
646 // r0: array
647 // r1: index
648 index_check(r0, r1); // leaves index in r1, kills rscratch1
649 __ lea(r1, Address(r0, r1, Address::uxtw(2)));
650 __ ldrs(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
651 }
652
653 void TemplateTable::daload()
654 {
655 transition(itos, dtos);
656 __ mov(r1, r0);
657 __ pop_ptr(r0);
658 // r0: array
659 // r1: index
660 index_check(r0, r1); // leaves index in r1, kills rscratch1
661 __ lea(r1, Address(r0, r1, Address::uxtw(3)));
662 __ ldrd(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
663 }
664
665 void TemplateTable::aaload()
666 {
667 transition(itos, atos);
668 __ mov(r1, r0);
669 __ pop_ptr(r0);
670 // r0: array
671 // r1: index
672 index_check(r0, r1); // leaves index in r1, kills rscratch1
673 int s = (UseCompressedOops ? 2 : 3);
674 __ lea(r1, Address(r0, r1, Address::uxtw(s)));
675 do_oop_load(_masm,
676 Address(r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
677 r0,
678 ACCESS_IN_HEAP | ACCESS_IN_HEAP_ARRAY);
679 }
680
681 void TemplateTable::baload()
682 {
683 transition(itos, itos);
684 __ mov(r1, r0);
685 __ pop_ptr(r0);
686 // r0: array
687 // r1: index
688 index_check(r0, r1); // leaves index in r1, kills rscratch1
689 __ lea(r1, Address(r0, r1, Address::uxtw(0)));
690 __ load_signed_byte(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
691 }
692
693 void TemplateTable::caload()
694 {
695 transition(itos, itos);
696 __ mov(r1, r0);
697 __ pop_ptr(r0);
698 // r0: array
699 // r1: index
700 index_check(r0, r1); // leaves index in r1, kills rscratch1
701 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
702 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
703 }
704
705 // iload followed by caload frequent pair
706 void TemplateTable::fast_icaload()
707 {
708 transition(vtos, itos);
709 // load index out of locals
710 locals_index(r2);
711 __ ldr(r1, iaddress(r2));
712
713 __ pop_ptr(r0);
714
715 // r0: array
716 // r1: index
717 index_check(r0, r1); // leaves index in r1, kills rscratch1
718 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
719 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
720 }
721
722 void TemplateTable::saload()
723 {
724 transition(itos, itos);
725 __ mov(r1, r0);
726 __ pop_ptr(r0);
727 // r0: array
728 // r1: index
729 index_check(r0, r1); // leaves index in r1, kills rscratch1
730 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
731 __ load_signed_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
732 }
733
734 void TemplateTable::iload(int n)
735 {
736 transition(vtos, itos);
737 __ ldr(r0, iaddress(n));
738 }
739
740 void TemplateTable::lload(int n)
741 {
742 transition(vtos, ltos);
743 __ ldr(r0, laddress(n));
744 }
745
746 void TemplateTable::fload(int n)
747 {
748 transition(vtos, ftos);
749 __ ldrs(v0, faddress(n));
750 }
751
752 void TemplateTable::dload(int n)
753 {
754 transition(vtos, dtos);
755 __ ldrd(v0, daddress(n));
756 }
757
758 void TemplateTable::aload(int n)
759 {
760 transition(vtos, atos);
761 __ ldr(r0, iaddress(n));
762 }
763
764 void TemplateTable::aload_0() {
765 aload_0_internal();
766 }
767
768 void TemplateTable::nofast_aload_0() {
769 aload_0_internal(may_not_rewrite);
770 }
771
772 void TemplateTable::aload_0_internal(RewriteControl rc) {
773 // According to bytecode histograms, the pairs:
774 //
775 // _aload_0, _fast_igetfield
776 // _aload_0, _fast_agetfield
777 // _aload_0, _fast_fgetfield
778 //
779 // occur frequently. If RewriteFrequentPairs is set, the (slow)
780 // _aload_0 bytecode checks if the next bytecode is either
781 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
782 // rewrites the current bytecode into a pair bytecode; otherwise it
783 // rewrites the current bytecode into _fast_aload_0 that doesn't do
784 // the pair check anymore.
785 //
786 // Note: If the next bytecode is _getfield, the rewrite must be
787 // delayed, otherwise we may miss an opportunity for a pair.
788 //
789 // Also rewrite frequent pairs
790 // aload_0, aload_1
791 // aload_0, iload_1
792 // These bytecodes with a small amount of code are most profitable
793 // to rewrite
794 if (RewriteFrequentPairs && rc == may_rewrite) {
795 Label rewrite, done;
796 const Register bc = r4;
797
798 // get next bytecode
799 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
800
801 // if _getfield then wait with rewrite
802 __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
803 __ br(Assembler::EQ, done);
804
805 // if _igetfield then rewrite to _fast_iaccess_0
806 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
807 __ cmpw(r1, Bytecodes::_fast_igetfield);
808 __ movw(bc, Bytecodes::_fast_iaccess_0);
809 __ br(Assembler::EQ, rewrite);
810
811 // if _agetfield then rewrite to _fast_aaccess_0
812 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
813 __ cmpw(r1, Bytecodes::_fast_agetfield);
814 __ movw(bc, Bytecodes::_fast_aaccess_0);
815 __ br(Assembler::EQ, rewrite);
816
817 // if _fgetfield then rewrite to _fast_faccess_0
818 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
819 __ cmpw(r1, Bytecodes::_fast_fgetfield);
820 __ movw(bc, Bytecodes::_fast_faccess_0);
821 __ br(Assembler::EQ, rewrite);
822
823 // else rewrite to _fast_aload0
824 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
825 __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
826
827 // rewrite
828 // bc: new bytecode
829 __ bind(rewrite);
830 patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
831
832 __ bind(done);
833 }
834
835 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
836 aload(0);
837 }
838
839 void TemplateTable::istore()
840 {
841 transition(itos, vtos);
842 locals_index(r1);
843 // FIXME: We're being very pernickerty here storing a jint in a
844 // local with strw, which costs an extra instruction over what we'd
845 // be able to do with a simple str. We should just store the whole
846 // word.
847 __ lea(rscratch1, iaddress(r1));
848 __ strw(r0, Address(rscratch1));
849 }
850
851 void TemplateTable::lstore()
852 {
853 transition(ltos, vtos);
854 locals_index(r1);
855 __ str(r0, laddress(r1, rscratch1, _masm));
856 }
857
858 void TemplateTable::fstore() {
859 transition(ftos, vtos);
860 locals_index(r1);
861 __ lea(rscratch1, iaddress(r1));
862 __ strs(v0, Address(rscratch1));
863 }
864
865 void TemplateTable::dstore() {
866 transition(dtos, vtos);
867 locals_index(r1);
868 __ strd(v0, daddress(r1, rscratch1, _masm));
869 }
870
871 void TemplateTable::astore()
872 {
873 transition(vtos, vtos);
874 __ pop_ptr(r0);
875 locals_index(r1);
876 __ str(r0, aaddress(r1));
877 }
878
879 void TemplateTable::wide_istore() {
880 transition(vtos, vtos);
881 __ pop_i();
882 locals_index_wide(r1);
883 __ lea(rscratch1, iaddress(r1));
884 __ strw(r0, Address(rscratch1));
885 }
886
887 void TemplateTable::wide_lstore() {
888 transition(vtos, vtos);
889 __ pop_l();
890 locals_index_wide(r1);
891 __ str(r0, laddress(r1, rscratch1, _masm));
892 }
893
894 void TemplateTable::wide_fstore() {
895 transition(vtos, vtos);
896 __ pop_f();
897 locals_index_wide(r1);
898 __ lea(rscratch1, faddress(r1));
899 __ strs(v0, rscratch1);
900 }
901
902 void TemplateTable::wide_dstore() {
903 transition(vtos, vtos);
904 __ pop_d();
905 locals_index_wide(r1);
906 __ strd(v0, daddress(r1, rscratch1, _masm));
907 }
908
909 void TemplateTable::wide_astore() {
910 transition(vtos, vtos);
911 __ pop_ptr(r0);
912 locals_index_wide(r1);
913 __ str(r0, aaddress(r1));
914 }
915
916 void TemplateTable::iastore() {
917 transition(itos, vtos);
918 __ pop_i(r1);
919 __ pop_ptr(r3);
920 // r0: value
921 // r1: index
922 // r3: array
923 index_check(r3, r1); // prefer index in r1
924 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
925 __ strw(r0, Address(rscratch1,
926 arrayOopDesc::base_offset_in_bytes(T_INT)));
927 }
928
929 void TemplateTable::lastore() {
930 transition(ltos, vtos);
931 __ pop_i(r1);
932 __ pop_ptr(r3);
933 // r0: value
934 // r1: index
935 // r3: array
936 index_check(r3, r1); // prefer index in r1
937 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
938 __ str(r0, Address(rscratch1,
939 arrayOopDesc::base_offset_in_bytes(T_LONG)));
940 }
941
942 void TemplateTable::fastore() {
943 transition(ftos, vtos);
944 __ pop_i(r1);
945 __ pop_ptr(r3);
946 // v0: value
947 // r1: index
948 // r3: array
949 index_check(r3, r1); // prefer index in r1
950 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
951 __ strs(v0, Address(rscratch1,
952 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
953 }
954
955 void TemplateTable::dastore() {
956 transition(dtos, vtos);
957 __ pop_i(r1);
958 __ pop_ptr(r3);
959 // v0: value
960 // r1: index
961 // r3: array
962 index_check(r3, r1); // prefer index in r1
963 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
964 __ strd(v0, Address(rscratch1,
965 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
966 }
967
968 void TemplateTable::aastore() {
969 Label is_null, ok_is_subtype, done;
970 transition(vtos, vtos);
971 // stack: ..., array, index, value
972 __ ldr(r0, at_tos()); // value
973 __ ldr(r2, at_tos_p1()); // index
974 __ ldr(r3, at_tos_p2()); // array
975
976 Address element_address(r4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
977
978 index_check(r3, r2); // kills r1
979 __ lea(r4, Address(r3, r2, Address::uxtw(UseCompressedOops? 2 : 3)));
980
981 // do array store check - check for NULL value first
982 __ cbz(r0, is_null);
983
984 // Move subklass into r1
985 __ load_klass(r1, r0);
986 // Move superklass into r0
987 __ load_klass(r0, r3);
988 __ ldr(r0, Address(r0,
989 ObjArrayKlass::element_klass_offset()));
990 // Compress array + index*oopSize + 12 into a single register. Frees r2.
991
992 // Generate subtype check. Blows r2, r5
993 // Superklass in r0. Subklass in r1.
994 __ gen_subtype_check(r1, ok_is_subtype);
995
996 // Come here on failure
997 // object is at TOS
998 __ b(Interpreter::_throw_ArrayStoreException_entry);
999
1000 // Come here on success
1001 __ bind(ok_is_subtype);
1002
1003 // Get the value we will store
1004 __ ldr(r0, at_tos());
1005 // Now store using the appropriate barrier
1006 do_oop_store(_masm, element_address, r0, ACCESS_IN_HEAP);
1007 __ b(done);
1008
1009 // Have a NULL in r0, r3=array, r2=index. Store NULL at ary[idx]
1010 __ bind(is_null);
1011 __ profile_null_seen(r2);
1012
1013 // Store a NULL
1014 do_oop_store(_masm, element_address, noreg, ACCESS_IN_HEAP);
1015
1016 // Pop stack arguments
1017 __ bind(done);
1018 __ add(esp, esp, 3 * Interpreter::stackElementSize);
1019 }
1020
1021 void TemplateTable::bastore()
1022 {
1023 transition(itos, vtos);
1024 __ pop_i(r1);
1025 __ pop_ptr(r3);
1026 // r0: value
1027 // r1: index
1028 // r3: array
1029 index_check(r3, r1); // prefer index in r1
1030
1031 // Need to check whether array is boolean or byte
1032 // since both types share the bastore bytecode.
1033 __ load_klass(r2, r3);
1034 __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1035 int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit());
1036 Label L_skip;
1037 __ tbz(r2, diffbit_index, L_skip);
1038 __ andw(r0, r0, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1039 __ bind(L_skip);
1040
1041 __ lea(rscratch1, Address(r3, r1, Address::uxtw(0)));
1042 __ strb(r0, Address(rscratch1,
1043 arrayOopDesc::base_offset_in_bytes(T_BYTE)));
1044 }
1045
1046 void TemplateTable::castore()
1047 {
1048 transition(itos, vtos);
1049 __ pop_i(r1);
1050 __ pop_ptr(r3);
1051 // r0: value
1052 // r1: index
1053 // r3: array
1054 index_check(r3, r1); // prefer index in r1
1055 __ lea(rscratch1, Address(r3, r1, Address::uxtw(1)));
1056 __ strh(r0, Address(rscratch1,
1057 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1058 }
1059
1060 void TemplateTable::sastore()
1061 {
1062 castore();
1063 }
1064
1065 void TemplateTable::istore(int n)
1066 {
1067 transition(itos, vtos);
1068 __ str(r0, iaddress(n));
1069 }
1070
1071 void TemplateTable::lstore(int n)
1072 {
1073 transition(ltos, vtos);
1074 __ str(r0, laddress(n));
1075 }
1076
1077 void TemplateTable::fstore(int n)
1078 {
1079 transition(ftos, vtos);
1080 __ strs(v0, faddress(n));
1081 }
1082
1083 void TemplateTable::dstore(int n)
1084 {
1085 transition(dtos, vtos);
1086 __ strd(v0, daddress(n));
1087 }
1088
1089 void TemplateTable::astore(int n)
1090 {
1091 transition(vtos, vtos);
1092 __ pop_ptr(r0);
1093 __ str(r0, iaddress(n));
1094 }
1095
1096 void TemplateTable::pop()
1097 {
1098 transition(vtos, vtos);
1099 __ add(esp, esp, Interpreter::stackElementSize);
1100 }
1101
1102 void TemplateTable::pop2()
1103 {
1104 transition(vtos, vtos);
1105 __ add(esp, esp, 2 * Interpreter::stackElementSize);
1106 }
1107
1108 void TemplateTable::dup()
1109 {
1110 transition(vtos, vtos);
1111 __ ldr(r0, Address(esp, 0));
1112 __ push(r0);
1113 // stack: ..., a, a
1114 }
1115
1116 void TemplateTable::dup_x1()
1117 {
1118 transition(vtos, vtos);
1119 // stack: ..., a, b
1120 __ ldr(r0, at_tos()); // load b
1121 __ ldr(r2, at_tos_p1()); // load a
1122 __ str(r0, at_tos_p1()); // store b
1123 __ str(r2, at_tos()); // store a
1124 __ push(r0); // push b
1125 // stack: ..., b, a, b
1126 }
1127
1128 void TemplateTable::dup_x2()
1129 {
1130 transition(vtos, vtos);
1131 // stack: ..., a, b, c
1132 __ ldr(r0, at_tos()); // load c
1133 __ ldr(r2, at_tos_p2()); // load a
1134 __ str(r0, at_tos_p2()); // store c in a
1135 __ push(r0); // push c
1136 // stack: ..., c, b, c, c
1137 __ ldr(r0, at_tos_p2()); // load b
1138 __ str(r2, at_tos_p2()); // store a in b
1139 // stack: ..., c, a, c, c
1140 __ str(r0, at_tos_p1()); // store b in c
1141 // stack: ..., c, a, b, c
1142 }
1143
1144 void TemplateTable::dup2()
1145 {
1146 transition(vtos, vtos);
1147 // stack: ..., a, b
1148 __ ldr(r0, at_tos_p1()); // load a
1149 __ push(r0); // push a
1150 __ ldr(r0, at_tos_p1()); // load b
1151 __ push(r0); // push b
1152 // stack: ..., a, b, a, b
1153 }
1154
1155 void TemplateTable::dup2_x1()
1156 {
1157 transition(vtos, vtos);
1158 // stack: ..., a, b, c
1159 __ ldr(r2, at_tos()); // load c
1160 __ ldr(r0, at_tos_p1()); // load b
1161 __ push(r0); // push b
1162 __ push(r2); // push c
1163 // stack: ..., a, b, c, b, c
1164 __ str(r2, at_tos_p3()); // store c in b
1165 // stack: ..., a, c, c, b, c
1166 __ ldr(r2, at_tos_p4()); // load a
1167 __ str(r2, at_tos_p2()); // store a in 2nd c
1168 // stack: ..., a, c, a, b, c
1169 __ str(r0, at_tos_p4()); // store b in a
1170 // stack: ..., b, c, a, b, c
1171 }
1172
1173 void TemplateTable::dup2_x2()
1174 {
1175 transition(vtos, vtos);
1176 // stack: ..., a, b, c, d
1177 __ ldr(r2, at_tos()); // load d
1178 __ ldr(r0, at_tos_p1()); // load c
1179 __ push(r0) ; // push c
1180 __ push(r2); // push d
1181 // stack: ..., a, b, c, d, c, d
1182 __ ldr(r0, at_tos_p4()); // load b
1183 __ str(r0, at_tos_p2()); // store b in d
1184 __ str(r2, at_tos_p4()); // store d in b
1185 // stack: ..., a, d, c, b, c, d
1186 __ ldr(r2, at_tos_p5()); // load a
1187 __ ldr(r0, at_tos_p3()); // load c
1188 __ str(r2, at_tos_p3()); // store a in c
1189 __ str(r0, at_tos_p5()); // store c in a
1190 // stack: ..., c, d, a, b, c, d
1191 }
1192
1193 void TemplateTable::swap()
1194 {
1195 transition(vtos, vtos);
1196 // stack: ..., a, b
1197 __ ldr(r2, at_tos_p1()); // load a
1198 __ ldr(r0, at_tos()); // load b
1199 __ str(r2, at_tos()); // store a in b
1200 __ str(r0, at_tos_p1()); // store b in a
1201 // stack: ..., b, a
1202 }
1203
1204 void TemplateTable::iop2(Operation op)
1205 {
1206 transition(itos, itos);
1207 // r0 <== r1 op r0
1208 __ pop_i(r1);
1209 switch (op) {
1210 case add : __ addw(r0, r1, r0); break;
1211 case sub : __ subw(r0, r1, r0); break;
1212 case mul : __ mulw(r0, r1, r0); break;
1213 case _and : __ andw(r0, r1, r0); break;
1214 case _or : __ orrw(r0, r1, r0); break;
1215 case _xor : __ eorw(r0, r1, r0); break;
1216 case shl : __ lslvw(r0, r1, r0); break;
1217 case shr : __ asrvw(r0, r1, r0); break;
1218 case ushr : __ lsrvw(r0, r1, r0);break;
1219 default : ShouldNotReachHere();
1220 }
1221 }
1222
1223 void TemplateTable::lop2(Operation op)
1224 {
1225 transition(ltos, ltos);
1226 // r0 <== r1 op r0
1227 __ pop_l(r1);
1228 switch (op) {
1229 case add : __ add(r0, r1, r0); break;
1230 case sub : __ sub(r0, r1, r0); break;
1231 case mul : __ mul(r0, r1, r0); break;
1232 case _and : __ andr(r0, r1, r0); break;
1233 case _or : __ orr(r0, r1, r0); break;
1234 case _xor : __ eor(r0, r1, r0); break;
1235 default : ShouldNotReachHere();
1236 }
1237 }
1238
1239 void TemplateTable::idiv()
1240 {
1241 transition(itos, itos);
1242 // explicitly check for div0
1243 Label no_div0;
1244 __ cbnzw(r0, no_div0);
1245 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1246 __ br(rscratch1);
1247 __ bind(no_div0);
1248 __ pop_i(r1);
1249 // r0 <== r1 idiv r0
1250 __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1251 }
1252
1253 void TemplateTable::irem()
1254 {
1255 transition(itos, itos);
1256 // explicitly check for div0
1257 Label no_div0;
1258 __ cbnzw(r0, no_div0);
1259 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1260 __ br(rscratch1);
1261 __ bind(no_div0);
1262 __ pop_i(r1);
1263 // r0 <== r1 irem r0
1264 __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1265 }
1266
1267 void TemplateTable::lmul()
1268 {
1269 transition(ltos, ltos);
1270 __ pop_l(r1);
1271 __ mul(r0, r0, r1);
1272 }
1273
1274 void TemplateTable::ldiv()
1275 {
1276 transition(ltos, ltos);
1277 // explicitly check for div0
1278 Label no_div0;
1279 __ cbnz(r0, no_div0);
1280 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1281 __ br(rscratch1);
1282 __ bind(no_div0);
1283 __ pop_l(r1);
1284 // r0 <== r1 ldiv r0
1285 __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1286 }
1287
1288 void TemplateTable::lrem()
1289 {
1290 transition(ltos, ltos);
1291 // explicitly check for div0
1292 Label no_div0;
1293 __ cbnz(r0, no_div0);
1294 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1295 __ br(rscratch1);
1296 __ bind(no_div0);
1297 __ pop_l(r1);
1298 // r0 <== r1 lrem r0
1299 __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1300 }
1301
1302 void TemplateTable::lshl()
1303 {
1304 transition(itos, ltos);
1305 // shift count is in r0
1306 __ pop_l(r1);
1307 __ lslv(r0, r1, r0);
1308 }
1309
1310 void TemplateTable::lshr()
1311 {
1312 transition(itos, ltos);
1313 // shift count is in r0
1314 __ pop_l(r1);
1315 __ asrv(r0, r1, r0);
1316 }
1317
1318 void TemplateTable::lushr()
1319 {
1320 transition(itos, ltos);
1321 // shift count is in r0
1322 __ pop_l(r1);
1323 __ lsrv(r0, r1, r0);
1324 }
1325
1326 void TemplateTable::fop2(Operation op)
1327 {
1328 transition(ftos, ftos);
1329 switch (op) {
1330 case add:
1331 // n.b. use ldrd because this is a 64 bit slot
1332 __ pop_f(v1);
1333 __ fadds(v0, v1, v0);
1334 break;
1335 case sub:
1336 __ pop_f(v1);
1337 __ fsubs(v0, v1, v0);
1338 break;
1339 case mul:
1340 __ pop_f(v1);
1341 __ fmuls(v0, v1, v0);
1342 break;
1343 case div:
1344 __ pop_f(v1);
1345 __ fdivs(v0, v1, v0);
1346 break;
1347 case rem:
1348 __ fmovs(v1, v0);
1349 __ pop_f(v0);
1350 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::frem),
1351 0, 2, MacroAssembler::ret_type_float);
1352 break;
1353 default:
1354 ShouldNotReachHere();
1355 break;
1356 }
1357 }
1358
1359 void TemplateTable::dop2(Operation op)
1360 {
1361 transition(dtos, dtos);
1362 switch (op) {
1363 case add:
1364 // n.b. use ldrd because this is a 64 bit slot
1365 __ pop_d(v1);
1366 __ faddd(v0, v1, v0);
1367 break;
1368 case sub:
1369 __ pop_d(v1);
1370 __ fsubd(v0, v1, v0);
1371 break;
1372 case mul:
1373 __ pop_d(v1);
1374 __ fmuld(v0, v1, v0);
1375 break;
1376 case div:
1377 __ pop_d(v1);
1378 __ fdivd(v0, v1, v0);
1379 break;
1380 case rem:
1381 __ fmovd(v1, v0);
1382 __ pop_d(v0);
1383 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::drem),
1384 0, 2, MacroAssembler::ret_type_double);
1385 break;
1386 default:
1387 ShouldNotReachHere();
1388 break;
1389 }
1390 }
1391
1392 void TemplateTable::ineg()
1393 {
1394 transition(itos, itos);
1395 __ negw(r0, r0);
1396
1397 }
1398
1399 void TemplateTable::lneg()
1400 {
1401 transition(ltos, ltos);
1402 __ neg(r0, r0);
1403 }
1404
1405 void TemplateTable::fneg()
1406 {
1407 transition(ftos, ftos);
1408 __ fnegs(v0, v0);
1409 }
1410
1411 void TemplateTable::dneg()
1412 {
1413 transition(dtos, dtos);
1414 __ fnegd(v0, v0);
1415 }
1416
1417 void TemplateTable::iinc()
1418 {
1419 transition(vtos, vtos);
1420 __ load_signed_byte(r1, at_bcp(2)); // get constant
1421 locals_index(r2);
1422 __ ldr(r0, iaddress(r2));
1423 __ addw(r0, r0, r1);
1424 __ str(r0, iaddress(r2));
1425 }
1426
1427 void TemplateTable::wide_iinc()
1428 {
1429 transition(vtos, vtos);
1430 // __ mov(r1, zr);
1431 __ ldrw(r1, at_bcp(2)); // get constant and index
1432 __ rev16(r1, r1);
1433 __ ubfx(r2, r1, 0, 16);
1434 __ neg(r2, r2);
1435 __ sbfx(r1, r1, 16, 16);
1436 __ ldr(r0, iaddress(r2));
1437 __ addw(r0, r0, r1);
1438 __ str(r0, iaddress(r2));
1439 }
1440
1441 void TemplateTable::convert()
1442 {
1443 // Checking
1444 #ifdef ASSERT
1445 {
1446 TosState tos_in = ilgl;
1447 TosState tos_out = ilgl;
1448 switch (bytecode()) {
1449 case Bytecodes::_i2l: // fall through
1450 case Bytecodes::_i2f: // fall through
1451 case Bytecodes::_i2d: // fall through
1452 case Bytecodes::_i2b: // fall through
1453 case Bytecodes::_i2c: // fall through
1454 case Bytecodes::_i2s: tos_in = itos; break;
1455 case Bytecodes::_l2i: // fall through
1456 case Bytecodes::_l2f: // fall through
1457 case Bytecodes::_l2d: tos_in = ltos; break;
1458 case Bytecodes::_f2i: // fall through
1459 case Bytecodes::_f2l: // fall through
1460 case Bytecodes::_f2d: tos_in = ftos; break;
1461 case Bytecodes::_d2i: // fall through
1462 case Bytecodes::_d2l: // fall through
1463 case Bytecodes::_d2f: tos_in = dtos; break;
1464 default : ShouldNotReachHere();
1465 }
1466 switch (bytecode()) {
1467 case Bytecodes::_l2i: // fall through
1468 case Bytecodes::_f2i: // fall through
1469 case Bytecodes::_d2i: // fall through
1470 case Bytecodes::_i2b: // fall through
1471 case Bytecodes::_i2c: // fall through
1472 case Bytecodes::_i2s: tos_out = itos; break;
1473 case Bytecodes::_i2l: // fall through
1474 case Bytecodes::_f2l: // fall through
1475 case Bytecodes::_d2l: tos_out = ltos; break;
1476 case Bytecodes::_i2f: // fall through
1477 case Bytecodes::_l2f: // fall through
1478 case Bytecodes::_d2f: tos_out = ftos; break;
1479 case Bytecodes::_i2d: // fall through
1480 case Bytecodes::_l2d: // fall through
1481 case Bytecodes::_f2d: tos_out = dtos; break;
1482 default : ShouldNotReachHere();
1483 }
1484 transition(tos_in, tos_out);
1485 }
1486 #endif // ASSERT
1487 // static const int64_t is_nan = 0x8000000000000000L;
1488
1489 // Conversion
1490 switch (bytecode()) {
1491 case Bytecodes::_i2l:
1492 __ sxtw(r0, r0);
1493 break;
1494 case Bytecodes::_i2f:
1495 __ scvtfws(v0, r0);
1496 break;
1497 case Bytecodes::_i2d:
1498 __ scvtfwd(v0, r0);
1499 break;
1500 case Bytecodes::_i2b:
1501 __ sxtbw(r0, r0);
1502 break;
1503 case Bytecodes::_i2c:
1504 __ uxthw(r0, r0);
1505 break;
1506 case Bytecodes::_i2s:
1507 __ sxthw(r0, r0);
1508 break;
1509 case Bytecodes::_l2i:
1510 __ uxtw(r0, r0);
1511 break;
1512 case Bytecodes::_l2f:
1513 __ scvtfs(v0, r0);
1514 break;
1515 case Bytecodes::_l2d:
1516 __ scvtfd(v0, r0);
1517 break;
1518 case Bytecodes::_f2i:
1519 {
1520 Label L_Okay;
1521 __ clear_fpsr();
1522 __ fcvtzsw(r0, v0);
1523 __ get_fpsr(r1);
1524 __ cbzw(r1, L_Okay);
1525 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2i),
1526 0, 1, MacroAssembler::ret_type_integral);
1527 __ bind(L_Okay);
1528 }
1529 break;
1530 case Bytecodes::_f2l:
1531 {
1532 Label L_Okay;
1533 __ clear_fpsr();
1534 __ fcvtzs(r0, v0);
1535 __ get_fpsr(r1);
1536 __ cbzw(r1, L_Okay);
1537 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2l),
1538 0, 1, MacroAssembler::ret_type_integral);
1539 __ bind(L_Okay);
1540 }
1541 break;
1542 case Bytecodes::_f2d:
1543 __ fcvts(v0, v0);
1544 break;
1545 case Bytecodes::_d2i:
1546 {
1547 Label L_Okay;
1548 __ clear_fpsr();
1549 __ fcvtzdw(r0, v0);
1550 __ get_fpsr(r1);
1551 __ cbzw(r1, L_Okay);
1552 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2i),
1553 0, 1, MacroAssembler::ret_type_integral);
1554 __ bind(L_Okay);
1555 }
1556 break;
1557 case Bytecodes::_d2l:
1558 {
1559 Label L_Okay;
1560 __ clear_fpsr();
1561 __ fcvtzd(r0, v0);
1562 __ get_fpsr(r1);
1563 __ cbzw(r1, L_Okay);
1564 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2l),
1565 0, 1, MacroAssembler::ret_type_integral);
1566 __ bind(L_Okay);
1567 }
1568 break;
1569 case Bytecodes::_d2f:
1570 __ fcvtd(v0, v0);
1571 break;
1572 default:
1573 ShouldNotReachHere();
1574 }
1575 }
1576
1577 void TemplateTable::lcmp()
1578 {
1579 transition(ltos, itos);
1580 Label done;
1581 __ pop_l(r1);
1582 __ cmp(r1, r0);
1583 __ mov(r0, (u_int64_t)-1L);
1584 __ br(Assembler::LT, done);
1585 // __ mov(r0, 1UL);
1586 // __ csel(r0, r0, zr, Assembler::NE);
1587 // and here is a faster way
1588 __ csinc(r0, zr, zr, Assembler::EQ);
1589 __ bind(done);
1590 }
1591
1592 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1593 {
1594 Label done;
1595 if (is_float) {
1596 // XXX get rid of pop here, use ... reg, mem32
1597 __ pop_f(v1);
1598 __ fcmps(v1, v0);
1599 } else {
1600 // XXX get rid of pop here, use ... reg, mem64
1601 __ pop_d(v1);
1602 __ fcmpd(v1, v0);
1603 }
1604 if (unordered_result < 0) {
1605 // we want -1 for unordered or less than, 0 for equal and 1 for
1606 // greater than.
1607 __ mov(r0, (u_int64_t)-1L);
1608 // for FP LT tests less than or unordered
1609 __ br(Assembler::LT, done);
1610 // install 0 for EQ otherwise 1
1611 __ csinc(r0, zr, zr, Assembler::EQ);
1612 } else {
1613 // we want -1 for less than, 0 for equal and 1 for unordered or
1614 // greater than.
1615 __ mov(r0, 1L);
1616 // for FP HI tests greater than or unordered
1617 __ br(Assembler::HI, done);
1618 // install 0 for EQ otherwise ~0
1619 __ csinv(r0, zr, zr, Assembler::EQ);
1620
1621 }
1622 __ bind(done);
1623 }
1624
1625 void TemplateTable::branch(bool is_jsr, bool is_wide)
1626 {
1627 // We might be moving to a safepoint. The thread which calls
1628 // Interpreter::notice_safepoints() will effectively flush its cache
1629 // when it makes a system call, but we need to do something to
1630 // ensure that we see the changed dispatch table.
1631 __ membar(MacroAssembler::LoadLoad);
1632
1633 __ profile_taken_branch(r0, r1);
1634 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1635 InvocationCounter::counter_offset();
1636 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1637 InvocationCounter::counter_offset();
1638
1639 // load branch displacement
1640 if (!is_wide) {
1641 __ ldrh(r2, at_bcp(1));
1642 __ rev16(r2, r2);
1643 // sign extend the 16 bit value in r2
1644 __ sbfm(r2, r2, 0, 15);
1645 } else {
1646 __ ldrw(r2, at_bcp(1));
1647 __ revw(r2, r2);
1648 // sign extend the 32 bit value in r2
1649 __ sbfm(r2, r2, 0, 31);
1650 }
1651
1652 // Handle all the JSR stuff here, then exit.
1653 // It's much shorter and cleaner than intermingling with the non-JSR
1654 // normal-branch stuff occurring below.
1655
1656 if (is_jsr) {
1657 // Pre-load the next target bytecode into rscratch1
1658 __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1659 // compute return address as bci
1660 __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1661 __ add(rscratch2, rscratch2,
1662 in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1663 __ sub(r1, rbcp, rscratch2);
1664 __ push_i(r1);
1665 // Adjust the bcp by the 16-bit displacement in r2
1666 __ add(rbcp, rbcp, r2);
1667 __ dispatch_only(vtos);
1668 return;
1669 }
1670
1671 // Normal (non-jsr) branch handling
1672
1673 // Adjust the bcp by the displacement in r2
1674 __ add(rbcp, rbcp, r2);
1675
1676 assert(UseLoopCounter || !UseOnStackReplacement,
1677 "on-stack-replacement requires loop counters");
1678 Label backedge_counter_overflow;
1679 Label profile_method;
1680 Label dispatch;
1681 if (UseLoopCounter) {
1682 // increment backedge counter for backward branches
1683 // r0: MDO
1684 // w1: MDO bumped taken-count
1685 // r2: target offset
1686 __ cmp(r2, zr);
1687 __ br(Assembler::GT, dispatch); // count only if backward branch
1688
1689 // ECN: FIXME: This code smells
1690 // check if MethodCounters exists
1691 Label has_counters;
1692 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1693 __ cbnz(rscratch1, has_counters);
1694 __ push(r0);
1695 __ push(r1);
1696 __ push(r2);
1697 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1698 InterpreterRuntime::build_method_counters), rmethod);
1699 __ pop(r2);
1700 __ pop(r1);
1701 __ pop(r0);
1702 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1703 __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1704 __ bind(has_counters);
1705
1706 if (TieredCompilation) {
1707 Label no_mdo;
1708 int increment = InvocationCounter::count_increment;
1709 if (ProfileInterpreter) {
1710 // Are we profiling?
1711 __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1712 __ cbz(r1, no_mdo);
1713 // Increment the MDO backedge counter
1714 const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1715 in_bytes(InvocationCounter::counter_offset()));
1716 const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1717 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1718 r0, rscratch1, false, Assembler::EQ, &backedge_counter_overflow);
1719 __ b(dispatch);
1720 }
1721 __ bind(no_mdo);
1722 // Increment backedge counter in MethodCounters*
1723 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1724 const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1725 __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1726 r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow);
1727 } else { // not TieredCompilation
1728 // increment counter
1729 __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1730 __ ldrw(r0, Address(rscratch2, be_offset)); // load backedge counter
1731 __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1732 __ strw(rscratch1, Address(rscratch2, be_offset)); // store counter
1733
1734 __ ldrw(r0, Address(rscratch2, inv_offset)); // load invocation counter
1735 __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1736 __ addw(r0, r0, rscratch1); // add both counters
1737
1738 if (ProfileInterpreter) {
1739 // Test to see if we should create a method data oop
1740 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
1741 __ cmpw(r0, rscratch1);
1742 __ br(Assembler::LT, dispatch);
1743
1744 // if no method data exists, go to profile method
1745 __ test_method_data_pointer(r0, profile_method);
1746
1747 if (UseOnStackReplacement) {
1748 // check for overflow against w1 which is the MDO taken count
1749 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1750 __ cmpw(r1, rscratch1);
1751 __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1752
1753 // When ProfileInterpreter is on, the backedge_count comes
1754 // from the MethodData*, which value does not get reset on
1755 // the call to frequency_counter_overflow(). To avoid
1756 // excessive calls to the overflow routine while the method is
1757 // being compiled, add a second test to make sure the overflow
1758 // function is called only once every overflow_frequency.
1759 const int overflow_frequency = 1024;
1760 __ andsw(r1, r1, overflow_frequency - 1);
1761 __ br(Assembler::EQ, backedge_counter_overflow);
1762
1763 }
1764 } else {
1765 if (UseOnStackReplacement) {
1766 // check for overflow against w0, which is the sum of the
1767 // counters
1768 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1769 __ cmpw(r0, rscratch1);
1770 __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1771 }
1772 }
1773 }
1774 }
1775 __ bind(dispatch);
1776
1777 // Pre-load the next target bytecode into rscratch1
1778 __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1779
1780 // continue with the bytecode @ target
1781 // rscratch1: target bytecode
1782 // rbcp: target bcp
1783 __ dispatch_only(vtos);
1784
1785 if (UseLoopCounter) {
1786 if (ProfileInterpreter) {
1787 // Out-of-line code to allocate method data oop.
1788 __ bind(profile_method);
1789 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1790 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
1791 __ set_method_data_pointer_for_bcp();
1792 __ b(dispatch);
1793 }
1794
1795 if (TieredCompilation || UseOnStackReplacement) {
1796 // invocation counter overflow
1797 __ bind(backedge_counter_overflow);
1798 __ neg(r2, r2);
1799 __ add(r2, r2, rbcp); // branch bcp
1800 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1801 __ call_VM(noreg,
1802 CAST_FROM_FN_PTR(address,
1803 InterpreterRuntime::frequency_counter_overflow),
1804 r2);
1805 if (!UseOnStackReplacement)
1806 __ b(dispatch);
1807 }
1808
1809 if (UseOnStackReplacement) {
1810 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
1811
1812 // r0: osr nmethod (osr ok) or NULL (osr not possible)
1813 // w1: target bytecode
1814 // r2: scratch
1815 __ cbz(r0, dispatch); // test result -- no osr if null
1816 // nmethod may have been invalidated (VM may block upon call_VM return)
1817 __ ldrb(r2, Address(r0, nmethod::state_offset()));
1818 if (nmethod::in_use != 0)
1819 __ sub(r2, r2, nmethod::in_use);
1820 __ cbnz(r2, dispatch);
1821
1822 // We have the address of an on stack replacement routine in r0
1823 // We need to prepare to execute the OSR method. First we must
1824 // migrate the locals and monitors off of the stack.
1825
1826 __ mov(r19, r0); // save the nmethod
1827
1828 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1829
1830 // r0 is OSR buffer, move it to expected parameter location
1831 __ mov(j_rarg0, r0);
1832
1833 // remove activation
1834 // get sender esp
1835 __ ldr(esp,
1836 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1837 // remove frame anchor
1838 __ leave();
1839 // Ensure compiled code always sees stack at proper alignment
1840 __ andr(sp, esp, -16);
1841
1842 // and begin the OSR nmethod
1843 __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1844 __ br(rscratch1);
1845 }
1846 }
1847 }
1848
1849
1850 void TemplateTable::if_0cmp(Condition cc)
1851 {
1852 transition(itos, vtos);
1853 // assume branch is more often taken than not (loops use backward branches)
1854 Label not_taken;
1855 if (cc == equal)
1856 __ cbnzw(r0, not_taken);
1857 else if (cc == not_equal)
1858 __ cbzw(r0, not_taken);
1859 else {
1860 __ andsw(zr, r0, r0);
1861 __ br(j_not(cc), not_taken);
1862 }
1863
1864 branch(false, false);
1865 __ bind(not_taken);
1866 __ profile_not_taken_branch(r0);
1867 }
1868
1869 void TemplateTable::if_icmp(Condition cc)
1870 {
1871 transition(itos, vtos);
1872 // assume branch is more often taken than not (loops use backward branches)
1873 Label not_taken;
1874 __ pop_i(r1);
1875 __ cmpw(r1, r0, Assembler::LSL);
1876 __ br(j_not(cc), not_taken);
1877 branch(false, false);
1878 __ bind(not_taken);
1879 __ profile_not_taken_branch(r0);
1880 }
1881
1882 void TemplateTable::if_nullcmp(Condition cc)
1883 {
1884 transition(atos, vtos);
1885 // assume branch is more often taken than not (loops use backward branches)
1886 Label not_taken;
1887 if (cc == equal)
1888 __ cbnz(r0, not_taken);
1889 else
1890 __ cbz(r0, not_taken);
1891 branch(false, false);
1892 __ bind(not_taken);
1893 __ profile_not_taken_branch(r0);
1894 }
1895
1896 void TemplateTable::if_acmp(Condition cc)
1897 {
1898 transition(atos, vtos);
1899 // assume branch is more often taken than not (loops use backward branches)
1900 Label not_taken;
1901 __ pop_ptr(r1);
1902 __ cmp(r1, r0);
1903 __ br(j_not(cc), not_taken);
1904 branch(false, false);
1905 __ bind(not_taken);
1906 __ profile_not_taken_branch(r0);
1907 }
1908
1909 void TemplateTable::ret() {
1910 transition(vtos, vtos);
1911 // We might be moving to a safepoint. The thread which calls
1912 // Interpreter::notice_safepoints() will effectively flush its cache
1913 // when it makes a system call, but we need to do something to
1914 // ensure that we see the changed dispatch table.
1915 __ membar(MacroAssembler::LoadLoad);
1916
1917 locals_index(r1);
1918 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1919 __ profile_ret(r1, r2);
1920 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1921 __ lea(rbcp, Address(rbcp, r1));
1922 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1923 __ dispatch_next(vtos);
1924 }
1925
1926 void TemplateTable::wide_ret() {
1927 transition(vtos, vtos);
1928 locals_index_wide(r1);
1929 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1930 __ profile_ret(r1, r2);
1931 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1932 __ lea(rbcp, Address(rbcp, r1));
1933 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1934 __ dispatch_next(vtos);
1935 }
1936
1937
1938 void TemplateTable::tableswitch() {
1939 Label default_case, continue_execution;
1940 transition(itos, vtos);
1941 // align rbcp
1942 __ lea(r1, at_bcp(BytesPerInt));
1943 __ andr(r1, r1, -BytesPerInt);
1944 // load lo & hi
1945 __ ldrw(r2, Address(r1, BytesPerInt));
1946 __ ldrw(r3, Address(r1, 2 * BytesPerInt));
1947 __ rev32(r2, r2);
1948 __ rev32(r3, r3);
1949 // check against lo & hi
1950 __ cmpw(r0, r2);
1951 __ br(Assembler::LT, default_case);
1952 __ cmpw(r0, r3);
1953 __ br(Assembler::GT, default_case);
1954 // lookup dispatch offset
1955 __ subw(r0, r0, r2);
1956 __ lea(r3, Address(r1, r0, Address::uxtw(2)));
1957 __ ldrw(r3, Address(r3, 3 * BytesPerInt));
1958 __ profile_switch_case(r0, r1, r2);
1959 // continue execution
1960 __ bind(continue_execution);
1961 __ rev32(r3, r3);
1962 __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
1963 __ add(rbcp, rbcp, r3, ext::sxtw);
1964 __ dispatch_only(vtos);
1965 // handle default
1966 __ bind(default_case);
1967 __ profile_switch_default(r0);
1968 __ ldrw(r3, Address(r1, 0));
1969 __ b(continue_execution);
1970 }
1971
1972 void TemplateTable::lookupswitch() {
1973 transition(itos, itos);
1974 __ stop("lookupswitch bytecode should have been rewritten");
1975 }
1976
1977 void TemplateTable::fast_linearswitch() {
1978 transition(itos, vtos);
1979 Label loop_entry, loop, found, continue_execution;
1980 // bswap r0 so we can avoid bswapping the table entries
1981 __ rev32(r0, r0);
1982 // align rbcp
1983 __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
1984 // this instruction (change offsets
1985 // below)
1986 __ andr(r19, r19, -BytesPerInt);
1987 // set counter
1988 __ ldrw(r1, Address(r19, BytesPerInt));
1989 __ rev32(r1, r1);
1990 __ b(loop_entry);
1991 // table search
1992 __ bind(loop);
1993 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
1994 __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
1995 __ cmpw(r0, rscratch1);
1996 __ br(Assembler::EQ, found);
1997 __ bind(loop_entry);
1998 __ subs(r1, r1, 1);
1999 __ br(Assembler::PL, loop);
2000 // default case
2001 __ profile_switch_default(r0);
2002 __ ldrw(r3, Address(r19, 0));
2003 __ b(continue_execution);
2004 // entry found -> get offset
2005 __ bind(found);
2006 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2007 __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2008 __ profile_switch_case(r1, r0, r19);
2009 // continue execution
2010 __ bind(continue_execution);
2011 __ rev32(r3, r3);
2012 __ add(rbcp, rbcp, r3, ext::sxtw);
2013 __ ldrb(rscratch1, Address(rbcp, 0));
2014 __ dispatch_only(vtos);
2015 }
2016
2017 void TemplateTable::fast_binaryswitch() {
2018 transition(itos, vtos);
2019 // Implementation using the following core algorithm:
2020 //
2021 // int binary_search(int key, LookupswitchPair* array, int n) {
2022 // // Binary search according to "Methodik des Programmierens" by
2023 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2024 // int i = 0;
2025 // int j = n;
2026 // while (i+1 < j) {
2027 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2028 // // with Q: for all i: 0 <= i < n: key < a[i]
2029 // // where a stands for the array and assuming that the (inexisting)
2030 // // element a[n] is infinitely big.
2031 // int h = (i + j) >> 1;
2032 // // i < h < j
2033 // if (key < array[h].fast_match()) {
2034 // j = h;
2035 // } else {
2036 // i = h;
2037 // }
2038 // }
2039 // // R: a[i] <= key < a[i+1] or Q
2040 // // (i.e., if key is within array, i is the correct index)
2041 // return i;
2042 // }
2043
2044 // Register allocation
2045 const Register key = r0; // already set (tosca)
2046 const Register array = r1;
2047 const Register i = r2;
2048 const Register j = r3;
2049 const Register h = rscratch1;
2050 const Register temp = rscratch2;
2051
2052 // Find array start
2053 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2054 // get rid of this
2055 // instruction (change
2056 // offsets below)
2057 __ andr(array, array, -BytesPerInt);
2058
2059 // Initialize i & j
2060 __ mov(i, 0); // i = 0;
2061 __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2062
2063 // Convert j into native byteordering
2064 __ rev32(j, j);
2065
2066 // And start
2067 Label entry;
2068 __ b(entry);
2069
2070 // binary search loop
2071 {
2072 Label loop;
2073 __ bind(loop);
2074 // int h = (i + j) >> 1;
2075 __ addw(h, i, j); // h = i + j;
2076 __ lsrw(h, h, 1); // h = (i + j) >> 1;
2077 // if (key < array[h].fast_match()) {
2078 // j = h;
2079 // } else {
2080 // i = h;
2081 // }
2082 // Convert array[h].match to native byte-ordering before compare
2083 __ ldr(temp, Address(array, h, Address::lsl(3)));
2084 __ rev32(temp, temp);
2085 __ cmpw(key, temp);
2086 // j = h if (key < array[h].fast_match())
2087 __ csel(j, h, j, Assembler::LT);
2088 // i = h if (key >= array[h].fast_match())
2089 __ csel(i, h, i, Assembler::GE);
2090 // while (i+1 < j)
2091 __ bind(entry);
2092 __ addw(h, i, 1); // i+1
2093 __ cmpw(h, j); // i+1 < j
2094 __ br(Assembler::LT, loop);
2095 }
2096
2097 // end of binary search, result index is i (must check again!)
2098 Label default_case;
2099 // Convert array[i].match to native byte-ordering before compare
2100 __ ldr(temp, Address(array, i, Address::lsl(3)));
2101 __ rev32(temp, temp);
2102 __ cmpw(key, temp);
2103 __ br(Assembler::NE, default_case);
2104
2105 // entry found -> j = offset
2106 __ add(j, array, i, ext::uxtx, 3);
2107 __ ldrw(j, Address(j, BytesPerInt));
2108 __ profile_switch_case(i, key, array);
2109 __ rev32(j, j);
2110 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2111 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2112 __ dispatch_only(vtos);
2113
2114 // default case -> j = default offset
2115 __ bind(default_case);
2116 __ profile_switch_default(i);
2117 __ ldrw(j, Address(array, -2 * BytesPerInt));
2118 __ rev32(j, j);
2119 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2120 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2121 __ dispatch_only(vtos);
2122 }
2123
2124
2125 void TemplateTable::_return(TosState state)
2126 {
2127 transition(state, state);
2128 assert(_desc->calls_vm(),
2129 "inconsistent calls_vm information"); // call in remove_activation
2130
2131 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2132 assert(state == vtos, "only valid state");
2133
2134 __ ldr(c_rarg1, aaddress(0));
2135 __ load_klass(r3, c_rarg1);
2136 __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2137 Label skip_register_finalizer;
2138 __ tbz(r3, exact_log2(JVM_ACC_HAS_FINALIZER), skip_register_finalizer);
2139
2140 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2141
2142 __ bind(skip_register_finalizer);
2143 }
2144
2145 // Issue a StoreStore barrier after all stores but before return
2146 // from any constructor for any class with a final field. We don't
2147 // know if this is a finalizer, so we always do so.
2148 if (_desc->bytecode() == Bytecodes::_return)
2149 __ membar(MacroAssembler::StoreStore);
2150
2151 // Narrow result if state is itos but result type is smaller.
2152 // Need to narrow in the return bytecode rather than in generate_return_entry
2153 // since compiled code callers expect the result to already be narrowed.
2154 if (state == itos) {
2155 __ narrow(r0);
2156 }
2157
2158 __ remove_activation(state);
2159 __ ret(lr);
2160 }
2161
2162 // ----------------------------------------------------------------------------
2163 // Volatile variables demand their effects be made known to all CPU's
2164 // in order. Store buffers on most chips allow reads & writes to
2165 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2166 // without some kind of memory barrier (i.e., it's not sufficient that
2167 // the interpreter does not reorder volatile references, the hardware
2168 // also must not reorder them).
2169 //
2170 // According to the new Java Memory Model (JMM):
2171 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2172 // writes act as aquire & release, so:
2173 // (2) A read cannot let unrelated NON-volatile memory refs that
2174 // happen after the read float up to before the read. It's OK for
2175 // non-volatile memory refs that happen before the volatile read to
2176 // float down below it.
2177 // (3) Similar a volatile write cannot let unrelated NON-volatile
2178 // memory refs that happen BEFORE the write float down to after the
2179 // write. It's OK for non-volatile memory refs that happen after the
2180 // volatile write to float up before it.
2181 //
2182 // We only put in barriers around volatile refs (they are expensive),
2183 // not _between_ memory refs (that would require us to track the
2184 // flavor of the previous memory refs). Requirements (2) and (3)
2185 // require some barriers before volatile stores and after volatile
2186 // loads. These nearly cover requirement (1) but miss the
2187 // volatile-store-volatile-load case. This final case is placed after
2188 // volatile-stores although it could just as well go before
2189 // volatile-loads.
2190
2191 void TemplateTable::resolve_cache_and_index(int byte_no,
2192 Register Rcache,
2193 Register index,
2194 size_t index_size) {
2195 const Register temp = r19;
2196 assert_different_registers(Rcache, index, temp);
2197
2198 Label resolved;
2199
2200 Bytecodes::Code code = bytecode();
2201 switch (code) {
2202 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2203 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2204 }
2205
2206 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2207 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2208 __ cmp(temp, (int) code); // have we resolved this bytecode?
2209 __ br(Assembler::EQ, resolved);
2210
2211 // resolve first time through
2212 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2213 __ mov(temp, (int) code);
2214 __ call_VM(noreg, entry, temp);
2215
2216 // Update registers with resolved info
2217 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2218 // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2219 // so all clients ofthis method must be modified accordingly
2220 __ bind(resolved);
2221 }
2222
2223 // The Rcache and index registers must be set before call
2224 // n.b unlike x86 cache already includes the index offset
2225 void TemplateTable::load_field_cp_cache_entry(Register obj,
2226 Register cache,
2227 Register index,
2228 Register off,
2229 Register flags,
2230 bool is_static = false) {
2231 assert_different_registers(cache, index, flags, off);
2232
2233 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2234 // Field offset
2235 __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2236 ConstantPoolCacheEntry::f2_offset())));
2237 // Flags
2238 __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2239 ConstantPoolCacheEntry::flags_offset())));
2240
2241 // klass overwrite register
2242 if (is_static) {
2243 __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2244 ConstantPoolCacheEntry::f1_offset())));
2245 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2246 __ ldr(obj, Address(obj, mirror_offset));
2247 }
2248 }
2249
2250 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2251 Register method,
2252 Register itable_index,
2253 Register flags,
2254 bool is_invokevirtual,
2255 bool is_invokevfinal, /*unused*/
2256 bool is_invokedynamic) {
2257 // setup registers
2258 const Register cache = rscratch2;
2259 const Register index = r4;
2260 assert_different_registers(method, flags);
2261 assert_different_registers(method, cache, index);
2262 assert_different_registers(itable_index, flags);
2263 assert_different_registers(itable_index, cache, index);
2264 // determine constant pool cache field offsets
2265 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2266 const int method_offset = in_bytes(
2267 ConstantPoolCache::base_offset() +
2268 (is_invokevirtual
2269 ? ConstantPoolCacheEntry::f2_offset()
2270 : ConstantPoolCacheEntry::f1_offset()));
2271 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2272 ConstantPoolCacheEntry::flags_offset());
2273 // access constant pool cache fields
2274 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2275 ConstantPoolCacheEntry::f2_offset());
2276
2277 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2278 resolve_cache_and_index(byte_no, cache, index, index_size);
2279 __ ldr(method, Address(cache, method_offset));
2280
2281 if (itable_index != noreg) {
2282 __ ldr(itable_index, Address(cache, index_offset));
2283 }
2284 __ ldrw(flags, Address(cache, flags_offset));
2285 }
2286
2287
2288 // The registers cache and index expected to be set before call.
2289 // Correct values of the cache and index registers are preserved.
2290 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2291 bool is_static, bool has_tos) {
2292 // do the JVMTI work here to avoid disturbing the register state below
2293 // We use c_rarg registers here because we want to use the register used in
2294 // the call to the VM
2295 if (JvmtiExport::can_post_field_access()) {
2296 // Check to see if a field access watch has been set before we
2297 // take the time to call into the VM.
2298 Label L1;
2299 assert_different_registers(cache, index, r0);
2300 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2301 __ ldrw(r0, Address(rscratch1));
2302 __ cbzw(r0, L1);
2303
2304 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2305 __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2306
2307 if (is_static) {
2308 __ mov(c_rarg1, zr); // NULL object reference
2309 } else {
2310 __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2311 __ verify_oop(c_rarg1);
2312 }
2313 // c_rarg1: object pointer or NULL
2314 // c_rarg2: cache entry pointer
2315 // c_rarg3: jvalue object on the stack
2316 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2317 InterpreterRuntime::post_field_access),
2318 c_rarg1, c_rarg2, c_rarg3);
2319 __ get_cache_and_index_at_bcp(cache, index, 1);
2320 __ bind(L1);
2321 }
2322 }
2323
2324 void TemplateTable::pop_and_check_object(Register r)
2325 {
2326 __ pop_ptr(r);
2327 __ null_check(r); // for field access must check obj.
2328 __ verify_oop(r);
2329 }
2330
2331 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2332 {
2333 const Register cache = r2;
2334 const Register index = r3;
2335 const Register obj = r4;
2336 const Register off = r19;
2337 const Register flags = r0;
2338 const Register raw_flags = r6;
2339 const Register bc = r4; // uses same reg as obj, so don't mix them
2340
2341 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2342 jvmti_post_field_access(cache, index, is_static, false);
2343 load_field_cp_cache_entry(obj, cache, index, off, raw_flags, is_static);
2344
2345 if (!is_static) {
2346 // obj is on the stack
2347 pop_and_check_object(obj);
2348 }
2349
2350 // 8179954: We need to make sure that the code generated for
2351 // volatile accesses forms a sequentially-consistent set of
2352 // operations when combined with STLR and LDAR. Without a leading
2353 // membar it's possible for a simple Dekker test to fail if loads
2354 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
2355 // the stores in one method and we interpret the loads in another.
2356 if (! UseBarriersForVolatile) {
2357 Label notVolatile;
2358 __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2359 __ membar(MacroAssembler::AnyAny);
2360 __ bind(notVolatile);
2361 }
2362
2363 const Address field(obj, off);
2364
2365 Label Done, notByte, notBool, notInt, notShort, notChar,
2366 notLong, notFloat, notObj, notDouble;
2367
2368 // x86 uses a shift and mask or wings it with a shift plus assert
2369 // the mask is not needed. aarch64 just uses bitfield extract
2370 __ ubfxw(flags, raw_flags, ConstantPoolCacheEntry::tos_state_shift,
2371 ConstantPoolCacheEntry::tos_state_bits);
2372
2373 assert(btos == 0, "change code, btos != 0");
2374 __ cbnz(flags, notByte);
2375
2376 // Don't rewrite getstatic, only getfield
2377 if (is_static) rc = may_not_rewrite;
2378
2379 // btos
2380 __ load_signed_byte(r0, field);
2381 __ push(btos);
2382 // Rewrite bytecode to be faster
2383 if (rc == may_rewrite) {
2384 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2385 }
2386 __ b(Done);
2387
2388 __ bind(notByte);
2389 __ cmp(flags, ztos);
2390 __ br(Assembler::NE, notBool);
2391
2392 // ztos (same code as btos)
2393 __ ldrsb(r0, field);
2394 __ push(ztos);
2395 // Rewrite bytecode to be faster
2396 if (rc == may_rewrite) {
2397 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2398 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2399 }
2400 __ b(Done);
2401
2402 __ bind(notBool);
2403 __ cmp(flags, atos);
2404 __ br(Assembler::NE, notObj);
2405 // atos
2406 do_oop_load(_masm, field, r0, ACCESS_IN_HEAP);
2407 __ push(atos);
2408 if (rc == may_rewrite) {
2409 patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2410 }
2411 __ b(Done);
2412
2413 __ bind(notObj);
2414 __ cmp(flags, itos);
2415 __ br(Assembler::NE, notInt);
2416 // itos
2417 __ ldrw(r0, field);
2418 __ push(itos);
2419 // Rewrite bytecode to be faster
2420 if (rc == may_rewrite) {
2421 patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2422 }
2423 __ b(Done);
2424
2425 __ bind(notInt);
2426 __ cmp(flags, ctos);
2427 __ br(Assembler::NE, notChar);
2428 // ctos
2429 __ load_unsigned_short(r0, field);
2430 __ push(ctos);
2431 // Rewrite bytecode to be faster
2432 if (rc == may_rewrite) {
2433 patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2434 }
2435 __ b(Done);
2436
2437 __ bind(notChar);
2438 __ cmp(flags, stos);
2439 __ br(Assembler::NE, notShort);
2440 // stos
2441 __ load_signed_short(r0, field);
2442 __ push(stos);
2443 // Rewrite bytecode to be faster
2444 if (rc == may_rewrite) {
2445 patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2446 }
2447 __ b(Done);
2448
2449 __ bind(notShort);
2450 __ cmp(flags, ltos);
2451 __ br(Assembler::NE, notLong);
2452 // ltos
2453 __ ldr(r0, field);
2454 __ push(ltos);
2455 // Rewrite bytecode to be faster
2456 if (rc == may_rewrite) {
2457 patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2458 }
2459 __ b(Done);
2460
2461 __ bind(notLong);
2462 __ cmp(flags, ftos);
2463 __ br(Assembler::NE, notFloat);
2464 // ftos
2465 __ ldrs(v0, field);
2466 __ push(ftos);
2467 // Rewrite bytecode to be faster
2468 if (rc == may_rewrite) {
2469 patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2470 }
2471 __ b(Done);
2472
2473 __ bind(notFloat);
2474 #ifdef ASSERT
2475 __ cmp(flags, dtos);
2476 __ br(Assembler::NE, notDouble);
2477 #endif
2478 // dtos
2479 __ ldrd(v0, field);
2480 __ push(dtos);
2481 // Rewrite bytecode to be faster
2482 if (rc == may_rewrite) {
2483 patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2484 }
2485 #ifdef ASSERT
2486 __ b(Done);
2487
2488 __ bind(notDouble);
2489 __ stop("Bad state");
2490 #endif
2491
2492 __ bind(Done);
2493
2494 Label notVolatile;
2495 __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2496 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2497 __ bind(notVolatile);
2498 }
2499
2500
2501 void TemplateTable::getfield(int byte_no)
2502 {
2503 getfield_or_static(byte_no, false);
2504 }
2505
2506 void TemplateTable::nofast_getfield(int byte_no) {
2507 getfield_or_static(byte_no, false, may_not_rewrite);
2508 }
2509
2510 void TemplateTable::getstatic(int byte_no)
2511 {
2512 getfield_or_static(byte_no, true);
2513 }
2514
2515 // The registers cache and index expected to be set before call.
2516 // The function may destroy various registers, just not the cache and index registers.
2517 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2518 transition(vtos, vtos);
2519
2520 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2521
2522 if (JvmtiExport::can_post_field_modification()) {
2523 // Check to see if a field modification watch has been set before
2524 // we take the time to call into the VM.
2525 Label L1;
2526 assert_different_registers(cache, index, r0);
2527 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2528 __ ldrw(r0, Address(rscratch1));
2529 __ cbz(r0, L1);
2530
2531 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2532
2533 if (is_static) {
2534 // Life is simple. Null out the object pointer.
2535 __ mov(c_rarg1, zr);
2536 } else {
2537 // Life is harder. The stack holds the value on top, followed by
2538 // the object. We don't know the size of the value, though; it
2539 // could be one or two words depending on its type. As a result,
2540 // we must find the type to determine where the object is.
2541 __ ldrw(c_rarg3, Address(c_rarg2,
2542 in_bytes(cp_base_offset +
2543 ConstantPoolCacheEntry::flags_offset())));
2544 __ lsr(c_rarg3, c_rarg3,
2545 ConstantPoolCacheEntry::tos_state_shift);
2546 ConstantPoolCacheEntry::verify_tos_state_shift();
2547 Label nope2, done, ok;
2548 __ ldr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2549 __ cmpw(c_rarg3, ltos);
2550 __ br(Assembler::EQ, ok);
2551 __ cmpw(c_rarg3, dtos);
2552 __ br(Assembler::NE, nope2);
2553 __ bind(ok);
2554 __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2555 __ bind(nope2);
2556 }
2557 // cache entry pointer
2558 __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2559 // object (tos)
2560 __ mov(c_rarg3, esp);
2561 // c_rarg1: object pointer set up above (NULL if static)
2562 // c_rarg2: cache entry pointer
2563 // c_rarg3: jvalue object on the stack
2564 __ call_VM(noreg,
2565 CAST_FROM_FN_PTR(address,
2566 InterpreterRuntime::post_field_modification),
2567 c_rarg1, c_rarg2, c_rarg3);
2568 __ get_cache_and_index_at_bcp(cache, index, 1);
2569 __ bind(L1);
2570 }
2571 }
2572
2573 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2574 transition(vtos, vtos);
2575
2576 const Register cache = r2;
2577 const Register index = r3;
2578 const Register obj = r2;
2579 const Register off = r19;
2580 const Register flags = r0;
2581 const Register bc = r4;
2582
2583 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2584 jvmti_post_field_mod(cache, index, is_static);
2585 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2586
2587 Label Done;
2588 __ mov(r5, flags);
2589
2590 {
2591 Label notVolatile;
2592 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2593 __ membar(MacroAssembler::StoreStore);
2594 __ bind(notVolatile);
2595 }
2596
2597 // field address
2598 const Address field(obj, off);
2599
2600 Label notByte, notBool, notInt, notShort, notChar,
2601 notLong, notFloat, notObj, notDouble;
2602
2603 // x86 uses a shift and mask or wings it with a shift plus assert
2604 // the mask is not needed. aarch64 just uses bitfield extract
2605 __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
2606
2607 assert(btos == 0, "change code, btos != 0");
2608 __ cbnz(flags, notByte);
2609
2610 // Don't rewrite putstatic, only putfield
2611 if (is_static) rc = may_not_rewrite;
2612
2613 // btos
2614 {
2615 __ pop(btos);
2616 if (!is_static) pop_and_check_object(obj);
2617 __ strb(r0, field);
2618 if (rc == may_rewrite) {
2619 patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2620 }
2621 __ b(Done);
2622 }
2623
2624 __ bind(notByte);
2625 __ cmp(flags, ztos);
2626 __ br(Assembler::NE, notBool);
2627
2628 // ztos
2629 {
2630 __ pop(ztos);
2631 if (!is_static) pop_and_check_object(obj);
2632 __ andw(r0, r0, 0x1);
2633 __ strb(r0, field);
2634 if (rc == may_rewrite) {
2635 patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2636 }
2637 __ b(Done);
2638 }
2639
2640 __ bind(notBool);
2641 __ cmp(flags, atos);
2642 __ br(Assembler::NE, notObj);
2643
2644 // atos
2645 {
2646 __ pop(atos);
2647 if (!is_static) pop_and_check_object(obj);
2648 // Store into the field
2649 do_oop_store(_masm, field, r0, ACCESS_IN_HEAP);
2650 if (rc == may_rewrite) {
2651 patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2652 }
2653 __ b(Done);
2654 }
2655
2656 __ bind(notObj);
2657 __ cmp(flags, itos);
2658 __ br(Assembler::NE, notInt);
2659
2660 // itos
2661 {
2662 __ pop(itos);
2663 if (!is_static) pop_and_check_object(obj);
2664 __ strw(r0, field);
2665 if (rc == may_rewrite) {
2666 patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2667 }
2668 __ b(Done);
2669 }
2670
2671 __ bind(notInt);
2672 __ cmp(flags, ctos);
2673 __ br(Assembler::NE, notChar);
2674
2675 // ctos
2676 {
2677 __ pop(ctos);
2678 if (!is_static) pop_and_check_object(obj);
2679 __ strh(r0, field);
2680 if (rc == may_rewrite) {
2681 patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2682 }
2683 __ b(Done);
2684 }
2685
2686 __ bind(notChar);
2687 __ cmp(flags, stos);
2688 __ br(Assembler::NE, notShort);
2689
2690 // stos
2691 {
2692 __ pop(stos);
2693 if (!is_static) pop_and_check_object(obj);
2694 __ strh(r0, field);
2695 if (rc == may_rewrite) {
2696 patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2697 }
2698 __ b(Done);
2699 }
2700
2701 __ bind(notShort);
2702 __ cmp(flags, ltos);
2703 __ br(Assembler::NE, notLong);
2704
2705 // ltos
2706 {
2707 __ pop(ltos);
2708 if (!is_static) pop_and_check_object(obj);
2709 __ str(r0, field);
2710 if (rc == may_rewrite) {
2711 patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2712 }
2713 __ b(Done);
2714 }
2715
2716 __ bind(notLong);
2717 __ cmp(flags, ftos);
2718 __ br(Assembler::NE, notFloat);
2719
2720 // ftos
2721 {
2722 __ pop(ftos);
2723 if (!is_static) pop_and_check_object(obj);
2724 __ strs(v0, field);
2725 if (rc == may_rewrite) {
2726 patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2727 }
2728 __ b(Done);
2729 }
2730
2731 __ bind(notFloat);
2732 #ifdef ASSERT
2733 __ cmp(flags, dtos);
2734 __ br(Assembler::NE, notDouble);
2735 #endif
2736
2737 // dtos
2738 {
2739 __ pop(dtos);
2740 if (!is_static) pop_and_check_object(obj);
2741 __ strd(v0, field);
2742 if (rc == may_rewrite) {
2743 patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2744 }
2745 }
2746
2747 #ifdef ASSERT
2748 __ b(Done);
2749
2750 __ bind(notDouble);
2751 __ stop("Bad state");
2752 #endif
2753
2754 __ bind(Done);
2755
2756 {
2757 Label notVolatile;
2758 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2759 __ membar(MacroAssembler::StoreLoad);
2760 __ bind(notVolatile);
2761 }
2762 }
2763
2764 void TemplateTable::putfield(int byte_no)
2765 {
2766 putfield_or_static(byte_no, false);
2767 }
2768
2769 void TemplateTable::nofast_putfield(int byte_no) {
2770 putfield_or_static(byte_no, false, may_not_rewrite);
2771 }
2772
2773 void TemplateTable::putstatic(int byte_no) {
2774 putfield_or_static(byte_no, true);
2775 }
2776
2777 void TemplateTable::jvmti_post_fast_field_mod()
2778 {
2779 if (JvmtiExport::can_post_field_modification()) {
2780 // Check to see if a field modification watch has been set before
2781 // we take the time to call into the VM.
2782 Label L2;
2783 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2784 __ ldrw(c_rarg3, Address(rscratch1));
2785 __ cbzw(c_rarg3, L2);
2786 __ pop_ptr(r19); // copy the object pointer from tos
2787 __ verify_oop(r19);
2788 __ push_ptr(r19); // put the object pointer back on tos
2789 // Save tos values before call_VM() clobbers them. Since we have
2790 // to do it for every data type, we use the saved values as the
2791 // jvalue object.
2792 switch (bytecode()) { // load values into the jvalue object
2793 case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2794 case Bytecodes::_fast_bputfield: // fall through
2795 case Bytecodes::_fast_zputfield: // fall through
2796 case Bytecodes::_fast_sputfield: // fall through
2797 case Bytecodes::_fast_cputfield: // fall through
2798 case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2799 case Bytecodes::_fast_dputfield: __ push_d(); break;
2800 case Bytecodes::_fast_fputfield: __ push_f(); break;
2801 case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2802
2803 default:
2804 ShouldNotReachHere();
2805 }
2806 __ mov(c_rarg3, esp); // points to jvalue on the stack
2807 // access constant pool cache entry
2808 __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2809 __ verify_oop(r19);
2810 // r19: object pointer copied above
2811 // c_rarg2: cache entry pointer
2812 // c_rarg3: jvalue object on the stack
2813 __ call_VM(noreg,
2814 CAST_FROM_FN_PTR(address,
2815 InterpreterRuntime::post_field_modification),
2816 r19, c_rarg2, c_rarg3);
2817
2818 switch (bytecode()) { // restore tos values
2819 case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2820 case Bytecodes::_fast_bputfield: // fall through
2821 case Bytecodes::_fast_zputfield: // fall through
2822 case Bytecodes::_fast_sputfield: // fall through
2823 case Bytecodes::_fast_cputfield: // fall through
2824 case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2825 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2826 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2827 case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2828 }
2829 __ bind(L2);
2830 }
2831 }
2832
2833 void TemplateTable::fast_storefield(TosState state)
2834 {
2835 transition(state, vtos);
2836
2837 ByteSize base = ConstantPoolCache::base_offset();
2838
2839 jvmti_post_fast_field_mod();
2840
2841 // access constant pool cache
2842 __ get_cache_and_index_at_bcp(r2, r1, 1);
2843
2844 // test for volatile with r3
2845 __ ldrw(r3, Address(r2, in_bytes(base +
2846 ConstantPoolCacheEntry::flags_offset())));
2847
2848 // replace index with field offset from cache entry
2849 __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2850
2851 {
2852 Label notVolatile;
2853 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2854 __ membar(MacroAssembler::StoreStore);
2855 __ bind(notVolatile);
2856 }
2857
2858 Label notVolatile;
2859
2860 // Get object from stack
2861 pop_and_check_object(r2);
2862
2863 // field address
2864 const Address field(r2, r1);
2865
2866 // access field
2867 switch (bytecode()) {
2868 case Bytecodes::_fast_aputfield:
2869 do_oop_store(_masm, field, r0, ACCESS_IN_HEAP);
2870 break;
2871 case Bytecodes::_fast_lputfield:
2872 __ str(r0, field);
2873 break;
2874 case Bytecodes::_fast_iputfield:
2875 __ strw(r0, field);
2876 break;
2877 case Bytecodes::_fast_zputfield:
2878 __ andw(r0, r0, 0x1); // boolean is true if LSB is 1
2879 // fall through to bputfield
2880 case Bytecodes::_fast_bputfield:
2881 __ strb(r0, field);
2882 break;
2883 case Bytecodes::_fast_sputfield:
2884 // fall through
2885 case Bytecodes::_fast_cputfield:
2886 __ strh(r0, field);
2887 break;
2888 case Bytecodes::_fast_fputfield:
2889 __ strs(v0, field);
2890 break;
2891 case Bytecodes::_fast_dputfield:
2892 __ strd(v0, field);
2893 break;
2894 default:
2895 ShouldNotReachHere();
2896 }
2897
2898 {
2899 Label notVolatile;
2900 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2901 __ membar(MacroAssembler::StoreLoad);
2902 __ bind(notVolatile);
2903 }
2904 }
2905
2906
2907 void TemplateTable::fast_accessfield(TosState state)
2908 {
2909 transition(atos, state);
2910 // Do the JVMTI work here to avoid disturbing the register state below
2911 if (JvmtiExport::can_post_field_access()) {
2912 // Check to see if a field access watch has been set before we
2913 // take the time to call into the VM.
2914 Label L1;
2915 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2916 __ ldrw(r2, Address(rscratch1));
2917 __ cbzw(r2, L1);
2918 // access constant pool cache entry
2919 __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
2920 __ verify_oop(r0);
2921 __ push_ptr(r0); // save object pointer before call_VM() clobbers it
2922 __ mov(c_rarg1, r0);
2923 // c_rarg1: object pointer copied above
2924 // c_rarg2: cache entry pointer
2925 __ call_VM(noreg,
2926 CAST_FROM_FN_PTR(address,
2927 InterpreterRuntime::post_field_access),
2928 c_rarg1, c_rarg2);
2929 __ pop_ptr(r0); // restore object pointer
2930 __ bind(L1);
2931 }
2932
2933 // access constant pool cache
2934 __ get_cache_and_index_at_bcp(r2, r1, 1);
2935 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2936 ConstantPoolCacheEntry::f2_offset())));
2937 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2938 ConstantPoolCacheEntry::flags_offset())));
2939
2940 // r0: object
2941 __ verify_oop(r0);
2942 __ null_check(r0);
2943 const Address field(r0, r1);
2944
2945 // 8179954: We need to make sure that the code generated for
2946 // volatile accesses forms a sequentially-consistent set of
2947 // operations when combined with STLR and LDAR. Without a leading
2948 // membar it's possible for a simple Dekker test to fail if loads
2949 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
2950 // the stores in one method and we interpret the loads in another.
2951 if (! UseBarriersForVolatile) {
2952 Label notVolatile;
2953 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2954 __ membar(MacroAssembler::AnyAny);
2955 __ bind(notVolatile);
2956 }
2957
2958 // access field
2959 switch (bytecode()) {
2960 case Bytecodes::_fast_agetfield:
2961 do_oop_load(_masm, field, r0, ACCESS_IN_HEAP);
2962 __ verify_oop(r0);
2963 break;
2964 case Bytecodes::_fast_lgetfield:
2965 __ ldr(r0, field);
2966 break;
2967 case Bytecodes::_fast_igetfield:
2968 __ ldrw(r0, field);
2969 break;
2970 case Bytecodes::_fast_bgetfield:
2971 __ load_signed_byte(r0, field);
2972 break;
2973 case Bytecodes::_fast_sgetfield:
2974 __ load_signed_short(r0, field);
2975 break;
2976 case Bytecodes::_fast_cgetfield:
2977 __ load_unsigned_short(r0, field);
2978 break;
2979 case Bytecodes::_fast_fgetfield:
2980 __ ldrs(v0, field);
2981 break;
2982 case Bytecodes::_fast_dgetfield:
2983 __ ldrd(v0, field);
2984 break;
2985 default:
2986 ShouldNotReachHere();
2987 }
2988 {
2989 Label notVolatile;
2990 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2991 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2992 __ bind(notVolatile);
2993 }
2994 }
2995
2996 void TemplateTable::fast_xaccess(TosState state)
2997 {
2998 transition(vtos, state);
2999
3000 // get receiver
3001 __ ldr(r0, aaddress(0));
3002 // access constant pool cache
3003 __ get_cache_and_index_at_bcp(r2, r3, 2);
3004 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3005 ConstantPoolCacheEntry::f2_offset())));
3006
3007 // 8179954: We need to make sure that the code generated for
3008 // volatile accesses forms a sequentially-consistent set of
3009 // operations when combined with STLR and LDAR. Without a leading
3010 // membar it's possible for a simple Dekker test to fail if loads
3011 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
3012 // the stores in one method and we interpret the loads in another.
3013 if (! UseBarriersForVolatile) {
3014 Label notVolatile;
3015 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3016 ConstantPoolCacheEntry::flags_offset())));
3017 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3018 __ membar(MacroAssembler::AnyAny);
3019 __ bind(notVolatile);
3020 }
3021
3022 // make sure exception is reported in correct bcp range (getfield is
3023 // next instruction)
3024 __ increment(rbcp);
3025 __ null_check(r0);
3026 switch (state) {
3027 case itos:
3028 __ ldrw(r0, Address(r0, r1, Address::lsl(0)));
3029 break;
3030 case atos:
3031 do_oop_load(_masm, Address(r0, r1, Address::lsl(0)), r0, ACCESS_IN_HEAP);
3032 __ verify_oop(r0);
3033 break;
3034 case ftos:
3035 __ ldrs(v0, Address(r0, r1, Address::lsl(0)));
3036 break;
3037 default:
3038 ShouldNotReachHere();
3039 }
3040
3041 {
3042 Label notVolatile;
3043 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3044 ConstantPoolCacheEntry::flags_offset())));
3045 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3046 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3047 __ bind(notVolatile);
3048 }
3049
3050 __ decrement(rbcp);
3051 }
3052
3053
3054
3055 //-----------------------------------------------------------------------------
3056 // Calls
3057
3058 void TemplateTable::count_calls(Register method, Register temp)
3059 {
3060 __ call_Unimplemented();
3061 }
3062
3063 void TemplateTable::prepare_invoke(int byte_no,
3064 Register method, // linked method (or i-klass)
3065 Register index, // itable index, MethodType, etc.
3066 Register recv, // if caller wants to see it
3067 Register flags // if caller wants to test it
3068 ) {
3069 // determine flags
3070 Bytecodes::Code code = bytecode();
3071 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3072 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3073 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3074 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3075 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3076 const bool load_receiver = (recv != noreg);
3077 const bool save_flags = (flags != noreg);
3078 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3079 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3080 assert(flags == noreg || flags == r3, "");
3081 assert(recv == noreg || recv == r2, "");
3082
3083 // setup registers & access constant pool cache
3084 if (recv == noreg) recv = r2;
3085 if (flags == noreg) flags = r3;
3086 assert_different_registers(method, index, recv, flags);
3087
3088 // save 'interpreter return address'
3089 __ save_bcp();
3090
3091 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3092
3093 // maybe push appendix to arguments (just before return address)
3094 if (is_invokedynamic || is_invokehandle) {
3095 Label L_no_push;
3096 __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3097 // Push the appendix as a trailing parameter.
3098 // This must be done before we get the receiver,
3099 // since the parameter_size includes it.
3100 __ push(r19);
3101 __ mov(r19, index);
3102 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3103 __ load_resolved_reference_at_index(index, r19);
3104 __ pop(r19);
3105 __ push(index); // push appendix (MethodType, CallSite, etc.)
3106 __ bind(L_no_push);
3107 }
3108
3109 // load receiver if needed (note: no return address pushed yet)
3110 if (load_receiver) {
3111 __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3112 // FIXME -- is this actually correct? looks like it should be 2
3113 // const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3114 // const int receiver_is_at_end = -1; // back off one slot to get receiver
3115 // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3116 // __ movptr(recv, recv_addr);
3117 __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3118 __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3119 __ verify_oop(recv);
3120 }
3121
3122 // compute return type
3123 // x86 uses a shift and mask or wings it with a shift plus assert
3124 // the mask is not needed. aarch64 just uses bitfield extract
3125 __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
3126 // load return address
3127 {
3128 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3129 __ mov(rscratch1, table_addr);
3130 __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3131 }
3132 }
3133
3134
3135 void TemplateTable::invokevirtual_helper(Register index,
3136 Register recv,
3137 Register flags)
3138 {
3139 // Uses temporary registers r0, r3
3140 assert_different_registers(index, recv, r0, r3);
3141 // Test for an invoke of a final method
3142 Label notFinal;
3143 __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3144
3145 const Register method = index; // method must be rmethod
3146 assert(method == rmethod,
3147 "methodOop must be rmethod for interpreter calling convention");
3148
3149 // do the call - the index is actually the method to call
3150 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3151
3152 // It's final, need a null check here!
3153 __ null_check(recv);
3154
3155 // profile this call
3156 __ profile_final_call(r0);
3157 __ profile_arguments_type(r0, method, r4, true);
3158
3159 __ jump_from_interpreted(method, r0);
3160
3161 __ bind(notFinal);
3162
3163 // get receiver klass
3164 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3165 __ load_klass(r0, recv);
3166
3167 // profile this call
3168 __ profile_virtual_call(r0, rlocals, r3);
3169
3170 // get target methodOop & entry point
3171 __ lookup_virtual_method(r0, index, method);
3172 __ profile_arguments_type(r3, method, r4, true);
3173 // FIXME -- this looks completely redundant. is it?
3174 // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3175 __ jump_from_interpreted(method, r3);
3176 }
3177
3178 void TemplateTable::invokevirtual(int byte_no)
3179 {
3180 transition(vtos, vtos);
3181 assert(byte_no == f2_byte, "use this argument");
3182
3183 prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3184
3185 // rmethod: index (actually a Method*)
3186 // r2: receiver
3187 // r3: flags
3188
3189 invokevirtual_helper(rmethod, r2, r3);
3190 }
3191
3192 void TemplateTable::invokespecial(int byte_no)
3193 {
3194 transition(vtos, vtos);
3195 assert(byte_no == f1_byte, "use this argument");
3196
3197 prepare_invoke(byte_no, rmethod, noreg, // get f1 Method*
3198 r2); // get receiver also for null check
3199 __ verify_oop(r2);
3200 __ null_check(r2);
3201 // do the call
3202 __ profile_call(r0);
3203 __ profile_arguments_type(r0, rmethod, rbcp, false);
3204 __ jump_from_interpreted(rmethod, r0);
3205 }
3206
3207 void TemplateTable::invokestatic(int byte_no)
3208 {
3209 transition(vtos, vtos);
3210 assert(byte_no == f1_byte, "use this argument");
3211
3212 prepare_invoke(byte_no, rmethod); // get f1 Method*
3213 // do the call
3214 __ profile_call(r0);
3215 __ profile_arguments_type(r0, rmethod, r4, false);
3216 __ jump_from_interpreted(rmethod, r0);
3217 }
3218
3219 void TemplateTable::fast_invokevfinal(int byte_no)
3220 {
3221 __ call_Unimplemented();
3222 }
3223
3224 void TemplateTable::invokeinterface(int byte_no) {
3225 transition(vtos, vtos);
3226 assert(byte_no == f1_byte, "use this argument");
3227
3228 prepare_invoke(byte_no, r0, rmethod, // get f1 Klass*, f2 itable index
3229 r2, r3); // recv, flags
3230
3231 // r0: interface klass (from f1)
3232 // rmethod: itable index (from f2)
3233 // r2: receiver
3234 // r3: flags
3235
3236 // Special case of invokeinterface called for virtual method of
3237 // java.lang.Object. See cpCacheOop.cpp for details.
3238 // This code isn't produced by javac, but could be produced by
3239 // another compliant java compiler.
3240 Label notMethod;
3241 __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notMethod);
3242
3243 invokevirtual_helper(rmethod, r2, r3);
3244 __ bind(notMethod);
3245
3246 // Get receiver klass into r3 - also a null check
3247 __ restore_locals();
3248 __ null_check(r2, oopDesc::klass_offset_in_bytes());
3249 __ load_klass(r3, r2);
3250
3251 // profile this call
3252 __ profile_virtual_call(r3, r13, r19);
3253
3254 Label no_such_interface, no_such_method;
3255
3256 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3257 r3, r0, rmethod,
3258 // outputs: method, scan temp. reg
3259 rmethod, r13,
3260 no_such_interface);
3261
3262 // rmethod,: methodOop to call
3263 // r2: receiver
3264 // Check for abstract method error
3265 // Note: This should be done more efficiently via a throw_abstract_method_error
3266 // interpreter entry point and a conditional jump to it in case of a null
3267 // method.
3268 __ cbz(rmethod, no_such_method);
3269
3270 __ profile_arguments_type(r3, rmethod, r13, true);
3271
3272 // do the call
3273 // r2: receiver
3274 // rmethod,: methodOop
3275 __ jump_from_interpreted(rmethod, r3);
3276 __ should_not_reach_here();
3277
3278 // exception handling code follows...
3279 // note: must restore interpreter registers to canonical
3280 // state for exception handling to work correctly!
3281
3282 __ bind(no_such_method);
3283 // throw exception
3284 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3285 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3286 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3287 // the call_VM checks for exception, so we should never return here.
3288 __ should_not_reach_here();
3289
3290 __ bind(no_such_interface);
3291 // throw exception
3292 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3293 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3294 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3295 InterpreterRuntime::throw_IncompatibleClassChangeError));
3296 // the call_VM checks for exception, so we should never return here.
3297 __ should_not_reach_here();
3298 return;
3299 }
3300
3301 void TemplateTable::invokehandle(int byte_no) {
3302 transition(vtos, vtos);
3303 assert(byte_no == f1_byte, "use this argument");
3304
3305 prepare_invoke(byte_no, rmethod, r0, r2);
3306 __ verify_method_ptr(r2);
3307 __ verify_oop(r2);
3308 __ null_check(r2);
3309
3310 // FIXME: profile the LambdaForm also
3311
3312 // r13 is safe to use here as a scratch reg because it is about to
3313 // be clobbered by jump_from_interpreted().
3314 __ profile_final_call(r13);
3315 __ profile_arguments_type(r13, rmethod, r4, true);
3316
3317 __ jump_from_interpreted(rmethod, r0);
3318 }
3319
3320 void TemplateTable::invokedynamic(int byte_no) {
3321 transition(vtos, vtos);
3322 assert(byte_no == f1_byte, "use this argument");
3323
3324 prepare_invoke(byte_no, rmethod, r0);
3325
3326 // r0: CallSite object (from cpool->resolved_references[])
3327 // rmethod: MH.linkToCallSite method (from f2)
3328
3329 // Note: r0_callsite is already pushed by prepare_invoke
3330
3331 // %%% should make a type profile for any invokedynamic that takes a ref argument
3332 // profile this call
3333 __ profile_call(rbcp);
3334 __ profile_arguments_type(r3, rmethod, r13, false);
3335
3336 __ verify_oop(r0);
3337
3338 __ jump_from_interpreted(rmethod, r0);
3339 }
3340
3341
3342 //-----------------------------------------------------------------------------
3343 // Allocation
3344
3345 void TemplateTable::_new() {
3346 transition(vtos, atos);
3347
3348 __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3349 Label slow_case;
3350 Label done;
3351 Label initialize_header;
3352 Label initialize_object; // including clearing the fields
3353 Label allocate_shared;
3354
3355 __ get_cpool_and_tags(r4, r0);
3356 // Make sure the class we're about to instantiate has been resolved.
3357 // This is done before loading InstanceKlass to be consistent with the order
3358 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3359 const int tags_offset = Array<u1>::base_offset_in_bytes();
3360 __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3361 __ lea(rscratch1, Address(rscratch1, tags_offset));
3362 __ ldarb(rscratch1, rscratch1);
3363 __ cmp(rscratch1, JVM_CONSTANT_Class);
3364 __ br(Assembler::NE, slow_case);
3365
3366 // get InstanceKlass
3367 __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
3368
3369 // make sure klass is initialized & doesn't have finalizer
3370 // make sure klass is fully initialized
3371 __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3372 __ cmp(rscratch1, InstanceKlass::fully_initialized);
3373 __ br(Assembler::NE, slow_case);
3374
3375 // get instance_size in InstanceKlass (scaled to a count of bytes)
3376 __ ldrw(r3,
3377 Address(r4,
3378 Klass::layout_helper_offset()));
3379 // test to see if it has a finalizer or is malformed in some way
3380 __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3381
3382 // Allocate the instance
3383 // 1) Try to allocate in the TLAB
3384 // 2) if fail and the object is large allocate in the shared Eden
3385 // 3) if the above fails (or is not applicable), go to a slow case
3386 // (creates a new TLAB, etc.)
3387
3388 const bool allow_shared_alloc =
3389 Universe::heap()->supports_inline_contig_alloc();
3390
3391 if (UseTLAB) {
3392 __ tlab_allocate(r0, r3, 0, noreg, r1,
3393 allow_shared_alloc ? allocate_shared : slow_case);
3394
3395 if (ZeroTLAB) {
3396 // the fields have been already cleared
3397 __ b(initialize_header);
3398 } else {
3399 // initialize both the header and fields
3400 __ b(initialize_object);
3401 }
3402 }
3403
3404 // Allocation in the shared Eden, if allowed.
3405 //
3406 // r3: instance size in bytes
3407 if (allow_shared_alloc) {
3408 __ bind(allocate_shared);
3409
3410 __ eden_allocate(r0, r3, 0, r10, slow_case);
3411 __ incr_allocated_bytes(rthread, r3, 0, rscratch1);
3412 }
3413
3414 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3415 // The object is initialized before the header. If the object size is
3416 // zero, go directly to the header initialization.
3417 __ bind(initialize_object);
3418 __ sub(r3, r3, sizeof(oopDesc));
3419 __ cbz(r3, initialize_header);
3420
3421 // Initialize object fields
3422 {
3423 __ add(r2, r0, sizeof(oopDesc));
3424 Label loop;
3425 __ bind(loop);
3426 __ str(zr, Address(__ post(r2, BytesPerLong)));
3427 __ sub(r3, r3, BytesPerLong);
3428 __ cbnz(r3, loop);
3429 }
3430
3431 // initialize object header only.
3432 __ bind(initialize_header);
3433 if (UseBiasedLocking) {
3434 __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3435 } else {
3436 __ mov(rscratch1, (intptr_t)markOopDesc::prototype());
3437 }
3438 __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3439 __ store_klass_gap(r0, zr); // zero klass gap for compressed oops
3440 __ store_klass(r0, r4); // store klass last
3441
3442 {
3443 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3444 // Trigger dtrace event for fastpath
3445 __ push(atos); // save the return value
3446 __ call_VM_leaf(
3447 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3448 __ pop(atos); // restore the return value
3449
3450 }
3451 __ b(done);
3452 }
3453
3454 // slow case
3455 __ bind(slow_case);
3456 __ get_constant_pool(c_rarg1);
3457 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3458 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3459 __ verify_oop(r0);
3460
3461 // continue
3462 __ bind(done);
3463 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3464 __ membar(Assembler::StoreStore);
3465 }
3466
3467 void TemplateTable::newarray() {
3468 transition(itos, atos);
3469 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3470 __ mov(c_rarg2, r0);
3471 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3472 c_rarg1, c_rarg2);
3473 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3474 __ membar(Assembler::StoreStore);
3475 }
3476
3477 void TemplateTable::anewarray() {
3478 transition(itos, atos);
3479 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3480 __ get_constant_pool(c_rarg1);
3481 __ mov(c_rarg3, r0);
3482 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3483 c_rarg1, c_rarg2, c_rarg3);
3484 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3485 __ membar(Assembler::StoreStore);
3486 }
3487
3488 void TemplateTable::arraylength() {
3489 transition(atos, itos);
3490 __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3491 __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3492 }
3493
3494 void TemplateTable::checkcast()
3495 {
3496 transition(atos, atos);
3497 Label done, is_null, ok_is_subtype, quicked, resolved;
3498 __ cbz(r0, is_null);
3499
3500 // Get cpool & tags index
3501 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3502 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3503 // See if bytecode has already been quicked
3504 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3505 __ lea(r1, Address(rscratch1, r19));
3506 __ ldarb(r1, r1);
3507 __ cmp(r1, JVM_CONSTANT_Class);
3508 __ br(Assembler::EQ, quicked);
3509
3510 __ push(atos); // save receiver for result, and for GC
3511 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3512 // vm_result_2 has metadata result
3513 __ get_vm_result_2(r0, rthread);
3514 __ pop(r3); // restore receiver
3515 __ b(resolved);
3516
3517 // Get superklass in r0 and subklass in r3
3518 __ bind(quicked);
3519 __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3520 __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass
3521
3522 __ bind(resolved);
3523 __ load_klass(r19, r3);
3524
3525 // Generate subtype check. Blows r2, r5. Object in r3.
3526 // Superklass in r0. Subklass in r19.
3527 __ gen_subtype_check(r19, ok_is_subtype);
3528
3529 // Come here on failure
3530 __ push(r3);
3531 // object is at TOS
3532 __ b(Interpreter::_throw_ClassCastException_entry);
3533
3534 // Come here on success
3535 __ bind(ok_is_subtype);
3536 __ mov(r0, r3); // Restore object in r3
3537
3538 // Collect counts on whether this test sees NULLs a lot or not.
3539 if (ProfileInterpreter) {
3540 __ b(done);
3541 __ bind(is_null);
3542 __ profile_null_seen(r2);
3543 } else {
3544 __ bind(is_null); // same as 'done'
3545 }
3546 __ bind(done);
3547 }
3548
3549 void TemplateTable::instanceof() {
3550 transition(atos, itos);
3551 Label done, is_null, ok_is_subtype, quicked, resolved;
3552 __ cbz(r0, is_null);
3553
3554 // Get cpool & tags index
3555 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3556 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3557 // See if bytecode has already been quicked
3558 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3559 __ lea(r1, Address(rscratch1, r19));
3560 __ ldarb(r1, r1);
3561 __ cmp(r1, JVM_CONSTANT_Class);
3562 __ br(Assembler::EQ, quicked);
3563
3564 __ push(atos); // save receiver for result, and for GC
3565 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3566 // vm_result_2 has metadata result
3567 __ get_vm_result_2(r0, rthread);
3568 __ pop(r3); // restore receiver
3569 __ verify_oop(r3);
3570 __ load_klass(r3, r3);
3571 __ b(resolved);
3572
3573 // Get superklass in r0 and subklass in r3
3574 __ bind(quicked);
3575 __ load_klass(r3, r0);
3576 __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1);
3577
3578 __ bind(resolved);
3579
3580 // Generate subtype check. Blows r2, r5
3581 // Superklass in r0. Subklass in r3.
3582 __ gen_subtype_check(r3, ok_is_subtype);
3583
3584 // Come here on failure
3585 __ mov(r0, 0);
3586 __ b(done);
3587 // Come here on success
3588 __ bind(ok_is_subtype);
3589 __ mov(r0, 1);
3590
3591 // Collect counts on whether this test sees NULLs a lot or not.
3592 if (ProfileInterpreter) {
3593 __ b(done);
3594 __ bind(is_null);
3595 __ profile_null_seen(r2);
3596 } else {
3597 __ bind(is_null); // same as 'done'
3598 }
3599 __ bind(done);
3600 // r0 = 0: obj == NULL or obj is not an instanceof the specified klass
3601 // r0 = 1: obj != NULL and obj is an instanceof the specified klass
3602 }
3603
3604 //-----------------------------------------------------------------------------
3605 // Breakpoints
3606 void TemplateTable::_breakpoint() {
3607 // Note: We get here even if we are single stepping..
3608 // jbug inists on setting breakpoints at every bytecode
3609 // even if we are in single step mode.
3610
3611 transition(vtos, vtos);
3612
3613 // get the unpatched byte code
3614 __ get_method(c_rarg1);
3615 __ call_VM(noreg,
3616 CAST_FROM_FN_PTR(address,
3617 InterpreterRuntime::get_original_bytecode_at),
3618 c_rarg1, rbcp);
3619 __ mov(r19, r0);
3620
3621 // post the breakpoint event
3622 __ call_VM(noreg,
3623 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3624 rmethod, rbcp);
3625
3626 // complete the execution of original bytecode
3627 __ mov(rscratch1, r19);
3628 __ dispatch_only_normal(vtos);
3629 }
3630
3631 //-----------------------------------------------------------------------------
3632 // Exceptions
3633
3634 void TemplateTable::athrow() {
3635 transition(atos, vtos);
3636 __ null_check(r0);
3637 __ b(Interpreter::throw_exception_entry());
3638 }
3639
3640 //-----------------------------------------------------------------------------
3641 // Synchronization
3642 //
3643 // Note: monitorenter & exit are symmetric routines; which is reflected
3644 // in the assembly code structure as well
3645 //
3646 // Stack layout:
3647 //
3648 // [expressions ] <--- esp = expression stack top
3649 // ..
3650 // [expressions ]
3651 // [monitor entry] <--- monitor block top = expression stack bot
3652 // ..
3653 // [monitor entry]
3654 // [frame data ] <--- monitor block bot
3655 // ...
3656 // [saved rbp ] <--- rbp
3657 void TemplateTable::monitorenter()
3658 {
3659 transition(atos, vtos);
3660
3661 // check for NULL object
3662 __ null_check(r0);
3663
3664 const Address monitor_block_top(
3665 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3666 const Address monitor_block_bot(
3667 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3668 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3669
3670 Label allocated;
3671
3672 // initialize entry pointer
3673 __ mov(c_rarg1, zr); // points to free slot or NULL
3674
3675 // find a free slot in the monitor block (result in c_rarg1)
3676 {
3677 Label entry, loop, exit;
3678 __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3679 // starting with top-most entry
3680 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3681
3682 __ b(entry);
3683
3684 __ bind(loop);
3685 // check if current entry is used
3686 // if not used then remember entry in c_rarg1
3687 __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3688 __ cmp(zr, rscratch1);
3689 __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3690 // check if current entry is for same object
3691 __ cmp(r0, rscratch1);
3692 // if same object then stop searching
3693 __ br(Assembler::EQ, exit);
3694 // otherwise advance to next entry
3695 __ add(c_rarg3, c_rarg3, entry_size);
3696 __ bind(entry);
3697 // check if bottom reached
3698 __ cmp(c_rarg3, c_rarg2);
3699 // if not at bottom then check this entry
3700 __ br(Assembler::NE, loop);
3701 __ bind(exit);
3702 }
3703
3704 __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3705 // if found, continue with that on
3706
3707 // allocate one if there's no free slot
3708 {
3709 Label entry, loop;
3710 // 1. compute new pointers // rsp: old expression stack top
3711 __ ldr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
3712 __ sub(esp, esp, entry_size); // move expression stack top
3713 __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3714 __ mov(c_rarg3, esp); // set start value for copy loop
3715 __ str(c_rarg1, monitor_block_bot); // set new monitor block bottom
3716
3717 __ sub(sp, sp, entry_size); // make room for the monitor
3718
3719 __ b(entry);
3720 // 2. move expression stack contents
3721 __ bind(loop);
3722 __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3723 // word from old location
3724 __ str(c_rarg2, Address(c_rarg3, 0)); // and store it at new location
3725 __ add(c_rarg3, c_rarg3, wordSize); // advance to next word
3726 __ bind(entry);
3727 __ cmp(c_rarg3, c_rarg1); // check if bottom reached
3728 __ br(Assembler::NE, loop); // if not at bottom then
3729 // copy next word
3730 }
3731
3732 // call run-time routine
3733 // c_rarg1: points to monitor entry
3734 __ bind(allocated);
3735
3736 // Increment bcp to point to the next bytecode, so exception
3737 // handling for async. exceptions work correctly.
3738 // The object has already been poped from the stack, so the
3739 // expression stack looks correct.
3740 __ increment(rbcp);
3741
3742 // store object
3743 __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3744 __ lock_object(c_rarg1);
3745
3746 // check to make sure this monitor doesn't cause stack overflow after locking
3747 __ save_bcp(); // in case of exception
3748 __ generate_stack_overflow_check(0);
3749
3750 // The bcp has already been incremented. Just need to dispatch to
3751 // next instruction.
3752 __ dispatch_next(vtos);
3753 }
3754
3755
3756 void TemplateTable::monitorexit()
3757 {
3758 transition(atos, vtos);
3759
3760 // check for NULL object
3761 __ null_check(r0);
3762
3763 const Address monitor_block_top(
3764 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3765 const Address monitor_block_bot(
3766 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3767 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3768
3769 Label found;
3770
3771 // find matching slot
3772 {
3773 Label entry, loop;
3774 __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3775 // starting with top-most entry
3776 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3777 // of monitor block
3778 __ b(entry);
3779
3780 __ bind(loop);
3781 // check if current entry is for same object
3782 __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3783 __ cmp(r0, rscratch1);
3784 // if same object then stop searching
3785 __ br(Assembler::EQ, found);
3786 // otherwise advance to next entry
3787 __ add(c_rarg1, c_rarg1, entry_size);
3788 __ bind(entry);
3789 // check if bottom reached
3790 __ cmp(c_rarg1, c_rarg2);
3791 // if not at bottom then check this entry
3792 __ br(Assembler::NE, loop);
3793 }
3794
3795 // error handling. Unlocking was not block-structured
3796 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3797 InterpreterRuntime::throw_illegal_monitor_state_exception));
3798 __ should_not_reach_here();
3799
3800 // call run-time routine
3801 __ bind(found);
3802 __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3803 __ unlock_object(c_rarg1);
3804 __ pop_ptr(r0); // discard object
3805 }
3806
3807
3808 // Wide instructions
3809 void TemplateTable::wide()
3810 {
3811 __ load_unsigned_byte(r19, at_bcp(1));
3812 __ mov(rscratch1, (address)Interpreter::_wentry_point);
3813 __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
3814 __ br(rscratch1);
3815 }
3816
3817
3818 // Multi arrays
3819 void TemplateTable::multianewarray() {
3820 transition(vtos, atos);
3821 __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
3822 // last dim is on top of stack; we want address of first one:
3823 // first_addr = last_addr + (ndims - 1) * wordSize
3824 __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
3825 __ sub(c_rarg1, c_rarg1, wordSize);
3826 call_VM(r0,
3827 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3828 c_rarg1);
3829 __ load_unsigned_byte(r1, at_bcp(3));
3830 __ lea(esp, Address(esp, r1, Address::uxtw(3)));
3831 }