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 "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "interpreter/interp_masm.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "memory/universe.inline.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/method.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "prims/methodHandles.hpp"
38 #include "runtime/sharedRuntime.hpp"
39 #include "runtime/stubRoutines.hpp"
40 #include "runtime/synchronizer.hpp"
41
42 #define __ _masm->
43
44 // Platform-dependent initialization
45
46 void TemplateTable::pd_initialize() {
47 // No aarch64 specific initialization
48 }
49
50 // Address computation: local variables
51
52 static inline Address iaddress(int n) {
53 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
54 }
55
56 static inline Address laddress(int n) {
57 return iaddress(n + 1);
58 }
59
60 static inline Address faddress(int n) {
61 return iaddress(n);
62 }
63
64 static inline Address daddress(int n) {
65 return laddress(n);
66 }
67
68 static inline Address aaddress(int n) {
69 return iaddress(n);
70 }
71
72 static inline Address iaddress(Register r) {
73 return Address(rlocals, r, Address::lsl(3));
74 }
75
76 static inline Address laddress(Register r, Register scratch,
77 InterpreterMacroAssembler* _masm) {
78 __ lea(scratch, Address(rlocals, r, Address::lsl(3)));
79 return Address(scratch, Interpreter::local_offset_in_bytes(1));
80 }
81
82 static inline Address faddress(Register r) {
83 return iaddress(r);
84 }
85
86 static inline Address daddress(Register r, Register scratch,
87 InterpreterMacroAssembler* _masm) {
88 return laddress(r, scratch, _masm);
89 }
90
91 static inline Address aaddress(Register r) {
92 return iaddress(r);
93 }
94
95 static inline Address at_rsp() {
96 return Address(esp, 0);
97 }
98
99 // At top of Java expression stack which may be different than esp(). It
100 // isn't for category 1 objects.
101 static inline Address at_tos () {
102 return Address(esp, Interpreter::expr_offset_in_bytes(0));
103 }
104
105 static inline Address at_tos_p1() {
106 return Address(esp, Interpreter::expr_offset_in_bytes(1));
107 }
108
109 static inline Address at_tos_p2() {
110 return Address(esp, Interpreter::expr_offset_in_bytes(2));
111 }
112
113 static inline Address at_tos_p3() {
114 return Address(esp, Interpreter::expr_offset_in_bytes(3));
115 }
116
117 static inline Address at_tos_p4() {
118 return Address(esp, Interpreter::expr_offset_in_bytes(4));
119 }
120
121 static inline Address at_tos_p5() {
122 return Address(esp, Interpreter::expr_offset_in_bytes(5));
123 }
124
125 // Condition conversion
126 static Assembler::Condition j_not(TemplateTable::Condition cc) {
127 switch (cc) {
128 case TemplateTable::equal : return Assembler::NE;
129 case TemplateTable::not_equal : return Assembler::EQ;
130 case TemplateTable::less : return Assembler::GE;
131 case TemplateTable::less_equal : return Assembler::GT;
132 case TemplateTable::greater : return Assembler::LE;
133 case TemplateTable::greater_equal: return Assembler::LT;
134 }
135 ShouldNotReachHere();
136 return Assembler::EQ;
137 }
138
139
140 // Miscelaneous helper routines
141 // Store an oop (or NULL) at the Address described by obj.
142 // If val == noreg this means store a NULL
143 static void do_oop_store(InterpreterMacroAssembler* _masm,
144 Address obj,
145 Register val,
146 BarrierSet::Name barrier,
147 bool precise) {
148 assert(val == noreg || val == r0, "parameter is just for looks");
149 switch (barrier) {
150 #if INCLUDE_ALL_GCS
151 case BarrierSet::G1SATBCTLogging:
152 {
153 // flatten object address if needed
154 if (obj.index() == noreg && obj.offset() == 0) {
155 if (obj.base() != r3) {
156 __ mov(r3, obj.base());
157 }
158 } else {
159 __ lea(r3, obj);
160 }
161 __ g1_write_barrier_pre(r3 /* obj */,
162 r1 /* pre_val */,
163 rthread /* thread */,
164 r10 /* tmp */,
165 val != noreg /* tosca_live */,
166 false /* expand_call */);
167 if (val == noreg) {
168 __ store_heap_oop_null(Address(r3, 0));
169 } else {
170 // G1 barrier needs uncompressed oop for region cross check.
171 Register new_val = val;
172 if (UseCompressedOops) {
173 new_val = rscratch1;
174 __ mov(new_val, val);
175 }
176 __ store_heap_oop(Address(r3, 0), val);
177 __ g1_write_barrier_post(r3 /* store_adr */,
178 new_val /* new_val */,
179 rthread /* thread */,
180 r10 /* tmp */,
181 r1 /* tmp2 */);
182 }
183
184 }
185 break;
186 #endif // INCLUDE_ALL_GCS
187 case BarrierSet::CardTableForRS:
188 case BarrierSet::CardTableExtension:
189 {
190 if (val == noreg) {
191 __ store_heap_oop_null(obj);
192 } else {
193 __ store_heap_oop(obj, val);
194 // flatten object address if needed
195 if (!precise || (obj.index() == noreg && obj.offset() == 0)) {
196 __ store_check(obj.base());
197 } else {
198 __ lea(r3, obj);
199 __ store_check(r3);
200 }
201 }
202 }
203 break;
204 case BarrierSet::ModRef:
205 if (val == noreg) {
206 __ store_heap_oop_null(obj);
207 } else {
208 __ store_heap_oop(obj, val);
209 }
210 break;
211 default :
212 ShouldNotReachHere();
213
214 }
215 }
216
217 Address TemplateTable::at_bcp(int offset) {
218 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
219 return Address(rbcp, offset);
220 }
221
222 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
223 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
224 int byte_no)
225 {
226 if (!RewriteBytecodes) return;
227 Label L_patch_done;
228
229 switch (bc) {
230 case Bytecodes::_fast_aputfield:
231 case Bytecodes::_fast_bputfield:
232 case Bytecodes::_fast_zputfield:
233 case Bytecodes::_fast_cputfield:
234 case Bytecodes::_fast_dputfield:
235 case Bytecodes::_fast_fputfield:
236 case Bytecodes::_fast_iputfield:
237 case Bytecodes::_fast_lputfield:
238 case Bytecodes::_fast_sputfield:
239 {
240 // We skip bytecode quickening for putfield instructions when
241 // the put_code written to the constant pool cache is zero.
242 // This is required so that every execution of this instruction
243 // calls out to InterpreterRuntime::resolve_get_put to do
244 // additional, required work.
245 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
246 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
247 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
248 __ movw(bc_reg, bc);
249 __ cbzw(temp_reg, L_patch_done); // don't patch
250 }
251 break;
252 default:
253 assert(byte_no == -1, "sanity");
254 // the pair bytecodes have already done the load.
255 if (load_bc_into_bc_reg) {
256 __ movw(bc_reg, bc);
257 }
258 }
259
260 if (JvmtiExport::can_post_breakpoint()) {
261 Label L_fast_patch;
262 // if a breakpoint is present we can't rewrite the stream directly
263 __ load_unsigned_byte(temp_reg, at_bcp(0));
264 __ cmpw(temp_reg, Bytecodes::_breakpoint);
265 __ br(Assembler::NE, L_fast_patch);
266 // Let breakpoint table handling rewrite to quicker bytecode
267 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), rmethod, rbcp, bc_reg);
268 __ b(L_patch_done);
269 __ bind(L_fast_patch);
270 }
271
272 #ifdef ASSERT
273 Label L_okay;
274 __ load_unsigned_byte(temp_reg, at_bcp(0));
275 __ cmpw(temp_reg, (int) Bytecodes::java_code(bc));
276 __ br(Assembler::EQ, L_okay);
277 __ cmpw(temp_reg, bc_reg);
278 __ br(Assembler::EQ, L_okay);
279 __ stop("patching the wrong bytecode");
280 __ bind(L_okay);
281 #endif
282
283 // patch bytecode
284 __ strb(bc_reg, at_bcp(0));
285 __ bind(L_patch_done);
286 }
287
288
289 // Individual instructions
290
291 void TemplateTable::nop() {
292 transition(vtos, vtos);
293 // nothing to do
294 }
295
296 void TemplateTable::shouldnotreachhere() {
297 transition(vtos, vtos);
298 __ stop("shouldnotreachhere bytecode");
299 }
300
301 void TemplateTable::aconst_null()
302 {
303 transition(vtos, atos);
304 __ mov(r0, 0);
305 }
306
307 void TemplateTable::iconst(int value)
308 {
309 transition(vtos, itos);
310 __ mov(r0, value);
311 }
312
313 void TemplateTable::lconst(int value)
314 {
315 __ mov(r0, value);
316 }
317
318 void TemplateTable::fconst(int value)
319 {
320 transition(vtos, ftos);
321 switch (value) {
322 case 0:
323 __ fmovs(v0, zr);
324 break;
325 case 1:
326 __ fmovs(v0, 1.0);
327 break;
328 case 2:
329 __ fmovs(v0, 2.0);
330 break;
331 default:
332 ShouldNotReachHere();
333 break;
334 }
335 }
336
337 void TemplateTable::dconst(int value)
338 {
339 transition(vtos, dtos);
340 switch (value) {
341 case 0:
342 __ fmovd(v0, zr);
343 break;
344 case 1:
345 __ fmovd(v0, 1.0);
346 break;
347 case 2:
348 __ fmovd(v0, 2.0);
349 break;
350 default:
351 ShouldNotReachHere();
352 break;
353 }
354 }
355
356 void TemplateTable::bipush()
357 {
358 transition(vtos, itos);
359 __ load_signed_byte32(r0, at_bcp(1));
360 }
361
362 void TemplateTable::sipush()
363 {
364 transition(vtos, itos);
365 __ load_unsigned_short(r0, at_bcp(1));
366 __ revw(r0, r0);
367 __ asrw(r0, r0, 16);
368 }
369
370 void TemplateTable::ldc(bool wide)
371 {
372 transition(vtos, vtos);
373 Label call_ldc, notFloat, notClass, Done;
374
375 if (wide) {
376 __ get_unsigned_2_byte_index_at_bcp(r1, 1);
377 } else {
378 __ load_unsigned_byte(r1, at_bcp(1));
379 }
380 __ get_cpool_and_tags(r2, r0);
381
382 const int base_offset = ConstantPool::header_size() * wordSize;
383 const int tags_offset = Array<u1>::base_offset_in_bytes();
384
385 // get type
386 __ add(r3, r1, tags_offset);
387 __ lea(r3, Address(r0, r3));
388 __ ldarb(r3, r3);
389
390 // unresolved class - get the resolved class
391 __ cmp(r3, JVM_CONSTANT_UnresolvedClass);
392 __ br(Assembler::EQ, call_ldc);
393
394 // unresolved class in error state - call into runtime to throw the error
395 // from the first resolution attempt
396 __ cmp(r3, JVM_CONSTANT_UnresolvedClassInError);
397 __ br(Assembler::EQ, call_ldc);
398
399 // resolved class - need to call vm to get java mirror of the class
400 __ cmp(r3, JVM_CONSTANT_Class);
401 __ br(Assembler::NE, notClass);
402
403 __ bind(call_ldc);
404 __ mov(c_rarg1, wide);
405 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), c_rarg1);
406 __ push_ptr(r0);
407 __ verify_oop(r0);
408 __ b(Done);
409
410 __ bind(notClass);
411 __ cmp(r3, JVM_CONSTANT_Float);
412 __ br(Assembler::NE, notFloat);
413 // ftos
414 __ adds(r1, r2, r1, Assembler::LSL, 3);
415 __ ldrs(v0, Address(r1, base_offset));
416 __ push_f();
417 __ b(Done);
418
419 __ bind(notFloat);
420 #ifdef ASSERT
421 {
422 Label L;
423 __ cmp(r3, JVM_CONSTANT_Integer);
424 __ br(Assembler::EQ, L);
425 // String and Object are rewritten to fast_aldc
426 __ stop("unexpected tag type in ldc");
427 __ bind(L);
428 }
429 #endif
430 // itos JVM_CONSTANT_Integer only
431 __ adds(r1, r2, r1, Assembler::LSL, 3);
432 __ ldrw(r0, Address(r1, base_offset));
433 __ push_i(r0);
434 __ bind(Done);
435 }
436
437 // Fast path for caching oop constants.
438 void TemplateTable::fast_aldc(bool wide)
439 {
440 transition(vtos, atos);
441
442 Register result = r0;
443 Register tmp = r1;
444 int index_size = wide ? sizeof(u2) : sizeof(u1);
445
446 Label resolved;
447
448 // We are resolved if the resolved reference cache entry contains a
449 // non-null object (String, MethodType, etc.)
450 assert_different_registers(result, tmp);
451 __ get_cache_index_at_bcp(tmp, 1, index_size);
452 __ load_resolved_reference_at_index(result, tmp);
453 __ cbnz(result, resolved);
454
455 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
456
457 // first time invocation - must resolve first
458 __ mov(tmp, (int)bytecode());
459 __ call_VM(result, entry, tmp);
460
461 __ bind(resolved);
462
463 if (VerifyOops) {
464 __ verify_oop(result);
465 }
466 }
467
468 void TemplateTable::ldc2_w()
469 {
470 transition(vtos, vtos);
471 Label Long, Done;
472 __ get_unsigned_2_byte_index_at_bcp(r0, 1);
473
474 __ get_cpool_and_tags(r1, r2);
475 const int base_offset = ConstantPool::header_size() * wordSize;
476 const int tags_offset = Array<u1>::base_offset_in_bytes();
477
478 // get type
479 __ lea(r2, Address(r2, r0, Address::lsl(0)));
480 __ load_unsigned_byte(r2, Address(r2, tags_offset));
481 __ cmpw(r2, (int)JVM_CONSTANT_Double);
482 __ br(Assembler::NE, Long);
483 // dtos
484 __ lea (r2, Address(r1, r0, Address::lsl(3)));
485 __ ldrd(v0, Address(r2, base_offset));
486 __ push_d();
487 __ b(Done);
488
489 __ bind(Long);
490 // ltos
491 __ lea(r0, Address(r1, r0, Address::lsl(3)));
492 __ ldr(r0, Address(r0, base_offset));
493 __ push_l();
494
495 __ bind(Done);
496 }
497
498 void TemplateTable::locals_index(Register reg, int offset)
499 {
500 __ ldrb(reg, at_bcp(offset));
501 __ neg(reg, reg);
502 }
503
504 void TemplateTable::iload() {
505 iload_internal();
506 }
507
508 void TemplateTable::nofast_iload() {
509 iload_internal(may_not_rewrite);
510 }
511
512 void TemplateTable::iload_internal(RewriteControl rc) {
513 transition(vtos, itos);
514 if (RewriteFrequentPairs && rc == may_rewrite) {
515 Label rewrite, done;
516 Register bc = r4;
517
518 // get next bytecode
519 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
520
521 // if _iload, wait to rewrite to iload2. We only want to rewrite the
522 // last two iloads in a pair. Comparing against fast_iload means that
523 // the next bytecode is neither an iload or a caload, and therefore
524 // an iload pair.
525 __ cmpw(r1, Bytecodes::_iload);
526 __ br(Assembler::EQ, done);
527
528 // if _fast_iload rewrite to _fast_iload2
529 __ cmpw(r1, Bytecodes::_fast_iload);
530 __ movw(bc, Bytecodes::_fast_iload2);
531 __ br(Assembler::EQ, rewrite);
532
533 // if _caload rewrite to _fast_icaload
534 __ cmpw(r1, Bytecodes::_caload);
535 __ movw(bc, Bytecodes::_fast_icaload);
536 __ br(Assembler::EQ, rewrite);
537
538 // else rewrite to _fast_iload
539 __ movw(bc, Bytecodes::_fast_iload);
540
541 // rewrite
542 // bc: new bytecode
543 __ bind(rewrite);
544 patch_bytecode(Bytecodes::_iload, bc, r1, false);
545 __ bind(done);
546
547 }
548
549 // do iload, get the local value into tos
550 locals_index(r1);
551 __ ldr(r0, iaddress(r1));
552
553 }
554
555 void TemplateTable::fast_iload2()
556 {
557 transition(vtos, itos);
558 locals_index(r1);
559 __ ldr(r0, iaddress(r1));
560 __ push(itos);
561 locals_index(r1, 3);
562 __ ldr(r0, iaddress(r1));
563 }
564
565 void TemplateTable::fast_iload()
566 {
567 transition(vtos, itos);
568 locals_index(r1);
569 __ ldr(r0, iaddress(r1));
570 }
571
572 void TemplateTable::lload()
573 {
574 transition(vtos, ltos);
575 __ ldrb(r1, at_bcp(1));
576 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
577 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
578 }
579
580 void TemplateTable::fload()
581 {
582 transition(vtos, ftos);
583 locals_index(r1);
584 // n.b. we use ldrd here because this is a 64 bit slot
585 // this is comparable to the iload case
586 __ ldrd(v0, faddress(r1));
587 }
588
589 void TemplateTable::dload()
590 {
591 transition(vtos, dtos);
592 __ ldrb(r1, at_bcp(1));
593 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
594 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
595 }
596
597 void TemplateTable::aload()
598 {
599 transition(vtos, atos);
600 locals_index(r1);
601 __ ldr(r0, iaddress(r1));
602 }
603
604 void TemplateTable::locals_index_wide(Register reg) {
605 __ ldrh(reg, at_bcp(2));
606 __ rev16w(reg, reg);
607 __ neg(reg, reg);
608 }
609
610 void TemplateTable::wide_iload() {
611 transition(vtos, itos);
612 locals_index_wide(r1);
613 __ ldr(r0, iaddress(r1));
614 }
615
616 void TemplateTable::wide_lload()
617 {
618 transition(vtos, ltos);
619 __ ldrh(r1, at_bcp(2));
620 __ rev16w(r1, r1);
621 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
622 __ ldr(r0, Address(r1, Interpreter::local_offset_in_bytes(1)));
623 }
624
625 void TemplateTable::wide_fload()
626 {
627 transition(vtos, ftos);
628 locals_index_wide(r1);
629 // n.b. we use ldrd here because this is a 64 bit slot
630 // this is comparable to the iload case
631 __ ldrd(v0, faddress(r1));
632 }
633
634 void TemplateTable::wide_dload()
635 {
636 transition(vtos, dtos);
637 __ ldrh(r1, at_bcp(2));
638 __ rev16w(r1, r1);
639 __ sub(r1, rlocals, r1, ext::uxtw, LogBytesPerWord);
640 __ ldrd(v0, Address(r1, Interpreter::local_offset_in_bytes(1)));
641 }
642
643 void TemplateTable::wide_aload()
644 {
645 transition(vtos, atos);
646 locals_index_wide(r1);
647 __ ldr(r0, aaddress(r1));
648 }
649
650 void TemplateTable::index_check(Register array, Register index)
651 {
652 // destroys r1, rscratch1
653 // check array
654 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
655 // sign extend index for use by indexed load
656 // __ movl2ptr(index, index);
657 // check index
658 Register length = rscratch1;
659 __ ldrw(length, Address(array, arrayOopDesc::length_offset_in_bytes()));
660 __ cmpw(index, length);
661 if (index != r1) {
662 // ??? convention: move aberrant index into r1 for exception message
663 assert(r1 != array, "different registers");
664 __ mov(r1, index);
665 }
666 Label ok;
667 __ br(Assembler::LO, ok);
668 __ mov(rscratch1, Interpreter::_throw_ArrayIndexOutOfBoundsException_entry);
669 __ br(rscratch1);
670 __ bind(ok);
671 }
672
673 void TemplateTable::iaload()
674 {
675 transition(itos, itos);
676 __ mov(r1, r0);
677 __ pop_ptr(r0);
678 // r0: array
679 // r1: index
680 index_check(r0, r1); // leaves index in r1, kills rscratch1
681 __ lea(r1, Address(r0, r1, Address::uxtw(2)));
682 __ ldrw(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_INT)));
683 }
684
685 void TemplateTable::laload()
686 {
687 transition(itos, ltos);
688 __ mov(r1, r0);
689 __ pop_ptr(r0);
690 // r0: array
691 // r1: index
692 index_check(r0, r1); // leaves index in r1, kills rscratch1
693 __ lea(r1, Address(r0, r1, Address::uxtw(3)));
694 __ ldr(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_LONG)));
695 }
696
697 void TemplateTable::faload()
698 {
699 transition(itos, ftos);
700 __ mov(r1, r0);
701 __ pop_ptr(r0);
702 // r0: array
703 // r1: index
704 index_check(r0, r1); // leaves index in r1, kills rscratch1
705 __ lea(r1, Address(r0, r1, Address::uxtw(2)));
706 __ ldrs(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
707 }
708
709 void TemplateTable::daload()
710 {
711 transition(itos, dtos);
712 __ mov(r1, r0);
713 __ pop_ptr(r0);
714 // r0: array
715 // r1: index
716 index_check(r0, r1); // leaves index in r1, kills rscratch1
717 __ lea(r1, Address(r0, r1, Address::uxtw(3)));
718 __ ldrd(v0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
719 }
720
721 void TemplateTable::aaload()
722 {
723 transition(itos, atos);
724 __ mov(r1, r0);
725 __ pop_ptr(r0);
726 // r0: array
727 // r1: index
728 index_check(r0, r1); // leaves index in r1, kills rscratch1
729 int s = (UseCompressedOops ? 2 : 3);
730 __ lea(r1, Address(r0, r1, Address::uxtw(s)));
731 __ load_heap_oop(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
732 }
733
734 void TemplateTable::baload()
735 {
736 transition(itos, itos);
737 __ mov(r1, r0);
738 __ pop_ptr(r0);
739 // r0: array
740 // r1: index
741 index_check(r0, r1); // leaves index in r1, kills rscratch1
742 __ lea(r1, Address(r0, r1, Address::uxtw(0)));
743 __ load_signed_byte(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
744 }
745
746 void TemplateTable::caload()
747 {
748 transition(itos, itos);
749 __ mov(r1, r0);
750 __ pop_ptr(r0);
751 // r0: array
752 // r1: index
753 index_check(r0, r1); // leaves index in r1, kills rscratch1
754 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
755 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
756 }
757
758 // iload followed by caload frequent pair
759 void TemplateTable::fast_icaload()
760 {
761 transition(vtos, itos);
762 // load index out of locals
763 locals_index(r2);
764 __ ldr(r1, iaddress(r2));
765
766 __ pop_ptr(r0);
767
768 // r0: array
769 // r1: index
770 index_check(r0, r1); // leaves index in r1, kills rscratch1
771 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
772 __ load_unsigned_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
773 }
774
775 void TemplateTable::saload()
776 {
777 transition(itos, itos);
778 __ mov(r1, r0);
779 __ pop_ptr(r0);
780 // r0: array
781 // r1: index
782 index_check(r0, r1); // leaves index in r1, kills rscratch1
783 __ lea(r1, Address(r0, r1, Address::uxtw(1)));
784 __ load_signed_short(r0, Address(r1, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
785 }
786
787 void TemplateTable::iload(int n)
788 {
789 transition(vtos, itos);
790 __ ldr(r0, iaddress(n));
791 }
792
793 void TemplateTable::lload(int n)
794 {
795 transition(vtos, ltos);
796 __ ldr(r0, laddress(n));
797 }
798
799 void TemplateTable::fload(int n)
800 {
801 transition(vtos, ftos);
802 __ ldrs(v0, faddress(n));
803 }
804
805 void TemplateTable::dload(int n)
806 {
807 transition(vtos, dtos);
808 __ ldrd(v0, daddress(n));
809 }
810
811 void TemplateTable::aload(int n)
812 {
813 transition(vtos, atos);
814 __ ldr(r0, iaddress(n));
815 }
816
817 void TemplateTable::aload_0() {
818 aload_0_internal();
819 }
820
821 void TemplateTable::nofast_aload_0() {
822 aload_0_internal(may_not_rewrite);
823 }
824
825 void TemplateTable::aload_0_internal(RewriteControl rc) {
826 // According to bytecode histograms, the pairs:
827 //
828 // _aload_0, _fast_igetfield
829 // _aload_0, _fast_agetfield
830 // _aload_0, _fast_fgetfield
831 //
832 // occur frequently. If RewriteFrequentPairs is set, the (slow)
833 // _aload_0 bytecode checks if the next bytecode is either
834 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
835 // rewrites the current bytecode into a pair bytecode; otherwise it
836 // rewrites the current bytecode into _fast_aload_0 that doesn't do
837 // the pair check anymore.
838 //
839 // Note: If the next bytecode is _getfield, the rewrite must be
840 // delayed, otherwise we may miss an opportunity for a pair.
841 //
842 // Also rewrite frequent pairs
843 // aload_0, aload_1
844 // aload_0, iload_1
845 // These bytecodes with a small amount of code are most profitable
846 // to rewrite
847 if (RewriteFrequentPairs && rc == may_rewrite) {
848 Label rewrite, done;
849 const Register bc = r4;
850
851 // get next bytecode
852 __ load_unsigned_byte(r1, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
853
854 // if _getfield then wait with rewrite
855 __ cmpw(r1, Bytecodes::Bytecodes::_getfield);
856 __ br(Assembler::EQ, done);
857
858 // if _igetfield then rewrite to _fast_iaccess_0
859 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
860 __ cmpw(r1, Bytecodes::_fast_igetfield);
861 __ movw(bc, Bytecodes::_fast_iaccess_0);
862 __ br(Assembler::EQ, rewrite);
863
864 // if _agetfield then rewrite to _fast_aaccess_0
865 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
866 __ cmpw(r1, Bytecodes::_fast_agetfield);
867 __ movw(bc, Bytecodes::_fast_aaccess_0);
868 __ br(Assembler::EQ, rewrite);
869
870 // if _fgetfield then rewrite to _fast_faccess_0
871 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
872 __ cmpw(r1, Bytecodes::_fast_fgetfield);
873 __ movw(bc, Bytecodes::_fast_faccess_0);
874 __ br(Assembler::EQ, rewrite);
875
876 // else rewrite to _fast_aload0
877 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
878 __ movw(bc, Bytecodes::Bytecodes::_fast_aload_0);
879
880 // rewrite
881 // bc: new bytecode
882 __ bind(rewrite);
883 patch_bytecode(Bytecodes::_aload_0, bc, r1, false);
884
885 __ bind(done);
886 }
887
888 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
889 aload(0);
890 }
891
892 void TemplateTable::istore()
893 {
894 transition(itos, vtos);
895 locals_index(r1);
896 // FIXME: We're being very pernickerty here storing a jint in a
897 // local with strw, which costs an extra instruction over what we'd
898 // be able to do with a simple str. We should just store the whole
899 // word.
900 __ lea(rscratch1, iaddress(r1));
901 __ strw(r0, Address(rscratch1));
902 }
903
904 void TemplateTable::lstore()
905 {
906 transition(ltos, vtos);
907 locals_index(r1);
908 __ str(r0, laddress(r1, rscratch1, _masm));
909 }
910
911 void TemplateTable::fstore() {
912 transition(ftos, vtos);
913 locals_index(r1);
914 __ lea(rscratch1, iaddress(r1));
915 __ strs(v0, Address(rscratch1));
916 }
917
918 void TemplateTable::dstore() {
919 transition(dtos, vtos);
920 locals_index(r1);
921 __ strd(v0, daddress(r1, rscratch1, _masm));
922 }
923
924 void TemplateTable::astore()
925 {
926 transition(vtos, vtos);
927 __ pop_ptr(r0);
928 locals_index(r1);
929 __ str(r0, aaddress(r1));
930 }
931
932 void TemplateTable::wide_istore() {
933 transition(vtos, vtos);
934 __ pop_i();
935 locals_index_wide(r1);
936 __ lea(rscratch1, iaddress(r1));
937 __ strw(r0, Address(rscratch1));
938 }
939
940 void TemplateTable::wide_lstore() {
941 transition(vtos, vtos);
942 __ pop_l();
943 locals_index_wide(r1);
944 __ str(r0, laddress(r1, rscratch1, _masm));
945 }
946
947 void TemplateTable::wide_fstore() {
948 transition(vtos, vtos);
949 __ pop_f();
950 locals_index_wide(r1);
951 __ lea(rscratch1, faddress(r1));
952 __ strs(v0, rscratch1);
953 }
954
955 void TemplateTable::wide_dstore() {
956 transition(vtos, vtos);
957 __ pop_d();
958 locals_index_wide(r1);
959 __ strd(v0, daddress(r1, rscratch1, _masm));
960 }
961
962 void TemplateTable::wide_astore() {
963 transition(vtos, vtos);
964 __ pop_ptr(r0);
965 locals_index_wide(r1);
966 __ str(r0, aaddress(r1));
967 }
968
969 void TemplateTable::iastore() {
970 transition(itos, vtos);
971 __ pop_i(r1);
972 __ pop_ptr(r3);
973 // r0: value
974 // r1: index
975 // r3: array
976 index_check(r3, r1); // prefer index in r1
977 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
978 __ strw(r0, Address(rscratch1,
979 arrayOopDesc::base_offset_in_bytes(T_INT)));
980 }
981
982 void TemplateTable::lastore() {
983 transition(ltos, vtos);
984 __ pop_i(r1);
985 __ pop_ptr(r3);
986 // r0: value
987 // r1: index
988 // r3: array
989 index_check(r3, r1); // prefer index in r1
990 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
991 __ str(r0, Address(rscratch1,
992 arrayOopDesc::base_offset_in_bytes(T_LONG)));
993 }
994
995 void TemplateTable::fastore() {
996 transition(ftos, vtos);
997 __ pop_i(r1);
998 __ pop_ptr(r3);
999 // v0: value
1000 // r1: index
1001 // r3: array
1002 index_check(r3, r1); // prefer index in r1
1003 __ lea(rscratch1, Address(r3, r1, Address::uxtw(2)));
1004 __ strs(v0, Address(rscratch1,
1005 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1006 }
1007
1008 void TemplateTable::dastore() {
1009 transition(dtos, vtos);
1010 __ pop_i(r1);
1011 __ pop_ptr(r3);
1012 // v0: value
1013 // r1: index
1014 // r3: array
1015 index_check(r3, r1); // prefer index in r1
1016 __ lea(rscratch1, Address(r3, r1, Address::uxtw(3)));
1017 __ strd(v0, Address(rscratch1,
1018 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1019 }
1020
1021 void TemplateTable::aastore() {
1022 Label is_null, ok_is_subtype, done;
1023 transition(vtos, vtos);
1024 // stack: ..., array, index, value
1025 __ ldr(r0, at_tos()); // value
1026 __ ldr(r2, at_tos_p1()); // index
1027 __ ldr(r3, at_tos_p2()); // array
1028
1029 Address element_address(r4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1030
1031 index_check(r3, r2); // kills r1
1032 __ lea(r4, Address(r3, r2, Address::uxtw(UseCompressedOops? 2 : 3)));
1033
1034 // do array store check - check for NULL value first
1035 __ cbz(r0, is_null);
1036
1037 // Move subklass into r1
1038 __ load_klass(r1, r0);
1039 // Move superklass into r0
1040 __ load_klass(r0, r3);
1041 __ ldr(r0, Address(r0,
1042 ObjArrayKlass::element_klass_offset()));
1043 // Compress array + index*oopSize + 12 into a single register. Frees r2.
1044
1045 // Generate subtype check. Blows r2, r5
1046 // Superklass in r0. Subklass in r1.
1047 __ gen_subtype_check(r1, ok_is_subtype);
1048
1049 // Come here on failure
1050 // object is at TOS
1051 __ b(Interpreter::_throw_ArrayStoreException_entry);
1052
1053 // Come here on success
1054 __ bind(ok_is_subtype);
1055
1056 // Get the value we will store
1057 __ ldr(r0, at_tos());
1058 // Now store using the appropriate barrier
1059 do_oop_store(_masm, element_address, r0, _bs->kind(), true);
1060 __ b(done);
1061
1062 // Have a NULL in r0, r3=array, r2=index. Store NULL at ary[idx]
1063 __ bind(is_null);
1064 __ profile_null_seen(r2);
1065
1066 // Store a NULL
1067 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1068
1069 // Pop stack arguments
1070 __ bind(done);
1071 __ add(esp, esp, 3 * Interpreter::stackElementSize);
1072 }
1073
1074 void TemplateTable::bastore()
1075 {
1076 transition(itos, vtos);
1077 __ pop_i(r1);
1078 __ pop_ptr(r3);
1079 // r0: value
1080 // r1: index
1081 // r3: array
1082 index_check(r3, r1); // prefer index in r1
1083
1084 // Need to check whether array is boolean or byte
1085 // since both types share the bastore bytecode.
1086 __ load_klass(r2, r3);
1087 __ ldrw(r2, Address(r2, Klass::layout_helper_offset()));
1088 int diffbit_index = exact_log2(Klass::layout_helper_boolean_diffbit());
1089 Label L_skip;
1090 __ tbz(r2, diffbit_index, L_skip);
1091 __ andw(r0, r0, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1092 __ bind(L_skip);
1093
1094 __ lea(rscratch1, Address(r3, r1, Address::uxtw(0)));
1095 __ strb(r0, Address(rscratch1,
1096 arrayOopDesc::base_offset_in_bytes(T_BYTE)));
1097 }
1098
1099 void TemplateTable::castore()
1100 {
1101 transition(itos, vtos);
1102 __ pop_i(r1);
1103 __ pop_ptr(r3);
1104 // r0: value
1105 // r1: index
1106 // r3: array
1107 index_check(r3, r1); // prefer index in r1
1108 __ lea(rscratch1, Address(r3, r1, Address::uxtw(1)));
1109 __ strh(r0, Address(rscratch1,
1110 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
1111 }
1112
1113 void TemplateTable::sastore()
1114 {
1115 castore();
1116 }
1117
1118 void TemplateTable::istore(int n)
1119 {
1120 transition(itos, vtos);
1121 __ str(r0, iaddress(n));
1122 }
1123
1124 void TemplateTable::lstore(int n)
1125 {
1126 transition(ltos, vtos);
1127 __ str(r0, laddress(n));
1128 }
1129
1130 void TemplateTable::fstore(int n)
1131 {
1132 transition(ftos, vtos);
1133 __ strs(v0, faddress(n));
1134 }
1135
1136 void TemplateTable::dstore(int n)
1137 {
1138 transition(dtos, vtos);
1139 __ strd(v0, daddress(n));
1140 }
1141
1142 void TemplateTable::astore(int n)
1143 {
1144 transition(vtos, vtos);
1145 __ pop_ptr(r0);
1146 __ str(r0, iaddress(n));
1147 }
1148
1149 void TemplateTable::pop()
1150 {
1151 transition(vtos, vtos);
1152 __ add(esp, esp, Interpreter::stackElementSize);
1153 }
1154
1155 void TemplateTable::pop2()
1156 {
1157 transition(vtos, vtos);
1158 __ add(esp, esp, 2 * Interpreter::stackElementSize);
1159 }
1160
1161 void TemplateTable::dup()
1162 {
1163 transition(vtos, vtos);
1164 __ ldr(r0, Address(esp, 0));
1165 __ push(r0);
1166 // stack: ..., a, a
1167 }
1168
1169 void TemplateTable::dup_x1()
1170 {
1171 transition(vtos, vtos);
1172 // stack: ..., a, b
1173 __ ldr(r0, at_tos()); // load b
1174 __ ldr(r2, at_tos_p1()); // load a
1175 __ str(r0, at_tos_p1()); // store b
1176 __ str(r2, at_tos()); // store a
1177 __ push(r0); // push b
1178 // stack: ..., b, a, b
1179 }
1180
1181 void TemplateTable::dup_x2()
1182 {
1183 transition(vtos, vtos);
1184 // stack: ..., a, b, c
1185 __ ldr(r0, at_tos()); // load c
1186 __ ldr(r2, at_tos_p2()); // load a
1187 __ str(r0, at_tos_p2()); // store c in a
1188 __ push(r0); // push c
1189 // stack: ..., c, b, c, c
1190 __ ldr(r0, at_tos_p2()); // load b
1191 __ str(r2, at_tos_p2()); // store a in b
1192 // stack: ..., c, a, c, c
1193 __ str(r0, at_tos_p1()); // store b in c
1194 // stack: ..., c, a, b, c
1195 }
1196
1197 void TemplateTable::dup2()
1198 {
1199 transition(vtos, vtos);
1200 // stack: ..., a, b
1201 __ ldr(r0, at_tos_p1()); // load a
1202 __ push(r0); // push a
1203 __ ldr(r0, at_tos_p1()); // load b
1204 __ push(r0); // push b
1205 // stack: ..., a, b, a, b
1206 }
1207
1208 void TemplateTable::dup2_x1()
1209 {
1210 transition(vtos, vtos);
1211 // stack: ..., a, b, c
1212 __ ldr(r2, at_tos()); // load c
1213 __ ldr(r0, at_tos_p1()); // load b
1214 __ push(r0); // push b
1215 __ push(r2); // push c
1216 // stack: ..., a, b, c, b, c
1217 __ str(r2, at_tos_p3()); // store c in b
1218 // stack: ..., a, c, c, b, c
1219 __ ldr(r2, at_tos_p4()); // load a
1220 __ str(r2, at_tos_p2()); // store a in 2nd c
1221 // stack: ..., a, c, a, b, c
1222 __ str(r0, at_tos_p4()); // store b in a
1223 // stack: ..., b, c, a, b, c
1224 }
1225
1226 void TemplateTable::dup2_x2()
1227 {
1228 transition(vtos, vtos);
1229 // stack: ..., a, b, c, d
1230 __ ldr(r2, at_tos()); // load d
1231 __ ldr(r0, at_tos_p1()); // load c
1232 __ push(r0) ; // push c
1233 __ push(r2); // push d
1234 // stack: ..., a, b, c, d, c, d
1235 __ ldr(r0, at_tos_p4()); // load b
1236 __ str(r0, at_tos_p2()); // store b in d
1237 __ str(r2, at_tos_p4()); // store d in b
1238 // stack: ..., a, d, c, b, c, d
1239 __ ldr(r2, at_tos_p5()); // load a
1240 __ ldr(r0, at_tos_p3()); // load c
1241 __ str(r2, at_tos_p3()); // store a in c
1242 __ str(r0, at_tos_p5()); // store c in a
1243 // stack: ..., c, d, a, b, c, d
1244 }
1245
1246 void TemplateTable::swap()
1247 {
1248 transition(vtos, vtos);
1249 // stack: ..., a, b
1250 __ ldr(r2, at_tos_p1()); // load a
1251 __ ldr(r0, at_tos()); // load b
1252 __ str(r2, at_tos()); // store a in b
1253 __ str(r0, at_tos_p1()); // store b in a
1254 // stack: ..., b, a
1255 }
1256
1257 void TemplateTable::iop2(Operation op)
1258 {
1259 transition(itos, itos);
1260 // r0 <== r1 op r0
1261 __ pop_i(r1);
1262 switch (op) {
1263 case add : __ addw(r0, r1, r0); break;
1264 case sub : __ subw(r0, r1, r0); break;
1265 case mul : __ mulw(r0, r1, r0); break;
1266 case _and : __ andw(r0, r1, r0); break;
1267 case _or : __ orrw(r0, r1, r0); break;
1268 case _xor : __ eorw(r0, r1, r0); break;
1269 case shl : __ lslvw(r0, r1, r0); break;
1270 case shr : __ asrvw(r0, r1, r0); break;
1271 case ushr : __ lsrvw(r0, r1, r0);break;
1272 default : ShouldNotReachHere();
1273 }
1274 }
1275
1276 void TemplateTable::lop2(Operation op)
1277 {
1278 transition(ltos, ltos);
1279 // r0 <== r1 op r0
1280 __ pop_l(r1);
1281 switch (op) {
1282 case add : __ add(r0, r1, r0); break;
1283 case sub : __ sub(r0, r1, r0); break;
1284 case mul : __ mul(r0, r1, r0); break;
1285 case _and : __ andr(r0, r1, r0); break;
1286 case _or : __ orr(r0, r1, r0); break;
1287 case _xor : __ eor(r0, r1, r0); break;
1288 default : ShouldNotReachHere();
1289 }
1290 }
1291
1292 void TemplateTable::idiv()
1293 {
1294 transition(itos, itos);
1295 // explicitly check for div0
1296 Label no_div0;
1297 __ cbnzw(r0, no_div0);
1298 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1299 __ br(rscratch1);
1300 __ bind(no_div0);
1301 __ pop_i(r1);
1302 // r0 <== r1 idiv r0
1303 __ corrected_idivl(r0, r1, r0, /* want_remainder */ false);
1304 }
1305
1306 void TemplateTable::irem()
1307 {
1308 transition(itos, itos);
1309 // explicitly check for div0
1310 Label no_div0;
1311 __ cbnzw(r0, no_div0);
1312 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1313 __ br(rscratch1);
1314 __ bind(no_div0);
1315 __ pop_i(r1);
1316 // r0 <== r1 irem r0
1317 __ corrected_idivl(r0, r1, r0, /* want_remainder */ true);
1318 }
1319
1320 void TemplateTable::lmul()
1321 {
1322 transition(ltos, ltos);
1323 __ pop_l(r1);
1324 __ mul(r0, r0, r1);
1325 }
1326
1327 void TemplateTable::ldiv()
1328 {
1329 transition(ltos, ltos);
1330 // explicitly check for div0
1331 Label no_div0;
1332 __ cbnz(r0, no_div0);
1333 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1334 __ br(rscratch1);
1335 __ bind(no_div0);
1336 __ pop_l(r1);
1337 // r0 <== r1 ldiv r0
1338 __ corrected_idivq(r0, r1, r0, /* want_remainder */ false);
1339 }
1340
1341 void TemplateTable::lrem()
1342 {
1343 transition(ltos, ltos);
1344 // explicitly check for div0
1345 Label no_div0;
1346 __ cbnz(r0, no_div0);
1347 __ mov(rscratch1, Interpreter::_throw_ArithmeticException_entry);
1348 __ br(rscratch1);
1349 __ bind(no_div0);
1350 __ pop_l(r1);
1351 // r0 <== r1 lrem r0
1352 __ corrected_idivq(r0, r1, r0, /* want_remainder */ true);
1353 }
1354
1355 void TemplateTable::lshl()
1356 {
1357 transition(itos, ltos);
1358 // shift count is in r0
1359 __ pop_l(r1);
1360 __ lslv(r0, r1, r0);
1361 }
1362
1363 void TemplateTable::lshr()
1364 {
1365 transition(itos, ltos);
1366 // shift count is in r0
1367 __ pop_l(r1);
1368 __ asrv(r0, r1, r0);
1369 }
1370
1371 void TemplateTable::lushr()
1372 {
1373 transition(itos, ltos);
1374 // shift count is in r0
1375 __ pop_l(r1);
1376 __ lsrv(r0, r1, r0);
1377 }
1378
1379 void TemplateTable::fop2(Operation op)
1380 {
1381 transition(ftos, ftos);
1382 switch (op) {
1383 case add:
1384 // n.b. use ldrd because this is a 64 bit slot
1385 __ pop_f(v1);
1386 __ fadds(v0, v1, v0);
1387 break;
1388 case sub:
1389 __ pop_f(v1);
1390 __ fsubs(v0, v1, v0);
1391 break;
1392 case mul:
1393 __ pop_f(v1);
1394 __ fmuls(v0, v1, v0);
1395 break;
1396 case div:
1397 __ pop_f(v1);
1398 __ fdivs(v0, v1, v0);
1399 break;
1400 case rem:
1401 __ fmovs(v1, v0);
1402 __ pop_f(v0);
1403 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::frem),
1404 0, 2, MacroAssembler::ret_type_float);
1405 break;
1406 default:
1407 ShouldNotReachHere();
1408 break;
1409 }
1410 }
1411
1412 void TemplateTable::dop2(Operation op)
1413 {
1414 transition(dtos, dtos);
1415 switch (op) {
1416 case add:
1417 // n.b. use ldrd because this is a 64 bit slot
1418 __ pop_d(v1);
1419 __ faddd(v0, v1, v0);
1420 break;
1421 case sub:
1422 __ pop_d(v1);
1423 __ fsubd(v0, v1, v0);
1424 break;
1425 case mul:
1426 __ pop_d(v1);
1427 __ fmuld(v0, v1, v0);
1428 break;
1429 case div:
1430 __ pop_d(v1);
1431 __ fdivd(v0, v1, v0);
1432 break;
1433 case rem:
1434 __ fmovd(v1, v0);
1435 __ pop_d(v0);
1436 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::drem),
1437 0, 2, MacroAssembler::ret_type_double);
1438 break;
1439 default:
1440 ShouldNotReachHere();
1441 break;
1442 }
1443 }
1444
1445 void TemplateTable::ineg()
1446 {
1447 transition(itos, itos);
1448 __ negw(r0, r0);
1449
1450 }
1451
1452 void TemplateTable::lneg()
1453 {
1454 transition(ltos, ltos);
1455 __ neg(r0, r0);
1456 }
1457
1458 void TemplateTable::fneg()
1459 {
1460 transition(ftos, ftos);
1461 __ fnegs(v0, v0);
1462 }
1463
1464 void TemplateTable::dneg()
1465 {
1466 transition(dtos, dtos);
1467 __ fnegd(v0, v0);
1468 }
1469
1470 void TemplateTable::iinc()
1471 {
1472 transition(vtos, vtos);
1473 __ load_signed_byte(r1, at_bcp(2)); // get constant
1474 locals_index(r2);
1475 __ ldr(r0, iaddress(r2));
1476 __ addw(r0, r0, r1);
1477 __ str(r0, iaddress(r2));
1478 }
1479
1480 void TemplateTable::wide_iinc()
1481 {
1482 transition(vtos, vtos);
1483 // __ mov(r1, zr);
1484 __ ldrw(r1, at_bcp(2)); // get constant and index
1485 __ rev16(r1, r1);
1486 __ ubfx(r2, r1, 0, 16);
1487 __ neg(r2, r2);
1488 __ sbfx(r1, r1, 16, 16);
1489 __ ldr(r0, iaddress(r2));
1490 __ addw(r0, r0, r1);
1491 __ str(r0, iaddress(r2));
1492 }
1493
1494 void TemplateTable::convert()
1495 {
1496 // Checking
1497 #ifdef ASSERT
1498 {
1499 TosState tos_in = ilgl;
1500 TosState tos_out = ilgl;
1501 switch (bytecode()) {
1502 case Bytecodes::_i2l: // fall through
1503 case Bytecodes::_i2f: // fall through
1504 case Bytecodes::_i2d: // fall through
1505 case Bytecodes::_i2b: // fall through
1506 case Bytecodes::_i2c: // fall through
1507 case Bytecodes::_i2s: tos_in = itos; break;
1508 case Bytecodes::_l2i: // fall through
1509 case Bytecodes::_l2f: // fall through
1510 case Bytecodes::_l2d: tos_in = ltos; break;
1511 case Bytecodes::_f2i: // fall through
1512 case Bytecodes::_f2l: // fall through
1513 case Bytecodes::_f2d: tos_in = ftos; break;
1514 case Bytecodes::_d2i: // fall through
1515 case Bytecodes::_d2l: // fall through
1516 case Bytecodes::_d2f: tos_in = dtos; break;
1517 default : ShouldNotReachHere();
1518 }
1519 switch (bytecode()) {
1520 case Bytecodes::_l2i: // fall through
1521 case Bytecodes::_f2i: // fall through
1522 case Bytecodes::_d2i: // fall through
1523 case Bytecodes::_i2b: // fall through
1524 case Bytecodes::_i2c: // fall through
1525 case Bytecodes::_i2s: tos_out = itos; break;
1526 case Bytecodes::_i2l: // fall through
1527 case Bytecodes::_f2l: // fall through
1528 case Bytecodes::_d2l: tos_out = ltos; break;
1529 case Bytecodes::_i2f: // fall through
1530 case Bytecodes::_l2f: // fall through
1531 case Bytecodes::_d2f: tos_out = ftos; break;
1532 case Bytecodes::_i2d: // fall through
1533 case Bytecodes::_l2d: // fall through
1534 case Bytecodes::_f2d: tos_out = dtos; break;
1535 default : ShouldNotReachHere();
1536 }
1537 transition(tos_in, tos_out);
1538 }
1539 #endif // ASSERT
1540 // static const int64_t is_nan = 0x8000000000000000L;
1541
1542 // Conversion
1543 switch (bytecode()) {
1544 case Bytecodes::_i2l:
1545 __ sxtw(r0, r0);
1546 break;
1547 case Bytecodes::_i2f:
1548 __ scvtfws(v0, r0);
1549 break;
1550 case Bytecodes::_i2d:
1551 __ scvtfwd(v0, r0);
1552 break;
1553 case Bytecodes::_i2b:
1554 __ sxtbw(r0, r0);
1555 break;
1556 case Bytecodes::_i2c:
1557 __ uxthw(r0, r0);
1558 break;
1559 case Bytecodes::_i2s:
1560 __ sxthw(r0, r0);
1561 break;
1562 case Bytecodes::_l2i:
1563 __ uxtw(r0, r0);
1564 break;
1565 case Bytecodes::_l2f:
1566 __ scvtfs(v0, r0);
1567 break;
1568 case Bytecodes::_l2d:
1569 __ scvtfd(v0, r0);
1570 break;
1571 case Bytecodes::_f2i:
1572 {
1573 Label L_Okay;
1574 __ clear_fpsr();
1575 __ fcvtzsw(r0, v0);
1576 __ get_fpsr(r1);
1577 __ cbzw(r1, L_Okay);
1578 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2i),
1579 0, 1, MacroAssembler::ret_type_integral);
1580 __ bind(L_Okay);
1581 }
1582 break;
1583 case Bytecodes::_f2l:
1584 {
1585 Label L_Okay;
1586 __ clear_fpsr();
1587 __ fcvtzs(r0, v0);
1588 __ get_fpsr(r1);
1589 __ cbzw(r1, L_Okay);
1590 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::f2l),
1591 0, 1, MacroAssembler::ret_type_integral);
1592 __ bind(L_Okay);
1593 }
1594 break;
1595 case Bytecodes::_f2d:
1596 __ fcvts(v0, v0);
1597 break;
1598 case Bytecodes::_d2i:
1599 {
1600 Label L_Okay;
1601 __ clear_fpsr();
1602 __ fcvtzdw(r0, v0);
1603 __ get_fpsr(r1);
1604 __ cbzw(r1, L_Okay);
1605 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2i),
1606 0, 1, MacroAssembler::ret_type_integral);
1607 __ bind(L_Okay);
1608 }
1609 break;
1610 case Bytecodes::_d2l:
1611 {
1612 Label L_Okay;
1613 __ clear_fpsr();
1614 __ fcvtzd(r0, v0);
1615 __ get_fpsr(r1);
1616 __ cbzw(r1, L_Okay);
1617 __ call_VM_leaf_base1(CAST_FROM_FN_PTR(address, SharedRuntime::d2l),
1618 0, 1, MacroAssembler::ret_type_integral);
1619 __ bind(L_Okay);
1620 }
1621 break;
1622 case Bytecodes::_d2f:
1623 __ fcvtd(v0, v0);
1624 break;
1625 default:
1626 ShouldNotReachHere();
1627 }
1628 }
1629
1630 void TemplateTable::lcmp()
1631 {
1632 transition(ltos, itos);
1633 Label done;
1634 __ pop_l(r1);
1635 __ cmp(r1, r0);
1636 __ mov(r0, (u_int64_t)-1L);
1637 __ br(Assembler::LT, done);
1638 // __ mov(r0, 1UL);
1639 // __ csel(r0, r0, zr, Assembler::NE);
1640 // and here is a faster way
1641 __ csinc(r0, zr, zr, Assembler::EQ);
1642 __ bind(done);
1643 }
1644
1645 void TemplateTable::float_cmp(bool is_float, int unordered_result)
1646 {
1647 Label done;
1648 if (is_float) {
1649 // XXX get rid of pop here, use ... reg, mem32
1650 __ pop_f(v1);
1651 __ fcmps(v1, v0);
1652 } else {
1653 // XXX get rid of pop here, use ... reg, mem64
1654 __ pop_d(v1);
1655 __ fcmpd(v1, v0);
1656 }
1657 if (unordered_result < 0) {
1658 // we want -1 for unordered or less than, 0 for equal and 1 for
1659 // greater than.
1660 __ mov(r0, (u_int64_t)-1L);
1661 // for FP LT tests less than or unordered
1662 __ br(Assembler::LT, done);
1663 // install 0 for EQ otherwise 1
1664 __ csinc(r0, zr, zr, Assembler::EQ);
1665 } else {
1666 // we want -1 for less than, 0 for equal and 1 for unordered or
1667 // greater than.
1668 __ mov(r0, 1L);
1669 // for FP HI tests greater than or unordered
1670 __ br(Assembler::HI, done);
1671 // install 0 for EQ otherwise ~0
1672 __ csinv(r0, zr, zr, Assembler::EQ);
1673
1674 }
1675 __ bind(done);
1676 }
1677
1678 void TemplateTable::branch(bool is_jsr, bool is_wide)
1679 {
1680 // We might be moving to a safepoint. The thread which calls
1681 // Interpreter::notice_safepoints() will effectively flush its cache
1682 // when it makes a system call, but we need to do something to
1683 // ensure that we see the changed dispatch table.
1684 __ membar(MacroAssembler::LoadLoad);
1685
1686 __ profile_taken_branch(r0, r1);
1687 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1688 InvocationCounter::counter_offset();
1689 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1690 InvocationCounter::counter_offset();
1691
1692 // load branch displacement
1693 if (!is_wide) {
1694 __ ldrh(r2, at_bcp(1));
1695 __ rev16(r2, r2);
1696 // sign extend the 16 bit value in r2
1697 __ sbfm(r2, r2, 0, 15);
1698 } else {
1699 __ ldrw(r2, at_bcp(1));
1700 __ revw(r2, r2);
1701 // sign extend the 32 bit value in r2
1702 __ sbfm(r2, r2, 0, 31);
1703 }
1704
1705 // Handle all the JSR stuff here, then exit.
1706 // It's much shorter and cleaner than intermingling with the non-JSR
1707 // normal-branch stuff occurring below.
1708
1709 if (is_jsr) {
1710 // Pre-load the next target bytecode into rscratch1
1711 __ load_unsigned_byte(rscratch1, Address(rbcp, r2));
1712 // compute return address as bci
1713 __ ldr(rscratch2, Address(rmethod, Method::const_offset()));
1714 __ add(rscratch2, rscratch2,
1715 in_bytes(ConstMethod::codes_offset()) - (is_wide ? 5 : 3));
1716 __ sub(r1, rbcp, rscratch2);
1717 __ push_i(r1);
1718 // Adjust the bcp by the 16-bit displacement in r2
1719 __ add(rbcp, rbcp, r2);
1720 __ dispatch_only(vtos);
1721 return;
1722 }
1723
1724 // Normal (non-jsr) branch handling
1725
1726 // Adjust the bcp by the displacement in r2
1727 __ add(rbcp, rbcp, r2);
1728
1729 assert(UseLoopCounter || !UseOnStackReplacement,
1730 "on-stack-replacement requires loop counters");
1731 Label backedge_counter_overflow;
1732 Label profile_method;
1733 Label dispatch;
1734 if (UseLoopCounter) {
1735 // increment backedge counter for backward branches
1736 // r0: MDO
1737 // w1: MDO bumped taken-count
1738 // r2: target offset
1739 __ cmp(r2, zr);
1740 __ br(Assembler::GT, dispatch); // count only if backward branch
1741
1742 // ECN: FIXME: This code smells
1743 // check if MethodCounters exists
1744 Label has_counters;
1745 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1746 __ cbnz(rscratch1, has_counters);
1747 __ push(r0);
1748 __ push(r1);
1749 __ push(r2);
1750 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
1751 InterpreterRuntime::build_method_counters), rmethod);
1752 __ pop(r2);
1753 __ pop(r1);
1754 __ pop(r0);
1755 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1756 __ cbz(rscratch1, dispatch); // No MethodCounters allocated, OutOfMemory
1757 __ bind(has_counters);
1758
1759 if (TieredCompilation) {
1760 Label no_mdo;
1761 int increment = InvocationCounter::count_increment;
1762 if (ProfileInterpreter) {
1763 // Are we profiling?
1764 __ ldr(r1, Address(rmethod, in_bytes(Method::method_data_offset())));
1765 __ cbz(r1, no_mdo);
1766 // Increment the MDO backedge counter
1767 const Address mdo_backedge_counter(r1, in_bytes(MethodData::backedge_counter_offset()) +
1768 in_bytes(InvocationCounter::counter_offset()));
1769 const Address mask(r1, in_bytes(MethodData::backedge_mask_offset()));
1770 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1771 r0, rscratch1, false, Assembler::EQ, &backedge_counter_overflow);
1772 __ b(dispatch);
1773 }
1774 __ bind(no_mdo);
1775 // Increment backedge counter in MethodCounters*
1776 __ ldr(rscratch1, Address(rmethod, Method::method_counters_offset()));
1777 const Address mask(rscratch1, in_bytes(MethodCounters::backedge_mask_offset()));
1778 __ increment_mask_and_jump(Address(rscratch1, be_offset), increment, mask,
1779 r0, rscratch2, false, Assembler::EQ, &backedge_counter_overflow);
1780 } else { // not TieredCompilation
1781 // increment counter
1782 __ ldr(rscratch2, Address(rmethod, Method::method_counters_offset()));
1783 __ ldrw(r0, Address(rscratch2, be_offset)); // load backedge counter
1784 __ addw(rscratch1, r0, InvocationCounter::count_increment); // increment counter
1785 __ strw(rscratch1, Address(rscratch2, be_offset)); // store counter
1786
1787 __ ldrw(r0, Address(rscratch2, inv_offset)); // load invocation counter
1788 __ andw(r0, r0, (unsigned)InvocationCounter::count_mask_value); // and the status bits
1789 __ addw(r0, r0, rscratch1); // add both counters
1790
1791 if (ProfileInterpreter) {
1792 // Test to see if we should create a method data oop
1793 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
1794 __ cmpw(r0, rscratch1);
1795 __ br(Assembler::LT, dispatch);
1796
1797 // if no method data exists, go to profile method
1798 __ test_method_data_pointer(r0, profile_method);
1799
1800 if (UseOnStackReplacement) {
1801 // check for overflow against w1 which is the MDO taken count
1802 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1803 __ cmpw(r1, rscratch1);
1804 __ br(Assembler::LO, dispatch); // Intel == Assembler::below
1805
1806 // When ProfileInterpreter is on, the backedge_count comes
1807 // from the MethodData*, which value does not get reset on
1808 // the call to frequency_counter_overflow(). To avoid
1809 // excessive calls to the overflow routine while the method is
1810 // being compiled, add a second test to make sure the overflow
1811 // function is called only once every overflow_frequency.
1812 const int overflow_frequency = 1024;
1813 __ andsw(r1, r1, overflow_frequency - 1);
1814 __ br(Assembler::EQ, backedge_counter_overflow);
1815
1816 }
1817 } else {
1818 if (UseOnStackReplacement) {
1819 // check for overflow against w0, which is the sum of the
1820 // counters
1821 __ ldrw(rscratch1, Address(rscratch2, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
1822 __ cmpw(r0, rscratch1);
1823 __ br(Assembler::HS, backedge_counter_overflow); // Intel == Assembler::aboveEqual
1824 }
1825 }
1826 }
1827 }
1828 __ bind(dispatch);
1829
1830 // Pre-load the next target bytecode into rscratch1
1831 __ load_unsigned_byte(rscratch1, Address(rbcp, 0));
1832
1833 // continue with the bytecode @ target
1834 // rscratch1: target bytecode
1835 // rbcp: target bcp
1836 __ dispatch_only(vtos);
1837
1838 if (UseLoopCounter) {
1839 if (ProfileInterpreter) {
1840 // Out-of-line code to allocate method data oop.
1841 __ bind(profile_method);
1842 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1843 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
1844 __ set_method_data_pointer_for_bcp();
1845 __ b(dispatch);
1846 }
1847
1848 if (TieredCompilation || UseOnStackReplacement) {
1849 // invocation counter overflow
1850 __ bind(backedge_counter_overflow);
1851 __ neg(r2, r2);
1852 __ add(r2, r2, rbcp); // branch bcp
1853 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
1854 __ call_VM(noreg,
1855 CAST_FROM_FN_PTR(address,
1856 InterpreterRuntime::frequency_counter_overflow),
1857 r2);
1858 if (!UseOnStackReplacement)
1859 __ b(dispatch);
1860 }
1861
1862 if (UseOnStackReplacement) {
1863 __ load_unsigned_byte(r1, Address(rbcp, 0)); // restore target bytecode
1864
1865 // r0: osr nmethod (osr ok) or NULL (osr not possible)
1866 // w1: target bytecode
1867 // r2: scratch
1868 __ cbz(r0, dispatch); // test result -- no osr if null
1869 // nmethod may have been invalidated (VM may block upon call_VM return)
1870 __ ldrb(r2, Address(r0, nmethod::state_offset()));
1871 if (nmethod::in_use != 0)
1872 __ sub(r2, r2, nmethod::in_use);
1873 __ cbnz(r2, dispatch);
1874
1875 // We have the address of an on stack replacement routine in r0
1876 // We need to prepare to execute the OSR method. First we must
1877 // migrate the locals and monitors off of the stack.
1878
1879 __ mov(r19, r0); // save the nmethod
1880
1881 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1882
1883 // r0 is OSR buffer, move it to expected parameter location
1884 __ mov(j_rarg0, r0);
1885
1886 // remove activation
1887 // get sender esp
1888 __ ldr(esp,
1889 Address(rfp, frame::interpreter_frame_sender_sp_offset * wordSize));
1890 // remove frame anchor
1891 __ leave();
1892 // Ensure compiled code always sees stack at proper alignment
1893 __ andr(sp, esp, -16);
1894
1895 // and begin the OSR nmethod
1896 __ ldr(rscratch1, Address(r19, nmethod::osr_entry_point_offset()));
1897 __ br(rscratch1);
1898 }
1899 }
1900 }
1901
1902
1903 void TemplateTable::if_0cmp(Condition cc)
1904 {
1905 transition(itos, vtos);
1906 // assume branch is more often taken than not (loops use backward branches)
1907 Label not_taken;
1908 if (cc == equal)
1909 __ cbnzw(r0, not_taken);
1910 else if (cc == not_equal)
1911 __ cbzw(r0, not_taken);
1912 else {
1913 __ andsw(zr, r0, r0);
1914 __ br(j_not(cc), not_taken);
1915 }
1916
1917 branch(false, false);
1918 __ bind(not_taken);
1919 __ profile_not_taken_branch(r0);
1920 }
1921
1922 void TemplateTable::if_icmp(Condition cc)
1923 {
1924 transition(itos, vtos);
1925 // assume branch is more often taken than not (loops use backward branches)
1926 Label not_taken;
1927 __ pop_i(r1);
1928 __ cmpw(r1, r0, Assembler::LSL);
1929 __ br(j_not(cc), not_taken);
1930 branch(false, false);
1931 __ bind(not_taken);
1932 __ profile_not_taken_branch(r0);
1933 }
1934
1935 void TemplateTable::if_nullcmp(Condition cc)
1936 {
1937 transition(atos, vtos);
1938 // assume branch is more often taken than not (loops use backward branches)
1939 Label not_taken;
1940 if (cc == equal)
1941 __ cbnz(r0, not_taken);
1942 else
1943 __ cbz(r0, not_taken);
1944 branch(false, false);
1945 __ bind(not_taken);
1946 __ profile_not_taken_branch(r0);
1947 }
1948
1949 void TemplateTable::if_acmp(Condition cc)
1950 {
1951 transition(atos, vtos);
1952 // assume branch is more often taken than not (loops use backward branches)
1953 Label not_taken;
1954 __ pop_ptr(r1);
1955 __ cmp(r1, r0);
1956 __ br(j_not(cc), not_taken);
1957 branch(false, false);
1958 __ bind(not_taken);
1959 __ profile_not_taken_branch(r0);
1960 }
1961
1962 void TemplateTable::ret() {
1963 transition(vtos, vtos);
1964 // We might be moving to a safepoint. The thread which calls
1965 // Interpreter::notice_safepoints() will effectively flush its cache
1966 // when it makes a system call, but we need to do something to
1967 // ensure that we see the changed dispatch table.
1968 __ membar(MacroAssembler::LoadLoad);
1969
1970 locals_index(r1);
1971 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1972 __ profile_ret(r1, r2);
1973 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1974 __ lea(rbcp, Address(rbcp, r1));
1975 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1976 __ dispatch_next(vtos);
1977 }
1978
1979 void TemplateTable::wide_ret() {
1980 transition(vtos, vtos);
1981 locals_index_wide(r1);
1982 __ ldr(r1, aaddress(r1)); // get return bci, compute return bcp
1983 __ profile_ret(r1, r2);
1984 __ ldr(rbcp, Address(rmethod, Method::const_offset()));
1985 __ lea(rbcp, Address(rbcp, r1));
1986 __ add(rbcp, rbcp, in_bytes(ConstMethod::codes_offset()));
1987 __ dispatch_next(vtos);
1988 }
1989
1990
1991 void TemplateTable::tableswitch() {
1992 Label default_case, continue_execution;
1993 transition(itos, vtos);
1994 // align rbcp
1995 __ lea(r1, at_bcp(BytesPerInt));
1996 __ andr(r1, r1, -BytesPerInt);
1997 // load lo & hi
1998 __ ldrw(r2, Address(r1, BytesPerInt));
1999 __ ldrw(r3, Address(r1, 2 * BytesPerInt));
2000 __ rev32(r2, r2);
2001 __ rev32(r3, r3);
2002 // check against lo & hi
2003 __ cmpw(r0, r2);
2004 __ br(Assembler::LT, default_case);
2005 __ cmpw(r0, r3);
2006 __ br(Assembler::GT, default_case);
2007 // lookup dispatch offset
2008 __ subw(r0, r0, r2);
2009 __ lea(r3, Address(r1, r0, Address::uxtw(2)));
2010 __ ldrw(r3, Address(r3, 3 * BytesPerInt));
2011 __ profile_switch_case(r0, r1, r2);
2012 // continue execution
2013 __ bind(continue_execution);
2014 __ rev32(r3, r3);
2015 __ load_unsigned_byte(rscratch1, Address(rbcp, r3, Address::sxtw(0)));
2016 __ add(rbcp, rbcp, r3, ext::sxtw);
2017 __ dispatch_only(vtos);
2018 // handle default
2019 __ bind(default_case);
2020 __ profile_switch_default(r0);
2021 __ ldrw(r3, Address(r1, 0));
2022 __ b(continue_execution);
2023 }
2024
2025 void TemplateTable::lookupswitch() {
2026 transition(itos, itos);
2027 __ stop("lookupswitch bytecode should have been rewritten");
2028 }
2029
2030 void TemplateTable::fast_linearswitch() {
2031 transition(itos, vtos);
2032 Label loop_entry, loop, found, continue_execution;
2033 // bswap r0 so we can avoid bswapping the table entries
2034 __ rev32(r0, r0);
2035 // align rbcp
2036 __ lea(r19, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2037 // this instruction (change offsets
2038 // below)
2039 __ andr(r19, r19, -BytesPerInt);
2040 // set counter
2041 __ ldrw(r1, Address(r19, BytesPerInt));
2042 __ rev32(r1, r1);
2043 __ b(loop_entry);
2044 // table search
2045 __ bind(loop);
2046 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2047 __ ldrw(rscratch1, Address(rscratch1, 2 * BytesPerInt));
2048 __ cmpw(r0, rscratch1);
2049 __ br(Assembler::EQ, found);
2050 __ bind(loop_entry);
2051 __ subs(r1, r1, 1);
2052 __ br(Assembler::PL, loop);
2053 // default case
2054 __ profile_switch_default(r0);
2055 __ ldrw(r3, Address(r19, 0));
2056 __ b(continue_execution);
2057 // entry found -> get offset
2058 __ bind(found);
2059 __ lea(rscratch1, Address(r19, r1, Address::lsl(3)));
2060 __ ldrw(r3, Address(rscratch1, 3 * BytesPerInt));
2061 __ profile_switch_case(r1, r0, r19);
2062 // continue execution
2063 __ bind(continue_execution);
2064 __ rev32(r3, r3);
2065 __ add(rbcp, rbcp, r3, ext::sxtw);
2066 __ ldrb(rscratch1, Address(rbcp, 0));
2067 __ dispatch_only(vtos);
2068 }
2069
2070 void TemplateTable::fast_binaryswitch() {
2071 transition(itos, vtos);
2072 // Implementation using the following core algorithm:
2073 //
2074 // int binary_search(int key, LookupswitchPair* array, int n) {
2075 // // Binary search according to "Methodik des Programmierens" by
2076 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2077 // int i = 0;
2078 // int j = n;
2079 // while (i+1 < j) {
2080 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2081 // // with Q: for all i: 0 <= i < n: key < a[i]
2082 // // where a stands for the array and assuming that the (inexisting)
2083 // // element a[n] is infinitely big.
2084 // int h = (i + j) >> 1;
2085 // // i < h < j
2086 // if (key < array[h].fast_match()) {
2087 // j = h;
2088 // } else {
2089 // i = h;
2090 // }
2091 // }
2092 // // R: a[i] <= key < a[i+1] or Q
2093 // // (i.e., if key is within array, i is the correct index)
2094 // return i;
2095 // }
2096
2097 // Register allocation
2098 const Register key = r0; // already set (tosca)
2099 const Register array = r1;
2100 const Register i = r2;
2101 const Register j = r3;
2102 const Register h = rscratch1;
2103 const Register temp = rscratch2;
2104
2105 // Find array start
2106 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2107 // get rid of this
2108 // instruction (change
2109 // offsets below)
2110 __ andr(array, array, -BytesPerInt);
2111
2112 // Initialize i & j
2113 __ mov(i, 0); // i = 0;
2114 __ ldrw(j, Address(array, -BytesPerInt)); // j = length(array);
2115
2116 // Convert j into native byteordering
2117 __ rev32(j, j);
2118
2119 // And start
2120 Label entry;
2121 __ b(entry);
2122
2123 // binary search loop
2124 {
2125 Label loop;
2126 __ bind(loop);
2127 // int h = (i + j) >> 1;
2128 __ addw(h, i, j); // h = i + j;
2129 __ lsrw(h, h, 1); // h = (i + j) >> 1;
2130 // if (key < array[h].fast_match()) {
2131 // j = h;
2132 // } else {
2133 // i = h;
2134 // }
2135 // Convert array[h].match to native byte-ordering before compare
2136 __ ldr(temp, Address(array, h, Address::lsl(3)));
2137 __ rev32(temp, temp);
2138 __ cmpw(key, temp);
2139 // j = h if (key < array[h].fast_match())
2140 __ csel(j, h, j, Assembler::LT);
2141 // i = h if (key >= array[h].fast_match())
2142 __ csel(i, h, i, Assembler::GE);
2143 // while (i+1 < j)
2144 __ bind(entry);
2145 __ addw(h, i, 1); // i+1
2146 __ cmpw(h, j); // i+1 < j
2147 __ br(Assembler::LT, loop);
2148 }
2149
2150 // end of binary search, result index is i (must check again!)
2151 Label default_case;
2152 // Convert array[i].match to native byte-ordering before compare
2153 __ ldr(temp, Address(array, i, Address::lsl(3)));
2154 __ rev32(temp, temp);
2155 __ cmpw(key, temp);
2156 __ br(Assembler::NE, default_case);
2157
2158 // entry found -> j = offset
2159 __ add(j, array, i, ext::uxtx, 3);
2160 __ ldrw(j, Address(j, BytesPerInt));
2161 __ profile_switch_case(i, key, array);
2162 __ rev32(j, j);
2163 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2164 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2165 __ dispatch_only(vtos);
2166
2167 // default case -> j = default offset
2168 __ bind(default_case);
2169 __ profile_switch_default(i);
2170 __ ldrw(j, Address(array, -2 * BytesPerInt));
2171 __ rev32(j, j);
2172 __ load_unsigned_byte(rscratch1, Address(rbcp, j, Address::sxtw(0)));
2173 __ lea(rbcp, Address(rbcp, j, Address::sxtw(0)));
2174 __ dispatch_only(vtos);
2175 }
2176
2177
2178 void TemplateTable::_return(TosState state)
2179 {
2180 transition(state, state);
2181 assert(_desc->calls_vm(),
2182 "inconsistent calls_vm information"); // call in remove_activation
2183
2184 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2185 assert(state == vtos, "only valid state");
2186
2187 __ ldr(c_rarg1, aaddress(0));
2188 __ load_klass(r3, c_rarg1);
2189 __ ldrw(r3, Address(r3, Klass::access_flags_offset()));
2190 Label skip_register_finalizer;
2191 __ tbz(r3, exact_log2(JVM_ACC_HAS_FINALIZER), skip_register_finalizer);
2192
2193 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), c_rarg1);
2194
2195 __ bind(skip_register_finalizer);
2196 }
2197
2198 // Issue a StoreStore barrier after all stores but before return
2199 // from any constructor for any class with a final field. We don't
2200 // know if this is a finalizer, so we always do so.
2201 if (_desc->bytecode() == Bytecodes::_return)
2202 __ membar(MacroAssembler::StoreStore);
2203
2204 // Narrow result if state is itos but result type is smaller.
2205 // Need to narrow in the return bytecode rather than in generate_return_entry
2206 // since compiled code callers expect the result to already be narrowed.
2207 if (state == itos) {
2208 __ narrow(r0);
2209 }
2210
2211 __ remove_activation(state);
2212 __ ret(lr);
2213 }
2214
2215 // ----------------------------------------------------------------------------
2216 // Volatile variables demand their effects be made known to all CPU's
2217 // in order. Store buffers on most chips allow reads & writes to
2218 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2219 // without some kind of memory barrier (i.e., it's not sufficient that
2220 // the interpreter does not reorder volatile references, the hardware
2221 // also must not reorder them).
2222 //
2223 // According to the new Java Memory Model (JMM):
2224 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2225 // writes act as aquire & release, so:
2226 // (2) A read cannot let unrelated NON-volatile memory refs that
2227 // happen after the read float up to before the read. It's OK for
2228 // non-volatile memory refs that happen before the volatile read to
2229 // float down below it.
2230 // (3) Similar a volatile write cannot let unrelated NON-volatile
2231 // memory refs that happen BEFORE the write float down to after the
2232 // write. It's OK for non-volatile memory refs that happen after the
2233 // volatile write to float up before it.
2234 //
2235 // We only put in barriers around volatile refs (they are expensive),
2236 // not _between_ memory refs (that would require us to track the
2237 // flavor of the previous memory refs). Requirements (2) and (3)
2238 // require some barriers before volatile stores and after volatile
2239 // loads. These nearly cover requirement (1) but miss the
2240 // volatile-store-volatile-load case. This final case is placed after
2241 // volatile-stores although it could just as well go before
2242 // volatile-loads.
2243
2244 void TemplateTable::resolve_cache_and_index(int byte_no,
2245 Register Rcache,
2246 Register index,
2247 size_t index_size) {
2248 const Register temp = r19;
2249 assert_different_registers(Rcache, index, temp);
2250
2251 Label resolved;
2252
2253 Bytecodes::Code code = bytecode();
2254 switch (code) {
2255 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2256 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2257 }
2258
2259 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2260 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2261 __ cmp(temp, (int) code); // have we resolved this bytecode?
2262 __ br(Assembler::EQ, resolved);
2263
2264 // resolve first time through
2265 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2266 __ mov(temp, (int) code);
2267 __ call_VM(noreg, entry, temp);
2268
2269 // Update registers with resolved info
2270 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2271 // n.b. unlike x86 Rcache is now rcpool plus the indexed offset
2272 // so all clients ofthis method must be modified accordingly
2273 __ bind(resolved);
2274 }
2275
2276 // The Rcache and index registers must be set before call
2277 // n.b unlike x86 cache already includes the index offset
2278 void TemplateTable::load_field_cp_cache_entry(Register obj,
2279 Register cache,
2280 Register index,
2281 Register off,
2282 Register flags,
2283 bool is_static = false) {
2284 assert_different_registers(cache, index, flags, off);
2285
2286 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2287 // Field offset
2288 __ ldr(off, Address(cache, in_bytes(cp_base_offset +
2289 ConstantPoolCacheEntry::f2_offset())));
2290 // Flags
2291 __ ldrw(flags, Address(cache, in_bytes(cp_base_offset +
2292 ConstantPoolCacheEntry::flags_offset())));
2293
2294 // klass overwrite register
2295 if (is_static) {
2296 __ ldr(obj, Address(cache, in_bytes(cp_base_offset +
2297 ConstantPoolCacheEntry::f1_offset())));
2298 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2299 __ ldr(obj, Address(obj, mirror_offset));
2300 }
2301 }
2302
2303 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2304 Register method,
2305 Register itable_index,
2306 Register flags,
2307 bool is_invokevirtual,
2308 bool is_invokevfinal, /*unused*/
2309 bool is_invokedynamic) {
2310 // setup registers
2311 const Register cache = rscratch2;
2312 const Register index = r4;
2313 assert_different_registers(method, flags);
2314 assert_different_registers(method, cache, index);
2315 assert_different_registers(itable_index, flags);
2316 assert_different_registers(itable_index, cache, index);
2317 // determine constant pool cache field offsets
2318 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2319 const int method_offset = in_bytes(
2320 ConstantPoolCache::base_offset() +
2321 (is_invokevirtual
2322 ? ConstantPoolCacheEntry::f2_offset()
2323 : ConstantPoolCacheEntry::f1_offset()));
2324 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2325 ConstantPoolCacheEntry::flags_offset());
2326 // access constant pool cache fields
2327 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2328 ConstantPoolCacheEntry::f2_offset());
2329
2330 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2331 resolve_cache_and_index(byte_no, cache, index, index_size);
2332 __ ldr(method, Address(cache, method_offset));
2333
2334 if (itable_index != noreg) {
2335 __ ldr(itable_index, Address(cache, index_offset));
2336 }
2337 __ ldrw(flags, Address(cache, flags_offset));
2338 }
2339
2340
2341 // The registers cache and index expected to be set before call.
2342 // Correct values of the cache and index registers are preserved.
2343 void TemplateTable::jvmti_post_field_access(Register cache, Register index,
2344 bool is_static, bool has_tos) {
2345 // do the JVMTI work here to avoid disturbing the register state below
2346 // We use c_rarg registers here because we want to use the register used in
2347 // the call to the VM
2348 if (JvmtiExport::can_post_field_access()) {
2349 // Check to see if a field access watch has been set before we
2350 // take the time to call into the VM.
2351 Label L1;
2352 assert_different_registers(cache, index, r0);
2353 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2354 __ ldrw(r0, Address(rscratch1));
2355 __ cbzw(r0, L1);
2356
2357 __ get_cache_and_index_at_bcp(c_rarg2, c_rarg3, 1);
2358 __ lea(c_rarg2, Address(c_rarg2, in_bytes(ConstantPoolCache::base_offset())));
2359
2360 if (is_static) {
2361 __ mov(c_rarg1, zr); // NULL object reference
2362 } else {
2363 __ ldr(c_rarg1, at_tos()); // get object pointer without popping it
2364 __ verify_oop(c_rarg1);
2365 }
2366 // c_rarg1: object pointer or NULL
2367 // c_rarg2: cache entry pointer
2368 // c_rarg3: jvalue object on the stack
2369 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2370 InterpreterRuntime::post_field_access),
2371 c_rarg1, c_rarg2, c_rarg3);
2372 __ get_cache_and_index_at_bcp(cache, index, 1);
2373 __ bind(L1);
2374 }
2375 }
2376
2377 void TemplateTable::pop_and_check_object(Register r)
2378 {
2379 __ pop_ptr(r);
2380 __ null_check(r); // for field access must check obj.
2381 __ verify_oop(r);
2382 }
2383
2384 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc)
2385 {
2386 const Register cache = r2;
2387 const Register index = r3;
2388 const Register obj = r4;
2389 const Register off = r19;
2390 const Register flags = r0;
2391 const Register raw_flags = r6;
2392 const Register bc = r4; // uses same reg as obj, so don't mix them
2393
2394 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2395 jvmti_post_field_access(cache, index, is_static, false);
2396 load_field_cp_cache_entry(obj, cache, index, off, raw_flags, is_static);
2397
2398 if (!is_static) {
2399 // obj is on the stack
2400 pop_and_check_object(obj);
2401 }
2402
2403 // 8179954: We need to make sure that the code generated for
2404 // volatile accesses forms a sequentially-consistent set of
2405 // operations when combined with STLR and LDAR. Without a leading
2406 // membar it's possible for a simple Dekker test to fail if loads
2407 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
2408 // the stores in one method and we interpret the loads in another.
2409 if (! UseBarriersForVolatile) {
2410 Label notVolatile;
2411 __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2412 __ membar(MacroAssembler::AnyAny);
2413 __ bind(notVolatile);
2414 }
2415
2416 const Address field(obj, off);
2417
2418 Label Done, notByte, notBool, notInt, notShort, notChar,
2419 notLong, notFloat, notObj, notDouble;
2420
2421 // x86 uses a shift and mask or wings it with a shift plus assert
2422 // the mask is not needed. aarch64 just uses bitfield extract
2423 __ ubfxw(flags, raw_flags, ConstantPoolCacheEntry::tos_state_shift,
2424 ConstantPoolCacheEntry::tos_state_bits);
2425
2426 assert(btos == 0, "change code, btos != 0");
2427 __ cbnz(flags, notByte);
2428
2429 // Don't rewrite getstatic, only getfield
2430 if (is_static) rc = may_not_rewrite;
2431
2432 // btos
2433 __ load_signed_byte(r0, field);
2434 __ push(btos);
2435 // Rewrite bytecode to be faster
2436 if (rc == may_rewrite) {
2437 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2438 }
2439 __ b(Done);
2440
2441 __ bind(notByte);
2442 __ cmp(flags, ztos);
2443 __ br(Assembler::NE, notBool);
2444
2445 // ztos (same code as btos)
2446 __ ldrsb(r0, field);
2447 __ push(ztos);
2448 // Rewrite bytecode to be faster
2449 if (rc == may_rewrite) {
2450 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2451 patch_bytecode(Bytecodes::_fast_bgetfield, bc, r1);
2452 }
2453 __ b(Done);
2454
2455 __ bind(notBool);
2456 __ cmp(flags, atos);
2457 __ br(Assembler::NE, notObj);
2458 // atos
2459 __ load_heap_oop(r0, field);
2460 __ push(atos);
2461 if (rc == may_rewrite) {
2462 patch_bytecode(Bytecodes::_fast_agetfield, bc, r1);
2463 }
2464 __ b(Done);
2465
2466 __ bind(notObj);
2467 __ cmp(flags, itos);
2468 __ br(Assembler::NE, notInt);
2469 // itos
2470 __ ldrw(r0, field);
2471 __ push(itos);
2472 // Rewrite bytecode to be faster
2473 if (rc == may_rewrite) {
2474 patch_bytecode(Bytecodes::_fast_igetfield, bc, r1);
2475 }
2476 __ b(Done);
2477
2478 __ bind(notInt);
2479 __ cmp(flags, ctos);
2480 __ br(Assembler::NE, notChar);
2481 // ctos
2482 __ load_unsigned_short(r0, field);
2483 __ push(ctos);
2484 // Rewrite bytecode to be faster
2485 if (rc == may_rewrite) {
2486 patch_bytecode(Bytecodes::_fast_cgetfield, bc, r1);
2487 }
2488 __ b(Done);
2489
2490 __ bind(notChar);
2491 __ cmp(flags, stos);
2492 __ br(Assembler::NE, notShort);
2493 // stos
2494 __ load_signed_short(r0, field);
2495 __ push(stos);
2496 // Rewrite bytecode to be faster
2497 if (rc == may_rewrite) {
2498 patch_bytecode(Bytecodes::_fast_sgetfield, bc, r1);
2499 }
2500 __ b(Done);
2501
2502 __ bind(notShort);
2503 __ cmp(flags, ltos);
2504 __ br(Assembler::NE, notLong);
2505 // ltos
2506 __ ldr(r0, field);
2507 __ push(ltos);
2508 // Rewrite bytecode to be faster
2509 if (rc == may_rewrite) {
2510 patch_bytecode(Bytecodes::_fast_lgetfield, bc, r1);
2511 }
2512 __ b(Done);
2513
2514 __ bind(notLong);
2515 __ cmp(flags, ftos);
2516 __ br(Assembler::NE, notFloat);
2517 // ftos
2518 __ ldrs(v0, field);
2519 __ push(ftos);
2520 // Rewrite bytecode to be faster
2521 if (rc == may_rewrite) {
2522 patch_bytecode(Bytecodes::_fast_fgetfield, bc, r1);
2523 }
2524 __ b(Done);
2525
2526 __ bind(notFloat);
2527 #ifdef ASSERT
2528 __ cmp(flags, dtos);
2529 __ br(Assembler::NE, notDouble);
2530 #endif
2531 // dtos
2532 __ ldrd(v0, field);
2533 __ push(dtos);
2534 // Rewrite bytecode to be faster
2535 if (rc == may_rewrite) {
2536 patch_bytecode(Bytecodes::_fast_dgetfield, bc, r1);
2537 }
2538 #ifdef ASSERT
2539 __ b(Done);
2540
2541 __ bind(notDouble);
2542 __ stop("Bad state");
2543 #endif
2544
2545 __ bind(Done);
2546
2547 Label notVolatile;
2548 __ tbz(raw_flags, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2549 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
2550 __ bind(notVolatile);
2551 }
2552
2553
2554 void TemplateTable::getfield(int byte_no)
2555 {
2556 getfield_or_static(byte_no, false);
2557 }
2558
2559 void TemplateTable::nofast_getfield(int byte_no) {
2560 getfield_or_static(byte_no, false, may_not_rewrite);
2561 }
2562
2563 void TemplateTable::getstatic(int byte_no)
2564 {
2565 getfield_or_static(byte_no, true);
2566 }
2567
2568 // The registers cache and index expected to be set before call.
2569 // The function may destroy various registers, just not the cache and index registers.
2570 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2571 transition(vtos, vtos);
2572
2573 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2574
2575 if (JvmtiExport::can_post_field_modification()) {
2576 // Check to see if a field modification watch has been set before
2577 // we take the time to call into the VM.
2578 Label L1;
2579 assert_different_registers(cache, index, r0);
2580 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2581 __ ldrw(r0, Address(rscratch1));
2582 __ cbz(r0, L1);
2583
2584 __ get_cache_and_index_at_bcp(c_rarg2, rscratch1, 1);
2585
2586 if (is_static) {
2587 // Life is simple. Null out the object pointer.
2588 __ mov(c_rarg1, zr);
2589 } else {
2590 // Life is harder. The stack holds the value on top, followed by
2591 // the object. We don't know the size of the value, though; it
2592 // could be one or two words depending on its type. As a result,
2593 // we must find the type to determine where the object is.
2594 __ ldrw(c_rarg3, Address(c_rarg2,
2595 in_bytes(cp_base_offset +
2596 ConstantPoolCacheEntry::flags_offset())));
2597 __ lsr(c_rarg3, c_rarg3,
2598 ConstantPoolCacheEntry::tos_state_shift);
2599 ConstantPoolCacheEntry::verify_tos_state_shift();
2600 Label nope2, done, ok;
2601 __ ldr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2602 __ cmpw(c_rarg3, ltos);
2603 __ br(Assembler::EQ, ok);
2604 __ cmpw(c_rarg3, dtos);
2605 __ br(Assembler::NE, nope2);
2606 __ bind(ok);
2607 __ ldr(c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2608 __ bind(nope2);
2609 }
2610 // cache entry pointer
2611 __ add(c_rarg2, c_rarg2, in_bytes(cp_base_offset));
2612 // object (tos)
2613 __ mov(c_rarg3, esp);
2614 // c_rarg1: object pointer set up above (NULL if static)
2615 // c_rarg2: cache entry pointer
2616 // c_rarg3: jvalue object on the stack
2617 __ call_VM(noreg,
2618 CAST_FROM_FN_PTR(address,
2619 InterpreterRuntime::post_field_modification),
2620 c_rarg1, c_rarg2, c_rarg3);
2621 __ get_cache_and_index_at_bcp(cache, index, 1);
2622 __ bind(L1);
2623 }
2624 }
2625
2626 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2627 transition(vtos, vtos);
2628
2629 const Register cache = r2;
2630 const Register index = r3;
2631 const Register obj = r2;
2632 const Register off = r19;
2633 const Register flags = r0;
2634 const Register bc = r4;
2635
2636 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2637 jvmti_post_field_mod(cache, index, is_static);
2638 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2639
2640 Label Done;
2641 __ mov(r5, flags);
2642
2643 {
2644 Label notVolatile;
2645 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2646 __ membar(MacroAssembler::StoreStore);
2647 __ bind(notVolatile);
2648 }
2649
2650 // field address
2651 const Address field(obj, off);
2652
2653 Label notByte, notBool, notInt, notShort, notChar,
2654 notLong, notFloat, notObj, notDouble;
2655
2656 // x86 uses a shift and mask or wings it with a shift plus assert
2657 // the mask is not needed. aarch64 just uses bitfield extract
2658 __ ubfxw(flags, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
2659
2660 assert(btos == 0, "change code, btos != 0");
2661 __ cbnz(flags, notByte);
2662
2663 // Don't rewrite putstatic, only putfield
2664 if (is_static) rc = may_not_rewrite;
2665
2666 // btos
2667 {
2668 __ pop(btos);
2669 if (!is_static) pop_and_check_object(obj);
2670 __ strb(r0, field);
2671 if (rc == may_rewrite) {
2672 patch_bytecode(Bytecodes::_fast_bputfield, bc, r1, true, byte_no);
2673 }
2674 __ b(Done);
2675 }
2676
2677 __ bind(notByte);
2678 __ cmp(flags, ztos);
2679 __ br(Assembler::NE, notBool);
2680
2681 // ztos
2682 {
2683 __ pop(ztos);
2684 if (!is_static) pop_and_check_object(obj);
2685 __ andw(r0, r0, 0x1);
2686 __ strb(r0, field);
2687 if (rc == may_rewrite) {
2688 patch_bytecode(Bytecodes::_fast_zputfield, bc, r1, true, byte_no);
2689 }
2690 __ b(Done);
2691 }
2692
2693 __ bind(notBool);
2694 __ cmp(flags, atos);
2695 __ br(Assembler::NE, notObj);
2696
2697 // atos
2698 {
2699 __ pop(atos);
2700 if (!is_static) pop_and_check_object(obj);
2701 // Store into the field
2702 do_oop_store(_masm, field, r0, _bs->kind(), false);
2703 if (rc == may_rewrite) {
2704 patch_bytecode(Bytecodes::_fast_aputfield, bc, r1, true, byte_no);
2705 }
2706 __ b(Done);
2707 }
2708
2709 __ bind(notObj);
2710 __ cmp(flags, itos);
2711 __ br(Assembler::NE, notInt);
2712
2713 // itos
2714 {
2715 __ pop(itos);
2716 if (!is_static) pop_and_check_object(obj);
2717 __ strw(r0, field);
2718 if (rc == may_rewrite) {
2719 patch_bytecode(Bytecodes::_fast_iputfield, bc, r1, true, byte_no);
2720 }
2721 __ b(Done);
2722 }
2723
2724 __ bind(notInt);
2725 __ cmp(flags, ctos);
2726 __ br(Assembler::NE, notChar);
2727
2728 // ctos
2729 {
2730 __ pop(ctos);
2731 if (!is_static) pop_and_check_object(obj);
2732 __ strh(r0, field);
2733 if (rc == may_rewrite) {
2734 patch_bytecode(Bytecodes::_fast_cputfield, bc, r1, true, byte_no);
2735 }
2736 __ b(Done);
2737 }
2738
2739 __ bind(notChar);
2740 __ cmp(flags, stos);
2741 __ br(Assembler::NE, notShort);
2742
2743 // stos
2744 {
2745 __ pop(stos);
2746 if (!is_static) pop_and_check_object(obj);
2747 __ strh(r0, field);
2748 if (rc == may_rewrite) {
2749 patch_bytecode(Bytecodes::_fast_sputfield, bc, r1, true, byte_no);
2750 }
2751 __ b(Done);
2752 }
2753
2754 __ bind(notShort);
2755 __ cmp(flags, ltos);
2756 __ br(Assembler::NE, notLong);
2757
2758 // ltos
2759 {
2760 __ pop(ltos);
2761 if (!is_static) pop_and_check_object(obj);
2762 __ str(r0, field);
2763 if (rc == may_rewrite) {
2764 patch_bytecode(Bytecodes::_fast_lputfield, bc, r1, true, byte_no);
2765 }
2766 __ b(Done);
2767 }
2768
2769 __ bind(notLong);
2770 __ cmp(flags, ftos);
2771 __ br(Assembler::NE, notFloat);
2772
2773 // ftos
2774 {
2775 __ pop(ftos);
2776 if (!is_static) pop_and_check_object(obj);
2777 __ strs(v0, field);
2778 if (rc == may_rewrite) {
2779 patch_bytecode(Bytecodes::_fast_fputfield, bc, r1, true, byte_no);
2780 }
2781 __ b(Done);
2782 }
2783
2784 __ bind(notFloat);
2785 #ifdef ASSERT
2786 __ cmp(flags, dtos);
2787 __ br(Assembler::NE, notDouble);
2788 #endif
2789
2790 // dtos
2791 {
2792 __ pop(dtos);
2793 if (!is_static) pop_and_check_object(obj);
2794 __ strd(v0, field);
2795 if (rc == may_rewrite) {
2796 patch_bytecode(Bytecodes::_fast_dputfield, bc, r1, true, byte_no);
2797 }
2798 }
2799
2800 #ifdef ASSERT
2801 __ b(Done);
2802
2803 __ bind(notDouble);
2804 __ stop("Bad state");
2805 #endif
2806
2807 __ bind(Done);
2808
2809 {
2810 Label notVolatile;
2811 __ tbz(r5, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2812 __ membar(MacroAssembler::StoreLoad);
2813 __ bind(notVolatile);
2814 }
2815 }
2816
2817 void TemplateTable::putfield(int byte_no)
2818 {
2819 putfield_or_static(byte_no, false);
2820 }
2821
2822 void TemplateTable::nofast_putfield(int byte_no) {
2823 putfield_or_static(byte_no, false, may_not_rewrite);
2824 }
2825
2826 void TemplateTable::putstatic(int byte_no) {
2827 putfield_or_static(byte_no, true);
2828 }
2829
2830 void TemplateTable::jvmti_post_fast_field_mod()
2831 {
2832 if (JvmtiExport::can_post_field_modification()) {
2833 // Check to see if a field modification watch has been set before
2834 // we take the time to call into the VM.
2835 Label L2;
2836 __ lea(rscratch1, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2837 __ ldrw(c_rarg3, Address(rscratch1));
2838 __ cbzw(c_rarg3, L2);
2839 __ pop_ptr(r19); // copy the object pointer from tos
2840 __ verify_oop(r19);
2841 __ push_ptr(r19); // put the object pointer back on tos
2842 // Save tos values before call_VM() clobbers them. Since we have
2843 // to do it for every data type, we use the saved values as the
2844 // jvalue object.
2845 switch (bytecode()) { // load values into the jvalue object
2846 case Bytecodes::_fast_aputfield: __ push_ptr(r0); break;
2847 case Bytecodes::_fast_bputfield: // fall through
2848 case Bytecodes::_fast_zputfield: // fall through
2849 case Bytecodes::_fast_sputfield: // fall through
2850 case Bytecodes::_fast_cputfield: // fall through
2851 case Bytecodes::_fast_iputfield: __ push_i(r0); break;
2852 case Bytecodes::_fast_dputfield: __ push_d(); break;
2853 case Bytecodes::_fast_fputfield: __ push_f(); break;
2854 case Bytecodes::_fast_lputfield: __ push_l(r0); break;
2855
2856 default:
2857 ShouldNotReachHere();
2858 }
2859 __ mov(c_rarg3, esp); // points to jvalue on the stack
2860 // access constant pool cache entry
2861 __ get_cache_entry_pointer_at_bcp(c_rarg2, r0, 1);
2862 __ verify_oop(r19);
2863 // r19: object pointer copied above
2864 // c_rarg2: cache entry pointer
2865 // c_rarg3: jvalue object on the stack
2866 __ call_VM(noreg,
2867 CAST_FROM_FN_PTR(address,
2868 InterpreterRuntime::post_field_modification),
2869 r19, c_rarg2, c_rarg3);
2870
2871 switch (bytecode()) { // restore tos values
2872 case Bytecodes::_fast_aputfield: __ pop_ptr(r0); break;
2873 case Bytecodes::_fast_bputfield: // fall through
2874 case Bytecodes::_fast_zputfield: // fall through
2875 case Bytecodes::_fast_sputfield: // fall through
2876 case Bytecodes::_fast_cputfield: // fall through
2877 case Bytecodes::_fast_iputfield: __ pop_i(r0); break;
2878 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2879 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2880 case Bytecodes::_fast_lputfield: __ pop_l(r0); break;
2881 }
2882 __ bind(L2);
2883 }
2884 }
2885
2886 void TemplateTable::fast_storefield(TosState state)
2887 {
2888 transition(state, vtos);
2889
2890 ByteSize base = ConstantPoolCache::base_offset();
2891
2892 jvmti_post_fast_field_mod();
2893
2894 // access constant pool cache
2895 __ get_cache_and_index_at_bcp(r2, r1, 1);
2896
2897 // test for volatile with r3
2898 __ ldrw(r3, Address(r2, in_bytes(base +
2899 ConstantPoolCacheEntry::flags_offset())));
2900
2901 // replace index with field offset from cache entry
2902 __ ldr(r1, Address(r2, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2903
2904 {
2905 Label notVolatile;
2906 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2907 __ membar(MacroAssembler::StoreStore);
2908 __ bind(notVolatile);
2909 }
2910
2911 Label notVolatile;
2912
2913 // Get object from stack
2914 pop_and_check_object(r2);
2915
2916 // field address
2917 const Address field(r2, r1);
2918
2919 // access field
2920 switch (bytecode()) {
2921 case Bytecodes::_fast_aputfield:
2922 do_oop_store(_masm, field, r0, _bs->kind(), false);
2923 break;
2924 case Bytecodes::_fast_lputfield:
2925 __ str(r0, field);
2926 break;
2927 case Bytecodes::_fast_iputfield:
2928 __ strw(r0, field);
2929 break;
2930 case Bytecodes::_fast_zputfield:
2931 __ andw(r0, r0, 0x1); // boolean is true if LSB is 1
2932 // fall through to bputfield
2933 case Bytecodes::_fast_bputfield:
2934 __ strb(r0, field);
2935 break;
2936 case Bytecodes::_fast_sputfield:
2937 // fall through
2938 case Bytecodes::_fast_cputfield:
2939 __ strh(r0, field);
2940 break;
2941 case Bytecodes::_fast_fputfield:
2942 __ strs(v0, field);
2943 break;
2944 case Bytecodes::_fast_dputfield:
2945 __ strd(v0, field);
2946 break;
2947 default:
2948 ShouldNotReachHere();
2949 }
2950
2951 {
2952 Label notVolatile;
2953 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
2954 __ membar(MacroAssembler::StoreLoad);
2955 __ bind(notVolatile);
2956 }
2957 }
2958
2959
2960 void TemplateTable::fast_accessfield(TosState state)
2961 {
2962 transition(atos, state);
2963 // Do the JVMTI work here to avoid disturbing the register state below
2964 if (JvmtiExport::can_post_field_access()) {
2965 // Check to see if a field access watch has been set before we
2966 // take the time to call into the VM.
2967 Label L1;
2968 __ lea(rscratch1, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2969 __ ldrw(r2, Address(rscratch1));
2970 __ cbzw(r2, L1);
2971 // access constant pool cache entry
2972 __ get_cache_entry_pointer_at_bcp(c_rarg2, rscratch2, 1);
2973 __ verify_oop(r0);
2974 __ push_ptr(r0); // save object pointer before call_VM() clobbers it
2975 __ mov(c_rarg1, r0);
2976 // c_rarg1: object pointer copied above
2977 // c_rarg2: cache entry pointer
2978 __ call_VM(noreg,
2979 CAST_FROM_FN_PTR(address,
2980 InterpreterRuntime::post_field_access),
2981 c_rarg1, c_rarg2);
2982 __ pop_ptr(r0); // restore object pointer
2983 __ bind(L1);
2984 }
2985
2986 // access constant pool cache
2987 __ get_cache_and_index_at_bcp(r2, r1, 1);
2988 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2989 ConstantPoolCacheEntry::f2_offset())));
2990 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
2991 ConstantPoolCacheEntry::flags_offset())));
2992
2993 // r0: object
2994 __ verify_oop(r0);
2995 __ null_check(r0);
2996 const Address field(r0, r1);
2997
2998 // 8179954: We need to make sure that the code generated for
2999 // volatile accesses forms a sequentially-consistent set of
3000 // operations when combined with STLR and LDAR. Without a leading
3001 // membar it's possible for a simple Dekker test to fail if loads
3002 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
3003 // the stores in one method and we interpret the loads in another.
3004 if (! UseBarriersForVolatile) {
3005 Label notVolatile;
3006 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3007 __ membar(MacroAssembler::AnyAny);
3008 __ bind(notVolatile);
3009 }
3010
3011 // access field
3012 switch (bytecode()) {
3013 case Bytecodes::_fast_agetfield:
3014 __ load_heap_oop(r0, field);
3015 __ verify_oop(r0);
3016 break;
3017 case Bytecodes::_fast_lgetfield:
3018 __ ldr(r0, field);
3019 break;
3020 case Bytecodes::_fast_igetfield:
3021 __ ldrw(r0, field);
3022 break;
3023 case Bytecodes::_fast_bgetfield:
3024 __ load_signed_byte(r0, field);
3025 break;
3026 case Bytecodes::_fast_sgetfield:
3027 __ load_signed_short(r0, field);
3028 break;
3029 case Bytecodes::_fast_cgetfield:
3030 __ load_unsigned_short(r0, field);
3031 break;
3032 case Bytecodes::_fast_fgetfield:
3033 __ ldrs(v0, field);
3034 break;
3035 case Bytecodes::_fast_dgetfield:
3036 __ ldrd(v0, field);
3037 break;
3038 default:
3039 ShouldNotReachHere();
3040 }
3041 {
3042 Label notVolatile;
3043 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3044 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3045 __ bind(notVolatile);
3046 }
3047 }
3048
3049 void TemplateTable::fast_xaccess(TosState state)
3050 {
3051 transition(vtos, state);
3052
3053 // get receiver
3054 __ ldr(r0, aaddress(0));
3055 // access constant pool cache
3056 __ get_cache_and_index_at_bcp(r2, r3, 2);
3057 __ ldr(r1, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3058 ConstantPoolCacheEntry::f2_offset())));
3059
3060 // 8179954: We need to make sure that the code generated for
3061 // volatile accesses forms a sequentially-consistent set of
3062 // operations when combined with STLR and LDAR. Without a leading
3063 // membar it's possible for a simple Dekker test to fail if loads
3064 // use LDR;DMB but stores use STLR. This can happen if C2 compiles
3065 // the stores in one method and we interpret the loads in another.
3066 if (! UseBarriersForVolatile) {
3067 Label notVolatile;
3068 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3069 ConstantPoolCacheEntry::flags_offset())));
3070 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3071 __ membar(MacroAssembler::AnyAny);
3072 __ bind(notVolatile);
3073 }
3074
3075 // make sure exception is reported in correct bcp range (getfield is
3076 // next instruction)
3077 __ increment(rbcp);
3078 __ null_check(r0);
3079 switch (state) {
3080 case itos:
3081 __ ldrw(r0, Address(r0, r1, Address::lsl(0)));
3082 break;
3083 case atos:
3084 __ load_heap_oop(r0, Address(r0, r1, Address::lsl(0)));
3085 __ verify_oop(r0);
3086 break;
3087 case ftos:
3088 __ ldrs(v0, Address(r0, r1, Address::lsl(0)));
3089 break;
3090 default:
3091 ShouldNotReachHere();
3092 }
3093
3094 {
3095 Label notVolatile;
3096 __ ldrw(r3, Address(r2, in_bytes(ConstantPoolCache::base_offset() +
3097 ConstantPoolCacheEntry::flags_offset())));
3098 __ tbz(r3, ConstantPoolCacheEntry::is_volatile_shift, notVolatile);
3099 __ membar(MacroAssembler::LoadLoad | MacroAssembler::LoadStore);
3100 __ bind(notVolatile);
3101 }
3102
3103 __ decrement(rbcp);
3104 }
3105
3106
3107
3108 //-----------------------------------------------------------------------------
3109 // Calls
3110
3111 void TemplateTable::count_calls(Register method, Register temp)
3112 {
3113 __ call_Unimplemented();
3114 }
3115
3116 void TemplateTable::prepare_invoke(int byte_no,
3117 Register method, // linked method (or i-klass)
3118 Register index, // itable index, MethodType, etc.
3119 Register recv, // if caller wants to see it
3120 Register flags // if caller wants to test it
3121 ) {
3122 // determine flags
3123 Bytecodes::Code code = bytecode();
3124 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3125 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3126 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3127 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3128 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3129 const bool load_receiver = (recv != noreg);
3130 const bool save_flags = (flags != noreg);
3131 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3132 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3133 assert(flags == noreg || flags == r3, "");
3134 assert(recv == noreg || recv == r2, "");
3135
3136 // setup registers & access constant pool cache
3137 if (recv == noreg) recv = r2;
3138 if (flags == noreg) flags = r3;
3139 assert_different_registers(method, index, recv, flags);
3140
3141 // save 'interpreter return address'
3142 __ save_bcp();
3143
3144 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3145
3146 // maybe push appendix to arguments (just before return address)
3147 if (is_invokedynamic || is_invokehandle) {
3148 Label L_no_push;
3149 __ tbz(flags, ConstantPoolCacheEntry::has_appendix_shift, L_no_push);
3150 // Push the appendix as a trailing parameter.
3151 // This must be done before we get the receiver,
3152 // since the parameter_size includes it.
3153 __ push(r19);
3154 __ mov(r19, index);
3155 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3156 __ load_resolved_reference_at_index(index, r19);
3157 __ pop(r19);
3158 __ push(index); // push appendix (MethodType, CallSite, etc.)
3159 __ bind(L_no_push);
3160 }
3161
3162 // load receiver if needed (note: no return address pushed yet)
3163 if (load_receiver) {
3164 __ andw(recv, flags, ConstantPoolCacheEntry::parameter_size_mask);
3165 // FIXME -- is this actually correct? looks like it should be 2
3166 // const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3167 // const int receiver_is_at_end = -1; // back off one slot to get receiver
3168 // Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3169 // __ movptr(recv, recv_addr);
3170 __ add(rscratch1, esp, recv, ext::uxtx, 3); // FIXME: uxtb here?
3171 __ ldr(recv, Address(rscratch1, -Interpreter::expr_offset_in_bytes(1)));
3172 __ verify_oop(recv);
3173 }
3174
3175 // compute return type
3176 // x86 uses a shift and mask or wings it with a shift plus assert
3177 // the mask is not needed. aarch64 just uses bitfield extract
3178 __ ubfxw(rscratch2, flags, ConstantPoolCacheEntry::tos_state_shift, ConstantPoolCacheEntry::tos_state_bits);
3179 // load return address
3180 {
3181 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3182 __ mov(rscratch1, table_addr);
3183 __ ldr(lr, Address(rscratch1, rscratch2, Address::lsl(3)));
3184 }
3185 }
3186
3187
3188 void TemplateTable::invokevirtual_helper(Register index,
3189 Register recv,
3190 Register flags)
3191 {
3192 // Uses temporary registers r0, r3
3193 assert_different_registers(index, recv, r0, r3);
3194 // Test for an invoke of a final method
3195 Label notFinal;
3196 __ tbz(flags, ConstantPoolCacheEntry::is_vfinal_shift, notFinal);
3197
3198 const Register method = index; // method must be rmethod
3199 assert(method == rmethod,
3200 "methodOop must be rmethod for interpreter calling convention");
3201
3202 // do the call - the index is actually the method to call
3203 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3204
3205 // It's final, need a null check here!
3206 __ null_check(recv);
3207
3208 // profile this call
3209 __ profile_final_call(r0);
3210 __ profile_arguments_type(r0, method, r4, true);
3211
3212 __ jump_from_interpreted(method, r0);
3213
3214 __ bind(notFinal);
3215
3216 // get receiver klass
3217 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3218 __ load_klass(r0, recv);
3219
3220 // profile this call
3221 __ profile_virtual_call(r0, rlocals, r3);
3222
3223 // get target methodOop & entry point
3224 __ lookup_virtual_method(r0, index, method);
3225 __ profile_arguments_type(r3, method, r4, true);
3226 // FIXME -- this looks completely redundant. is it?
3227 // __ ldr(r3, Address(method, Method::interpreter_entry_offset()));
3228 __ jump_from_interpreted(method, r3);
3229 }
3230
3231 void TemplateTable::invokevirtual(int byte_no)
3232 {
3233 transition(vtos, vtos);
3234 assert(byte_no == f2_byte, "use this argument");
3235
3236 prepare_invoke(byte_no, rmethod, noreg, r2, r3);
3237
3238 // rmethod: index (actually a Method*)
3239 // r2: receiver
3240 // r3: flags
3241
3242 invokevirtual_helper(rmethod, r2, r3);
3243 }
3244
3245 void TemplateTable::invokespecial(int byte_no)
3246 {
3247 transition(vtos, vtos);
3248 assert(byte_no == f1_byte, "use this argument");
3249
3250 prepare_invoke(byte_no, rmethod, noreg, // get f1 Method*
3251 r2); // get receiver also for null check
3252 __ verify_oop(r2);
3253 __ null_check(r2);
3254 // do the call
3255 __ profile_call(r0);
3256 __ profile_arguments_type(r0, rmethod, rbcp, false);
3257 __ jump_from_interpreted(rmethod, r0);
3258 }
3259
3260 void TemplateTable::invokestatic(int byte_no)
3261 {
3262 transition(vtos, vtos);
3263 assert(byte_no == f1_byte, "use this argument");
3264
3265 prepare_invoke(byte_no, rmethod); // get f1 Method*
3266 // do the call
3267 __ profile_call(r0);
3268 __ profile_arguments_type(r0, rmethod, r4, false);
3269 __ jump_from_interpreted(rmethod, r0);
3270 }
3271
3272 void TemplateTable::fast_invokevfinal(int byte_no)
3273 {
3274 __ call_Unimplemented();
3275 }
3276
3277 void TemplateTable::invokeinterface(int byte_no) {
3278 transition(vtos, vtos);
3279 assert(byte_no == f1_byte, "use this argument");
3280
3281 prepare_invoke(byte_no, r0, rmethod, // get f1 Klass*, f2 itable index
3282 r2, r3); // recv, flags
3283
3284 // r0: interface klass (from f1)
3285 // rmethod: itable index (from f2)
3286 // r2: receiver
3287 // r3: flags
3288
3289 // Special case of invokeinterface called for virtual method of
3290 // java.lang.Object. See cpCacheOop.cpp for details.
3291 // This code isn't produced by javac, but could be produced by
3292 // another compliant java compiler.
3293 Label notMethod;
3294 __ tbz(r3, ConstantPoolCacheEntry::is_forced_virtual_shift, notMethod);
3295
3296 invokevirtual_helper(rmethod, r2, r3);
3297 __ bind(notMethod);
3298
3299 // Get receiver klass into r3 - also a null check
3300 __ restore_locals();
3301 __ null_check(r2, oopDesc::klass_offset_in_bytes());
3302 __ load_klass(r3, r2);
3303
3304 // profile this call
3305 __ profile_virtual_call(r3, r13, r19);
3306
3307 Label no_such_interface, no_such_method;
3308
3309 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3310 r3, r0, rmethod,
3311 // outputs: method, scan temp. reg
3312 rmethod, r13,
3313 no_such_interface);
3314
3315 // rmethod,: methodOop to call
3316 // r2: receiver
3317 // Check for abstract method error
3318 // Note: This should be done more efficiently via a throw_abstract_method_error
3319 // interpreter entry point and a conditional jump to it in case of a null
3320 // method.
3321 __ cbz(rmethod, no_such_method);
3322
3323 __ profile_arguments_type(r3, rmethod, r13, true);
3324
3325 // do the call
3326 // r2: receiver
3327 // rmethod,: methodOop
3328 __ jump_from_interpreted(rmethod, r3);
3329 __ should_not_reach_here();
3330
3331 // exception handling code follows...
3332 // note: must restore interpreter registers to canonical
3333 // state for exception handling to work correctly!
3334
3335 __ bind(no_such_method);
3336 // throw exception
3337 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3338 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3339 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3340 // the call_VM checks for exception, so we should never return here.
3341 __ should_not_reach_here();
3342
3343 __ bind(no_such_interface);
3344 // throw exception
3345 __ restore_bcp(); // bcp must be correct for exception handler (was destroyed)
3346 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3347 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3348 InterpreterRuntime::throw_IncompatibleClassChangeError));
3349 // the call_VM checks for exception, so we should never return here.
3350 __ should_not_reach_here();
3351 return;
3352 }
3353
3354 void TemplateTable::invokehandle(int byte_no) {
3355 transition(vtos, vtos);
3356 assert(byte_no == f1_byte, "use this argument");
3357
3358 prepare_invoke(byte_no, rmethod, r0, r2);
3359 __ verify_method_ptr(r2);
3360 __ verify_oop(r2);
3361 __ null_check(r2);
3362
3363 // FIXME: profile the LambdaForm also
3364
3365 // r13 is safe to use here as a scratch reg because it is about to
3366 // be clobbered by jump_from_interpreted().
3367 __ profile_final_call(r13);
3368 __ profile_arguments_type(r13, rmethod, r4, true);
3369
3370 __ jump_from_interpreted(rmethod, r0);
3371 }
3372
3373 void TemplateTable::invokedynamic(int byte_no) {
3374 transition(vtos, vtos);
3375 assert(byte_no == f1_byte, "use this argument");
3376
3377 prepare_invoke(byte_no, rmethod, r0);
3378
3379 // r0: CallSite object (from cpool->resolved_references[])
3380 // rmethod: MH.linkToCallSite method (from f2)
3381
3382 // Note: r0_callsite is already pushed by prepare_invoke
3383
3384 // %%% should make a type profile for any invokedynamic that takes a ref argument
3385 // profile this call
3386 __ profile_call(rbcp);
3387 __ profile_arguments_type(r3, rmethod, r13, false);
3388
3389 __ verify_oop(r0);
3390
3391 __ jump_from_interpreted(rmethod, r0);
3392 }
3393
3394
3395 //-----------------------------------------------------------------------------
3396 // Allocation
3397
3398 void TemplateTable::_new() {
3399 transition(vtos, atos);
3400
3401 __ get_unsigned_2_byte_index_at_bcp(r3, 1);
3402 Label slow_case;
3403 Label done;
3404 Label initialize_header;
3405 Label initialize_object; // including clearing the fields
3406 Label allocate_shared;
3407
3408 __ get_cpool_and_tags(r4, r0);
3409 // Make sure the class we're about to instantiate has been resolved.
3410 // This is done before loading InstanceKlass to be consistent with the order
3411 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3412 const int tags_offset = Array<u1>::base_offset_in_bytes();
3413 __ lea(rscratch1, Address(r0, r3, Address::lsl(0)));
3414 __ lea(rscratch1, Address(rscratch1, tags_offset));
3415 __ ldarb(rscratch1, rscratch1);
3416 __ cmp(rscratch1, JVM_CONSTANT_Class);
3417 __ br(Assembler::NE, slow_case);
3418
3419 // get InstanceKlass
3420 __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
3421
3422 // make sure klass is initialized & doesn't have finalizer
3423 // make sure klass is fully initialized
3424 __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
3425 __ cmp(rscratch1, InstanceKlass::fully_initialized);
3426 __ br(Assembler::NE, slow_case);
3427
3428 // get instance_size in InstanceKlass (scaled to a count of bytes)
3429 __ ldrw(r3,
3430 Address(r4,
3431 Klass::layout_helper_offset()));
3432 // test to see if it has a finalizer or is malformed in some way
3433 __ tbnz(r3, exact_log2(Klass::_lh_instance_slow_path_bit), slow_case);
3434
3435 // Allocate the instance
3436 // 1) Try to allocate in the TLAB
3437 // 2) if fail and the object is large allocate in the shared Eden
3438 // 3) if the above fails (or is not applicable), go to a slow case
3439 // (creates a new TLAB, etc.)
3440
3441 const bool allow_shared_alloc =
3442 Universe::heap()->supports_inline_contig_alloc();
3443
3444 if (UseTLAB) {
3445 __ tlab_allocate(r0, r3, 0, noreg, r1,
3446 allow_shared_alloc ? allocate_shared : slow_case);
3447
3448 if (ZeroTLAB) {
3449 // the fields have been already cleared
3450 __ b(initialize_header);
3451 } else {
3452 // initialize both the header and fields
3453 __ b(initialize_object);
3454 }
3455 }
3456
3457 // Allocation in the shared Eden, if allowed.
3458 //
3459 // r3: instance size in bytes
3460 if (allow_shared_alloc) {
3461 __ bind(allocate_shared);
3462
3463 __ eden_allocate(r0, r3, 0, r10, slow_case);
3464 __ incr_allocated_bytes(rthread, r3, 0, rscratch1);
3465 }
3466
3467 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3468 // The object is initialized before the header. If the object size is
3469 // zero, go directly to the header initialization.
3470 __ bind(initialize_object);
3471 __ sub(r3, r3, sizeof(oopDesc));
3472 __ cbz(r3, initialize_header);
3473
3474 // Initialize object fields
3475 {
3476 __ add(r2, r0, sizeof(oopDesc));
3477 Label loop;
3478 __ bind(loop);
3479 __ str(zr, Address(__ post(r2, BytesPerLong)));
3480 __ sub(r3, r3, BytesPerLong);
3481 __ cbnz(r3, loop);
3482 }
3483
3484 // initialize object header only.
3485 __ bind(initialize_header);
3486 if (UseBiasedLocking) {
3487 __ ldr(rscratch1, Address(r4, Klass::prototype_header_offset()));
3488 } else {
3489 __ mov(rscratch1, (intptr_t)markOopDesc::prototype());
3490 }
3491 __ str(rscratch1, Address(r0, oopDesc::mark_offset_in_bytes()));
3492 __ store_klass_gap(r0, zr); // zero klass gap for compressed oops
3493 __ store_klass(r0, r4); // store klass last
3494
3495 {
3496 SkipIfEqual skip(_masm, &DTraceAllocProbes, false);
3497 // Trigger dtrace event for fastpath
3498 __ push(atos); // save the return value
3499 __ call_VM_leaf(
3500 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), r0);
3501 __ pop(atos); // restore the return value
3502
3503 }
3504 __ b(done);
3505 }
3506
3507 // slow case
3508 __ bind(slow_case);
3509 __ get_constant_pool(c_rarg1);
3510 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3511 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), c_rarg1, c_rarg2);
3512 __ verify_oop(r0);
3513
3514 // continue
3515 __ bind(done);
3516 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3517 __ membar(Assembler::StoreStore);
3518 }
3519
3520 void TemplateTable::newarray() {
3521 transition(itos, atos);
3522 __ load_unsigned_byte(c_rarg1, at_bcp(1));
3523 __ mov(c_rarg2, r0);
3524 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3525 c_rarg1, c_rarg2);
3526 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3527 __ membar(Assembler::StoreStore);
3528 }
3529
3530 void TemplateTable::anewarray() {
3531 transition(itos, atos);
3532 __ get_unsigned_2_byte_index_at_bcp(c_rarg2, 1);
3533 __ get_constant_pool(c_rarg1);
3534 __ mov(c_rarg3, r0);
3535 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3536 c_rarg1, c_rarg2, c_rarg3);
3537 // Must prevent reordering of stores for object initialization with stores that publish the new object.
3538 __ membar(Assembler::StoreStore);
3539 }
3540
3541 void TemplateTable::arraylength() {
3542 transition(atos, itos);
3543 __ null_check(r0, arrayOopDesc::length_offset_in_bytes());
3544 __ ldrw(r0, Address(r0, arrayOopDesc::length_offset_in_bytes()));
3545 }
3546
3547 void TemplateTable::checkcast()
3548 {
3549 transition(atos, atos);
3550 Label done, is_null, ok_is_subtype, quicked, resolved;
3551 __ cbz(r0, is_null);
3552
3553 // Get cpool & tags index
3554 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3555 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3556 // See if bytecode has already been quicked
3557 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3558 __ lea(r1, Address(rscratch1, r19));
3559 __ ldarb(r1, r1);
3560 __ cmp(r1, JVM_CONSTANT_Class);
3561 __ br(Assembler::EQ, quicked);
3562
3563 __ push(atos); // save receiver for result, and for GC
3564 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3565 // vm_result_2 has metadata result
3566 __ get_vm_result_2(r0, rthread);
3567 __ pop(r3); // restore receiver
3568 __ b(resolved);
3569
3570 // Get superklass in r0 and subklass in r3
3571 __ bind(quicked);
3572 __ mov(r3, r0); // Save object in r3; r0 needed for subtype check
3573 __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1); // r0 = klass
3574
3575 __ bind(resolved);
3576 __ load_klass(r19, r3);
3577
3578 // Generate subtype check. Blows r2, r5. Object in r3.
3579 // Superklass in r0. Subklass in r19.
3580 __ gen_subtype_check(r19, ok_is_subtype);
3581
3582 // Come here on failure
3583 __ push(r3);
3584 // object is at TOS
3585 __ b(Interpreter::_throw_ClassCastException_entry);
3586
3587 // Come here on success
3588 __ bind(ok_is_subtype);
3589 __ mov(r0, r3); // Restore object in r3
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 }
3601
3602 void TemplateTable::instanceof() {
3603 transition(atos, itos);
3604 Label done, is_null, ok_is_subtype, quicked, resolved;
3605 __ cbz(r0, is_null);
3606
3607 // Get cpool & tags index
3608 __ get_cpool_and_tags(r2, r3); // r2=cpool, r3=tags array
3609 __ get_unsigned_2_byte_index_at_bcp(r19, 1); // r19=index
3610 // See if bytecode has already been quicked
3611 __ add(rscratch1, r3, Array<u1>::base_offset_in_bytes());
3612 __ lea(r1, Address(rscratch1, r19));
3613 __ ldarb(r1, r1);
3614 __ cmp(r1, JVM_CONSTANT_Class);
3615 __ br(Assembler::EQ, quicked);
3616
3617 __ push(atos); // save receiver for result, and for GC
3618 call_VM(r0, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3619 // vm_result_2 has metadata result
3620 __ get_vm_result_2(r0, rthread);
3621 __ pop(r3); // restore receiver
3622 __ verify_oop(r3);
3623 __ load_klass(r3, r3);
3624 __ b(resolved);
3625
3626 // Get superklass in r0 and subklass in r3
3627 __ bind(quicked);
3628 __ load_klass(r3, r0);
3629 __ load_resolved_klass_at_offset(r2, r19, r0, rscratch1);
3630
3631 __ bind(resolved);
3632
3633 // Generate subtype check. Blows r2, r5
3634 // Superklass in r0. Subklass in r3.
3635 __ gen_subtype_check(r3, ok_is_subtype);
3636
3637 // Come here on failure
3638 __ mov(r0, 0);
3639 __ b(done);
3640 // Come here on success
3641 __ bind(ok_is_subtype);
3642 __ mov(r0, 1);
3643
3644 // Collect counts on whether this test sees NULLs a lot or not.
3645 if (ProfileInterpreter) {
3646 __ b(done);
3647 __ bind(is_null);
3648 __ profile_null_seen(r2);
3649 } else {
3650 __ bind(is_null); // same as 'done'
3651 }
3652 __ bind(done);
3653 // r0 = 0: obj == NULL or obj is not an instanceof the specified klass
3654 // r0 = 1: obj != NULL and obj is an instanceof the specified klass
3655 }
3656
3657 //-----------------------------------------------------------------------------
3658 // Breakpoints
3659 void TemplateTable::_breakpoint() {
3660 // Note: We get here even if we are single stepping..
3661 // jbug inists on setting breakpoints at every bytecode
3662 // even if we are in single step mode.
3663
3664 transition(vtos, vtos);
3665
3666 // get the unpatched byte code
3667 __ get_method(c_rarg1);
3668 __ call_VM(noreg,
3669 CAST_FROM_FN_PTR(address,
3670 InterpreterRuntime::get_original_bytecode_at),
3671 c_rarg1, rbcp);
3672 __ mov(r19, r0);
3673
3674 // post the breakpoint event
3675 __ call_VM(noreg,
3676 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
3677 rmethod, rbcp);
3678
3679 // complete the execution of original bytecode
3680 __ mov(rscratch1, r19);
3681 __ dispatch_only_normal(vtos);
3682 }
3683
3684 //-----------------------------------------------------------------------------
3685 // Exceptions
3686
3687 void TemplateTable::athrow() {
3688 transition(atos, vtos);
3689 __ null_check(r0);
3690 __ b(Interpreter::throw_exception_entry());
3691 }
3692
3693 //-----------------------------------------------------------------------------
3694 // Synchronization
3695 //
3696 // Note: monitorenter & exit are symmetric routines; which is reflected
3697 // in the assembly code structure as well
3698 //
3699 // Stack layout:
3700 //
3701 // [expressions ] <--- esp = expression stack top
3702 // ..
3703 // [expressions ]
3704 // [monitor entry] <--- monitor block top = expression stack bot
3705 // ..
3706 // [monitor entry]
3707 // [frame data ] <--- monitor block bot
3708 // ...
3709 // [saved rbp ] <--- rbp
3710 void TemplateTable::monitorenter()
3711 {
3712 transition(atos, vtos);
3713
3714 // check for NULL object
3715 __ null_check(r0);
3716
3717 const Address monitor_block_top(
3718 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3719 const Address monitor_block_bot(
3720 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3721 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3722
3723 Label allocated;
3724
3725 // initialize entry pointer
3726 __ mov(c_rarg1, zr); // points to free slot or NULL
3727
3728 // find a free slot in the monitor block (result in c_rarg1)
3729 {
3730 Label entry, loop, exit;
3731 __ ldr(c_rarg3, monitor_block_top); // points to current entry,
3732 // starting with top-most entry
3733 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3734
3735 __ b(entry);
3736
3737 __ bind(loop);
3738 // check if current entry is used
3739 // if not used then remember entry in c_rarg1
3740 __ ldr(rscratch1, Address(c_rarg3, BasicObjectLock::obj_offset_in_bytes()));
3741 __ cmp(zr, rscratch1);
3742 __ csel(c_rarg1, c_rarg3, c_rarg1, Assembler::EQ);
3743 // check if current entry is for same object
3744 __ cmp(r0, rscratch1);
3745 // if same object then stop searching
3746 __ br(Assembler::EQ, exit);
3747 // otherwise advance to next entry
3748 __ add(c_rarg3, c_rarg3, entry_size);
3749 __ bind(entry);
3750 // check if bottom reached
3751 __ cmp(c_rarg3, c_rarg2);
3752 // if not at bottom then check this entry
3753 __ br(Assembler::NE, loop);
3754 __ bind(exit);
3755 }
3756
3757 __ cbnz(c_rarg1, allocated); // check if a slot has been found and
3758 // if found, continue with that on
3759
3760 // allocate one if there's no free slot
3761 {
3762 Label entry, loop;
3763 // 1. compute new pointers // rsp: old expression stack top
3764 __ ldr(c_rarg1, monitor_block_bot); // c_rarg1: old expression stack bottom
3765 __ sub(esp, esp, entry_size); // move expression stack top
3766 __ sub(c_rarg1, c_rarg1, entry_size); // move expression stack bottom
3767 __ mov(c_rarg3, esp); // set start value for copy loop
3768 __ str(c_rarg1, monitor_block_bot); // set new monitor block bottom
3769
3770 __ sub(sp, sp, entry_size); // make room for the monitor
3771
3772 __ b(entry);
3773 // 2. move expression stack contents
3774 __ bind(loop);
3775 __ ldr(c_rarg2, Address(c_rarg3, entry_size)); // load expression stack
3776 // word from old location
3777 __ str(c_rarg2, Address(c_rarg3, 0)); // and store it at new location
3778 __ add(c_rarg3, c_rarg3, wordSize); // advance to next word
3779 __ bind(entry);
3780 __ cmp(c_rarg3, c_rarg1); // check if bottom reached
3781 __ br(Assembler::NE, loop); // if not at bottom then
3782 // copy next word
3783 }
3784
3785 // call run-time routine
3786 // c_rarg1: points to monitor entry
3787 __ bind(allocated);
3788
3789 // Increment bcp to point to the next bytecode, so exception
3790 // handling for async. exceptions work correctly.
3791 // The object has already been poped from the stack, so the
3792 // expression stack looks correct.
3793 __ increment(rbcp);
3794
3795 // store object
3796 __ str(r0, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3797 __ lock_object(c_rarg1);
3798
3799 // check to make sure this monitor doesn't cause stack overflow after locking
3800 __ save_bcp(); // in case of exception
3801 __ generate_stack_overflow_check(0);
3802
3803 // The bcp has already been incremented. Just need to dispatch to
3804 // next instruction.
3805 __ dispatch_next(vtos);
3806 }
3807
3808
3809 void TemplateTable::monitorexit()
3810 {
3811 transition(atos, vtos);
3812
3813 // check for NULL object
3814 __ null_check(r0);
3815
3816 const Address monitor_block_top(
3817 rfp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3818 const Address monitor_block_bot(
3819 rfp, frame::interpreter_frame_initial_sp_offset * wordSize);
3820 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
3821
3822 Label found;
3823
3824 // find matching slot
3825 {
3826 Label entry, loop;
3827 __ ldr(c_rarg1, monitor_block_top); // points to current entry,
3828 // starting with top-most entry
3829 __ lea(c_rarg2, monitor_block_bot); // points to word before bottom
3830 // of monitor block
3831 __ b(entry);
3832
3833 __ bind(loop);
3834 // check if current entry is for same object
3835 __ ldr(rscratch1, Address(c_rarg1, BasicObjectLock::obj_offset_in_bytes()));
3836 __ cmp(r0, rscratch1);
3837 // if same object then stop searching
3838 __ br(Assembler::EQ, found);
3839 // otherwise advance to next entry
3840 __ add(c_rarg1, c_rarg1, entry_size);
3841 __ bind(entry);
3842 // check if bottom reached
3843 __ cmp(c_rarg1, c_rarg2);
3844 // if not at bottom then check this entry
3845 __ br(Assembler::NE, loop);
3846 }
3847
3848 // error handling. Unlocking was not block-structured
3849 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3850 InterpreterRuntime::throw_illegal_monitor_state_exception));
3851 __ should_not_reach_here();
3852
3853 // call run-time routine
3854 __ bind(found);
3855 __ push_ptr(r0); // make sure object is on stack (contract with oopMaps)
3856 __ unlock_object(c_rarg1);
3857 __ pop_ptr(r0); // discard object
3858 }
3859
3860
3861 // Wide instructions
3862 void TemplateTable::wide()
3863 {
3864 __ load_unsigned_byte(r19, at_bcp(1));
3865 __ mov(rscratch1, (address)Interpreter::_wentry_point);
3866 __ ldr(rscratch1, Address(rscratch1, r19, Address::uxtw(3)));
3867 __ br(rscratch1);
3868 }
3869
3870
3871 // Multi arrays
3872 void TemplateTable::multianewarray() {
3873 transition(vtos, atos);
3874 __ load_unsigned_byte(r0, at_bcp(3)); // get number of dimensions
3875 // last dim is on top of stack; we want address of first one:
3876 // first_addr = last_addr + (ndims - 1) * wordSize
3877 __ lea(c_rarg1, Address(esp, r0, Address::uxtw(3)));
3878 __ sub(c_rarg1, c_rarg1, wordSize);
3879 call_VM(r0,
3880 CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray),
3881 c_rarg1);
3882 __ load_unsigned_byte(r1, at_bcp(3));
3883 __ lea(esp, Address(esp, r1, Address::uxtw(3)));
3884 }