1 /*
2 * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.inline.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #define __ _masm->
42
43 // Global Register Names
44 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
45 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
46
47 // Platform-dependent initialization
48 void TemplateTable::pd_initialize() {
49 // No x86 specific initialization
50 }
51
52 // Address Computation: local variables
53 static inline Address iaddress(int n) {
54 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
55 }
56
57 static inline Address laddress(int n) {
58 return iaddress(n + 1);
59 }
60
61 #ifndef _LP64
62 static inline Address haddress(int n) {
63 return iaddress(n + 0);
64 }
65 #endif
66
67 static inline Address faddress(int n) {
68 return iaddress(n);
69 }
70
71 static inline Address daddress(int n) {
72 return laddress(n);
73 }
74
75 static inline Address aaddress(int n) {
76 return iaddress(n);
77 }
78
79 static inline Address iaddress(Register r) {
80 return Address(rlocals, r, Address::times_ptr);
81 }
82
83 static inline Address laddress(Register r) {
84 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
85 }
86
87 #ifndef _LP64
88 static inline Address haddress(Register r) {
89 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
90 }
91 #endif
92
93 static inline Address faddress(Register r) {
94 return iaddress(r);
95 }
96
97 static inline Address daddress(Register r) {
98 return laddress(r);
99 }
100
101 static inline Address aaddress(Register r) {
102 return iaddress(r);
103 }
104
105
106 // expression stack
107 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
108 // data beyond the rsp which is potentially unsafe in an MT environment;
109 // an interrupt may overwrite that data.)
110 static inline Address at_rsp () {
111 return Address(rsp, 0);
112 }
113
114 // At top of Java expression stack which may be different than esp(). It
115 // isn't for category 1 objects.
116 static inline Address at_tos () {
117 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
118 }
119
120 static inline Address at_tos_p1() {
121 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
122 }
123
124 static inline Address at_tos_p2() {
125 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
126 }
127
128 // Condition conversion
129 static Assembler::Condition j_not(TemplateTable::Condition cc) {
130 switch (cc) {
131 case TemplateTable::equal : return Assembler::notEqual;
132 case TemplateTable::not_equal : return Assembler::equal;
133 case TemplateTable::less : return Assembler::greaterEqual;
134 case TemplateTable::less_equal : return Assembler::greater;
135 case TemplateTable::greater : return Assembler::lessEqual;
136 case TemplateTable::greater_equal: return Assembler::less;
137 }
138 ShouldNotReachHere();
139 return Assembler::zero;
140 }
141
142
143
144 // Miscelaneous helper routines
145 // Store an oop (or NULL) at the address described by obj.
146 // If val == noreg this means store a NULL
147
148
149 static void do_oop_store(InterpreterMacroAssembler* _masm,
150 Address obj,
151 Register val,
152 BarrierSet::Name barrier,
153 bool precise) {
154 assert(val == noreg || val == rax, "parameter is just for looks");
155 switch (barrier) {
156 #if INCLUDE_ALL_GCS
157 case BarrierSet::G1SATBCTLogging:
158 {
159 // flatten object address if needed
160 // We do it regardless of precise because we need the registers
161 if (obj.index() == noreg && obj.disp() == 0) {
162 if (obj.base() != rdx) {
163 __ movptr(rdx, obj.base());
164 }
165 } else {
166 __ lea(rdx, obj);
167 }
168
169 Register rtmp = LP64_ONLY(r8) NOT_LP64(rsi);
170 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
171
172 NOT_LP64(__ get_thread(rcx));
173 NOT_LP64(__ save_bcp());
174
175 __ g1_write_barrier_pre(rdx /* obj */,
176 rbx /* pre_val */,
177 rthread /* thread */,
178 rtmp /* tmp */,
179 val != noreg /* tosca_live */,
180 false /* expand_call */);
181 if (val == noreg) {
182 __ store_heap_oop_null(Address(rdx, 0));
183 } else {
184 // G1 barrier needs uncompressed oop for region cross check.
185 Register new_val = val;
186 if (UseCompressedOops) {
187 new_val = rbx;
188 __ movptr(new_val, val);
189 }
190 __ store_heap_oop(Address(rdx, 0), val);
191 __ g1_write_barrier_post(rdx /* store_adr */,
192 new_val /* new_val */,
193 rthread /* thread */,
194 rtmp /* tmp */,
195 rbx /* tmp2 */);
196 }
197 NOT_LP64( __ restore_bcp());
198 }
199 break;
200 #endif // INCLUDE_ALL_GCS
201 case BarrierSet::CardTableModRef:
202 {
203 if (val == noreg) {
204 __ store_heap_oop_null(obj);
205 } else {
206 __ store_heap_oop(obj, val);
207 // flatten object address if needed
208 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
209 __ store_check(obj.base());
210 } else {
211 __ lea(rdx, obj);
212 __ store_check(rdx);
213 }
214 }
215 }
216 break;
217 case BarrierSet::ModRef:
218 if (val == noreg) {
219 __ store_heap_oop_null(obj);
220 } else {
221 __ store_heap_oop(obj, val);
222 }
223 break;
224 default :
225 ShouldNotReachHere();
226
227 }
228 }
229
230 Address TemplateTable::at_bcp(int offset) {
231 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
232 return Address(rbcp, offset);
233 }
234
235
236 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
237 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
238 int byte_no) {
239 if (!RewriteBytecodes) return;
240 Label L_patch_done;
241
242 switch (bc) {
243 case Bytecodes::_fast_qputfield:
244 case Bytecodes::_fast_aputfield:
245 case Bytecodes::_fast_bputfield:
246 case Bytecodes::_fast_zputfield:
247 case Bytecodes::_fast_cputfield:
248 case Bytecodes::_fast_dputfield:
249 case Bytecodes::_fast_fputfield:
250 case Bytecodes::_fast_iputfield:
251 case Bytecodes::_fast_lputfield:
252 case Bytecodes::_fast_sputfield:
253 {
254 // We skip bytecode quickening for putfield instructions when
255 // the put_code written to the constant pool cache is zero.
256 // This is required so that every execution of this instruction
257 // calls out to InterpreterRuntime::resolve_get_put to do
258 // additional, required work.
259 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
260 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
261 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
262 __ movl(bc_reg, bc);
263 __ cmpl(temp_reg, (int) 0);
264 __ jcc(Assembler::zero, L_patch_done); // don't patch
265 }
266 break;
267 default:
268 assert(byte_no == -1, "sanity");
269 // the pair bytecodes have already done the load.
270 if (load_bc_into_bc_reg) {
271 __ movl(bc_reg, bc);
272 }
273 }
274
275 if (JvmtiExport::can_post_breakpoint()) {
276 Label L_fast_patch;
277 // if a breakpoint is present we can't rewrite the stream directly
278 __ movzbl(temp_reg, at_bcp(0));
279 __ cmpl(temp_reg, Bytecodes::_breakpoint);
280 __ jcc(Assembler::notEqual, L_fast_patch);
281 __ get_method(temp_reg);
282 // Let breakpoint table handling rewrite to quicker bytecode
283 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
284 #ifndef ASSERT
285 __ jmpb(L_patch_done);
286 #else
287 __ jmp(L_patch_done);
288 #endif
289 __ bind(L_fast_patch);
290 }
291
292 #ifdef ASSERT
293 Label L_okay;
294 __ load_unsigned_byte(temp_reg, at_bcp(0));
295 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
296 __ jcc(Assembler::equal, L_okay);
297 __ cmpl(temp_reg, bc_reg);
298 __ jcc(Assembler::equal, L_okay);
299 __ stop("patching the wrong bytecode");
300 __ bind(L_okay);
301 #endif
302
303 // patch bytecode
304 __ movb(at_bcp(0), bc_reg);
305 __ bind(L_patch_done);
306 }
307 // Individual instructions
308
309
310 void TemplateTable::nop() {
311 transition(vtos, vtos);
312 // nothing to do
313 }
314
315 void TemplateTable::shouldnotreachhere() {
316 transition(vtos, vtos);
317 __ stop("shouldnotreachhere bytecode");
318 }
319
320 void TemplateTable::aconst_null() {
321 transition(vtos, atos);
322 __ xorl(rax, rax);
323 }
324
325 void TemplateTable::iconst(int value) {
326 transition(vtos, itos);
327 if (value == 0) {
328 __ xorl(rax, rax);
329 } else {
330 __ movl(rax, value);
331 }
332 }
333
334 void TemplateTable::lconst(int value) {
335 transition(vtos, ltos);
336 if (value == 0) {
337 __ xorl(rax, rax);
338 } else {
339 __ movl(rax, value);
340 }
341 #ifndef _LP64
342 assert(value >= 0, "check this code");
343 __ xorptr(rdx, rdx);
344 #endif
345 }
346
347
348
349 void TemplateTable::fconst(int value) {
350 transition(vtos, ftos);
351 if (UseSSE >= 1) {
352 static float one = 1.0f, two = 2.0f;
353 switch (value) {
354 case 0:
355 __ xorps(xmm0, xmm0);
356 break;
357 case 1:
358 __ movflt(xmm0, ExternalAddress((address) &one));
359 break;
360 case 2:
361 __ movflt(xmm0, ExternalAddress((address) &two));
362 break;
363 default:
364 ShouldNotReachHere();
365 break;
366 }
367 } else {
368 #ifdef _LP64
369 ShouldNotReachHere();
370 #else
371 if (value == 0) { __ fldz();
372 } else if (value == 1) { __ fld1();
373 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
374 } else { ShouldNotReachHere();
375 }
376 #endif // _LP64
377 }
378 }
379
380 void TemplateTable::dconst(int value) {
381 transition(vtos, dtos);
382 if (UseSSE >= 2) {
383 static double one = 1.0;
384 switch (value) {
385 case 0:
386 __ xorpd(xmm0, xmm0);
387 break;
388 case 1:
389 __ movdbl(xmm0, ExternalAddress((address) &one));
390 break;
391 default:
392 ShouldNotReachHere();
393 break;
394 }
395 } else {
396 #ifdef _LP64
397 ShouldNotReachHere();
398 #else
399 if (value == 0) { __ fldz();
400 } else if (value == 1) { __ fld1();
401 } else { ShouldNotReachHere();
402 }
403 #endif
404 }
405 }
406
407 void TemplateTable::bipush() {
408 transition(vtos, itos);
409 __ load_signed_byte(rax, at_bcp(1));
410 }
411
412 void TemplateTable::sipush() {
413 transition(vtos, itos);
414 __ load_unsigned_short(rax, at_bcp(1));
415 __ bswapl(rax);
416 __ sarl(rax, 16);
417 }
418
419 void TemplateTable::ldc(bool wide) {
420 transition(vtos, vtos);
421 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
422 Label call_ldc, notFloat, notClass, notInt, Done;
423
424 if (wide) {
425 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
426 } else {
427 __ load_unsigned_byte(rbx, at_bcp(1));
428 }
429
430 __ get_cpool_and_tags(rcx, rax);
431 const int base_offset = ConstantPool::header_size() * wordSize;
432 const int tags_offset = Array<u1>::base_offset_in_bytes();
433
434 // get type
435 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
436
437 // unresolved class - get the resolved class
438 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
439 __ jccb(Assembler::equal, call_ldc);
440
441 // unresolved class in error state - call into runtime to throw the error
442 // from the first resolution attempt
443 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
444 __ jccb(Assembler::equal, call_ldc);
445
446 // resolved class - need to call vm to get java mirror of the class
447 __ cmpl(rdx, JVM_CONSTANT_Class);
448 __ jcc(Assembler::notEqual, notClass);
449
450 __ bind(call_ldc);
451
452 __ movl(rarg, wide);
453 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
454
455 __ push(atos);
456 __ jmp(Done);
457
458 __ bind(notClass);
459 __ cmpl(rdx, JVM_CONSTANT_Float);
460 __ jccb(Assembler::notEqual, notFloat);
461
462 // ftos
463 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
464 __ push(ftos);
465 __ jmp(Done);
466
467 __ bind(notFloat);
468 __ cmpl(rdx, JVM_CONSTANT_Integer);
469 __ jccb(Assembler::notEqual, notInt);
470
471 // itos
472 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
473 __ push(itos);
474 __ jmp(Done);
475
476 // assume the tag is for condy; if not, the VM runtime will tell us
477 __ bind(notInt);
478 condy_helper(Done);
479
480 __ bind(Done);
481 }
482
483 // Fast path for caching oop constants.
484 void TemplateTable::fast_aldc(bool wide) {
485 transition(vtos, atos);
486
487 Register result = rax;
488 Register tmp = rdx;
489 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
490 int index_size = wide ? sizeof(u2) : sizeof(u1);
491
492 Label resolved;
493
494 // We are resolved if the resolved reference cache entry contains a
495 // non-null object (String, MethodType, etc.)
496 assert_different_registers(result, tmp);
497 __ get_cache_index_at_bcp(tmp, 1, index_size);
498 __ load_resolved_reference_at_index(result, tmp);
499 __ testptr(result, result);
500 __ jcc(Assembler::notZero, resolved);
501
502 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
503
504 // first time invocation - must resolve first
505 __ movl(rarg, (int)bytecode());
506 __ call_VM(result, entry, rarg);
507 __ bind(resolved);
508
509 { // Check for the null sentinel.
510 // If we just called the VM, that already did the mapping for us,
511 // but it's harmless to retry.
512 Label notNull;
513 ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
514 __ movptr(tmp, null_sentinel);
515 __ cmpptr(tmp, result);
516 __ jccb(Assembler::notEqual, notNull);
517 __ xorptr(result, result); // NULL object reference
518 __ bind(notNull);
519 }
520
521 if (VerifyOops) {
522 __ verify_oop(result);
523 }
524 }
525
526 void TemplateTable::ldc2_w() {
527 transition(vtos, vtos);
528 Label notDouble, notLong, Done;
529 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
530
531 __ get_cpool_and_tags(rcx, rax);
532 const int base_offset = ConstantPool::header_size() * wordSize;
533 const int tags_offset = Array<u1>::base_offset_in_bytes();
534
535 // get type
536 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
537 __ cmpl(rdx, JVM_CONSTANT_Double);
538 __ jccb(Assembler::notEqual, notDouble);
539
540 // dtos
541 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
542 __ push(dtos);
543
544 __ jmp(Done);
545 __ bind(notDouble);
546 __ cmpl(rdx, JVM_CONSTANT_Long);
547 __ jccb(Assembler::notEqual, notLong);
548
549 // ltos
550 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
551 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
552 __ push(ltos);
553 __ jmp(Done);
554
555 __ bind(notLong);
556 condy_helper(Done);
557
558 __ bind(Done);
559 }
560
561 void TemplateTable::condy_helper(Label& Done) {
562 const Register obj = rax;
563 const Register off = rbx;
564 const Register flags = rcx;
565 const Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
566 __ movl(rarg, (int)bytecode());
567 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
568 #ifndef _LP64
569 // borrow rdi from locals
570 __ get_thread(rdi);
571 __ get_vm_result_2(flags, rdi);
572 __ restore_locals();
573 #else
574 __ get_vm_result_2(flags, r15_thread);
575 #endif
576 // VMr = obj = base address to find primitive value to push
577 // VMr2 = flags = (tos, off) using format of CPCE::_flags
578 __ movl(off, flags);
579 __ andl(off, ConstantPoolCacheEntry::field_index_mask);
580 const Address field(obj, off, Address::times_1, 0*wordSize);
581
582 // What sort of thing are we loading?
583 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
584 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
585
586 switch (bytecode()) {
587 case Bytecodes::_ldc:
588 case Bytecodes::_ldc_w:
589 {
590 // tos in (itos, ftos, stos, btos, ctos, ztos)
591 Label notInt, notFloat, notShort, notByte, notChar, notBool;
592 __ cmpl(flags, itos);
593 __ jcc(Assembler::notEqual, notInt);
594 // itos
595 __ movl(rax, field);
596 __ push(itos);
597 __ jmp(Done);
598
599 __ bind(notInt);
600 __ cmpl(flags, ftos);
601 __ jcc(Assembler::notEqual, notFloat);
602 // ftos
603 __ load_float(field);
604 __ push(ftos);
605 __ jmp(Done);
606
607 __ bind(notFloat);
608 __ cmpl(flags, stos);
609 __ jcc(Assembler::notEqual, notShort);
610 // stos
611 __ load_signed_short(rax, field);
612 __ push(stos);
613 __ jmp(Done);
614
615 __ bind(notShort);
616 __ cmpl(flags, btos);
617 __ jcc(Assembler::notEqual, notByte);
618 // btos
619 __ load_signed_byte(rax, field);
620 __ push(btos);
621 __ jmp(Done);
622
623 __ bind(notByte);
624 __ cmpl(flags, ctos);
625 __ jcc(Assembler::notEqual, notChar);
626 // ctos
627 __ load_unsigned_short(rax, field);
628 __ push(ctos);
629 __ jmp(Done);
630
631 __ bind(notChar);
632 __ cmpl(flags, ztos);
633 __ jcc(Assembler::notEqual, notBool);
634 // ztos
635 __ load_signed_byte(rax, field);
636 __ push(ztos);
637 __ jmp(Done);
638
639 __ bind(notBool);
640 break;
641 }
642
643 case Bytecodes::_ldc2_w:
644 {
645 Label notLong, notDouble;
646 __ cmpl(flags, ltos);
647 __ jcc(Assembler::notEqual, notLong);
648 // ltos
649 __ movptr(rax, field);
650 NOT_LP64(__ movptr(rdx, field.plus_disp(4)));
651 __ push(ltos);
652 __ jmp(Done);
653
654 __ bind(notLong);
655 __ cmpl(flags, dtos);
656 __ jcc(Assembler::notEqual, notDouble);
657 // dtos
658 __ load_double(field);
659 __ push(dtos);
660 __ jmp(Done);
661
662 __ bind(notDouble);
663 break;
664 }
665
666 default:
667 ShouldNotReachHere();
668 }
669
670 __ stop("bad ldc/condy");
671 }
672
673 void TemplateTable::locals_index(Register reg, int offset) {
674 __ load_unsigned_byte(reg, at_bcp(offset));
675 __ negptr(reg);
676 }
677
678 void TemplateTable::iload() {
679 iload_internal();
680 }
681
682 void TemplateTable::nofast_iload() {
683 iload_internal(may_not_rewrite);
684 }
685
686 void TemplateTable::iload_internal(RewriteControl rc) {
687 transition(vtos, itos);
688 if (RewriteFrequentPairs && rc == may_rewrite) {
689 Label rewrite, done;
690 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
691 LP64_ONLY(assert(rbx != bc, "register damaged"));
692
693 // get next byte
694 __ load_unsigned_byte(rbx,
695 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
696 // if _iload, wait to rewrite to iload2. We only want to rewrite the
697 // last two iloads in a pair. Comparing against fast_iload means that
698 // the next bytecode is neither an iload or a caload, and therefore
699 // an iload pair.
700 __ cmpl(rbx, Bytecodes::_iload);
701 __ jcc(Assembler::equal, done);
702
703 __ cmpl(rbx, Bytecodes::_fast_iload);
704 __ movl(bc, Bytecodes::_fast_iload2);
705
706 __ jccb(Assembler::equal, rewrite);
707
708 // if _caload, rewrite to fast_icaload
709 __ cmpl(rbx, Bytecodes::_caload);
710 __ movl(bc, Bytecodes::_fast_icaload);
711 __ jccb(Assembler::equal, rewrite);
712
713 // rewrite so iload doesn't check again.
714 __ movl(bc, Bytecodes::_fast_iload);
715
716 // rewrite
717 // bc: fast bytecode
718 __ bind(rewrite);
719 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
720 __ bind(done);
721 }
722
723 // Get the local value into tos
724 locals_index(rbx);
725 __ movl(rax, iaddress(rbx));
726 }
727
728 void TemplateTable::fast_iload2() {
729 transition(vtos, itos);
730 locals_index(rbx);
731 __ movl(rax, iaddress(rbx));
732 __ push(itos);
733 locals_index(rbx, 3);
734 __ movl(rax, iaddress(rbx));
735 }
736
737 void TemplateTable::fast_iload() {
738 transition(vtos, itos);
739 locals_index(rbx);
740 __ movl(rax, iaddress(rbx));
741 }
742
743 void TemplateTable::lload() {
744 transition(vtos, ltos);
745 locals_index(rbx);
746 __ movptr(rax, laddress(rbx));
747 NOT_LP64(__ movl(rdx, haddress(rbx)));
748 }
749
750 void TemplateTable::fload() {
751 transition(vtos, ftos);
752 locals_index(rbx);
753 __ load_float(faddress(rbx));
754 }
755
756 void TemplateTable::dload() {
757 transition(vtos, dtos);
758 locals_index(rbx);
759 __ load_double(daddress(rbx));
760 }
761
762 void TemplateTable::aload() {
763 transition(vtos, atos);
764 locals_index(rbx);
765 __ movptr(rax, aaddress(rbx));
766 }
767
768 void TemplateTable::locals_index_wide(Register reg) {
769 __ load_unsigned_short(reg, at_bcp(2));
770 __ bswapl(reg);
771 __ shrl(reg, 16);
772 __ negptr(reg);
773 }
774
775 void TemplateTable::wide_iload() {
776 transition(vtos, itos);
777 locals_index_wide(rbx);
778 __ movl(rax, iaddress(rbx));
779 }
780
781 void TemplateTable::wide_lload() {
782 transition(vtos, ltos);
783 locals_index_wide(rbx);
784 __ movptr(rax, laddress(rbx));
785 NOT_LP64(__ movl(rdx, haddress(rbx)));
786 }
787
788 void TemplateTable::wide_fload() {
789 transition(vtos, ftos);
790 locals_index_wide(rbx);
791 __ load_float(faddress(rbx));
792 }
793
794 void TemplateTable::wide_dload() {
795 transition(vtos, dtos);
796 locals_index_wide(rbx);
797 __ load_double(daddress(rbx));
798 }
799
800 void TemplateTable::wide_aload() {
801 transition(vtos, atos);
802 locals_index_wide(rbx);
803 __ movptr(rax, aaddress(rbx));
804 }
805
806 void TemplateTable::index_check(Register array, Register index) {
807 // Pop ptr into array
808 __ pop_ptr(array);
809 index_check_without_pop(array, index);
810 }
811
812 void TemplateTable::index_check_without_pop(Register array, Register index) {
813 // destroys rbx
814 // check array
815 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
816 // sign extend index for use by indexed load
817 __ movl2ptr(index, index);
818 // check index
819 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
820 if (index != rbx) {
821 // ??? convention: move aberrant index into rbx for exception message
822 assert(rbx != array, "different registers");
823 __ movl(rbx, index);
824 }
825 __ jump_cc(Assembler::aboveEqual,
826 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
827 }
828
829
830 void TemplateTable::iaload() {
831 transition(itos, itos);
832 // rax: index
833 // rdx: array
834 index_check(rdx, rax); // kills rbx
835 __ movl(rax, Address(rdx, rax,
836 Address::times_4,
837 arrayOopDesc::base_offset_in_bytes(T_INT)));
838 }
839
840 void TemplateTable::laload() {
841 transition(itos, ltos);
842 // rax: index
843 // rdx: array
844 index_check(rdx, rax); // kills rbx
845 NOT_LP64(__ mov(rbx, rax));
846 // rbx,: index
847 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
848 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
849 }
850
851
852
853 void TemplateTable::faload() {
854 transition(itos, ftos);
855 // rax: index
856 // rdx: array
857 index_check(rdx, rax); // kills rbx
858 __ load_float(Address(rdx, rax,
859 Address::times_4,
860 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
861 }
862
863 void TemplateTable::daload() {
864 transition(itos, dtos);
865 // rax: index
866 // rdx: array
867 index_check(rdx, rax); // kills rbx
868 __ load_double(Address(rdx, rax,
869 Address::times_8,
870 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
871 }
872
873 void TemplateTable::aaload() {
874 transition(itos, atos);
875
876 Register array = rcx;
877 Register index = rax;
878
879 index_check(array, index); // kills rbx
880 if (EnableValhalla) {
881 Label is_flat_array, done;
882 __ test_flat_array_oop(array, rbx, is_flat_array);
883 __ load_heap_oop(rax, Address(array, index,
884 UseCompressedOops ? Address::times_4 : Address::times_ptr,
885 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
886 __ jmp(done);
887 __ bind(is_flat_array);
888 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_load), array, index);
889 __ bind(done);
890 } else {
891 __ load_heap_oop(rax, Address(array, index,
892 UseCompressedOops ? Address::times_4 : Address::times_ptr,
893 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
894 }
895 }
896
897 void TemplateTable::baload() {
898 transition(itos, itos);
899 // rax: index
900 // rdx: array
901 index_check(rdx, rax); // kills rbx
902 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
903 }
904
905 void TemplateTable::caload() {
906 transition(itos, itos);
907 // rax: index
908 // rdx: array
909 index_check(rdx, rax); // kills rbx
910 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
911 }
912
913 // iload followed by caload frequent pair
914 void TemplateTable::fast_icaload() {
915 transition(vtos, itos);
916 // load index out of locals
917 locals_index(rbx);
918 __ movl(rax, iaddress(rbx));
919
920 // rax: index
921 // rdx: array
922 index_check(rdx, rax); // kills rbx
923 __ load_unsigned_short(rax,
924 Address(rdx, rax,
925 Address::times_2,
926 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
927 }
928
929
930 void TemplateTable::saload() {
931 transition(itos, itos);
932 // rax: index
933 // rdx: array
934 index_check(rdx, rax); // kills rbx
935 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
936 }
937
938 void TemplateTable::iload(int n) {
939 transition(vtos, itos);
940 __ movl(rax, iaddress(n));
941 }
942
943 void TemplateTable::lload(int n) {
944 transition(vtos, ltos);
945 __ movptr(rax, laddress(n));
946 NOT_LP64(__ movptr(rdx, haddress(n)));
947 }
948
949 void TemplateTable::fload(int n) {
950 transition(vtos, ftos);
951 __ load_float(faddress(n));
952 }
953
954 void TemplateTable::dload(int n) {
955 transition(vtos, dtos);
956 __ load_double(daddress(n));
957 }
958
959 void TemplateTable::aload(int n) {
960 transition(vtos, atos);
961 __ movptr(rax, aaddress(n));
962 }
963
964 void TemplateTable::aload_0() {
965 aload_0_internal();
966 }
967
968 void TemplateTable::nofast_aload_0() {
969 aload_0_internal(may_not_rewrite);
970 }
971
972 void TemplateTable::aload_0_internal(RewriteControl rc) {
973 transition(vtos, atos);
974 // According to bytecode histograms, the pairs:
975 //
976 // _aload_0, _fast_igetfield
977 // _aload_0, _fast_agetfield
978 // _aload_0, _fast_fgetfield
979 //
980 // occur frequently. If RewriteFrequentPairs is set, the (slow)
981 // _aload_0 bytecode checks if the next bytecode is either
982 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
983 // rewrites the current bytecode into a pair bytecode; otherwise it
984 // rewrites the current bytecode into _fast_aload_0 that doesn't do
985 // the pair check anymore.
986 //
987 // Note: If the next bytecode is _getfield, the rewrite must be
988 // delayed, otherwise we may miss an opportunity for a pair.
989 //
990 // Also rewrite frequent pairs
991 // aload_0, aload_1
992 // aload_0, iload_1
993 // These bytecodes with a small amount of code are most profitable
994 // to rewrite
995 if (RewriteFrequentPairs && rc == may_rewrite) {
996 Label rewrite, done;
997
998 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
999 LP64_ONLY(assert(rbx != bc, "register damaged"));
1000
1001 // get next byte
1002 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
1003
1004 // if _getfield then wait with rewrite
1005 __ cmpl(rbx, Bytecodes::_getfield);
1006 __ jcc(Assembler::equal, done);
1007
1008 // if _igetfield then rewrite to _fast_iaccess_0
1009 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
1010 __ cmpl(rbx, Bytecodes::_fast_igetfield);
1011 __ movl(bc, Bytecodes::_fast_iaccess_0);
1012 __ jccb(Assembler::equal, rewrite);
1013
1014 // if _agetfield then rewrite to _fast_aaccess_0
1015 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
1016 assert(ValueTypesBufferMaxMemory == 0, "Such rewritting doesn't support flattened values yet");
1017 __ cmpl(rbx, Bytecodes::_fast_agetfield);
1018 __ movl(bc, Bytecodes::_fast_aaccess_0);
1019 __ jccb(Assembler::equal, rewrite);
1020
1021 // if _fgetfield then rewrite to _fast_faccess_0
1022 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
1023 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
1024 __ movl(bc, Bytecodes::_fast_faccess_0);
1025 __ jccb(Assembler::equal, rewrite);
1026
1027 // else rewrite to _fast_aload0
1028 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
1029 __ movl(bc, Bytecodes::_fast_aload_0);
1030
1031 // rewrite
1032 // bc: fast bytecode
1033 __ bind(rewrite);
1034 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
1035
1036 __ bind(done);
1037 }
1038
1039 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
1040 aload(0);
1041 }
1042
1043 void TemplateTable::istore() {
1044 transition(itos, vtos);
1045 locals_index(rbx);
1046 __ movl(iaddress(rbx), rax);
1047 }
1048
1049
1050 void TemplateTable::lstore() {
1051 transition(ltos, vtos);
1052 locals_index(rbx);
1053 __ movptr(laddress(rbx), rax);
1054 NOT_LP64(__ movptr(haddress(rbx), rdx));
1055 }
1056
1057 void TemplateTable::fstore() {
1058 transition(ftos, vtos);
1059 locals_index(rbx);
1060 __ store_float(faddress(rbx));
1061 }
1062
1063 void TemplateTable::dstore() {
1064 transition(dtos, vtos);
1065 locals_index(rbx);
1066 __ store_double(daddress(rbx));
1067 }
1068
1069 void TemplateTable::astore() {
1070 transition(vtos, vtos);
1071 __ pop_ptr(rax);
1072 locals_index(rbx);
1073 __ movptr(aaddress(rbx), rax);
1074 }
1075
1076 void TemplateTable::wide_istore() {
1077 transition(vtos, vtos);
1078 __ pop_i();
1079 locals_index_wide(rbx);
1080 __ movl(iaddress(rbx), rax);
1081 }
1082
1083 void TemplateTable::wide_lstore() {
1084 transition(vtos, vtos);
1085 NOT_LP64(__ pop_l(rax, rdx));
1086 LP64_ONLY(__ pop_l());
1087 locals_index_wide(rbx);
1088 __ movptr(laddress(rbx), rax);
1089 NOT_LP64(__ movl(haddress(rbx), rdx));
1090 }
1091
1092 void TemplateTable::wide_fstore() {
1093 #ifdef _LP64
1094 transition(vtos, vtos);
1095 __ pop_f(xmm0);
1096 locals_index_wide(rbx);
1097 __ movflt(faddress(rbx), xmm0);
1098 #else
1099 wide_istore();
1100 #endif
1101 }
1102
1103 void TemplateTable::wide_dstore() {
1104 #ifdef _LP64
1105 transition(vtos, vtos);
1106 __ pop_d(xmm0);
1107 locals_index_wide(rbx);
1108 __ movdbl(daddress(rbx), xmm0);
1109 #else
1110 wide_lstore();
1111 #endif
1112 }
1113
1114 void TemplateTable::wide_astore() {
1115 transition(vtos, vtos);
1116 __ pop_ptr(rax);
1117 locals_index_wide(rbx);
1118 __ movptr(aaddress(rbx), rax);
1119 }
1120
1121 void TemplateTable::iastore() {
1122 transition(itos, vtos);
1123 __ pop_i(rbx);
1124 // rax: value
1125 // rbx: index
1126 // rdx: array
1127 index_check(rdx, rbx); // prefer index in rbx
1128 __ movl(Address(rdx, rbx,
1129 Address::times_4,
1130 arrayOopDesc::base_offset_in_bytes(T_INT)),
1131 rax);
1132 }
1133
1134 void TemplateTable::lastore() {
1135 transition(ltos, vtos);
1136 __ pop_i(rbx);
1137 // rax,: low(value)
1138 // rcx: array
1139 // rdx: high(value)
1140 index_check(rcx, rbx); // prefer index in rbx,
1141 // rbx,: index
1142 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1143 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1144 }
1145
1146
1147 void TemplateTable::fastore() {
1148 transition(ftos, vtos);
1149 __ pop_i(rbx);
1150 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1151 // rbx: index
1152 // rdx: array
1153 index_check(rdx, rbx); // prefer index in rbx
1154 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1155 }
1156
1157 void TemplateTable::dastore() {
1158 transition(dtos, vtos);
1159 __ pop_i(rbx);
1160 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1161 // rbx: index
1162 // rdx: array
1163 index_check(rdx, rbx); // prefer index in rbx
1164 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1165 }
1166
1167 void TemplateTable::aastore() {
1168 Label is_null, is_flat_array, ok_is_subtype, done;
1169 transition(vtos, vtos);
1170 // stack: ..., array, index, value
1171 __ movptr(rax, at_tos()); // value
1172 __ movl(rcx, at_tos_p1()); // index
1173 __ movptr(rdx, at_tos_p2()); // array
1174
1175 Address element_address(rdx, rcx,
1176 UseCompressedOops? Address::times_4 : Address::times_ptr,
1177 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1178
1179 index_check_without_pop(rdx, rcx); // kills rbx
1180
1181 __ testptr(rax, rax);
1182 __ jcc(Assembler::zero, is_null);
1183
1184 // Move array class to rdi
1185 __ load_klass(rdi, rdx);
1186 if (EnableValhalla) {
1187 __ test_flat_array_klass(rdi, rbx, is_flat_array);
1188 }
1189
1190 // Move subklass into rbx
1191 __ load_klass(rbx, rax);
1192 // Move array element superklass into rax
1193 __ movptr(rax, Address(rdi,
1194 ObjArrayKlass::element_klass_offset()));
1195 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1196 __ lea(rdx, element_address);
1197
1198 // Generate subtype check. Blows rcx, rdi
1199 // Superklass in rax. Subklass in rbx.
1200 // is "rbx <: rax" ? (value subclass <: array element superclass)
1201 __ gen_subtype_check(rbx, ok_is_subtype);
1202
1203 // Come here on failure
1204 // object is at TOS
1205 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1206
1207 // Come here on success
1208 __ bind(ok_is_subtype);
1209
1210 // Get the value we will store
1211 __ movptr(rax, at_tos());
1212 if (ValueTypesBufferMaxMemory > 0) {
1213 Label is_on_heap;
1214 __ test_value_is_not_buffered(rax, rbx, is_on_heap);
1215 __ push(rdx); // save precomputed element address, and convert buffer oop to heap oop
1216 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_heap_copy), rax);
1217 __ pop(rdx);
1218 __ bind(is_on_heap);
1219 }
1220 // Now store using the appropriate barrier
1221 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1222 __ jmp(done);
1223
1224 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1225 __ bind(is_null);
1226 __ profile_null_seen(rbx);
1227 if (EnableValhalla) {
1228 Label is_null_into_value_array_npe, store_null;
1229
1230 __ load_klass(rdi, rdx);
1231 // No way to store null in flat array
1232 __ test_flat_array_klass(rdi, rbx, is_null_into_value_array_npe);
1233
1234 // Use case for storing values in objArray where element_klass is specifically
1235 // a value type because they could not be flattened "for reasons",
1236 // these need to have the same semantics as flat arrays, i.e. NPE
1237 __ movptr(rdi, Address(rdi, ObjArrayKlass::element_klass_offset()));
1238 __ test_klass_is_value(rdi, rdi, is_null_into_value_array_npe);
1239 __ jmp(store_null);
1240
1241 __ bind(is_null_into_value_array_npe);
1242 __ jump(ExternalAddress(Interpreter::_throw_NullPointerException_entry));
1243
1244 __ bind(store_null);
1245 }
1246 // Store a NULL
1247 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1248 __ jmp(done);
1249
1250 if (EnableValhalla) {
1251 Label is_type_ok;
1252 __ bind(is_flat_array); // Store non-null value to flat
1253
1254 // Simplistic type check...
1255
1256 // Profile the not-null value's klass.
1257 __ load_klass(rbx, rax);
1258 __ profile_typecheck(rcx, rbx, rax); // blows rcx, and rax
1259 // Move element klass into rax
1260 __ movptr(rax, Address(rdi, ArrayKlass::element_klass_offset()));
1261 // flat value array needs exact type match
1262 // is "rax == rbx" (value subclass == array element superclass)
1263 __ cmpptr(rax, rbx);
1264 __ jccb(Assembler::equal, is_type_ok);
1265
1266 __ profile_typecheck_failed(rcx);
1267 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1268
1269 __ bind(is_type_ok);
1270 __ movptr(rax, at_tos()); // value
1271 __ movl(rcx, at_tos_p1()); // index
1272 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_store), rax, rdx, rcx);
1273 }
1274 // Pop stack arguments
1275 __ bind(done);
1276 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1277 }
1278
1279 void TemplateTable::bastore() {
1280 transition(itos, vtos);
1281 __ pop_i(rbx);
1282 // rax: value
1283 // rbx: index
1284 // rdx: array
1285 index_check(rdx, rbx); // prefer index in rbx
1286 // Need to check whether array is boolean or byte
1287 // since both types share the bastore bytecode.
1288 __ load_klass(rcx, rdx);
1289 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1290 int diffbit = Klass::layout_helper_boolean_diffbit();
1291 __ testl(rcx, diffbit);
1292 Label L_skip;
1293 __ jccb(Assembler::zero, L_skip);
1294 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1295 __ bind(L_skip);
1296 __ movb(Address(rdx, rbx,
1297 Address::times_1,
1298 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1299 rax);
1300 }
1301
1302 void TemplateTable::castore() {
1303 transition(itos, vtos);
1304 __ pop_i(rbx);
1305 // rax: value
1306 // rbx: index
1307 // rdx: array
1308 index_check(rdx, rbx); // prefer index in rbx
1309 __ movw(Address(rdx, rbx,
1310 Address::times_2,
1311 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1312 rax);
1313 }
1314
1315
1316 void TemplateTable::sastore() {
1317 castore();
1318 }
1319
1320 void TemplateTable::istore(int n) {
1321 transition(itos, vtos);
1322 __ movl(iaddress(n), rax);
1323 }
1324
1325 void TemplateTable::lstore(int n) {
1326 transition(ltos, vtos);
1327 __ movptr(laddress(n), rax);
1328 NOT_LP64(__ movptr(haddress(n), rdx));
1329 }
1330
1331 void TemplateTable::fstore(int n) {
1332 transition(ftos, vtos);
1333 __ store_float(faddress(n));
1334 }
1335
1336 void TemplateTable::dstore(int n) {
1337 transition(dtos, vtos);
1338 __ store_double(daddress(n));
1339 }
1340
1341
1342 void TemplateTable::astore(int n) {
1343 transition(vtos, vtos);
1344 __ pop_ptr(rax);
1345 __ movptr(aaddress(n), rax);
1346 }
1347
1348 void TemplateTable::pop() {
1349 transition(vtos, vtos);
1350 __ addptr(rsp, Interpreter::stackElementSize);
1351 }
1352
1353 void TemplateTable::pop2() {
1354 transition(vtos, vtos);
1355 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1356 }
1357
1358
1359 void TemplateTable::dup() {
1360 transition(vtos, vtos);
1361 __ load_ptr(0, rax);
1362 __ push_ptr(rax);
1363 // stack: ..., a, a
1364 }
1365
1366 void TemplateTable::dup_x1() {
1367 transition(vtos, vtos);
1368 // stack: ..., a, b
1369 __ load_ptr( 0, rax); // load b
1370 __ load_ptr( 1, rcx); // load a
1371 __ store_ptr(1, rax); // store b
1372 __ store_ptr(0, rcx); // store a
1373 __ push_ptr(rax); // push b
1374 // stack: ..., b, a, b
1375 }
1376
1377 void TemplateTable::dup_x2() {
1378 transition(vtos, vtos);
1379 // stack: ..., a, b, c
1380 __ load_ptr( 0, rax); // load c
1381 __ load_ptr( 2, rcx); // load a
1382 __ store_ptr(2, rax); // store c in a
1383 __ push_ptr(rax); // push c
1384 // stack: ..., c, b, c, c
1385 __ load_ptr( 2, rax); // load b
1386 __ store_ptr(2, rcx); // store a in b
1387 // stack: ..., c, a, c, c
1388 __ store_ptr(1, rax); // store b in c
1389 // stack: ..., c, a, b, c
1390 }
1391
1392 void TemplateTable::dup2() {
1393 transition(vtos, vtos);
1394 // stack: ..., a, b
1395 __ load_ptr(1, rax); // load a
1396 __ push_ptr(rax); // push a
1397 __ load_ptr(1, rax); // load b
1398 __ push_ptr(rax); // push b
1399 // stack: ..., a, b, a, b
1400 }
1401
1402
1403 void TemplateTable::dup2_x1() {
1404 transition(vtos, vtos);
1405 // stack: ..., a, b, c
1406 __ load_ptr( 0, rcx); // load c
1407 __ load_ptr( 1, rax); // load b
1408 __ push_ptr(rax); // push b
1409 __ push_ptr(rcx); // push c
1410 // stack: ..., a, b, c, b, c
1411 __ store_ptr(3, rcx); // store c in b
1412 // stack: ..., a, c, c, b, c
1413 __ load_ptr( 4, rcx); // load a
1414 __ store_ptr(2, rcx); // store a in 2nd c
1415 // stack: ..., a, c, a, b, c
1416 __ store_ptr(4, rax); // store b in a
1417 // stack: ..., b, c, a, b, c
1418 }
1419
1420 void TemplateTable::dup2_x2() {
1421 transition(vtos, vtos);
1422 // stack: ..., a, b, c, d
1423 __ load_ptr( 0, rcx); // load d
1424 __ load_ptr( 1, rax); // load c
1425 __ push_ptr(rax); // push c
1426 __ push_ptr(rcx); // push d
1427 // stack: ..., a, b, c, d, c, d
1428 __ load_ptr( 4, rax); // load b
1429 __ store_ptr(2, rax); // store b in d
1430 __ store_ptr(4, rcx); // store d in b
1431 // stack: ..., a, d, c, b, c, d
1432 __ load_ptr( 5, rcx); // load a
1433 __ load_ptr( 3, rax); // load c
1434 __ store_ptr(3, rcx); // store a in c
1435 __ store_ptr(5, rax); // store c in a
1436 // stack: ..., c, d, a, b, c, d
1437 }
1438
1439 void TemplateTable::swap() {
1440 transition(vtos, vtos);
1441 // stack: ..., a, b
1442 __ load_ptr( 1, rcx); // load a
1443 __ load_ptr( 0, rax); // load b
1444 __ store_ptr(0, rcx); // store a in b
1445 __ store_ptr(1, rax); // store b in a
1446 // stack: ..., b, a
1447 }
1448
1449 void TemplateTable::iop2(Operation op) {
1450 transition(itos, itos);
1451 switch (op) {
1452 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1453 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1454 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1455 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1456 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1457 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1458 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1459 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1460 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1461 default : ShouldNotReachHere();
1462 }
1463 }
1464
1465 void TemplateTable::lop2(Operation op) {
1466 transition(ltos, ltos);
1467 #ifdef _LP64
1468 switch (op) {
1469 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1470 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1471 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1472 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1473 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1474 default : ShouldNotReachHere();
1475 }
1476 #else
1477 __ pop_l(rbx, rcx);
1478 switch (op) {
1479 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1480 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1481 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1482 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1483 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1484 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1485 default : ShouldNotReachHere();
1486 }
1487 #endif
1488 }
1489
1490 void TemplateTable::idiv() {
1491 transition(itos, itos);
1492 __ movl(rcx, rax);
1493 __ pop_i(rax);
1494 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1495 // they are not equal, one could do a normal division (no correction
1496 // needed), which may speed up this implementation for the common case.
1497 // (see also JVM spec., p.243 & p.271)
1498 __ corrected_idivl(rcx);
1499 }
1500
1501 void TemplateTable::irem() {
1502 transition(itos, itos);
1503 __ movl(rcx, rax);
1504 __ pop_i(rax);
1505 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1506 // they are not equal, one could do a normal division (no correction
1507 // needed), which may speed up this implementation for the common case.
1508 // (see also JVM spec., p.243 & p.271)
1509 __ corrected_idivl(rcx);
1510 __ movl(rax, rdx);
1511 }
1512
1513 void TemplateTable::lmul() {
1514 transition(ltos, ltos);
1515 #ifdef _LP64
1516 __ pop_l(rdx);
1517 __ imulq(rax, rdx);
1518 #else
1519 __ pop_l(rbx, rcx);
1520 __ push(rcx); __ push(rbx);
1521 __ push(rdx); __ push(rax);
1522 __ lmul(2 * wordSize, 0);
1523 __ addptr(rsp, 4 * wordSize); // take off temporaries
1524 #endif
1525 }
1526
1527 void TemplateTable::ldiv() {
1528 transition(ltos, ltos);
1529 #ifdef _LP64
1530 __ mov(rcx, rax);
1531 __ pop_l(rax);
1532 // generate explicit div0 check
1533 __ testq(rcx, rcx);
1534 __ jump_cc(Assembler::zero,
1535 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1536 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1537 // they are not equal, one could do a normal division (no correction
1538 // needed), which may speed up this implementation for the common case.
1539 // (see also JVM spec., p.243 & p.271)
1540 __ corrected_idivq(rcx); // kills rbx
1541 #else
1542 __ pop_l(rbx, rcx);
1543 __ push(rcx); __ push(rbx);
1544 __ push(rdx); __ push(rax);
1545 // check if y = 0
1546 __ orl(rax, rdx);
1547 __ jump_cc(Assembler::zero,
1548 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1549 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1550 __ addptr(rsp, 4 * wordSize); // take off temporaries
1551 #endif
1552 }
1553
1554 void TemplateTable::lrem() {
1555 transition(ltos, ltos);
1556 #ifdef _LP64
1557 __ mov(rcx, rax);
1558 __ pop_l(rax);
1559 __ testq(rcx, rcx);
1560 __ jump_cc(Assembler::zero,
1561 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1562 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1563 // they are not equal, one could do a normal division (no correction
1564 // needed), which may speed up this implementation for the common case.
1565 // (see also JVM spec., p.243 & p.271)
1566 __ corrected_idivq(rcx); // kills rbx
1567 __ mov(rax, rdx);
1568 #else
1569 __ pop_l(rbx, rcx);
1570 __ push(rcx); __ push(rbx);
1571 __ push(rdx); __ push(rax);
1572 // check if y = 0
1573 __ orl(rax, rdx);
1574 __ jump_cc(Assembler::zero,
1575 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1576 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1577 __ addptr(rsp, 4 * wordSize);
1578 #endif
1579 }
1580
1581 void TemplateTable::lshl() {
1582 transition(itos, ltos);
1583 __ movl(rcx, rax); // get shift count
1584 #ifdef _LP64
1585 __ pop_l(rax); // get shift value
1586 __ shlq(rax);
1587 #else
1588 __ pop_l(rax, rdx); // get shift value
1589 __ lshl(rdx, rax);
1590 #endif
1591 }
1592
1593 void TemplateTable::lshr() {
1594 #ifdef _LP64
1595 transition(itos, ltos);
1596 __ movl(rcx, rax); // get shift count
1597 __ pop_l(rax); // get shift value
1598 __ sarq(rax);
1599 #else
1600 transition(itos, ltos);
1601 __ mov(rcx, rax); // get shift count
1602 __ pop_l(rax, rdx); // get shift value
1603 __ lshr(rdx, rax, true);
1604 #endif
1605 }
1606
1607 void TemplateTable::lushr() {
1608 transition(itos, ltos);
1609 #ifdef _LP64
1610 __ movl(rcx, rax); // get shift count
1611 __ pop_l(rax); // get shift value
1612 __ shrq(rax);
1613 #else
1614 __ mov(rcx, rax); // get shift count
1615 __ pop_l(rax, rdx); // get shift value
1616 __ lshr(rdx, rax);
1617 #endif
1618 }
1619
1620 void TemplateTable::fop2(Operation op) {
1621 transition(ftos, ftos);
1622
1623 if (UseSSE >= 1) {
1624 switch (op) {
1625 case add:
1626 __ addss(xmm0, at_rsp());
1627 __ addptr(rsp, Interpreter::stackElementSize);
1628 break;
1629 case sub:
1630 __ movflt(xmm1, xmm0);
1631 __ pop_f(xmm0);
1632 __ subss(xmm0, xmm1);
1633 break;
1634 case mul:
1635 __ mulss(xmm0, at_rsp());
1636 __ addptr(rsp, Interpreter::stackElementSize);
1637 break;
1638 case div:
1639 __ movflt(xmm1, xmm0);
1640 __ pop_f(xmm0);
1641 __ divss(xmm0, xmm1);
1642 break;
1643 case rem:
1644 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1645 // modulo operation. The frem method calls the function
1646 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1647 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1648 // (signalling or quiet) is returned.
1649 //
1650 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1651 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1652 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1653 // The fprem instruction used on x86_32 is functionally equivalent to
1654 // SharedRuntime::frem in that it returns a NaN.
1655 #ifdef _LP64
1656 __ movflt(xmm1, xmm0);
1657 __ pop_f(xmm0);
1658 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1659 #else
1660 __ push_f(xmm0);
1661 __ pop_f();
1662 __ fld_s(at_rsp());
1663 __ fremr(rax);
1664 __ f2ieee();
1665 __ pop(rax); // pop second operand off the stack
1666 __ push_f();
1667 __ pop_f(xmm0);
1668 #endif
1669 break;
1670 default:
1671 ShouldNotReachHere();
1672 break;
1673 }
1674 } else {
1675 #ifdef _LP64
1676 ShouldNotReachHere();
1677 #else
1678 switch (op) {
1679 case add: __ fadd_s (at_rsp()); break;
1680 case sub: __ fsubr_s(at_rsp()); break;
1681 case mul: __ fmul_s (at_rsp()); break;
1682 case div: __ fdivr_s(at_rsp()); break;
1683 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1684 default : ShouldNotReachHere();
1685 }
1686 __ f2ieee();
1687 __ pop(rax); // pop second operand off the stack
1688 #endif // _LP64
1689 }
1690 }
1691
1692 void TemplateTable::dop2(Operation op) {
1693 transition(dtos, dtos);
1694 if (UseSSE >= 2) {
1695 switch (op) {
1696 case add:
1697 __ addsd(xmm0, at_rsp());
1698 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1699 break;
1700 case sub:
1701 __ movdbl(xmm1, xmm0);
1702 __ pop_d(xmm0);
1703 __ subsd(xmm0, xmm1);
1704 break;
1705 case mul:
1706 __ mulsd(xmm0, at_rsp());
1707 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1708 break;
1709 case div:
1710 __ movdbl(xmm1, xmm0);
1711 __ pop_d(xmm0);
1712 __ divsd(xmm0, xmm1);
1713 break;
1714 case rem:
1715 // Similar to fop2(), the modulo operation is performed using the
1716 // SharedRuntime::drem method (on x86_64 platforms) or using the
1717 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1718 #ifdef _LP64
1719 __ movdbl(xmm1, xmm0);
1720 __ pop_d(xmm0);
1721 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1722 #else
1723 __ push_d(xmm0);
1724 __ pop_d();
1725 __ fld_d(at_rsp());
1726 __ fremr(rax);
1727 __ d2ieee();
1728 __ pop(rax);
1729 __ pop(rdx);
1730 __ push_d();
1731 __ pop_d(xmm0);
1732 #endif
1733 break;
1734 default:
1735 ShouldNotReachHere();
1736 break;
1737 }
1738 } else {
1739 #ifdef _LP64
1740 ShouldNotReachHere();
1741 #else
1742 switch (op) {
1743 case add: __ fadd_d (at_rsp()); break;
1744 case sub: __ fsubr_d(at_rsp()); break;
1745 case mul: {
1746 Label L_strict;
1747 Label L_join;
1748 const Address access_flags (rcx, Method::access_flags_offset());
1749 __ get_method(rcx);
1750 __ movl(rcx, access_flags);
1751 __ testl(rcx, JVM_ACC_STRICT);
1752 __ jccb(Assembler::notZero, L_strict);
1753 __ fmul_d (at_rsp());
1754 __ jmpb(L_join);
1755 __ bind(L_strict);
1756 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1757 __ fmulp();
1758 __ fmul_d (at_rsp());
1759 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1760 __ fmulp();
1761 __ bind(L_join);
1762 break;
1763 }
1764 case div: {
1765 Label L_strict;
1766 Label L_join;
1767 const Address access_flags (rcx, Method::access_flags_offset());
1768 __ get_method(rcx);
1769 __ movl(rcx, access_flags);
1770 __ testl(rcx, JVM_ACC_STRICT);
1771 __ jccb(Assembler::notZero, L_strict);
1772 __ fdivr_d(at_rsp());
1773 __ jmp(L_join);
1774 __ bind(L_strict);
1775 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1776 __ fmul_d (at_rsp());
1777 __ fdivrp();
1778 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1779 __ fmulp();
1780 __ bind(L_join);
1781 break;
1782 }
1783 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1784 default : ShouldNotReachHere();
1785 }
1786 __ d2ieee();
1787 // Pop double precision number from rsp.
1788 __ pop(rax);
1789 __ pop(rdx);
1790 #endif
1791 }
1792 }
1793
1794 void TemplateTable::ineg() {
1795 transition(itos, itos);
1796 __ negl(rax);
1797 }
1798
1799 void TemplateTable::lneg() {
1800 transition(ltos, ltos);
1801 LP64_ONLY(__ negq(rax));
1802 NOT_LP64(__ lneg(rdx, rax));
1803 }
1804
1805 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1806 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1807 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1808 // of 128-bits operands for SSE instructions.
1809 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1810 // Store the value to a 128-bits operand.
1811 operand[0] = lo;
1812 operand[1] = hi;
1813 return operand;
1814 }
1815
1816 // Buffer for 128-bits masks used by SSE instructions.
1817 static jlong float_signflip_pool[2*2];
1818 static jlong double_signflip_pool[2*2];
1819
1820 void TemplateTable::fneg() {
1821 transition(ftos, ftos);
1822 if (UseSSE >= 1) {
1823 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1824 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1825 } else {
1826 LP64_ONLY(ShouldNotReachHere());
1827 NOT_LP64(__ fchs());
1828 }
1829 }
1830
1831 void TemplateTable::dneg() {
1832 transition(dtos, dtos);
1833 if (UseSSE >= 2) {
1834 static jlong *double_signflip =
1835 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1836 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1837 } else {
1838 #ifdef _LP64
1839 ShouldNotReachHere();
1840 #else
1841 __ fchs();
1842 #endif
1843 }
1844 }
1845
1846 void TemplateTable::iinc() {
1847 transition(vtos, vtos);
1848 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1849 locals_index(rbx);
1850 __ addl(iaddress(rbx), rdx);
1851 }
1852
1853 void TemplateTable::wide_iinc() {
1854 transition(vtos, vtos);
1855 __ movl(rdx, at_bcp(4)); // get constant
1856 locals_index_wide(rbx);
1857 __ bswapl(rdx); // swap bytes & sign-extend constant
1858 __ sarl(rdx, 16);
1859 __ addl(iaddress(rbx), rdx);
1860 // Note: should probably use only one movl to get both
1861 // the index and the constant -> fix this
1862 }
1863
1864 void TemplateTable::convert() {
1865 #ifdef _LP64
1866 // Checking
1867 #ifdef ASSERT
1868 {
1869 TosState tos_in = ilgl;
1870 TosState tos_out = ilgl;
1871 switch (bytecode()) {
1872 case Bytecodes::_i2l: // fall through
1873 case Bytecodes::_i2f: // fall through
1874 case Bytecodes::_i2d: // fall through
1875 case Bytecodes::_i2b: // fall through
1876 case Bytecodes::_i2c: // fall through
1877 case Bytecodes::_i2s: tos_in = itos; break;
1878 case Bytecodes::_l2i: // fall through
1879 case Bytecodes::_l2f: // fall through
1880 case Bytecodes::_l2d: tos_in = ltos; break;
1881 case Bytecodes::_f2i: // fall through
1882 case Bytecodes::_f2l: // fall through
1883 case Bytecodes::_f2d: tos_in = ftos; break;
1884 case Bytecodes::_d2i: // fall through
1885 case Bytecodes::_d2l: // fall through
1886 case Bytecodes::_d2f: tos_in = dtos; break;
1887 default : ShouldNotReachHere();
1888 }
1889 switch (bytecode()) {
1890 case Bytecodes::_l2i: // fall through
1891 case Bytecodes::_f2i: // fall through
1892 case Bytecodes::_d2i: // fall through
1893 case Bytecodes::_i2b: // fall through
1894 case Bytecodes::_i2c: // fall through
1895 case Bytecodes::_i2s: tos_out = itos; break;
1896 case Bytecodes::_i2l: // fall through
1897 case Bytecodes::_f2l: // fall through
1898 case Bytecodes::_d2l: tos_out = ltos; break;
1899 case Bytecodes::_i2f: // fall through
1900 case Bytecodes::_l2f: // fall through
1901 case Bytecodes::_d2f: tos_out = ftos; break;
1902 case Bytecodes::_i2d: // fall through
1903 case Bytecodes::_l2d: // fall through
1904 case Bytecodes::_f2d: tos_out = dtos; break;
1905 default : ShouldNotReachHere();
1906 }
1907 transition(tos_in, tos_out);
1908 }
1909 #endif // ASSERT
1910
1911 static const int64_t is_nan = 0x8000000000000000L;
1912
1913 // Conversion
1914 switch (bytecode()) {
1915 case Bytecodes::_i2l:
1916 __ movslq(rax, rax);
1917 break;
1918 case Bytecodes::_i2f:
1919 __ cvtsi2ssl(xmm0, rax);
1920 break;
1921 case Bytecodes::_i2d:
1922 __ cvtsi2sdl(xmm0, rax);
1923 break;
1924 case Bytecodes::_i2b:
1925 __ movsbl(rax, rax);
1926 break;
1927 case Bytecodes::_i2c:
1928 __ movzwl(rax, rax);
1929 break;
1930 case Bytecodes::_i2s:
1931 __ movswl(rax, rax);
1932 break;
1933 case Bytecodes::_l2i:
1934 __ movl(rax, rax);
1935 break;
1936 case Bytecodes::_l2f:
1937 __ cvtsi2ssq(xmm0, rax);
1938 break;
1939 case Bytecodes::_l2d:
1940 __ cvtsi2sdq(xmm0, rax);
1941 break;
1942 case Bytecodes::_f2i:
1943 {
1944 Label L;
1945 __ cvttss2sil(rax, xmm0);
1946 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1947 __ jcc(Assembler::notEqual, L);
1948 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1949 __ bind(L);
1950 }
1951 break;
1952 case Bytecodes::_f2l:
1953 {
1954 Label L;
1955 __ cvttss2siq(rax, xmm0);
1956 // NaN or overflow/underflow?
1957 __ cmp64(rax, ExternalAddress((address) &is_nan));
1958 __ jcc(Assembler::notEqual, L);
1959 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1960 __ bind(L);
1961 }
1962 break;
1963 case Bytecodes::_f2d:
1964 __ cvtss2sd(xmm0, xmm0);
1965 break;
1966 case Bytecodes::_d2i:
1967 {
1968 Label L;
1969 __ cvttsd2sil(rax, xmm0);
1970 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1971 __ jcc(Assembler::notEqual, L);
1972 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1973 __ bind(L);
1974 }
1975 break;
1976 case Bytecodes::_d2l:
1977 {
1978 Label L;
1979 __ cvttsd2siq(rax, xmm0);
1980 // NaN or overflow/underflow?
1981 __ cmp64(rax, ExternalAddress((address) &is_nan));
1982 __ jcc(Assembler::notEqual, L);
1983 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1984 __ bind(L);
1985 }
1986 break;
1987 case Bytecodes::_d2f:
1988 __ cvtsd2ss(xmm0, xmm0);
1989 break;
1990 default:
1991 ShouldNotReachHere();
1992 }
1993 #else
1994 // Checking
1995 #ifdef ASSERT
1996 { TosState tos_in = ilgl;
1997 TosState tos_out = ilgl;
1998 switch (bytecode()) {
1999 case Bytecodes::_i2l: // fall through
2000 case Bytecodes::_i2f: // fall through
2001 case Bytecodes::_i2d: // fall through
2002 case Bytecodes::_i2b: // fall through
2003 case Bytecodes::_i2c: // fall through
2004 case Bytecodes::_i2s: tos_in = itos; break;
2005 case Bytecodes::_l2i: // fall through
2006 case Bytecodes::_l2f: // fall through
2007 case Bytecodes::_l2d: tos_in = ltos; break;
2008 case Bytecodes::_f2i: // fall through
2009 case Bytecodes::_f2l: // fall through
2010 case Bytecodes::_f2d: tos_in = ftos; break;
2011 case Bytecodes::_d2i: // fall through
2012 case Bytecodes::_d2l: // fall through
2013 case Bytecodes::_d2f: tos_in = dtos; break;
2014 default : ShouldNotReachHere();
2015 }
2016 switch (bytecode()) {
2017 case Bytecodes::_l2i: // fall through
2018 case Bytecodes::_f2i: // fall through
2019 case Bytecodes::_d2i: // fall through
2020 case Bytecodes::_i2b: // fall through
2021 case Bytecodes::_i2c: // fall through
2022 case Bytecodes::_i2s: tos_out = itos; break;
2023 case Bytecodes::_i2l: // fall through
2024 case Bytecodes::_f2l: // fall through
2025 case Bytecodes::_d2l: tos_out = ltos; break;
2026 case Bytecodes::_i2f: // fall through
2027 case Bytecodes::_l2f: // fall through
2028 case Bytecodes::_d2f: tos_out = ftos; break;
2029 case Bytecodes::_i2d: // fall through
2030 case Bytecodes::_l2d: // fall through
2031 case Bytecodes::_f2d: tos_out = dtos; break;
2032 default : ShouldNotReachHere();
2033 }
2034 transition(tos_in, tos_out);
2035 }
2036 #endif // ASSERT
2037
2038 // Conversion
2039 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
2040 switch (bytecode()) {
2041 case Bytecodes::_i2l:
2042 __ extend_sign(rdx, rax);
2043 break;
2044 case Bytecodes::_i2f:
2045 if (UseSSE >= 1) {
2046 __ cvtsi2ssl(xmm0, rax);
2047 } else {
2048 __ push(rax); // store int on tos
2049 __ fild_s(at_rsp()); // load int to ST0
2050 __ f2ieee(); // truncate to float size
2051 __ pop(rcx); // adjust rsp
2052 }
2053 break;
2054 case Bytecodes::_i2d:
2055 if (UseSSE >= 2) {
2056 __ cvtsi2sdl(xmm0, rax);
2057 } else {
2058 __ push(rax); // add one slot for d2ieee()
2059 __ push(rax); // store int on tos
2060 __ fild_s(at_rsp()); // load int to ST0
2061 __ d2ieee(); // truncate to double size
2062 __ pop(rcx); // adjust rsp
2063 __ pop(rcx);
2064 }
2065 break;
2066 case Bytecodes::_i2b:
2067 __ shll(rax, 24); // truncate upper 24 bits
2068 __ sarl(rax, 24); // and sign-extend byte
2069 LP64_ONLY(__ movsbl(rax, rax));
2070 break;
2071 case Bytecodes::_i2c:
2072 __ andl(rax, 0xFFFF); // truncate upper 16 bits
2073 LP64_ONLY(__ movzwl(rax, rax));
2074 break;
2075 case Bytecodes::_i2s:
2076 __ shll(rax, 16); // truncate upper 16 bits
2077 __ sarl(rax, 16); // and sign-extend short
2078 LP64_ONLY(__ movswl(rax, rax));
2079 break;
2080 case Bytecodes::_l2i:
2081 /* nothing to do */
2082 break;
2083 case Bytecodes::_l2f:
2084 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
2085 // 64-bit long values to floats. On 32-bit platforms it is not possible
2086 // to use that instruction with 64-bit operands, therefore the FPU is
2087 // used to perform the conversion.
2088 __ push(rdx); // store long on tos
2089 __ push(rax);
2090 __ fild_d(at_rsp()); // load long to ST0
2091 __ f2ieee(); // truncate to float size
2092 __ pop(rcx); // adjust rsp
2093 __ pop(rcx);
2094 if (UseSSE >= 1) {
2095 __ push_f();
2096 __ pop_f(xmm0);
2097 }
2098 break;
2099 case Bytecodes::_l2d:
2100 // On 32-bit platforms the FPU is used for conversion because on
2101 // 32-bit platforms it is not not possible to use the cvtsi2sdq
2102 // instruction with 64-bit operands.
2103 __ push(rdx); // store long on tos
2104 __ push(rax);
2105 __ fild_d(at_rsp()); // load long to ST0
2106 __ d2ieee(); // truncate to double size
2107 __ pop(rcx); // adjust rsp
2108 __ pop(rcx);
2109 if (UseSSE >= 2) {
2110 __ push_d();
2111 __ pop_d(xmm0);
2112 }
2113 break;
2114 case Bytecodes::_f2i:
2115 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
2116 // as it returns 0 for any NaN.
2117 if (UseSSE >= 1) {
2118 __ push_f(xmm0);
2119 } else {
2120 __ push(rcx); // reserve space for argument
2121 __ fstp_s(at_rsp()); // pass float argument on stack
2122 }
2123 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
2124 break;
2125 case Bytecodes::_f2l:
2126 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
2127 // as it returns 0 for any NaN.
2128 if (UseSSE >= 1) {
2129 __ push_f(xmm0);
2130 } else {
2131 __ push(rcx); // reserve space for argument
2132 __ fstp_s(at_rsp()); // pass float argument on stack
2133 }
2134 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2135 break;
2136 case Bytecodes::_f2d:
2137 if (UseSSE < 1) {
2138 /* nothing to do */
2139 } else if (UseSSE == 1) {
2140 __ push_f(xmm0);
2141 __ pop_f();
2142 } else { // UseSSE >= 2
2143 __ cvtss2sd(xmm0, xmm0);
2144 }
2145 break;
2146 case Bytecodes::_d2i:
2147 if (UseSSE >= 2) {
2148 __ push_d(xmm0);
2149 } else {
2150 __ push(rcx); // reserve space for argument
2151 __ push(rcx);
2152 __ fstp_d(at_rsp()); // pass double argument on stack
2153 }
2154 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2155 break;
2156 case Bytecodes::_d2l:
2157 if (UseSSE >= 2) {
2158 __ push_d(xmm0);
2159 } else {
2160 __ push(rcx); // reserve space for argument
2161 __ push(rcx);
2162 __ fstp_d(at_rsp()); // pass double argument on stack
2163 }
2164 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2165 break;
2166 case Bytecodes::_d2f:
2167 if (UseSSE <= 1) {
2168 __ push(rcx); // reserve space for f2ieee()
2169 __ f2ieee(); // truncate to float size
2170 __ pop(rcx); // adjust rsp
2171 if (UseSSE == 1) {
2172 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2173 // the conversion is performed using the FPU in this case.
2174 __ push_f();
2175 __ pop_f(xmm0);
2176 }
2177 } else { // UseSSE >= 2
2178 __ cvtsd2ss(xmm0, xmm0);
2179 }
2180 break;
2181 default :
2182 ShouldNotReachHere();
2183 }
2184 #endif
2185 }
2186
2187 void TemplateTable::lcmp() {
2188 transition(ltos, itos);
2189 #ifdef _LP64
2190 Label done;
2191 __ pop_l(rdx);
2192 __ cmpq(rdx, rax);
2193 __ movl(rax, -1);
2194 __ jccb(Assembler::less, done);
2195 __ setb(Assembler::notEqual, rax);
2196 __ movzbl(rax, rax);
2197 __ bind(done);
2198 #else
2199
2200 // y = rdx:rax
2201 __ pop_l(rbx, rcx); // get x = rcx:rbx
2202 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2203 __ mov(rax, rcx);
2204 #endif
2205 }
2206
2207 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2208 if ((is_float && UseSSE >= 1) ||
2209 (!is_float && UseSSE >= 2)) {
2210 Label done;
2211 if (is_float) {
2212 // XXX get rid of pop here, use ... reg, mem32
2213 __ pop_f(xmm1);
2214 __ ucomiss(xmm1, xmm0);
2215 } else {
2216 // XXX get rid of pop here, use ... reg, mem64
2217 __ pop_d(xmm1);
2218 __ ucomisd(xmm1, xmm0);
2219 }
2220 if (unordered_result < 0) {
2221 __ movl(rax, -1);
2222 __ jccb(Assembler::parity, done);
2223 __ jccb(Assembler::below, done);
2224 __ setb(Assembler::notEqual, rdx);
2225 __ movzbl(rax, rdx);
2226 } else {
2227 __ movl(rax, 1);
2228 __ jccb(Assembler::parity, done);
2229 __ jccb(Assembler::above, done);
2230 __ movl(rax, 0);
2231 __ jccb(Assembler::equal, done);
2232 __ decrementl(rax);
2233 }
2234 __ bind(done);
2235 } else {
2236 #ifdef _LP64
2237 ShouldNotReachHere();
2238 #else
2239 if (is_float) {
2240 __ fld_s(at_rsp());
2241 } else {
2242 __ fld_d(at_rsp());
2243 __ pop(rdx);
2244 }
2245 __ pop(rcx);
2246 __ fcmp2int(rax, unordered_result < 0);
2247 #endif // _LP64
2248 }
2249 }
2250
2251 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2252 if (ValueTypesThreadLocalRecycling) {
2253 Label no_vt_recycling, no_fixing_required;
2254 const Register thread1 = NOT_LP64(rbx) LP64_ONLY(r15_thread);
2255 NOT_LP64(__ get_thread(thread1));
2256 __ movptr(rbx, Address(thread1, in_bytes(JavaThread::vt_alloc_ptr_offset())));
2257 __ testptr(rbx, rbx);
2258 __ jcc(Assembler::zero, no_vt_recycling);
2259 __ movptr(rcx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize));
2260 __ testptr(rcx, rcx);
2261 __ jcc(Assembler::notZero, no_fixing_required);
2262 // vt_alloc_ptr in JavaThread is non-null but frame vt_alloc_ptr is null
2263 // which means frame vt_alloc_ptr needs to be initialized
2264 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::fix_frame_vt_alloc_ptr));
2265 __ movptr(rcx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize));
2266 __ bind(no_fixing_required);
2267 __ testptr(rcx, rbx);
2268 __ jcc(Assembler::equal, no_vt_recycling);
2269 __ andptr(rcx, VTBufferChunk::chunk_mask());
2270 __ movl(rcx, Address(rcx, VTBufferChunk::index_offset()));
2271 __ andptr(rbx, VTBufferChunk::chunk_mask());
2272 __ movl(rbx, Address(rbx, VTBufferChunk::index_offset()));
2273 __ subl(rbx, rcx);
2274 __ get_method(rcx);
2275 __ movl(rcx, Address(rcx, Method::max_vt_buffer_offset()));
2276 __ cmpl(rbx, rcx);
2277 __ jcc(Assembler::lessEqual, no_vt_recycling);
2278 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::recycle_buffered_values));
2279 __ bind(no_vt_recycling);
2280 }
2281
2282 __ get_method(rcx); // rcx holds method
2283 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2284 // holds bumped taken count
2285
2286 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2287 InvocationCounter::counter_offset();
2288 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2289 InvocationCounter::counter_offset();
2290
2291 // Load up edx with the branch displacement
2292 if (is_wide) {
2293 __ movl(rdx, at_bcp(1));
2294 } else {
2295 __ load_signed_short(rdx, at_bcp(1));
2296 }
2297 __ bswapl(rdx);
2298
2299 if (!is_wide) {
2300 __ sarl(rdx, 16);
2301 }
2302 LP64_ONLY(__ movl2ptr(rdx, rdx));
2303
2304 // Handle all the JSR stuff here, then exit.
2305 // It's much shorter and cleaner than intermingling with the non-JSR
2306 // normal-branch stuff occurring below.
2307 if (is_jsr) {
2308 // Pre-load the next target bytecode into rbx
2309 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2310
2311 // compute return address as bci in rax
2312 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2313 in_bytes(ConstMethod::codes_offset())));
2314 __ subptr(rax, Address(rcx, Method::const_offset()));
2315 // Adjust the bcp in r13 by the displacement in rdx
2316 __ addptr(rbcp, rdx);
2317 // jsr returns atos that is not an oop
2318 __ push_i(rax);
2319 __ dispatch_only(vtos, true);
2320 return;
2321 }
2322
2323 // Normal (non-jsr) branch handling
2324
2325 // Adjust the bcp in r13 by the displacement in rdx
2326 __ addptr(rbcp, rdx);
2327
2328 assert(UseLoopCounter || !UseOnStackReplacement,
2329 "on-stack-replacement requires loop counters");
2330 Label backedge_counter_overflow;
2331 Label profile_method;
2332 Label dispatch;
2333 if (UseLoopCounter) {
2334 // increment backedge counter for backward branches
2335 // rax: MDO
2336 // rbx: MDO bumped taken-count
2337 // rcx: method
2338 // rdx: target offset
2339 // r13: target bcp
2340 // r14: locals pointer
2341 __ testl(rdx, rdx); // check if forward or backward branch
2342 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2343
2344 // check if MethodCounters exists
2345 Label has_counters;
2346 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2347 __ testptr(rax, rax);
2348 __ jcc(Assembler::notZero, has_counters);
2349 __ push(rdx);
2350 __ push(rcx);
2351 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2352 rcx);
2353 __ pop(rcx);
2354 __ pop(rdx);
2355 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2356 __ testptr(rax, rax);
2357 __ jcc(Assembler::zero, dispatch);
2358 __ bind(has_counters);
2359
2360 if (TieredCompilation) {
2361 Label no_mdo;
2362 int increment = InvocationCounter::count_increment;
2363 if (ProfileInterpreter) {
2364 // Are we profiling?
2365 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2366 __ testptr(rbx, rbx);
2367 __ jccb(Assembler::zero, no_mdo);
2368 // Increment the MDO backedge counter
2369 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2370 in_bytes(InvocationCounter::counter_offset()));
2371 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2372 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2373 rax, false, Assembler::zero, &backedge_counter_overflow);
2374 __ jmp(dispatch);
2375 }
2376 __ bind(no_mdo);
2377 // Increment backedge counter in MethodCounters*
2378 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2379 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2380 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2381 rax, false, Assembler::zero, &backedge_counter_overflow);
2382 } else { // not TieredCompilation
2383 // increment counter
2384 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2385 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2386 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2387 __ movl(Address(rcx, be_offset), rax); // store counter
2388
2389 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2390
2391 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2392 __ addl(rax, Address(rcx, be_offset)); // add both counters
2393
2394 if (ProfileInterpreter) {
2395 // Test to see if we should create a method data oop
2396 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2397 __ jcc(Assembler::less, dispatch);
2398
2399 // if no method data exists, go to profile method
2400 __ test_method_data_pointer(rax, profile_method);
2401
2402 if (UseOnStackReplacement) {
2403 // check for overflow against rbx which is the MDO taken count
2404 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2405 __ jcc(Assembler::below, dispatch);
2406
2407 // When ProfileInterpreter is on, the backedge_count comes
2408 // from the MethodData*, which value does not get reset on
2409 // the call to frequency_counter_overflow(). To avoid
2410 // excessive calls to the overflow routine while the method is
2411 // being compiled, add a second test to make sure the overflow
2412 // function is called only once every overflow_frequency.
2413 const int overflow_frequency = 1024;
2414 __ andl(rbx, overflow_frequency - 1);
2415 __ jcc(Assembler::zero, backedge_counter_overflow);
2416
2417 }
2418 } else {
2419 if (UseOnStackReplacement) {
2420 // check for overflow against rax, which is the sum of the
2421 // counters
2422 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2423 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2424
2425 }
2426 }
2427 }
2428 __ bind(dispatch);
2429 }
2430
2431 // Pre-load the next target bytecode into rbx
2432 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2433
2434 // continue with the bytecode @ target
2435 // rax: return bci for jsr's, unused otherwise
2436 // rbx: target bytecode
2437 // r13: target bcp
2438 __ dispatch_only(vtos, true);
2439
2440 if (UseLoopCounter) {
2441 if (ProfileInterpreter) {
2442 // Out-of-line code to allocate method data oop.
2443 __ bind(profile_method);
2444 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2445 __ set_method_data_pointer_for_bcp();
2446 __ jmp(dispatch);
2447 }
2448
2449 if (UseOnStackReplacement) {
2450 // invocation counter overflow
2451 __ bind(backedge_counter_overflow);
2452 __ negptr(rdx);
2453 __ addptr(rdx, rbcp); // branch bcp
2454 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2455 __ call_VM(noreg,
2456 CAST_FROM_FN_PTR(address,
2457 InterpreterRuntime::frequency_counter_overflow),
2458 rdx);
2459
2460 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2461 // rdx: scratch
2462 // r14: locals pointer
2463 // r13: bcp
2464 __ testptr(rax, rax); // test result
2465 __ jcc(Assembler::zero, dispatch); // no osr if null
2466 // nmethod may have been invalidated (VM may block upon call_VM return)
2467 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2468 __ jcc(Assembler::notEqual, dispatch);
2469
2470 // We have the address of an on stack replacement routine in rax.
2471 // In preparation of invoking it, first we must migrate the locals
2472 // and monitors from off the interpreter frame on the stack.
2473 // Ensure to save the osr nmethod over the migration call,
2474 // it will be preserved in rbx.
2475 __ mov(rbx, rax);
2476
2477 NOT_LP64(__ get_thread(rcx));
2478
2479 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2480
2481 // rax is OSR buffer, move it to expected parameter location
2482 LP64_ONLY(__ mov(j_rarg0, rax));
2483 NOT_LP64(__ mov(rcx, rax));
2484 // We use j_rarg definitions here so that registers don't conflict as parameter
2485 // registers change across platforms as we are in the midst of a calling
2486 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2487
2488 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2489 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2490
2491 // pop the interpreter frame
2492 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2493 __ leave(); // remove frame anchor
2494 __ pop(retaddr); // get return address
2495 __ mov(rsp, sender_sp); // set sp to sender sp
2496 // Ensure compiled code always sees stack at proper alignment
2497 __ andptr(rsp, -(StackAlignmentInBytes));
2498
2499 // unlike x86 we need no specialized return from compiled code
2500 // to the interpreter or the call stub.
2501
2502 // push the return address
2503 __ push(retaddr);
2504
2505 // and begin the OSR nmethod
2506 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2507 }
2508 }
2509 }
2510
2511 void TemplateTable::if_0cmp(Condition cc) {
2512 transition(itos, vtos);
2513 // assume branch is more often taken than not (loops use backward branches)
2514 Label not_taken;
2515 __ testl(rax, rax);
2516 __ jcc(j_not(cc), not_taken);
2517 branch(false, false);
2518 __ bind(not_taken);
2519 __ profile_not_taken_branch(rax);
2520 }
2521
2522 void TemplateTable::if_icmp(Condition cc) {
2523 transition(itos, vtos);
2524 // assume branch is more often taken than not (loops use backward branches)
2525 Label not_taken;
2526 __ pop_i(rdx);
2527 __ cmpl(rdx, rax);
2528 __ jcc(j_not(cc), not_taken);
2529 branch(false, false);
2530 __ bind(not_taken);
2531 __ profile_not_taken_branch(rax);
2532 }
2533
2534 void TemplateTable::if_nullcmp(Condition cc) {
2535 transition(atos, vtos);
2536 // assume branch is more often taken than not (loops use backward branches)
2537 Label not_taken;
2538 __ testptr(rax, rax);
2539 __ jcc(j_not(cc), not_taken);
2540 branch(false, false);
2541 __ bind(not_taken);
2542 __ profile_not_taken_branch(rax);
2543 }
2544
2545 void TemplateTable::if_acmp(Condition cc) {
2546 transition(atos, vtos);
2547 // assume branch is more often taken than not (loops use backward branches)
2548 Label not_taken;
2549 __ pop_ptr(rdx);
2550 __ cmpoop(rdx, rax);
2551 __ jcc(j_not(cc), not_taken);
2552 branch(false, false);
2553 __ bind(not_taken);
2554 __ profile_not_taken_branch(rax);
2555 }
2556
2557 void TemplateTable::ret() {
2558 transition(vtos, vtos);
2559 locals_index(rbx);
2560 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2561 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2562 __ profile_ret(rbx, rcx);
2563 __ get_method(rax);
2564 __ movptr(rbcp, Address(rax, Method::const_offset()));
2565 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2566 ConstMethod::codes_offset()));
2567 __ dispatch_next(vtos, 0, true);
2568 }
2569
2570 void TemplateTable::wide_ret() {
2571 transition(vtos, vtos);
2572 locals_index_wide(rbx);
2573 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2574 __ profile_ret(rbx, rcx);
2575 __ get_method(rax);
2576 __ movptr(rbcp, Address(rax, Method::const_offset()));
2577 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2578 __ dispatch_next(vtos, 0, true);
2579 }
2580
2581 void TemplateTable::tableswitch() {
2582 Label default_case, continue_execution;
2583 transition(itos, vtos);
2584
2585 // align r13/rsi
2586 __ lea(rbx, at_bcp(BytesPerInt));
2587 __ andptr(rbx, -BytesPerInt);
2588 // load lo & hi
2589 __ movl(rcx, Address(rbx, BytesPerInt));
2590 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2591 __ bswapl(rcx);
2592 __ bswapl(rdx);
2593 // check against lo & hi
2594 __ cmpl(rax, rcx);
2595 __ jcc(Assembler::less, default_case);
2596 __ cmpl(rax, rdx);
2597 __ jcc(Assembler::greater, default_case);
2598 // lookup dispatch offset
2599 __ subl(rax, rcx);
2600 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2601 __ profile_switch_case(rax, rbx, rcx);
2602 // continue execution
2603 __ bind(continue_execution);
2604 __ bswapl(rdx);
2605 LP64_ONLY(__ movl2ptr(rdx, rdx));
2606 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2607 __ addptr(rbcp, rdx);
2608 __ dispatch_only(vtos, true);
2609 // handle default
2610 __ bind(default_case);
2611 __ profile_switch_default(rax);
2612 __ movl(rdx, Address(rbx, 0));
2613 __ jmp(continue_execution);
2614 }
2615
2616 void TemplateTable::lookupswitch() {
2617 transition(itos, itos);
2618 __ stop("lookupswitch bytecode should have been rewritten");
2619 }
2620
2621 void TemplateTable::fast_linearswitch() {
2622 transition(itos, vtos);
2623 Label loop_entry, loop, found, continue_execution;
2624 // bswap rax so we can avoid bswapping the table entries
2625 __ bswapl(rax);
2626 // align r13
2627 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2628 // this instruction (change offsets
2629 // below)
2630 __ andptr(rbx, -BytesPerInt);
2631 // set counter
2632 __ movl(rcx, Address(rbx, BytesPerInt));
2633 __ bswapl(rcx);
2634 __ jmpb(loop_entry);
2635 // table search
2636 __ bind(loop);
2637 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2638 __ jcc(Assembler::equal, found);
2639 __ bind(loop_entry);
2640 __ decrementl(rcx);
2641 __ jcc(Assembler::greaterEqual, loop);
2642 // default case
2643 __ profile_switch_default(rax);
2644 __ movl(rdx, Address(rbx, 0));
2645 __ jmp(continue_execution);
2646 // entry found -> get offset
2647 __ bind(found);
2648 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2649 __ profile_switch_case(rcx, rax, rbx);
2650 // continue execution
2651 __ bind(continue_execution);
2652 __ bswapl(rdx);
2653 __ movl2ptr(rdx, rdx);
2654 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2655 __ addptr(rbcp, rdx);
2656 __ dispatch_only(vtos, true);
2657 }
2658
2659 void TemplateTable::fast_binaryswitch() {
2660 transition(itos, vtos);
2661 // Implementation using the following core algorithm:
2662 //
2663 // int binary_search(int key, LookupswitchPair* array, int n) {
2664 // // Binary search according to "Methodik des Programmierens" by
2665 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2666 // int i = 0;
2667 // int j = n;
2668 // while (i+1 < j) {
2669 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2670 // // with Q: for all i: 0 <= i < n: key < a[i]
2671 // // where a stands for the array and assuming that the (inexisting)
2672 // // element a[n] is infinitely big.
2673 // int h = (i + j) >> 1;
2674 // // i < h < j
2675 // if (key < array[h].fast_match()) {
2676 // j = h;
2677 // } else {
2678 // i = h;
2679 // }
2680 // }
2681 // // R: a[i] <= key < a[i+1] or Q
2682 // // (i.e., if key is within array, i is the correct index)
2683 // return i;
2684 // }
2685
2686 // Register allocation
2687 const Register key = rax; // already set (tosca)
2688 const Register array = rbx;
2689 const Register i = rcx;
2690 const Register j = rdx;
2691 const Register h = rdi;
2692 const Register temp = rsi;
2693
2694 // Find array start
2695 NOT_LP64(__ save_bcp());
2696
2697 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2698 // get rid of this
2699 // instruction (change
2700 // offsets below)
2701 __ andptr(array, -BytesPerInt);
2702
2703 // Initialize i & j
2704 __ xorl(i, i); // i = 0;
2705 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2706
2707 // Convert j into native byteordering
2708 __ bswapl(j);
2709
2710 // And start
2711 Label entry;
2712 __ jmp(entry);
2713
2714 // binary search loop
2715 {
2716 Label loop;
2717 __ bind(loop);
2718 // int h = (i + j) >> 1;
2719 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2720 __ sarl(h, 1); // h = (i + j) >> 1;
2721 // if (key < array[h].fast_match()) {
2722 // j = h;
2723 // } else {
2724 // i = h;
2725 // }
2726 // Convert array[h].match to native byte-ordering before compare
2727 __ movl(temp, Address(array, h, Address::times_8));
2728 __ bswapl(temp);
2729 __ cmpl(key, temp);
2730 // j = h if (key < array[h].fast_match())
2731 __ cmov32(Assembler::less, j, h);
2732 // i = h if (key >= array[h].fast_match())
2733 __ cmov32(Assembler::greaterEqual, i, h);
2734 // while (i+1 < j)
2735 __ bind(entry);
2736 __ leal(h, Address(i, 1)); // i+1
2737 __ cmpl(h, j); // i+1 < j
2738 __ jcc(Assembler::less, loop);
2739 }
2740
2741 // end of binary search, result index is i (must check again!)
2742 Label default_case;
2743 // Convert array[i].match to native byte-ordering before compare
2744 __ movl(temp, Address(array, i, Address::times_8));
2745 __ bswapl(temp);
2746 __ cmpl(key, temp);
2747 __ jcc(Assembler::notEqual, default_case);
2748
2749 // entry found -> j = offset
2750 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2751 __ profile_switch_case(i, key, array);
2752 __ bswapl(j);
2753 LP64_ONLY(__ movslq(j, j));
2754
2755 NOT_LP64(__ restore_bcp());
2756 NOT_LP64(__ restore_locals()); // restore rdi
2757
2758 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2759 __ addptr(rbcp, j);
2760 __ dispatch_only(vtos, true);
2761
2762 // default case -> j = default offset
2763 __ bind(default_case);
2764 __ profile_switch_default(i);
2765 __ movl(j, Address(array, -2 * BytesPerInt));
2766 __ bswapl(j);
2767 LP64_ONLY(__ movslq(j, j));
2768
2769 NOT_LP64(__ restore_bcp());
2770 NOT_LP64(__ restore_locals());
2771
2772 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2773 __ addptr(rbcp, j);
2774 __ dispatch_only(vtos, true);
2775 }
2776
2777 void TemplateTable::_return(TosState state) {
2778 transition(state, state);
2779
2780 assert(_desc->calls_vm(),
2781 "inconsistent calls_vm information"); // call in remove_activation
2782
2783 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2784 assert(state == vtos, "only valid state");
2785 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2786 __ movptr(robj, aaddress(0));
2787 __ load_klass(rdi, robj);
2788 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2789 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2790 Label skip_register_finalizer;
2791 __ jcc(Assembler::zero, skip_register_finalizer);
2792
2793 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2794
2795 __ bind(skip_register_finalizer);
2796 }
2797
2798 if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2799 Label no_safepoint;
2800 NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2801 #ifdef _LP64
2802 __ testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2803 #else
2804 const Register thread = rdi;
2805 __ get_thread(thread);
2806 __ testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2807 #endif
2808 __ jcc(Assembler::zero, no_safepoint);
2809 __ push(state);
2810 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2811 InterpreterRuntime::at_safepoint));
2812 __ pop(state);
2813 __ bind(no_safepoint);
2814 }
2815
2816 if (state == atos) {
2817 if (ValueTypesBufferMaxMemory > 0) {
2818 Label notBuffered;
2819
2820 __ test_value_is_not_buffered(rax, rbx, notBuffered);
2821 const Register thread1 = NOT_LP64(rcx) LP64_ONLY(r15_thread);
2822 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::return_value), rax);
2823 NOT_LP64(__ get_thread(thread1));
2824 __ get_vm_result(rax, thread1);
2825 __ bind(notBuffered);
2826 }
2827 }
2828
2829 // Narrow result if state is itos but result type is smaller.
2830 // Need to narrow in the return bytecode rather than in generate_return_entry
2831 // since compiled code callers expect the result to already be narrowed.
2832 if (state == itos) {
2833 __ narrow(rax);
2834 }
2835
2836 __ remove_activation(state, rbcp, true, true, true, /*state == qtos*/ false && ValueTypeReturnedAsFields);
2837
2838 __ jmp(rbcp);
2839 }
2840
2841 // ----------------------------------------------------------------------------
2842 // Volatile variables demand their effects be made known to all CPU's
2843 // in order. Store buffers on most chips allow reads & writes to
2844 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2845 // without some kind of memory barrier (i.e., it's not sufficient that
2846 // the interpreter does not reorder volatile references, the hardware
2847 // also must not reorder them).
2848 //
2849 // According to the new Java Memory Model (JMM):
2850 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2851 // writes act as aquire & release, so:
2852 // (2) A read cannot let unrelated NON-volatile memory refs that
2853 // happen after the read float up to before the read. It's OK for
2854 // non-volatile memory refs that happen before the volatile read to
2855 // float down below it.
2856 // (3) Similar a volatile write cannot let unrelated NON-volatile
2857 // memory refs that happen BEFORE the write float down to after the
2858 // write. It's OK for non-volatile memory refs that happen after the
2859 // volatile write to float up before it.
2860 //
2861 // We only put in barriers around volatile refs (they are expensive),
2862 // not _between_ memory refs (that would require us to track the
2863 // flavor of the previous memory refs). Requirements (2) and (3)
2864 // require some barriers before volatile stores and after volatile
2865 // loads. These nearly cover requirement (1) but miss the
2866 // volatile-store-volatile-load case. This final case is placed after
2867 // volatile-stores although it could just as well go before
2868 // volatile-loads.
2869
2870 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2871 // Helper function to insert a is-volatile test and memory barrier
2872 if(!os::is_MP()) return; // Not needed on single CPU
2873 __ membar(order_constraint);
2874 }
2875
2876 void TemplateTable::resolve_cache_and_index(int byte_no,
2877 Register Rcache,
2878 Register index,
2879 size_t index_size) {
2880 const Register temp = rbx;
2881 assert_different_registers(Rcache, index, temp);
2882
2883 Label resolved;
2884
2885 Bytecodes::Code code = bytecode();
2886 switch (code) {
2887 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2888 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2889 default: break;
2890 }
2891
2892 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2893 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2894 __ cmpl(temp, code); // have we resolved this bytecode?
2895 __ jcc(Assembler::equal, resolved);
2896
2897 // resolve first time through
2898 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2899 __ movl(temp, code);
2900 __ call_VM(noreg, entry, temp);
2901 // Update registers with resolved info
2902 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2903 __ bind(resolved);
2904 }
2905
2906 // The cache and index registers must be set before call
2907 void TemplateTable::load_field_cp_cache_entry(Register obj,
2908 Register cache,
2909 Register index,
2910 Register off,
2911 Register flags,
2912 bool is_static = false) {
2913 assert_different_registers(cache, index, flags, off);
2914
2915 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2916 // Field offset
2917 __ movptr(off, Address(cache, index, Address::times_ptr,
2918 in_bytes(cp_base_offset +
2919 ConstantPoolCacheEntry::f2_offset())));
2920 // Flags
2921 __ movl(flags, Address(cache, index, Address::times_ptr,
2922 in_bytes(cp_base_offset +
2923 ConstantPoolCacheEntry::flags_offset())));
2924
2925 // klass overwrite register
2926 if (is_static) {
2927 __ movptr(obj, Address(cache, index, Address::times_ptr,
2928 in_bytes(cp_base_offset +
2929 ConstantPoolCacheEntry::f1_offset())));
2930 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2931 __ movptr(obj, Address(obj, mirror_offset));
2932 __ resolve_oop_handle(obj);
2933 }
2934 }
2935
2936 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2937 Register method,
2938 Register itable_index,
2939 Register flags,
2940 bool is_invokevirtual,
2941 bool is_invokevfinal, /*unused*/
2942 bool is_invokedynamic) {
2943 // setup registers
2944 const Register cache = rcx;
2945 const Register index = rdx;
2946 assert_different_registers(method, flags);
2947 assert_different_registers(method, cache, index);
2948 assert_different_registers(itable_index, flags);
2949 assert_different_registers(itable_index, cache, index);
2950 // determine constant pool cache field offsets
2951 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2952 const int method_offset = in_bytes(
2953 ConstantPoolCache::base_offset() +
2954 ((byte_no == f2_byte)
2955 ? ConstantPoolCacheEntry::f2_offset()
2956 : ConstantPoolCacheEntry::f1_offset()));
2957 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2958 ConstantPoolCacheEntry::flags_offset());
2959 // access constant pool cache fields
2960 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2961 ConstantPoolCacheEntry::f2_offset());
2962
2963 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2964 resolve_cache_and_index(byte_no, cache, index, index_size);
2965 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2966
2967 if (itable_index != noreg) {
2968 // pick up itable or appendix index from f2 also:
2969 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2970 }
2971 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2972 }
2973
2974 // The registers cache and index expected to be set before call.
2975 // Correct values of the cache and index registers are preserved.
2976 void TemplateTable::jvmti_post_field_access(Register cache,
2977 Register index,
2978 bool is_static,
2979 bool has_tos) {
2980 if (JvmtiExport::can_post_field_access()) {
2981 // Check to see if a field access watch has been set before we take
2982 // the time to call into the VM.
2983 Label L1;
2984 assert_different_registers(cache, index, rax);
2985 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2986 __ testl(rax,rax);
2987 __ jcc(Assembler::zero, L1);
2988
2989 // cache entry pointer
2990 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2991 __ shll(index, LogBytesPerWord);
2992 __ addptr(cache, index);
2993 if (is_static) {
2994 __ xorptr(rax, rax); // NULL object reference
2995 } else {
2996 __ pop(atos); // Get the object
2997 __ verify_oop(rax);
2998 __ push(atos); // Restore stack state
2999 }
3000 // rax,: object pointer or NULL
3001 // cache: cache entry pointer
3002 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
3003 rax, cache);
3004 __ get_cache_and_index_at_bcp(cache, index, 1);
3005 __ bind(L1);
3006 }
3007 }
3008
3009 void TemplateTable::pop_and_check_object(Register r) {
3010 __ pop_ptr(r);
3011 __ null_check(r); // for field access must check obj.
3012 __ verify_oop(r);
3013 }
3014
3015 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3016 transition(vtos, vtos);
3017
3018 const Register cache = rcx;
3019 const Register index = rdx;
3020 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3021 const Register off = rbx;
3022 const Register flags = rax;
3023 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
3024 const Register flags2 = rdx;
3025
3026 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3027 jvmti_post_field_access(cache, index, is_static, false);
3028 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3029
3030 const Address field(obj, off, Address::times_1, 0*wordSize);
3031 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
3032
3033 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notValueType, notDouble;
3034
3035 if (!is_static) {
3036 __ movptr(rcx, Address(cache, index, Address::times_ptr,
3037 in_bytes(ConstantPoolCache::base_offset() +
3038 ConstantPoolCacheEntry::f1_offset())));
3039 }
3040
3041 __ movl(flags2, flags);
3042
3043 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3044 // Make sure we don't need to mask edx after the above shift
3045 assert(btos == 0, "change code, btos != 0");
3046
3047 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3048
3049 __ jcc(Assembler::notZero, notByte);
3050 // btos
3051 if (!is_static) pop_and_check_object(obj);
3052 __ load_signed_byte(rax, field);
3053 __ push(btos);
3054 // Rewrite bytecode to be faster
3055 if (!is_static && rc == may_rewrite) {
3056 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
3057 }
3058 __ jmp(Done);
3059
3060 __ bind(notByte);
3061
3062 __ cmpl(flags, ztos);
3063 __ jcc(Assembler::notEqual, notBool);
3064 if (!is_static) pop_and_check_object(obj);
3065 // ztos (same code as btos)
3066 __ load_signed_byte(rax, field);
3067 __ push(ztos);
3068 // Rewrite bytecode to be faster
3069 if (!is_static && rc == may_rewrite) {
3070 // use btos rewriting, no truncating to t/f bit is needed for getfield.
3071 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
3072 }
3073 __ jmp(Done);
3074
3075 __ bind(notBool);
3076 __ cmpl(flags, atos);
3077 __ jcc(Assembler::notEqual, notObj);
3078 // atos
3079 if (!EnableValhalla) {
3080 if (!is_static) pop_and_check_object(obj);
3081 __ load_heap_oop(rax, field);
3082 __ push(atos);
3083 if (!is_static && rc == may_rewrite) {
3084 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
3085 }
3086 __ jmp(Done);
3087 } else {
3088 if (is_static) {
3089 __ load_heap_oop(rax, field);
3090 Label isFlattenable, uninitialized;
3091 // Issue below if the static field has not been initialized yet
3092 __ test_field_is_flattenable(flags2, rscratch1, isFlattenable);
3093 // Not flattenable case
3094 __ push(atos);
3095 __ jmp(Done);
3096 // Flattenable case, must not return null even if uninitialized
3097 __ bind(isFlattenable);
3098 __ testptr(rax, rax);
3099 __ jcc(Assembler::zero, uninitialized);
3100 __ push(atos);
3101 __ jmp(Done);
3102 __ bind(uninitialized);
3103 __ andl(flags2, ConstantPoolCacheEntry::field_index_mask);
3104 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_static_value_field),
3105 obj, flags2);
3106 __ verify_oop(rax);
3107 __ push(atos);
3108 __ jmp(Done);
3109 } else {
3110 Label isFlattened, nonnull, isFlattenable, rewriteFlattenable;
3111 __ test_field_is_flattenable(flags2, rscratch1, isFlattenable);
3112 // Non-flattenable field case, also covers the object case
3113 pop_and_check_object(obj);
3114 __ load_heap_oop(rax, field);
3115 __ push(atos);
3116 if (rc == may_rewrite) {
3117 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
3118 }
3119 __ jmp(Done);
3120 __ bind(isFlattenable);
3121 __ test_field_is_flattened(flags2, rscratch1, isFlattened);
3122 // Non-flattened field case
3123 pop_and_check_object(obj);
3124 __ load_heap_oop(rax, field);
3125 __ testptr(rax, rax);
3126 __ jcc(Assembler::notZero, nonnull);
3127 __ andl(flags2, ConstantPoolCacheEntry::field_index_mask);
3128 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_instance_value_field),
3129 obj, flags2);
3130 __ bind(nonnull);
3131 __ verify_oop(rax);
3132 __ push(atos);
3133 __ jmp(rewriteFlattenable);
3134 __ bind(isFlattened);
3135 __ andl(flags2, ConstantPoolCacheEntry::field_index_mask);
3136 pop_and_check_object(rbx);
3137 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flattened_field),
3138 rbx, flags2, rcx);
3139 __ verify_oop(rax);
3140 __ push(atos);
3141 __ bind(rewriteFlattenable);
3142 if (rc == may_rewrite) {
3143 patch_bytecode(Bytecodes::_fast_qgetfield, bc, rbx);
3144 }
3145 __ jmp(Done);
3146 }
3147 }
3148
3149 __ bind(notObj);
3150
3151 if (!is_static) pop_and_check_object(obj);
3152
3153 __ cmpl(flags, itos);
3154 __ jcc(Assembler::notEqual, notInt);
3155 // itos
3156 __ movl(rax, field);
3157 __ push(itos);
3158 // Rewrite bytecode to be faster
3159 if (!is_static && rc == may_rewrite) {
3160 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
3161 }
3162 __ jmp(Done);
3163
3164 __ bind(notInt);
3165 __ cmpl(flags, ctos);
3166 __ jcc(Assembler::notEqual, notChar);
3167 // ctos
3168 __ load_unsigned_short(rax, field);
3169 __ push(ctos);
3170 // Rewrite bytecode to be faster
3171 if (!is_static && rc == may_rewrite) {
3172 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
3173 }
3174 __ jmp(Done);
3175
3176 __ bind(notChar);
3177 __ cmpl(flags, stos);
3178 __ jcc(Assembler::notEqual, notShort);
3179 // stos
3180 __ load_signed_short(rax, field);
3181 __ push(stos);
3182 // Rewrite bytecode to be faster
3183 if (!is_static && rc == may_rewrite) {
3184 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
3185 }
3186 __ jmp(Done);
3187
3188 __ bind(notShort);
3189 __ cmpl(flags, ltos);
3190 __ jcc(Assembler::notEqual, notLong);
3191 // ltos
3192
3193 #ifndef _LP64
3194 // Generate code as if volatile. There just aren't enough registers to
3195 // save that information and this code is faster than the test.
3196 __ fild_d(field); // Must load atomically
3197 __ subptr(rsp,2*wordSize); // Make space for store
3198 __ fistp_d(Address(rsp,0));
3199 __ pop(rax);
3200 __ pop(rdx);
3201 #else
3202 __ movq(rax, field);
3203 #endif
3204
3205 __ push(ltos);
3206 // Rewrite bytecode to be faster
3207 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
3208 __ jmp(Done);
3209
3210 __ bind(notLong);
3211 __ cmpl(flags, ftos);
3212 __ jcc(Assembler::notEqual, notFloat);
3213 // ftos
3214
3215 __ load_float(field);
3216 __ push(ftos);
3217 // Rewrite bytecode to be faster
3218 if (!is_static && rc == may_rewrite) {
3219 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
3220 }
3221 __ jmp(Done);
3222
3223 __ bind(notFloat);
3224 #ifdef ASSERT
3225 __ cmpl(flags, dtos);
3226 __ jcc(Assembler::notEqual, notDouble);
3227 #endif
3228 // dtos
3229 __ load_double(field);
3230 __ push(dtos);
3231 // Rewrite bytecode to be faster
3232 if (!is_static && rc == may_rewrite) {
3233 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
3234 }
3235 #ifdef ASSERT
3236 __ jmp(Done);
3237
3238
3239 __ bind(notDouble);
3240 __ stop("Bad state");
3241 #endif
3242
3243 __ bind(Done);
3244 // [jk] not needed currently
3245 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3246 // Assembler::LoadStore));
3247 }
3248
3249 void TemplateTable::getfield(int byte_no) {
3250 getfield_or_static(byte_no, false);
3251 }
3252
3253 void TemplateTable::nofast_getfield(int byte_no) {
3254 getfield_or_static(byte_no, false, may_not_rewrite);
3255 }
3256
3257 void TemplateTable::getstatic(int byte_no) {
3258 getfield_or_static(byte_no, true);
3259 }
3260
3261 void TemplateTable::withfield() {
3262 transition(vtos, atos);
3263
3264 Register cache = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
3265 Register index = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3266
3267 resolve_cache_and_index(f2_byte, cache, index, sizeof(u2));
3268
3269 call_VM(rbx, CAST_FROM_FN_PTR(address, InterpreterRuntime::withfield), cache);
3270 // new value type is returned in rbx
3271 // stack adjustement is returned in rax
3272 __ verify_oop(rbx);
3273 __ addptr(rsp, rax);
3274 __ movptr(rax, rbx);
3275 }
3276
3277 // The registers cache and index expected to be set before call.
3278 // The function may destroy various registers, just not the cache and index registers.
3279 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3280
3281 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
3282 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
3283 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3284 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
3285
3286 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3287
3288 if (JvmtiExport::can_post_field_modification()) {
3289 // Check to see if a field modification watch has been set before
3290 // we take the time to call into the VM.
3291 Label L1;
3292 assert_different_registers(cache, index, rax);
3293 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3294 __ testl(rax, rax);
3295 __ jcc(Assembler::zero, L1);
3296
3297 __ get_cache_and_index_at_bcp(robj, RDX, 1);
3298
3299
3300 if (is_static) {
3301 // Life is simple. Null out the object pointer.
3302 __ xorl(RBX, RBX);
3303
3304 } else {
3305 // Life is harder. The stack holds the value on top, followed by
3306 // the object. We don't know the size of the value, though; it
3307 // could be one or two words depending on its type. As a result,
3308 // we must find the type to determine where the object is.
3309 #ifndef _LP64
3310 Label two_word, valsize_known;
3311 #endif
3312 __ movl(RCX, Address(robj, RDX,
3313 Address::times_ptr,
3314 in_bytes(cp_base_offset +
3315 ConstantPoolCacheEntry::flags_offset())));
3316 NOT_LP64(__ mov(rbx, rsp));
3317 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3318
3319 // Make sure we don't need to mask rcx after the above shift
3320 ConstantPoolCacheEntry::verify_tos_state_shift();
3321 #ifdef _LP64
3322 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
3323 __ cmpl(c_rarg3, ltos);
3324 __ cmovptr(Assembler::equal,
3325 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3326 __ cmpl(c_rarg3, dtos);
3327 __ cmovptr(Assembler::equal,
3328 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3329 #else
3330 __ cmpl(rcx, ltos);
3331 __ jccb(Assembler::equal, two_word);
3332 __ cmpl(rcx, dtos);
3333 __ jccb(Assembler::equal, two_word);
3334 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3335 __ jmpb(valsize_known);
3336
3337 __ bind(two_word);
3338 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3339
3340 __ bind(valsize_known);
3341 // setup object pointer
3342 __ movptr(rbx, Address(rbx, 0));
3343 #endif
3344 }
3345 // cache entry pointer
3346 __ addptr(robj, in_bytes(cp_base_offset));
3347 __ shll(RDX, LogBytesPerWord);
3348 __ addptr(robj, RDX);
3349 // object (tos)
3350 __ mov(RCX, rsp);
3351 // c_rarg1: object pointer set up above (NULL if static)
3352 // c_rarg2: cache entry pointer
3353 // c_rarg3: jvalue object on the stack
3354 __ call_VM(noreg,
3355 CAST_FROM_FN_PTR(address,
3356 InterpreterRuntime::post_field_modification),
3357 RBX, robj, RCX);
3358 __ get_cache_and_index_at_bcp(cache, index, 1);
3359 __ bind(L1);
3360 }
3361 }
3362
3363 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3364 transition(vtos, vtos);
3365
3366 const Register cache = rcx;
3367 const Register index = rdx;
3368 const Register obj = rcx;
3369 const Register off = rbx;
3370 const Register flags = rax;
3371 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3372 const Register flags2 = rdx;
3373
3374 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3375 jvmti_post_field_mod(cache, index, is_static);
3376 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3377
3378 // [jk] not needed currently
3379 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3380 // Assembler::StoreStore));
3381
3382 Label notVolatile, Done;
3383
3384 __ movl(flags2, flags);
3385
3386 // field addresses
3387 const Address field(obj, off, Address::times_1, 0*wordSize);
3388 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3389
3390 Label notByte, notBool, notInt, notShort, notChar,
3391 notLong, notFloat, notObj, notValueType, notDouble;
3392
3393 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3394
3395 assert(btos == 0, "change code, btos != 0");
3396 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3397 __ jcc(Assembler::notZero, notByte);
3398
3399 // btos
3400 {
3401 __ pop(btos);
3402 if (!is_static) pop_and_check_object(obj);
3403 __ movb(field, rax);
3404 if (!is_static && rc == may_rewrite) {
3405 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3406 }
3407 __ jmp(Done);
3408 }
3409
3410 __ bind(notByte);
3411 __ cmpl(flags, ztos);
3412 __ jcc(Assembler::notEqual, notBool);
3413
3414 // ztos
3415 {
3416 __ pop(ztos);
3417 if (!is_static) pop_and_check_object(obj);
3418 __ andl(rax, 0x1);
3419 __ movb(field, rax);
3420 if (!is_static && rc == may_rewrite) {
3421 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3422 }
3423 __ jmp(Done);
3424 }
3425
3426 __ bind(notBool);
3427 __ cmpl(flags, atos);
3428 __ jcc(Assembler::notEqual, notObj);
3429
3430 // atos
3431 {
3432 if (!EnableValhalla) {
3433 __ pop(atos);
3434 if (!is_static) pop_and_check_object(obj);
3435 // Store into the field
3436 do_oop_store(_masm, field, rax, _bs->kind(), false);
3437 if (!is_static && rc == may_rewrite) {
3438 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3439 }
3440 __ jmp(Done);
3441 } else {
3442 __ pop(atos);
3443 if (is_static) {
3444 Label notFlattenable, notBuffered;
3445 __ test_field_is_not_flattenable(flags2, rscratch1, notFlattenable);
3446 __ null_check(rax);
3447 __ bind(notFlattenable);
3448 if (ValueTypesBufferMaxMemory > 0) {
3449 __ test_value_is_not_buffered(rax, rscratch1, notBuffered);
3450 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_heap_copy),
3451 rax, off, obj);
3452 __ jmp(Done);
3453 __ bind(notBuffered);
3454 }
3455 do_oop_store(_masm, field, rax, _bs->kind(), false);
3456 __ jmp(Done);
3457 } else {
3458 Label isFlattenable, isFlattened, notBuffered, notBuffered2, rewriteNotFlattenable, rewriteFlattenable;
3459 __ test_field_is_flattenable(flags2, rscratch1, isFlattenable);
3460 // Not flattenable case, covers not flattenable values and objects
3461 pop_and_check_object(obj);
3462 // Store into the field
3463 if (ValueTypesBufferMaxMemory > 0) {
3464 __ test_value_is_not_buffered(rax, rscratch1, notBuffered);
3465 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_heap_copy),
3466 rax, off, obj);
3467 __ jmp(rewriteNotFlattenable);
3468 __ bind(notBuffered);
3469 }
3470 do_oop_store(_masm, field, rax, _bs->kind(), false);
3471 __ bind(rewriteNotFlattenable);
3472 if (rc == may_rewrite) {
3473 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3474 }
3475 __ jmp(Done);
3476 // Implementation of the flattenable semantic
3477 __ bind(isFlattenable);
3478 __ null_check(rax);
3479 __ test_field_is_flattened(flags2, rscratch1, isFlattened);
3480 // Not flattened case
3481 if (ValueTypesBufferMaxMemory > 0) {
3482 __ test_value_is_not_buffered(rax, rscratch1, notBuffered2);
3483 pop_and_check_object(obj);
3484 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_heap_copy),
3485 rax, off, obj);
3486 __ jmp(rewriteFlattenable);
3487 __ bind(notBuffered2);
3488 }
3489 pop_and_check_object(obj);
3490 // Store into the field
3491 do_oop_store(_masm, field, rax, _bs->kind(), false);
3492 __ jmp(rewriteFlattenable);
3493 __ bind(isFlattened);
3494 pop_and_check_object(obj);
3495 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_flattened_value),
3496 rax, off, obj);
3497 __ bind(rewriteFlattenable);
3498 if (rc == may_rewrite) {
3499 patch_bytecode(Bytecodes::_fast_qputfield, bc, rbx, true, byte_no);
3500 }
3501 __ jmp(Done);
3502 }
3503 }
3504 }
3505
3506 __ bind(notObj);
3507 __ cmpl(flags, itos);
3508 __ jcc(Assembler::notEqual, notInt);
3509
3510 // itos
3511 {
3512 __ pop(itos);
3513 if (!is_static) pop_and_check_object(obj);
3514 __ movl(field, rax);
3515 if (!is_static && rc == may_rewrite) {
3516 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3517 }
3518 __ jmp(Done);
3519 }
3520
3521 __ bind(notInt);
3522 __ cmpl(flags, ctos);
3523 __ jcc(Assembler::notEqual, notChar);
3524
3525 // ctos
3526 {
3527 __ pop(ctos);
3528 if (!is_static) pop_and_check_object(obj);
3529 __ movw(field, rax);
3530 if (!is_static && rc == may_rewrite) {
3531 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3532 }
3533 __ jmp(Done);
3534 }
3535
3536 __ bind(notChar);
3537 __ cmpl(flags, stos);
3538 __ jcc(Assembler::notEqual, notShort);
3539
3540 // stos
3541 {
3542 __ pop(stos);
3543 if (!is_static) pop_and_check_object(obj);
3544 __ movw(field, rax);
3545 if (!is_static && rc == may_rewrite) {
3546 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3547 }
3548 __ jmp(Done);
3549 }
3550
3551 __ bind(notShort);
3552 __ cmpl(flags, ltos);
3553 __ jcc(Assembler::notEqual, notLong);
3554
3555 // ltos
3556 #ifdef _LP64
3557 {
3558 __ pop(ltos);
3559 if (!is_static) pop_and_check_object(obj);
3560 __ movq(field, rax);
3561 if (!is_static && rc == may_rewrite) {
3562 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3563 }
3564 __ jmp(Done);
3565 }
3566 #else
3567 {
3568 Label notVolatileLong;
3569 __ testl(rdx, rdx);
3570 __ jcc(Assembler::zero, notVolatileLong);
3571
3572 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3573 if (!is_static) pop_and_check_object(obj);
3574
3575 // Replace with real volatile test
3576 __ push(rdx);
3577 __ push(rax); // Must update atomically with FIST
3578 __ fild_d(Address(rsp,0)); // So load into FPU register
3579 __ fistp_d(field); // and put into memory atomically
3580 __ addptr(rsp, 2*wordSize);
3581 // volatile_barrier();
3582 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3583 Assembler::StoreStore));
3584 // Don't rewrite volatile version
3585 __ jmp(notVolatile);
3586
3587 __ bind(notVolatileLong);
3588
3589 __ pop(ltos); // overwrites rdx
3590 if (!is_static) pop_and_check_object(obj);
3591 __ movptr(hi, rdx);
3592 __ movptr(field, rax);
3593 // Don't rewrite to _fast_lputfield for potential volatile case.
3594 __ jmp(notVolatile);
3595 }
3596 #endif // _LP64
3597
3598 __ bind(notLong);
3599 __ cmpl(flags, ftos);
3600 __ jcc(Assembler::notEqual, notFloat);
3601
3602 // ftos
3603 {
3604 __ pop(ftos);
3605 if (!is_static) pop_and_check_object(obj);
3606 __ store_float(field);
3607 if (!is_static && rc == may_rewrite) {
3608 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3609 }
3610 __ jmp(Done);
3611 }
3612
3613 __ bind(notFloat);
3614 #ifdef ASSERT
3615 __ cmpl(flags, dtos);
3616 __ jcc(Assembler::notEqual, notDouble);
3617 #endif
3618
3619 // dtos
3620 {
3621 __ pop(dtos);
3622 if (!is_static) pop_and_check_object(obj);
3623 __ store_double(field);
3624 if (!is_static && rc == may_rewrite) {
3625 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3626 }
3627 }
3628
3629 #ifdef ASSERT
3630 __ jmp(Done);
3631
3632 __ bind(notDouble);
3633 __ stop("Bad state");
3634 #endif
3635
3636 __ bind(Done);
3637
3638 __ shrl(flags2, ConstantPoolCacheEntry::is_volatile_shift);
3639 __ andl(flags2, 0x1);
3640
3641 // Check for volatile store
3642 __ testl(flags2, flags2);
3643 __ jcc(Assembler::zero, notVolatile);
3644 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3645 Assembler::StoreStore));
3646 __ bind(notVolatile);
3647 }
3648
3649 void TemplateTable::putfield(int byte_no) {
3650 putfield_or_static(byte_no, false);
3651 }
3652
3653 void TemplateTable::nofast_putfield(int byte_no) {
3654 putfield_or_static(byte_no, false, may_not_rewrite);
3655 }
3656
3657 void TemplateTable::putstatic(int byte_no) {
3658 putfield_or_static(byte_no, true);
3659 }
3660
3661 void TemplateTable::jvmti_post_fast_field_mod() {
3662
3663 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3664
3665 if (JvmtiExport::can_post_field_modification()) {
3666 // Check to see if a field modification watch has been set before
3667 // we take the time to call into the VM.
3668 Label L2;
3669 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3670 __ testl(scratch, scratch);
3671 __ jcc(Assembler::zero, L2);
3672 __ pop_ptr(rbx); // copy the object pointer from tos
3673 __ verify_oop(rbx);
3674 __ push_ptr(rbx); // put the object pointer back on tos
3675 // Save tos values before call_VM() clobbers them. Since we have
3676 // to do it for every data type, we use the saved values as the
3677 // jvalue object.
3678 switch (bytecode()) { // load values into the jvalue object
3679 case Bytecodes::_fast_qputfield: //fall through
3680 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3681 case Bytecodes::_fast_bputfield: // fall through
3682 case Bytecodes::_fast_zputfield: // fall through
3683 case Bytecodes::_fast_sputfield: // fall through
3684 case Bytecodes::_fast_cputfield: // fall through
3685 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3686 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3687 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3688 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3689
3690 default:
3691 ShouldNotReachHere();
3692 }
3693 __ mov(scratch, rsp); // points to jvalue on the stack
3694 // access constant pool cache entry
3695 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3696 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3697 __ verify_oop(rbx);
3698 // rbx: object pointer copied above
3699 // c_rarg2: cache entry pointer
3700 // c_rarg3: jvalue object on the stack
3701 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3702 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3703
3704 switch (bytecode()) { // restore tos values
3705 case Bytecodes::_fast_qputfield: // fall through
3706 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3707 case Bytecodes::_fast_bputfield: // fall through
3708 case Bytecodes::_fast_zputfield: // fall through
3709 case Bytecodes::_fast_sputfield: // fall through
3710 case Bytecodes::_fast_cputfield: // fall through
3711 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3712 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3713 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3714 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3715 default: break;
3716 }
3717 __ bind(L2);
3718 }
3719 }
3720
3721 void TemplateTable::fast_storefield(TosState state) {
3722 transition(state, vtos);
3723
3724 ByteSize base = ConstantPoolCache::base_offset();
3725
3726 jvmti_post_fast_field_mod();
3727
3728 // access constant pool cache
3729 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3730
3731 // test for volatile with rdx but rdx is tos register for lputfield.
3732 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3733 in_bytes(base +
3734 ConstantPoolCacheEntry::flags_offset())));
3735
3736 // replace index with field offset from cache entry
3737 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3738 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3739
3740 // [jk] not needed currently
3741 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3742 // Assembler::StoreStore));
3743
3744 if (bytecode() == Bytecodes::_fast_qputfield) {
3745 __ movl(rscratch2, rdx);
3746 }
3747
3748 Label notVolatile;
3749 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3750 __ andl(rdx, 0x1);
3751
3752 // Get object from stack
3753 pop_and_check_object(rcx);
3754
3755 // field address
3756 const Address field(rcx, rbx, Address::times_1);
3757
3758 // access field
3759 switch (bytecode()) {
3760 case Bytecodes::_fast_qputfield:
3761 {
3762 Label isFlattened, notBuffered, done;
3763 __ null_check(rax);
3764 __ test_field_is_flattened(rscratch2, rscratch1, isFlattened);
3765 // No Flattened case
3766 if (ValueTypesBufferMaxMemory > 0) {
3767 __ test_value_is_not_buffered(rax, rscratch1, notBuffered);
3768 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_heap_copy),
3769 rax, rbx, rcx);
3770 __ jmp(done);
3771 __ bind(notBuffered);
3772 }
3773 do_oop_store(_masm, field, rax, _bs->kind(), false);
3774 __ jmp(done);
3775 __ bind(isFlattened);
3776 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_flattened_value),
3777 rax, rbx, rcx);
3778 __ bind(done);
3779 }
3780 break;
3781 case Bytecodes::_fast_aputfield:
3782 {
3783 Label notBuffered, done;
3784 if (ValueTypesBufferMaxMemory > 0) {
3785 __ test_value_is_not_buffered(rax, rscratch1, notBuffered);
3786 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_heap_copy),
3787 rax, rbx, rcx);
3788 __ jmp(done);
3789 __ bind(notBuffered);
3790 }
3791 do_oop_store(_masm, field, rax, _bs->kind(), false);
3792 __ bind(done);
3793 }
3794 break;
3795 case Bytecodes::_fast_lputfield:
3796 #ifdef _LP64
3797 __ movq(field, rax);
3798 #else
3799 __ stop("should not be rewritten");
3800 #endif
3801 break;
3802 case Bytecodes::_fast_iputfield:
3803 __ movl(field, rax);
3804 break;
3805 case Bytecodes::_fast_zputfield:
3806 __ andl(rax, 0x1); // boolean is true if LSB is 1
3807 // fall through to bputfield
3808 case Bytecodes::_fast_bputfield:
3809 __ movb(field, rax);
3810 break;
3811 case Bytecodes::_fast_sputfield:
3812 // fall through
3813 case Bytecodes::_fast_cputfield:
3814 __ movw(field, rax);
3815 break;
3816 case Bytecodes::_fast_fputfield:
3817 __ store_float(field);
3818 break;
3819 case Bytecodes::_fast_dputfield:
3820 __ store_double(field);
3821 break;
3822 default:
3823 ShouldNotReachHere();
3824 }
3825
3826 // Check for volatile store
3827 __ testl(rdx, rdx);
3828 __ jcc(Assembler::zero, notVolatile);
3829 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3830 Assembler::StoreStore));
3831 __ bind(notVolatile);
3832 }
3833
3834 void TemplateTable::fast_accessfield(TosState state) {
3835 transition(atos, state);
3836
3837 // Do the JVMTI work here to avoid disturbing the register state below
3838 if (JvmtiExport::can_post_field_access()) {
3839 // Check to see if a field access watch has been set before we
3840 // take the time to call into the VM.
3841 Label L1;
3842 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3843 __ testl(rcx, rcx);
3844 __ jcc(Assembler::zero, L1);
3845 // access constant pool cache entry
3846 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3847 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3848 __ verify_oop(rax);
3849 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3850 LP64_ONLY(__ mov(c_rarg1, rax));
3851 // c_rarg1: object pointer copied above
3852 // c_rarg2: cache entry pointer
3853 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3854 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3855 __ pop_ptr(rax); // restore object pointer
3856 __ bind(L1);
3857 }
3858
3859 // access constant pool cache
3860 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3861 // replace index with field offset from cache entry
3862 // [jk] not needed currently
3863 // if (os::is_MP()) {
3864 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3865 // in_bytes(ConstantPoolCache::base_offset() +
3866 // ConstantPoolCacheEntry::flags_offset())));
3867 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3868 // __ andl(rdx, 0x1);
3869 // }
3870 __ movptr(rdx, Address(rcx, rbx, Address::times_ptr,
3871 in_bytes(ConstantPoolCache::base_offset() +
3872 ConstantPoolCacheEntry::f2_offset())));
3873
3874 // rax: object
3875 __ verify_oop(rax);
3876 __ null_check(rax);
3877 Address field(rax, rdx, Address::times_1);
3878
3879 // access field
3880 switch (bytecode()) {
3881 case Bytecodes::_fast_qgetfield:
3882 {
3883 Label isFlattened, nonnull, Done;
3884 __ movptr(rscratch1, Address(rcx, rbx, Address::times_ptr,
3885 in_bytes(ConstantPoolCache::base_offset() +
3886 ConstantPoolCacheEntry::flags_offset())));
3887 __ test_field_is_flattened(rscratch1, rscratch2, isFlattened);
3888 // Non-flattened field case
3889 __ movptr(rscratch1, rax);
3890 __ load_heap_oop(rax, field);
3891 __ testptr(rax, rax);
3892 __ jcc(Assembler::notZero, nonnull);
3893 __ movptr(rax, rscratch1);
3894 __ movl(rcx, Address(rcx, rbx, Address::times_ptr,
3895 in_bytes(ConstantPoolCache::base_offset() +
3896 ConstantPoolCacheEntry::flags_offset())));
3897 __ andl(rcx, ConstantPoolCacheEntry::field_index_mask);
3898 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_instance_value_field),
3899 rax, rcx);
3900 __ bind(nonnull);
3901 __ verify_oop(rax);
3902 __ jmp(Done);
3903 __ bind(isFlattened);
3904 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3905 in_bytes(ConstantPoolCache::base_offset() +
3906 ConstantPoolCacheEntry::flags_offset())));
3907 __ andl(rdx, ConstantPoolCacheEntry::field_index_mask);
3908 __ movptr(rcx, Address(rcx, rbx, Address::times_ptr,
3909 in_bytes(ConstantPoolCache::base_offset() +
3910 ConstantPoolCacheEntry::f1_offset())));
3911 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flattened_field),
3912 rax, rdx, rcx);
3913 __ verify_oop(rax);
3914 __ bind(Done);
3915 }
3916 break;
3917 case Bytecodes::_fast_agetfield:
3918 __ load_heap_oop(rax, field);
3919 __ verify_oop(rax);
3920 break;
3921 case Bytecodes::_fast_lgetfield:
3922 #ifdef _LP64
3923 __ movq(rax, field);
3924 #else
3925 __ stop("should not be rewritten");
3926 #endif
3927 break;
3928 case Bytecodes::_fast_igetfield:
3929 __ movl(rax, field);
3930 break;
3931 case Bytecodes::_fast_bgetfield:
3932 __ movsbl(rax, field);
3933 break;
3934 case Bytecodes::_fast_sgetfield:
3935 __ load_signed_short(rax, field);
3936 break;
3937 case Bytecodes::_fast_cgetfield:
3938 __ load_unsigned_short(rax, field);
3939 break;
3940 case Bytecodes::_fast_fgetfield:
3941 __ load_float(field);
3942 break;
3943 case Bytecodes::_fast_dgetfield:
3944 __ load_double(field);
3945 break;
3946 default:
3947 ShouldNotReachHere();
3948 }
3949 // [jk] not needed currently
3950 // if (os::is_MP()) {
3951 // Label notVolatile;
3952 // __ testl(rdx, rdx);
3953 // __ jcc(Assembler::zero, notVolatile);
3954 // __ membar(Assembler::LoadLoad);
3955 // __ bind(notVolatile);
3956 //};
3957 }
3958
3959 void TemplateTable::fast_xaccess(TosState state) {
3960 transition(vtos, state);
3961
3962 // get receiver
3963 __ movptr(rax, aaddress(0));
3964 // access constant pool cache
3965 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3966 __ movptr(rbx,
3967 Address(rcx, rdx, Address::times_ptr,
3968 in_bytes(ConstantPoolCache::base_offset() +
3969 ConstantPoolCacheEntry::f2_offset())));
3970 // make sure exception is reported in correct bcp range (getfield is
3971 // next instruction)
3972 __ increment(rbcp);
3973 __ null_check(rax);
3974 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3975 switch (state) {
3976 case itos:
3977 __ movl(rax, field);
3978 break;
3979 case atos:
3980 __ load_heap_oop(rax, field);
3981 __ verify_oop(rax);
3982 break;
3983 case ftos:
3984 __ load_float(field);
3985 break;
3986 default:
3987 ShouldNotReachHere();
3988 }
3989
3990 // [jk] not needed currently
3991 // if (os::is_MP()) {
3992 // Label notVolatile;
3993 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3994 // in_bytes(ConstantPoolCache::base_offset() +
3995 // ConstantPoolCacheEntry::flags_offset())));
3996 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3997 // __ testl(rdx, 0x1);
3998 // __ jcc(Assembler::zero, notVolatile);
3999 // __ membar(Assembler::LoadLoad);
4000 // __ bind(notVolatile);
4001 // }
4002
4003 __ decrement(rbcp);
4004 }
4005
4006 //-----------------------------------------------------------------------------
4007 // Calls
4008
4009 void TemplateTable::count_calls(Register method, Register temp) {
4010 // implemented elsewhere
4011 ShouldNotReachHere();
4012 }
4013
4014 void TemplateTable::prepare_invoke(int byte_no,
4015 Register method, // linked method (or i-klass)
4016 Register index, // itable index, MethodType, etc.
4017 Register recv, // if caller wants to see it
4018 Register flags // if caller wants to test it
4019 ) {
4020 // determine flags
4021 const Bytecodes::Code code = bytecode();
4022 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
4023 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
4024 const bool is_invokehandle = code == Bytecodes::_invokehandle;
4025 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
4026 const bool is_invokespecial = code == Bytecodes::_invokespecial;
4027 const bool load_receiver = (recv != noreg);
4028 const bool save_flags = (flags != noreg);
4029 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
4030 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
4031 assert(flags == noreg || flags == rdx, "");
4032 assert(recv == noreg || recv == rcx, "");
4033
4034 // setup registers & access constant pool cache
4035 if (recv == noreg) recv = rcx;
4036 if (flags == noreg) flags = rdx;
4037 assert_different_registers(method, index, recv, flags);
4038
4039 // save 'interpreter return address'
4040 __ save_bcp();
4041
4042 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
4043
4044 // maybe push appendix to arguments (just before return address)
4045 if (is_invokedynamic || is_invokehandle) {
4046 Label L_no_push;
4047 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
4048 __ jcc(Assembler::zero, L_no_push);
4049 // Push the appendix as a trailing parameter.
4050 // This must be done before we get the receiver,
4051 // since the parameter_size includes it.
4052 __ push(rbx);
4053 __ mov(rbx, index);
4054 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
4055 __ load_resolved_reference_at_index(index, rbx);
4056 __ pop(rbx);
4057 __ push(index); // push appendix (MethodType, CallSite, etc.)
4058 __ bind(L_no_push);
4059 }
4060
4061 // load receiver if needed (after appendix is pushed so parameter size is correct)
4062 // Note: no return address pushed yet
4063 if (load_receiver) {
4064 __ movl(recv, flags);
4065 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
4066 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
4067 const int receiver_is_at_end = -1; // back off one slot to get receiver
4068 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
4069 __ movptr(recv, recv_addr);
4070 __ verify_oop(recv);
4071 }
4072
4073 if (save_flags) {
4074 __ movl(rbcp, flags);
4075 }
4076
4077 // compute return type
4078 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
4079 // Make sure we don't need to mask flags after the above shift
4080 ConstantPoolCacheEntry::verify_tos_state_shift();
4081 // load return address
4082 {
4083 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
4084 ExternalAddress table(table_addr);
4085 LP64_ONLY(__ lea(rscratch1, table));
4086 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
4087 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
4088 }
4089
4090 // push return address
4091 __ push(flags);
4092
4093 // Restore flags value from the constant pool cache, and restore rsi
4094 // for later null checks. r13 is the bytecode pointer
4095 if (save_flags) {
4096 __ movl(flags, rbcp);
4097 __ restore_bcp();
4098 }
4099 }
4100
4101 void TemplateTable::invokevirtual_helper(Register index,
4102 Register recv,
4103 Register flags) {
4104 // Uses temporary registers rax, rdx
4105 assert_different_registers(index, recv, rax, rdx);
4106 assert(index == rbx, "");
4107 assert(recv == rcx, "");
4108
4109 // Test for an invoke of a final method
4110 Label notFinal;
4111 __ movl(rax, flags);
4112 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
4113 __ jcc(Assembler::zero, notFinal);
4114
4115 const Register method = index; // method must be rbx
4116 assert(method == rbx,
4117 "Method* must be rbx for interpreter calling convention");
4118
4119 // do the call - the index is actually the method to call
4120 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
4121
4122 // It's final, need a null check here!
4123 __ null_check(recv);
4124
4125 // profile this call
4126 __ profile_final_call(rax);
4127 __ profile_arguments_type(rax, method, rbcp, true);
4128
4129 __ jump_from_interpreted(method, rax);
4130
4131 __ bind(notFinal);
4132
4133 // get receiver klass
4134 __ null_check(recv, oopDesc::klass_offset_in_bytes());
4135 __ load_klass(rax, recv);
4136
4137 // profile this call
4138 __ profile_virtual_call(rax, rlocals, rdx);
4139 // get target Method* & entry point
4140 __ lookup_virtual_method(rax, index, method);
4141 __ profile_called_method(method, rdx, rbcp);
4142
4143 __ profile_arguments_type(rdx, method, rbcp, true);
4144 __ jump_from_interpreted(method, rdx);
4145 }
4146
4147 void TemplateTable::invokevirtual(int byte_no) {
4148 transition(vtos, vtos);
4149 assert(byte_no == f2_byte, "use this argument");
4150 prepare_invoke(byte_no,
4151 rbx, // method or vtable index
4152 noreg, // unused itable index
4153 rcx, rdx); // recv, flags
4154
4155 // rbx: index
4156 // rcx: receiver
4157 // rdx: flags
4158
4159 invokevirtual_helper(rbx, rcx, rdx);
4160 }
4161
4162 void TemplateTable::invokespecial(int byte_no) {
4163 transition(vtos, vtos);
4164 assert(byte_no == f1_byte, "use this argument");
4165 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
4166 rcx); // get receiver also for null check
4167 __ verify_oop(rcx);
4168 __ null_check(rcx);
4169 // do the call
4170 __ profile_call(rax);
4171 __ profile_arguments_type(rax, rbx, rbcp, false);
4172 __ jump_from_interpreted(rbx, rax);
4173 }
4174
4175 void TemplateTable::invokestatic(int byte_no) {
4176 transition(vtos, vtos);
4177 assert(byte_no == f1_byte, "use this argument");
4178 prepare_invoke(byte_no, rbx); // get f1 Method*
4179 // do the call
4180 __ profile_call(rax);
4181 __ profile_arguments_type(rax, rbx, rbcp, false);
4182 __ jump_from_interpreted(rbx, rax);
4183 }
4184
4185
4186 void TemplateTable::fast_invokevfinal(int byte_no) {
4187 transition(vtos, vtos);
4188 assert(byte_no == f2_byte, "use this argument");
4189 __ stop("fast_invokevfinal not used on x86");
4190 }
4191
4192
4193 void TemplateTable::invokeinterface(int byte_no) {
4194 transition(vtos, vtos);
4195 assert(byte_no == f1_byte, "use this argument");
4196 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 Method*
4197 rcx, rdx); // recv, flags
4198
4199 // rax: reference klass (from f1)
4200 // rbx: method (from f2)
4201 // rcx: receiver
4202 // rdx: flags
4203
4204 // Special case of invokeinterface called for virtual method of
4205 // java.lang.Object. See cpCacheOop.cpp for details.
4206 // This code isn't produced by javac, but could be produced by
4207 // another compliant java compiler.
4208 Label notMethod;
4209 __ movl(rlocals, rdx);
4210 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
4211
4212 __ jcc(Assembler::zero, notMethod);
4213
4214 invokevirtual_helper(rbx, rcx, rdx);
4215 __ bind(notMethod);
4216
4217 // Get receiver klass into rdx - also a null check
4218 __ restore_locals(); // restore r14
4219 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
4220 __ load_klass(rdx, rcx);
4221
4222 Label no_such_interface, no_such_method;
4223
4224 // Receiver subtype check against REFC.
4225 // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
4226 __ lookup_interface_method(// inputs: rec. class, interface, itable index
4227 rdx, rax, noreg,
4228 // outputs: scan temp. reg, scan temp. reg
4229 rbcp, rlocals,
4230 no_such_interface,
4231 /*return_method=*/false);
4232
4233 // profile this call
4234 __ restore_bcp(); // rbcp was destroyed by receiver type check
4235 __ profile_virtual_call(rdx, rbcp, rlocals);
4236
4237 // Get declaring interface class from method, and itable index
4238 __ movptr(rax, Address(rbx, Method::const_offset()));
4239 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
4240 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
4241 __ movl(rbx, Address(rbx, Method::itable_index_offset()));
4242 __ subl(rbx, Method::itable_index_max);
4243 __ negl(rbx);
4244
4245 __ lookup_interface_method(// inputs: rec. class, interface, itable index
4246 rdx, rax, rbx,
4247 // outputs: method, scan temp. reg
4248 rbx, rbcp,
4249 no_such_interface);
4250
4251 // rbx: Method* to call
4252 // rcx: receiver
4253 // Check for abstract method error
4254 // Note: This should be done more efficiently via a throw_abstract_method_error
4255 // interpreter entry point and a conditional jump to it in case of a null
4256 // method.
4257 __ testptr(rbx, rbx);
4258 __ jcc(Assembler::zero, no_such_method);
4259
4260 __ profile_called_method(rbx, rbcp, rdx);
4261 __ profile_arguments_type(rdx, rbx, rbcp, true);
4262
4263 // do the call
4264 // rcx: receiver
4265 // rbx,: Method*
4266 __ jump_from_interpreted(rbx, rdx);
4267 __ should_not_reach_here();
4268
4269 // exception handling code follows...
4270 // note: must restore interpreter registers to canonical
4271 // state for exception handling to work correctly!
4272
4273 __ bind(no_such_method);
4274 // throw exception
4275 __ pop(rbx); // pop return address (pushed by prepare_invoke)
4276 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
4277 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
4278 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
4279 // the call_VM checks for exception, so we should never return here.
4280 __ should_not_reach_here();
4281
4282 __ bind(no_such_interface);
4283 // throw exception
4284 __ pop(rbx); // pop return address (pushed by prepare_invoke)
4285 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
4286 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
4287 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4288 InterpreterRuntime::throw_IncompatibleClassChangeError));
4289 // the call_VM checks for exception, so we should never return here.
4290 __ should_not_reach_here();
4291 }
4292
4293 void TemplateTable::invokehandle(int byte_no) {
4294 transition(vtos, vtos);
4295 assert(byte_no == f1_byte, "use this argument");
4296 const Register rbx_method = rbx;
4297 const Register rax_mtype = rax;
4298 const Register rcx_recv = rcx;
4299 const Register rdx_flags = rdx;
4300
4301 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
4302 __ verify_method_ptr(rbx_method);
4303 __ verify_oop(rcx_recv);
4304 __ null_check(rcx_recv);
4305
4306 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
4307 // rbx: MH.invokeExact_MT method (from f2)
4308
4309 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
4310
4311 // FIXME: profile the LambdaForm also
4312 __ profile_final_call(rax);
4313 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
4314
4315 __ jump_from_interpreted(rbx_method, rdx);
4316 }
4317
4318 void TemplateTable::invokedynamic(int byte_no) {
4319 transition(vtos, vtos);
4320 assert(byte_no == f1_byte, "use this argument");
4321
4322 const Register rbx_method = rbx;
4323 const Register rax_callsite = rax;
4324
4325 prepare_invoke(byte_no, rbx_method, rax_callsite);
4326
4327 // rax: CallSite object (from cpool->resolved_references[f1])
4328 // rbx: MH.linkToCallSite method (from f2)
4329
4330 // Note: rax_callsite is already pushed by prepare_invoke
4331
4332 // %%% should make a type profile for any invokedynamic that takes a ref argument
4333 // profile this call
4334 __ profile_call(rbcp);
4335 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
4336
4337 __ verify_oop(rax_callsite);
4338
4339 __ jump_from_interpreted(rbx_method, rdx);
4340 }
4341
4342 //-----------------------------------------------------------------------------
4343 // Allocation
4344
4345 void TemplateTable::_new() {
4346 transition(vtos, atos);
4347 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
4348 Label slow_case;
4349 Label slow_case_no_pop;
4350 Label done;
4351 Label initialize_header;
4352 Label initialize_object; // including clearing the fields
4353
4354 __ get_cpool_and_tags(rcx, rax);
4355
4356 // Make sure the class we're about to instantiate has been resolved.
4357 // This is done before loading InstanceKlass to be consistent with the order
4358 // how Constant Pool is updated (see ConstantPool::klass_at_put)
4359 const int tags_offset = Array<u1>::base_offset_in_bytes();
4360 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
4361 __ jcc(Assembler::notEqual, slow_case_no_pop);
4362
4363 // get InstanceKlass
4364 __ load_resolved_klass_at_index(rcx, rdx, rcx);
4365 __ push(rcx); // save the contexts of klass for initializing the header
4366
4367 // make sure klass is initialized & doesn't have finalizer
4368 // make sure klass is fully initialized
4369 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4370 __ jcc(Assembler::notEqual, slow_case);
4371
4372 // get instance_size in InstanceKlass (scaled to a count of bytes)
4373 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
4374 // test to see if it has a finalizer or is malformed in some way
4375 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
4376 __ jcc(Assembler::notZero, slow_case);
4377
4378 // Allocate the instance:
4379 // If TLAB is enabled:
4380 // Try to allocate in the TLAB.
4381 // If fails, go to the slow path.
4382 // Else If inline contiguous allocations are enabled:
4383 // Try to allocate in eden.
4384 // If fails due to heap end, go to slow path.
4385 //
4386 // If TLAB is enabled OR inline contiguous is enabled:
4387 // Initialize the allocation.
4388 // Exit.
4389 //
4390 // Go to slow path.
4391
4392 const bool allow_shared_alloc =
4393 Universe::heap()->supports_inline_contig_alloc();
4394
4395 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4396 #ifndef _LP64
4397 if (UseTLAB || allow_shared_alloc) {
4398 __ get_thread(thread);
4399 }
4400 #endif // _LP64
4401
4402 if (UseTLAB) {
4403 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
4404 __ lea(rbx, Address(rax, rdx, Address::times_1));
4405 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
4406 __ jcc(Assembler::above, slow_case);
4407 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
4408 if (ZeroTLAB) {
4409 // the fields have been already cleared
4410 __ jmp(initialize_header);
4411 } else {
4412 // initialize both the header and fields
4413 __ jmp(initialize_object);
4414 }
4415 } else {
4416 // Allocation in the shared Eden, if allowed.
4417 //
4418 // rdx: instance size in bytes
4419 if (allow_shared_alloc) {
4420 ExternalAddress heap_top((address)Universe::heap()->top_addr());
4421 ExternalAddress heap_end((address)Universe::heap()->end_addr());
4422
4423 Label retry;
4424 __ bind(retry);
4425 __ movptr(rax, heap_top);
4426 __ lea(rbx, Address(rax, rdx, Address::times_1));
4427 __ cmpptr(rbx, heap_end);
4428 __ jcc(Assembler::above, slow_case);
4429
4430 // Compare rax, with the top addr, and if still equal, store the new
4431 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
4432 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
4433 //
4434 // rax,: object begin
4435 // rbx,: object end
4436 // rdx: instance size in bytes
4437 __ locked_cmpxchgptr(rbx, heap_top);
4438
4439 // if someone beat us on the allocation, try again, otherwise continue
4440 __ jcc(Assembler::notEqual, retry);
4441
4442 __ incr_allocated_bytes(thread, rdx, 0);
4443 }
4444 }
4445
4446 // If UseTLAB or allow_shared_alloc are true, the object is created above and
4447 // there is an initialize need. Otherwise, skip and go to the slow path.
4448 if (UseTLAB || allow_shared_alloc) {
4449 // The object is initialized before the header. If the object size is
4450 // zero, go directly to the header initialization.
4451 __ bind(initialize_object);
4452 __ decrement(rdx, sizeof(oopDesc));
4453 __ jcc(Assembler::zero, initialize_header);
4454
4455 // Initialize topmost object field, divide rdx by 8, check if odd and
4456 // test if zero.
4457 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
4458 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4459
4460 // rdx must have been multiple of 8
4461 #ifdef ASSERT
4462 // make sure rdx was multiple of 8
4463 Label L;
4464 // Ignore partial flag stall after shrl() since it is debug VM
4465 __ jccb(Assembler::carryClear, L);
4466 __ stop("object size is not multiple of 2 - adjust this code");
4467 __ bind(L);
4468 // rdx must be > 0, no extra check needed here
4469 #endif
4470
4471 // initialize remaining object fields: rdx was a multiple of 8
4472 { Label loop;
4473 __ bind(loop);
4474 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4475 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4476 __ decrement(rdx);
4477 __ jcc(Assembler::notZero, loop);
4478 }
4479
4480 // initialize object header only.
4481 __ bind(initialize_header);
4482 if (UseBiasedLocking) {
4483 __ pop(rcx); // get saved klass back in the register.
4484 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4485 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4486 } else {
4487 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4488 (intptr_t)markOopDesc::prototype()); // header
4489 __ pop(rcx); // get saved klass back in the register.
4490 }
4491 #ifdef _LP64
4492 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4493 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
4494 #endif
4495 __ store_klass(rax, rcx); // klass
4496
4497 {
4498 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4499 // Trigger dtrace event for fastpath
4500 __ push(atos);
4501 __ call_VM_leaf(
4502 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4503 __ pop(atos);
4504 }
4505
4506 __ jmp(done);
4507 }
4508
4509 // slow case
4510 __ bind(slow_case);
4511 __ pop(rcx); // restore stack pointer to what it was when we came in.
4512 __ bind(slow_case_no_pop);
4513
4514 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4515 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4516
4517 __ get_constant_pool(rarg1);
4518 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4519 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4520 __ verify_oop(rax);
4521
4522 // continue
4523 __ bind(done);
4524 }
4525
4526 void TemplateTable::defaultvalue() {
4527 transition(vtos, atos);
4528
4529 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4530 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4531
4532 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4533 __ get_constant_pool(rarg1);
4534
4535 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::defaultvalue),
4536 rarg1, rarg2);
4537 __ verify_oop(rax);
4538 }
4539
4540 void TemplateTable::newarray() {
4541 transition(itos, atos);
4542 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4543 __ load_unsigned_byte(rarg1, at_bcp(1));
4544 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4545 rarg1, rax);
4546 }
4547
4548 void TemplateTable::anewarray() {
4549 transition(itos, atos);
4550
4551 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4552 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4553
4554 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4555 __ get_constant_pool(rarg1);
4556 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4557 rarg1, rarg2, rax);
4558 }
4559
4560 void TemplateTable::arraylength() {
4561 transition(atos, itos);
4562 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4563 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4564 }
4565
4566 void TemplateTable::checkcast() {
4567 transition(atos, atos);
4568 Label done, is_null, ok_is_subtype, quicked, resolved;
4569 __ testptr(rax, rax); // object is in rax
4570 __ jcc(Assembler::zero, is_null);
4571
4572 // Get cpool & tags index
4573 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4574 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4575 // See if bytecode has already been quicked
4576 __ cmpb(Address(rdx, rbx,
4577 Address::times_1,
4578 Array<u1>::base_offset_in_bytes()),
4579 JVM_CONSTANT_Class);
4580 __ jcc(Assembler::equal, quicked);
4581 __ push(atos); // save receiver for result, and for GC
4582 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4583
4584 // vm_result_2 has metadata result
4585 #ifndef _LP64
4586 // borrow rdi from locals
4587 __ get_thread(rdi);
4588 __ get_vm_result_2(rax, rdi);
4589 __ restore_locals();
4590 #else
4591 __ get_vm_result_2(rax, r15_thread);
4592 #endif
4593
4594 __ pop_ptr(rdx); // restore receiver
4595 __ jmpb(resolved);
4596
4597 // Get superklass in rax and subklass in rbx
4598 __ bind(quicked);
4599 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4600 __ load_resolved_klass_at_index(rcx, rbx, rax);
4601
4602 __ bind(resolved);
4603 __ load_klass(rbx, rdx);
4604
4605 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4606 // Superklass in rax. Subklass in rbx.
4607 __ gen_subtype_check(rbx, ok_is_subtype);
4608
4609 // Come here on failure
4610 __ push_ptr(rdx);
4611 // object is at TOS
4612 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4613
4614 // Come here on success
4615 __ bind(ok_is_subtype);
4616 __ mov(rax, rdx); // Restore object in rdx
4617
4618 // Collect counts on whether this check-cast sees NULLs a lot or not.
4619 if (ProfileInterpreter) {
4620 __ jmp(done);
4621 __ bind(is_null);
4622 __ profile_null_seen(rcx);
4623 } else {
4624 __ bind(is_null); // same as 'done'
4625 }
4626 __ bind(done);
4627 }
4628
4629 void TemplateTable::instanceof() {
4630 transition(atos, itos);
4631 Label done, is_null, ok_is_subtype, quicked, resolved;
4632 __ testptr(rax, rax);
4633 __ jcc(Assembler::zero, is_null);
4634
4635 // Get cpool & tags index
4636 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4637 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4638 // See if bytecode has already been quicked
4639 __ cmpb(Address(rdx, rbx,
4640 Address::times_1,
4641 Array<u1>::base_offset_in_bytes()),
4642 JVM_CONSTANT_Class);
4643 __ jcc(Assembler::equal, quicked);
4644
4645 __ push(atos); // save receiver for result, and for GC
4646 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4647 // vm_result_2 has metadata result
4648
4649 #ifndef _LP64
4650 // borrow rdi from locals
4651 __ get_thread(rdi);
4652 __ get_vm_result_2(rax, rdi);
4653 __ restore_locals();
4654 #else
4655 __ get_vm_result_2(rax, r15_thread);
4656 #endif
4657
4658 __ pop_ptr(rdx); // restore receiver
4659 __ verify_oop(rdx);
4660 __ load_klass(rdx, rdx);
4661 __ jmpb(resolved);
4662
4663 // Get superklass in rax and subklass in rdx
4664 __ bind(quicked);
4665 __ load_klass(rdx, rax);
4666 __ load_resolved_klass_at_index(rcx, rbx, rax);
4667
4668 __ bind(resolved);
4669
4670 // Generate subtype check. Blows rcx, rdi
4671 // Superklass in rax. Subklass in rdx.
4672 __ gen_subtype_check(rdx, ok_is_subtype);
4673
4674 // Come here on failure
4675 __ xorl(rax, rax);
4676 __ jmpb(done);
4677 // Come here on success
4678 __ bind(ok_is_subtype);
4679 __ movl(rax, 1);
4680
4681 // Collect counts on whether this test sees NULLs a lot or not.
4682 if (ProfileInterpreter) {
4683 __ jmp(done);
4684 __ bind(is_null);
4685 __ profile_null_seen(rcx);
4686 } else {
4687 __ bind(is_null); // same as 'done'
4688 }
4689 __ bind(done);
4690 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4691 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4692 }
4693
4694 //----------------------------------------------------------------------------------------------------
4695 // Breakpoints
4696 void TemplateTable::_breakpoint() {
4697 // Note: We get here even if we are single stepping..
4698 // jbug insists on setting breakpoints at every bytecode
4699 // even if we are in single step mode.
4700
4701 transition(vtos, vtos);
4702
4703 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4704
4705 // get the unpatched byte code
4706 __ get_method(rarg);
4707 __ call_VM(noreg,
4708 CAST_FROM_FN_PTR(address,
4709 InterpreterRuntime::get_original_bytecode_at),
4710 rarg, rbcp);
4711 __ mov(rbx, rax); // why?
4712
4713 // post the breakpoint event
4714 __ get_method(rarg);
4715 __ call_VM(noreg,
4716 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4717 rarg, rbcp);
4718
4719 // complete the execution of original bytecode
4720 __ dispatch_only_normal(vtos);
4721 }
4722
4723 //-----------------------------------------------------------------------------
4724 // Exceptions
4725
4726 void TemplateTable::athrow() {
4727 transition(atos, vtos);
4728 __ null_check(rax);
4729 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4730 }
4731
4732 //-----------------------------------------------------------------------------
4733 // Synchronization
4734 //
4735 // Note: monitorenter & exit are symmetric routines; which is reflected
4736 // in the assembly code structure as well
4737 //
4738 // Stack layout:
4739 //
4740 // [expressions ] <--- rsp = expression stack top
4741 // ..
4742 // [expressions ]
4743 // [monitor entry] <--- monitor block top = expression stack bot
4744 // ..
4745 // [monitor entry]
4746 // [frame data ] <--- monitor block bot
4747 // ...
4748 // [saved rbp ] <--- rbp
4749 void TemplateTable::monitorenter() {
4750 transition(atos, vtos);
4751
4752 // check for NULL object
4753 __ null_check(rax);
4754
4755 const Address monitor_block_top(
4756 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4757 const Address monitor_block_bot(
4758 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4759 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4760
4761 Label allocated;
4762
4763 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4764 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4765 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4766
4767 if (EnableValhalla) {
4768 Label is_not_value;
4769
4770 __ test_oop_is_not_value(rax, rmon, is_not_value);
4771 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4772 InterpreterRuntime::throw_illegal_monitor_state_exception));
4773 __ should_not_reach_here();
4774
4775 __ bind(is_not_value);
4776
4777 }
4778 // initialize entry pointer
4779 __ xorl(rmon, rmon); // points to free slot or NULL
4780
4781 // find a free slot in the monitor block (result in rmon)
4782 {
4783 Label entry, loop, exit;
4784 __ movptr(rtop, monitor_block_top); // points to current entry,
4785 // starting with top-most entry
4786 __ lea(rbot, monitor_block_bot); // points to word before bottom
4787 // of monitor block
4788 __ jmpb(entry);
4789
4790 __ bind(loop);
4791 // check if current entry is used
4792 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4793 // if not used then remember entry in rmon
4794 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4795 // check if current entry is for same object
4796 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4797 // if same object then stop searching
4798 __ jccb(Assembler::equal, exit);
4799 // otherwise advance to next entry
4800 __ addptr(rtop, entry_size);
4801 __ bind(entry);
4802 // check if bottom reached
4803 __ cmpptr(rtop, rbot);
4804 // if not at bottom then check this entry
4805 __ jcc(Assembler::notEqual, loop);
4806 __ bind(exit);
4807 }
4808
4809 __ testptr(rmon, rmon); // check if a slot has been found
4810 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4811
4812 // allocate one if there's no free slot
4813 {
4814 Label entry, loop;
4815 // 1. compute new pointers // rsp: old expression stack top
4816 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4817 __ subptr(rsp, entry_size); // move expression stack top
4818 __ subptr(rmon, entry_size); // move expression stack bottom
4819 __ mov(rtop, rsp); // set start value for copy loop
4820 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4821 __ jmp(entry);
4822 // 2. move expression stack contents
4823 __ bind(loop);
4824 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4825 // word from old location
4826 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4827 __ addptr(rtop, wordSize); // advance to next word
4828 __ bind(entry);
4829 __ cmpptr(rtop, rmon); // check if bottom reached
4830 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4831 // copy next word
4832 }
4833
4834 // call run-time routine
4835 // rmon: points to monitor entry
4836 __ bind(allocated);
4837
4838 // Increment bcp to point to the next bytecode, so exception
4839 // handling for async. exceptions work correctly.
4840 // The object has already been poped from the stack, so the
4841 // expression stack looks correct.
4842 __ increment(rbcp);
4843
4844 // store object
4845 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4846 __ lock_object(rmon);
4847
4848 // check to make sure this monitor doesn't cause stack overflow after locking
4849 __ save_bcp(); // in case of exception
4850 __ generate_stack_overflow_check(0);
4851
4852 // The bcp has already been incremented. Just need to dispatch to
4853 // next instruction.
4854 __ dispatch_next(vtos);
4855 }
4856
4857 void TemplateTable::monitorexit() {
4858 transition(atos, vtos);
4859
4860 // check for NULL object
4861 __ null_check(rax);
4862
4863 const Address monitor_block_top(
4864 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4865 const Address monitor_block_bot(
4866 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4867 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4868
4869 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4870 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4871
4872 Label found;
4873
4874 // find matching slot
4875 {
4876 Label entry, loop;
4877 __ movptr(rtop, monitor_block_top); // points to current entry,
4878 // starting with top-most entry
4879 __ lea(rbot, monitor_block_bot); // points to word before bottom
4880 // of monitor block
4881 __ jmpb(entry);
4882
4883 __ bind(loop);
4884 // check if current entry is for same object
4885 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4886 // if same object then stop searching
4887 __ jcc(Assembler::equal, found);
4888 // otherwise advance to next entry
4889 __ addptr(rtop, entry_size);
4890 __ bind(entry);
4891 // check if bottom reached
4892 __ cmpptr(rtop, rbot);
4893 // if not at bottom then check this entry
4894 __ jcc(Assembler::notEqual, loop);
4895 }
4896
4897 // error handling. Unlocking was not block-structured
4898 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4899 InterpreterRuntime::throw_illegal_monitor_state_exception));
4900 __ should_not_reach_here();
4901
4902 // call run-time routine
4903 __ bind(found);
4904 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4905 __ unlock_object(rtop);
4906 __ pop_ptr(rax); // discard object
4907 }
4908
4909 // Wide instructions
4910 void TemplateTable::wide() {
4911 transition(vtos, vtos);
4912 __ load_unsigned_byte(rbx, at_bcp(1));
4913 ExternalAddress wtable((address)Interpreter::_wentry_point);
4914 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4915 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4916 }
4917
4918 // Multi arrays
4919 void TemplateTable::multianewarray() {
4920 transition(vtos, atos);
4921
4922 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4923 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4924 // last dim is on top of stack; we want address of first one:
4925 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4926 // the latter wordSize to point to the beginning of the array.
4927 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4928 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4929 __ load_unsigned_byte(rbx, at_bcp(3));
4930 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4931 }