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