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