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