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