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