1 /*
2 * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.inline.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #define __ _masm->
42
43 // Global Register Names
44 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
45 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
46
47 // Platform-dependent initialization
48 void TemplateTable::pd_initialize() {
49 // No x86 specific initialization
50 }
51
52 // Address Computation: local variables
53 static inline Address iaddress(int n) {
54 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
55 }
56
57 static inline Address laddress(int n) {
58 return iaddress(n + 1);
59 }
60
61 #ifndef _LP64
62 static inline Address haddress(int n) {
63 return iaddress(n + 0);
64 }
65 #endif
66
67 static inline Address faddress(int n) {
68 return iaddress(n);
69 }
70
71 static inline Address daddress(int n) {
72 return laddress(n);
73 }
74
75 static inline Address aaddress(int n) {
76 return iaddress(n);
77 }
78
79 static inline Address iaddress(Register r) {
80 return Address(rlocals, r, Address::times_ptr);
81 }
82
83 static inline Address laddress(Register r) {
84 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
85 }
86
87 #ifndef _LP64
88 static inline Address haddress(Register r) {
89 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
90 }
91 #endif
92
93 static inline Address faddress(Register r) {
94 return iaddress(r);
95 }
96
97 static inline Address daddress(Register r) {
98 return laddress(r);
99 }
100
101 static inline Address aaddress(Register r) {
102 return iaddress(r);
103 }
104
105
106 // expression stack
107 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
108 // data beyond the rsp which is potentially unsafe in an MT environment;
109 // an interrupt may overwrite that data.)
110 static inline Address at_rsp () {
111 return Address(rsp, 0);
112 }
113
114 // At top of Java expression stack which may be different than esp(). It
115 // isn't for category 1 objects.
116 static inline Address at_tos () {
117 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
118 }
119
120 static inline Address at_tos_p1() {
121 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
122 }
123
124 static inline Address at_tos_p2() {
125 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
126 }
127
128 // Condition conversion
129 static Assembler::Condition j_not(TemplateTable::Condition cc) {
130 switch (cc) {
131 case TemplateTable::equal : return Assembler::notEqual;
132 case TemplateTable::not_equal : return Assembler::equal;
133 case TemplateTable::less : return Assembler::greaterEqual;
134 case TemplateTable::less_equal : return Assembler::greater;
135 case TemplateTable::greater : return Assembler::lessEqual;
136 case TemplateTable::greater_equal: return Assembler::less;
137 }
138 ShouldNotReachHere();
139 return Assembler::zero;
140 }
141
142
143
144 // Miscelaneous helper routines
145 // Store an oop (or NULL) at the address described by obj.
146 // If val == noreg this means store a NULL
147
148
149 static void do_oop_store(InterpreterMacroAssembler* _masm,
150 Address obj,
151 Register val,
152 BarrierSet::Name barrier,
153 bool precise) {
154 assert(val == noreg || val == rax, "parameter is just for looks");
155 switch (barrier) {
156 #if INCLUDE_ALL_GCS
157 case BarrierSet::G1SATBCTLogging:
158 {
159 // flatten object address if needed
160 // We do it regardless of precise because we need the registers
161 if (obj.index() == noreg && obj.disp() == 0) {
162 if (obj.base() != rdx) {
163 __ movptr(rdx, obj.base());
164 }
165 } else {
166 __ lea(rdx, obj);
167 }
168
169 Register rtmp = LP64_ONLY(r8) NOT_LP64(rsi);
170 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
171
172 NOT_LP64(__ get_thread(rcx));
173 NOT_LP64(__ save_bcp());
174
175 __ g1_write_barrier_pre(rdx /* obj */,
176 rbx /* pre_val */,
177 rthread /* thread */,
178 rtmp /* tmp */,
179 val != noreg /* tosca_live */,
180 false /* expand_call */);
181 if (val == noreg) {
182 __ store_heap_oop_null(Address(rdx, 0));
183 } else {
184 // G1 barrier needs uncompressed oop for region cross check.
185 Register new_val = val;
186 if (UseCompressedOops) {
187 new_val = rbx;
188 __ movptr(new_val, val);
189 }
190 __ store_heap_oop(Address(rdx, 0), val);
191 __ g1_write_barrier_post(rdx /* store_adr */,
192 new_val /* new_val */,
193 rthread /* thread */,
194 rtmp /* tmp */,
195 rbx /* tmp2 */);
196 }
197 NOT_LP64( __ restore_bcp());
198 }
199 break;
200 #endif // INCLUDE_ALL_GCS
201 case BarrierSet::CardTableForRS:
202 case BarrierSet::CardTableExtension:
203 {
204 if (val == noreg) {
205 __ store_heap_oop_null(obj);
206 } else {
207 __ store_heap_oop(obj, val);
208 // flatten object address if needed
209 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
210 __ store_check(obj.base());
211 } else {
212 __ lea(rdx, obj);
213 __ store_check(rdx);
214 }
215 }
216 }
217 break;
218 case BarrierSet::ModRef:
219 if (val == noreg) {
220 __ store_heap_oop_null(obj);
221 } else {
222 __ store_heap_oop(obj, val);
223 }
224 break;
225 default :
226 ShouldNotReachHere();
227
228 }
229 }
230
231 Address TemplateTable::at_bcp(int offset) {
232 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
233 return Address(rbcp, offset);
234 }
235
236
237 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
238 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
239 int byte_no) {
240 if (!RewriteBytecodes) return;
241 Label L_patch_done;
242
243 switch (bc) {
244 case Bytecodes::_fast_qputfield:
245 __ jmp(L_patch_done); // don't patch yet
246 break;
247 case Bytecodes::_fast_aputfield:
248 case Bytecodes::_fast_bputfield:
249 case Bytecodes::_fast_zputfield:
250 case Bytecodes::_fast_cputfield:
251 case Bytecodes::_fast_dputfield:
252 case Bytecodes::_fast_fputfield:
253 case Bytecodes::_fast_iputfield:
254 case Bytecodes::_fast_lputfield:
255 case Bytecodes::_fast_sputfield:
256 {
257 // We skip bytecode quickening for putfield instructions when
258 // the put_code written to the constant pool cache is zero.
259 // This is required so that every execution of this instruction
260 // calls out to InterpreterRuntime::resolve_get_put to do
261 // additional, required work.
262 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
263 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
264 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
265 __ movl(bc_reg, bc);
266 __ cmpl(temp_reg, (int) 0);
267 __ jcc(Assembler::zero, L_patch_done); // don't patch
268 }
269 break;
270 default:
271 assert(byte_no == -1, "sanity");
272 // the pair bytecodes have already done the load.
273 if (load_bc_into_bc_reg) {
274 __ movl(bc_reg, bc);
275 }
276 }
277
278 if (JvmtiExport::can_post_breakpoint()) {
279 Label L_fast_patch;
280 // if a breakpoint is present we can't rewrite the stream directly
281 __ movzbl(temp_reg, at_bcp(0));
282 __ cmpl(temp_reg, Bytecodes::_breakpoint);
283 __ jcc(Assembler::notEqual, L_fast_patch);
284 __ get_method(temp_reg);
285 // Let breakpoint table handling rewrite to quicker bytecode
286 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
287 #ifndef ASSERT
288 __ jmpb(L_patch_done);
289 #else
290 __ jmp(L_patch_done);
291 #endif
292 __ bind(L_fast_patch);
293 }
294
295 #ifdef ASSERT
296 Label L_okay;
297 __ load_unsigned_byte(temp_reg, at_bcp(0));
298 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
299 __ jcc(Assembler::equal, L_okay);
300 __ cmpl(temp_reg, bc_reg);
301 __ jcc(Assembler::equal, L_okay);
302 __ stop("patching the wrong bytecode");
303 __ bind(L_okay);
304 #endif
305
306 // patch bytecode
307 __ movb(at_bcp(0), bc_reg);
308 __ bind(L_patch_done);
309 }
310 // Individual instructions
311
312
313 void TemplateTable::nop() {
314 transition(vtos, vtos);
315 // nothing to do
316 }
317
318 void TemplateTable::shouldnotreachhere() {
319 transition(vtos, vtos);
320 __ stop("shouldnotreachhere bytecode");
321 }
322
323 void TemplateTable::aconst_null() {
324 transition(vtos, atos);
325 __ xorl(rax, rax);
326 }
327
328 void TemplateTable::iconst(int value) {
329 transition(vtos, itos);
330 if (value == 0) {
331 __ xorl(rax, rax);
332 } else {
333 __ movl(rax, value);
334 }
335 }
336
337 void TemplateTable::lconst(int value) {
338 transition(vtos, ltos);
339 if (value == 0) {
340 __ xorl(rax, rax);
341 } else {
342 __ movl(rax, value);
343 }
344 #ifndef _LP64
345 assert(value >= 0, "check this code");
346 __ xorptr(rdx, rdx);
347 #endif
348 }
349
350
351
352 void TemplateTable::fconst(int value) {
353 transition(vtos, ftos);
354 if (UseSSE >= 1) {
355 static float one = 1.0f, two = 2.0f;
356 switch (value) {
357 case 0:
358 __ xorps(xmm0, xmm0);
359 break;
360 case 1:
361 __ movflt(xmm0, ExternalAddress((address) &one));
362 break;
363 case 2:
364 __ movflt(xmm0, ExternalAddress((address) &two));
365 break;
366 default:
367 ShouldNotReachHere();
368 break;
369 }
370 } else {
371 #ifdef _LP64
372 ShouldNotReachHere();
373 #else
374 if (value == 0) { __ fldz();
375 } else if (value == 1) { __ fld1();
376 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
377 } else { ShouldNotReachHere();
378 }
379 #endif // _LP64
380 }
381 }
382
383 void TemplateTable::dconst(int value) {
384 transition(vtos, dtos);
385 if (UseSSE >= 2) {
386 static double one = 1.0;
387 switch (value) {
388 case 0:
389 __ xorpd(xmm0, xmm0);
390 break;
391 case 1:
392 __ movdbl(xmm0, ExternalAddress((address) &one));
393 break;
394 default:
395 ShouldNotReachHere();
396 break;
397 }
398 } else {
399 #ifdef _LP64
400 ShouldNotReachHere();
401 #else
402 if (value == 0) { __ fldz();
403 } else if (value == 1) { __ fld1();
404 } else { ShouldNotReachHere();
405 }
406 #endif
407 }
408 }
409
410 void TemplateTable::bipush() {
411 transition(vtos, itos);
412 __ load_signed_byte(rax, at_bcp(1));
413 }
414
415 void TemplateTable::sipush() {
416 transition(vtos, itos);
417 __ load_unsigned_short(rax, at_bcp(1));
418 __ bswapl(rax);
419 __ sarl(rax, 16);
420 }
421
422 void TemplateTable::ldc(bool wide) {
423 transition(vtos, vtos);
424 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
425 Label call_ldc, notFloat, notClass, Done;
426
427 if (wide) {
428 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
429 } else {
430 __ load_unsigned_byte(rbx, at_bcp(1));
431 }
432
433 __ get_cpool_and_tags(rcx, rax);
434 const int base_offset = ConstantPool::header_size() * wordSize;
435 const int tags_offset = Array<u1>::base_offset_in_bytes();
436
437 // get type
438 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
439
440 // unresolved class - get the resolved class
441 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
442 __ jccb(Assembler::equal, call_ldc);
443
444 // unresolved class in error state - call into runtime to throw the error
445 // from the first resolution attempt
446 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
447 __ jccb(Assembler::equal, call_ldc);
448
449 // resolved class - need to call vm to get java mirror of the class
450 __ cmpl(rdx, JVM_CONSTANT_Class);
451 __ jcc(Assembler::notEqual, notClass);
452
453 __ bind(call_ldc);
454
455 __ movl(rarg, wide);
456 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
457
458 __ push(atos);
459 __ jmp(Done);
460
461 __ bind(notClass);
462 __ cmpl(rdx, JVM_CONSTANT_Float);
463 __ jccb(Assembler::notEqual, notFloat);
464
465 // ftos
466 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
467 __ push(ftos);
468 __ jmp(Done);
469
470 __ bind(notFloat);
471 #ifdef ASSERT
472 {
473 Label L;
474 __ cmpl(rdx, JVM_CONSTANT_Integer);
475 __ jcc(Assembler::equal, L);
476 // String and Object are rewritten to fast_aldc
477 __ stop("unexpected tag type in ldc");
478 __ bind(L);
479 }
480 #endif
481 // itos JVM_CONSTANT_Integer only
482 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
483 __ push(itos);
484 __ bind(Done);
485 }
486
487 // Fast path for caching oop constants.
488 void TemplateTable::fast_aldc(bool wide) {
489 transition(vtos, atos);
490
491 Register result = rax;
492 Register tmp = rdx;
493 int index_size = wide ? sizeof(u2) : sizeof(u1);
494
495 Label resolved;
496
497 // We are resolved if the resolved reference cache entry contains a
498 // non-null object (String, MethodType, etc.)
499 assert_different_registers(result, tmp);
500 __ get_cache_index_at_bcp(tmp, 1, index_size);
501 __ load_resolved_reference_at_index(result, tmp);
502 __ testl(result, result);
503 __ jcc(Assembler::notZero, resolved);
504
505 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
506
507 // first time invocation - must resolve first
508 __ movl(tmp, (int)bytecode());
509 __ call_VM(result, entry, tmp);
510
511 __ bind(resolved);
512
513 if (VerifyOops) {
514 __ verify_oop(result);
515 }
516 }
517
518 void TemplateTable::ldc2_w() {
519 transition(vtos, vtos);
520 Label Long, Done;
521 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
522
523 __ get_cpool_and_tags(rcx, rax);
524 const int base_offset = ConstantPool::header_size() * wordSize;
525 const int tags_offset = Array<u1>::base_offset_in_bytes();
526
527 // get type
528 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
529 JVM_CONSTANT_Double);
530 __ jccb(Assembler::notEqual, Long);
531
532 // dtos
533 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
534 __ push(dtos);
535
536 __ jmpb(Done);
537 __ bind(Long);
538
539 // ltos
540 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
541 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
542 __ push(ltos);
543
544 __ bind(Done);
545 }
546
547 void TemplateTable::locals_index(Register reg, int offset) {
548 __ load_unsigned_byte(reg, at_bcp(offset));
549 __ negptr(reg);
550 }
551
552 void TemplateTable::iload() {
553 iload_internal();
554 }
555
556 void TemplateTable::nofast_iload() {
557 iload_internal(may_not_rewrite);
558 }
559
560 void TemplateTable::iload_internal(RewriteControl rc) {
561 transition(vtos, itos);
562 if (RewriteFrequentPairs && rc == may_rewrite) {
563 Label rewrite, done;
564 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
565 LP64_ONLY(assert(rbx != bc, "register damaged"));
566
567 // get next byte
568 __ load_unsigned_byte(rbx,
569 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
570 // if _iload, wait to rewrite to iload2. We only want to rewrite the
571 // last two iloads in a pair. Comparing against fast_iload means that
572 // the next bytecode is neither an iload or a caload, and therefore
573 // an iload pair.
574 __ cmpl(rbx, Bytecodes::_iload);
575 __ jcc(Assembler::equal, done);
576
577 __ cmpl(rbx, Bytecodes::_fast_iload);
578 __ movl(bc, Bytecodes::_fast_iload2);
579
580 __ jccb(Assembler::equal, rewrite);
581
582 // if _caload, rewrite to fast_icaload
583 __ cmpl(rbx, Bytecodes::_caload);
584 __ movl(bc, Bytecodes::_fast_icaload);
585 __ jccb(Assembler::equal, rewrite);
586
587 // rewrite so iload doesn't check again.
588 __ movl(bc, Bytecodes::_fast_iload);
589
590 // rewrite
591 // bc: fast bytecode
592 __ bind(rewrite);
593 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
594 __ bind(done);
595 }
596
597 // Get the local value into tos
598 locals_index(rbx);
599 __ movl(rax, iaddress(rbx));
600 }
601
602 void TemplateTable::fast_iload2() {
603 transition(vtos, itos);
604 locals_index(rbx);
605 __ movl(rax, iaddress(rbx));
606 __ push(itos);
607 locals_index(rbx, 3);
608 __ movl(rax, iaddress(rbx));
609 }
610
611 void TemplateTable::fast_iload() {
612 transition(vtos, itos);
613 locals_index(rbx);
614 __ movl(rax, iaddress(rbx));
615 }
616
617 void TemplateTable::lload() {
618 transition(vtos, ltos);
619 locals_index(rbx);
620 __ movptr(rax, laddress(rbx));
621 NOT_LP64(__ movl(rdx, haddress(rbx)));
622 }
623
624 void TemplateTable::fload() {
625 transition(vtos, ftos);
626 locals_index(rbx);
627 __ load_float(faddress(rbx));
628 }
629
630 void TemplateTable::dload() {
631 transition(vtos, dtos);
632 locals_index(rbx);
633 __ load_double(daddress(rbx));
634 }
635
636 void TemplateTable::aload() {
637 transition(vtos, atos);
638 locals_index(rbx);
639 __ movptr(rax, aaddress(rbx));
640 }
641
642 void TemplateTable::vload() {
643 transition(vtos, qtos);
644 locals_index(rbx);
645 __ movptr(rax, aaddress(rbx));
646 }
647
648 void TemplateTable::locals_index_wide(Register reg) {
649 __ load_unsigned_short(reg, at_bcp(2));
650 __ bswapl(reg);
651 __ shrl(reg, 16);
652 __ negptr(reg);
653 }
654
655 void TemplateTable::wide_iload() {
656 transition(vtos, itos);
657 locals_index_wide(rbx);
658 __ movl(rax, iaddress(rbx));
659 }
660
661 void TemplateTable::wide_lload() {
662 transition(vtos, ltos);
663 locals_index_wide(rbx);
664 __ movptr(rax, laddress(rbx));
665 NOT_LP64(__ movl(rdx, haddress(rbx)));
666 }
667
668 void TemplateTable::wide_fload() {
669 transition(vtos, ftos);
670 locals_index_wide(rbx);
671 __ load_float(faddress(rbx));
672 }
673
674 void TemplateTable::wide_dload() {
675 transition(vtos, dtos);
676 locals_index_wide(rbx);
677 __ load_double(daddress(rbx));
678 }
679
680 void TemplateTable::wide_aload() {
681 transition(vtos, atos);
682 locals_index_wide(rbx);
683 __ movptr(rax, aaddress(rbx));
684 }
685
686 void TemplateTable::wide_vload() {
687 transition(vtos, qtos);
688 locals_index_wide(rbx);
689 __ movptr(rax, aaddress(rbx));
690 }
691
692 void TemplateTable::index_check(Register array, Register index) {
693 // Pop ptr into array
694 __ pop_ptr(array);
695 index_check_without_pop(array, index);
696 }
697
698 void TemplateTable::index_check_without_pop(Register array, Register index) {
699 // destroys rbx
700 // check array
701 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
702 // sign extend index for use by indexed load
703 __ movl2ptr(index, index);
704 // check index
705 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
706 if (index != rbx) {
707 // ??? convention: move aberrant index into rbx for exception message
708 assert(rbx != array, "different registers");
709 __ movl(rbx, index);
710 }
711 __ jump_cc(Assembler::aboveEqual,
712 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
713 }
714
715
716 void TemplateTable::iaload() {
717 transition(itos, itos);
718 // rax: index
719 // rdx: array
720 index_check(rdx, rax); // kills rbx
721 __ movl(rax, Address(rdx, rax,
722 Address::times_4,
723 arrayOopDesc::base_offset_in_bytes(T_INT)));
724 }
725
726 void TemplateTable::laload() {
727 transition(itos, ltos);
728 // rax: index
729 // rdx: array
730 index_check(rdx, rax); // kills rbx
731 NOT_LP64(__ mov(rbx, rax));
732 // rbx,: index
733 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
734 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
735 }
736
737
738
739 void TemplateTable::faload() {
740 transition(itos, ftos);
741 // rax: index
742 // rdx: array
743 index_check(rdx, rax); // kills rbx
744 __ load_float(Address(rdx, rax,
745 Address::times_4,
746 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
747 }
748
749 void TemplateTable::daload() {
750 transition(itos, dtos);
751 // rax: index
752 // rdx: array
753 index_check(rdx, rax); // kills rbx
754 __ load_double(Address(rdx, rax,
755 Address::times_8,
756 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
757 }
758
759 void TemplateTable::aaload() {
760 transition(itos, atos);
761 // rax: index
762 // rdx: array
763 index_check(rdx, rax); // kills rbx
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::vaload() {
770 transition(itos, qtos);
771
772 Register array = rcx;
773 Register index = rax;
774
775 index_check(array, index); // kills rbx, pops array
776
777 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_load) , array, index);
778 }
779
780 void TemplateTable::baload() {
781 transition(itos, itos);
782 // rax: index
783 // rdx: array
784 index_check(rdx, rax); // kills rbx
785 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
786 }
787
788 void TemplateTable::caload() {
789 transition(itos, itos);
790 // rax: index
791 // rdx: array
792 index_check(rdx, rax); // kills rbx
793 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
794 }
795
796 // iload followed by caload frequent pair
797 void TemplateTable::fast_icaload() {
798 transition(vtos, itos);
799 // load index out of locals
800 locals_index(rbx);
801 __ movl(rax, iaddress(rbx));
802
803 // rax: index
804 // rdx: array
805 index_check(rdx, rax); // kills rbx
806 __ load_unsigned_short(rax,
807 Address(rdx, rax,
808 Address::times_2,
809 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
810 }
811
812
813 void TemplateTable::saload() {
814 transition(itos, itos);
815 // rax: index
816 // rdx: array
817 index_check(rdx, rax); // kills rbx
818 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
819 }
820
821 void TemplateTable::iload(int n) {
822 transition(vtos, itos);
823 __ movl(rax, iaddress(n));
824 }
825
826 void TemplateTable::lload(int n) {
827 transition(vtos, ltos);
828 __ movptr(rax, laddress(n));
829 NOT_LP64(__ movptr(rdx, haddress(n)));
830 }
831
832 void TemplateTable::fload(int n) {
833 transition(vtos, ftos);
834 __ load_float(faddress(n));
835 }
836
837 void TemplateTable::dload(int n) {
838 transition(vtos, dtos);
839 __ load_double(daddress(n));
840 }
841
842 void TemplateTable::aload(int n) {
843 transition(vtos, atos);
844 __ movptr(rax, aaddress(n));
845 }
846
847 void TemplateTable::aload_0() {
848 aload_0_internal();
849 }
850
851 void TemplateTable::nofast_aload_0() {
852 aload_0_internal(may_not_rewrite);
853 }
854
855 void TemplateTable::aload_0_internal(RewriteControl rc) {
856 transition(vtos, atos);
857 // According to bytecode histograms, the pairs:
858 //
859 // _aload_0, _fast_igetfield
860 // _aload_0, _fast_agetfield
861 // _aload_0, _fast_fgetfield
862 //
863 // occur frequently. If RewriteFrequentPairs is set, the (slow)
864 // _aload_0 bytecode checks if the next bytecode is either
865 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
866 // rewrites the current bytecode into a pair bytecode; otherwise it
867 // rewrites the current bytecode into _fast_aload_0 that doesn't do
868 // the pair check anymore.
869 //
870 // Note: If the next bytecode is _getfield, the rewrite must be
871 // delayed, otherwise we may miss an opportunity for a pair.
872 //
873 // Also rewrite frequent pairs
874 // aload_0, aload_1
875 // aload_0, iload_1
876 // These bytecodes with a small amount of code are most profitable
877 // to rewrite
878 if (RewriteFrequentPairs && rc == may_rewrite) {
879 Label rewrite, done;
880
881 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
882 LP64_ONLY(assert(rbx != bc, "register damaged"));
883
884 // get next byte
885 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
886
887 // if _getfield then wait with rewrite
888 __ cmpl(rbx, Bytecodes::_getfield);
889 __ jcc(Assembler::equal, done);
890
891 // if _igetfield then rewrite to _fast_iaccess_0
892 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
893 __ cmpl(rbx, Bytecodes::_fast_igetfield);
894 __ movl(bc, Bytecodes::_fast_iaccess_0);
895 __ jccb(Assembler::equal, rewrite);
896
897 // if _agetfield then rewrite to _fast_aaccess_0
898 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
899 __ cmpl(rbx, Bytecodes::_fast_agetfield);
900 __ movl(bc, Bytecodes::_fast_aaccess_0);
901 __ jccb(Assembler::equal, rewrite);
902
903 // if _fgetfield then rewrite to _fast_faccess_0
904 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
905 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
906 __ movl(bc, Bytecodes::_fast_faccess_0);
907 __ jccb(Assembler::equal, rewrite);
908
909 // else rewrite to _fast_aload0
910 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
911 __ movl(bc, Bytecodes::_fast_aload_0);
912
913 // rewrite
914 // bc: fast bytecode
915 __ bind(rewrite);
916 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
917
918 __ bind(done);
919 }
920
921 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
922 aload(0);
923 }
924
925 void TemplateTable::istore() {
926 transition(itos, vtos);
927 locals_index(rbx);
928 __ movl(iaddress(rbx), rax);
929 }
930
931
932 void TemplateTable::lstore() {
933 transition(ltos, vtos);
934 locals_index(rbx);
935 __ movptr(laddress(rbx), rax);
936 NOT_LP64(__ movptr(haddress(rbx), rdx));
937 }
938
939 void TemplateTable::fstore() {
940 transition(ftos, vtos);
941 locals_index(rbx);
942 __ store_float(faddress(rbx));
943 }
944
945 void TemplateTable::dstore() {
946 transition(dtos, vtos);
947 locals_index(rbx);
948 __ store_double(daddress(rbx));
949 }
950
951 void TemplateTable::astore() {
952 transition(vtos, vtos);
953 __ pop_ptr(rax);
954 locals_index(rbx);
955 __ movptr(aaddress(rbx), rax);
956 }
957
958 void TemplateTable::vstore() {
959 transition(vtos, vtos);
960 __ pop_ptr(rax);
961 locals_index(rbx);
962 __ movptr(aaddress(rbx), rax);
963 }
964
965 void TemplateTable::wide_istore() {
966 transition(vtos, vtos);
967 __ pop_i();
968 locals_index_wide(rbx);
969 __ movl(iaddress(rbx), rax);
970 }
971
972 void TemplateTable::wide_lstore() {
973 transition(vtos, vtos);
974 NOT_LP64(__ pop_l(rax, rdx));
975 LP64_ONLY(__ pop_l());
976 locals_index_wide(rbx);
977 __ movptr(laddress(rbx), rax);
978 NOT_LP64(__ movl(haddress(rbx), rdx));
979 }
980
981 void TemplateTable::wide_fstore() {
982 #ifdef _LP64
983 transition(vtos, vtos);
984 __ pop_f(xmm0);
985 locals_index_wide(rbx);
986 __ movflt(faddress(rbx), xmm0);
987 #else
988 wide_istore();
989 #endif
990 }
991
992 void TemplateTable::wide_dstore() {
993 #ifdef _LP64
994 transition(vtos, vtos);
995 __ pop_d(xmm0);
996 locals_index_wide(rbx);
997 __ movdbl(daddress(rbx), xmm0);
998 #else
999 wide_lstore();
1000 #endif
1001 }
1002
1003 void TemplateTable::wide_astore() {
1004 transition(vtos, vtos);
1005 __ pop_ptr(rax);
1006 locals_index_wide(rbx);
1007 __ movptr(aaddress(rbx), rax);
1008 }
1009
1010 void TemplateTable::wide_vstore() {
1011 transition(vtos, vtos);
1012 __ pop_ptr(rax);
1013 locals_index_wide(rbx);
1014 __ movptr(aaddress(rbx), rax);
1015 }
1016
1017 void TemplateTable::iastore() {
1018 transition(itos, vtos);
1019 __ pop_i(rbx);
1020 // rax: value
1021 // rbx: index
1022 // rdx: array
1023 index_check(rdx, rbx); // prefer index in rbx
1024 __ movl(Address(rdx, rbx,
1025 Address::times_4,
1026 arrayOopDesc::base_offset_in_bytes(T_INT)),
1027 rax);
1028 }
1029
1030 void TemplateTable::lastore() {
1031 transition(ltos, vtos);
1032 __ pop_i(rbx);
1033 // rax,: low(value)
1034 // rcx: array
1035 // rdx: high(value)
1036 index_check(rcx, rbx); // prefer index in rbx,
1037 // rbx,: index
1038 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1039 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1040 }
1041
1042
1043 void TemplateTable::fastore() {
1044 transition(ftos, vtos);
1045 __ pop_i(rbx);
1046 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1047 // rbx: index
1048 // rdx: array
1049 index_check(rdx, rbx); // prefer index in rbx
1050 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1051 }
1052
1053 void TemplateTable::dastore() {
1054 transition(dtos, vtos);
1055 __ pop_i(rbx);
1056 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1057 // rbx: index
1058 // rdx: array
1059 index_check(rdx, rbx); // prefer index in rbx
1060 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1061 }
1062
1063 void TemplateTable::aastore() {
1064 Label is_null, ok_is_subtype, done;
1065 transition(vtos, vtos);
1066 // stack: ..., array, index, value
1067 __ movptr(rax, at_tos()); // value
1068 __ movl(rcx, at_tos_p1()); // index
1069 __ movptr(rdx, at_tos_p2()); // array
1070
1071 Address element_address(rdx, rcx,
1072 UseCompressedOops? Address::times_4 : Address::times_ptr,
1073 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1074
1075 index_check_without_pop(rdx, rcx); // kills rbx
1076 __ testptr(rax, rax);
1077 __ jcc(Assembler::zero, is_null);
1078
1079 // Move subklass into rbx
1080 __ load_klass(rbx, rax);
1081 // Move superklass into rax
1082 __ load_klass(rax, rdx);
1083 __ movptr(rax, Address(rax,
1084 ObjArrayKlass::element_klass_offset()));
1085 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1086 __ lea(rdx, element_address);
1087
1088 // Generate subtype check. Blows rcx, rdi
1089 // Superklass in rax. Subklass in rbx.
1090 __ gen_subtype_check(rbx, ok_is_subtype);
1091
1092 // Come here on failure
1093 // object is at TOS
1094 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1095
1096 // Come here on success
1097 __ bind(ok_is_subtype);
1098
1099 // Get the value we will store
1100 __ movptr(rax, at_tos());
1101 // Now store using the appropriate barrier
1102 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1103 __ jmp(done);
1104
1105 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1106 __ bind(is_null);
1107 __ profile_null_seen(rbx);
1108
1109 // Store a NULL
1110 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1111
1112 // Pop stack arguments
1113 __ bind(done);
1114 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1115 }
1116
1117 void TemplateTable::vastore() {
1118 transition(vtos, vtos);
1119
1120 Register value = rcx;
1121 Register index = rbx;
1122 Register array = rax;
1123
1124 // stack: ..., array, index, value
1125 __ pop_ptr(value);
1126 __ pop_i(index);
1127 __ pop_ptr(array);
1128
1129 index_check_without_pop(array, index);
1130
1131 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_store), array, index, value);
1132 }
1133
1134 void TemplateTable::bastore() {
1135 transition(itos, vtos);
1136 __ pop_i(rbx);
1137 // rax: value
1138 // rbx: index
1139 // rdx: array
1140 index_check(rdx, rbx); // prefer index in rbx
1141 // Need to check whether array is boolean or byte
1142 // since both types share the bastore bytecode.
1143 __ load_klass(rcx, rdx);
1144 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1145 int diffbit = Klass::layout_helper_boolean_diffbit();
1146 __ testl(rcx, diffbit);
1147 Label L_skip;
1148 __ jccb(Assembler::zero, L_skip);
1149 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1150 __ bind(L_skip);
1151 __ movb(Address(rdx, rbx,
1152 Address::times_1,
1153 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1154 rax);
1155 }
1156
1157 void TemplateTable::castore() {
1158 transition(itos, vtos);
1159 __ pop_i(rbx);
1160 // rax: value
1161 // rbx: index
1162 // rdx: array
1163 index_check(rdx, rbx); // prefer index in rbx
1164 __ movw(Address(rdx, rbx,
1165 Address::times_2,
1166 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1167 rax);
1168 }
1169
1170
1171 void TemplateTable::sastore() {
1172 castore();
1173 }
1174
1175 void TemplateTable::istore(int n) {
1176 transition(itos, vtos);
1177 __ movl(iaddress(n), rax);
1178 }
1179
1180 void TemplateTable::lstore(int n) {
1181 transition(ltos, vtos);
1182 __ movptr(laddress(n), rax);
1183 NOT_LP64(__ movptr(haddress(n), rdx));
1184 }
1185
1186 void TemplateTable::fstore(int n) {
1187 transition(ftos, vtos);
1188 __ store_float(faddress(n));
1189 }
1190
1191 void TemplateTable::dstore(int n) {
1192 transition(dtos, vtos);
1193 __ store_double(daddress(n));
1194 }
1195
1196
1197 void TemplateTable::astore(int n) {
1198 transition(vtos, vtos);
1199 __ pop_ptr(rax);
1200 __ movptr(aaddress(n), rax);
1201 }
1202
1203 void TemplateTable::pop() {
1204 transition(vtos, vtos);
1205 __ addptr(rsp, Interpreter::stackElementSize);
1206 }
1207
1208 void TemplateTable::pop2() {
1209 transition(vtos, vtos);
1210 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1211 }
1212
1213
1214 void TemplateTable::dup() {
1215 transition(vtos, vtos);
1216 __ load_ptr(0, rax);
1217 __ push_ptr(rax);
1218 // stack: ..., a, a
1219 }
1220
1221 void TemplateTable::dup_x1() {
1222 transition(vtos, vtos);
1223 // stack: ..., a, b
1224 __ load_ptr( 0, rax); // load b
1225 __ load_ptr( 1, rcx); // load a
1226 __ store_ptr(1, rax); // store b
1227 __ store_ptr(0, rcx); // store a
1228 __ push_ptr(rax); // push b
1229 // stack: ..., b, a, b
1230 }
1231
1232 void TemplateTable::dup_x2() {
1233 transition(vtos, vtos);
1234 // stack: ..., a, b, c
1235 __ load_ptr( 0, rax); // load c
1236 __ load_ptr( 2, rcx); // load a
1237 __ store_ptr(2, rax); // store c in a
1238 __ push_ptr(rax); // push c
1239 // stack: ..., c, b, c, c
1240 __ load_ptr( 2, rax); // load b
1241 __ store_ptr(2, rcx); // store a in b
1242 // stack: ..., c, a, c, c
1243 __ store_ptr(1, rax); // store b in c
1244 // stack: ..., c, a, b, c
1245 }
1246
1247 void TemplateTable::dup2() {
1248 transition(vtos, vtos);
1249 // stack: ..., a, b
1250 __ load_ptr(1, rax); // load a
1251 __ push_ptr(rax); // push a
1252 __ load_ptr(1, rax); // load b
1253 __ push_ptr(rax); // push b
1254 // stack: ..., a, b, a, b
1255 }
1256
1257
1258 void TemplateTable::dup2_x1() {
1259 transition(vtos, vtos);
1260 // stack: ..., a, b, c
1261 __ load_ptr( 0, rcx); // load c
1262 __ load_ptr( 1, rax); // load b
1263 __ push_ptr(rax); // push b
1264 __ push_ptr(rcx); // push c
1265 // stack: ..., a, b, c, b, c
1266 __ store_ptr(3, rcx); // store c in b
1267 // stack: ..., a, c, c, b, c
1268 __ load_ptr( 4, rcx); // load a
1269 __ store_ptr(2, rcx); // store a in 2nd c
1270 // stack: ..., a, c, a, b, c
1271 __ store_ptr(4, rax); // store b in a
1272 // stack: ..., b, c, a, b, c
1273 }
1274
1275 void TemplateTable::dup2_x2() {
1276 transition(vtos, vtos);
1277 // stack: ..., a, b, c, d
1278 __ load_ptr( 0, rcx); // load d
1279 __ load_ptr( 1, rax); // load c
1280 __ push_ptr(rax); // push c
1281 __ push_ptr(rcx); // push d
1282 // stack: ..., a, b, c, d, c, d
1283 __ load_ptr( 4, rax); // load b
1284 __ store_ptr(2, rax); // store b in d
1285 __ store_ptr(4, rcx); // store d in b
1286 // stack: ..., a, d, c, b, c, d
1287 __ load_ptr( 5, rcx); // load a
1288 __ load_ptr( 3, rax); // load c
1289 __ store_ptr(3, rcx); // store a in c
1290 __ store_ptr(5, rax); // store c in a
1291 // stack: ..., c, d, a, b, c, d
1292 }
1293
1294 void TemplateTable::swap() {
1295 transition(vtos, vtos);
1296 // stack: ..., a, b
1297 __ load_ptr( 1, rcx); // load a
1298 __ load_ptr( 0, rax); // load b
1299 __ store_ptr(0, rcx); // store a in b
1300 __ store_ptr(1, rax); // store b in a
1301 // stack: ..., b, a
1302 }
1303
1304 void TemplateTable::iop2(Operation op) {
1305 transition(itos, itos);
1306 switch (op) {
1307 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1308 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1309 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1310 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1311 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1312 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1313 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1314 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1315 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1316 default : ShouldNotReachHere();
1317 }
1318 }
1319
1320 void TemplateTable::lop2(Operation op) {
1321 transition(ltos, ltos);
1322 #ifdef _LP64
1323 switch (op) {
1324 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1325 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1326 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1327 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1328 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1329 default : ShouldNotReachHere();
1330 }
1331 #else
1332 __ pop_l(rbx, rcx);
1333 switch (op) {
1334 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1335 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1336 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1337 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1338 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1339 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1340 default : ShouldNotReachHere();
1341 }
1342 #endif
1343 }
1344
1345 void TemplateTable::idiv() {
1346 transition(itos, itos);
1347 __ movl(rcx, rax);
1348 __ pop_i(rax);
1349 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1350 // they are not equal, one could do a normal division (no correction
1351 // needed), which may speed up this implementation for the common case.
1352 // (see also JVM spec., p.243 & p.271)
1353 __ corrected_idivl(rcx);
1354 }
1355
1356 void TemplateTable::irem() {
1357 transition(itos, itos);
1358 __ movl(rcx, rax);
1359 __ pop_i(rax);
1360 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1361 // they are not equal, one could do a normal division (no correction
1362 // needed), which may speed up this implementation for the common case.
1363 // (see also JVM spec., p.243 & p.271)
1364 __ corrected_idivl(rcx);
1365 __ movl(rax, rdx);
1366 }
1367
1368 void TemplateTable::lmul() {
1369 transition(ltos, ltos);
1370 #ifdef _LP64
1371 __ pop_l(rdx);
1372 __ imulq(rax, rdx);
1373 #else
1374 __ pop_l(rbx, rcx);
1375 __ push(rcx); __ push(rbx);
1376 __ push(rdx); __ push(rax);
1377 __ lmul(2 * wordSize, 0);
1378 __ addptr(rsp, 4 * wordSize); // take off temporaries
1379 #endif
1380 }
1381
1382 void TemplateTable::ldiv() {
1383 transition(ltos, ltos);
1384 #ifdef _LP64
1385 __ mov(rcx, rax);
1386 __ pop_l(rax);
1387 // generate explicit div0 check
1388 __ testq(rcx, rcx);
1389 __ jump_cc(Assembler::zero,
1390 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1391 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1392 // they are not equal, one could do a normal division (no correction
1393 // needed), which may speed up this implementation for the common case.
1394 // (see also JVM spec., p.243 & p.271)
1395 __ corrected_idivq(rcx); // kills rbx
1396 #else
1397 __ pop_l(rbx, rcx);
1398 __ push(rcx); __ push(rbx);
1399 __ push(rdx); __ push(rax);
1400 // check if y = 0
1401 __ orl(rax, rdx);
1402 __ jump_cc(Assembler::zero,
1403 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1404 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1405 __ addptr(rsp, 4 * wordSize); // take off temporaries
1406 #endif
1407 }
1408
1409 void TemplateTable::lrem() {
1410 transition(ltos, ltos);
1411 #ifdef _LP64
1412 __ mov(rcx, rax);
1413 __ pop_l(rax);
1414 __ testq(rcx, rcx);
1415 __ jump_cc(Assembler::zero,
1416 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1417 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1418 // they are not equal, one could do a normal division (no correction
1419 // needed), which may speed up this implementation for the common case.
1420 // (see also JVM spec., p.243 & p.271)
1421 __ corrected_idivq(rcx); // kills rbx
1422 __ mov(rax, rdx);
1423 #else
1424 __ pop_l(rbx, rcx);
1425 __ push(rcx); __ push(rbx);
1426 __ push(rdx); __ push(rax);
1427 // check if y = 0
1428 __ orl(rax, rdx);
1429 __ jump_cc(Assembler::zero,
1430 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1431 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1432 __ addptr(rsp, 4 * wordSize);
1433 #endif
1434 }
1435
1436 void TemplateTable::lshl() {
1437 transition(itos, ltos);
1438 __ movl(rcx, rax); // get shift count
1439 #ifdef _LP64
1440 __ pop_l(rax); // get shift value
1441 __ shlq(rax);
1442 #else
1443 __ pop_l(rax, rdx); // get shift value
1444 __ lshl(rdx, rax);
1445 #endif
1446 }
1447
1448 void TemplateTable::lshr() {
1449 #ifdef _LP64
1450 transition(itos, ltos);
1451 __ movl(rcx, rax); // get shift count
1452 __ pop_l(rax); // get shift value
1453 __ sarq(rax);
1454 #else
1455 transition(itos, ltos);
1456 __ mov(rcx, rax); // get shift count
1457 __ pop_l(rax, rdx); // get shift value
1458 __ lshr(rdx, rax, true);
1459 #endif
1460 }
1461
1462 void TemplateTable::lushr() {
1463 transition(itos, ltos);
1464 #ifdef _LP64
1465 __ movl(rcx, rax); // get shift count
1466 __ pop_l(rax); // get shift value
1467 __ shrq(rax);
1468 #else
1469 __ mov(rcx, rax); // get shift count
1470 __ pop_l(rax, rdx); // get shift value
1471 __ lshr(rdx, rax);
1472 #endif
1473 }
1474
1475 void TemplateTable::fop2(Operation op) {
1476 transition(ftos, ftos);
1477
1478 if (UseSSE >= 1) {
1479 switch (op) {
1480 case add:
1481 __ addss(xmm0, at_rsp());
1482 __ addptr(rsp, Interpreter::stackElementSize);
1483 break;
1484 case sub:
1485 __ movflt(xmm1, xmm0);
1486 __ pop_f(xmm0);
1487 __ subss(xmm0, xmm1);
1488 break;
1489 case mul:
1490 __ mulss(xmm0, at_rsp());
1491 __ addptr(rsp, Interpreter::stackElementSize);
1492 break;
1493 case div:
1494 __ movflt(xmm1, xmm0);
1495 __ pop_f(xmm0);
1496 __ divss(xmm0, xmm1);
1497 break;
1498 case rem:
1499 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1500 // modulo operation. The frem method calls the function
1501 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1502 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1503 // (signalling or quiet) is returned.
1504 //
1505 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1506 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1507 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1508 // The fprem instruction used on x86_32 is functionally equivalent to
1509 // SharedRuntime::frem in that it returns a NaN.
1510 #ifdef _LP64
1511 __ movflt(xmm1, xmm0);
1512 __ pop_f(xmm0);
1513 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1514 #else
1515 __ push_f(xmm0);
1516 __ pop_f();
1517 __ fld_s(at_rsp());
1518 __ fremr(rax);
1519 __ f2ieee();
1520 __ pop(rax); // pop second operand off the stack
1521 __ push_f();
1522 __ pop_f(xmm0);
1523 #endif
1524 break;
1525 default:
1526 ShouldNotReachHere();
1527 break;
1528 }
1529 } else {
1530 #ifdef _LP64
1531 ShouldNotReachHere();
1532 #else
1533 switch (op) {
1534 case add: __ fadd_s (at_rsp()); break;
1535 case sub: __ fsubr_s(at_rsp()); break;
1536 case mul: __ fmul_s (at_rsp()); break;
1537 case div: __ fdivr_s(at_rsp()); break;
1538 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1539 default : ShouldNotReachHere();
1540 }
1541 __ f2ieee();
1542 __ pop(rax); // pop second operand off the stack
1543 #endif // _LP64
1544 }
1545 }
1546
1547 void TemplateTable::dop2(Operation op) {
1548 transition(dtos, dtos);
1549 if (UseSSE >= 2) {
1550 switch (op) {
1551 case add:
1552 __ addsd(xmm0, at_rsp());
1553 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1554 break;
1555 case sub:
1556 __ movdbl(xmm1, xmm0);
1557 __ pop_d(xmm0);
1558 __ subsd(xmm0, xmm1);
1559 break;
1560 case mul:
1561 __ mulsd(xmm0, at_rsp());
1562 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1563 break;
1564 case div:
1565 __ movdbl(xmm1, xmm0);
1566 __ pop_d(xmm0);
1567 __ divsd(xmm0, xmm1);
1568 break;
1569 case rem:
1570 // Similar to fop2(), the modulo operation is performed using the
1571 // SharedRuntime::drem method (on x86_64 platforms) or using the
1572 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1573 #ifdef _LP64
1574 __ movdbl(xmm1, xmm0);
1575 __ pop_d(xmm0);
1576 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1577 #else
1578 __ push_d(xmm0);
1579 __ pop_d();
1580 __ fld_d(at_rsp());
1581 __ fremr(rax);
1582 __ d2ieee();
1583 __ pop(rax);
1584 __ pop(rdx);
1585 __ push_d();
1586 __ pop_d(xmm0);
1587 #endif
1588 break;
1589 default:
1590 ShouldNotReachHere();
1591 break;
1592 }
1593 } else {
1594 #ifdef _LP64
1595 ShouldNotReachHere();
1596 #else
1597 switch (op) {
1598 case add: __ fadd_d (at_rsp()); break;
1599 case sub: __ fsubr_d(at_rsp()); break;
1600 case mul: {
1601 Label L_strict;
1602 Label L_join;
1603 const Address access_flags (rcx, Method::access_flags_offset());
1604 __ get_method(rcx);
1605 __ movl(rcx, access_flags);
1606 __ testl(rcx, JVM_ACC_STRICT);
1607 __ jccb(Assembler::notZero, L_strict);
1608 __ fmul_d (at_rsp());
1609 __ jmpb(L_join);
1610 __ bind(L_strict);
1611 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1612 __ fmulp();
1613 __ fmul_d (at_rsp());
1614 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1615 __ fmulp();
1616 __ bind(L_join);
1617 break;
1618 }
1619 case div: {
1620 Label L_strict;
1621 Label L_join;
1622 const Address access_flags (rcx, Method::access_flags_offset());
1623 __ get_method(rcx);
1624 __ movl(rcx, access_flags);
1625 __ testl(rcx, JVM_ACC_STRICT);
1626 __ jccb(Assembler::notZero, L_strict);
1627 __ fdivr_d(at_rsp());
1628 __ jmp(L_join);
1629 __ bind(L_strict);
1630 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1631 __ fmul_d (at_rsp());
1632 __ fdivrp();
1633 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1634 __ fmulp();
1635 __ bind(L_join);
1636 break;
1637 }
1638 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1639 default : ShouldNotReachHere();
1640 }
1641 __ d2ieee();
1642 // Pop double precision number from rsp.
1643 __ pop(rax);
1644 __ pop(rdx);
1645 #endif
1646 }
1647 }
1648
1649 void TemplateTable::ineg() {
1650 transition(itos, itos);
1651 __ negl(rax);
1652 }
1653
1654 void TemplateTable::lneg() {
1655 transition(ltos, ltos);
1656 LP64_ONLY(__ negq(rax));
1657 NOT_LP64(__ lneg(rdx, rax));
1658 }
1659
1660 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1661 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1662 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1663 // of 128-bits operands for SSE instructions.
1664 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1665 // Store the value to a 128-bits operand.
1666 operand[0] = lo;
1667 operand[1] = hi;
1668 return operand;
1669 }
1670
1671 // Buffer for 128-bits masks used by SSE instructions.
1672 static jlong float_signflip_pool[2*2];
1673 static jlong double_signflip_pool[2*2];
1674
1675 void TemplateTable::fneg() {
1676 transition(ftos, ftos);
1677 if (UseSSE >= 1) {
1678 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1679 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1680 } else {
1681 LP64_ONLY(ShouldNotReachHere());
1682 NOT_LP64(__ fchs());
1683 }
1684 }
1685
1686 void TemplateTable::dneg() {
1687 transition(dtos, dtos);
1688 if (UseSSE >= 2) {
1689 static jlong *double_signflip =
1690 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1691 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1692 } else {
1693 #ifdef _LP64
1694 ShouldNotReachHere();
1695 #else
1696 __ fchs();
1697 #endif
1698 }
1699 }
1700
1701 void TemplateTable::iinc() {
1702 transition(vtos, vtos);
1703 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1704 locals_index(rbx);
1705 __ addl(iaddress(rbx), rdx);
1706 }
1707
1708 void TemplateTable::wide_iinc() {
1709 transition(vtos, vtos);
1710 __ movl(rdx, at_bcp(4)); // get constant
1711 locals_index_wide(rbx);
1712 __ bswapl(rdx); // swap bytes & sign-extend constant
1713 __ sarl(rdx, 16);
1714 __ addl(iaddress(rbx), rdx);
1715 // Note: should probably use only one movl to get both
1716 // the index and the constant -> fix this
1717 }
1718
1719 void TemplateTable::convert() {
1720 #ifdef _LP64
1721 // Checking
1722 #ifdef ASSERT
1723 {
1724 TosState tos_in = ilgl;
1725 TosState tos_out = ilgl;
1726 switch (bytecode()) {
1727 case Bytecodes::_i2l: // fall through
1728 case Bytecodes::_i2f: // fall through
1729 case Bytecodes::_i2d: // fall through
1730 case Bytecodes::_i2b: // fall through
1731 case Bytecodes::_i2c: // fall through
1732 case Bytecodes::_i2s: tos_in = itos; break;
1733 case Bytecodes::_l2i: // fall through
1734 case Bytecodes::_l2f: // fall through
1735 case Bytecodes::_l2d: tos_in = ltos; break;
1736 case Bytecodes::_f2i: // fall through
1737 case Bytecodes::_f2l: // fall through
1738 case Bytecodes::_f2d: tos_in = ftos; break;
1739 case Bytecodes::_d2i: // fall through
1740 case Bytecodes::_d2l: // fall through
1741 case Bytecodes::_d2f: tos_in = dtos; break;
1742 default : ShouldNotReachHere();
1743 }
1744 switch (bytecode()) {
1745 case Bytecodes::_l2i: // fall through
1746 case Bytecodes::_f2i: // fall through
1747 case Bytecodes::_d2i: // fall through
1748 case Bytecodes::_i2b: // fall through
1749 case Bytecodes::_i2c: // fall through
1750 case Bytecodes::_i2s: tos_out = itos; break;
1751 case Bytecodes::_i2l: // fall through
1752 case Bytecodes::_f2l: // fall through
1753 case Bytecodes::_d2l: tos_out = ltos; break;
1754 case Bytecodes::_i2f: // fall through
1755 case Bytecodes::_l2f: // fall through
1756 case Bytecodes::_d2f: tos_out = ftos; break;
1757 case Bytecodes::_i2d: // fall through
1758 case Bytecodes::_l2d: // fall through
1759 case Bytecodes::_f2d: tos_out = dtos; break;
1760 default : ShouldNotReachHere();
1761 }
1762 transition(tos_in, tos_out);
1763 }
1764 #endif // ASSERT
1765
1766 static const int64_t is_nan = 0x8000000000000000L;
1767
1768 // Conversion
1769 switch (bytecode()) {
1770 case Bytecodes::_i2l:
1771 __ movslq(rax, rax);
1772 break;
1773 case Bytecodes::_i2f:
1774 __ cvtsi2ssl(xmm0, rax);
1775 break;
1776 case Bytecodes::_i2d:
1777 __ cvtsi2sdl(xmm0, rax);
1778 break;
1779 case Bytecodes::_i2b:
1780 __ movsbl(rax, rax);
1781 break;
1782 case Bytecodes::_i2c:
1783 __ movzwl(rax, rax);
1784 break;
1785 case Bytecodes::_i2s:
1786 __ movswl(rax, rax);
1787 break;
1788 case Bytecodes::_l2i:
1789 __ movl(rax, rax);
1790 break;
1791 case Bytecodes::_l2f:
1792 __ cvtsi2ssq(xmm0, rax);
1793 break;
1794 case Bytecodes::_l2d:
1795 __ cvtsi2sdq(xmm0, rax);
1796 break;
1797 case Bytecodes::_f2i:
1798 {
1799 Label L;
1800 __ cvttss2sil(rax, xmm0);
1801 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1802 __ jcc(Assembler::notEqual, L);
1803 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1804 __ bind(L);
1805 }
1806 break;
1807 case Bytecodes::_f2l:
1808 {
1809 Label L;
1810 __ cvttss2siq(rax, xmm0);
1811 // NaN or overflow/underflow?
1812 __ cmp64(rax, ExternalAddress((address) &is_nan));
1813 __ jcc(Assembler::notEqual, L);
1814 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1815 __ bind(L);
1816 }
1817 break;
1818 case Bytecodes::_f2d:
1819 __ cvtss2sd(xmm0, xmm0);
1820 break;
1821 case Bytecodes::_d2i:
1822 {
1823 Label L;
1824 __ cvttsd2sil(rax, xmm0);
1825 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1826 __ jcc(Assembler::notEqual, L);
1827 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1828 __ bind(L);
1829 }
1830 break;
1831 case Bytecodes::_d2l:
1832 {
1833 Label L;
1834 __ cvttsd2siq(rax, xmm0);
1835 // NaN or overflow/underflow?
1836 __ cmp64(rax, ExternalAddress((address) &is_nan));
1837 __ jcc(Assembler::notEqual, L);
1838 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1839 __ bind(L);
1840 }
1841 break;
1842 case Bytecodes::_d2f:
1843 __ cvtsd2ss(xmm0, xmm0);
1844 break;
1845 default:
1846 ShouldNotReachHere();
1847 }
1848 #else
1849 // Checking
1850 #ifdef ASSERT
1851 { TosState tos_in = ilgl;
1852 TosState tos_out = ilgl;
1853 switch (bytecode()) {
1854 case Bytecodes::_i2l: // fall through
1855 case Bytecodes::_i2f: // fall through
1856 case Bytecodes::_i2d: // fall through
1857 case Bytecodes::_i2b: // fall through
1858 case Bytecodes::_i2c: // fall through
1859 case Bytecodes::_i2s: tos_in = itos; break;
1860 case Bytecodes::_l2i: // fall through
1861 case Bytecodes::_l2f: // fall through
1862 case Bytecodes::_l2d: tos_in = ltos; break;
1863 case Bytecodes::_f2i: // fall through
1864 case Bytecodes::_f2l: // fall through
1865 case Bytecodes::_f2d: tos_in = ftos; break;
1866 case Bytecodes::_d2i: // fall through
1867 case Bytecodes::_d2l: // fall through
1868 case Bytecodes::_d2f: tos_in = dtos; break;
1869 default : ShouldNotReachHere();
1870 }
1871 switch (bytecode()) {
1872 case Bytecodes::_l2i: // fall through
1873 case Bytecodes::_f2i: // fall through
1874 case Bytecodes::_d2i: // fall through
1875 case Bytecodes::_i2b: // fall through
1876 case Bytecodes::_i2c: // fall through
1877 case Bytecodes::_i2s: tos_out = itos; break;
1878 case Bytecodes::_i2l: // fall through
1879 case Bytecodes::_f2l: // fall through
1880 case Bytecodes::_d2l: tos_out = ltos; break;
1881 case Bytecodes::_i2f: // fall through
1882 case Bytecodes::_l2f: // fall through
1883 case Bytecodes::_d2f: tos_out = ftos; break;
1884 case Bytecodes::_i2d: // fall through
1885 case Bytecodes::_l2d: // fall through
1886 case Bytecodes::_f2d: tos_out = dtos; break;
1887 default : ShouldNotReachHere();
1888 }
1889 transition(tos_in, tos_out);
1890 }
1891 #endif // ASSERT
1892
1893 // Conversion
1894 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1895 switch (bytecode()) {
1896 case Bytecodes::_i2l:
1897 __ extend_sign(rdx, rax);
1898 break;
1899 case Bytecodes::_i2f:
1900 if (UseSSE >= 1) {
1901 __ cvtsi2ssl(xmm0, rax);
1902 } else {
1903 __ push(rax); // store int on tos
1904 __ fild_s(at_rsp()); // load int to ST0
1905 __ f2ieee(); // truncate to float size
1906 __ pop(rcx); // adjust rsp
1907 }
1908 break;
1909 case Bytecodes::_i2d:
1910 if (UseSSE >= 2) {
1911 __ cvtsi2sdl(xmm0, rax);
1912 } else {
1913 __ push(rax); // add one slot for d2ieee()
1914 __ push(rax); // store int on tos
1915 __ fild_s(at_rsp()); // load int to ST0
1916 __ d2ieee(); // truncate to double size
1917 __ pop(rcx); // adjust rsp
1918 __ pop(rcx);
1919 }
1920 break;
1921 case Bytecodes::_i2b:
1922 __ shll(rax, 24); // truncate upper 24 bits
1923 __ sarl(rax, 24); // and sign-extend byte
1924 LP64_ONLY(__ movsbl(rax, rax));
1925 break;
1926 case Bytecodes::_i2c:
1927 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1928 LP64_ONLY(__ movzwl(rax, rax));
1929 break;
1930 case Bytecodes::_i2s:
1931 __ shll(rax, 16); // truncate upper 16 bits
1932 __ sarl(rax, 16); // and sign-extend short
1933 LP64_ONLY(__ movswl(rax, rax));
1934 break;
1935 case Bytecodes::_l2i:
1936 /* nothing to do */
1937 break;
1938 case Bytecodes::_l2f:
1939 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1940 // 64-bit long values to floats. On 32-bit platforms it is not possible
1941 // to use that instruction with 64-bit operands, therefore the FPU is
1942 // used to perform the conversion.
1943 __ push(rdx); // store long on tos
1944 __ push(rax);
1945 __ fild_d(at_rsp()); // load long to ST0
1946 __ f2ieee(); // truncate to float size
1947 __ pop(rcx); // adjust rsp
1948 __ pop(rcx);
1949 if (UseSSE >= 1) {
1950 __ push_f();
1951 __ pop_f(xmm0);
1952 }
1953 break;
1954 case Bytecodes::_l2d:
1955 // On 32-bit platforms the FPU is used for conversion because on
1956 // 32-bit platforms it is not not possible to use the cvtsi2sdq
1957 // instruction with 64-bit operands.
1958 __ push(rdx); // store long on tos
1959 __ push(rax);
1960 __ fild_d(at_rsp()); // load long to ST0
1961 __ d2ieee(); // truncate to double size
1962 __ pop(rcx); // adjust rsp
1963 __ pop(rcx);
1964 if (UseSSE >= 2) {
1965 __ push_d();
1966 __ pop_d(xmm0);
1967 }
1968 break;
1969 case Bytecodes::_f2i:
1970 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1971 // as it returns 0 for any NaN.
1972 if (UseSSE >= 1) {
1973 __ push_f(xmm0);
1974 } else {
1975 __ push(rcx); // reserve space for argument
1976 __ fstp_s(at_rsp()); // pass float argument on stack
1977 }
1978 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1979 break;
1980 case Bytecodes::_f2l:
1981 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
1982 // as it returns 0 for any NaN.
1983 if (UseSSE >= 1) {
1984 __ push_f(xmm0);
1985 } else {
1986 __ push(rcx); // reserve space for argument
1987 __ fstp_s(at_rsp()); // pass float argument on stack
1988 }
1989 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1990 break;
1991 case Bytecodes::_f2d:
1992 if (UseSSE < 1) {
1993 /* nothing to do */
1994 } else if (UseSSE == 1) {
1995 __ push_f(xmm0);
1996 __ pop_f();
1997 } else { // UseSSE >= 2
1998 __ cvtss2sd(xmm0, xmm0);
1999 }
2000 break;
2001 case Bytecodes::_d2i:
2002 if (UseSSE >= 2) {
2003 __ push_d(xmm0);
2004 } else {
2005 __ push(rcx); // reserve space for argument
2006 __ push(rcx);
2007 __ fstp_d(at_rsp()); // pass double argument on stack
2008 }
2009 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2010 break;
2011 case Bytecodes::_d2l:
2012 if (UseSSE >= 2) {
2013 __ push_d(xmm0);
2014 } else {
2015 __ push(rcx); // reserve space for argument
2016 __ push(rcx);
2017 __ fstp_d(at_rsp()); // pass double argument on stack
2018 }
2019 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2020 break;
2021 case Bytecodes::_d2f:
2022 if (UseSSE <= 1) {
2023 __ push(rcx); // reserve space for f2ieee()
2024 __ f2ieee(); // truncate to float size
2025 __ pop(rcx); // adjust rsp
2026 if (UseSSE == 1) {
2027 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2028 // the conversion is performed using the FPU in this case.
2029 __ push_f();
2030 __ pop_f(xmm0);
2031 }
2032 } else { // UseSSE >= 2
2033 __ cvtsd2ss(xmm0, xmm0);
2034 }
2035 break;
2036 default :
2037 ShouldNotReachHere();
2038 }
2039 #endif
2040 }
2041
2042 void TemplateTable::lcmp() {
2043 transition(ltos, itos);
2044 #ifdef _LP64
2045 Label done;
2046 __ pop_l(rdx);
2047 __ cmpq(rdx, rax);
2048 __ movl(rax, -1);
2049 __ jccb(Assembler::less, done);
2050 __ setb(Assembler::notEqual, rax);
2051 __ movzbl(rax, rax);
2052 __ bind(done);
2053 #else
2054
2055 // y = rdx:rax
2056 __ pop_l(rbx, rcx); // get x = rcx:rbx
2057 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2058 __ mov(rax, rcx);
2059 #endif
2060 }
2061
2062 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2063 if ((is_float && UseSSE >= 1) ||
2064 (!is_float && UseSSE >= 2)) {
2065 Label done;
2066 if (is_float) {
2067 // XXX get rid of pop here, use ... reg, mem32
2068 __ pop_f(xmm1);
2069 __ ucomiss(xmm1, xmm0);
2070 } else {
2071 // XXX get rid of pop here, use ... reg, mem64
2072 __ pop_d(xmm1);
2073 __ ucomisd(xmm1, xmm0);
2074 }
2075 if (unordered_result < 0) {
2076 __ movl(rax, -1);
2077 __ jccb(Assembler::parity, done);
2078 __ jccb(Assembler::below, done);
2079 __ setb(Assembler::notEqual, rdx);
2080 __ movzbl(rax, rdx);
2081 } else {
2082 __ movl(rax, 1);
2083 __ jccb(Assembler::parity, done);
2084 __ jccb(Assembler::above, done);
2085 __ movl(rax, 0);
2086 __ jccb(Assembler::equal, done);
2087 __ decrementl(rax);
2088 }
2089 __ bind(done);
2090 } else {
2091 #ifdef _LP64
2092 ShouldNotReachHere();
2093 #else
2094 if (is_float) {
2095 __ fld_s(at_rsp());
2096 } else {
2097 __ fld_d(at_rsp());
2098 __ pop(rdx);
2099 }
2100 __ pop(rcx);
2101 __ fcmp2int(rax, unordered_result < 0);
2102 #endif // _LP64
2103 }
2104 }
2105
2106 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2107 __ get_method(rcx); // rcx holds method
2108 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2109 // holds bumped taken count
2110
2111 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2112 InvocationCounter::counter_offset();
2113 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2114 InvocationCounter::counter_offset();
2115
2116 // Load up edx with the branch displacement
2117 if (is_wide) {
2118 __ movl(rdx, at_bcp(1));
2119 } else {
2120 __ load_signed_short(rdx, at_bcp(1));
2121 }
2122 __ bswapl(rdx);
2123
2124 if (!is_wide) {
2125 __ sarl(rdx, 16);
2126 }
2127 LP64_ONLY(__ movl2ptr(rdx, rdx));
2128
2129 // Handle all the JSR stuff here, then exit.
2130 // It's much shorter and cleaner than intermingling with the non-JSR
2131 // normal-branch stuff occurring below.
2132 if (is_jsr) {
2133 // Pre-load the next target bytecode into rbx
2134 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2135
2136 // compute return address as bci in rax
2137 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2138 in_bytes(ConstMethod::codes_offset())));
2139 __ subptr(rax, Address(rcx, Method::const_offset()));
2140 // Adjust the bcp in r13 by the displacement in rdx
2141 __ addptr(rbcp, rdx);
2142 // jsr returns atos that is not an oop
2143 __ push_i(rax);
2144 __ dispatch_only(vtos);
2145 return;
2146 }
2147
2148 // Normal (non-jsr) branch handling
2149
2150 // Adjust the bcp in r13 by the displacement in rdx
2151 __ addptr(rbcp, rdx);
2152
2153 assert(UseLoopCounter || !UseOnStackReplacement,
2154 "on-stack-replacement requires loop counters");
2155 Label backedge_counter_overflow;
2156 Label profile_method;
2157 Label dispatch;
2158 if (UseLoopCounter) {
2159 // increment backedge counter for backward branches
2160 // rax: MDO
2161 // rbx: MDO bumped taken-count
2162 // rcx: method
2163 // rdx: target offset
2164 // r13: target bcp
2165 // r14: locals pointer
2166 __ testl(rdx, rdx); // check if forward or backward branch
2167 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2168
2169 // check if MethodCounters exists
2170 Label has_counters;
2171 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2172 __ testptr(rax, rax);
2173 __ jcc(Assembler::notZero, has_counters);
2174 __ push(rdx);
2175 __ push(rcx);
2176 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2177 rcx);
2178 __ pop(rcx);
2179 __ pop(rdx);
2180 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2181 __ testptr(rax, rax);
2182 __ jcc(Assembler::zero, dispatch);
2183 __ bind(has_counters);
2184
2185 if (TieredCompilation) {
2186 Label no_mdo;
2187 int increment = InvocationCounter::count_increment;
2188 if (ProfileInterpreter) {
2189 // Are we profiling?
2190 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2191 __ testptr(rbx, rbx);
2192 __ jccb(Assembler::zero, no_mdo);
2193 // Increment the MDO backedge counter
2194 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2195 in_bytes(InvocationCounter::counter_offset()));
2196 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2197 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2198 rax, false, Assembler::zero, &backedge_counter_overflow);
2199 __ jmp(dispatch);
2200 }
2201 __ bind(no_mdo);
2202 // Increment backedge counter in MethodCounters*
2203 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2204 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2205 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2206 rax, false, Assembler::zero, &backedge_counter_overflow);
2207 } else { // not TieredCompilation
2208 // increment counter
2209 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2210 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2211 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2212 __ movl(Address(rcx, be_offset), rax); // store counter
2213
2214 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2215
2216 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2217 __ addl(rax, Address(rcx, be_offset)); // add both counters
2218
2219 if (ProfileInterpreter) {
2220 // Test to see if we should create a method data oop
2221 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2222 __ jcc(Assembler::less, dispatch);
2223
2224 // if no method data exists, go to profile method
2225 __ test_method_data_pointer(rax, profile_method);
2226
2227 if (UseOnStackReplacement) {
2228 // check for overflow against rbx which is the MDO taken count
2229 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2230 __ jcc(Assembler::below, dispatch);
2231
2232 // When ProfileInterpreter is on, the backedge_count comes
2233 // from the MethodData*, which value does not get reset on
2234 // the call to frequency_counter_overflow(). To avoid
2235 // excessive calls to the overflow routine while the method is
2236 // being compiled, add a second test to make sure the overflow
2237 // function is called only once every overflow_frequency.
2238 const int overflow_frequency = 1024;
2239 __ andl(rbx, overflow_frequency - 1);
2240 __ jcc(Assembler::zero, backedge_counter_overflow);
2241
2242 }
2243 } else {
2244 if (UseOnStackReplacement) {
2245 // check for overflow against rax, which is the sum of the
2246 // counters
2247 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2248 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2249
2250 }
2251 }
2252 }
2253 __ bind(dispatch);
2254 }
2255
2256 // Pre-load the next target bytecode into rbx
2257 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2258
2259 // continue with the bytecode @ target
2260 // rax: return bci for jsr's, unused otherwise
2261 // rbx: target bytecode
2262 // r13: target bcp
2263 __ dispatch_only(vtos);
2264
2265 if (UseLoopCounter) {
2266 if (ProfileInterpreter) {
2267 // Out-of-line code to allocate method data oop.
2268 __ bind(profile_method);
2269 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2270 __ set_method_data_pointer_for_bcp();
2271 __ jmp(dispatch);
2272 }
2273
2274 if (UseOnStackReplacement) {
2275 // invocation counter overflow
2276 __ bind(backedge_counter_overflow);
2277 __ negptr(rdx);
2278 __ addptr(rdx, rbcp); // branch bcp
2279 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2280 __ call_VM(noreg,
2281 CAST_FROM_FN_PTR(address,
2282 InterpreterRuntime::frequency_counter_overflow),
2283 rdx);
2284
2285 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2286 // rdx: scratch
2287 // r14: locals pointer
2288 // r13: bcp
2289 __ testptr(rax, rax); // test result
2290 __ jcc(Assembler::zero, dispatch); // no osr if null
2291 // nmethod may have been invalidated (VM may block upon call_VM return)
2292 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2293 __ jcc(Assembler::notEqual, dispatch);
2294
2295 // We have the address of an on stack replacement routine in rax.
2296 // In preparation of invoking it, first we must migrate the locals
2297 // and monitors from off the interpreter frame on the stack.
2298 // Ensure to save the osr nmethod over the migration call,
2299 // it will be preserved in rbx.
2300 __ mov(rbx, rax);
2301
2302 NOT_LP64(__ get_thread(rcx));
2303
2304 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2305
2306 // rax is OSR buffer, move it to expected parameter location
2307 LP64_ONLY(__ mov(j_rarg0, rax));
2308 NOT_LP64(__ mov(rcx, rax));
2309 // We use j_rarg definitions here so that registers don't conflict as parameter
2310 // registers change across platforms as we are in the midst of a calling
2311 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2312
2313 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2314 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2315
2316 // pop the interpreter frame
2317 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2318 __ leave(); // remove frame anchor
2319 __ pop(retaddr); // get return address
2320 __ mov(rsp, sender_sp); // set sp to sender sp
2321 // Ensure compiled code always sees stack at proper alignment
2322 __ andptr(rsp, -(StackAlignmentInBytes));
2323
2324 // unlike x86 we need no specialized return from compiled code
2325 // to the interpreter or the call stub.
2326
2327 // push the return address
2328 __ push(retaddr);
2329
2330 // and begin the OSR nmethod
2331 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2332 }
2333 }
2334 }
2335
2336 void TemplateTable::if_0cmp(Condition cc) {
2337 transition(itos, vtos);
2338 // assume branch is more often taken than not (loops use backward branches)
2339 Label not_taken;
2340 __ testl(rax, rax);
2341 __ jcc(j_not(cc), not_taken);
2342 branch(false, false);
2343 __ bind(not_taken);
2344 __ profile_not_taken_branch(rax);
2345 }
2346
2347 void TemplateTable::if_icmp(Condition cc) {
2348 transition(itos, vtos);
2349 // assume branch is more often taken than not (loops use backward branches)
2350 Label not_taken;
2351 __ pop_i(rdx);
2352 __ cmpl(rdx, rax);
2353 __ jcc(j_not(cc), not_taken);
2354 branch(false, false);
2355 __ bind(not_taken);
2356 __ profile_not_taken_branch(rax);
2357 }
2358
2359 void TemplateTable::if_nullcmp(Condition cc) {
2360 transition(atos, vtos);
2361 // assume branch is more often taken than not (loops use backward branches)
2362 Label not_taken;
2363 __ testptr(rax, rax);
2364 __ jcc(j_not(cc), not_taken);
2365 branch(false, false);
2366 __ bind(not_taken);
2367 __ profile_not_taken_branch(rax);
2368 }
2369
2370 void TemplateTable::if_acmp(Condition cc) {
2371 transition(atos, vtos);
2372 // assume branch is more often taken than not (loops use backward branches)
2373 Label not_taken;
2374 __ pop_ptr(rdx);
2375 __ cmpptr(rdx, rax);
2376 __ jcc(j_not(cc), not_taken);
2377 branch(false, false);
2378 __ bind(not_taken);
2379 __ profile_not_taken_branch(rax);
2380 }
2381
2382 void TemplateTable::ret() {
2383 transition(vtos, vtos);
2384 locals_index(rbx);
2385 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2386 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2387 __ profile_ret(rbx, rcx);
2388 __ get_method(rax);
2389 __ movptr(rbcp, Address(rax, Method::const_offset()));
2390 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2391 ConstMethod::codes_offset()));
2392 __ dispatch_next(vtos);
2393 }
2394
2395 void TemplateTable::wide_ret() {
2396 transition(vtos, vtos);
2397 locals_index_wide(rbx);
2398 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2399 __ profile_ret(rbx, rcx);
2400 __ get_method(rax);
2401 __ movptr(rbcp, Address(rax, Method::const_offset()));
2402 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2403 __ dispatch_next(vtos);
2404 }
2405
2406 void TemplateTable::tableswitch() {
2407 Label default_case, continue_execution;
2408 transition(itos, vtos);
2409
2410 // align r13/rsi
2411 __ lea(rbx, at_bcp(BytesPerInt));
2412 __ andptr(rbx, -BytesPerInt);
2413 // load lo & hi
2414 __ movl(rcx, Address(rbx, BytesPerInt));
2415 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2416 __ bswapl(rcx);
2417 __ bswapl(rdx);
2418 // check against lo & hi
2419 __ cmpl(rax, rcx);
2420 __ jcc(Assembler::less, default_case);
2421 __ cmpl(rax, rdx);
2422 __ jcc(Assembler::greater, default_case);
2423 // lookup dispatch offset
2424 __ subl(rax, rcx);
2425 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2426 __ profile_switch_case(rax, rbx, rcx);
2427 // continue execution
2428 __ bind(continue_execution);
2429 __ bswapl(rdx);
2430 LP64_ONLY(__ movl2ptr(rdx, rdx));
2431 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2432 __ addptr(rbcp, rdx);
2433 __ dispatch_only(vtos);
2434 // handle default
2435 __ bind(default_case);
2436 __ profile_switch_default(rax);
2437 __ movl(rdx, Address(rbx, 0));
2438 __ jmp(continue_execution);
2439 }
2440
2441 void TemplateTable::lookupswitch() {
2442 transition(itos, itos);
2443 __ stop("lookupswitch bytecode should have been rewritten");
2444 }
2445
2446 void TemplateTable::fast_linearswitch() {
2447 transition(itos, vtos);
2448 Label loop_entry, loop, found, continue_execution;
2449 // bswap rax so we can avoid bswapping the table entries
2450 __ bswapl(rax);
2451 // align r13
2452 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2453 // this instruction (change offsets
2454 // below)
2455 __ andptr(rbx, -BytesPerInt);
2456 // set counter
2457 __ movl(rcx, Address(rbx, BytesPerInt));
2458 __ bswapl(rcx);
2459 __ jmpb(loop_entry);
2460 // table search
2461 __ bind(loop);
2462 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2463 __ jcc(Assembler::equal, found);
2464 __ bind(loop_entry);
2465 __ decrementl(rcx);
2466 __ jcc(Assembler::greaterEqual, loop);
2467 // default case
2468 __ profile_switch_default(rax);
2469 __ movl(rdx, Address(rbx, 0));
2470 __ jmp(continue_execution);
2471 // entry found -> get offset
2472 __ bind(found);
2473 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2474 __ profile_switch_case(rcx, rax, rbx);
2475 // continue execution
2476 __ bind(continue_execution);
2477 __ bswapl(rdx);
2478 __ movl2ptr(rdx, rdx);
2479 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2480 __ addptr(rbcp, rdx);
2481 __ dispatch_only(vtos);
2482 }
2483
2484 void TemplateTable::fast_binaryswitch() {
2485 transition(itos, vtos);
2486 // Implementation using the following core algorithm:
2487 //
2488 // int binary_search(int key, LookupswitchPair* array, int n) {
2489 // // Binary search according to "Methodik des Programmierens" by
2490 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2491 // int i = 0;
2492 // int j = n;
2493 // while (i+1 < j) {
2494 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2495 // // with Q: for all i: 0 <= i < n: key < a[i]
2496 // // where a stands for the array and assuming that the (inexisting)
2497 // // element a[n] is infinitely big.
2498 // int h = (i + j) >> 1;
2499 // // i < h < j
2500 // if (key < array[h].fast_match()) {
2501 // j = h;
2502 // } else {
2503 // i = h;
2504 // }
2505 // }
2506 // // R: a[i] <= key < a[i+1] or Q
2507 // // (i.e., if key is within array, i is the correct index)
2508 // return i;
2509 // }
2510
2511 // Register allocation
2512 const Register key = rax; // already set (tosca)
2513 const Register array = rbx;
2514 const Register i = rcx;
2515 const Register j = rdx;
2516 const Register h = rdi;
2517 const Register temp = rsi;
2518
2519 // Find array start
2520 NOT_LP64(__ save_bcp());
2521
2522 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2523 // get rid of this
2524 // instruction (change
2525 // offsets below)
2526 __ andptr(array, -BytesPerInt);
2527
2528 // Initialize i & j
2529 __ xorl(i, i); // i = 0;
2530 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2531
2532 // Convert j into native byteordering
2533 __ bswapl(j);
2534
2535 // And start
2536 Label entry;
2537 __ jmp(entry);
2538
2539 // binary search loop
2540 {
2541 Label loop;
2542 __ bind(loop);
2543 // int h = (i + j) >> 1;
2544 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2545 __ sarl(h, 1); // h = (i + j) >> 1;
2546 // if (key < array[h].fast_match()) {
2547 // j = h;
2548 // } else {
2549 // i = h;
2550 // }
2551 // Convert array[h].match to native byte-ordering before compare
2552 __ movl(temp, Address(array, h, Address::times_8));
2553 __ bswapl(temp);
2554 __ cmpl(key, temp);
2555 // j = h if (key < array[h].fast_match())
2556 __ cmov32(Assembler::less, j, h);
2557 // i = h if (key >= array[h].fast_match())
2558 __ cmov32(Assembler::greaterEqual, i, h);
2559 // while (i+1 < j)
2560 __ bind(entry);
2561 __ leal(h, Address(i, 1)); // i+1
2562 __ cmpl(h, j); // i+1 < j
2563 __ jcc(Assembler::less, loop);
2564 }
2565
2566 // end of binary search, result index is i (must check again!)
2567 Label default_case;
2568 // Convert array[i].match to native byte-ordering before compare
2569 __ movl(temp, Address(array, i, Address::times_8));
2570 __ bswapl(temp);
2571 __ cmpl(key, temp);
2572 __ jcc(Assembler::notEqual, default_case);
2573
2574 // entry found -> j = offset
2575 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2576 __ profile_switch_case(i, key, array);
2577 __ bswapl(j);
2578 LP64_ONLY(__ movslq(j, j));
2579
2580 NOT_LP64(__ restore_bcp());
2581 NOT_LP64(__ restore_locals()); // restore rdi
2582
2583 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2584 __ addptr(rbcp, j);
2585 __ dispatch_only(vtos);
2586
2587 // default case -> j = default offset
2588 __ bind(default_case);
2589 __ profile_switch_default(i);
2590 __ movl(j, Address(array, -2 * BytesPerInt));
2591 __ bswapl(j);
2592 LP64_ONLY(__ movslq(j, j));
2593
2594 NOT_LP64(__ restore_bcp());
2595 NOT_LP64(__ restore_locals());
2596
2597 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2598 __ addptr(rbcp, j);
2599 __ dispatch_only(vtos);
2600 }
2601
2602 void TemplateTable::_return(TosState state) {
2603 transition(state, state);
2604
2605 assert(_desc->calls_vm(),
2606 "inconsistent calls_vm information"); // call in remove_activation
2607
2608 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2609 assert(state == vtos, "only valid state");
2610 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2611 __ movptr(robj, aaddress(0));
2612 __ load_klass(rdi, robj);
2613 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2614 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2615 Label skip_register_finalizer;
2616 __ jcc(Assembler::zero, skip_register_finalizer);
2617
2618 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2619
2620 __ bind(skip_register_finalizer);
2621 }
2622
2623 // Narrow result if state is itos but result type is smaller.
2624 // Need to narrow in the return bytecode rather than in generate_return_entry
2625 // since compiled code callers expect the result to already be narrowed.
2626 if (state == itos) {
2627 __ narrow(rax);
2628 }
2629 __ remove_activation(state, rbcp, true, true, true, state == qtos && ValueTypeReturnedAsFields);
2630
2631 __ jmp(rbcp);
2632 }
2633
2634 // ----------------------------------------------------------------------------
2635 // Volatile variables demand their effects be made known to all CPU's
2636 // in order. Store buffers on most chips allow reads & writes to
2637 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2638 // without some kind of memory barrier (i.e., it's not sufficient that
2639 // the interpreter does not reorder volatile references, the hardware
2640 // also must not reorder them).
2641 //
2642 // According to the new Java Memory Model (JMM):
2643 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2644 // writes act as aquire & release, so:
2645 // (2) A read cannot let unrelated NON-volatile memory refs that
2646 // happen after the read float up to before the read. It's OK for
2647 // non-volatile memory refs that happen before the volatile read to
2648 // float down below it.
2649 // (3) Similar a volatile write cannot let unrelated NON-volatile
2650 // memory refs that happen BEFORE the write float down to after the
2651 // write. It's OK for non-volatile memory refs that happen after the
2652 // volatile write to float up before it.
2653 //
2654 // We only put in barriers around volatile refs (they are expensive),
2655 // not _between_ memory refs (that would require us to track the
2656 // flavor of the previous memory refs). Requirements (2) and (3)
2657 // require some barriers before volatile stores and after volatile
2658 // loads. These nearly cover requirement (1) but miss the
2659 // volatile-store-volatile-load case. This final case is placed after
2660 // volatile-stores although it could just as well go before
2661 // volatile-loads.
2662
2663 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2664 // Helper function to insert a is-volatile test and memory barrier
2665 if(!os::is_MP()) return; // Not needed on single CPU
2666 __ membar(order_constraint);
2667 }
2668
2669 void TemplateTable::resolve_cache_and_index(int byte_no,
2670 Register Rcache,
2671 Register index,
2672 size_t index_size) {
2673 const Register temp = rbx;
2674 assert_different_registers(Rcache, index, temp);
2675
2676 Label resolved;
2677
2678 Bytecodes::Code code = bytecode();
2679 switch (code) {
2680 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2681 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2682 }
2683
2684 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2685 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2686 __ cmpl(temp, code); // have we resolved this bytecode?
2687 __ jcc(Assembler::equal, resolved);
2688
2689 // resolve first time through
2690 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2691 __ movl(temp, code);
2692 __ call_VM(noreg, entry, temp);
2693 // Update registers with resolved info
2694 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2695 __ bind(resolved);
2696 }
2697
2698 // The cache and index registers must be set before call
2699 void TemplateTable::load_field_cp_cache_entry(Register obj,
2700 Register cache,
2701 Register index,
2702 Register off,
2703 Register flags,
2704 bool is_static = false) {
2705 assert_different_registers(cache, index, flags, off);
2706
2707 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2708 // Field offset
2709 __ movptr(off, Address(cache, index, Address::times_ptr,
2710 in_bytes(cp_base_offset +
2711 ConstantPoolCacheEntry::f2_offset())));
2712 // Flags
2713 __ movl(flags, Address(cache, index, Address::times_ptr,
2714 in_bytes(cp_base_offset +
2715 ConstantPoolCacheEntry::flags_offset())));
2716
2717 // klass overwrite register
2718 if (is_static) {
2719 __ movptr(obj, Address(cache, index, Address::times_ptr,
2720 in_bytes(cp_base_offset +
2721 ConstantPoolCacheEntry::f1_offset())));
2722 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2723 __ movptr(obj, Address(obj, mirror_offset));
2724 }
2725 }
2726
2727 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2728 Register method,
2729 Register itable_index,
2730 Register flags,
2731 bool is_invokevirtual,
2732 bool is_invokevfinal, /*unused*/
2733 bool is_invokedynamic) {
2734 // setup registers
2735 const Register cache = rcx;
2736 const Register index = rdx;
2737 assert_different_registers(method, flags);
2738 assert_different_registers(method, cache, index);
2739 assert_different_registers(itable_index, flags);
2740 assert_different_registers(itable_index, cache, index);
2741 // determine constant pool cache field offsets
2742 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2743 const int method_offset = in_bytes(
2744 ConstantPoolCache::base_offset() +
2745 ((byte_no == f2_byte)
2746 ? ConstantPoolCacheEntry::f2_offset()
2747 : ConstantPoolCacheEntry::f1_offset()));
2748 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2749 ConstantPoolCacheEntry::flags_offset());
2750 // access constant pool cache fields
2751 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2752 ConstantPoolCacheEntry::f2_offset());
2753
2754 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2755 resolve_cache_and_index(byte_no, cache, index, index_size);
2756 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2757
2758 if (itable_index != noreg) {
2759 // pick up itable or appendix index from f2 also:
2760 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2761 }
2762 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2763 }
2764
2765 // The registers cache and index expected to be set before call.
2766 // Correct values of the cache and index registers are preserved.
2767 void TemplateTable::jvmti_post_field_access(Register cache,
2768 Register index,
2769 bool is_static,
2770 bool has_tos) {
2771 if (JvmtiExport::can_post_field_access()) {
2772 // Check to see if a field access watch has been set before we take
2773 // the time to call into the VM.
2774 Label L1;
2775 assert_different_registers(cache, index, rax);
2776 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2777 __ testl(rax,rax);
2778 __ jcc(Assembler::zero, L1);
2779
2780 // cache entry pointer
2781 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2782 __ shll(index, LogBytesPerWord);
2783 __ addptr(cache, index);
2784 if (is_static) {
2785 __ xorptr(rax, rax); // NULL object reference
2786 } else {
2787 __ pop(atos); // Get the object
2788 __ verify_oop(rax);
2789 __ push(atos); // Restore stack state
2790 }
2791 // rax,: object pointer or NULL
2792 // cache: cache entry pointer
2793 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2794 rax, cache);
2795 __ get_cache_and_index_at_bcp(cache, index, 1);
2796 __ bind(L1);
2797 }
2798 }
2799
2800 void TemplateTable::pop_and_check_object(Register r) {
2801 __ pop_ptr(r);
2802 __ null_check(r); // for field access must check obj.
2803 __ verify_oop(r);
2804 }
2805
2806 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc, bool is_vgetfield) {
2807 transition(vtos, vtos);
2808
2809 const Register cache = rcx;
2810 const Register index = rdx;
2811 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2812 const Register off = rbx;
2813 const Register flags = rax;
2814 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2815
2816 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2817 jvmti_post_field_access(cache, index, is_static, false);
2818 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2819
2820 const Address field(obj, off, Address::times_1, 0*wordSize);
2821 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2822
2823 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notValueType, notDouble;
2824
2825 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2826 // Make sure we don't need to mask edx after the above shift
2827 assert(btos == 0, "change code, btos != 0");
2828
2829 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2830
2831 __ jcc(Assembler::notZero, notByte);
2832 // btos
2833 if (!is_static) pop_and_check_object(obj);
2834 __ load_signed_byte(rax, field);
2835 __ push(btos);
2836 // Rewrite bytecode to be faster
2837 if (!is_static && rc == may_rewrite) {
2838 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2839 }
2840 __ jmp(Done);
2841
2842 __ bind(notByte);
2843
2844 __ cmpl(flags, qtos);
2845 __ jcc(Assembler::notEqual, notValueType);
2846 // qtos
2847 if (is_static) {
2848 __ load_heap_oop(rax, field);
2849 __ push(qtos);
2850 // if (!is_static && !is_vgetfield) {
2851 // patch_bytecode(Bytecodes::_fast_qgetfield, bc, rbx);
2852 // }
2853 } else {
2854
2855 // cp cache entry pointer
2856 // __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2857 // __ shll(index, LogBytesPerWord);
2858 // __ addptr(cache, index);
2859
2860 pop_and_check_object(obj);
2861
2862 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::qgetfield),
2863 obj, off);
2864 __ verify_oop(rax);
2865 __ push(qtos);
2866 // Bytecode rewrite?
2867 }
2868 __ jmp(Done);
2869
2870 __ bind(notValueType);
2871
2872 if (!is_static) pop_and_check_object(obj);
2873
2874 __ cmpl(flags, ztos);
2875 __ jcc(Assembler::notEqual, notBool);
2876
2877 // ztos (same code as btos)
2878 __ load_signed_byte(rax, field);
2879 __ push(ztos);
2880 // Rewrite bytecode to be faster
2881 if (!is_static && rc == may_rewrite) {
2882 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2883 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2884 }
2885 __ jmp(Done);
2886
2887 __ bind(notBool);
2888 __ cmpl(flags, atos);
2889 __ jcc(Assembler::notEqual, notObj);
2890 // atos
2891 __ load_heap_oop(rax, field);
2892 __ push(atos);
2893 if (!is_static && rc == may_rewrite) {
2894 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2895 }
2896 __ jmp(Done);
2897
2898 __ bind(notObj);
2899 __ cmpl(flags, itos);
2900 __ jcc(Assembler::notEqual, notInt);
2901 // itos
2902 __ movl(rax, field);
2903 __ push(itos);
2904 // Rewrite bytecode to be faster
2905 if (!is_static && rc == may_rewrite) {
2906 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2907 }
2908 __ jmp(Done);
2909
2910 __ bind(notInt);
2911 __ cmpl(flags, ctos);
2912 __ jcc(Assembler::notEqual, notChar);
2913 // ctos
2914 __ load_unsigned_short(rax, field);
2915 __ push(ctos);
2916 // Rewrite bytecode to be faster
2917 if (!is_static && rc == may_rewrite) {
2918 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2919 }
2920 __ jmp(Done);
2921
2922 __ bind(notChar);
2923 __ cmpl(flags, stos);
2924 __ jcc(Assembler::notEqual, notShort);
2925 // stos
2926 __ load_signed_short(rax, field);
2927 __ push(stos);
2928 // Rewrite bytecode to be faster
2929 if (!is_static && rc == may_rewrite) {
2930 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2931 }
2932 __ jmp(Done);
2933
2934 __ bind(notShort);
2935 __ cmpl(flags, ltos);
2936 __ jcc(Assembler::notEqual, notLong);
2937 // ltos
2938
2939 #ifndef _LP64
2940 // Generate code as if volatile. There just aren't enough registers to
2941 // save that information and this code is faster than the test.
2942 __ fild_d(field); // Must load atomically
2943 __ subptr(rsp,2*wordSize); // Make space for store
2944 __ fistp_d(Address(rsp,0));
2945 __ pop(rax);
2946 __ pop(rdx);
2947 #else
2948 __ movq(rax, field);
2949 #endif
2950
2951 __ push(ltos);
2952 // Rewrite bytecode to be faster
2953 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2954 __ jmp(Done);
2955
2956 __ bind(notLong);
2957 __ cmpl(flags, ftos);
2958 __ jcc(Assembler::notEqual, notFloat);
2959 // ftos
2960
2961 __ load_float(field);
2962 __ push(ftos);
2963 // Rewrite bytecode to be faster
2964 if (!is_static && rc == may_rewrite) {
2965 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2966 }
2967 __ jmp(Done);
2968
2969 __ bind(notFloat);
2970 #ifdef ASSERT
2971 __ cmpl(flags, dtos);
2972 __ jcc(Assembler::notEqual, notDouble);
2973 #endif
2974 // dtos
2975 __ load_double(field);
2976 __ push(dtos);
2977 // Rewrite bytecode to be faster
2978 if (!is_static && rc == may_rewrite) {
2979 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2980 }
2981 #ifdef ASSERT
2982 __ jmp(Done);
2983
2984
2985 __ bind(notDouble);
2986 __ stop("Bad state");
2987 #endif
2988
2989 __ bind(Done);
2990 // [jk] not needed currently
2991 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2992 // Assembler::LoadStore));
2993 }
2994
2995 void TemplateTable::getfield(int byte_no) {
2996 getfield_or_static(byte_no, false);
2997 }
2998
2999 void TemplateTable::nofast_getfield(int byte_no) {
3000 getfield_or_static(byte_no, false, may_not_rewrite);
3001 }
3002
3003 void TemplateTable::getstatic(int byte_no) {
3004 getfield_or_static(byte_no, true);
3005 }
3006
3007 void TemplateTable::vgetfield() {
3008 getfield_or_static(f1_byte, false, may_not_rewrite, true);
3009 }
3010
3011 void TemplateTable::vwithfield() {
3012 transition(vtos, qtos);
3013
3014 Register cache = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
3015 Register index = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3016
3017 resolve_cache_and_index(f2_byte, cache, index, sizeof(u2));
3018
3019 call_VM(rbx, CAST_FROM_FN_PTR(address, InterpreterRuntime::vwithfield), cache);
3020 // new value type is returned in rbx
3021 // stack adjustement is returned in rax
3022 __ verify_oop(rbx);
3023 __ addptr(rsp, rax);
3024 __ movptr(rax, rbx);
3025 }
3026
3027 // The registers cache and index expected to be set before call.
3028 // The function may destroy various registers, just not the cache and index registers.
3029 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3030
3031 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
3032 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
3033 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3034 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
3035
3036 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3037
3038 if (JvmtiExport::can_post_field_modification()) {
3039 // Check to see if a field modification watch has been set before
3040 // we take the time to call into the VM.
3041 Label L1;
3042 assert_different_registers(cache, index, rax);
3043 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3044 __ testl(rax, rax);
3045 __ jcc(Assembler::zero, L1);
3046
3047 __ get_cache_and_index_at_bcp(robj, RDX, 1);
3048
3049
3050 if (is_static) {
3051 // Life is simple. Null out the object pointer.
3052 __ xorl(RBX, RBX);
3053
3054 } else {
3055 // Life is harder. The stack holds the value on top, followed by
3056 // the object. We don't know the size of the value, though; it
3057 // could be one or two words depending on its type. As a result,
3058 // we must find the type to determine where the object is.
3059 #ifndef _LP64
3060 Label two_word, valsize_known;
3061 #endif
3062 __ movl(RCX, Address(robj, RDX,
3063 Address::times_ptr,
3064 in_bytes(cp_base_offset +
3065 ConstantPoolCacheEntry::flags_offset())));
3066 NOT_LP64(__ mov(rbx, rsp));
3067 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3068
3069 // Make sure we don't need to mask rcx after the above shift
3070 ConstantPoolCacheEntry::verify_tos_state_shift();
3071 #ifdef _LP64
3072 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
3073 __ cmpl(c_rarg3, ltos);
3074 __ cmovptr(Assembler::equal,
3075 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3076 __ cmpl(c_rarg3, dtos);
3077 __ cmovptr(Assembler::equal,
3078 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3079 #else
3080 __ cmpl(rcx, ltos);
3081 __ jccb(Assembler::equal, two_word);
3082 __ cmpl(rcx, dtos);
3083 __ jccb(Assembler::equal, two_word);
3084 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3085 __ jmpb(valsize_known);
3086
3087 __ bind(two_word);
3088 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3089
3090 __ bind(valsize_known);
3091 // setup object pointer
3092 __ movptr(rbx, Address(rbx, 0));
3093 #endif
3094 }
3095 // cache entry pointer
3096 __ addptr(robj, in_bytes(cp_base_offset));
3097 __ shll(RDX, LogBytesPerWord);
3098 __ addptr(robj, RDX);
3099 // object (tos)
3100 __ mov(RCX, rsp);
3101 // c_rarg1: object pointer set up above (NULL if static)
3102 // c_rarg2: cache entry pointer
3103 // c_rarg3: jvalue object on the stack
3104 __ call_VM(noreg,
3105 CAST_FROM_FN_PTR(address,
3106 InterpreterRuntime::post_field_modification),
3107 RBX, robj, RCX);
3108 __ get_cache_and_index_at_bcp(cache, index, 1);
3109 __ bind(L1);
3110 }
3111 }
3112
3113 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3114 transition(vtos, vtos);
3115
3116 const Register cache = rcx;
3117 const Register index = rdx;
3118 const Register obj = rcx;
3119 const Register off = rbx;
3120 const Register flags = rax;
3121 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3122
3123 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3124 jvmti_post_field_mod(cache, index, is_static);
3125 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3126
3127 // [jk] not needed currently
3128 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3129 // Assembler::StoreStore));
3130
3131 Label notVolatile, Done;
3132 __ movl(rdx, flags);
3133 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3134 __ andl(rdx, 0x1);
3135
3136 // field addresses
3137 const Address field(obj, off, Address::times_1, 0*wordSize);
3138 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3139
3140 Label notByte, notBool, notInt, notShort, notChar,
3141 notLong, notFloat, notObj, notValueType, notDouble;
3142
3143 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3144
3145 assert(btos == 0, "change code, btos != 0");
3146 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3147 __ jcc(Assembler::notZero, notByte);
3148
3149 // btos
3150 {
3151 __ pop(btos);
3152 if (!is_static) pop_and_check_object(obj);
3153 __ movb(field, rax);
3154 if (!is_static && rc == may_rewrite) {
3155 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3156 }
3157 __ jmp(Done);
3158 }
3159
3160 __ bind(notByte);
3161 __ cmpl(flags, ztos);
3162 __ jcc(Assembler::notEqual, notBool);
3163
3164 // ztos
3165 {
3166 __ pop(ztos);
3167 if (!is_static) pop_and_check_object(obj);
3168 __ andl(rax, 0x1);
3169 __ movb(field, rax);
3170 if (!is_static && rc == may_rewrite) {
3171 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3172 }
3173 __ jmp(Done);
3174 }
3175
3176 __ bind(notBool);
3177 __ cmpl(flags, atos);
3178 __ jcc(Assembler::notEqual, notObj);
3179
3180 // atos
3181 {
3182 __ pop(atos);
3183 if (!is_static) pop_and_check_object(obj);
3184 // Store into the field
3185 do_oop_store(_masm, field, rax, _bs->kind(), false);
3186 if (!is_static && rc == may_rewrite) {
3187 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3188 }
3189 __ jmp(Done);
3190 }
3191
3192 __ bind(notObj);
3193 __ cmpl(flags, qtos);
3194 __ jcc(Assembler::notEqual, notValueType);
3195
3196 // qtos
3197 {
3198 __ pop(qtos); // => rax == value
3199 if (!is_static) {
3200 // value types in non-static fields are embedded
3201 __ movl(rcx, off);
3202 pop_and_check_object(rbx);
3203 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::qputfield),
3204 rbx, rax, rcx);
3205 __ jmp(notVolatile); // value types are never volatile
3206 } else {
3207 // Store into the static field
3208 // Value types in static fields are currently handled with indirection
3209 do_oop_store(_masm, field, rax, _bs->kind(), false);
3210 }
3211 __ jmp(Done);
3212 }
3213
3214 __ bind(notValueType);
3215 __ cmpl(flags, itos);
3216 __ jcc(Assembler::notEqual, notInt);
3217
3218 // itos
3219 {
3220 __ pop(itos);
3221 if (!is_static) pop_and_check_object(obj);
3222 __ movl(field, rax);
3223 if (!is_static && rc == may_rewrite) {
3224 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3225 }
3226 __ jmp(Done);
3227 }
3228
3229 __ bind(notInt);
3230 __ cmpl(flags, ctos);
3231 __ jcc(Assembler::notEqual, notChar);
3232
3233 // ctos
3234 {
3235 __ pop(ctos);
3236 if (!is_static) pop_and_check_object(obj);
3237 __ movw(field, rax);
3238 if (!is_static && rc == may_rewrite) {
3239 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3240 }
3241 __ jmp(Done);
3242 }
3243
3244 __ bind(notChar);
3245 __ cmpl(flags, stos);
3246 __ jcc(Assembler::notEqual, notShort);
3247
3248 // stos
3249 {
3250 __ pop(stos);
3251 if (!is_static) pop_and_check_object(obj);
3252 __ movw(field, rax);
3253 if (!is_static && rc == may_rewrite) {
3254 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3255 }
3256 __ jmp(Done);
3257 }
3258
3259 __ bind(notShort);
3260 __ cmpl(flags, ltos);
3261 __ jcc(Assembler::notEqual, notLong);
3262
3263 // ltos
3264 #ifdef _LP64
3265 {
3266 __ pop(ltos);
3267 if (!is_static) pop_and_check_object(obj);
3268 __ movq(field, rax);
3269 if (!is_static && rc == may_rewrite) {
3270 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3271 }
3272 __ jmp(Done);
3273 }
3274 #else
3275 {
3276 Label notVolatileLong;
3277 __ testl(rdx, rdx);
3278 __ jcc(Assembler::zero, notVolatileLong);
3279
3280 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3281 if (!is_static) pop_and_check_object(obj);
3282
3283 // Replace with real volatile test
3284 __ push(rdx);
3285 __ push(rax); // Must update atomically with FIST
3286 __ fild_d(Address(rsp,0)); // So load into FPU register
3287 __ fistp_d(field); // and put into memory atomically
3288 __ addptr(rsp, 2*wordSize);
3289 // volatile_barrier();
3290 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3291 Assembler::StoreStore));
3292 // Don't rewrite volatile version
3293 __ jmp(notVolatile);
3294
3295 __ bind(notVolatileLong);
3296
3297 __ pop(ltos); // overwrites rdx
3298 if (!is_static) pop_and_check_object(obj);
3299 __ movptr(hi, rdx);
3300 __ movptr(field, rax);
3301 // Don't rewrite to _fast_lputfield for potential volatile case.
3302 __ jmp(notVolatile);
3303 }
3304 #endif // _LP64
3305
3306 __ bind(notLong);
3307 __ cmpl(flags, ftos);
3308 __ jcc(Assembler::notEqual, notFloat);
3309
3310 // ftos
3311 {
3312 __ pop(ftos);
3313 if (!is_static) pop_and_check_object(obj);
3314 __ store_float(field);
3315 if (!is_static && rc == may_rewrite) {
3316 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3317 }
3318 __ jmp(Done);
3319 }
3320
3321 __ bind(notFloat);
3322 #ifdef ASSERT
3323 __ cmpl(flags, dtos);
3324 __ jcc(Assembler::notEqual, notDouble);
3325 #endif
3326
3327 // dtos
3328 {
3329 __ pop(dtos);
3330 if (!is_static) pop_and_check_object(obj);
3331 __ store_double(field);
3332 if (!is_static && rc == may_rewrite) {
3333 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3334 }
3335 }
3336
3337 #ifdef ASSERT
3338 __ jmp(Done);
3339
3340 __ bind(notDouble);
3341 __ stop("Bad state");
3342 #endif
3343
3344 __ bind(Done);
3345
3346 // Check for volatile store
3347 __ testl(rdx, rdx);
3348 __ jcc(Assembler::zero, notVolatile);
3349 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3350 Assembler::StoreStore));
3351 __ bind(notVolatile);
3352 }
3353
3354 void TemplateTable::putfield(int byte_no) {
3355 putfield_or_static(byte_no, false);
3356 }
3357
3358 void TemplateTable::nofast_putfield(int byte_no) {
3359 putfield_or_static(byte_no, false, may_not_rewrite);
3360 }
3361
3362 void TemplateTable::putstatic(int byte_no) {
3363 putfield_or_static(byte_no, true);
3364 }
3365
3366 void TemplateTable::jvmti_post_fast_field_mod() {
3367
3368 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3369
3370 if (JvmtiExport::can_post_field_modification()) {
3371 // Check to see if a field modification watch has been set before
3372 // we take the time to call into the VM.
3373 Label L2;
3374 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3375 __ testl(scratch, scratch);
3376 __ jcc(Assembler::zero, L2);
3377 __ pop_ptr(rbx); // copy the object pointer from tos
3378 __ verify_oop(rbx);
3379 __ push_ptr(rbx); // put the object pointer back on tos
3380 // Save tos values before call_VM() clobbers them. Since we have
3381 // to do it for every data type, we use the saved values as the
3382 // jvalue object.
3383 switch (bytecode()) { // load values into the jvalue object
3384 case Bytecodes::_fast_qputfield: //fall through
3385 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3386 case Bytecodes::_fast_bputfield: // fall through
3387 case Bytecodes::_fast_zputfield: // fall through
3388 case Bytecodes::_fast_sputfield: // fall through
3389 case Bytecodes::_fast_cputfield: // fall through
3390 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3391 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3392 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3393 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3394
3395 default:
3396 ShouldNotReachHere();
3397 }
3398 __ mov(scratch, rsp); // points to jvalue on the stack
3399 // access constant pool cache entry
3400 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3401 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3402 __ verify_oop(rbx);
3403 // rbx: object pointer copied above
3404 // c_rarg2: cache entry pointer
3405 // c_rarg3: jvalue object on the stack
3406 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3407 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3408
3409 switch (bytecode()) { // restore tos values
3410 case Bytecodes::_fast_qputfield: // fall through
3411 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3412 case Bytecodes::_fast_bputfield: // fall through
3413 case Bytecodes::_fast_zputfield: // fall through
3414 case Bytecodes::_fast_sputfield: // fall through
3415 case Bytecodes::_fast_cputfield: // fall through
3416 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3417 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3418 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3419 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3420 }
3421 __ bind(L2);
3422 }
3423 }
3424
3425 void TemplateTable::fast_storefield(TosState state) {
3426 transition(state, vtos);
3427
3428 ByteSize base = ConstantPoolCache::base_offset();
3429
3430 jvmti_post_fast_field_mod();
3431
3432 // access constant pool cache
3433 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3434
3435 // test for volatile with rdx but rdx is tos register for lputfield.
3436 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3437 in_bytes(base +
3438 ConstantPoolCacheEntry::flags_offset())));
3439
3440 // replace index with field offset from cache entry
3441 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3442 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3443
3444 // [jk] not needed currently
3445 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3446 // Assembler::StoreStore));
3447
3448 Label notVolatile;
3449 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3450 __ andl(rdx, 0x1);
3451
3452 // Get object from stack
3453 pop_and_check_object(rcx);
3454
3455 // field address
3456 const Address field(rcx, rbx, Address::times_1);
3457
3458 // access field
3459 switch (bytecode()) {
3460 case Bytecodes::_fast_qputfield:
3461 __ stop("should not be rewritten yet");
3462 break;
3463 case Bytecodes::_fast_aputfield:
3464 do_oop_store(_masm, field, rax, _bs->kind(), false);
3465 break;
3466 case Bytecodes::_fast_lputfield:
3467 #ifdef _LP64
3468 __ movq(field, rax);
3469 #else
3470 __ stop("should not be rewritten");
3471 #endif
3472 break;
3473 case Bytecodes::_fast_iputfield:
3474 __ movl(field, rax);
3475 break;
3476 case Bytecodes::_fast_zputfield:
3477 __ andl(rax, 0x1); // boolean is true if LSB is 1
3478 // fall through to bputfield
3479 case Bytecodes::_fast_bputfield:
3480 __ movb(field, rax);
3481 break;
3482 case Bytecodes::_fast_sputfield:
3483 // fall through
3484 case Bytecodes::_fast_cputfield:
3485 __ movw(field, rax);
3486 break;
3487 case Bytecodes::_fast_fputfield:
3488 __ store_float(field);
3489 break;
3490 case Bytecodes::_fast_dputfield:
3491 __ store_double(field);
3492 break;
3493 default:
3494 ShouldNotReachHere();
3495 }
3496
3497 // Check for volatile store
3498 __ testl(rdx, rdx);
3499 __ jcc(Assembler::zero, notVolatile);
3500 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3501 Assembler::StoreStore));
3502 __ bind(notVolatile);
3503 }
3504
3505 void TemplateTable::fast_accessfield(TosState state) {
3506 transition(atos, state);
3507
3508 // Do the JVMTI work here to avoid disturbing the register state below
3509 if (JvmtiExport::can_post_field_access()) {
3510 // Check to see if a field access watch has been set before we
3511 // take the time to call into the VM.
3512 Label L1;
3513 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3514 __ testl(rcx, rcx);
3515 __ jcc(Assembler::zero, L1);
3516 // access constant pool cache entry
3517 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3518 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3519 __ verify_oop(rax);
3520 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3521 LP64_ONLY(__ mov(c_rarg1, rax));
3522 // c_rarg1: object pointer copied above
3523 // c_rarg2: cache entry pointer
3524 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3525 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3526 __ pop_ptr(rax); // restore object pointer
3527 __ bind(L1);
3528 }
3529
3530 // access constant pool cache
3531 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3532 // replace index with field offset from cache entry
3533 // [jk] not needed currently
3534 // if (os::is_MP()) {
3535 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3536 // in_bytes(ConstantPoolCache::base_offset() +
3537 // ConstantPoolCacheEntry::flags_offset())));
3538 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3539 // __ andl(rdx, 0x1);
3540 // }
3541 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3542 in_bytes(ConstantPoolCache::base_offset() +
3543 ConstantPoolCacheEntry::f2_offset())));
3544
3545 // rax: object
3546 __ verify_oop(rax);
3547 __ null_check(rax);
3548 Address field(rax, rbx, Address::times_1);
3549
3550 // access field
3551 switch (bytecode()) {
3552 case Bytecodes::_fast_qgetfield:
3553 __ stop("should not be rewritten yet");
3554 break;
3555 case Bytecodes::_fast_agetfield:
3556 __ load_heap_oop(rax, field);
3557 __ verify_oop(rax);
3558 break;
3559 case Bytecodes::_fast_lgetfield:
3560 #ifdef _LP64
3561 __ movq(rax, field);
3562 #else
3563 __ stop("should not be rewritten");
3564 #endif
3565 break;
3566 case Bytecodes::_fast_igetfield:
3567 __ movl(rax, field);
3568 break;
3569 case Bytecodes::_fast_bgetfield:
3570 __ movsbl(rax, field);
3571 break;
3572 case Bytecodes::_fast_sgetfield:
3573 __ load_signed_short(rax, field);
3574 break;
3575 case Bytecodes::_fast_cgetfield:
3576 __ load_unsigned_short(rax, field);
3577 break;
3578 case Bytecodes::_fast_fgetfield:
3579 __ load_float(field);
3580 break;
3581 case Bytecodes::_fast_dgetfield:
3582 __ load_double(field);
3583 break;
3584 default:
3585 ShouldNotReachHere();
3586 }
3587 // [jk] not needed currently
3588 // if (os::is_MP()) {
3589 // Label notVolatile;
3590 // __ testl(rdx, rdx);
3591 // __ jcc(Assembler::zero, notVolatile);
3592 // __ membar(Assembler::LoadLoad);
3593 // __ bind(notVolatile);
3594 //};
3595 }
3596
3597 void TemplateTable::fast_xaccess(TosState state) {
3598 transition(vtos, state);
3599
3600 // get receiver
3601 __ movptr(rax, aaddress(0));
3602 // access constant pool cache
3603 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3604 __ movptr(rbx,
3605 Address(rcx, rdx, Address::times_ptr,
3606 in_bytes(ConstantPoolCache::base_offset() +
3607 ConstantPoolCacheEntry::f2_offset())));
3608 // make sure exception is reported in correct bcp range (getfield is
3609 // next instruction)
3610 __ increment(rbcp);
3611 __ null_check(rax);
3612 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3613 switch (state) {
3614 case itos:
3615 __ movl(rax, field);
3616 break;
3617 case atos:
3618 __ load_heap_oop(rax, field);
3619 __ verify_oop(rax);
3620 break;
3621 case ftos:
3622 __ load_float(field);
3623 break;
3624 default:
3625 ShouldNotReachHere();
3626 }
3627
3628 // [jk] not needed currently
3629 // if (os::is_MP()) {
3630 // Label notVolatile;
3631 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3632 // in_bytes(ConstantPoolCache::base_offset() +
3633 // ConstantPoolCacheEntry::flags_offset())));
3634 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3635 // __ testl(rdx, 0x1);
3636 // __ jcc(Assembler::zero, notVolatile);
3637 // __ membar(Assembler::LoadLoad);
3638 // __ bind(notVolatile);
3639 // }
3640
3641 __ decrement(rbcp);
3642 }
3643
3644 //-----------------------------------------------------------------------------
3645 // Calls
3646
3647 void TemplateTable::count_calls(Register method, Register temp) {
3648 // implemented elsewhere
3649 ShouldNotReachHere();
3650 }
3651
3652 void TemplateTable::prepare_invoke(int byte_no,
3653 Register method, // linked method (or i-klass)
3654 Register index, // itable index, MethodType, etc.
3655 Register recv, // if caller wants to see it
3656 Register flags // if caller wants to test it
3657 ) {
3658 // determine flags
3659 const Bytecodes::Code code = bytecode();
3660 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3661 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3662 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3663 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3664 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3665 const bool load_receiver = (recv != noreg);
3666 const bool save_flags = (flags != noreg);
3667 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3668 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3669 assert(flags == noreg || flags == rdx, "");
3670 assert(recv == noreg || recv == rcx, "");
3671
3672 // setup registers & access constant pool cache
3673 if (recv == noreg) recv = rcx;
3674 if (flags == noreg) flags = rdx;
3675 assert_different_registers(method, index, recv, flags);
3676
3677 // save 'interpreter return address'
3678 __ save_bcp();
3679
3680 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3681
3682 // maybe push appendix to arguments (just before return address)
3683 if (is_invokedynamic || is_invokehandle) {
3684 Label L_no_push;
3685 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3686 __ jcc(Assembler::zero, L_no_push);
3687 // Push the appendix as a trailing parameter.
3688 // This must be done before we get the receiver,
3689 // since the parameter_size includes it.
3690 __ push(rbx);
3691 __ mov(rbx, index);
3692 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3693 __ load_resolved_reference_at_index(index, rbx);
3694 __ pop(rbx);
3695 __ push(index); // push appendix (MethodType, CallSite, etc.)
3696 __ bind(L_no_push);
3697 }
3698
3699 // load receiver if needed (after appendix is pushed so parameter size is correct)
3700 // Note: no return address pushed yet
3701 if (load_receiver) {
3702 __ movl(recv, flags);
3703 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3704 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3705 const int receiver_is_at_end = -1; // back off one slot to get receiver
3706 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3707 __ movptr(recv, recv_addr);
3708 __ verify_oop(recv);
3709 }
3710
3711 if (save_flags) {
3712 __ movl(rbcp, flags);
3713 }
3714
3715 // compute return type
3716 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3717 // Make sure we don't need to mask flags after the above shift
3718 ConstantPoolCacheEntry::verify_tos_state_shift();
3719 // load return address
3720 {
3721 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3722 ExternalAddress table(table_addr);
3723 LP64_ONLY(__ lea(rscratch1, table));
3724 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3725 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3726 }
3727
3728 // push return address
3729 __ push(flags);
3730
3731 // Restore flags value from the constant pool cache, and restore rsi
3732 // for later null checks. r13 is the bytecode pointer
3733 if (save_flags) {
3734 __ movl(flags, rbcp);
3735 __ restore_bcp();
3736 }
3737 }
3738
3739 void TemplateTable::invokevirtual_helper(Register index,
3740 Register recv,
3741 Register flags) {
3742 // Uses temporary registers rax, rdx
3743 assert_different_registers(index, recv, rax, rdx);
3744 assert(index == rbx, "");
3745 assert(recv == rcx, "");
3746
3747 // Test for an invoke of a final method
3748 Label notFinal;
3749 __ movl(rax, flags);
3750 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3751 __ jcc(Assembler::zero, notFinal);
3752
3753 const Register method = index; // method must be rbx
3754 assert(method == rbx,
3755 "Method* must be rbx for interpreter calling convention");
3756
3757 // do the call - the index is actually the method to call
3758 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3759
3760 // It's final, need a null check here!
3761 __ null_check(recv);
3762
3763 // profile this call
3764 __ profile_final_call(rax);
3765 __ profile_arguments_type(rax, method, rbcp, true);
3766
3767 __ jump_from_interpreted(method, rax);
3768
3769 __ bind(notFinal);
3770
3771 // get receiver klass
3772 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3773 __ load_klass(rax, recv);
3774
3775 // profile this call
3776 __ profile_virtual_call(rax, rlocals, rdx);
3777 // get target Method* & entry point
3778 __ lookup_virtual_method(rax, index, method);
3779 __ profile_called_method(method, rdx, rbcp);
3780
3781 __ profile_arguments_type(rdx, method, rbcp, true);
3782 __ jump_from_interpreted(method, rdx);
3783 }
3784
3785 void TemplateTable::invokevirtual(int byte_no) {
3786 transition(vtos, vtos);
3787 assert(byte_no == f2_byte, "use this argument");
3788 prepare_invoke(byte_no,
3789 rbx, // method or vtable index
3790 noreg, // unused itable index
3791 rcx, rdx); // recv, flags
3792
3793 // rbx: index
3794 // rcx: receiver
3795 // rdx: flags
3796
3797 invokevirtual_helper(rbx, rcx, rdx);
3798 }
3799
3800 void TemplateTable::invokespecial(int byte_no) {
3801 transition(vtos, vtos);
3802 assert(byte_no == f1_byte, "use this argument");
3803 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3804 rcx); // get receiver also for null check
3805 __ verify_oop(rcx);
3806 __ null_check(rcx);
3807 // do the call
3808 __ profile_call(rax);
3809 __ profile_arguments_type(rax, rbx, rbcp, false);
3810 __ jump_from_interpreted(rbx, rax);
3811 }
3812
3813 void TemplateTable::invokestatic(int byte_no) {
3814 transition(vtos, vtos);
3815 assert(byte_no == f1_byte, "use this argument");
3816 prepare_invoke(byte_no, rbx); // get f1 Method*
3817 // do the call
3818 __ profile_call(rax);
3819 __ profile_arguments_type(rax, rbx, rbcp, false);
3820 __ jump_from_interpreted(rbx, rax);
3821 }
3822
3823
3824 void TemplateTable::fast_invokevfinal(int byte_no) {
3825 transition(vtos, vtos);
3826 assert(byte_no == f2_byte, "use this argument");
3827 __ stop("fast_invokevfinal not used on x86");
3828 }
3829
3830
3831 void TemplateTable::invokeinterface(int byte_no) {
3832 transition(vtos, vtos);
3833 assert(byte_no == f1_byte, "use this argument");
3834 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3835 rcx, rdx); // recv, flags
3836
3837 // rax: interface klass (from f1)
3838 // rbx: itable index (from f2)
3839 // rcx: receiver
3840 // rdx: flags
3841
3842 // Special case of invokeinterface called for virtual method of
3843 // java.lang.Object. See cpCacheOop.cpp for details.
3844 // This code isn't produced by javac, but could be produced by
3845 // another compliant java compiler.
3846 Label notMethod;
3847 __ movl(rlocals, rdx);
3848 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3849
3850 __ jcc(Assembler::zero, notMethod);
3851
3852 invokevirtual_helper(rbx, rcx, rdx);
3853 __ bind(notMethod);
3854
3855 // Get receiver klass into rdx - also a null check
3856 __ restore_locals(); // restore r14
3857 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3858 __ load_klass(rdx, rcx);
3859
3860 // profile this call
3861 __ profile_virtual_call(rdx, rbcp, rlocals);
3862
3863 Label no_such_interface, no_such_method;
3864
3865 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3866 rdx, rax, rbx,
3867 // outputs: method, scan temp. reg
3868 rbx, rbcp,
3869 no_such_interface);
3870
3871 // rbx: Method* to call
3872 // rcx: receiver
3873 // Check for abstract method error
3874 // Note: This should be done more efficiently via a throw_abstract_method_error
3875 // interpreter entry point and a conditional jump to it in case of a null
3876 // method.
3877 __ testptr(rbx, rbx);
3878 __ jcc(Assembler::zero, no_such_method);
3879
3880 __ profile_called_method(rbx, rbcp, rdx);
3881 __ profile_arguments_type(rdx, rbx, rbcp, true);
3882
3883 // do the call
3884 // rcx: receiver
3885 // rbx,: Method*
3886 __ jump_from_interpreted(rbx, rdx);
3887 __ should_not_reach_here();
3888
3889 // exception handling code follows...
3890 // note: must restore interpreter registers to canonical
3891 // state for exception handling to work correctly!
3892
3893 __ bind(no_such_method);
3894 // throw exception
3895 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3896 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3897 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3898 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3899 // the call_VM checks for exception, so we should never return here.
3900 __ should_not_reach_here();
3901
3902 __ bind(no_such_interface);
3903 // throw exception
3904 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3905 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3906 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3907 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3908 InterpreterRuntime::throw_IncompatibleClassChangeError));
3909 // the call_VM checks for exception, so we should never return here.
3910 __ should_not_reach_here();
3911 }
3912
3913 void TemplateTable::invokehandle(int byte_no) {
3914 transition(vtos, vtos);
3915 assert(byte_no == f1_byte, "use this argument");
3916 const Register rbx_method = rbx;
3917 const Register rax_mtype = rax;
3918 const Register rcx_recv = rcx;
3919 const Register rdx_flags = rdx;
3920
3921 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3922 __ verify_method_ptr(rbx_method);
3923 __ verify_oop(rcx_recv);
3924 __ null_check(rcx_recv);
3925
3926 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3927 // rbx: MH.invokeExact_MT method (from f2)
3928
3929 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3930
3931 // FIXME: profile the LambdaForm also
3932 __ profile_final_call(rax);
3933 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3934
3935 __ jump_from_interpreted(rbx_method, rdx);
3936 }
3937
3938 void TemplateTable::invokedynamic(int byte_no) {
3939 transition(vtos, vtos);
3940 assert(byte_no == f1_byte, "use this argument");
3941
3942 const Register rbx_method = rbx;
3943 const Register rax_callsite = rax;
3944
3945 prepare_invoke(byte_no, rbx_method, rax_callsite);
3946
3947 // rax: CallSite object (from cpool->resolved_references[f1])
3948 // rbx: MH.linkToCallSite method (from f2)
3949
3950 // Note: rax_callsite is already pushed by prepare_invoke
3951
3952 // %%% should make a type profile for any invokedynamic that takes a ref argument
3953 // profile this call
3954 __ profile_call(rbcp);
3955 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3956
3957 __ verify_oop(rax_callsite);
3958
3959 __ jump_from_interpreted(rbx_method, rdx);
3960 }
3961
3962 //-----------------------------------------------------------------------------
3963 // Allocation
3964
3965 void TemplateTable::_new() {
3966 transition(vtos, atos);
3967 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3968 Label slow_case;
3969 Label slow_case_no_pop;
3970 Label done;
3971 Label initialize_header;
3972 Label initialize_object; // including clearing the fields
3973 Label allocate_shared;
3974
3975 __ get_cpool_and_tags(rcx, rax);
3976
3977 // Make sure the class we're about to instantiate has been resolved.
3978 // This is done before loading InstanceKlass to be consistent with the order
3979 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3980 const int tags_offset = Array<u1>::base_offset_in_bytes();
3981 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3982 __ jcc(Assembler::notEqual, slow_case_no_pop);
3983
3984 // get InstanceKlass
3985 __ load_resolved_klass_at_index(rcx, rdx, rcx);
3986 __ push(rcx); // save the contexts of klass for initializing the header
3987
3988 // make sure klass is initialized & doesn't have finalizer
3989 // make sure klass is fully initialized
3990 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3991 __ jcc(Assembler::notEqual, slow_case);
3992
3993 // get instance_size in InstanceKlass (scaled to a count of bytes)
3994 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3995 // test to see if it has a finalizer or is malformed in some way
3996 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3997 __ jcc(Assembler::notZero, slow_case);
3998
3999 //
4000 // Allocate the instance
4001 // 1) Try to allocate in the TLAB
4002 // 2) if fail and the object is large allocate in the shared Eden
4003 // 3) if the above fails (or is not applicable), go to a slow case
4004 // (creates a new TLAB, etc.)
4005
4006 const bool allow_shared_alloc =
4007 Universe::heap()->supports_inline_contig_alloc();
4008
4009 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4010 #ifndef _LP64
4011 if (UseTLAB || allow_shared_alloc) {
4012 __ get_thread(thread);
4013 }
4014 #endif // _LP64
4015
4016 if (UseTLAB) {
4017 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
4018 __ lea(rbx, Address(rax, rdx, Address::times_1));
4019 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
4020 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
4021 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
4022 if (ZeroTLAB) {
4023 // the fields have been already cleared
4024 __ jmp(initialize_header);
4025 } else {
4026 // initialize both the header and fields
4027 __ jmp(initialize_object);
4028 }
4029 }
4030
4031 // Allocation in the shared Eden, if allowed.
4032 //
4033 // rdx: instance size in bytes
4034 if (allow_shared_alloc) {
4035 __ bind(allocate_shared);
4036
4037 ExternalAddress heap_top((address)Universe::heap()->top_addr());
4038 ExternalAddress heap_end((address)Universe::heap()->end_addr());
4039
4040 Label retry;
4041 __ bind(retry);
4042 __ movptr(rax, heap_top);
4043 __ lea(rbx, Address(rax, rdx, Address::times_1));
4044 __ cmpptr(rbx, heap_end);
4045 __ jcc(Assembler::above, slow_case);
4046
4047 // Compare rax, with the top addr, and if still equal, store the new
4048 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
4049 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
4050 //
4051 // rax,: object begin
4052 // rbx,: object end
4053 // rdx: instance size in bytes
4054 __ locked_cmpxchgptr(rbx, heap_top);
4055
4056 // if someone beat us on the allocation, try again, otherwise continue
4057 __ jcc(Assembler::notEqual, retry);
4058
4059 __ incr_allocated_bytes(thread, rdx, 0);
4060 }
4061
4062 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
4063 // The object is initialized before the header. If the object size is
4064 // zero, go directly to the header initialization.
4065 __ bind(initialize_object);
4066 __ decrement(rdx, sizeof(oopDesc));
4067 __ jcc(Assembler::zero, initialize_header);
4068
4069 // Initialize topmost object field, divide rdx by 8, check if odd and
4070 // test if zero.
4071 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
4072 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4073
4074 // rdx must have been multiple of 8
4075 #ifdef ASSERT
4076 // make sure rdx was multiple of 8
4077 Label L;
4078 // Ignore partial flag stall after shrl() since it is debug VM
4079 __ jccb(Assembler::carryClear, L);
4080 __ stop("object size is not multiple of 2 - adjust this code");
4081 __ bind(L);
4082 // rdx must be > 0, no extra check needed here
4083 #endif
4084
4085 // initialize remaining object fields: rdx was a multiple of 8
4086 { Label loop;
4087 __ bind(loop);
4088 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4089 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4090 __ decrement(rdx);
4091 __ jcc(Assembler::notZero, loop);
4092 }
4093
4094 // initialize object header only.
4095 __ bind(initialize_header);
4096 if (UseBiasedLocking) {
4097 __ pop(rcx); // get saved klass back in the register.
4098 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4099 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4100 } else {
4101 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4102 (intptr_t)markOopDesc::prototype()); // header
4103 __ pop(rcx); // get saved klass back in the register.
4104 }
4105 #ifdef _LP64
4106 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4107 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
4108 #endif
4109 __ store_klass(rax, rcx); // klass
4110
4111 {
4112 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4113 // Trigger dtrace event for fastpath
4114 __ push(atos);
4115 __ call_VM_leaf(
4116 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4117 __ pop(atos);
4118 }
4119
4120 __ jmp(done);
4121 }
4122
4123 // slow case
4124 __ bind(slow_case);
4125 __ pop(rcx); // restore stack pointer to what it was when we came in.
4126 __ bind(slow_case_no_pop);
4127
4128 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4129 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4130
4131 __ get_constant_pool(rarg1);
4132 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4133 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4134 __ verify_oop(rax);
4135
4136 // continue
4137 __ bind(done);
4138 }
4139
4140 void TemplateTable::vdefault() {
4141 transition(vtos, qtos);
4142
4143 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4144 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4145
4146 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4147 __ get_constant_pool(rarg1);
4148
4149 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::vdefault),
4150 rarg1, rarg2);
4151 __ verify_oop(rax);
4152 }
4153
4154 void TemplateTable::newarray() {
4155 transition(itos, atos);
4156 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4157 __ load_unsigned_byte(rarg1, at_bcp(1));
4158 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4159 rarg1, rax);
4160 }
4161
4162 void TemplateTable::anewarray() {
4163 transition(itos, atos);
4164
4165 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4166 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4167
4168 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4169 __ get_constant_pool(rarg1);
4170 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4171 rarg1, rarg2, rax);
4172 }
4173
4174 void TemplateTable::arraylength() {
4175 transition(atos, itos);
4176 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4177 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4178 }
4179
4180 void TemplateTable::checkcast() {
4181 transition(atos, atos);
4182 Label done, is_null, ok_is_subtype, quicked, resolved;
4183 __ testptr(rax, rax); // object is in rax
4184 __ jcc(Assembler::zero, is_null);
4185
4186 // Get cpool & tags index
4187 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4188 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4189 // See if bytecode has already been quicked
4190 __ cmpb(Address(rdx, rbx,
4191 Address::times_1,
4192 Array<u1>::base_offset_in_bytes()),
4193 JVM_CONSTANT_Class);
4194 __ jcc(Assembler::equal, quicked);
4195 __ push(atos); // save receiver for result, and for GC
4196 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4197
4198 // vm_result_2 has metadata result
4199 #ifndef _LP64
4200 // borrow rdi from locals
4201 __ get_thread(rdi);
4202 __ get_vm_result_2(rax, rdi);
4203 __ restore_locals();
4204 #else
4205 __ get_vm_result_2(rax, r15_thread);
4206 #endif
4207
4208 __ pop_ptr(rdx); // restore receiver
4209 __ jmpb(resolved);
4210
4211 // Get superklass in rax and subklass in rbx
4212 __ bind(quicked);
4213 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4214 __ load_resolved_klass_at_index(rcx, rbx, rax);
4215
4216 __ bind(resolved);
4217 __ load_klass(rbx, rdx);
4218
4219 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4220 // Superklass in rax. Subklass in rbx.
4221 __ gen_subtype_check(rbx, ok_is_subtype);
4222
4223 // Come here on failure
4224 __ push_ptr(rdx);
4225 // object is at TOS
4226 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4227
4228 // Come here on success
4229 __ bind(ok_is_subtype);
4230 __ mov(rax, rdx); // Restore object in rdx
4231
4232 // Collect counts on whether this check-cast sees NULLs a lot or not.
4233 if (ProfileInterpreter) {
4234 __ jmp(done);
4235 __ bind(is_null);
4236 __ profile_null_seen(rcx);
4237 } else {
4238 __ bind(is_null); // same as 'done'
4239 }
4240 __ bind(done);
4241 }
4242
4243 void TemplateTable::instanceof() {
4244 transition(atos, itos);
4245 Label done, is_null, ok_is_subtype, quicked, resolved;
4246 __ testptr(rax, rax);
4247 __ jcc(Assembler::zero, is_null);
4248
4249 // Get cpool & tags index
4250 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4251 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4252 // See if bytecode has already been quicked
4253 __ cmpb(Address(rdx, rbx,
4254 Address::times_1,
4255 Array<u1>::base_offset_in_bytes()),
4256 JVM_CONSTANT_Class);
4257 __ jcc(Assembler::equal, quicked);
4258
4259 __ push(atos); // save receiver for result, and for GC
4260 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4261 // vm_result_2 has metadata result
4262
4263 #ifndef _LP64
4264 // borrow rdi from locals
4265 __ get_thread(rdi);
4266 __ get_vm_result_2(rax, rdi);
4267 __ restore_locals();
4268 #else
4269 __ get_vm_result_2(rax, r15_thread);
4270 #endif
4271
4272 __ pop_ptr(rdx); // restore receiver
4273 __ verify_oop(rdx);
4274 __ load_klass(rdx, rdx);
4275 __ jmpb(resolved);
4276
4277 // Get superklass in rax and subklass in rdx
4278 __ bind(quicked);
4279 __ load_klass(rdx, rax);
4280 __ load_resolved_klass_at_index(rcx, rbx, rax);
4281
4282 __ bind(resolved);
4283
4284 // Generate subtype check. Blows rcx, rdi
4285 // Superklass in rax. Subklass in rdx.
4286 __ gen_subtype_check(rdx, ok_is_subtype);
4287
4288 // Come here on failure
4289 __ xorl(rax, rax);
4290 __ jmpb(done);
4291 // Come here on success
4292 __ bind(ok_is_subtype);
4293 __ movl(rax, 1);
4294
4295 // Collect counts on whether this test sees NULLs a lot or not.
4296 if (ProfileInterpreter) {
4297 __ jmp(done);
4298 __ bind(is_null);
4299 __ profile_null_seen(rcx);
4300 } else {
4301 __ bind(is_null); // same as 'done'
4302 }
4303 __ bind(done);
4304 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4305 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4306 }
4307
4308 void TemplateTable::_vbox() {
4309 transition(qtos, atos);
4310
4311 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4312 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4313
4314 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4315 __ get_constant_pool(rarg1);
4316 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::vbox),
4317 rarg1, rarg2, rax);
4318 }
4319
4320 void TemplateTable::_vunbox() {
4321 transition(atos, qtos);
4322
4323 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4324 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4325
4326 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4327 __ get_constant_pool(rarg1);
4328 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::vunbox),
4329 rarg1, rarg2, rax);
4330 }
4331
4332 //----------------------------------------------------------------------------------------------------
4333 // Breakpoints
4334 void TemplateTable::_breakpoint() {
4335 // Note: We get here even if we are single stepping..
4336 // jbug insists on setting breakpoints at every bytecode
4337 // even if we are in single step mode.
4338
4339 transition(vtos, vtos);
4340
4341 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4342
4343 // get the unpatched byte code
4344 __ get_method(rarg);
4345 __ call_VM(noreg,
4346 CAST_FROM_FN_PTR(address,
4347 InterpreterRuntime::get_original_bytecode_at),
4348 rarg, rbcp);
4349 __ mov(rbx, rax); // why?
4350
4351 // post the breakpoint event
4352 __ get_method(rarg);
4353 __ call_VM(noreg,
4354 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4355 rarg, rbcp);
4356
4357 // complete the execution of original bytecode
4358 __ dispatch_only_normal(vtos);
4359 }
4360
4361 //-----------------------------------------------------------------------------
4362 // Exceptions
4363
4364 void TemplateTable::athrow() {
4365 transition(atos, vtos);
4366 __ null_check(rax);
4367 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4368 }
4369
4370 //-----------------------------------------------------------------------------
4371 // Synchronization
4372 //
4373 // Note: monitorenter & exit are symmetric routines; which is reflected
4374 // in the assembly code structure as well
4375 //
4376 // Stack layout:
4377 //
4378 // [expressions ] <--- rsp = expression stack top
4379 // ..
4380 // [expressions ]
4381 // [monitor entry] <--- monitor block top = expression stack bot
4382 // ..
4383 // [monitor entry]
4384 // [frame data ] <--- monitor block bot
4385 // ...
4386 // [saved rbp ] <--- rbp
4387 void TemplateTable::monitorenter() {
4388 transition(atos, vtos);
4389
4390 // check for NULL object
4391 __ null_check(rax);
4392
4393 const Address monitor_block_top(
4394 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4395 const Address monitor_block_bot(
4396 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4397 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4398
4399 Label allocated;
4400
4401 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4402 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4403 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4404
4405 // initialize entry pointer
4406 __ xorl(rmon, rmon); // points to free slot or NULL
4407
4408 // find a free slot in the monitor block (result in rmon)
4409 {
4410 Label entry, loop, exit;
4411 __ movptr(rtop, monitor_block_top); // points to current entry,
4412 // starting with top-most entry
4413 __ lea(rbot, monitor_block_bot); // points to word before bottom
4414 // of monitor block
4415 __ jmpb(entry);
4416
4417 __ bind(loop);
4418 // check if current entry is used
4419 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4420 // if not used then remember entry in rmon
4421 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4422 // check if current entry is for same object
4423 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4424 // if same object then stop searching
4425 __ jccb(Assembler::equal, exit);
4426 // otherwise advance to next entry
4427 __ addptr(rtop, entry_size);
4428 __ bind(entry);
4429 // check if bottom reached
4430 __ cmpptr(rtop, rbot);
4431 // if not at bottom then check this entry
4432 __ jcc(Assembler::notEqual, loop);
4433 __ bind(exit);
4434 }
4435
4436 __ testptr(rmon, rmon); // check if a slot has been found
4437 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4438
4439 // allocate one if there's no free slot
4440 {
4441 Label entry, loop;
4442 // 1. compute new pointers // rsp: old expression stack top
4443 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4444 __ subptr(rsp, entry_size); // move expression stack top
4445 __ subptr(rmon, entry_size); // move expression stack bottom
4446 __ mov(rtop, rsp); // set start value for copy loop
4447 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4448 __ jmp(entry);
4449 // 2. move expression stack contents
4450 __ bind(loop);
4451 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4452 // word from old location
4453 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4454 __ addptr(rtop, wordSize); // advance to next word
4455 __ bind(entry);
4456 __ cmpptr(rtop, rmon); // check if bottom reached
4457 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4458 // copy next word
4459 }
4460
4461 // call run-time routine
4462 // rmon: points to monitor entry
4463 __ bind(allocated);
4464
4465 // Increment bcp to point to the next bytecode, so exception
4466 // handling for async. exceptions work correctly.
4467 // The object has already been poped from the stack, so the
4468 // expression stack looks correct.
4469 __ increment(rbcp);
4470
4471 // store object
4472 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4473 __ lock_object(rmon);
4474
4475 // check to make sure this monitor doesn't cause stack overflow after locking
4476 __ save_bcp(); // in case of exception
4477 __ generate_stack_overflow_check(0);
4478
4479 // The bcp has already been incremented. Just need to dispatch to
4480 // next instruction.
4481 __ dispatch_next(vtos);
4482 }
4483
4484 void TemplateTable::monitorexit() {
4485 transition(atos, vtos);
4486
4487 // check for NULL object
4488 __ null_check(rax);
4489
4490 const Address monitor_block_top(
4491 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4492 const Address monitor_block_bot(
4493 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4494 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4495
4496 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4497 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4498
4499 Label found;
4500
4501 // find matching slot
4502 {
4503 Label entry, loop;
4504 __ movptr(rtop, monitor_block_top); // points to current entry,
4505 // starting with top-most entry
4506 __ lea(rbot, monitor_block_bot); // points to word before bottom
4507 // of monitor block
4508 __ jmpb(entry);
4509
4510 __ bind(loop);
4511 // check if current entry is for same object
4512 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4513 // if same object then stop searching
4514 __ jcc(Assembler::equal, found);
4515 // otherwise advance to next entry
4516 __ addptr(rtop, entry_size);
4517 __ bind(entry);
4518 // check if bottom reached
4519 __ cmpptr(rtop, rbot);
4520 // if not at bottom then check this entry
4521 __ jcc(Assembler::notEqual, loop);
4522 }
4523
4524 // error handling. Unlocking was not block-structured
4525 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4526 InterpreterRuntime::throw_illegal_monitor_state_exception));
4527 __ should_not_reach_here();
4528
4529 // call run-time routine
4530 __ bind(found);
4531 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4532 __ unlock_object(rtop);
4533 __ pop_ptr(rax); // discard object
4534 }
4535
4536 // Wide instructions
4537 void TemplateTable::wide() {
4538 transition(vtos, vtos);
4539 __ load_unsigned_byte(rbx, at_bcp(1));
4540 ExternalAddress wtable((address)Interpreter::_wentry_point);
4541 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4542 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4543 }
4544
4545 // Multi arrays
4546 void TemplateTable::multianewarray() {
4547 transition(vtos, atos);
4548
4549 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4550 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4551 // last dim is on top of stack; we want address of first one:
4552 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4553 // the latter wordSize to point to the beginning of the array.
4554 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4555 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4556 __ load_unsigned_byte(rbx, at_bcp(3));
4557 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4558 }