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 case BarrierSet::ShenandoahBarrierSet:
159 {
160 // flatten object address if needed
161 // We do it regardless of precise because we need the registers
162 if (obj.index() == noreg && obj.disp() == 0) {
163 if (obj.base() != rdx) {
164 __ movptr(rdx, obj.base());
165 }
166 } else {
167 __ lea(rdx, obj);
168 }
169
170 Register rtmp = LP64_ONLY(r8) NOT_LP64(rsi);
171 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
172
173 NOT_LP64(__ get_thread(rcx));
174 NOT_LP64(__ save_bcp());
175
176 if (!UseShenandoahGC || ShenandoahSATBBarrier) {
177 __ g1_write_barrier_pre(rdx /* obj */,
178 rbx /* pre_val */,
179 rthread /* thread */,
180 rtmp /* tmp */,
181 val != noreg /* tosca_live */,
182 false /* expand_call */);
183 }
184
185 if (val == noreg) {
186 __ store_heap_oop_null(Address(rdx, 0));
187 } else {
188 // For Shenandoah, make sure we only store refs into to-space.
189 oopDesc::bs()->interpreter_read_barrier(_masm, val);
190
191 // G1 barrier needs uncompressed oop for region cross check.
192 Register new_val = val;
193 if (UseCompressedOops) {
194 new_val = rbx;
195 __ movptr(new_val, val);
196 }
197
198 __ store_heap_oop(Address(rdx, 0), val);
199 if (UseG1GC) {
200 __ g1_write_barrier_post(rdx /* store_adr */,
201 new_val /* new_val */,
202 rthread /* thread */,
203 rtmp /* tmp */,
204 rbx /* tmp2 */);
205 } else {
206 assert(UseShenandoahGC, "sanity");
207 __ shenandoah_write_barrier_post(rdx /* store_adr */,
208 new_val /* new_val */,
209 rthread /* thread */,
210 rtmp /* tmp */,
211 rbx /* tmp2 */);
212 }
213 }
214 NOT_LP64( __ restore_bcp());
215 }
216 break;
217 #endif // INCLUDE_ALL_GCS
218 case BarrierSet::CardTableForRS:
219 case BarrierSet::CardTableExtension:
220 {
221 if (val == noreg) {
222 __ store_heap_oop_null(obj);
223 } else {
224 __ store_heap_oop(obj, val);
225 // flatten object address if needed
226 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
227 __ store_check(obj.base());
228 } else {
229 __ lea(rdx, obj);
230 __ store_check(rdx);
231 }
232 }
233 }
234 break;
235 case BarrierSet::ModRef:
236 if (val == noreg) {
237 __ store_heap_oop_null(obj);
238 } else {
239 __ store_heap_oop(obj, val);
240 }
241 break;
242 default :
243 ShouldNotReachHere();
244
245 }
246 }
247
248 Address TemplateTable::at_bcp(int offset) {
249 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
250 return Address(rbcp, offset);
251 }
252
253
254 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
255 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
256 int byte_no) {
257 if (!RewriteBytecodes) return;
258 Label L_patch_done;
259
260 switch (bc) {
261 case Bytecodes::_fast_aputfield:
262 case Bytecodes::_fast_bputfield:
263 case Bytecodes::_fast_zputfield:
264 case Bytecodes::_fast_cputfield:
265 case Bytecodes::_fast_dputfield:
266 case Bytecodes::_fast_fputfield:
267 case Bytecodes::_fast_iputfield:
268 case Bytecodes::_fast_lputfield:
269 case Bytecodes::_fast_sputfield:
270 {
271 // We skip bytecode quickening for putfield instructions when
272 // the put_code written to the constant pool cache is zero.
273 // This is required so that every execution of this instruction
274 // calls out to InterpreterRuntime::resolve_get_put to do
275 // additional, required work.
276 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
277 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
278 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
279 __ movl(bc_reg, bc);
280 __ cmpl(temp_reg, (int) 0);
281 __ jcc(Assembler::zero, L_patch_done); // don't patch
282 }
283 break;
284 default:
285 assert(byte_no == -1, "sanity");
286 // the pair bytecodes have already done the load.
287 if (load_bc_into_bc_reg) {
288 __ movl(bc_reg, bc);
289 }
290 }
291
292 if (JvmtiExport::can_post_breakpoint()) {
293 Label L_fast_patch;
294 // if a breakpoint is present we can't rewrite the stream directly
295 __ movzbl(temp_reg, at_bcp(0));
296 __ cmpl(temp_reg, Bytecodes::_breakpoint);
297 __ jcc(Assembler::notEqual, L_fast_patch);
298 __ get_method(temp_reg);
299 // Let breakpoint table handling rewrite to quicker bytecode
300 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
301 #ifndef ASSERT
302 __ jmpb(L_patch_done);
303 #else
304 __ jmp(L_patch_done);
305 #endif
306 __ bind(L_fast_patch);
307 }
308
309 #ifdef ASSERT
310 Label L_okay;
311 __ load_unsigned_byte(temp_reg, at_bcp(0));
312 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
313 __ jcc(Assembler::equal, L_okay);
314 __ cmpl(temp_reg, bc_reg);
315 __ jcc(Assembler::equal, L_okay);
316 __ stop("patching the wrong bytecode");
317 __ bind(L_okay);
318 #endif
319
320 // patch bytecode
321 __ movb(at_bcp(0), bc_reg);
322 __ bind(L_patch_done);
323 }
324 // Individual instructions
325
326
327 void TemplateTable::nop() {
328 transition(vtos, vtos);
329 // nothing to do
330 }
331
332 void TemplateTable::shouldnotreachhere() {
333 transition(vtos, vtos);
334 __ stop("shouldnotreachhere bytecode");
335 }
336
337 void TemplateTable::aconst_null() {
338 transition(vtos, atos);
339 __ xorl(rax, rax);
340 }
341
342 void TemplateTable::iconst(int value) {
343 transition(vtos, itos);
344 if (value == 0) {
345 __ xorl(rax, rax);
346 } else {
347 __ movl(rax, value);
348 }
349 }
350
351 void TemplateTable::lconst(int value) {
352 transition(vtos, ltos);
353 if (value == 0) {
354 __ xorl(rax, rax);
355 } else {
356 __ movl(rax, value);
357 }
358 #ifndef _LP64
359 assert(value >= 0, "check this code");
360 __ xorptr(rdx, rdx);
361 #endif
362 }
363
364
365
366 void TemplateTable::fconst(int value) {
367 transition(vtos, ftos);
368 if (UseSSE >= 1) {
369 static float one = 1.0f, two = 2.0f;
370 switch (value) {
371 case 0:
372 __ xorps(xmm0, xmm0);
373 break;
374 case 1:
375 __ movflt(xmm0, ExternalAddress((address) &one));
376 break;
377 case 2:
378 __ movflt(xmm0, ExternalAddress((address) &two));
379 break;
380 default:
381 ShouldNotReachHere();
382 break;
383 }
384 } else {
385 #ifdef _LP64
386 ShouldNotReachHere();
387 #else
388 if (value == 0) { __ fldz();
389 } else if (value == 1) { __ fld1();
390 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
391 } else { ShouldNotReachHere();
392 }
393 #endif // _LP64
394 }
395 }
396
397 void TemplateTable::dconst(int value) {
398 transition(vtos, dtos);
399 if (UseSSE >= 2) {
400 static double one = 1.0;
401 switch (value) {
402 case 0:
403 __ xorpd(xmm0, xmm0);
404 break;
405 case 1:
406 __ movdbl(xmm0, ExternalAddress((address) &one));
407 break;
408 default:
409 ShouldNotReachHere();
410 break;
411 }
412 } else {
413 #ifdef _LP64
414 ShouldNotReachHere();
415 #else
416 if (value == 0) { __ fldz();
417 } else if (value == 1) { __ fld1();
418 } else { ShouldNotReachHere();
419 }
420 #endif
421 }
422 }
423
424 void TemplateTable::bipush() {
425 transition(vtos, itos);
426 __ load_signed_byte(rax, at_bcp(1));
427 }
428
429 void TemplateTable::sipush() {
430 transition(vtos, itos);
431 __ load_unsigned_short(rax, at_bcp(1));
432 __ bswapl(rax);
433 __ sarl(rax, 16);
434 }
435
436 void TemplateTable::ldc(bool wide) {
437 transition(vtos, vtos);
438 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
439 Label call_ldc, notFloat, notClass, Done;
440
441 if (wide) {
442 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
443 } else {
444 __ load_unsigned_byte(rbx, at_bcp(1));
445 }
446
447 __ get_cpool_and_tags(rcx, rax);
448 const int base_offset = ConstantPool::header_size() * wordSize;
449 const int tags_offset = Array<u1>::base_offset_in_bytes();
450
451 // get type
452 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
453
454 // unresolved class - get the resolved class
455 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
456 __ jccb(Assembler::equal, call_ldc);
457
458 // unresolved class in error state - call into runtime to throw the error
459 // from the first resolution attempt
460 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
461 __ jccb(Assembler::equal, call_ldc);
462
463 // resolved class - need to call vm to get java mirror of the class
464 __ cmpl(rdx, JVM_CONSTANT_Class);
465 __ jcc(Assembler::notEqual, notClass);
466
467 __ bind(call_ldc);
468
469 __ movl(rarg, wide);
470 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
471
472 __ push(atos);
473 __ jmp(Done);
474
475 __ bind(notClass);
476 __ cmpl(rdx, JVM_CONSTANT_Float);
477 __ jccb(Assembler::notEqual, notFloat);
478
479 // ftos
480 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
481 __ push(ftos);
482 __ jmp(Done);
483
484 __ bind(notFloat);
485 #ifdef ASSERT
486 {
487 Label L;
488 __ cmpl(rdx, JVM_CONSTANT_Integer);
489 __ jcc(Assembler::equal, L);
490 // String and Object are rewritten to fast_aldc
491 __ stop("unexpected tag type in ldc");
492 __ bind(L);
493 }
494 #endif
495 // itos JVM_CONSTANT_Integer only
496 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
497 __ push(itos);
498 __ bind(Done);
499 }
500
501 // Fast path for caching oop constants.
502 void TemplateTable::fast_aldc(bool wide) {
503 transition(vtos, atos);
504
505 Register result = rax;
506 Register tmp = rdx;
507 int index_size = wide ? sizeof(u2) : sizeof(u1);
508
509 Label resolved;
510
511 // We are resolved if the resolved reference cache entry contains a
512 // non-null object (String, MethodType, etc.)
513 assert_different_registers(result, tmp);
514 __ get_cache_index_at_bcp(tmp, 1, index_size);
515 __ load_resolved_reference_at_index(result, tmp);
516 __ testl(result, result);
517 __ jcc(Assembler::notZero, resolved);
518
519 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
520
521 // first time invocation - must resolve first
522 __ movl(tmp, (int)bytecode());
523 __ call_VM(result, entry, tmp);
524
525 __ bind(resolved);
526
527 if (VerifyOops) {
528 __ verify_oop(result);
529 }
530 }
531
532 void TemplateTable::ldc2_w() {
533 transition(vtos, vtos);
534 Label Long, Done;
535 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
536
537 __ get_cpool_and_tags(rcx, rax);
538 const int base_offset = ConstantPool::header_size() * wordSize;
539 const int tags_offset = Array<u1>::base_offset_in_bytes();
540
541 // get type
542 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
543 JVM_CONSTANT_Double);
544 __ jccb(Assembler::notEqual, Long);
545
546 // dtos
547 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
548 __ push(dtos);
549
550 __ jmpb(Done);
551 __ bind(Long);
552
553 // ltos
554 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
555 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
556 __ push(ltos);
557
558 __ bind(Done);
559 }
560
561 void TemplateTable::locals_index(Register reg, int offset) {
562 __ load_unsigned_byte(reg, at_bcp(offset));
563 __ negptr(reg);
564 }
565
566 void TemplateTable::iload() {
567 iload_internal();
568 }
569
570 void TemplateTable::nofast_iload() {
571 iload_internal(may_not_rewrite);
572 }
573
574 void TemplateTable::iload_internal(RewriteControl rc) {
575 transition(vtos, itos);
576 if (RewriteFrequentPairs && rc == may_rewrite) {
577 Label rewrite, done;
578 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
579 LP64_ONLY(assert(rbx != bc, "register damaged"));
580
581 // get next byte
582 __ load_unsigned_byte(rbx,
583 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
584 // if _iload, wait to rewrite to iload2. We only want to rewrite the
585 // last two iloads in a pair. Comparing against fast_iload means that
586 // the next bytecode is neither an iload or a caload, and therefore
587 // an iload pair.
588 __ cmpl(rbx, Bytecodes::_iload);
589 __ jcc(Assembler::equal, done);
590
591 __ cmpl(rbx, Bytecodes::_fast_iload);
592 __ movl(bc, Bytecodes::_fast_iload2);
593
594 __ jccb(Assembler::equal, rewrite);
595
596 // if _caload, rewrite to fast_icaload
597 __ cmpl(rbx, Bytecodes::_caload);
598 __ movl(bc, Bytecodes::_fast_icaload);
599 __ jccb(Assembler::equal, rewrite);
600
601 // rewrite so iload doesn't check again.
602 __ movl(bc, Bytecodes::_fast_iload);
603
604 // rewrite
605 // bc: fast bytecode
606 __ bind(rewrite);
607 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
608 __ bind(done);
609 }
610
611 // Get the local value into tos
612 locals_index(rbx);
613 __ movl(rax, iaddress(rbx));
614 }
615
616 void TemplateTable::fast_iload2() {
617 transition(vtos, itos);
618 locals_index(rbx);
619 __ movl(rax, iaddress(rbx));
620 __ push(itos);
621 locals_index(rbx, 3);
622 __ movl(rax, iaddress(rbx));
623 }
624
625 void TemplateTable::fast_iload() {
626 transition(vtos, itos);
627 locals_index(rbx);
628 __ movl(rax, iaddress(rbx));
629 }
630
631 void TemplateTable::lload() {
632 transition(vtos, ltos);
633 locals_index(rbx);
634 __ movptr(rax, laddress(rbx));
635 NOT_LP64(__ movl(rdx, haddress(rbx)));
636 }
637
638 void TemplateTable::fload() {
639 transition(vtos, ftos);
640 locals_index(rbx);
641 __ load_float(faddress(rbx));
642 }
643
644 void TemplateTable::dload() {
645 transition(vtos, dtos);
646 locals_index(rbx);
647 __ load_double(daddress(rbx));
648 }
649
650 void TemplateTable::aload() {
651 transition(vtos, atos);
652 locals_index(rbx);
653 __ movptr(rax, aaddress(rbx));
654 }
655
656 void TemplateTable::locals_index_wide(Register reg) {
657 __ load_unsigned_short(reg, at_bcp(2));
658 __ bswapl(reg);
659 __ shrl(reg, 16);
660 __ negptr(reg);
661 }
662
663 void TemplateTable::wide_iload() {
664 transition(vtos, itos);
665 locals_index_wide(rbx);
666 __ movl(rax, iaddress(rbx));
667 }
668
669 void TemplateTable::wide_lload() {
670 transition(vtos, ltos);
671 locals_index_wide(rbx);
672 __ movptr(rax, laddress(rbx));
673 NOT_LP64(__ movl(rdx, haddress(rbx)));
674 }
675
676 void TemplateTable::wide_fload() {
677 transition(vtos, ftos);
678 locals_index_wide(rbx);
679 __ load_float(faddress(rbx));
680 }
681
682 void TemplateTable::wide_dload() {
683 transition(vtos, dtos);
684 locals_index_wide(rbx);
685 __ load_double(daddress(rbx));
686 }
687
688 void TemplateTable::wide_aload() {
689 transition(vtos, atos);
690 locals_index_wide(rbx);
691 __ movptr(rax, aaddress(rbx));
692 }
693
694 void TemplateTable::index_check(Register array, Register index) {
695 // Pop ptr into array
696 __ pop_ptr(array);
697 index_check_without_pop(array, index);
698 }
699
700 void TemplateTable::index_check_without_pop(Register array, Register index) {
701 // destroys rbx
702 // check array
703 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
704 // sign extend index for use by indexed load
705 __ movl2ptr(index, index);
706 // check index
707 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
708 if (index != rbx) {
709 // ??? convention: move aberrant index into rbx for exception message
710 assert(rbx != array, "different registers");
711 __ movl(rbx, index);
712 }
713 __ jump_cc(Assembler::aboveEqual,
714 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
715 }
716
717
718 void TemplateTable::iaload() {
719 transition(itos, itos);
720 // rax: index
721 // rdx: array
722 index_check(rdx, rax); // kills rbx
723 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
724 __ movl(rax, Address(rdx, rax,
725 Address::times_4,
726 arrayOopDesc::base_offset_in_bytes(T_INT)));
727 }
728
729 void TemplateTable::laload() {
730 transition(itos, ltos);
731 // rax: index
732 // rdx: array
733 index_check(rdx, rax); // kills rbx
734 NOT_LP64(__ mov(rbx, rax));
735 // rbx,: index
736 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
737 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
738 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
739 }
740
741
742
743 void TemplateTable::faload() {
744 transition(itos, ftos);
745 // rax: index
746 // rdx: array
747 index_check(rdx, rax); // kills rbx
748 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
749 __ load_float(Address(rdx, rax,
750 Address::times_4,
751 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
752 }
753
754 void TemplateTable::daload() {
755 transition(itos, dtos);
756 // rax: index
757 // rdx: array
758 index_check(rdx, rax); // kills rbx
759 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
760 __ load_double(Address(rdx, rax,
761 Address::times_8,
762 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
763 }
764
765 void TemplateTable::aaload() {
766 transition(itos, atos);
767 // rax: index
768 // rdx: array
769 index_check(rdx, rax); // kills rbx
770 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
771 __ load_heap_oop(rax, Address(rdx, rax,
772 UseCompressedOops ? Address::times_4 : Address::times_ptr,
773 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
774 }
775
776 void TemplateTable::baload() {
777 transition(itos, itos);
778 // rax: index
779 // rdx: array
780 index_check(rdx, rax); // kills rbx
781 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
782 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
783 }
784
785 void TemplateTable::caload() {
786 transition(itos, itos);
787 // rax: index
788 // rdx: array
789 index_check(rdx, rax); // kills rbx
790 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
791 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
792 }
793
794 // iload followed by caload frequent pair
795 void TemplateTable::fast_icaload() {
796 transition(vtos, itos);
797 // load index out of locals
798 locals_index(rbx);
799 __ movl(rax, iaddress(rbx));
800
801 // rax: index
802 // rdx: array
803 index_check(rdx, rax); // kills rbx
804 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
805 __ load_unsigned_short(rax,
806 Address(rdx, rax,
807 Address::times_2,
808 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
809 }
810
811
812 void TemplateTable::saload() {
813 transition(itos, itos);
814 // rax: index
815 // rdx: array
816 index_check(rdx, rax); // kills rbx
817 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
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::wide_istore() {
959 transition(vtos, vtos);
960 __ pop_i();
961 locals_index_wide(rbx);
962 __ movl(iaddress(rbx), rax);
963 }
964
965 void TemplateTable::wide_lstore() {
966 transition(vtos, vtos);
967 NOT_LP64(__ pop_l(rax, rdx));
968 LP64_ONLY(__ pop_l());
969 locals_index_wide(rbx);
970 __ movptr(laddress(rbx), rax);
971 NOT_LP64(__ movl(haddress(rbx), rdx));
972 }
973
974 void TemplateTable::wide_fstore() {
975 #ifdef _LP64
976 transition(vtos, vtos);
977 __ pop_f(xmm0);
978 locals_index_wide(rbx);
979 __ movflt(faddress(rbx), xmm0);
980 #else
981 wide_istore();
982 #endif
983 }
984
985 void TemplateTable::wide_dstore() {
986 #ifdef _LP64
987 transition(vtos, vtos);
988 __ pop_d(xmm0);
989 locals_index_wide(rbx);
990 __ movdbl(daddress(rbx), xmm0);
991 #else
992 wide_lstore();
993 #endif
994 }
995
996 void TemplateTable::wide_astore() {
997 transition(vtos, vtos);
998 __ pop_ptr(rax);
999 locals_index_wide(rbx);
1000 __ movptr(aaddress(rbx), rax);
1001 }
1002
1003 void TemplateTable::iastore() {
1004 transition(itos, vtos);
1005 __ pop_i(rbx);
1006 // rax: value
1007 // rbx: index
1008 // rdx: array
1009 index_check(rdx, rbx); // prefer index in rbx
1010 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1011 __ movl(Address(rdx, rbx,
1012 Address::times_4,
1013 arrayOopDesc::base_offset_in_bytes(T_INT)),
1014 rax);
1015 }
1016
1017 void TemplateTable::lastore() {
1018 transition(ltos, vtos);
1019 __ pop_i(rbx);
1020 // rax,: low(value)
1021 // rcx: array
1022 // rdx: high(value)
1023 index_check(rcx, rbx); // prefer index in rbx,
1024 // rbx,: index
1025 oopDesc::bs()->interpreter_write_barrier(_masm, rcx);
1026 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1027 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1028 }
1029
1030
1031 void TemplateTable::fastore() {
1032 transition(ftos, vtos);
1033 __ pop_i(rbx);
1034 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1035 // rbx: index
1036 // rdx: array
1037 index_check(rdx, rbx); // prefer index in rbx
1038 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1039 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1040 }
1041
1042 void TemplateTable::dastore() {
1043 transition(dtos, vtos);
1044 __ pop_i(rbx);
1045 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1046 // rbx: index
1047 // rdx: array
1048 index_check(rdx, rbx); // prefer index in rbx
1049 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1050 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1051 }
1052
1053 void TemplateTable::aastore() {
1054 Label is_null, ok_is_subtype, done;
1055 transition(vtos, vtos);
1056 // stack: ..., array, index, value
1057 __ movptr(rax, at_tos()); // value
1058 __ movl(rcx, at_tos_p1()); // index
1059 __ movptr(rdx, at_tos_p2()); // array
1060
1061 Address element_address(rdx, rcx,
1062 UseCompressedOops? Address::times_4 : Address::times_ptr,
1063 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1064
1065 index_check_without_pop(rdx, rcx); // kills rbx
1066 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1067 __ testptr(rax, rax);
1068 __ jcc(Assembler::zero, is_null);
1069
1070 // Move subklass into rbx
1071 __ load_klass(rbx, rax);
1072 // Move superklass into rax
1073 __ load_klass(rax, rdx);
1074 __ movptr(rax, Address(rax,
1075 ObjArrayKlass::element_klass_offset()));
1076 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1077 __ lea(rdx, element_address);
1078
1079 // Generate subtype check. Blows rcx, rdi
1080 // Superklass in rax. Subklass in rbx.
1081 __ gen_subtype_check(rbx, ok_is_subtype);
1082
1083 // Come here on failure
1084 // object is at TOS
1085 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1086
1087 // Come here on success
1088 __ bind(ok_is_subtype);
1089
1090 // Get the value we will store
1091 __ movptr(rax, at_tos());
1092 // Now store using the appropriate barrier
1093 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1094 __ jmp(done);
1095
1096 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1097 __ bind(is_null);
1098 __ profile_null_seen(rbx);
1099
1100 // Store a NULL
1101 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1102
1103 // Pop stack arguments
1104 __ bind(done);
1105 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1106 }
1107
1108 void TemplateTable::bastore() {
1109 transition(itos, vtos);
1110 __ pop_i(rbx);
1111 // rax: value
1112 // rbx: index
1113 // rdx: array
1114 index_check(rdx, rbx); // prefer index in rbx
1115 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1116 // Need to check whether array is boolean or byte
1117 // since both types share the bastore bytecode.
1118 __ load_klass(rcx, rdx);
1119 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1120 int diffbit = Klass::layout_helper_boolean_diffbit();
1121 __ testl(rcx, diffbit);
1122 Label L_skip;
1123 __ jccb(Assembler::zero, L_skip);
1124 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1125 __ bind(L_skip);
1126 __ movb(Address(rdx, rbx,
1127 Address::times_1,
1128 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1129 rax);
1130 }
1131
1132 void TemplateTable::castore() {
1133 transition(itos, vtos);
1134 __ pop_i(rbx);
1135 // rax: value
1136 // rbx: index
1137 // rdx: array
1138 index_check(rdx, rbx); // prefer index in rbx
1139 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1140 __ movw(Address(rdx, rbx,
1141 Address::times_2,
1142 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1143 rax);
1144 }
1145
1146
1147 void TemplateTable::sastore() {
1148 castore();
1149 }
1150
1151 void TemplateTable::istore(int n) {
1152 transition(itos, vtos);
1153 __ movl(iaddress(n), rax);
1154 }
1155
1156 void TemplateTable::lstore(int n) {
1157 transition(ltos, vtos);
1158 __ movptr(laddress(n), rax);
1159 NOT_LP64(__ movptr(haddress(n), rdx));
1160 }
1161
1162 void TemplateTable::fstore(int n) {
1163 transition(ftos, vtos);
1164 __ store_float(faddress(n));
1165 }
1166
1167 void TemplateTable::dstore(int n) {
1168 transition(dtos, vtos);
1169 __ store_double(daddress(n));
1170 }
1171
1172
1173 void TemplateTable::astore(int n) {
1174 transition(vtos, vtos);
1175 __ pop_ptr(rax);
1176 __ movptr(aaddress(n), rax);
1177 }
1178
1179 void TemplateTable::pop() {
1180 transition(vtos, vtos);
1181 __ addptr(rsp, Interpreter::stackElementSize);
1182 }
1183
1184 void TemplateTable::pop2() {
1185 transition(vtos, vtos);
1186 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1187 }
1188
1189
1190 void TemplateTable::dup() {
1191 transition(vtos, vtos);
1192 __ load_ptr(0, rax);
1193 __ push_ptr(rax);
1194 // stack: ..., a, a
1195 }
1196
1197 void TemplateTable::dup_x1() {
1198 transition(vtos, vtos);
1199 // stack: ..., a, b
1200 __ load_ptr( 0, rax); // load b
1201 __ load_ptr( 1, rcx); // load a
1202 __ store_ptr(1, rax); // store b
1203 __ store_ptr(0, rcx); // store a
1204 __ push_ptr(rax); // push b
1205 // stack: ..., b, a, b
1206 }
1207
1208 void TemplateTable::dup_x2() {
1209 transition(vtos, vtos);
1210 // stack: ..., a, b, c
1211 __ load_ptr( 0, rax); // load c
1212 __ load_ptr( 2, rcx); // load a
1213 __ store_ptr(2, rax); // store c in a
1214 __ push_ptr(rax); // push c
1215 // stack: ..., c, b, c, c
1216 __ load_ptr( 2, rax); // load b
1217 __ store_ptr(2, rcx); // store a in b
1218 // stack: ..., c, a, c, c
1219 __ store_ptr(1, rax); // store b in c
1220 // stack: ..., c, a, b, c
1221 }
1222
1223 void TemplateTable::dup2() {
1224 transition(vtos, vtos);
1225 // stack: ..., a, b
1226 __ load_ptr(1, rax); // load a
1227 __ push_ptr(rax); // push a
1228 __ load_ptr(1, rax); // load b
1229 __ push_ptr(rax); // push b
1230 // stack: ..., a, b, a, b
1231 }
1232
1233
1234 void TemplateTable::dup2_x1() {
1235 transition(vtos, vtos);
1236 // stack: ..., a, b, c
1237 __ load_ptr( 0, rcx); // load c
1238 __ load_ptr( 1, rax); // load b
1239 __ push_ptr(rax); // push b
1240 __ push_ptr(rcx); // push c
1241 // stack: ..., a, b, c, b, c
1242 __ store_ptr(3, rcx); // store c in b
1243 // stack: ..., a, c, c, b, c
1244 __ load_ptr( 4, rcx); // load a
1245 __ store_ptr(2, rcx); // store a in 2nd c
1246 // stack: ..., a, c, a, b, c
1247 __ store_ptr(4, rax); // store b in a
1248 // stack: ..., b, c, a, b, c
1249 }
1250
1251 void TemplateTable::dup2_x2() {
1252 transition(vtos, vtos);
1253 // stack: ..., a, b, c, d
1254 __ load_ptr( 0, rcx); // load d
1255 __ load_ptr( 1, rax); // load c
1256 __ push_ptr(rax); // push c
1257 __ push_ptr(rcx); // push d
1258 // stack: ..., a, b, c, d, c, d
1259 __ load_ptr( 4, rax); // load b
1260 __ store_ptr(2, rax); // store b in d
1261 __ store_ptr(4, rcx); // store d in b
1262 // stack: ..., a, d, c, b, c, d
1263 __ load_ptr( 5, rcx); // load a
1264 __ load_ptr( 3, rax); // load c
1265 __ store_ptr(3, rcx); // store a in c
1266 __ store_ptr(5, rax); // store c in a
1267 // stack: ..., c, d, a, b, c, d
1268 }
1269
1270 void TemplateTable::swap() {
1271 transition(vtos, vtos);
1272 // stack: ..., a, b
1273 __ load_ptr( 1, rcx); // load a
1274 __ load_ptr( 0, rax); // load b
1275 __ store_ptr(0, rcx); // store a in b
1276 __ store_ptr(1, rax); // store b in a
1277 // stack: ..., b, a
1278 }
1279
1280 void TemplateTable::iop2(Operation op) {
1281 transition(itos, itos);
1282 switch (op) {
1283 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1284 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1285 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1286 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1287 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1288 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1289 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1290 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1291 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1292 default : ShouldNotReachHere();
1293 }
1294 }
1295
1296 void TemplateTable::lop2(Operation op) {
1297 transition(ltos, ltos);
1298 #ifdef _LP64
1299 switch (op) {
1300 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1301 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1302 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1303 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1304 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1305 default : ShouldNotReachHere();
1306 }
1307 #else
1308 __ pop_l(rbx, rcx);
1309 switch (op) {
1310 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1311 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1312 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1313 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1314 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1315 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1316 default : ShouldNotReachHere();
1317 }
1318 #endif
1319 }
1320
1321 void TemplateTable::idiv() {
1322 transition(itos, itos);
1323 __ movl(rcx, rax);
1324 __ pop_i(rax);
1325 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1326 // they are not equal, one could do a normal division (no correction
1327 // needed), which may speed up this implementation for the common case.
1328 // (see also JVM spec., p.243 & p.271)
1329 __ corrected_idivl(rcx);
1330 }
1331
1332 void TemplateTable::irem() {
1333 transition(itos, itos);
1334 __ movl(rcx, rax);
1335 __ pop_i(rax);
1336 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1337 // they are not equal, one could do a normal division (no correction
1338 // needed), which may speed up this implementation for the common case.
1339 // (see also JVM spec., p.243 & p.271)
1340 __ corrected_idivl(rcx);
1341 __ movl(rax, rdx);
1342 }
1343
1344 void TemplateTable::lmul() {
1345 transition(ltos, ltos);
1346 #ifdef _LP64
1347 __ pop_l(rdx);
1348 __ imulq(rax, rdx);
1349 #else
1350 __ pop_l(rbx, rcx);
1351 __ push(rcx); __ push(rbx);
1352 __ push(rdx); __ push(rax);
1353 __ lmul(2 * wordSize, 0);
1354 __ addptr(rsp, 4 * wordSize); // take off temporaries
1355 #endif
1356 }
1357
1358 void TemplateTable::ldiv() {
1359 transition(ltos, ltos);
1360 #ifdef _LP64
1361 __ mov(rcx, rax);
1362 __ pop_l(rax);
1363 // generate explicit div0 check
1364 __ testq(rcx, rcx);
1365 __ jump_cc(Assembler::zero,
1366 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1367 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1368 // they are not equal, one could do a normal division (no correction
1369 // needed), which may speed up this implementation for the common case.
1370 // (see also JVM spec., p.243 & p.271)
1371 __ corrected_idivq(rcx); // kills rbx
1372 #else
1373 __ pop_l(rbx, rcx);
1374 __ push(rcx); __ push(rbx);
1375 __ push(rdx); __ push(rax);
1376 // check if y = 0
1377 __ orl(rax, rdx);
1378 __ jump_cc(Assembler::zero,
1379 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1380 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1381 __ addptr(rsp, 4 * wordSize); // take off temporaries
1382 #endif
1383 }
1384
1385 void TemplateTable::lrem() {
1386 transition(ltos, ltos);
1387 #ifdef _LP64
1388 __ mov(rcx, rax);
1389 __ pop_l(rax);
1390 __ testq(rcx, rcx);
1391 __ jump_cc(Assembler::zero,
1392 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1393 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1394 // they are not equal, one could do a normal division (no correction
1395 // needed), which may speed up this implementation for the common case.
1396 // (see also JVM spec., p.243 & p.271)
1397 __ corrected_idivq(rcx); // kills rbx
1398 __ mov(rax, rdx);
1399 #else
1400 __ pop_l(rbx, rcx);
1401 __ push(rcx); __ push(rbx);
1402 __ push(rdx); __ push(rax);
1403 // check if y = 0
1404 __ orl(rax, rdx);
1405 __ jump_cc(Assembler::zero,
1406 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1407 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1408 __ addptr(rsp, 4 * wordSize);
1409 #endif
1410 }
1411
1412 void TemplateTable::lshl() {
1413 transition(itos, ltos);
1414 __ movl(rcx, rax); // get shift count
1415 #ifdef _LP64
1416 __ pop_l(rax); // get shift value
1417 __ shlq(rax);
1418 #else
1419 __ pop_l(rax, rdx); // get shift value
1420 __ lshl(rdx, rax);
1421 #endif
1422 }
1423
1424 void TemplateTable::lshr() {
1425 #ifdef _LP64
1426 transition(itos, ltos);
1427 __ movl(rcx, rax); // get shift count
1428 __ pop_l(rax); // get shift value
1429 __ sarq(rax);
1430 #else
1431 transition(itos, ltos);
1432 __ mov(rcx, rax); // get shift count
1433 __ pop_l(rax, rdx); // get shift value
1434 __ lshr(rdx, rax, true);
1435 #endif
1436 }
1437
1438 void TemplateTable::lushr() {
1439 transition(itos, ltos);
1440 #ifdef _LP64
1441 __ movl(rcx, rax); // get shift count
1442 __ pop_l(rax); // get shift value
1443 __ shrq(rax);
1444 #else
1445 __ mov(rcx, rax); // get shift count
1446 __ pop_l(rax, rdx); // get shift value
1447 __ lshr(rdx, rax);
1448 #endif
1449 }
1450
1451 void TemplateTable::fop2(Operation op) {
1452 transition(ftos, ftos);
1453
1454 if (UseSSE >= 1) {
1455 switch (op) {
1456 case add:
1457 __ addss(xmm0, at_rsp());
1458 __ addptr(rsp, Interpreter::stackElementSize);
1459 break;
1460 case sub:
1461 __ movflt(xmm1, xmm0);
1462 __ pop_f(xmm0);
1463 __ subss(xmm0, xmm1);
1464 break;
1465 case mul:
1466 __ mulss(xmm0, at_rsp());
1467 __ addptr(rsp, Interpreter::stackElementSize);
1468 break;
1469 case div:
1470 __ movflt(xmm1, xmm0);
1471 __ pop_f(xmm0);
1472 __ divss(xmm0, xmm1);
1473 break;
1474 case rem:
1475 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1476 // modulo operation. The frem method calls the function
1477 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1478 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1479 // (signalling or quiet) is returned.
1480 //
1481 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1482 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1483 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1484 // The fprem instruction used on x86_32 is functionally equivalent to
1485 // SharedRuntime::frem in that it returns a NaN.
1486 #ifdef _LP64
1487 __ movflt(xmm1, xmm0);
1488 __ pop_f(xmm0);
1489 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1490 #else
1491 __ push_f(xmm0);
1492 __ pop_f();
1493 __ fld_s(at_rsp());
1494 __ fremr(rax);
1495 __ f2ieee();
1496 __ pop(rax); // pop second operand off the stack
1497 __ push_f();
1498 __ pop_f(xmm0);
1499 #endif
1500 break;
1501 default:
1502 ShouldNotReachHere();
1503 break;
1504 }
1505 } else {
1506 #ifdef _LP64
1507 ShouldNotReachHere();
1508 #else
1509 switch (op) {
1510 case add: __ fadd_s (at_rsp()); break;
1511 case sub: __ fsubr_s(at_rsp()); break;
1512 case mul: __ fmul_s (at_rsp()); break;
1513 case div: __ fdivr_s(at_rsp()); break;
1514 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1515 default : ShouldNotReachHere();
1516 }
1517 __ f2ieee();
1518 __ pop(rax); // pop second operand off the stack
1519 #endif // _LP64
1520 }
1521 }
1522
1523 void TemplateTable::dop2(Operation op) {
1524 transition(dtos, dtos);
1525 if (UseSSE >= 2) {
1526 switch (op) {
1527 case add:
1528 __ addsd(xmm0, at_rsp());
1529 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1530 break;
1531 case sub:
1532 __ movdbl(xmm1, xmm0);
1533 __ pop_d(xmm0);
1534 __ subsd(xmm0, xmm1);
1535 break;
1536 case mul:
1537 __ mulsd(xmm0, at_rsp());
1538 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1539 break;
1540 case div:
1541 __ movdbl(xmm1, xmm0);
1542 __ pop_d(xmm0);
1543 __ divsd(xmm0, xmm1);
1544 break;
1545 case rem:
1546 // Similar to fop2(), the modulo operation is performed using the
1547 // SharedRuntime::drem method (on x86_64 platforms) or using the
1548 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1549 #ifdef _LP64
1550 __ movdbl(xmm1, xmm0);
1551 __ pop_d(xmm0);
1552 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1553 #else
1554 __ push_d(xmm0);
1555 __ pop_d();
1556 __ fld_d(at_rsp());
1557 __ fremr(rax);
1558 __ d2ieee();
1559 __ pop(rax);
1560 __ pop(rdx);
1561 __ push_d();
1562 __ pop_d(xmm0);
1563 #endif
1564 break;
1565 default:
1566 ShouldNotReachHere();
1567 break;
1568 }
1569 } else {
1570 #ifdef _LP64
1571 ShouldNotReachHere();
1572 #else
1573 switch (op) {
1574 case add: __ fadd_d (at_rsp()); break;
1575 case sub: __ fsubr_d(at_rsp()); break;
1576 case mul: {
1577 Label L_strict;
1578 Label L_join;
1579 const Address access_flags (rcx, Method::access_flags_offset());
1580 __ get_method(rcx);
1581 __ movl(rcx, access_flags);
1582 __ testl(rcx, JVM_ACC_STRICT);
1583 __ jccb(Assembler::notZero, L_strict);
1584 __ fmul_d (at_rsp());
1585 __ jmpb(L_join);
1586 __ bind(L_strict);
1587 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1588 __ fmulp();
1589 __ fmul_d (at_rsp());
1590 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1591 __ fmulp();
1592 __ bind(L_join);
1593 break;
1594 }
1595 case div: {
1596 Label L_strict;
1597 Label L_join;
1598 const Address access_flags (rcx, Method::access_flags_offset());
1599 __ get_method(rcx);
1600 __ movl(rcx, access_flags);
1601 __ testl(rcx, JVM_ACC_STRICT);
1602 __ jccb(Assembler::notZero, L_strict);
1603 __ fdivr_d(at_rsp());
1604 __ jmp(L_join);
1605 __ bind(L_strict);
1606 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1607 __ fmul_d (at_rsp());
1608 __ fdivrp();
1609 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1610 __ fmulp();
1611 __ bind(L_join);
1612 break;
1613 }
1614 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1615 default : ShouldNotReachHere();
1616 }
1617 __ d2ieee();
1618 // Pop double precision number from rsp.
1619 __ pop(rax);
1620 __ pop(rdx);
1621 #endif
1622 }
1623 }
1624
1625 void TemplateTable::ineg() {
1626 transition(itos, itos);
1627 __ negl(rax);
1628 }
1629
1630 void TemplateTable::lneg() {
1631 transition(ltos, ltos);
1632 LP64_ONLY(__ negq(rax));
1633 NOT_LP64(__ lneg(rdx, rax));
1634 }
1635
1636 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1637 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1638 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1639 // of 128-bits operands for SSE instructions.
1640 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1641 // Store the value to a 128-bits operand.
1642 operand[0] = lo;
1643 operand[1] = hi;
1644 return operand;
1645 }
1646
1647 // Buffer for 128-bits masks used by SSE instructions.
1648 static jlong float_signflip_pool[2*2];
1649 static jlong double_signflip_pool[2*2];
1650
1651 void TemplateTable::fneg() {
1652 transition(ftos, ftos);
1653 if (UseSSE >= 1) {
1654 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1655 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1656 } else {
1657 LP64_ONLY(ShouldNotReachHere());
1658 NOT_LP64(__ fchs());
1659 }
1660 }
1661
1662 void TemplateTable::dneg() {
1663 transition(dtos, dtos);
1664 if (UseSSE >= 2) {
1665 static jlong *double_signflip =
1666 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1667 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1668 } else {
1669 #ifdef _LP64
1670 ShouldNotReachHere();
1671 #else
1672 __ fchs();
1673 #endif
1674 }
1675 }
1676
1677 void TemplateTable::iinc() {
1678 transition(vtos, vtos);
1679 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1680 locals_index(rbx);
1681 __ addl(iaddress(rbx), rdx);
1682 }
1683
1684 void TemplateTable::wide_iinc() {
1685 transition(vtos, vtos);
1686 __ movl(rdx, at_bcp(4)); // get constant
1687 locals_index_wide(rbx);
1688 __ bswapl(rdx); // swap bytes & sign-extend constant
1689 __ sarl(rdx, 16);
1690 __ addl(iaddress(rbx), rdx);
1691 // Note: should probably use only one movl to get both
1692 // the index and the constant -> fix this
1693 }
1694
1695 void TemplateTable::convert() {
1696 #ifdef _LP64
1697 // Checking
1698 #ifdef ASSERT
1699 {
1700 TosState tos_in = ilgl;
1701 TosState tos_out = ilgl;
1702 switch (bytecode()) {
1703 case Bytecodes::_i2l: // fall through
1704 case Bytecodes::_i2f: // fall through
1705 case Bytecodes::_i2d: // fall through
1706 case Bytecodes::_i2b: // fall through
1707 case Bytecodes::_i2c: // fall through
1708 case Bytecodes::_i2s: tos_in = itos; break;
1709 case Bytecodes::_l2i: // fall through
1710 case Bytecodes::_l2f: // fall through
1711 case Bytecodes::_l2d: tos_in = ltos; break;
1712 case Bytecodes::_f2i: // fall through
1713 case Bytecodes::_f2l: // fall through
1714 case Bytecodes::_f2d: tos_in = ftos; break;
1715 case Bytecodes::_d2i: // fall through
1716 case Bytecodes::_d2l: // fall through
1717 case Bytecodes::_d2f: tos_in = dtos; break;
1718 default : ShouldNotReachHere();
1719 }
1720 switch (bytecode()) {
1721 case Bytecodes::_l2i: // fall through
1722 case Bytecodes::_f2i: // fall through
1723 case Bytecodes::_d2i: // fall through
1724 case Bytecodes::_i2b: // fall through
1725 case Bytecodes::_i2c: // fall through
1726 case Bytecodes::_i2s: tos_out = itos; break;
1727 case Bytecodes::_i2l: // fall through
1728 case Bytecodes::_f2l: // fall through
1729 case Bytecodes::_d2l: tos_out = ltos; break;
1730 case Bytecodes::_i2f: // fall through
1731 case Bytecodes::_l2f: // fall through
1732 case Bytecodes::_d2f: tos_out = ftos; break;
1733 case Bytecodes::_i2d: // fall through
1734 case Bytecodes::_l2d: // fall through
1735 case Bytecodes::_f2d: tos_out = dtos; break;
1736 default : ShouldNotReachHere();
1737 }
1738 transition(tos_in, tos_out);
1739 }
1740 #endif // ASSERT
1741
1742 static const int64_t is_nan = 0x8000000000000000L;
1743
1744 // Conversion
1745 switch (bytecode()) {
1746 case Bytecodes::_i2l:
1747 __ movslq(rax, rax);
1748 break;
1749 case Bytecodes::_i2f:
1750 __ cvtsi2ssl(xmm0, rax);
1751 break;
1752 case Bytecodes::_i2d:
1753 __ cvtsi2sdl(xmm0, rax);
1754 break;
1755 case Bytecodes::_i2b:
1756 __ movsbl(rax, rax);
1757 break;
1758 case Bytecodes::_i2c:
1759 __ movzwl(rax, rax);
1760 break;
1761 case Bytecodes::_i2s:
1762 __ movswl(rax, rax);
1763 break;
1764 case Bytecodes::_l2i:
1765 __ movl(rax, rax);
1766 break;
1767 case Bytecodes::_l2f:
1768 __ cvtsi2ssq(xmm0, rax);
1769 break;
1770 case Bytecodes::_l2d:
1771 __ cvtsi2sdq(xmm0, rax);
1772 break;
1773 case Bytecodes::_f2i:
1774 {
1775 Label L;
1776 __ cvttss2sil(rax, xmm0);
1777 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1778 __ jcc(Assembler::notEqual, L);
1779 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1780 __ bind(L);
1781 }
1782 break;
1783 case Bytecodes::_f2l:
1784 {
1785 Label L;
1786 __ cvttss2siq(rax, xmm0);
1787 // NaN or overflow/underflow?
1788 __ cmp64(rax, ExternalAddress((address) &is_nan));
1789 __ jcc(Assembler::notEqual, L);
1790 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1791 __ bind(L);
1792 }
1793 break;
1794 case Bytecodes::_f2d:
1795 __ cvtss2sd(xmm0, xmm0);
1796 break;
1797 case Bytecodes::_d2i:
1798 {
1799 Label L;
1800 __ cvttsd2sil(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::d2i), 1);
1804 __ bind(L);
1805 }
1806 break;
1807 case Bytecodes::_d2l:
1808 {
1809 Label L;
1810 __ cvttsd2siq(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::d2l), 1);
1815 __ bind(L);
1816 }
1817 break;
1818 case Bytecodes::_d2f:
1819 __ cvtsd2ss(xmm0, xmm0);
1820 break;
1821 default:
1822 ShouldNotReachHere();
1823 }
1824 #else
1825 // Checking
1826 #ifdef ASSERT
1827 { TosState tos_in = ilgl;
1828 TosState tos_out = ilgl;
1829 switch (bytecode()) {
1830 case Bytecodes::_i2l: // fall through
1831 case Bytecodes::_i2f: // fall through
1832 case Bytecodes::_i2d: // fall through
1833 case Bytecodes::_i2b: // fall through
1834 case Bytecodes::_i2c: // fall through
1835 case Bytecodes::_i2s: tos_in = itos; break;
1836 case Bytecodes::_l2i: // fall through
1837 case Bytecodes::_l2f: // fall through
1838 case Bytecodes::_l2d: tos_in = ltos; break;
1839 case Bytecodes::_f2i: // fall through
1840 case Bytecodes::_f2l: // fall through
1841 case Bytecodes::_f2d: tos_in = ftos; break;
1842 case Bytecodes::_d2i: // fall through
1843 case Bytecodes::_d2l: // fall through
1844 case Bytecodes::_d2f: tos_in = dtos; break;
1845 default : ShouldNotReachHere();
1846 }
1847 switch (bytecode()) {
1848 case Bytecodes::_l2i: // fall through
1849 case Bytecodes::_f2i: // fall through
1850 case Bytecodes::_d2i: // fall through
1851 case Bytecodes::_i2b: // fall through
1852 case Bytecodes::_i2c: // fall through
1853 case Bytecodes::_i2s: tos_out = itos; break;
1854 case Bytecodes::_i2l: // fall through
1855 case Bytecodes::_f2l: // fall through
1856 case Bytecodes::_d2l: tos_out = ltos; break;
1857 case Bytecodes::_i2f: // fall through
1858 case Bytecodes::_l2f: // fall through
1859 case Bytecodes::_d2f: tos_out = ftos; break;
1860 case Bytecodes::_i2d: // fall through
1861 case Bytecodes::_l2d: // fall through
1862 case Bytecodes::_f2d: tos_out = dtos; break;
1863 default : ShouldNotReachHere();
1864 }
1865 transition(tos_in, tos_out);
1866 }
1867 #endif // ASSERT
1868
1869 // Conversion
1870 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1871 switch (bytecode()) {
1872 case Bytecodes::_i2l:
1873 __ extend_sign(rdx, rax);
1874 break;
1875 case Bytecodes::_i2f:
1876 if (UseSSE >= 1) {
1877 __ cvtsi2ssl(xmm0, rax);
1878 } else {
1879 __ push(rax); // store int on tos
1880 __ fild_s(at_rsp()); // load int to ST0
1881 __ f2ieee(); // truncate to float size
1882 __ pop(rcx); // adjust rsp
1883 }
1884 break;
1885 case Bytecodes::_i2d:
1886 if (UseSSE >= 2) {
1887 __ cvtsi2sdl(xmm0, rax);
1888 } else {
1889 __ push(rax); // add one slot for d2ieee()
1890 __ push(rax); // store int on tos
1891 __ fild_s(at_rsp()); // load int to ST0
1892 __ d2ieee(); // truncate to double size
1893 __ pop(rcx); // adjust rsp
1894 __ pop(rcx);
1895 }
1896 break;
1897 case Bytecodes::_i2b:
1898 __ shll(rax, 24); // truncate upper 24 bits
1899 __ sarl(rax, 24); // and sign-extend byte
1900 LP64_ONLY(__ movsbl(rax, rax));
1901 break;
1902 case Bytecodes::_i2c:
1903 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1904 LP64_ONLY(__ movzwl(rax, rax));
1905 break;
1906 case Bytecodes::_i2s:
1907 __ shll(rax, 16); // truncate upper 16 bits
1908 __ sarl(rax, 16); // and sign-extend short
1909 LP64_ONLY(__ movswl(rax, rax));
1910 break;
1911 case Bytecodes::_l2i:
1912 /* nothing to do */
1913 break;
1914 case Bytecodes::_l2f:
1915 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1916 // 64-bit long values to floats. On 32-bit platforms it is not possible
1917 // to use that instruction with 64-bit operands, therefore the FPU is
1918 // used to perform the conversion.
1919 __ push(rdx); // store long on tos
1920 __ push(rax);
1921 __ fild_d(at_rsp()); // load long to ST0
1922 __ f2ieee(); // truncate to float size
1923 __ pop(rcx); // adjust rsp
1924 __ pop(rcx);
1925 if (UseSSE >= 1) {
1926 __ push_f();
1927 __ pop_f(xmm0);
1928 }
1929 break;
1930 case Bytecodes::_l2d:
1931 // On 32-bit platforms the FPU is used for conversion because on
1932 // 32-bit platforms it is not not possible to use the cvtsi2sdq
1933 // instruction with 64-bit operands.
1934 __ push(rdx); // store long on tos
1935 __ push(rax);
1936 __ fild_d(at_rsp()); // load long to ST0
1937 __ d2ieee(); // truncate to double size
1938 __ pop(rcx); // adjust rsp
1939 __ pop(rcx);
1940 if (UseSSE >= 2) {
1941 __ push_d();
1942 __ pop_d(xmm0);
1943 }
1944 break;
1945 case Bytecodes::_f2i:
1946 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1947 // as it returns 0 for any NaN.
1948 if (UseSSE >= 1) {
1949 __ push_f(xmm0);
1950 } else {
1951 __ push(rcx); // reserve space for argument
1952 __ fstp_s(at_rsp()); // pass float argument on stack
1953 }
1954 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1955 break;
1956 case Bytecodes::_f2l:
1957 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
1958 // as it returns 0 for any NaN.
1959 if (UseSSE >= 1) {
1960 __ push_f(xmm0);
1961 } else {
1962 __ push(rcx); // reserve space for argument
1963 __ fstp_s(at_rsp()); // pass float argument on stack
1964 }
1965 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1966 break;
1967 case Bytecodes::_f2d:
1968 if (UseSSE < 1) {
1969 /* nothing to do */
1970 } else if (UseSSE == 1) {
1971 __ push_f(xmm0);
1972 __ pop_f();
1973 } else { // UseSSE >= 2
1974 __ cvtss2sd(xmm0, xmm0);
1975 }
1976 break;
1977 case Bytecodes::_d2i:
1978 if (UseSSE >= 2) {
1979 __ push_d(xmm0);
1980 } else {
1981 __ push(rcx); // reserve space for argument
1982 __ push(rcx);
1983 __ fstp_d(at_rsp()); // pass double argument on stack
1984 }
1985 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1986 break;
1987 case Bytecodes::_d2l:
1988 if (UseSSE >= 2) {
1989 __ push_d(xmm0);
1990 } else {
1991 __ push(rcx); // reserve space for argument
1992 __ push(rcx);
1993 __ fstp_d(at_rsp()); // pass double argument on stack
1994 }
1995 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1996 break;
1997 case Bytecodes::_d2f:
1998 if (UseSSE <= 1) {
1999 __ push(rcx); // reserve space for f2ieee()
2000 __ f2ieee(); // truncate to float size
2001 __ pop(rcx); // adjust rsp
2002 if (UseSSE == 1) {
2003 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2004 // the conversion is performed using the FPU in this case.
2005 __ push_f();
2006 __ pop_f(xmm0);
2007 }
2008 } else { // UseSSE >= 2
2009 __ cvtsd2ss(xmm0, xmm0);
2010 }
2011 break;
2012 default :
2013 ShouldNotReachHere();
2014 }
2015 #endif
2016 }
2017
2018 void TemplateTable::lcmp() {
2019 transition(ltos, itos);
2020 #ifdef _LP64
2021 Label done;
2022 __ pop_l(rdx);
2023 __ cmpq(rdx, rax);
2024 __ movl(rax, -1);
2025 __ jccb(Assembler::less, done);
2026 __ setb(Assembler::notEqual, rax);
2027 __ movzbl(rax, rax);
2028 __ bind(done);
2029 #else
2030
2031 // y = rdx:rax
2032 __ pop_l(rbx, rcx); // get x = rcx:rbx
2033 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2034 __ mov(rax, rcx);
2035 #endif
2036 }
2037
2038 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2039 if ((is_float && UseSSE >= 1) ||
2040 (!is_float && UseSSE >= 2)) {
2041 Label done;
2042 if (is_float) {
2043 // XXX get rid of pop here, use ... reg, mem32
2044 __ pop_f(xmm1);
2045 __ ucomiss(xmm1, xmm0);
2046 } else {
2047 // XXX get rid of pop here, use ... reg, mem64
2048 __ pop_d(xmm1);
2049 __ ucomisd(xmm1, xmm0);
2050 }
2051 if (unordered_result < 0) {
2052 __ movl(rax, -1);
2053 __ jccb(Assembler::parity, done);
2054 __ jccb(Assembler::below, done);
2055 __ setb(Assembler::notEqual, rdx);
2056 __ movzbl(rax, rdx);
2057 } else {
2058 __ movl(rax, 1);
2059 __ jccb(Assembler::parity, done);
2060 __ jccb(Assembler::above, done);
2061 __ movl(rax, 0);
2062 __ jccb(Assembler::equal, done);
2063 __ decrementl(rax);
2064 }
2065 __ bind(done);
2066 } else {
2067 #ifdef _LP64
2068 ShouldNotReachHere();
2069 #else
2070 if (is_float) {
2071 __ fld_s(at_rsp());
2072 } else {
2073 __ fld_d(at_rsp());
2074 __ pop(rdx);
2075 }
2076 __ pop(rcx);
2077 __ fcmp2int(rax, unordered_result < 0);
2078 #endif // _LP64
2079 }
2080 }
2081
2082 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2083 __ get_method(rcx); // rcx holds method
2084 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2085 // holds bumped taken count
2086
2087 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2088 InvocationCounter::counter_offset();
2089 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2090 InvocationCounter::counter_offset();
2091
2092 // Load up edx with the branch displacement
2093 if (is_wide) {
2094 __ movl(rdx, at_bcp(1));
2095 } else {
2096 __ load_signed_short(rdx, at_bcp(1));
2097 }
2098 __ bswapl(rdx);
2099
2100 if (!is_wide) {
2101 __ sarl(rdx, 16);
2102 }
2103 LP64_ONLY(__ movl2ptr(rdx, rdx));
2104
2105 // Handle all the JSR stuff here, then exit.
2106 // It's much shorter and cleaner than intermingling with the non-JSR
2107 // normal-branch stuff occurring below.
2108 if (is_jsr) {
2109 // Pre-load the next target bytecode into rbx
2110 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2111
2112 // compute return address as bci in rax
2113 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2114 in_bytes(ConstMethod::codes_offset())));
2115 __ subptr(rax, Address(rcx, Method::const_offset()));
2116 // Adjust the bcp in r13 by the displacement in rdx
2117 __ addptr(rbcp, rdx);
2118 // jsr returns atos that is not an oop
2119 __ push_i(rax);
2120 __ dispatch_only(vtos);
2121 return;
2122 }
2123
2124 // Normal (non-jsr) branch handling
2125
2126 // Adjust the bcp in r13 by the displacement in rdx
2127 __ addptr(rbcp, rdx);
2128
2129 assert(UseLoopCounter || !UseOnStackReplacement,
2130 "on-stack-replacement requires loop counters");
2131 Label backedge_counter_overflow;
2132 Label profile_method;
2133 Label dispatch;
2134 if (UseLoopCounter) {
2135 // increment backedge counter for backward branches
2136 // rax: MDO
2137 // rbx: MDO bumped taken-count
2138 // rcx: method
2139 // rdx: target offset
2140 // r13: target bcp
2141 // r14: locals pointer
2142 __ testl(rdx, rdx); // check if forward or backward branch
2143 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2144
2145 // check if MethodCounters exists
2146 Label has_counters;
2147 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2148 __ testptr(rax, rax);
2149 __ jcc(Assembler::notZero, has_counters);
2150 __ push(rdx);
2151 __ push(rcx);
2152 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2153 rcx);
2154 __ pop(rcx);
2155 __ pop(rdx);
2156 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2157 __ testptr(rax, rax);
2158 __ jcc(Assembler::zero, dispatch);
2159 __ bind(has_counters);
2160
2161 if (TieredCompilation) {
2162 Label no_mdo;
2163 int increment = InvocationCounter::count_increment;
2164 if (ProfileInterpreter) {
2165 // Are we profiling?
2166 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2167 __ testptr(rbx, rbx);
2168 __ jccb(Assembler::zero, no_mdo);
2169 // Increment the MDO backedge counter
2170 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2171 in_bytes(InvocationCounter::counter_offset()));
2172 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2173 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2174 rax, false, Assembler::zero, &backedge_counter_overflow);
2175 __ jmp(dispatch);
2176 }
2177 __ bind(no_mdo);
2178 // Increment backedge counter in MethodCounters*
2179 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2180 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2181 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2182 rax, false, Assembler::zero, &backedge_counter_overflow);
2183 } else { // not TieredCompilation
2184 // increment counter
2185 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2186 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2187 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2188 __ movl(Address(rcx, be_offset), rax); // store counter
2189
2190 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2191
2192 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2193 __ addl(rax, Address(rcx, be_offset)); // add both counters
2194
2195 if (ProfileInterpreter) {
2196 // Test to see if we should create a method data oop
2197 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2198 __ jcc(Assembler::less, dispatch);
2199
2200 // if no method data exists, go to profile method
2201 __ test_method_data_pointer(rax, profile_method);
2202
2203 if (UseOnStackReplacement) {
2204 // check for overflow against rbx which is the MDO taken count
2205 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2206 __ jcc(Assembler::below, dispatch);
2207
2208 // When ProfileInterpreter is on, the backedge_count comes
2209 // from the MethodData*, which value does not get reset on
2210 // the call to frequency_counter_overflow(). To avoid
2211 // excessive calls to the overflow routine while the method is
2212 // being compiled, add a second test to make sure the overflow
2213 // function is called only once every overflow_frequency.
2214 const int overflow_frequency = 1024;
2215 __ andl(rbx, overflow_frequency - 1);
2216 __ jcc(Assembler::zero, backedge_counter_overflow);
2217
2218 }
2219 } else {
2220 if (UseOnStackReplacement) {
2221 // check for overflow against rax, which is the sum of the
2222 // counters
2223 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2224 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2225
2226 }
2227 }
2228 }
2229 __ bind(dispatch);
2230 }
2231
2232 // Pre-load the next target bytecode into rbx
2233 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2234
2235 // continue with the bytecode @ target
2236 // rax: return bci for jsr's, unused otherwise
2237 // rbx: target bytecode
2238 // r13: target bcp
2239 __ dispatch_only(vtos);
2240
2241 if (UseLoopCounter) {
2242 if (ProfileInterpreter) {
2243 // Out-of-line code to allocate method data oop.
2244 __ bind(profile_method);
2245 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2246 __ set_method_data_pointer_for_bcp();
2247 __ jmp(dispatch);
2248 }
2249
2250 if (UseOnStackReplacement) {
2251 // invocation counter overflow
2252 __ bind(backedge_counter_overflow);
2253 __ negptr(rdx);
2254 __ addptr(rdx, rbcp); // branch bcp
2255 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2256 __ call_VM(noreg,
2257 CAST_FROM_FN_PTR(address,
2258 InterpreterRuntime::frequency_counter_overflow),
2259 rdx);
2260
2261 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2262 // rdx: scratch
2263 // r14: locals pointer
2264 // r13: bcp
2265 __ testptr(rax, rax); // test result
2266 __ jcc(Assembler::zero, dispatch); // no osr if null
2267 // nmethod may have been invalidated (VM may block upon call_VM return)
2268 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2269 __ jcc(Assembler::notEqual, dispatch);
2270
2271 // We have the address of an on stack replacement routine in rax.
2272 // In preparation of invoking it, first we must migrate the locals
2273 // and monitors from off the interpreter frame on the stack.
2274 // Ensure to save the osr nmethod over the migration call,
2275 // it will be preserved in rbx.
2276 __ mov(rbx, rax);
2277
2278 NOT_LP64(__ get_thread(rcx));
2279
2280 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2281
2282 // rax is OSR buffer, move it to expected parameter location
2283 LP64_ONLY(__ mov(j_rarg0, rax));
2284 NOT_LP64(__ mov(rcx, rax));
2285 // We use j_rarg definitions here so that registers don't conflict as parameter
2286 // registers change across platforms as we are in the midst of a calling
2287 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2288
2289 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2290 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2291
2292 // pop the interpreter frame
2293 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2294 __ leave(); // remove frame anchor
2295 __ pop(retaddr); // get return address
2296 __ mov(rsp, sender_sp); // set sp to sender sp
2297 // Ensure compiled code always sees stack at proper alignment
2298 __ andptr(rsp, -(StackAlignmentInBytes));
2299
2300 // unlike x86 we need no specialized return from compiled code
2301 // to the interpreter or the call stub.
2302
2303 // push the return address
2304 __ push(retaddr);
2305
2306 // and begin the OSR nmethod
2307 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2308 }
2309 }
2310 }
2311
2312 void TemplateTable::if_0cmp(Condition cc) {
2313 transition(itos, vtos);
2314 // assume branch is more often taken than not (loops use backward branches)
2315 Label not_taken;
2316 __ testl(rax, rax);
2317 __ jcc(j_not(cc), not_taken);
2318 branch(false, false);
2319 __ bind(not_taken);
2320 __ profile_not_taken_branch(rax);
2321 }
2322
2323 void TemplateTable::if_icmp(Condition cc) {
2324 transition(itos, vtos);
2325 // assume branch is more often taken than not (loops use backward branches)
2326 Label not_taken;
2327 __ pop_i(rdx);
2328 __ cmpl(rdx, rax);
2329 __ jcc(j_not(cc), not_taken);
2330 branch(false, false);
2331 __ bind(not_taken);
2332 __ profile_not_taken_branch(rax);
2333 }
2334
2335 void TemplateTable::if_nullcmp(Condition cc) {
2336 transition(atos, vtos);
2337 // assume branch is more often taken than not (loops use backward branches)
2338 Label not_taken;
2339 __ testptr(rax, rax);
2340 __ jcc(j_not(cc), not_taken);
2341 branch(false, false);
2342 __ bind(not_taken);
2343 __ profile_not_taken_branch(rax);
2344 }
2345
2346 void TemplateTable::if_acmp(Condition cc) {
2347 transition(atos, vtos);
2348 // assume branch is more often taken than not (loops use backward branches)
2349 Label not_taken;
2350 __ pop_ptr(rdx);
2351 __ cmpoopptr(rdx, rax);
2352 __ jcc(j_not(cc), not_taken);
2353 branch(false, false);
2354 __ bind(not_taken);
2355 __ profile_not_taken_branch(rax);
2356 }
2357
2358 void TemplateTable::ret() {
2359 transition(vtos, vtos);
2360 locals_index(rbx);
2361 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2362 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2363 __ profile_ret(rbx, rcx);
2364 __ get_method(rax);
2365 __ movptr(rbcp, Address(rax, Method::const_offset()));
2366 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2367 ConstMethod::codes_offset()));
2368 __ dispatch_next(vtos);
2369 }
2370
2371 void TemplateTable::wide_ret() {
2372 transition(vtos, vtos);
2373 locals_index_wide(rbx);
2374 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2375 __ profile_ret(rbx, rcx);
2376 __ get_method(rax);
2377 __ movptr(rbcp, Address(rax, Method::const_offset()));
2378 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2379 __ dispatch_next(vtos);
2380 }
2381
2382 void TemplateTable::tableswitch() {
2383 Label default_case, continue_execution;
2384 transition(itos, vtos);
2385
2386 // align r13/rsi
2387 __ lea(rbx, at_bcp(BytesPerInt));
2388 __ andptr(rbx, -BytesPerInt);
2389 // load lo & hi
2390 __ movl(rcx, Address(rbx, BytesPerInt));
2391 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2392 __ bswapl(rcx);
2393 __ bswapl(rdx);
2394 // check against lo & hi
2395 __ cmpl(rax, rcx);
2396 __ jcc(Assembler::less, default_case);
2397 __ cmpl(rax, rdx);
2398 __ jcc(Assembler::greater, default_case);
2399 // lookup dispatch offset
2400 __ subl(rax, rcx);
2401 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2402 __ profile_switch_case(rax, rbx, rcx);
2403 // continue execution
2404 __ bind(continue_execution);
2405 __ bswapl(rdx);
2406 LP64_ONLY(__ movl2ptr(rdx, rdx));
2407 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2408 __ addptr(rbcp, rdx);
2409 __ dispatch_only(vtos);
2410 // handle default
2411 __ bind(default_case);
2412 __ profile_switch_default(rax);
2413 __ movl(rdx, Address(rbx, 0));
2414 __ jmp(continue_execution);
2415 }
2416
2417 void TemplateTable::lookupswitch() {
2418 transition(itos, itos);
2419 __ stop("lookupswitch bytecode should have been rewritten");
2420 }
2421
2422 void TemplateTable::fast_linearswitch() {
2423 transition(itos, vtos);
2424 Label loop_entry, loop, found, continue_execution;
2425 // bswap rax so we can avoid bswapping the table entries
2426 __ bswapl(rax);
2427 // align r13
2428 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2429 // this instruction (change offsets
2430 // below)
2431 __ andptr(rbx, -BytesPerInt);
2432 // set counter
2433 __ movl(rcx, Address(rbx, BytesPerInt));
2434 __ bswapl(rcx);
2435 __ jmpb(loop_entry);
2436 // table search
2437 __ bind(loop);
2438 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2439 __ jcc(Assembler::equal, found);
2440 __ bind(loop_entry);
2441 __ decrementl(rcx);
2442 __ jcc(Assembler::greaterEqual, loop);
2443 // default case
2444 __ profile_switch_default(rax);
2445 __ movl(rdx, Address(rbx, 0));
2446 __ jmp(continue_execution);
2447 // entry found -> get offset
2448 __ bind(found);
2449 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2450 __ profile_switch_case(rcx, rax, rbx);
2451 // continue execution
2452 __ bind(continue_execution);
2453 __ bswapl(rdx);
2454 __ movl2ptr(rdx, rdx);
2455 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2456 __ addptr(rbcp, rdx);
2457 __ dispatch_only(vtos);
2458 }
2459
2460 void TemplateTable::fast_binaryswitch() {
2461 transition(itos, vtos);
2462 // Implementation using the following core algorithm:
2463 //
2464 // int binary_search(int key, LookupswitchPair* array, int n) {
2465 // // Binary search according to "Methodik des Programmierens" by
2466 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2467 // int i = 0;
2468 // int j = n;
2469 // while (i+1 < j) {
2470 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2471 // // with Q: for all i: 0 <= i < n: key < a[i]
2472 // // where a stands for the array and assuming that the (inexisting)
2473 // // element a[n] is infinitely big.
2474 // int h = (i + j) >> 1;
2475 // // i < h < j
2476 // if (key < array[h].fast_match()) {
2477 // j = h;
2478 // } else {
2479 // i = h;
2480 // }
2481 // }
2482 // // R: a[i] <= key < a[i+1] or Q
2483 // // (i.e., if key is within array, i is the correct index)
2484 // return i;
2485 // }
2486
2487 // Register allocation
2488 const Register key = rax; // already set (tosca)
2489 const Register array = rbx;
2490 const Register i = rcx;
2491 const Register j = rdx;
2492 const Register h = rdi;
2493 const Register temp = rsi;
2494
2495 // Find array start
2496 NOT_LP64(__ save_bcp());
2497
2498 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2499 // get rid of this
2500 // instruction (change
2501 // offsets below)
2502 __ andptr(array, -BytesPerInt);
2503
2504 // Initialize i & j
2505 __ xorl(i, i); // i = 0;
2506 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2507
2508 // Convert j into native byteordering
2509 __ bswapl(j);
2510
2511 // And start
2512 Label entry;
2513 __ jmp(entry);
2514
2515 // binary search loop
2516 {
2517 Label loop;
2518 __ bind(loop);
2519 // int h = (i + j) >> 1;
2520 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2521 __ sarl(h, 1); // h = (i + j) >> 1;
2522 // if (key < array[h].fast_match()) {
2523 // j = h;
2524 // } else {
2525 // i = h;
2526 // }
2527 // Convert array[h].match to native byte-ordering before compare
2528 __ movl(temp, Address(array, h, Address::times_8));
2529 __ bswapl(temp);
2530 __ cmpl(key, temp);
2531 // j = h if (key < array[h].fast_match())
2532 __ cmov32(Assembler::less, j, h);
2533 // i = h if (key >= array[h].fast_match())
2534 __ cmov32(Assembler::greaterEqual, i, h);
2535 // while (i+1 < j)
2536 __ bind(entry);
2537 __ leal(h, Address(i, 1)); // i+1
2538 __ cmpl(h, j); // i+1 < j
2539 __ jcc(Assembler::less, loop);
2540 }
2541
2542 // end of binary search, result index is i (must check again!)
2543 Label default_case;
2544 // Convert array[i].match to native byte-ordering before compare
2545 __ movl(temp, Address(array, i, Address::times_8));
2546 __ bswapl(temp);
2547 __ cmpl(key, temp);
2548 __ jcc(Assembler::notEqual, default_case);
2549
2550 // entry found -> j = offset
2551 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2552 __ profile_switch_case(i, key, array);
2553 __ bswapl(j);
2554 LP64_ONLY(__ movslq(j, j));
2555
2556 NOT_LP64(__ restore_bcp());
2557 NOT_LP64(__ restore_locals()); // restore rdi
2558
2559 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2560 __ addptr(rbcp, j);
2561 __ dispatch_only(vtos);
2562
2563 // default case -> j = default offset
2564 __ bind(default_case);
2565 __ profile_switch_default(i);
2566 __ movl(j, Address(array, -2 * BytesPerInt));
2567 __ bswapl(j);
2568 LP64_ONLY(__ movslq(j, j));
2569
2570 NOT_LP64(__ restore_bcp());
2571 NOT_LP64(__ restore_locals());
2572
2573 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2574 __ addptr(rbcp, j);
2575 __ dispatch_only(vtos);
2576 }
2577
2578 void TemplateTable::_return(TosState state) {
2579 transition(state, state);
2580
2581 assert(_desc->calls_vm(),
2582 "inconsistent calls_vm information"); // call in remove_activation
2583
2584 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2585 assert(state == vtos, "only valid state");
2586 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2587 __ movptr(robj, aaddress(0));
2588 __ load_klass(rdi, robj);
2589 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2590 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2591 Label skip_register_finalizer;
2592 __ jcc(Assembler::zero, skip_register_finalizer);
2593
2594 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2595
2596 __ bind(skip_register_finalizer);
2597 }
2598
2599 // Narrow result if state is itos but result type is smaller.
2600 // Need to narrow in the return bytecode rather than in generate_return_entry
2601 // since compiled code callers expect the result to already be narrowed.
2602 if (state == itos) {
2603 __ narrow(rax);
2604 }
2605 __ remove_activation(state, rbcp);
2606
2607 __ jmp(rbcp);
2608 }
2609
2610 // ----------------------------------------------------------------------------
2611 // Volatile variables demand their effects be made known to all CPU's
2612 // in order. Store buffers on most chips allow reads & writes to
2613 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2614 // without some kind of memory barrier (i.e., it's not sufficient that
2615 // the interpreter does not reorder volatile references, the hardware
2616 // also must not reorder them).
2617 //
2618 // According to the new Java Memory Model (JMM):
2619 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2620 // writes act as aquire & release, so:
2621 // (2) A read cannot let unrelated NON-volatile memory refs that
2622 // happen after the read float up to before the read. It's OK for
2623 // non-volatile memory refs that happen before the volatile read to
2624 // float down below it.
2625 // (3) Similar a volatile write cannot let unrelated NON-volatile
2626 // memory refs that happen BEFORE the write float down to after the
2627 // write. It's OK for non-volatile memory refs that happen after the
2628 // volatile write to float up before it.
2629 //
2630 // We only put in barriers around volatile refs (they are expensive),
2631 // not _between_ memory refs (that would require us to track the
2632 // flavor of the previous memory refs). Requirements (2) and (3)
2633 // require some barriers before volatile stores and after volatile
2634 // loads. These nearly cover requirement (1) but miss the
2635 // volatile-store-volatile-load case. This final case is placed after
2636 // volatile-stores although it could just as well go before
2637 // volatile-loads.
2638
2639 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2640 // Helper function to insert a is-volatile test and memory barrier
2641 if(!os::is_MP()) return; // Not needed on single CPU
2642 __ membar(order_constraint);
2643 }
2644
2645 void TemplateTable::resolve_cache_and_index(int byte_no,
2646 Register Rcache,
2647 Register index,
2648 size_t index_size) {
2649 const Register temp = rbx;
2650 assert_different_registers(Rcache, index, temp);
2651
2652 Label resolved;
2653
2654 Bytecodes::Code code = bytecode();
2655 switch (code) {
2656 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2657 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2658 default: break;
2659 }
2660
2661 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2662 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2663 __ cmpl(temp, code); // have we resolved this bytecode?
2664 __ jcc(Assembler::equal, resolved);
2665
2666 // resolve first time through
2667 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2668 __ movl(temp, code);
2669 __ call_VM(noreg, entry, temp);
2670 // Update registers with resolved info
2671 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2672 __ bind(resolved);
2673 }
2674
2675 // The cache and index registers must be set before call
2676 void TemplateTable::load_field_cp_cache_entry(Register obj,
2677 Register cache,
2678 Register index,
2679 Register off,
2680 Register flags,
2681 bool is_static = false) {
2682 assert_different_registers(cache, index, flags, off);
2683
2684 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2685 // Field offset
2686 __ movptr(off, Address(cache, index, Address::times_ptr,
2687 in_bytes(cp_base_offset +
2688 ConstantPoolCacheEntry::f2_offset())));
2689 // Flags
2690 __ movl(flags, Address(cache, index, Address::times_ptr,
2691 in_bytes(cp_base_offset +
2692 ConstantPoolCacheEntry::flags_offset())));
2693
2694 // klass overwrite register
2695 if (is_static) {
2696 __ movptr(obj, Address(cache, index, Address::times_ptr,
2697 in_bytes(cp_base_offset +
2698 ConstantPoolCacheEntry::f1_offset())));
2699 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2700 __ movptr(obj, Address(obj, mirror_offset));
2701 }
2702 }
2703
2704 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2705 Register method,
2706 Register itable_index,
2707 Register flags,
2708 bool is_invokevirtual,
2709 bool is_invokevfinal, /*unused*/
2710 bool is_invokedynamic) {
2711 // setup registers
2712 const Register cache = rcx;
2713 const Register index = rdx;
2714 assert_different_registers(method, flags);
2715 assert_different_registers(method, cache, index);
2716 assert_different_registers(itable_index, flags);
2717 assert_different_registers(itable_index, cache, index);
2718 // determine constant pool cache field offsets
2719 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2720 const int method_offset = in_bytes(
2721 ConstantPoolCache::base_offset() +
2722 ((byte_no == f2_byte)
2723 ? ConstantPoolCacheEntry::f2_offset()
2724 : ConstantPoolCacheEntry::f1_offset()));
2725 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2726 ConstantPoolCacheEntry::flags_offset());
2727 // access constant pool cache fields
2728 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2729 ConstantPoolCacheEntry::f2_offset());
2730
2731 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2732 resolve_cache_and_index(byte_no, cache, index, index_size);
2733 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2734
2735 if (itable_index != noreg) {
2736 // pick up itable or appendix index from f2 also:
2737 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2738 }
2739 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2740 }
2741
2742 // The registers cache and index expected to be set before call.
2743 // Correct values of the cache and index registers are preserved.
2744 void TemplateTable::jvmti_post_field_access(Register cache,
2745 Register index,
2746 bool is_static,
2747 bool has_tos) {
2748 if (JvmtiExport::can_post_field_access()) {
2749 // Check to see if a field access watch has been set before we take
2750 // the time to call into the VM.
2751 Label L1;
2752 assert_different_registers(cache, index, rax);
2753 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2754 __ testl(rax,rax);
2755 __ jcc(Assembler::zero, L1);
2756
2757 // cache entry pointer
2758 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2759 __ shll(index, LogBytesPerWord);
2760 __ addptr(cache, index);
2761 if (is_static) {
2762 __ xorptr(rax, rax); // NULL object reference
2763 } else {
2764 __ pop(atos); // Get the object
2765 __ verify_oop(rax);
2766 __ push(atos); // Restore stack state
2767 }
2768 // rax,: object pointer or NULL
2769 // cache: cache entry pointer
2770 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2771 rax, cache);
2772 __ get_cache_and_index_at_bcp(cache, index, 1);
2773 __ bind(L1);
2774 }
2775 }
2776
2777 void TemplateTable::pop_and_check_object(Register r) {
2778 __ pop_ptr(r);
2779 __ null_check(r); // for field access must check obj.
2780 __ verify_oop(r);
2781 }
2782
2783 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2784 transition(vtos, vtos);
2785
2786 const Register cache = rcx;
2787 const Register index = rdx;
2788 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2789 const Register off = rbx;
2790 const Register flags = rax;
2791 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2792
2793 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2794 jvmti_post_field_access(cache, index, is_static, false);
2795 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2796
2797 if (!is_static) pop_and_check_object(obj);
2798 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, obj);
2799
2800 const Address field(obj, off, Address::times_1, 0*wordSize);
2801 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2802
2803 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2804
2805 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2806 // Make sure we don't need to mask edx after the above shift
2807 assert(btos == 0, "change code, btos != 0");
2808
2809 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2810
2811 __ jcc(Assembler::notZero, notByte);
2812 // btos
2813 __ load_signed_byte(rax, field);
2814 __ push(btos);
2815 // Rewrite bytecode to be faster
2816 if (!is_static && rc == may_rewrite) {
2817 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2818 }
2819 __ jmp(Done);
2820
2821 __ bind(notByte);
2822 __ cmpl(flags, ztos);
2823 __ jcc(Assembler::notEqual, notBool);
2824
2825 // ztos (same code as btos)
2826 __ load_signed_byte(rax, field);
2827 __ push(ztos);
2828 // Rewrite bytecode to be faster
2829 if (!is_static && rc == may_rewrite) {
2830 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2831 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2832 }
2833 __ jmp(Done);
2834
2835 __ bind(notBool);
2836 __ cmpl(flags, atos);
2837 __ jcc(Assembler::notEqual, notObj);
2838 // atos
2839 __ load_heap_oop(rax, field);
2840 __ push(atos);
2841 if (!is_static && rc == may_rewrite) {
2842 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2843 }
2844 __ jmp(Done);
2845
2846 __ bind(notObj);
2847 __ cmpl(flags, itos);
2848 __ jcc(Assembler::notEqual, notInt);
2849 // itos
2850 __ movl(rax, field);
2851 __ push(itos);
2852 // Rewrite bytecode to be faster
2853 if (!is_static && rc == may_rewrite) {
2854 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2855 }
2856 __ jmp(Done);
2857
2858 __ bind(notInt);
2859 __ cmpl(flags, ctos);
2860 __ jcc(Assembler::notEqual, notChar);
2861 // ctos
2862 __ load_unsigned_short(rax, field);
2863 __ push(ctos);
2864 // Rewrite bytecode to be faster
2865 if (!is_static && rc == may_rewrite) {
2866 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2867 }
2868 __ jmp(Done);
2869
2870 __ bind(notChar);
2871 __ cmpl(flags, stos);
2872 __ jcc(Assembler::notEqual, notShort);
2873 // stos
2874 __ load_signed_short(rax, field);
2875 __ push(stos);
2876 // Rewrite bytecode to be faster
2877 if (!is_static && rc == may_rewrite) {
2878 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2879 }
2880 __ jmp(Done);
2881
2882 __ bind(notShort);
2883 __ cmpl(flags, ltos);
2884 __ jcc(Assembler::notEqual, notLong);
2885 // ltos
2886
2887 #ifndef _LP64
2888 // Generate code as if volatile. There just aren't enough registers to
2889 // save that information and this code is faster than the test.
2890 __ fild_d(field); // Must load atomically
2891 __ subptr(rsp,2*wordSize); // Make space for store
2892 __ fistp_d(Address(rsp,0));
2893 __ pop(rax);
2894 __ pop(rdx);
2895 #else
2896 __ movq(rax, field);
2897 #endif
2898
2899 __ push(ltos);
2900 // Rewrite bytecode to be faster
2901 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2902 __ jmp(Done);
2903
2904 __ bind(notLong);
2905 __ cmpl(flags, ftos);
2906 __ jcc(Assembler::notEqual, notFloat);
2907 // ftos
2908
2909 __ load_float(field);
2910 __ push(ftos);
2911 // Rewrite bytecode to be faster
2912 if (!is_static && rc == may_rewrite) {
2913 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2914 }
2915 __ jmp(Done);
2916
2917 __ bind(notFloat);
2918 #ifdef ASSERT
2919 __ cmpl(flags, dtos);
2920 __ jcc(Assembler::notEqual, notDouble);
2921 #endif
2922 // dtos
2923 __ load_double(field);
2924 __ push(dtos);
2925 // Rewrite bytecode to be faster
2926 if (!is_static && rc == may_rewrite) {
2927 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2928 }
2929 #ifdef ASSERT
2930 __ jmp(Done);
2931
2932
2933 __ bind(notDouble);
2934 __ stop("Bad state");
2935 #endif
2936
2937 __ bind(Done);
2938 // [jk] not needed currently
2939 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2940 // Assembler::LoadStore));
2941 }
2942
2943 void TemplateTable::getfield(int byte_no) {
2944 getfield_or_static(byte_no, false);
2945 }
2946
2947 void TemplateTable::nofast_getfield(int byte_no) {
2948 getfield_or_static(byte_no, false, may_not_rewrite);
2949 }
2950
2951 void TemplateTable::getstatic(int byte_no) {
2952 getfield_or_static(byte_no, true);
2953 }
2954
2955
2956 // The registers cache and index expected to be set before call.
2957 // The function may destroy various registers, just not the cache and index registers.
2958 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2959
2960 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
2961 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2962 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2963 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
2964
2965 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2966
2967 if (JvmtiExport::can_post_field_modification()) {
2968 // Check to see if a field modification watch has been set before
2969 // we take the time to call into the VM.
2970 Label L1;
2971 assert_different_registers(cache, index, rax);
2972 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2973 __ testl(rax, rax);
2974 __ jcc(Assembler::zero, L1);
2975
2976 __ get_cache_and_index_at_bcp(robj, RDX, 1);
2977
2978
2979 if (is_static) {
2980 // Life is simple. Null out the object pointer.
2981 __ xorl(RBX, RBX);
2982
2983 } else {
2984 // Life is harder. The stack holds the value on top, followed by
2985 // the object. We don't know the size of the value, though; it
2986 // could be one or two words depending on its type. As a result,
2987 // we must find the type to determine where the object is.
2988 #ifndef _LP64
2989 Label two_word, valsize_known;
2990 #endif
2991 __ movl(RCX, Address(robj, RDX,
2992 Address::times_ptr,
2993 in_bytes(cp_base_offset +
2994 ConstantPoolCacheEntry::flags_offset())));
2995 NOT_LP64(__ mov(rbx, rsp));
2996 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
2997
2998 // Make sure we don't need to mask rcx after the above shift
2999 ConstantPoolCacheEntry::verify_tos_state_shift();
3000 #ifdef _LP64
3001 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
3002 __ cmpl(c_rarg3, ltos);
3003 __ cmovptr(Assembler::equal,
3004 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3005 __ cmpl(c_rarg3, dtos);
3006 __ cmovptr(Assembler::equal,
3007 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3008 #else
3009 __ cmpl(rcx, ltos);
3010 __ jccb(Assembler::equal, two_word);
3011 __ cmpl(rcx, dtos);
3012 __ jccb(Assembler::equal, two_word);
3013 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3014 __ jmpb(valsize_known);
3015
3016 __ bind(two_word);
3017 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3018
3019 __ bind(valsize_known);
3020 // setup object pointer
3021 __ movptr(rbx, Address(rbx, 0));
3022 #endif
3023 }
3024 // cache entry pointer
3025 __ addptr(robj, in_bytes(cp_base_offset));
3026 __ shll(RDX, LogBytesPerWord);
3027 __ addptr(robj, RDX);
3028 // object (tos)
3029 __ mov(RCX, rsp);
3030 // c_rarg1: object pointer set up above (NULL if static)
3031 // c_rarg2: cache entry pointer
3032 // c_rarg3: jvalue object on the stack
3033 __ call_VM(noreg,
3034 CAST_FROM_FN_PTR(address,
3035 InterpreterRuntime::post_field_modification),
3036 RBX, robj, RCX);
3037 __ get_cache_and_index_at_bcp(cache, index, 1);
3038 __ bind(L1);
3039 }
3040 }
3041
3042 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3043 transition(vtos, vtos);
3044
3045 const Register cache = rcx;
3046 const Register index = rdx;
3047 const Register obj = rcx;
3048 const Register off = rbx;
3049 const Register flags = rax;
3050 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3051
3052 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3053 jvmti_post_field_mod(cache, index, is_static);
3054 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3055
3056 // [jk] not needed currently
3057 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3058 // Assembler::StoreStore));
3059
3060 Label notVolatile, Done;
3061 __ movl(rdx, flags);
3062 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3063 __ andl(rdx, 0x1);
3064
3065 // field addresses
3066 const Address field(obj, off, Address::times_1, 0*wordSize);
3067 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3068
3069 Label notByte, notBool, notInt, notShort, notChar,
3070 notLong, notFloat, notObj, notDouble;
3071
3072 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3073
3074 assert(btos == 0, "change code, btos != 0");
3075 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3076 __ jcc(Assembler::notZero, notByte);
3077
3078 // btos
3079 {
3080 __ pop(btos);
3081 if (!is_static) pop_and_check_object(obj);
3082 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3083 __ movb(field, rax);
3084 if (!is_static && rc == may_rewrite) {
3085 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3086 }
3087 __ jmp(Done);
3088 }
3089
3090 __ bind(notByte);
3091 __ cmpl(flags, ztos);
3092 __ jcc(Assembler::notEqual, notBool);
3093
3094 // ztos
3095 {
3096 __ pop(ztos);
3097 if (!is_static) pop_and_check_object(obj);
3098 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3099 __ andl(rax, 0x1);
3100 __ movb(field, rax);
3101 if (!is_static && rc == may_rewrite) {
3102 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3103 }
3104 __ jmp(Done);
3105 }
3106
3107 __ bind(notBool);
3108 __ cmpl(flags, atos);
3109 __ jcc(Assembler::notEqual, notObj);
3110
3111 // atos
3112 {
3113 __ pop(atos);
3114 if (!is_static) pop_and_check_object(obj);
3115 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3116 // Store into the field
3117 do_oop_store(_masm, field, rax, _bs->kind(), false);
3118 if (!is_static && rc == may_rewrite) {
3119 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3120 }
3121 __ jmp(Done);
3122 }
3123
3124 __ bind(notObj);
3125 __ cmpl(flags, itos);
3126 __ jcc(Assembler::notEqual, notInt);
3127
3128 // itos
3129 {
3130 __ pop(itos);
3131 if (!is_static) pop_and_check_object(obj);
3132 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3133 __ movl(field, rax);
3134 if (!is_static && rc == may_rewrite) {
3135 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3136 }
3137 __ jmp(Done);
3138 }
3139
3140 __ bind(notInt);
3141 __ cmpl(flags, ctos);
3142 __ jcc(Assembler::notEqual, notChar);
3143
3144 // ctos
3145 {
3146 __ pop(ctos);
3147 if (!is_static) pop_and_check_object(obj);
3148 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3149 __ movw(field, rax);
3150 if (!is_static && rc == may_rewrite) {
3151 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3152 }
3153 __ jmp(Done);
3154 }
3155
3156 __ bind(notChar);
3157 __ cmpl(flags, stos);
3158 __ jcc(Assembler::notEqual, notShort);
3159
3160 // stos
3161 {
3162 __ pop(stos);
3163 if (!is_static) pop_and_check_object(obj);
3164 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3165 __ movw(field, rax);
3166 if (!is_static && rc == may_rewrite) {
3167 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3168 }
3169 __ jmp(Done);
3170 }
3171
3172 __ bind(notShort);
3173 __ cmpl(flags, ltos);
3174 __ jcc(Assembler::notEqual, notLong);
3175
3176 // ltos
3177 #ifdef _LP64
3178 {
3179 __ pop(ltos);
3180 if (!is_static) pop_and_check_object(obj);
3181 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3182 __ movq(field, rax);
3183 if (!is_static && rc == may_rewrite) {
3184 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3185 }
3186 __ jmp(Done);
3187 }
3188 #else
3189 {
3190 Label notVolatileLong;
3191 __ testl(rdx, rdx);
3192 __ jcc(Assembler::zero, notVolatileLong);
3193
3194 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3195 if (!is_static) pop_and_check_object(obj);
3196
3197 // Replace with real volatile test
3198 __ push(rdx);
3199 __ push(rax); // Must update atomically with FIST
3200 __ fild_d(Address(rsp,0)); // So load into FPU register
3201 __ fistp_d(field); // and put into memory atomically
3202 __ addptr(rsp, 2*wordSize);
3203 // volatile_barrier();
3204 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3205 Assembler::StoreStore));
3206 // Don't rewrite volatile version
3207 __ jmp(notVolatile);
3208
3209 __ bind(notVolatileLong);
3210
3211 __ pop(ltos); // overwrites rdx
3212 if (!is_static) pop_and_check_object(obj);
3213 __ movptr(hi, rdx);
3214 __ movptr(field, rax);
3215 // Don't rewrite to _fast_lputfield for potential volatile case.
3216 __ jmp(notVolatile);
3217 }
3218 #endif // _LP64
3219
3220 __ bind(notLong);
3221 __ cmpl(flags, ftos);
3222 __ jcc(Assembler::notEqual, notFloat);
3223
3224 // ftos
3225 {
3226 __ pop(ftos);
3227 if (!is_static) pop_and_check_object(obj);
3228 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3229 __ store_float(field);
3230 if (!is_static && rc == may_rewrite) {
3231 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3232 }
3233 __ jmp(Done);
3234 }
3235
3236 __ bind(notFloat);
3237 #ifdef ASSERT
3238 __ cmpl(flags, dtos);
3239 __ jcc(Assembler::notEqual, notDouble);
3240 #endif
3241
3242 // dtos
3243 {
3244 __ pop(dtos);
3245 if (!is_static) pop_and_check_object(obj);
3246 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3247 __ store_double(field);
3248 if (!is_static && rc == may_rewrite) {
3249 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3250 }
3251 }
3252
3253 #ifdef ASSERT
3254 __ jmp(Done);
3255
3256 __ bind(notDouble);
3257 __ stop("Bad state");
3258 #endif
3259
3260 __ bind(Done);
3261
3262 // Check for volatile store
3263 __ testl(rdx, rdx);
3264 __ jcc(Assembler::zero, notVolatile);
3265 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3266 Assembler::StoreStore));
3267 __ bind(notVolatile);
3268 }
3269
3270 void TemplateTable::putfield(int byte_no) {
3271 putfield_or_static(byte_no, false);
3272 }
3273
3274 void TemplateTable::nofast_putfield(int byte_no) {
3275 putfield_or_static(byte_no, false, may_not_rewrite);
3276 }
3277
3278 void TemplateTable::putstatic(int byte_no) {
3279 putfield_or_static(byte_no, true);
3280 }
3281
3282 void TemplateTable::jvmti_post_fast_field_mod() {
3283
3284 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3285
3286 if (JvmtiExport::can_post_field_modification()) {
3287 // Check to see if a field modification watch has been set before
3288 // we take the time to call into the VM.
3289 Label L2;
3290 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3291 __ testl(scratch, scratch);
3292 __ jcc(Assembler::zero, L2);
3293 __ pop_ptr(rbx); // copy the object pointer from tos
3294 __ verify_oop(rbx);
3295 __ push_ptr(rbx); // put the object pointer back on tos
3296 // Save tos values before call_VM() clobbers them. Since we have
3297 // to do it for every data type, we use the saved values as the
3298 // jvalue object.
3299 switch (bytecode()) { // load values into the jvalue object
3300 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3301 case Bytecodes::_fast_bputfield: // fall through
3302 case Bytecodes::_fast_zputfield: // fall through
3303 case Bytecodes::_fast_sputfield: // fall through
3304 case Bytecodes::_fast_cputfield: // fall through
3305 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3306 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3307 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3308 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3309
3310 default:
3311 ShouldNotReachHere();
3312 }
3313 __ mov(scratch, rsp); // points to jvalue on the stack
3314 // access constant pool cache entry
3315 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3316 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3317 __ verify_oop(rbx);
3318 // rbx: object pointer copied above
3319 // c_rarg2: cache entry pointer
3320 // c_rarg3: jvalue object on the stack
3321 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3322 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3323
3324 switch (bytecode()) { // restore tos values
3325 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3326 case Bytecodes::_fast_bputfield: // fall through
3327 case Bytecodes::_fast_zputfield: // fall through
3328 case Bytecodes::_fast_sputfield: // fall through
3329 case Bytecodes::_fast_cputfield: // fall through
3330 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3331 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3332 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3333 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3334 default: break;
3335 }
3336 __ bind(L2);
3337 }
3338 }
3339
3340 void TemplateTable::fast_storefield(TosState state) {
3341 transition(state, vtos);
3342
3343 ByteSize base = ConstantPoolCache::base_offset();
3344
3345 jvmti_post_fast_field_mod();
3346
3347 // access constant pool cache
3348 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3349
3350 // test for volatile with rdx but rdx is tos register for lputfield.
3351 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3352 in_bytes(base +
3353 ConstantPoolCacheEntry::flags_offset())));
3354
3355 // replace index with field offset from cache entry
3356 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3357 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3358
3359 // [jk] not needed currently
3360 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3361 // Assembler::StoreStore));
3362
3363 Label notVolatile;
3364 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3365 __ andl(rdx, 0x1);
3366
3367 // Get object from stack
3368 pop_and_check_object(rcx);
3369 oopDesc::bs()->interpreter_write_barrier(_masm, rcx);
3370
3371 // field address
3372 const Address field(rcx, rbx, Address::times_1);
3373
3374 // access field
3375 switch (bytecode()) {
3376 case Bytecodes::_fast_aputfield:
3377 do_oop_store(_masm, field, rax, _bs->kind(), false);
3378 break;
3379 case Bytecodes::_fast_lputfield:
3380 #ifdef _LP64
3381 __ movq(field, rax);
3382 #else
3383 __ stop("should not be rewritten");
3384 #endif
3385 break;
3386 case Bytecodes::_fast_iputfield:
3387 __ movl(field, rax);
3388 break;
3389 case Bytecodes::_fast_zputfield:
3390 __ andl(rax, 0x1); // boolean is true if LSB is 1
3391 // fall through to bputfield
3392 case Bytecodes::_fast_bputfield:
3393 __ movb(field, rax);
3394 break;
3395 case Bytecodes::_fast_sputfield:
3396 // fall through
3397 case Bytecodes::_fast_cputfield:
3398 __ movw(field, rax);
3399 break;
3400 case Bytecodes::_fast_fputfield:
3401 __ store_float(field);
3402 break;
3403 case Bytecodes::_fast_dputfield:
3404 __ store_double(field);
3405 break;
3406 default:
3407 ShouldNotReachHere();
3408 }
3409
3410 // Check for volatile store
3411 __ testl(rdx, rdx);
3412 __ jcc(Assembler::zero, notVolatile);
3413 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3414 Assembler::StoreStore));
3415 __ bind(notVolatile);
3416 }
3417
3418 void TemplateTable::fast_accessfield(TosState state) {
3419 transition(atos, state);
3420
3421 // Do the JVMTI work here to avoid disturbing the register state below
3422 if (JvmtiExport::can_post_field_access()) {
3423 // Check to see if a field access watch has been set before we
3424 // take the time to call into the VM.
3425 Label L1;
3426 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3427 __ testl(rcx, rcx);
3428 __ jcc(Assembler::zero, L1);
3429 // access constant pool cache entry
3430 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3431 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3432 __ verify_oop(rax);
3433 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3434 LP64_ONLY(__ mov(c_rarg1, rax));
3435 // c_rarg1: object pointer copied above
3436 // c_rarg2: cache entry pointer
3437 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3438 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3439 __ pop_ptr(rax); // restore object pointer
3440 __ bind(L1);
3441 }
3442
3443 // access constant pool cache
3444 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3445 // replace index with field offset from cache entry
3446 // [jk] not needed currently
3447 // if (os::is_MP()) {
3448 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3449 // in_bytes(ConstantPoolCache::base_offset() +
3450 // ConstantPoolCacheEntry::flags_offset())));
3451 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3452 // __ andl(rdx, 0x1);
3453 // }
3454 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3455 in_bytes(ConstantPoolCache::base_offset() +
3456 ConstantPoolCacheEntry::f2_offset())));
3457
3458 // rax: object
3459 __ verify_oop(rax);
3460 __ null_check(rax);
3461 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rax);
3462 Address field(rax, rbx, Address::times_1);
3463
3464 // access field
3465 switch (bytecode()) {
3466 case Bytecodes::_fast_agetfield:
3467 __ load_heap_oop(rax, field);
3468 __ verify_oop(rax);
3469 break;
3470 case Bytecodes::_fast_lgetfield:
3471 #ifdef _LP64
3472 __ movq(rax, field);
3473 #else
3474 __ stop("should not be rewritten");
3475 #endif
3476 break;
3477 case Bytecodes::_fast_igetfield:
3478 __ movl(rax, field);
3479 break;
3480 case Bytecodes::_fast_bgetfield:
3481 __ movsbl(rax, field);
3482 break;
3483 case Bytecodes::_fast_sgetfield:
3484 __ load_signed_short(rax, field);
3485 break;
3486 case Bytecodes::_fast_cgetfield:
3487 __ load_unsigned_short(rax, field);
3488 break;
3489 case Bytecodes::_fast_fgetfield:
3490 __ load_float(field);
3491 break;
3492 case Bytecodes::_fast_dgetfield:
3493 __ load_double(field);
3494 break;
3495 default:
3496 ShouldNotReachHere();
3497 }
3498 // [jk] not needed currently
3499 // if (os::is_MP()) {
3500 // Label notVolatile;
3501 // __ testl(rdx, rdx);
3502 // __ jcc(Assembler::zero, notVolatile);
3503 // __ membar(Assembler::LoadLoad);
3504 // __ bind(notVolatile);
3505 //};
3506 }
3507
3508 void TemplateTable::fast_xaccess(TosState state) {
3509 transition(vtos, state);
3510
3511 // get receiver
3512 __ movptr(rax, aaddress(0));
3513 // access constant pool cache
3514 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3515 __ movptr(rbx,
3516 Address(rcx, rdx, Address::times_ptr,
3517 in_bytes(ConstantPoolCache::base_offset() +
3518 ConstantPoolCacheEntry::f2_offset())));
3519 // make sure exception is reported in correct bcp range (getfield is
3520 // next instruction)
3521 __ increment(rbcp);
3522 __ null_check(rax);
3523 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rax);
3524 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3525 switch (state) {
3526 case itos:
3527 __ movl(rax, field);
3528 break;
3529 case atos:
3530 __ load_heap_oop(rax, field);
3531 __ verify_oop(rax);
3532 break;
3533 case ftos:
3534 __ load_float(field);
3535 break;
3536 default:
3537 ShouldNotReachHere();
3538 }
3539
3540 // [jk] not needed currently
3541 // if (os::is_MP()) {
3542 // Label notVolatile;
3543 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3544 // in_bytes(ConstantPoolCache::base_offset() +
3545 // ConstantPoolCacheEntry::flags_offset())));
3546 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3547 // __ testl(rdx, 0x1);
3548 // __ jcc(Assembler::zero, notVolatile);
3549 // __ membar(Assembler::LoadLoad);
3550 // __ bind(notVolatile);
3551 // }
3552
3553 __ decrement(rbcp);
3554 }
3555
3556 //-----------------------------------------------------------------------------
3557 // Calls
3558
3559 void TemplateTable::count_calls(Register method, Register temp) {
3560 // implemented elsewhere
3561 ShouldNotReachHere();
3562 }
3563
3564 void TemplateTable::prepare_invoke(int byte_no,
3565 Register method, // linked method (or i-klass)
3566 Register index, // itable index, MethodType, etc.
3567 Register recv, // if caller wants to see it
3568 Register flags // if caller wants to test it
3569 ) {
3570 // determine flags
3571 const Bytecodes::Code code = bytecode();
3572 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3573 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3574 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3575 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3576 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3577 const bool load_receiver = (recv != noreg);
3578 const bool save_flags = (flags != noreg);
3579 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3580 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3581 assert(flags == noreg || flags == rdx, "");
3582 assert(recv == noreg || recv == rcx, "");
3583
3584 // setup registers & access constant pool cache
3585 if (recv == noreg) recv = rcx;
3586 if (flags == noreg) flags = rdx;
3587 assert_different_registers(method, index, recv, flags);
3588
3589 // save 'interpreter return address'
3590 __ save_bcp();
3591
3592 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3593
3594 // maybe push appendix to arguments (just before return address)
3595 if (is_invokedynamic || is_invokehandle) {
3596 Label L_no_push;
3597 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3598 __ jcc(Assembler::zero, L_no_push);
3599 // Push the appendix as a trailing parameter.
3600 // This must be done before we get the receiver,
3601 // since the parameter_size includes it.
3602 __ push(rbx);
3603 __ mov(rbx, index);
3604 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3605 __ load_resolved_reference_at_index(index, rbx);
3606 __ pop(rbx);
3607 __ push(index); // push appendix (MethodType, CallSite, etc.)
3608 __ bind(L_no_push);
3609 }
3610
3611 // load receiver if needed (after appendix is pushed so parameter size is correct)
3612 // Note: no return address pushed yet
3613 if (load_receiver) {
3614 __ movl(recv, flags);
3615 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3616 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3617 const int receiver_is_at_end = -1; // back off one slot to get receiver
3618 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3619 __ movptr(recv, recv_addr);
3620 __ verify_oop(recv);
3621 }
3622
3623 if (save_flags) {
3624 __ movl(rbcp, flags);
3625 }
3626
3627 // compute return type
3628 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3629 // Make sure we don't need to mask flags after the above shift
3630 ConstantPoolCacheEntry::verify_tos_state_shift();
3631 // load return address
3632 {
3633 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3634 ExternalAddress table(table_addr);
3635 LP64_ONLY(__ lea(rscratch1, table));
3636 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3637 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3638 }
3639
3640 // push return address
3641 __ push(flags);
3642
3643 // Restore flags value from the constant pool cache, and restore rsi
3644 // for later null checks. r13 is the bytecode pointer
3645 if (save_flags) {
3646 __ movl(flags, rbcp);
3647 __ restore_bcp();
3648 }
3649 }
3650
3651 void TemplateTable::invokevirtual_helper(Register index,
3652 Register recv,
3653 Register flags) {
3654 // Uses temporary registers rax, rdx
3655 assert_different_registers(index, recv, rax, rdx);
3656 assert(index == rbx, "");
3657 assert(recv == rcx, "");
3658
3659 // Test for an invoke of a final method
3660 Label notFinal;
3661 __ movl(rax, flags);
3662 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3663 __ jcc(Assembler::zero, notFinal);
3664
3665 const Register method = index; // method must be rbx
3666 assert(method == rbx,
3667 "Method* must be rbx for interpreter calling convention");
3668
3669 // do the call - the index is actually the method to call
3670 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3671
3672 // It's final, need a null check here!
3673 __ null_check(recv);
3674
3675 // profile this call
3676 __ profile_final_call(rax);
3677 __ profile_arguments_type(rax, method, rbcp, true);
3678
3679 __ jump_from_interpreted(method, rax);
3680
3681 __ bind(notFinal);
3682
3683 // get receiver klass
3684 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3685 __ load_klass(rax, recv);
3686
3687 // profile this call
3688 __ profile_virtual_call(rax, rlocals, rdx);
3689 // get target Method* & entry point
3690 __ lookup_virtual_method(rax, index, method);
3691 __ profile_called_method(method, rdx, rbcp);
3692
3693 __ profile_arguments_type(rdx, method, rbcp, true);
3694 __ jump_from_interpreted(method, rdx);
3695 }
3696
3697 void TemplateTable::invokevirtual(int byte_no) {
3698 transition(vtos, vtos);
3699 assert(byte_no == f2_byte, "use this argument");
3700 prepare_invoke(byte_no,
3701 rbx, // method or vtable index
3702 noreg, // unused itable index
3703 rcx, rdx); // recv, flags
3704
3705 // rbx: index
3706 // rcx: receiver
3707 // rdx: flags
3708
3709 invokevirtual_helper(rbx, rcx, rdx);
3710 }
3711
3712 void TemplateTable::invokespecial(int byte_no) {
3713 transition(vtos, vtos);
3714 assert(byte_no == f1_byte, "use this argument");
3715 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3716 rcx); // get receiver also for null check
3717 __ verify_oop(rcx);
3718 __ null_check(rcx);
3719 // do the call
3720 __ profile_call(rax);
3721 __ profile_arguments_type(rax, rbx, rbcp, false);
3722 __ jump_from_interpreted(rbx, rax);
3723 }
3724
3725 void TemplateTable::invokestatic(int byte_no) {
3726 transition(vtos, vtos);
3727 assert(byte_no == f1_byte, "use this argument");
3728 prepare_invoke(byte_no, rbx); // get f1 Method*
3729 // do the call
3730 __ profile_call(rax);
3731 __ profile_arguments_type(rax, rbx, rbcp, false);
3732 __ jump_from_interpreted(rbx, rax);
3733 }
3734
3735
3736 void TemplateTable::fast_invokevfinal(int byte_no) {
3737 transition(vtos, vtos);
3738 assert(byte_no == f2_byte, "use this argument");
3739 __ stop("fast_invokevfinal not used on x86");
3740 }
3741
3742
3743 void TemplateTable::invokeinterface(int byte_no) {
3744 transition(vtos, vtos);
3745 assert(byte_no == f1_byte, "use this argument");
3746 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3747 rcx, rdx); // recv, flags
3748
3749 // rax: interface klass (from f1)
3750 // rbx: itable index (from f2)
3751 // rcx: receiver
3752 // rdx: flags
3753
3754 // Special case of invokeinterface called for virtual method of
3755 // java.lang.Object. See cpCacheOop.cpp for details.
3756 // This code isn't produced by javac, but could be produced by
3757 // another compliant java compiler.
3758 Label notMethod;
3759 __ movl(rlocals, rdx);
3760 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3761
3762 __ jcc(Assembler::zero, notMethod);
3763
3764 invokevirtual_helper(rbx, rcx, rdx);
3765 __ bind(notMethod);
3766
3767 // Get receiver klass into rdx - also a null check
3768 __ restore_locals(); // restore r14
3769 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3770 __ load_klass(rdx, rcx);
3771
3772 // profile this call
3773 __ profile_virtual_call(rdx, rbcp, rlocals);
3774
3775 Label no_such_interface, no_such_method;
3776
3777 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3778 rdx, rax, rbx,
3779 // outputs: method, scan temp. reg
3780 rbx, rbcp,
3781 no_such_interface);
3782
3783 // rbx: Method* to call
3784 // rcx: receiver
3785 // Check for abstract method error
3786 // Note: This should be done more efficiently via a throw_abstract_method_error
3787 // interpreter entry point and a conditional jump to it in case of a null
3788 // method.
3789 __ testptr(rbx, rbx);
3790 __ jcc(Assembler::zero, no_such_method);
3791
3792 __ profile_called_method(rbx, rbcp, rdx);
3793 __ profile_arguments_type(rdx, rbx, rbcp, true);
3794
3795 // do the call
3796 // rcx: receiver
3797 // rbx,: Method*
3798 __ jump_from_interpreted(rbx, rdx);
3799 __ should_not_reach_here();
3800
3801 // exception handling code follows...
3802 // note: must restore interpreter registers to canonical
3803 // state for exception handling to work correctly!
3804
3805 __ bind(no_such_method);
3806 // throw exception
3807 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3808 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3809 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3810 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3811 // the call_VM checks for exception, so we should never return here.
3812 __ should_not_reach_here();
3813
3814 __ bind(no_such_interface);
3815 // throw exception
3816 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3817 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3818 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3819 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3820 InterpreterRuntime::throw_IncompatibleClassChangeError));
3821 // the call_VM checks for exception, so we should never return here.
3822 __ should_not_reach_here();
3823 }
3824
3825 void TemplateTable::invokehandle(int byte_no) {
3826 transition(vtos, vtos);
3827 assert(byte_no == f1_byte, "use this argument");
3828 const Register rbx_method = rbx;
3829 const Register rax_mtype = rax;
3830 const Register rcx_recv = rcx;
3831 const Register rdx_flags = rdx;
3832
3833 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3834 __ verify_method_ptr(rbx_method);
3835 __ verify_oop(rcx_recv);
3836 __ null_check(rcx_recv);
3837
3838 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3839 // rbx: MH.invokeExact_MT method (from f2)
3840
3841 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3842
3843 // FIXME: profile the LambdaForm also
3844 __ profile_final_call(rax);
3845 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3846
3847 __ jump_from_interpreted(rbx_method, rdx);
3848 }
3849
3850 void TemplateTable::invokedynamic(int byte_no) {
3851 transition(vtos, vtos);
3852 assert(byte_no == f1_byte, "use this argument");
3853
3854 const Register rbx_method = rbx;
3855 const Register rax_callsite = rax;
3856
3857 prepare_invoke(byte_no, rbx_method, rax_callsite);
3858
3859 // rax: CallSite object (from cpool->resolved_references[f1])
3860 // rbx: MH.linkToCallSite method (from f2)
3861
3862 // Note: rax_callsite is already pushed by prepare_invoke
3863
3864 // %%% should make a type profile for any invokedynamic that takes a ref argument
3865 // profile this call
3866 __ profile_call(rbcp);
3867 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3868
3869 __ verify_oop(rax_callsite);
3870
3871 __ jump_from_interpreted(rbx_method, rdx);
3872 }
3873
3874 //-----------------------------------------------------------------------------
3875 // Allocation
3876
3877 void TemplateTable::_new() {
3878 transition(vtos, atos);
3879 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3880 Label slow_case;
3881 Label slow_case_no_pop;
3882 Label done;
3883 Label initialize_header;
3884 Label initialize_object; // including clearing the fields
3885 Label allocate_shared;
3886
3887 __ get_cpool_and_tags(rcx, rax);
3888
3889 // Make sure the class we're about to instantiate has been resolved.
3890 // This is done before loading InstanceKlass to be consistent with the order
3891 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3892 const int tags_offset = Array<u1>::base_offset_in_bytes();
3893 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3894 __ jcc(Assembler::notEqual, slow_case_no_pop);
3895
3896 // get InstanceKlass
3897 __ load_resolved_klass_at_index(rcx, rdx, rcx);
3898 __ push(rcx); // save the contexts of klass for initializing the header
3899
3900 // make sure klass is initialized & doesn't have finalizer
3901 // make sure klass is fully initialized
3902 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3903 __ jcc(Assembler::notEqual, slow_case);
3904
3905 // get instance_size in InstanceKlass (scaled to a count of bytes)
3906 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3907 // test to see if it has a finalizer or is malformed in some way
3908 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3909 __ jcc(Assembler::notZero, slow_case);
3910
3911 //
3912 // Allocate the instance
3913 // 1) Try to allocate in the TLAB
3914 // 2) if fail and the object is large allocate in the shared Eden
3915 // 3) if the above fails (or is not applicable), go to a slow case
3916 // (creates a new TLAB, etc.)
3917
3918 const bool allow_shared_alloc =
3919 Universe::heap()->supports_inline_contig_alloc();
3920
3921 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
3922 #ifndef _LP64
3923 if (UseTLAB || allow_shared_alloc) {
3924 __ get_thread(thread);
3925 }
3926 #endif // _LP64
3927
3928 if (UseTLAB) {
3929 uint oop_extra_words = Universe::heap()->oop_extra_words();
3930 if (oop_extra_words > 0) {
3931 __ addptr(rdx, oop_extra_words * HeapWordSize);
3932 }
3933
3934 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3935 __ lea(rbx, Address(rax, rdx, Address::times_1));
3936 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3937 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3938 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3939 Universe::heap()->compile_prepare_oop(_masm, rax);
3940 if (ZeroTLAB) {
3941 // the fields have been already cleared
3942 __ jmp(initialize_header);
3943 } else {
3944 // initialize both the header and fields
3945 __ jmp(initialize_object);
3946 }
3947 }
3948
3949 // Allocation in the shared Eden, if allowed.
3950 //
3951 // rdx: instance size in bytes
3952 if (allow_shared_alloc) {
3953 __ bind(allocate_shared);
3954
3955 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3956 ExternalAddress heap_end((address)Universe::heap()->end_addr());
3957
3958 Label retry;
3959 __ bind(retry);
3960 __ movptr(rax, heap_top);
3961 __ lea(rbx, Address(rax, rdx, Address::times_1));
3962 __ cmpptr(rbx, heap_end);
3963 __ jcc(Assembler::above, slow_case);
3964
3965 // Compare rax, with the top addr, and if still equal, store the new
3966 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3967 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3968 //
3969 // rax,: object begin
3970 // rbx,: object end
3971 // rdx: instance size in bytes
3972 __ locked_cmpxchgptr(rbx, heap_top);
3973
3974 // if someone beat us on the allocation, try again, otherwise continue
3975 __ jcc(Assembler::notEqual, retry);
3976
3977 __ incr_allocated_bytes(thread, rdx, 0);
3978 }
3979
3980 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3981 // The object is initialized before the header. If the object size is
3982 // zero, go directly to the header initialization.
3983 __ bind(initialize_object);
3984 __ decrement(rdx, sizeof(oopDesc));
3985 __ jcc(Assembler::zero, initialize_header);
3986
3987 // Initialize topmost object field, divide rdx by 8, check if odd and
3988 // test if zero.
3989 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3990 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3991
3992 // rdx must have been multiple of 8
3993 #ifdef ASSERT
3994 // make sure rdx was multiple of 8
3995 Label L;
3996 // Ignore partial flag stall after shrl() since it is debug VM
3997 __ jccb(Assembler::carryClear, L);
3998 __ stop("object size is not multiple of 2 - adjust this code");
3999 __ bind(L);
4000 // rdx must be > 0, no extra check needed here
4001 #endif
4002
4003 // initialize remaining object fields: rdx was a multiple of 8
4004 { Label loop;
4005 __ bind(loop);
4006 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4007 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4008 __ decrement(rdx);
4009 __ jcc(Assembler::notZero, loop);
4010 }
4011
4012 // initialize object header only.
4013 __ bind(initialize_header);
4014 if (UseBiasedLocking) {
4015 __ pop(rcx); // get saved klass back in the register.
4016 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4017 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4018 } else {
4019 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4020 (intptr_t)markOopDesc::prototype()); // header
4021 __ pop(rcx); // get saved klass back in the register.
4022 }
4023 #ifdef _LP64
4024 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4025 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
4026 #endif
4027 __ store_klass(rax, rcx); // klass
4028
4029 {
4030 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4031 // Trigger dtrace event for fastpath
4032 __ push(atos);
4033 __ call_VM_leaf(
4034 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4035 __ pop(atos);
4036 }
4037
4038 __ jmp(done);
4039 }
4040
4041 // slow case
4042 __ bind(slow_case);
4043 __ pop(rcx); // restore stack pointer to what it was when we came in.
4044 __ bind(slow_case_no_pop);
4045
4046 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4047 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4048
4049 __ get_constant_pool(rarg1);
4050 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4051 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4052 __ verify_oop(rax);
4053
4054 // continue
4055 __ bind(done);
4056 }
4057
4058 void TemplateTable::newarray() {
4059 transition(itos, atos);
4060 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4061 __ load_unsigned_byte(rarg1, at_bcp(1));
4062 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4063 rarg1, rax);
4064 }
4065
4066 void TemplateTable::anewarray() {
4067 transition(itos, atos);
4068
4069 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4070 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4071
4072 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4073 __ get_constant_pool(rarg1);
4074 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4075 rarg1, rarg2, rax);
4076 }
4077
4078 void TemplateTable::arraylength() {
4079 transition(atos, itos);
4080 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4081 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4082 }
4083
4084 void TemplateTable::checkcast() {
4085 transition(atos, atos);
4086 Label done, is_null, ok_is_subtype, quicked, resolved;
4087 __ testptr(rax, rax); // object is in rax
4088 __ jcc(Assembler::zero, is_null);
4089
4090 // Get cpool & tags index
4091 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4092 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4093 // See if bytecode has already been quicked
4094 __ cmpb(Address(rdx, rbx,
4095 Address::times_1,
4096 Array<u1>::base_offset_in_bytes()),
4097 JVM_CONSTANT_Class);
4098 __ jcc(Assembler::equal, quicked);
4099 __ push(atos); // save receiver for result, and for GC
4100 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4101
4102 // vm_result_2 has metadata result
4103 #ifndef _LP64
4104 // borrow rdi from locals
4105 __ get_thread(rdi);
4106 __ get_vm_result_2(rax, rdi);
4107 __ restore_locals();
4108 #else
4109 __ get_vm_result_2(rax, r15_thread);
4110 #endif
4111
4112 __ pop_ptr(rdx); // restore receiver
4113 __ jmpb(resolved);
4114
4115 // Get superklass in rax and subklass in rbx
4116 __ bind(quicked);
4117 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4118 __ load_resolved_klass_at_index(rcx, rbx, rax);
4119
4120 __ bind(resolved);
4121 __ load_klass(rbx, rdx);
4122
4123 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4124 // Superklass in rax. Subklass in rbx.
4125 __ gen_subtype_check(rbx, ok_is_subtype);
4126
4127 // Come here on failure
4128 __ push_ptr(rdx);
4129 // object is at TOS
4130 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4131
4132 // Come here on success
4133 __ bind(ok_is_subtype);
4134 __ mov(rax, rdx); // Restore object in rdx
4135
4136 // Collect counts on whether this check-cast sees NULLs a lot or not.
4137 if (ProfileInterpreter) {
4138 __ jmp(done);
4139 __ bind(is_null);
4140 __ profile_null_seen(rcx);
4141 } else {
4142 __ bind(is_null); // same as 'done'
4143 }
4144 __ bind(done);
4145 }
4146
4147 void TemplateTable::instanceof() {
4148 transition(atos, itos);
4149 Label done, is_null, ok_is_subtype, quicked, resolved;
4150 __ testptr(rax, rax);
4151 __ jcc(Assembler::zero, is_null);
4152
4153 // Get cpool & tags index
4154 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4155 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4156 // See if bytecode has already been quicked
4157 __ cmpb(Address(rdx, rbx,
4158 Address::times_1,
4159 Array<u1>::base_offset_in_bytes()),
4160 JVM_CONSTANT_Class);
4161 __ jcc(Assembler::equal, quicked);
4162
4163 __ push(atos); // save receiver for result, and for GC
4164 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4165 // vm_result_2 has metadata result
4166
4167 #ifndef _LP64
4168 // borrow rdi from locals
4169 __ get_thread(rdi);
4170 __ get_vm_result_2(rax, rdi);
4171 __ restore_locals();
4172 #else
4173 __ get_vm_result_2(rax, r15_thread);
4174 #endif
4175
4176 __ pop_ptr(rdx); // restore receiver
4177 __ verify_oop(rdx);
4178 __ load_klass(rdx, rdx);
4179 __ jmpb(resolved);
4180
4181 // Get superklass in rax and subklass in rdx
4182 __ bind(quicked);
4183 __ load_klass(rdx, rax);
4184 __ load_resolved_klass_at_index(rcx, rbx, rax);
4185
4186 __ bind(resolved);
4187
4188 // Generate subtype check. Blows rcx, rdi
4189 // Superklass in rax. Subklass in rdx.
4190 __ gen_subtype_check(rdx, ok_is_subtype);
4191
4192 // Come here on failure
4193 __ xorl(rax, rax);
4194 __ jmpb(done);
4195 // Come here on success
4196 __ bind(ok_is_subtype);
4197 __ movl(rax, 1);
4198
4199 // Collect counts on whether this test sees NULLs a lot or not.
4200 if (ProfileInterpreter) {
4201 __ jmp(done);
4202 __ bind(is_null);
4203 __ profile_null_seen(rcx);
4204 } else {
4205 __ bind(is_null); // same as 'done'
4206 }
4207 __ bind(done);
4208 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4209 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4210 }
4211
4212
4213 //----------------------------------------------------------------------------------------------------
4214 // Breakpoints
4215 void TemplateTable::_breakpoint() {
4216 // Note: We get here even if we are single stepping..
4217 // jbug insists on setting breakpoints at every bytecode
4218 // even if we are in single step mode.
4219
4220 transition(vtos, vtos);
4221
4222 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4223
4224 // get the unpatched byte code
4225 __ get_method(rarg);
4226 __ call_VM(noreg,
4227 CAST_FROM_FN_PTR(address,
4228 InterpreterRuntime::get_original_bytecode_at),
4229 rarg, rbcp);
4230 __ mov(rbx, rax); // why?
4231
4232 // post the breakpoint event
4233 __ get_method(rarg);
4234 __ call_VM(noreg,
4235 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4236 rarg, rbcp);
4237
4238 // complete the execution of original bytecode
4239 __ dispatch_only_normal(vtos);
4240 }
4241
4242 //-----------------------------------------------------------------------------
4243 // Exceptions
4244
4245 void TemplateTable::athrow() {
4246 transition(atos, vtos);
4247 __ null_check(rax);
4248 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4249 }
4250
4251 //-----------------------------------------------------------------------------
4252 // Synchronization
4253 //
4254 // Note: monitorenter & exit are symmetric routines; which is reflected
4255 // in the assembly code structure as well
4256 //
4257 // Stack layout:
4258 //
4259 // [expressions ] <--- rsp = expression stack top
4260 // ..
4261 // [expressions ]
4262 // [monitor entry] <--- monitor block top = expression stack bot
4263 // ..
4264 // [monitor entry]
4265 // [frame data ] <--- monitor block bot
4266 // ...
4267 // [saved rbp ] <--- rbp
4268 void TemplateTable::monitorenter() {
4269 transition(atos, vtos);
4270
4271 // check for NULL object
4272 __ null_check(rax);
4273
4274 oopDesc::bs()->interpreter_write_barrier(_masm, rax);
4275
4276 const Address monitor_block_top(
4277 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4278 const Address monitor_block_bot(
4279 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4280 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4281
4282 Label allocated;
4283
4284 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4285 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4286 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4287
4288 // initialize entry pointer
4289 __ xorl(rmon, rmon); // points to free slot or NULL
4290
4291 // find a free slot in the monitor block (result in rmon)
4292 {
4293 Label entry, loop, exit;
4294 __ movptr(rtop, monitor_block_top); // points to current entry,
4295 // starting with top-most entry
4296 __ lea(rbot, monitor_block_bot); // points to word before bottom
4297 // of monitor block
4298 __ jmpb_if_possible(entry);
4299
4300 __ bind(loop);
4301 // check if current entry is used
4302 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4303 // if not used then remember entry in rmon
4304 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4305 // check if current entry is for same object
4306 __ shenandoah_lock_check(rtop); // Invariant
4307 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4308 // if same object then stop searching
4309 __ jccb(Assembler::equal, exit);
4310 // otherwise advance to next entry
4311 __ addptr(rtop, entry_size);
4312 __ bind(entry);
4313 // check if bottom reached
4314 __ cmpptr(rtop, rbot);
4315 // if not at bottom then check this entry
4316 __ jcc(Assembler::notEqual, loop);
4317 __ bind(exit);
4318 }
4319
4320 __ testptr(rmon, rmon); // check if a slot has been found
4321 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4322
4323 // allocate one if there's no free slot
4324 {
4325 Label entry, loop;
4326 // 1. compute new pointers // rsp: old expression stack top
4327 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4328 __ subptr(rsp, entry_size); // move expression stack top
4329 __ subptr(rmon, entry_size); // move expression stack bottom
4330 __ mov(rtop, rsp); // set start value for copy loop
4331 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4332 __ jmp(entry);
4333 // 2. move expression stack contents
4334 __ bind(loop);
4335 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4336 // word from old location
4337 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4338 __ addptr(rtop, wordSize); // advance to next word
4339 __ bind(entry);
4340 __ cmpptr(rtop, rmon); // check if bottom reached
4341 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4342 // copy next word
4343 }
4344
4345 // call run-time routine
4346 // rmon: points to monitor entry
4347 __ bind(allocated);
4348
4349 // Increment bcp to point to the next bytecode, so exception
4350 // handling for async. exceptions work correctly.
4351 // The object has already been poped from the stack, so the
4352 // expression stack looks correct.
4353 __ increment(rbcp);
4354
4355 // store object
4356 __ shenandoah_store_addr_check(rax); // Invariant
4357 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4358 __ lock_object(rmon);
4359
4360 // check to make sure this monitor doesn't cause stack overflow after locking
4361 __ save_bcp(); // in case of exception
4362 __ generate_stack_overflow_check(0);
4363
4364 // The bcp has already been incremented. Just need to dispatch to
4365 // next instruction.
4366 __ dispatch_next(vtos);
4367 }
4368
4369 void TemplateTable::monitorexit() {
4370 transition(atos, vtos);
4371
4372 // check for NULL object
4373 __ null_check(rax);
4374
4375 oopDesc::bs()->interpreter_write_barrier(_masm, rax);
4376
4377 const Address monitor_block_top(
4378 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4379 const Address monitor_block_bot(
4380 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4381 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4382
4383 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4384 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4385
4386 Label found;
4387
4388 // find matching slot
4389 {
4390 Label entry, loop;
4391 __ movptr(rtop, monitor_block_top); // points to current entry,
4392 // starting with top-most entry
4393 __ lea(rbot, monitor_block_bot); // points to word before bottom
4394 // of monitor block
4395 __ jmpb_if_possible(entry);
4396
4397 __ bind(loop);
4398 // check if current entry is for same object
4399 __ shenandoah_lock_check(rtop); // Invariant
4400 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4401 // if same object then stop searching
4402 __ jcc(Assembler::equal, found);
4403 // otherwise advance to next entry
4404 __ addptr(rtop, entry_size);
4405 __ bind(entry);
4406 // check if bottom reached
4407 __ cmpptr(rtop, rbot);
4408 // if not at bottom then check this entry
4409 __ jcc(Assembler::notEqual, loop);
4410 }
4411
4412 // error handling. Unlocking was not block-structured
4413 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4414 InterpreterRuntime::throw_illegal_monitor_state_exception));
4415 __ should_not_reach_here();
4416
4417 // call run-time routine
4418 __ bind(found);
4419 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4420 __ unlock_object(rtop);
4421 __ pop_ptr(rax); // discard object
4422 }
4423
4424 // Wide instructions
4425 void TemplateTable::wide() {
4426 transition(vtos, vtos);
4427 __ load_unsigned_byte(rbx, at_bcp(1));
4428 ExternalAddress wtable((address)Interpreter::_wentry_point);
4429 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4430 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4431 }
4432
4433 // Multi arrays
4434 void TemplateTable::multianewarray() {
4435 transition(vtos, atos);
4436
4437 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4438 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4439 // last dim is on top of stack; we want address of first one:
4440 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4441 // the latter wordSize to point to the beginning of the array.
4442 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4443 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4444 __ load_unsigned_byte(rbx, at_bcp(3));
4445 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4446 }