1 /*
2 * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.inline.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #ifndef CC_INTERP
42
43 #define __ _masm->
44
45 // Global Register Names
46 Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
47 Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
48
49 // Platform-dependent initialization
50 void TemplateTable::pd_initialize() {
51 // No x86 specific initialization
52 }
53
54 // Address Computation: local variables
55 static inline Address iaddress(int n) {
56 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
57 }
58
59 static inline Address laddress(int n) {
60 return iaddress(n + 1);
61 }
62
63 #ifndef _LP64
64 static inline Address haddress(int n) {
65 return iaddress(n + 0);
66 }
67 #endif
68
69 static inline Address faddress(int n) {
70 return iaddress(n);
71 }
72
73 static inline Address daddress(int n) {
74 return laddress(n);
75 }
76
77 static inline Address aaddress(int n) {
78 return iaddress(n);
79 }
80
81 static inline Address iaddress(Register r) {
82 return Address(rlocals, r, Address::times_ptr);
83 }
84
85 static inline Address laddress(Register r) {
86 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
87 }
88
89 #ifndef _LP64
90 static inline Address haddress(Register r) {
91 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
92 }
93 #endif
94
95 static inline Address faddress(Register r) {
96 return iaddress(r);
97 }
98
99 static inline Address daddress(Register r) {
100 return laddress(r);
101 }
102
103 static inline Address aaddress(Register r) {
104 return iaddress(r);
105 }
106
107
108 // expression stack
109 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
110 // data beyond the rsp which is potentially unsafe in an MT environment;
111 // an interrupt may overwrite that data.)
112 static inline Address at_rsp () {
113 return Address(rsp, 0);
114 }
115
116 // At top of Java expression stack which may be different than esp(). It
117 // isn't for category 1 objects.
118 static inline Address at_tos () {
119 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
120 }
121
122 static inline Address at_tos_p1() {
123 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
124 }
125
126 static inline Address at_tos_p2() {
127 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
128 }
129
130 // Condition conversion
131 static Assembler::Condition j_not(TemplateTable::Condition cc) {
132 switch (cc) {
133 case TemplateTable::equal : return Assembler::notEqual;
134 case TemplateTable::not_equal : return Assembler::equal;
135 case TemplateTable::less : return Assembler::greaterEqual;
136 case TemplateTable::less_equal : return Assembler::greater;
137 case TemplateTable::greater : return Assembler::lessEqual;
138 case TemplateTable::greater_equal: return Assembler::less;
139 }
140 ShouldNotReachHere();
141 return Assembler::zero;
142 }
143
144
145
146 // Miscelaneous helper routines
147 // Store an oop (or NULL) at the address described by obj.
148 // If val == noreg this means store a NULL
149
150
151 static void do_oop_store(InterpreterMacroAssembler* _masm,
152 Address obj,
153 Register val,
154 BarrierSet::Name barrier,
155 bool precise) {
156 assert(val == noreg || val == rax, "parameter is just for looks");
157 switch (barrier) {
158 #if INCLUDE_ALL_GCS
159 case BarrierSet::G1SATBCT:
160 case BarrierSet::G1SATBCTLogging:
161 {
162 // flatten object address if needed
163 // We do it regardless of precise because we need the registers
164 if (obj.index() == noreg && obj.disp() == 0) {
165 if (obj.base() != rdx) {
166 __ movptr(rdx, obj.base());
167 }
168 } else {
169 __ lea(rdx, obj);
170 }
171
172 Register rtmp = LP64_ONLY(r8) NOT_LP64(rsi);
173 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
174
175 NOT_LP64(__ get_thread(rcx));
176 NOT_LP64(__ save_bcp());
177
178 __ g1_write_barrier_pre(rdx /* obj */,
179 rbx /* pre_val */,
180 rthread /* thread */,
181 rtmp /* tmp */,
182 val != noreg /* tosca_live */,
183 false /* expand_call */);
184 if (val == noreg) {
185 __ store_heap_oop_null(Address(rdx, 0));
186 } else {
187 // G1 barrier needs uncompressed oop for region cross check.
188 Register new_val = val;
189 if (UseCompressedOops) {
190 new_val = rbx;
191 __ movptr(new_val, val);
192 }
193 __ store_heap_oop(Address(rdx, 0), val);
194 __ g1_write_barrier_post(rdx /* store_adr */,
195 new_val /* new_val */,
196 rthread /* thread */,
197 rtmp /* tmp */,
198 rbx /* tmp2 */);
199 }
200 NOT_LP64( __ restore_bcp());
201 }
202 break;
203 #endif // INCLUDE_ALL_GCS
204 case BarrierSet::CardTableModRef:
205 case BarrierSet::CardTableExtension:
206 {
207 if (val == noreg) {
208 __ store_heap_oop_null(obj);
209 } else {
210 __ store_heap_oop(obj, val);
211 // flatten object address if needed
212 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
213 __ store_check(obj.base());
214 } else {
215 __ lea(rdx, obj);
216 __ store_check(rdx);
217 }
218 }
219 }
220 break;
221 case BarrierSet::ModRef:
222 if (val == noreg) {
223 __ store_heap_oop_null(obj);
224 } else {
225 __ store_heap_oop(obj, val);
226 }
227 break;
228 default :
229 ShouldNotReachHere();
230
231 }
232 }
233
234 Address TemplateTable::at_bcp(int offset) {
235 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
236 return Address(rbcp, offset);
237 }
238
239
240 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
241 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
242 int byte_no) {
243 if (!RewriteBytecodes) return;
244 Label L_patch_done;
245
246 switch (bc) {
247 case Bytecodes::_fast_aputfield:
248 case Bytecodes::_fast_bputfield:
249 case Bytecodes::_fast_cputfield:
250 case Bytecodes::_fast_dputfield:
251 case Bytecodes::_fast_fputfield:
252 case Bytecodes::_fast_iputfield:
253 case Bytecodes::_fast_lputfield:
254 case Bytecodes::_fast_sputfield:
255 {
256 // We skip bytecode quickening for putfield instructions when
257 // the put_code written to the constant pool cache is zero.
258 // This is required so that every execution of this instruction
259 // calls out to InterpreterRuntime::resolve_get_put to do
260 // additional, required work.
261 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
262 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
263 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
264 __ movl(bc_reg, bc);
265 __ cmpl(temp_reg, (int) 0);
266 __ jcc(Assembler::zero, L_patch_done); // don't patch
267 }
268 break;
269 default:
270 assert(byte_no == -1, "sanity");
271 // the pair bytecodes have already done the load.
272 if (load_bc_into_bc_reg) {
273 __ movl(bc_reg, bc);
274 }
275 }
276
277 if (JvmtiExport::can_post_breakpoint()) {
278 Label L_fast_patch;
279 // if a breakpoint is present we can't rewrite the stream directly
280 __ movzbl(temp_reg, at_bcp(0));
281 __ cmpl(temp_reg, Bytecodes::_breakpoint);
282 __ jcc(Assembler::notEqual, L_fast_patch);
283 __ get_method(temp_reg);
284 // Let breakpoint table handling rewrite to quicker bytecode
285 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
286 #ifndef ASSERT
287 __ jmpb(L_patch_done);
288 #else
289 __ jmp(L_patch_done);
290 #endif
291 __ bind(L_fast_patch);
292 }
293
294 #ifdef ASSERT
295 Label L_okay;
296 __ load_unsigned_byte(temp_reg, at_bcp(0));
297 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
298 __ jcc(Assembler::equal, L_okay);
299 __ cmpl(temp_reg, bc_reg);
300 __ jcc(Assembler::equal, L_okay);
301 __ stop("patching the wrong bytecode");
302 __ bind(L_okay);
303 #endif
304
305 // patch bytecode
306 __ movb(at_bcp(0), bc_reg);
307 __ bind(L_patch_done);
308 }
309 // Individual instructions
310
311
312 void TemplateTable::nop() {
313 transition(vtos, vtos);
314 // nothing to do
315 }
316
317 void TemplateTable::shouldnotreachhere() {
318 transition(vtos, vtos);
319 __ stop("shouldnotreachhere bytecode");
320 }
321
322 void TemplateTable::aconst_null() {
323 transition(vtos, atos);
324 __ xorl(rax, rax);
325 }
326
327 void TemplateTable::iconst(int value) {
328 transition(vtos, itos);
329 if (value == 0) {
330 __ xorl(rax, rax);
331 } else {
332 __ movl(rax, value);
333 }
334 }
335
336 void TemplateTable::lconst(int value) {
337 transition(vtos, ltos);
338 if (value == 0) {
339 __ xorl(rax, rax);
340 } else {
341 __ movl(rax, value);
342 }
343 #ifndef _LP64
344 assert(value >= 0, "check this code");
345 __ xorptr(rdx, rdx);
346 #endif
347 }
348
349
350
351 void TemplateTable::fconst(int value) {
352 transition(vtos, ftos);
353 #ifdef _LP64
354 static float one = 1.0f, two = 2.0f;
355 switch (value) {
356 case 0:
357 __ xorps(xmm0, xmm0);
358 break;
359 case 1:
360 __ movflt(xmm0, ExternalAddress((address) &one));
361 break;
362 case 2:
363 __ movflt(xmm0, ExternalAddress((address) &two));
364 break;
365 default:
366 ShouldNotReachHere();
367 break;
368 }
369 #else
370 if (value == 0) { __ fldz();
371 } else if (value == 1) { __ fld1();
372 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
373 } else { ShouldNotReachHere();
374 }
375 #endif
376 }
377
378 void TemplateTable::dconst(int value) {
379 transition(vtos, dtos);
380 #ifdef _LP64
381 static double one = 1.0;
382 switch (value) {
383 case 0:
384 __ xorpd(xmm0, xmm0);
385 break;
386 case 1:
387 __ movdbl(xmm0, ExternalAddress((address) &one));
388 break;
389 default:
390 ShouldNotReachHere();
391 break;
392 }
393
394 #else
395 if (value == 0) { __ fldz();
396 } else if (value == 1) { __ fld1();
397 } else { ShouldNotReachHere();
398 }
399 #endif
400 }
401
402 void TemplateTable::bipush() {
403 transition(vtos, itos);
404 __ load_signed_byte(rax, at_bcp(1));
405 }
406
407 void TemplateTable::sipush() {
408 transition(vtos, itos);
409 __ load_unsigned_short(rax, at_bcp(1));
410 __ bswapl(rax);
411 __ sarl(rax, 16);
412 }
413
414 void TemplateTable::ldc(bool wide) {
415 transition(vtos, vtos);
416 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
417 Label call_ldc, notFloat, notClass, Done;
418
419 if (wide) {
420 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
421 } else {
422 __ load_unsigned_byte(rbx, at_bcp(1));
423 }
424
425 __ get_cpool_and_tags(rcx, rax);
426 const int base_offset = ConstantPool::header_size() * wordSize;
427 const int tags_offset = Array<u1>::base_offset_in_bytes();
428
429 // get type
430 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
431
432 // unresolved class - get the resolved class
433 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
434 __ jccb(Assembler::equal, call_ldc);
435
436 // unresolved class in error state - call into runtime to throw the error
437 // from the first resolution attempt
438 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
439 __ jccb(Assembler::equal, call_ldc);
440
441 // resolved class - need to call vm to get java mirror of the class
442 __ cmpl(rdx, JVM_CONSTANT_Class);
443 __ jcc(Assembler::notEqual, notClass);
444
445 __ bind(call_ldc);
446
447 __ movl(rarg, wide);
448 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
449
450 __ push(atos);
451 __ jmp(Done);
452
453 __ bind(notClass);
454 __ cmpl(rdx, JVM_CONSTANT_Float);
455 __ jccb(Assembler::notEqual, notFloat);
456
457 // ftos
458 LP64_ONLY(__ movflt(xmm0, Address(rcx, rbx, Address::times_8, base_offset)));
459 NOT_LP64(__ fld_s( Address(rcx, rbx, Address::times_ptr, base_offset)));
460 __ push(ftos);
461 __ jmp(Done);
462
463 __ bind(notFloat);
464 #ifdef ASSERT
465 {
466 Label L;
467 __ cmpl(rdx, JVM_CONSTANT_Integer);
468 __ jcc(Assembler::equal, L);
469 // String and Object are rewritten to fast_aldc
470 __ stop("unexpected tag type in ldc");
471 __ bind(L);
472 }
473 #endif
474 // itos JVM_CONSTANT_Integer only
475 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
476 __ push(itos);
477 __ bind(Done);
478 }
479
480 // Fast path for caching oop constants.
481 void TemplateTable::fast_aldc(bool wide) {
482 transition(vtos, atos);
483
484 Register result = rax;
485 Register tmp = rdx;
486 int index_size = wide ? sizeof(u2) : sizeof(u1);
487
488 Label resolved;
489
490 // We are resolved if the resolved reference cache entry contains a
491 // non-null object (String, MethodType, etc.)
492 assert_different_registers(result, tmp);
493 __ get_cache_index_at_bcp(tmp, 1, index_size);
494 __ load_resolved_reference_at_index(result, tmp);
495 __ testl(result, result);
496 __ jcc(Assembler::notZero, resolved);
497
498 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
499
500 // first time invocation - must resolve first
501 __ movl(tmp, (int)bytecode());
502 __ call_VM(result, entry, tmp);
503
504 __ bind(resolved);
505
506 if (VerifyOops) {
507 __ verify_oop(result);
508 }
509 }
510
511 void TemplateTable::ldc2_w() {
512 transition(vtos, vtos);
513 Label Long, Done;
514 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
515
516 __ get_cpool_and_tags(rcx, rax);
517 const int base_offset = ConstantPool::header_size() * wordSize;
518 const int tags_offset = Array<u1>::base_offset_in_bytes();
519
520 // get type
521 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
522 JVM_CONSTANT_Double);
523 __ jccb(Assembler::notEqual, Long);
524
525 // dtos
526 LP64_ONLY(__ movdbl(xmm0, Address(rcx, rbx, Address::times_8, base_offset)));
527 NOT_LP64(__ fld_d( Address(rcx, rbx, Address::times_ptr, base_offset)));
528 __ push(dtos);
529
530 __ jmpb(Done);
531 __ bind(Long);
532
533 // ltos
534 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
535 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
536 __ push(ltos);
537
538 __ bind(Done);
539 }
540
541 void TemplateTable::locals_index(Register reg, int offset) {
542 __ load_unsigned_byte(reg, at_bcp(offset));
543 __ negptr(reg);
544 }
545
546 void TemplateTable::iload() {
547 iload_internal();
548 }
549
550 void TemplateTable::nofast_iload() {
551 iload_internal(may_not_rewrite);
552 }
553
554 void TemplateTable::iload_internal(RewriteControl rc) {
555 transition(vtos, itos);
556 if (RewriteFrequentPairs && rc == may_rewrite) {
557 Label rewrite, done;
558 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
559 LP64_ONLY(assert(rbx != bc, "register damaged"));
560
561 // get next byte
562 __ load_unsigned_byte(rbx,
563 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
564 // if _iload, wait to rewrite to iload2. We only want to rewrite the
565 // last two iloads in a pair. Comparing against fast_iload means that
566 // the next bytecode is neither an iload or a caload, and therefore
567 // an iload pair.
568 __ cmpl(rbx, Bytecodes::_iload);
569 __ jcc(Assembler::equal, done);
570
571 __ cmpl(rbx, Bytecodes::_fast_iload);
572 __ movl(bc, Bytecodes::_fast_iload2);
573
574 __ jccb(Assembler::equal, rewrite);
575
576 // if _caload, rewrite to fast_icaload
577 __ cmpl(rbx, Bytecodes::_caload);
578 __ movl(bc, Bytecodes::_fast_icaload);
579 __ jccb(Assembler::equal, rewrite);
580
581 // rewrite so iload doesn't check again.
582 __ movl(bc, Bytecodes::_fast_iload);
583
584 // rewrite
585 // bc: fast bytecode
586 __ bind(rewrite);
587 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
588 __ bind(done);
589 }
590
591 // Get the local value into tos
592 locals_index(rbx);
593 __ movl(rax, iaddress(rbx));
594 }
595
596 void TemplateTable::fast_iload2() {
597 transition(vtos, itos);
598 locals_index(rbx);
599 __ movl(rax, iaddress(rbx));
600 __ push(itos);
601 locals_index(rbx, 3);
602 __ movl(rax, iaddress(rbx));
603 }
604
605 void TemplateTable::fast_iload() {
606 transition(vtos, itos);
607 locals_index(rbx);
608 __ movl(rax, iaddress(rbx));
609 }
610
611 void TemplateTable::lload() {
612 transition(vtos, ltos);
613 locals_index(rbx);
614 __ movptr(rax, laddress(rbx));
615 NOT_LP64(__ movl(rdx, haddress(rbx)));
616 }
617
618 void TemplateTable::fload() {
619 transition(vtos, ftos);
620 locals_index(rbx);
621 LP64_ONLY(__ movflt(xmm0, faddress(rbx)));
622 NOT_LP64(__ fld_s(faddress(rbx)));
623 }
624
625 void TemplateTable::dload() {
626 transition(vtos, dtos);
627 locals_index(rbx);
628 LP64_ONLY(__ movdbl(xmm0, daddress(rbx)));
629 NOT_LP64(__ fld_d(daddress(rbx)));
630 }
631
632 void TemplateTable::aload() {
633 transition(vtos, atos);
634 locals_index(rbx);
635 __ movptr(rax, aaddress(rbx));
636 }
637
638 void TemplateTable::locals_index_wide(Register reg) {
639 __ load_unsigned_short(reg, at_bcp(2));
640 __ bswapl(reg);
641 __ shrl(reg, 16);
642 __ negptr(reg);
643 }
644
645 void TemplateTable::wide_iload() {
646 transition(vtos, itos);
647 locals_index_wide(rbx);
648 __ movl(rax, iaddress(rbx));
649 }
650
651 void TemplateTable::wide_lload() {
652 transition(vtos, ltos);
653 locals_index_wide(rbx);
654 __ movptr(rax, laddress(rbx));
655 NOT_LP64(__ movl(rdx, haddress(rbx)));
656 }
657
658 void TemplateTable::wide_fload() {
659 transition(vtos, ftos);
660 locals_index_wide(rbx);
661 LP64_ONLY(__ movflt(xmm0, faddress(rbx)));
662 NOT_LP64(__ fld_s(faddress(rbx)));
663 }
664
665 void TemplateTable::wide_dload() {
666 transition(vtos, dtos);
667 locals_index_wide(rbx);
668 LP64_ONLY(__ movdbl(xmm0, daddress(rbx)));
669 NOT_LP64(__ fld_d(daddress(rbx)));
670 }
671
672 void TemplateTable::wide_aload() {
673 transition(vtos, atos);
674 locals_index_wide(rbx);
675 __ movptr(rax, aaddress(rbx));
676 }
677
678 void TemplateTable::index_check(Register array, Register index) {
679 // Pop ptr into array
680 __ pop_ptr(array);
681 index_check_without_pop(array, index);
682 }
683
684 void TemplateTable::index_check_without_pop(Register array, Register index) {
685 // destroys rbx
686 // check array
687 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
688 // sign extend index for use by indexed load
689 __ movl2ptr(index, index);
690 // check index
691 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
692 if (index != rbx) {
693 // ??? convention: move aberrant index into rbx for exception message
694 assert(rbx != array, "different registers");
695 __ movl(rbx, index);
696 }
697 __ jump_cc(Assembler::aboveEqual,
698 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
699 }
700
701
702 void TemplateTable::iaload() {
703 transition(itos, itos);
704 // rax: index
705 // rdx: array
706 index_check(rdx, rax); // kills rbx
707 __ movl(rax, Address(rdx, rax,
708 Address::times_4,
709 arrayOopDesc::base_offset_in_bytes(T_INT)));
710 }
711
712 void TemplateTable::laload() {
713 transition(itos, ltos);
714 // rax: index
715 // rdx: array
716 index_check(rdx, rax); // kills rbx
717 NOT_LP64(__ mov(rbx, rax));
718 // rbx,: index
719 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
720 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
721 }
722
723
724
725 void TemplateTable::faload() {
726 transition(itos, ftos);
727 // rax: index
728 // rdx: array
729 index_check(rdx, rax); // kills rbx
730 LP64_ONLY(__ movflt(xmm0, Address(rdx, rax,
731 Address::times_4,
732 arrayOopDesc::base_offset_in_bytes(T_FLOAT))));
733 NOT_LP64(__ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))));
734 }
735
736 void TemplateTable::daload() {
737 transition(itos, dtos);
738 // rax: index
739 // rdx: array
740 index_check(rdx, rax); // kills rbx
741 LP64_ONLY(__ movdbl(xmm0, Address(rdx, rax,
742 Address::times_8,
743 arrayOopDesc::base_offset_in_bytes(T_DOUBLE))));
744 NOT_LP64(__ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))));
745 }
746
747 void TemplateTable::aaload() {
748 transition(itos, atos);
749 // rax: index
750 // rdx: array
751 index_check(rdx, rax); // kills rbx
752 __ load_heap_oop(rax, Address(rdx, rax,
753 UseCompressedOops ? Address::times_4 : Address::times_ptr,
754 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
755 }
756
757 void TemplateTable::baload() {
758 transition(itos, itos);
759 // rax: index
760 // rdx: array
761 index_check(rdx, rax); // kills rbx
762 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
763 }
764
765 void TemplateTable::caload() {
766 transition(itos, itos);
767 // rax: index
768 // rdx: array
769 index_check(rdx, rax); // kills rbx
770 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
771 }
772
773 // iload followed by caload frequent pair
774 void TemplateTable::fast_icaload() {
775 transition(vtos, itos);
776 // load index out of locals
777 locals_index(rbx);
778 __ movl(rax, iaddress(rbx));
779
780 // rax: index
781 // rdx: array
782 index_check(rdx, rax); // kills rbx
783 __ load_unsigned_short(rax,
784 Address(rdx, rax,
785 Address::times_2,
786 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
787 }
788
789
790 void TemplateTable::saload() {
791 transition(itos, itos);
792 // rax: index
793 // rdx: array
794 index_check(rdx, rax); // kills rbx
795 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
796 }
797
798 void TemplateTable::iload(int n) {
799 transition(vtos, itos);
800 __ movl(rax, iaddress(n));
801 }
802
803 void TemplateTable::lload(int n) {
804 transition(vtos, ltos);
805 __ movptr(rax, laddress(n));
806 NOT_LP64(__ movptr(rdx, haddress(n)));
807 }
808
809 void TemplateTable::fload(int n) {
810 transition(vtos, ftos);
811 LP64_ONLY(__ movflt(xmm0, faddress(n)));
812 NOT_LP64(__ fld_s(faddress(n)));
813 }
814
815 void TemplateTable::dload(int n) {
816 transition(vtos, dtos);
817 LP64_ONLY(__ movdbl(xmm0, daddress(n)));
818 NOT_LP64(__ fld_d(daddress(n)));
819 }
820
821 void TemplateTable::aload(int n) {
822 transition(vtos, atos);
823 __ movptr(rax, aaddress(n));
824 }
825
826 void TemplateTable::aload_0() {
827 aload_0_internal();
828 }
829
830 void TemplateTable::nofast_aload_0() {
831 aload_0_internal(may_not_rewrite);
832 }
833
834 void TemplateTable::aload_0_internal(RewriteControl rc) {
835 transition(vtos, atos);
836 // According to bytecode histograms, the pairs:
837 //
838 // _aload_0, _fast_igetfield
839 // _aload_0, _fast_agetfield
840 // _aload_0, _fast_fgetfield
841 //
842 // occur frequently. If RewriteFrequentPairs is set, the (slow)
843 // _aload_0 bytecode checks if the next bytecode is either
844 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
845 // rewrites the current bytecode into a pair bytecode; otherwise it
846 // rewrites the current bytecode into _fast_aload_0 that doesn't do
847 // the pair check anymore.
848 //
849 // Note: If the next bytecode is _getfield, the rewrite must be
850 // delayed, otherwise we may miss an opportunity for a pair.
851 //
852 // Also rewrite frequent pairs
853 // aload_0, aload_1
854 // aload_0, iload_1
855 // These bytecodes with a small amount of code are most profitable
856 // to rewrite
857 if (RewriteFrequentPairs && rc == may_rewrite) {
858 Label rewrite, done;
859
860 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
861 LP64_ONLY(assert(rbx != bc, "register damaged"));
862
863 // get next byte
864 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
865
866 // do actual aload_0
867 aload(0);
868
869 // if _getfield then wait with rewrite
870 __ cmpl(rbx, Bytecodes::_getfield);
871 __ jcc(Assembler::equal, done);
872
873 // if _igetfield then reqrite to _fast_iaccess_0
874 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
875 __ cmpl(rbx, Bytecodes::_fast_igetfield);
876 __ movl(bc, Bytecodes::_fast_iaccess_0);
877 __ jccb(Assembler::equal, rewrite);
878
879 // if _agetfield then reqrite to _fast_aaccess_0
880 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
881 __ cmpl(rbx, Bytecodes::_fast_agetfield);
882 __ movl(bc, Bytecodes::_fast_aaccess_0);
883 __ jccb(Assembler::equal, rewrite);
884
885 // if _fgetfield then reqrite to _fast_faccess_0
886 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
887 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
888 __ movl(bc, Bytecodes::_fast_faccess_0);
889 __ jccb(Assembler::equal, rewrite);
890
891 // else rewrite to _fast_aload0
892 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
893 __ movl(bc, Bytecodes::_fast_aload_0);
894
895 // rewrite
896 // bc: fast bytecode
897 __ bind(rewrite);
898 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
899
900 __ bind(done);
901 } else {
902 aload(0);
903 }
904 }
905
906 void TemplateTable::istore() {
907 transition(itos, vtos);
908 locals_index(rbx);
909 __ movl(iaddress(rbx), rax);
910 }
911
912
913 void TemplateTable::lstore() {
914 transition(ltos, vtos);
915 locals_index(rbx);
916 __ movptr(laddress(rbx), rax);
917 NOT_LP64(__ movptr(haddress(rbx), rdx));
918 }
919
920 void TemplateTable::fstore() {
921 transition(ftos, vtos);
922 locals_index(rbx);
923 LP64_ONLY(__ movflt(faddress(rbx), xmm0));
924 NOT_LP64(__ fstp_s(faddress(rbx)));
925 }
926
927 void TemplateTable::dstore() {
928 transition(dtos, vtos);
929 locals_index(rbx);
930 LP64_ONLY(__ movdbl(daddress(rbx), xmm0));
931 NOT_LP64(__ fstp_d(daddress(rbx)));
932 }
933
934 void TemplateTable::astore() {
935 transition(vtos, vtos);
936 __ pop_ptr(rax);
937 locals_index(rbx);
938 __ movptr(aaddress(rbx), rax);
939 }
940
941 void TemplateTable::wide_istore() {
942 transition(vtos, vtos);
943 __ pop_i();
944 locals_index_wide(rbx);
945 __ movl(iaddress(rbx), rax);
946 }
947
948 void TemplateTable::wide_lstore() {
949 transition(vtos, vtos);
950 NOT_LP64(__ pop_l(rax, rdx));
951 LP64_ONLY(__ pop_l());
952 locals_index_wide(rbx);
953 __ movptr(laddress(rbx), rax);
954 NOT_LP64(__ movl(haddress(rbx), rdx));
955 }
956
957 void TemplateTable::wide_fstore() {
958 #ifdef _LP64
959 transition(vtos, vtos);
960 __ pop_f();
961 locals_index_wide(rbx);
962 __ movflt(faddress(rbx), xmm0);
963 #else
964 wide_istore();
965 #endif
966 }
967
968 void TemplateTable::wide_dstore() {
969 #ifdef _LP64
970 transition(vtos, vtos);
971 __ pop_d();
972 locals_index_wide(rbx);
973 __ movdbl(daddress(rbx), xmm0);
974 #else
975 wide_lstore();
976 #endif
977 }
978
979 void TemplateTable::wide_astore() {
980 transition(vtos, vtos);
981 __ pop_ptr(rax);
982 locals_index_wide(rbx);
983 __ movptr(aaddress(rbx), rax);
984 }
985
986 void TemplateTable::iastore() {
987 transition(itos, vtos);
988 __ pop_i(rbx);
989 // rax: value
990 // rbx: index
991 // rdx: array
992 index_check(rdx, rbx); // prefer index in rbx
993 __ movl(Address(rdx, rbx,
994 Address::times_4,
995 arrayOopDesc::base_offset_in_bytes(T_INT)),
996 rax);
997 }
998
999 void TemplateTable::lastore() {
1000 transition(ltos, vtos);
1001 __ pop_i(rbx);
1002 // rax,: low(value)
1003 // rcx: array
1004 // rdx: high(value)
1005 index_check(rcx, rbx); // prefer index in rbx,
1006 // rbx,: index
1007 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1008 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1009 }
1010
1011
1012 void TemplateTable::fastore() {
1013 transition(ftos, vtos);
1014 __ pop_i(rbx);
1015 // xmm0: value
1016 // rbx: index
1017 // rdx: array
1018 index_check(rdx, rbx); // prefer index in rbx
1019 LP64_ONLY(__ movflt(Address(rdx, rbx,
1020 Address::times_4,
1021 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
1022 xmm0));
1023 NOT_LP64(__ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT))));
1024 }
1025
1026 void TemplateTable::dastore() {
1027 transition(dtos, vtos);
1028 __ pop_i(rbx);
1029 // xmm0: value
1030 // rbx: index
1031 // rdx: array
1032 index_check(rdx, rbx); // prefer index in rbx
1033 LP64_ONLY(__ movdbl(Address(rdx, rbx,
1034 Address::times_8,
1035 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
1036 xmm0));
1037 NOT_LP64(__ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE))));
1038 }
1039
1040 void TemplateTable::aastore() {
1041 Label is_null, ok_is_subtype, done;
1042 transition(vtos, vtos);
1043 // stack: ..., array, index, value
1044 __ movptr(rax, at_tos()); // value
1045 __ movl(rcx, at_tos_p1()); // index
1046 __ movptr(rdx, at_tos_p2()); // array
1047
1048 Address element_address(rdx, rcx,
1049 UseCompressedOops? Address::times_4 : Address::times_ptr,
1050 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1051
1052 index_check_without_pop(rdx, rcx); // kills rbx
1053 __ testptr(rax, rax);
1054 __ jcc(Assembler::zero, is_null);
1055
1056 // Move subklass into rbx
1057 __ load_klass(rbx, rax);
1058 // Move superklass into rax
1059 __ load_klass(rax, rdx);
1060 __ movptr(rax, Address(rax,
1061 ObjArrayKlass::element_klass_offset()));
1062 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1063 __ lea(rdx, element_address);
1064
1065 // Generate subtype check. Blows rcx, rdi
1066 // Superklass in rax. Subklass in rbx.
1067 __ gen_subtype_check(rbx, ok_is_subtype);
1068
1069 // Come here on failure
1070 // object is at TOS
1071 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1072
1073 // Come here on success
1074 __ bind(ok_is_subtype);
1075
1076 // Get the value we will store
1077 __ movptr(rax, at_tos());
1078 // Now store using the appropriate barrier
1079 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1080 __ jmp(done);
1081
1082 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1083 __ bind(is_null);
1084 __ profile_null_seen(rbx);
1085
1086 // Store a NULL
1087 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1088
1089 // Pop stack arguments
1090 __ bind(done);
1091 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1092 }
1093
1094 void TemplateTable::bastore() {
1095 transition(itos, vtos);
1096 __ pop_i(rbx);
1097 // rax: value
1098 // rbx: index
1099 // rdx: array
1100 index_check(rdx, rbx); // prefer index in rbx
1101 __ movb(Address(rdx, rbx,
1102 Address::times_1,
1103 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1104 rax);
1105 }
1106
1107 void TemplateTable::castore() {
1108 transition(itos, vtos);
1109 __ pop_i(rbx);
1110 // rax: value
1111 // rbx: index
1112 // rdx: array
1113 index_check(rdx, rbx); // prefer index in rbx
1114 __ movw(Address(rdx, rbx,
1115 Address::times_2,
1116 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1117 rax);
1118 }
1119
1120
1121 void TemplateTable::sastore() {
1122 castore();
1123 }
1124
1125 void TemplateTable::istore(int n) {
1126 transition(itos, vtos);
1127 __ movl(iaddress(n), rax);
1128 }
1129
1130 void TemplateTable::lstore(int n) {
1131 transition(ltos, vtos);
1132 __ movptr(laddress(n), rax);
1133 NOT_LP64(__ movptr(haddress(n), rdx));
1134 }
1135
1136 void TemplateTable::fstore(int n) {
1137 transition(ftos, vtos);
1138 LP64_ONLY(__ movflt(faddress(n), xmm0));
1139 NOT_LP64(__ fstp_s(faddress(n)));
1140 }
1141
1142 void TemplateTable::dstore(int n) {
1143 transition(dtos, vtos);
1144 LP64_ONLY(__ movdbl(daddress(n), xmm0));
1145 NOT_LP64(__ fstp_d(daddress(n)));
1146 }
1147
1148
1149 void TemplateTable::astore(int n) {
1150 transition(vtos, vtos);
1151 __ pop_ptr(rax);
1152 __ movptr(aaddress(n), rax);
1153 }
1154
1155 void TemplateTable::pop() {
1156 transition(vtos, vtos);
1157 __ addptr(rsp, Interpreter::stackElementSize);
1158 }
1159
1160 void TemplateTable::pop2() {
1161 transition(vtos, vtos);
1162 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1163 }
1164
1165
1166 void TemplateTable::dup() {
1167 transition(vtos, vtos);
1168 __ load_ptr(0, rax);
1169 __ push_ptr(rax);
1170 // stack: ..., a, a
1171 }
1172
1173 void TemplateTable::dup_x1() {
1174 transition(vtos, vtos);
1175 // stack: ..., a, b
1176 __ load_ptr( 0, rax); // load b
1177 __ load_ptr( 1, rcx); // load a
1178 __ store_ptr(1, rax); // store b
1179 __ store_ptr(0, rcx); // store a
1180 __ push_ptr(rax); // push b
1181 // stack: ..., b, a, b
1182 }
1183
1184 void TemplateTable::dup_x2() {
1185 transition(vtos, vtos);
1186 // stack: ..., a, b, c
1187 __ load_ptr( 0, rax); // load c
1188 __ load_ptr( 2, rcx); // load a
1189 __ store_ptr(2, rax); // store c in a
1190 __ push_ptr(rax); // push c
1191 // stack: ..., c, b, c, c
1192 __ load_ptr( 2, rax); // load b
1193 __ store_ptr(2, rcx); // store a in b
1194 // stack: ..., c, a, c, c
1195 __ store_ptr(1, rax); // store b in c
1196 // stack: ..., c, a, b, c
1197 }
1198
1199 void TemplateTable::dup2() {
1200 transition(vtos, vtos);
1201 // stack: ..., a, b
1202 __ load_ptr(1, rax); // load a
1203 __ push_ptr(rax); // push a
1204 __ load_ptr(1, rax); // load b
1205 __ push_ptr(rax); // push b
1206 // stack: ..., a, b, a, b
1207 }
1208
1209
1210 void TemplateTable::dup2_x1() {
1211 transition(vtos, vtos);
1212 // stack: ..., a, b, c
1213 __ load_ptr( 0, rcx); // load c
1214 __ load_ptr( 1, rax); // load b
1215 __ push_ptr(rax); // push b
1216 __ push_ptr(rcx); // push c
1217 // stack: ..., a, b, c, b, c
1218 __ store_ptr(3, rcx); // store c in b
1219 // stack: ..., a, c, c, b, c
1220 __ load_ptr( 4, rcx); // load a
1221 __ store_ptr(2, rcx); // store a in 2nd c
1222 // stack: ..., a, c, a, b, c
1223 __ store_ptr(4, rax); // store b in a
1224 // stack: ..., b, c, a, b, c
1225 }
1226
1227 void TemplateTable::dup2_x2() {
1228 transition(vtos, vtos);
1229 // stack: ..., a, b, c, d
1230 __ load_ptr( 0, rcx); // load d
1231 __ load_ptr( 1, rax); // load c
1232 __ push_ptr(rax); // push c
1233 __ push_ptr(rcx); // push d
1234 // stack: ..., a, b, c, d, c, d
1235 __ load_ptr( 4, rax); // load b
1236 __ store_ptr(2, rax); // store b in d
1237 __ store_ptr(4, rcx); // store d in b
1238 // stack: ..., a, d, c, b, c, d
1239 __ load_ptr( 5, rcx); // load a
1240 __ load_ptr( 3, rax); // load c
1241 __ store_ptr(3, rcx); // store a in c
1242 __ store_ptr(5, rax); // store c in a
1243 // stack: ..., c, d, a, b, c, d
1244 }
1245
1246 void TemplateTable::swap() {
1247 transition(vtos, vtos);
1248 // stack: ..., a, b
1249 __ load_ptr( 1, rcx); // load a
1250 __ load_ptr( 0, rax); // load b
1251 __ store_ptr(0, rcx); // store a in b
1252 __ store_ptr(1, rax); // store b in a
1253 // stack: ..., b, a
1254 }
1255
1256 void TemplateTable::iop2(Operation op) {
1257 transition(itos, itos);
1258 switch (op) {
1259 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1260 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1261 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1262 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1263 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1264 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1265 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1266 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1267 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1268 default : ShouldNotReachHere();
1269 }
1270 }
1271
1272 void TemplateTable::lop2(Operation op) {
1273 transition(ltos, ltos);
1274 #ifdef _LP64
1275 switch (op) {
1276 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1277 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1278 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1279 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1280 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1281 default : ShouldNotReachHere();
1282 }
1283 #else
1284 __ pop_l(rbx, rcx);
1285 switch (op) {
1286 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1287 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1288 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1289 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1290 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1291 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1292 default : ShouldNotReachHere();
1293 }
1294 #endif
1295 }
1296
1297 void TemplateTable::idiv() {
1298 transition(itos, itos);
1299 __ movl(rcx, rax);
1300 __ pop_i(rax);
1301 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1302 // they are not equal, one could do a normal division (no correction
1303 // needed), which may speed up this implementation for the common case.
1304 // (see also JVM spec., p.243 & p.271)
1305 __ corrected_idivl(rcx);
1306 }
1307
1308 void TemplateTable::irem() {
1309 transition(itos, itos);
1310 __ movl(rcx, rax);
1311 __ pop_i(rax);
1312 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1313 // they are not equal, one could do a normal division (no correction
1314 // needed), which may speed up this implementation for the common case.
1315 // (see also JVM spec., p.243 & p.271)
1316 __ corrected_idivl(rcx);
1317 __ movl(rax, rdx);
1318 }
1319
1320 void TemplateTable::lmul() {
1321 transition(ltos, ltos);
1322 #ifdef _LP64
1323 __ pop_l(rdx);
1324 __ imulq(rax, rdx);
1325 #else
1326 __ pop_l(rbx, rcx);
1327 __ push(rcx); __ push(rbx);
1328 __ push(rdx); __ push(rax);
1329 __ lmul(2 * wordSize, 0);
1330 __ addptr(rsp, 4 * wordSize); // take off temporaries
1331 #endif
1332 }
1333
1334 void TemplateTable::ldiv() {
1335 transition(ltos, ltos);
1336 #ifdef _LP64
1337 __ mov(rcx, rax);
1338 __ pop_l(rax);
1339 // generate explicit div0 check
1340 __ testq(rcx, rcx);
1341 __ jump_cc(Assembler::zero,
1342 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1343 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1344 // they are not equal, one could do a normal division (no correction
1345 // needed), which may speed up this implementation for the common case.
1346 // (see also JVM spec., p.243 & p.271)
1347 __ corrected_idivq(rcx); // kills rbx
1348 #else
1349 __ pop_l(rbx, rcx);
1350 __ push(rcx); __ push(rbx);
1351 __ push(rdx); __ push(rax);
1352 // check if y = 0
1353 __ orl(rax, rdx);
1354 __ jump_cc(Assembler::zero,
1355 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1356 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1357 __ addptr(rsp, 4 * wordSize); // take off temporaries
1358 #endif
1359 }
1360
1361 void TemplateTable::lrem() {
1362 transition(ltos, ltos);
1363 #ifdef _LP64
1364 __ mov(rcx, rax);
1365 __ pop_l(rax);
1366 __ testq(rcx, rcx);
1367 __ jump_cc(Assembler::zero,
1368 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1369 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1370 // they are not equal, one could do a normal division (no correction
1371 // needed), which may speed up this implementation for the common case.
1372 // (see also JVM spec., p.243 & p.271)
1373 __ corrected_idivq(rcx); // kills rbx
1374 __ mov(rax, rdx);
1375 #else
1376 __ pop_l(rbx, rcx);
1377 __ push(rcx); __ push(rbx);
1378 __ push(rdx); __ push(rax);
1379 // check if y = 0
1380 __ orl(rax, rdx);
1381 __ jump_cc(Assembler::zero,
1382 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1383 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1384 __ addptr(rsp, 4 * wordSize);
1385 #endif
1386 }
1387
1388 void TemplateTable::lshl() {
1389 transition(itos, ltos);
1390 __ movl(rcx, rax); // get shift count
1391 #ifdef _LP64
1392 __ pop_l(rax); // get shift value
1393 __ shlq(rax);
1394 #else
1395 __ pop_l(rax, rdx); // get shift value
1396 __ lshl(rdx, rax);
1397 #endif
1398 }
1399
1400 void TemplateTable::lshr() {
1401 #ifdef _LP64
1402 transition(itos, ltos);
1403 __ movl(rcx, rax); // get shift count
1404 __ pop_l(rax); // get shift value
1405 __ sarq(rax);
1406 #else
1407 transition(itos, ltos);
1408 __ mov(rcx, rax); // get shift count
1409 __ pop_l(rax, rdx); // get shift value
1410 __ lshr(rdx, rax, true);
1411 #endif
1412 }
1413
1414 void TemplateTable::lushr() {
1415 transition(itos, ltos);
1416 #ifdef _LP64
1417 __ movl(rcx, rax); // get shift count
1418 __ pop_l(rax); // get shift value
1419 __ shrq(rax);
1420 #else
1421 __ mov(rcx, rax); // get shift count
1422 __ pop_l(rax, rdx); // get shift value
1423 __ lshr(rdx, rax);
1424 #endif
1425 }
1426
1427 void TemplateTable::fop2(Operation op) {
1428 transition(ftos, ftos);
1429 #ifdef _LP64
1430 switch (op) {
1431 case add:
1432 __ addss(xmm0, at_rsp());
1433 __ addptr(rsp, Interpreter::stackElementSize);
1434 break;
1435 case sub:
1436 __ movflt(xmm1, xmm0);
1437 __ pop_f(xmm0);
1438 __ subss(xmm0, xmm1);
1439 break;
1440 case mul:
1441 __ mulss(xmm0, at_rsp());
1442 __ addptr(rsp, Interpreter::stackElementSize);
1443 break;
1444 case div:
1445 __ movflt(xmm1, xmm0);
1446 __ pop_f(xmm0);
1447 __ divss(xmm0, xmm1);
1448 break;
1449 case rem:
1450 __ movflt(xmm1, xmm0);
1451 __ pop_f(xmm0);
1452 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1453 break;
1454 default:
1455 ShouldNotReachHere();
1456 break;
1457 }
1458 #else
1459 switch (op) {
1460 case add: __ fadd_s (at_rsp()); break;
1461 case sub: __ fsubr_s(at_rsp()); break;
1462 case mul: __ fmul_s (at_rsp()); break;
1463 case div: __ fdivr_s(at_rsp()); break;
1464 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1465 default : ShouldNotReachHere();
1466 }
1467 __ f2ieee();
1468 __ pop(rax); // pop float thing off
1469 #endif
1470 }
1471
1472 void TemplateTable::dop2(Operation op) {
1473 transition(dtos, dtos);
1474 #ifdef _LP64
1475 switch (op) {
1476 case add:
1477 __ addsd(xmm0, at_rsp());
1478 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1479 break;
1480 case sub:
1481 __ movdbl(xmm1, xmm0);
1482 __ pop_d(xmm0);
1483 __ subsd(xmm0, xmm1);
1484 break;
1485 case mul:
1486 __ mulsd(xmm0, at_rsp());
1487 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1488 break;
1489 case div:
1490 __ movdbl(xmm1, xmm0);
1491 __ pop_d(xmm0);
1492 __ divsd(xmm0, xmm1);
1493 break;
1494 case rem:
1495 __ movdbl(xmm1, xmm0);
1496 __ pop_d(xmm0);
1497 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1498 break;
1499 default:
1500 ShouldNotReachHere();
1501 break;
1502 }
1503 #else
1504 switch (op) {
1505 case add: __ fadd_d (at_rsp()); break;
1506 case sub: __ fsubr_d(at_rsp()); break;
1507 case mul: {
1508 Label L_strict;
1509 Label L_join;
1510 const Address access_flags (rcx, Method::access_flags_offset());
1511 __ get_method(rcx);
1512 __ movl(rcx, access_flags);
1513 __ testl(rcx, JVM_ACC_STRICT);
1514 __ jccb(Assembler::notZero, L_strict);
1515 __ fmul_d (at_rsp());
1516 __ jmpb(L_join);
1517 __ bind(L_strict);
1518 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1519 __ fmulp();
1520 __ fmul_d (at_rsp());
1521 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1522 __ fmulp();
1523 __ bind(L_join);
1524 break;
1525 }
1526 case div: {
1527 Label L_strict;
1528 Label L_join;
1529 const Address access_flags (rcx, Method::access_flags_offset());
1530 __ get_method(rcx);
1531 __ movl(rcx, access_flags);
1532 __ testl(rcx, JVM_ACC_STRICT);
1533 __ jccb(Assembler::notZero, L_strict);
1534 __ fdivr_d(at_rsp());
1535 __ jmp(L_join);
1536 __ bind(L_strict);
1537 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1538 __ fmul_d (at_rsp());
1539 __ fdivrp();
1540 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1541 __ fmulp();
1542 __ bind(L_join);
1543 break;
1544 }
1545 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1546 default : ShouldNotReachHere();
1547 }
1548 __ d2ieee();
1549 // Pop double precision number from rsp.
1550 __ pop(rax);
1551 __ pop(rdx);
1552 #endif
1553 }
1554
1555 void TemplateTable::ineg() {
1556 transition(itos, itos);
1557 __ negl(rax);
1558 }
1559
1560 void TemplateTable::lneg() {
1561 transition(ltos, ltos);
1562 LP64_ONLY(__ negq(rax));
1563 NOT_LP64(__ lneg(rdx, rax));
1564 }
1565
1566 #ifdef _LP64
1567 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1568 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1569 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1570 // of 128-bits operands for SSE instructions.
1571 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1572 // Store the value to a 128-bits operand.
1573 operand[0] = lo;
1574 operand[1] = hi;
1575 return operand;
1576 }
1577
1578 // Buffer for 128-bits masks used by SSE instructions.
1579 static jlong float_signflip_pool[2*2];
1580 static jlong double_signflip_pool[2*2];
1581 #endif
1582
1583 void TemplateTable::fneg() {
1584 transition(ftos, ftos);
1585 #ifdef _LP64
1586 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], 0x8000000080000000, 0x8000000080000000);
1587 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1588 #else
1589 __ fchs();
1590 #endif
1591 }
1592
1593 void TemplateTable::dneg() {
1594 transition(dtos, dtos);
1595 #ifdef _LP64
1596 static jlong *double_signflip = double_quadword(&double_signflip_pool[1], 0x8000000000000000, 0x8000000000000000);
1597 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1598 #else
1599 __ fchs();
1600 #endif
1601 }
1602
1603 void TemplateTable::iinc() {
1604 transition(vtos, vtos);
1605 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1606 locals_index(rbx);
1607 __ addl(iaddress(rbx), rdx);
1608 }
1609
1610 void TemplateTable::wide_iinc() {
1611 transition(vtos, vtos);
1612 __ movl(rdx, at_bcp(4)); // get constant
1613 locals_index_wide(rbx);
1614 __ bswapl(rdx); // swap bytes & sign-extend constant
1615 __ sarl(rdx, 16);
1616 __ addl(iaddress(rbx), rdx);
1617 // Note: should probably use only one movl to get both
1618 // the index and the constant -> fix this
1619 }
1620
1621 void TemplateTable::convert() {
1622 #ifdef _LP64
1623 // Checking
1624 #ifdef ASSERT
1625 {
1626 TosState tos_in = ilgl;
1627 TosState tos_out = ilgl;
1628 switch (bytecode()) {
1629 case Bytecodes::_i2l: // fall through
1630 case Bytecodes::_i2f: // fall through
1631 case Bytecodes::_i2d: // fall through
1632 case Bytecodes::_i2b: // fall through
1633 case Bytecodes::_i2c: // fall through
1634 case Bytecodes::_i2s: tos_in = itos; break;
1635 case Bytecodes::_l2i: // fall through
1636 case Bytecodes::_l2f: // fall through
1637 case Bytecodes::_l2d: tos_in = ltos; break;
1638 case Bytecodes::_f2i: // fall through
1639 case Bytecodes::_f2l: // fall through
1640 case Bytecodes::_f2d: tos_in = ftos; break;
1641 case Bytecodes::_d2i: // fall through
1642 case Bytecodes::_d2l: // fall through
1643 case Bytecodes::_d2f: tos_in = dtos; break;
1644 default : ShouldNotReachHere();
1645 }
1646 switch (bytecode()) {
1647 case Bytecodes::_l2i: // fall through
1648 case Bytecodes::_f2i: // fall through
1649 case Bytecodes::_d2i: // fall through
1650 case Bytecodes::_i2b: // fall through
1651 case Bytecodes::_i2c: // fall through
1652 case Bytecodes::_i2s: tos_out = itos; break;
1653 case Bytecodes::_i2l: // fall through
1654 case Bytecodes::_f2l: // fall through
1655 case Bytecodes::_d2l: tos_out = ltos; break;
1656 case Bytecodes::_i2f: // fall through
1657 case Bytecodes::_l2f: // fall through
1658 case Bytecodes::_d2f: tos_out = ftos; break;
1659 case Bytecodes::_i2d: // fall through
1660 case Bytecodes::_l2d: // fall through
1661 case Bytecodes::_f2d: tos_out = dtos; break;
1662 default : ShouldNotReachHere();
1663 }
1664 transition(tos_in, tos_out);
1665 }
1666 #endif // ASSERT
1667
1668 static const int64_t is_nan = 0x8000000000000000L;
1669
1670 // Conversion
1671 switch (bytecode()) {
1672 case Bytecodes::_i2l:
1673 __ movslq(rax, rax);
1674 break;
1675 case Bytecodes::_i2f:
1676 __ cvtsi2ssl(xmm0, rax);
1677 break;
1678 case Bytecodes::_i2d:
1679 __ cvtsi2sdl(xmm0, rax);
1680 break;
1681 case Bytecodes::_i2b:
1682 __ movsbl(rax, rax);
1683 break;
1684 case Bytecodes::_i2c:
1685 __ movzwl(rax, rax);
1686 break;
1687 case Bytecodes::_i2s:
1688 __ movswl(rax, rax);
1689 break;
1690 case Bytecodes::_l2i:
1691 __ movl(rax, rax);
1692 break;
1693 case Bytecodes::_l2f:
1694 __ cvtsi2ssq(xmm0, rax);
1695 break;
1696 case Bytecodes::_l2d:
1697 __ cvtsi2sdq(xmm0, rax);
1698 break;
1699 case Bytecodes::_f2i:
1700 {
1701 Label L;
1702 __ cvttss2sil(rax, xmm0);
1703 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1704 __ jcc(Assembler::notEqual, L);
1705 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1706 __ bind(L);
1707 }
1708 break;
1709 case Bytecodes::_f2l:
1710 {
1711 Label L;
1712 __ cvttss2siq(rax, xmm0);
1713 // NaN or overflow/underflow?
1714 __ cmp64(rax, ExternalAddress((address) &is_nan));
1715 __ jcc(Assembler::notEqual, L);
1716 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1717 __ bind(L);
1718 }
1719 break;
1720 case Bytecodes::_f2d:
1721 __ cvtss2sd(xmm0, xmm0);
1722 break;
1723 case Bytecodes::_d2i:
1724 {
1725 Label L;
1726 __ cvttsd2sil(rax, xmm0);
1727 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1728 __ jcc(Assembler::notEqual, L);
1729 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1730 __ bind(L);
1731 }
1732 break;
1733 case Bytecodes::_d2l:
1734 {
1735 Label L;
1736 __ cvttsd2siq(rax, xmm0);
1737 // NaN or overflow/underflow?
1738 __ cmp64(rax, ExternalAddress((address) &is_nan));
1739 __ jcc(Assembler::notEqual, L);
1740 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1741 __ bind(L);
1742 }
1743 break;
1744 case Bytecodes::_d2f:
1745 __ cvtsd2ss(xmm0, xmm0);
1746 break;
1747 default:
1748 ShouldNotReachHere();
1749 }
1750 #else
1751 // Checking
1752 #ifdef ASSERT
1753 { TosState tos_in = ilgl;
1754 TosState tos_out = ilgl;
1755 switch (bytecode()) {
1756 case Bytecodes::_i2l: // fall through
1757 case Bytecodes::_i2f: // fall through
1758 case Bytecodes::_i2d: // fall through
1759 case Bytecodes::_i2b: // fall through
1760 case Bytecodes::_i2c: // fall through
1761 case Bytecodes::_i2s: tos_in = itos; break;
1762 case Bytecodes::_l2i: // fall through
1763 case Bytecodes::_l2f: // fall through
1764 case Bytecodes::_l2d: tos_in = ltos; break;
1765 case Bytecodes::_f2i: // fall through
1766 case Bytecodes::_f2l: // fall through
1767 case Bytecodes::_f2d: tos_in = ftos; break;
1768 case Bytecodes::_d2i: // fall through
1769 case Bytecodes::_d2l: // fall through
1770 case Bytecodes::_d2f: tos_in = dtos; break;
1771 default : ShouldNotReachHere();
1772 }
1773 switch (bytecode()) {
1774 case Bytecodes::_l2i: // fall through
1775 case Bytecodes::_f2i: // fall through
1776 case Bytecodes::_d2i: // fall through
1777 case Bytecodes::_i2b: // fall through
1778 case Bytecodes::_i2c: // fall through
1779 case Bytecodes::_i2s: tos_out = itos; break;
1780 case Bytecodes::_i2l: // fall through
1781 case Bytecodes::_f2l: // fall through
1782 case Bytecodes::_d2l: tos_out = ltos; break;
1783 case Bytecodes::_i2f: // fall through
1784 case Bytecodes::_l2f: // fall through
1785 case Bytecodes::_d2f: tos_out = ftos; break;
1786 case Bytecodes::_i2d: // fall through
1787 case Bytecodes::_l2d: // fall through
1788 case Bytecodes::_f2d: tos_out = dtos; break;
1789 default : ShouldNotReachHere();
1790 }
1791 transition(tos_in, tos_out);
1792 }
1793 #endif // ASSERT
1794
1795 // Conversion
1796 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1797 switch (bytecode()) {
1798 case Bytecodes::_i2l:
1799 __ extend_sign(rdx, rax);
1800 break;
1801 case Bytecodes::_i2f:
1802 __ push(rax); // store int on tos
1803 __ fild_s(at_rsp()); // load int to ST0
1804 __ f2ieee(); // truncate to float size
1805 __ pop(rcx); // adjust rsp
1806 break;
1807 case Bytecodes::_i2d:
1808 __ push(rax); // add one slot for d2ieee()
1809 __ push(rax); // store int on tos
1810 __ fild_s(at_rsp()); // load int to ST0
1811 __ d2ieee(); // truncate to double size
1812 __ pop(rcx); // adjust rsp
1813 __ pop(rcx);
1814 break;
1815 case Bytecodes::_i2b:
1816 __ shll(rax, 24); // truncate upper 24 bits
1817 __ sarl(rax, 24); // and sign-extend byte
1818 LP64_ONLY(__ movsbl(rax, rax));
1819 break;
1820 case Bytecodes::_i2c:
1821 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1822 LP64_ONLY(__ movzwl(rax, rax));
1823 break;
1824 case Bytecodes::_i2s:
1825 __ shll(rax, 16); // truncate upper 16 bits
1826 __ sarl(rax, 16); // and sign-extend short
1827 LP64_ONLY(__ movswl(rax, rax));
1828 break;
1829 case Bytecodes::_l2i:
1830 /* nothing to do */
1831 break;
1832 case Bytecodes::_l2f:
1833 __ push(rdx); // store long on tos
1834 __ push(rax);
1835 __ fild_d(at_rsp()); // load long to ST0
1836 __ f2ieee(); // truncate to float size
1837 __ pop(rcx); // adjust rsp
1838 __ pop(rcx);
1839 break;
1840 case Bytecodes::_l2d:
1841 __ push(rdx); // store long on tos
1842 __ push(rax);
1843 __ fild_d(at_rsp()); // load long to ST0
1844 __ d2ieee(); // truncate to double size
1845 __ pop(rcx); // adjust rsp
1846 __ pop(rcx);
1847 break;
1848 case Bytecodes::_f2i:
1849 __ push(rcx); // reserve space for argument
1850 __ fstp_s(at_rsp()); // pass float argument on stack
1851 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1852 break;
1853 case Bytecodes::_f2l:
1854 __ push(rcx); // reserve space for argument
1855 __ fstp_s(at_rsp()); // pass float argument on stack
1856 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1857 break;
1858 case Bytecodes::_f2d:
1859 /* nothing to do */
1860 break;
1861 case Bytecodes::_d2i:
1862 __ push(rcx); // reserve space for argument
1863 __ push(rcx);
1864 __ fstp_d(at_rsp()); // pass double argument on stack
1865 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1866 break;
1867 case Bytecodes::_d2l:
1868 __ push(rcx); // reserve space for argument
1869 __ push(rcx);
1870 __ fstp_d(at_rsp()); // pass double argument on stack
1871 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1872 break;
1873 case Bytecodes::_d2f:
1874 __ push(rcx); // reserve space for f2ieee()
1875 __ f2ieee(); // truncate to float size
1876 __ pop(rcx); // adjust rsp
1877 break;
1878 default :
1879 ShouldNotReachHere();
1880 }
1881 #endif
1882 }
1883
1884 void TemplateTable::lcmp() {
1885 transition(ltos, itos);
1886 #ifdef _LP64
1887 Label done;
1888 __ pop_l(rdx);
1889 __ cmpq(rdx, rax);
1890 __ movl(rax, -1);
1891 __ jccb(Assembler::less, done);
1892 __ setb(Assembler::notEqual, rax);
1893 __ movzbl(rax, rax);
1894 __ bind(done);
1895 #else
1896
1897 // y = rdx:rax
1898 __ pop_l(rbx, rcx); // get x = rcx:rbx
1899 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1900 __ mov(rax, rcx);
1901 #endif
1902 }
1903
1904 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1905 #ifdef _LP64
1906 Label done;
1907 if (is_float) {
1908 // XXX get rid of pop here, use ... reg, mem32
1909 __ pop_f(xmm1);
1910 __ ucomiss(xmm1, xmm0);
1911 } else {
1912 // XXX get rid of pop here, use ... reg, mem64
1913 __ pop_d(xmm1);
1914 __ ucomisd(xmm1, xmm0);
1915 }
1916 if (unordered_result < 0) {
1917 __ movl(rax, -1);
1918 __ jccb(Assembler::parity, done);
1919 __ jccb(Assembler::below, done);
1920 __ setb(Assembler::notEqual, rdx);
1921 __ movzbl(rax, rdx);
1922 } else {
1923 __ movl(rax, 1);
1924 __ jccb(Assembler::parity, done);
1925 __ jccb(Assembler::above, done);
1926 __ movl(rax, 0);
1927 __ jccb(Assembler::equal, done);
1928 __ decrementl(rax);
1929 }
1930 __ bind(done);
1931 #else
1932 if (is_float) {
1933 __ fld_s(at_rsp());
1934 } else {
1935 __ fld_d(at_rsp());
1936 __ pop(rdx);
1937 }
1938 __ pop(rcx);
1939 __ fcmp2int(rax, unordered_result < 0);
1940 #endif
1941 }
1942
1943 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1944 __ get_method(rcx); // rcx holds method
1945 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
1946 // holds bumped taken count
1947
1948 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
1949 InvocationCounter::counter_offset();
1950 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
1951 InvocationCounter::counter_offset();
1952
1953 // Load up edx with the branch displacement
1954 if (is_wide) {
1955 __ movl(rdx, at_bcp(1));
1956 } else {
1957 __ load_signed_short(rdx, at_bcp(1));
1958 }
1959 __ bswapl(rdx);
1960
1961 if (!is_wide) {
1962 __ sarl(rdx, 16);
1963 }
1964 LP64_ONLY(__ movl2ptr(rdx, rdx));
1965
1966 // Handle all the JSR stuff here, then exit.
1967 // It's much shorter and cleaner than intermingling with the non-JSR
1968 // normal-branch stuff occurring below.
1969 if (is_jsr) {
1970 // Pre-load the next target bytecode into rbx
1971 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
1972
1973 // compute return address as bci in rax
1974 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
1975 in_bytes(ConstMethod::codes_offset())));
1976 __ subptr(rax, Address(rcx, Method::const_offset()));
1977 // Adjust the bcp in r13 by the displacement in rdx
1978 __ addptr(rbcp, rdx);
1979 // jsr returns atos that is not an oop
1980 __ push_i(rax);
1981 __ dispatch_only(vtos);
1982 return;
1983 }
1984
1985 // Normal (non-jsr) branch handling
1986
1987 // Adjust the bcp in r13 by the displacement in rdx
1988 __ addptr(rbcp, rdx);
1989
1990 assert(UseLoopCounter || !UseOnStackReplacement,
1991 "on-stack-replacement requires loop counters");
1992 Label backedge_counter_overflow;
1993 Label profile_method;
1994 Label dispatch;
1995 if (UseLoopCounter) {
1996 // increment backedge counter for backward branches
1997 // rax: MDO
1998 // rbx: MDO bumped taken-count
1999 // rcx: method
2000 // rdx: target offset
2001 // r13: target bcp
2002 // r14: locals pointer
2003 __ testl(rdx, rdx); // check if forward or backward branch
2004 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2005
2006 // check if MethodCounters exists
2007 Label has_counters;
2008 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2009 __ testptr(rax, rax);
2010 __ jcc(Assembler::notZero, has_counters);
2011 __ push(rdx);
2012 __ push(rcx);
2013 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2014 rcx);
2015 __ pop(rcx);
2016 __ pop(rdx);
2017 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2018 __ jcc(Assembler::zero, dispatch);
2019 __ bind(has_counters);
2020
2021 if (TieredCompilation) {
2022 Label no_mdo;
2023 int increment = InvocationCounter::count_increment;
2024 if (ProfileInterpreter) {
2025 // Are we profiling?
2026 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2027 __ testptr(rbx, rbx);
2028 __ jccb(Assembler::zero, no_mdo);
2029 // Increment the MDO backedge counter
2030 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2031 in_bytes(InvocationCounter::counter_offset()));
2032 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2033 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2034 rax, false, Assembler::zero, &backedge_counter_overflow);
2035 __ jmp(dispatch);
2036 }
2037 __ bind(no_mdo);
2038 // Increment backedge counter in MethodCounters*
2039 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2040 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2041 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2042 rax, false, Assembler::zero, &backedge_counter_overflow);
2043 } else { // not TieredCompilation
2044 // increment counter
2045 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2046 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2047 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2048 __ movl(Address(rcx, be_offset), rax); // store counter
2049
2050 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2051
2052 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2053 __ addl(rax, Address(rcx, be_offset)); // add both counters
2054
2055 if (ProfileInterpreter) {
2056 // Test to see if we should create a method data oop
2057 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2058 __ jcc(Assembler::less, dispatch);
2059
2060 // if no method data exists, go to profile method
2061 __ test_method_data_pointer(rax, profile_method);
2062
2063 if (UseOnStackReplacement) {
2064 // check for overflow against rbx which is the MDO taken count
2065 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2066 __ jcc(Assembler::below, dispatch);
2067
2068 // When ProfileInterpreter is on, the backedge_count comes
2069 // from the MethodData*, which value does not get reset on
2070 // the call to frequency_counter_overflow(). To avoid
2071 // excessive calls to the overflow routine while the method is
2072 // being compiled, add a second test to make sure the overflow
2073 // function is called only once every overflow_frequency.
2074 const int overflow_frequency = 1024;
2075 __ andl(rbx, overflow_frequency - 1);
2076 __ jcc(Assembler::zero, backedge_counter_overflow);
2077
2078 }
2079 } else {
2080 if (UseOnStackReplacement) {
2081 // check for overflow against rax, which is the sum of the
2082 // counters
2083 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2084 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2085
2086 }
2087 }
2088 }
2089 __ bind(dispatch);
2090 }
2091
2092 // Pre-load the next target bytecode into rbx
2093 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2094
2095 // continue with the bytecode @ target
2096 // rax: return bci for jsr's, unused otherwise
2097 // rbx: target bytecode
2098 // r13: target bcp
2099 __ dispatch_only(vtos);
2100
2101 if (UseLoopCounter) {
2102 if (ProfileInterpreter) {
2103 // Out-of-line code to allocate method data oop.
2104 __ bind(profile_method);
2105 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2106 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2107 __ set_method_data_pointer_for_bcp();
2108 __ jmp(dispatch);
2109 }
2110
2111 if (UseOnStackReplacement) {
2112 // invocation counter overflow
2113 __ bind(backedge_counter_overflow);
2114 __ negptr(rdx);
2115 __ addptr(rdx, rbcp); // branch bcp
2116 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2117 __ call_VM(noreg,
2118 CAST_FROM_FN_PTR(address,
2119 InterpreterRuntime::frequency_counter_overflow),
2120 rdx);
2121 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2122
2123 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2124 // rbx: target bytecode
2125 // rdx: scratch
2126 // r14: locals pointer
2127 // r13: bcp
2128 __ testptr(rax, rax); // test result
2129 __ jcc(Assembler::zero, dispatch); // no osr if null
2130 // nmethod may have been invalidated (VM may block upon call_VM return)
2131 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2132 __ jcc(Assembler::notEqual, dispatch);
2133
2134 // We have the address of an on stack replacement routine in rax
2135 // We need to prepare to execute the OSR method. First we must
2136 // migrate the locals and monitors off of the stack.
2137
2138 LP64_ONLY(__ mov(r13, rax)); // save the nmethod
2139 NOT_LP64(__ mov(rbx, rax)); // save the nmethod
2140 NOT_LP64(__ get_thread(rcx));
2141
2142 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2143
2144 // rax is OSR buffer, move it to expected parameter location
2145 LP64_ONLY(__ mov(j_rarg0, rax));
2146 NOT_LP64(__ mov(rcx, rax));
2147 // We use j_rarg definitions here so that registers don't conflict as parameter
2148 // registers change across platforms as we are in the midst of a calling
2149 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2150
2151 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2152 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2153
2154
2155 // pop the interpreter frame
2156 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2157 __ leave(); // remove frame anchor
2158 __ pop(retaddr); // get return address
2159 __ mov(rsp, sender_sp); // set sp to sender sp
2160 // Ensure compiled code always sees stack at proper alignment
2161 __ andptr(rsp, -(StackAlignmentInBytes));
2162
2163 // unlike x86 we need no specialized return from compiled code
2164 // to the interpreter or the call stub.
2165
2166 // push the return address
2167 __ push(retaddr);
2168
2169 // and begin the OSR nmethod
2170 LP64_ONLY(__ jmp(Address(r13, nmethod::osr_entry_point_offset())));
2171 NOT_LP64(__ jmp(Address(rbx, nmethod::osr_entry_point_offset())));
2172 }
2173 }
2174 }
2175
2176 void TemplateTable::if_0cmp(Condition cc) {
2177 transition(itos, vtos);
2178 // assume branch is more often taken than not (loops use backward branches)
2179 Label not_taken;
2180 __ testl(rax, rax);
2181 __ jcc(j_not(cc), not_taken);
2182 branch(false, false);
2183 __ bind(not_taken);
2184 __ profile_not_taken_branch(rax);
2185 }
2186
2187 void TemplateTable::if_icmp(Condition cc) {
2188 transition(itos, vtos);
2189 // assume branch is more often taken than not (loops use backward branches)
2190 Label not_taken;
2191 __ pop_i(rdx);
2192 __ cmpl(rdx, rax);
2193 __ jcc(j_not(cc), not_taken);
2194 branch(false, false);
2195 __ bind(not_taken);
2196 __ profile_not_taken_branch(rax);
2197 }
2198
2199 void TemplateTable::if_nullcmp(Condition cc) {
2200 transition(atos, vtos);
2201 // assume branch is more often taken than not (loops use backward branches)
2202 Label not_taken;
2203 __ testptr(rax, rax);
2204 __ jcc(j_not(cc), not_taken);
2205 branch(false, false);
2206 __ bind(not_taken);
2207 __ profile_not_taken_branch(rax);
2208 }
2209
2210 void TemplateTable::if_acmp(Condition cc) {
2211 transition(atos, vtos);
2212 // assume branch is more often taken than not (loops use backward branches)
2213 Label not_taken;
2214 __ pop_ptr(rdx);
2215 __ cmpptr(rdx, rax);
2216 __ jcc(j_not(cc), not_taken);
2217 branch(false, false);
2218 __ bind(not_taken);
2219 __ profile_not_taken_branch(rax);
2220 }
2221
2222 void TemplateTable::ret() {
2223 transition(vtos, vtos);
2224 locals_index(rbx);
2225 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2226 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2227 __ profile_ret(rbx, rcx);
2228 __ get_method(rax);
2229 __ movptr(rbcp, Address(rax, Method::const_offset()));
2230 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2231 ConstMethod::codes_offset()));
2232 __ dispatch_next(vtos);
2233 }
2234
2235 void TemplateTable::wide_ret() {
2236 transition(vtos, vtos);
2237 locals_index_wide(rbx);
2238 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2239 __ profile_ret(rbx, rcx);
2240 __ get_method(rax);
2241 __ movptr(rbcp, Address(rax, Method::const_offset()));
2242 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2243 __ dispatch_next(vtos);
2244 }
2245
2246 void TemplateTable::tableswitch() {
2247 Label default_case, continue_execution;
2248 transition(itos, vtos);
2249
2250 // align r13/rsi
2251 __ lea(rbx, at_bcp(BytesPerInt));
2252 __ andptr(rbx, -BytesPerInt);
2253 // load lo & hi
2254 __ movl(rcx, Address(rbx, BytesPerInt));
2255 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2256 __ bswapl(rcx);
2257 __ bswapl(rdx);
2258 // check against lo & hi
2259 __ cmpl(rax, rcx);
2260 __ jcc(Assembler::less, default_case);
2261 __ cmpl(rax, rdx);
2262 __ jcc(Assembler::greater, default_case);
2263 // lookup dispatch offset
2264 __ subl(rax, rcx);
2265 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2266 __ profile_switch_case(rax, rbx, rcx);
2267 // continue execution
2268 __ bind(continue_execution);
2269 __ bswapl(rdx);
2270 LP64_ONLY(__ movl2ptr(rdx, rdx));
2271 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2272 __ addptr(rbcp, rdx);
2273 __ dispatch_only(vtos);
2274 // handle default
2275 __ bind(default_case);
2276 __ profile_switch_default(rax);
2277 __ movl(rdx, Address(rbx, 0));
2278 __ jmp(continue_execution);
2279 }
2280
2281 void TemplateTable::lookupswitch() {
2282 transition(itos, itos);
2283 __ stop("lookupswitch bytecode should have been rewritten");
2284 }
2285
2286 void TemplateTable::fast_linearswitch() {
2287 transition(itos, vtos);
2288 Label loop_entry, loop, found, continue_execution;
2289 // bswap rax so we can avoid bswapping the table entries
2290 __ bswapl(rax);
2291 // align r13
2292 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2293 // this instruction (change offsets
2294 // below)
2295 __ andptr(rbx, -BytesPerInt);
2296 // set counter
2297 __ movl(rcx, Address(rbx, BytesPerInt));
2298 __ bswapl(rcx);
2299 __ jmpb(loop_entry);
2300 // table search
2301 __ bind(loop);
2302 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2303 __ jcc(Assembler::equal, found);
2304 __ bind(loop_entry);
2305 __ decrementl(rcx);
2306 __ jcc(Assembler::greaterEqual, loop);
2307 // default case
2308 __ profile_switch_default(rax);
2309 __ movl(rdx, Address(rbx, 0));
2310 __ jmp(continue_execution);
2311 // entry found -> get offset
2312 __ bind(found);
2313 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2314 __ profile_switch_case(rcx, rax, rbx);
2315 // continue execution
2316 __ bind(continue_execution);
2317 __ bswapl(rdx);
2318 __ movl2ptr(rdx, rdx);
2319 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2320 __ addptr(rbcp, rdx);
2321 __ dispatch_only(vtos);
2322 }
2323
2324 void TemplateTable::fast_binaryswitch() {
2325 transition(itos, vtos);
2326 // Implementation using the following core algorithm:
2327 //
2328 // int binary_search(int key, LookupswitchPair* array, int n) {
2329 // // Binary search according to "Methodik des Programmierens" by
2330 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2331 // int i = 0;
2332 // int j = n;
2333 // while (i+1 < j) {
2334 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2335 // // with Q: for all i: 0 <= i < n: key < a[i]
2336 // // where a stands for the array and assuming that the (inexisting)
2337 // // element a[n] is infinitely big.
2338 // int h = (i + j) >> 1;
2339 // // i < h < j
2340 // if (key < array[h].fast_match()) {
2341 // j = h;
2342 // } else {
2343 // i = h;
2344 // }
2345 // }
2346 // // R: a[i] <= key < a[i+1] or Q
2347 // // (i.e., if key is within array, i is the correct index)
2348 // return i;
2349 // }
2350
2351 // Register allocation
2352 const Register key = rax; // already set (tosca)
2353 const Register array = rbx;
2354 const Register i = rcx;
2355 const Register j = rdx;
2356 const Register h = rdi;
2357 const Register temp = rsi;
2358
2359 // Find array start
2360 NOT_LP64(__ save_bcp());
2361
2362 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2363 // get rid of this
2364 // instruction (change
2365 // offsets below)
2366 __ andptr(array, -BytesPerInt);
2367
2368 // Initialize i & j
2369 __ xorl(i, i); // i = 0;
2370 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2371
2372 // Convert j into native byteordering
2373 __ bswapl(j);
2374
2375 // And start
2376 Label entry;
2377 __ jmp(entry);
2378
2379 // binary search loop
2380 {
2381 Label loop;
2382 __ bind(loop);
2383 // int h = (i + j) >> 1;
2384 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2385 __ sarl(h, 1); // h = (i + j) >> 1;
2386 // if (key < array[h].fast_match()) {
2387 // j = h;
2388 // } else {
2389 // i = h;
2390 // }
2391 // Convert array[h].match to native byte-ordering before compare
2392 __ movl(temp, Address(array, h, Address::times_8));
2393 __ bswapl(temp);
2394 __ cmpl(key, temp);
2395 // j = h if (key < array[h].fast_match())
2396 __ cmov32(Assembler::less, j, h);
2397 // i = h if (key >= array[h].fast_match())
2398 __ cmov32(Assembler::greaterEqual, i, h);
2399 // while (i+1 < j)
2400 __ bind(entry);
2401 __ leal(h, Address(i, 1)); // i+1
2402 __ cmpl(h, j); // i+1 < j
2403 __ jcc(Assembler::less, loop);
2404 }
2405
2406 // end of binary search, result index is i (must check again!)
2407 Label default_case;
2408 // Convert array[i].match to native byte-ordering before compare
2409 __ movl(temp, Address(array, i, Address::times_8));
2410 __ bswapl(temp);
2411 __ cmpl(key, temp);
2412 __ jcc(Assembler::notEqual, default_case);
2413
2414 // entry found -> j = offset
2415 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2416 __ profile_switch_case(i, key, array);
2417 __ bswapl(j);
2418 LP64_ONLY(__ movslq(j, j));
2419
2420 NOT_LP64(__ restore_bcp());
2421 NOT_LP64(__ restore_locals()); // restore rdi
2422
2423 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2424 __ addptr(rbcp, j);
2425 __ dispatch_only(vtos);
2426
2427 // default case -> j = default offset
2428 __ bind(default_case);
2429 __ profile_switch_default(i);
2430 __ movl(j, Address(array, -2 * BytesPerInt));
2431 __ bswapl(j);
2432 LP64_ONLY(__ movslq(j, j));
2433
2434 NOT_LP64(__ restore_bcp());
2435 NOT_LP64(__ restore_locals());
2436
2437 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2438 __ addptr(rbcp, j);
2439 __ dispatch_only(vtos);
2440 }
2441
2442 void TemplateTable::_return(TosState state) {
2443 transition(state, state);
2444
2445 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2446
2447 assert(_desc->calls_vm(),
2448 "inconsistent calls_vm information"); // call in remove_activation
2449
2450 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2451 assert(state == vtos, "only valid state");
2452 __ movptr(robj, aaddress(0));
2453 __ load_klass(rdi, robj);
2454 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2455 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2456 Label skip_register_finalizer;
2457 __ jcc(Assembler::zero, skip_register_finalizer);
2458
2459 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2460
2461 __ bind(skip_register_finalizer);
2462 }
2463
2464 __ remove_activation(state, rbcp);
2465 __ jmp(rbcp);
2466 }
2467
2468 // ----------------------------------------------------------------------------
2469 // Volatile variables demand their effects be made known to all CPU's
2470 // in order. Store buffers on most chips allow reads & writes to
2471 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2472 // without some kind of memory barrier (i.e., it's not sufficient that
2473 // the interpreter does not reorder volatile references, the hardware
2474 // also must not reorder them).
2475 //
2476 // According to the new Java Memory Model (JMM):
2477 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2478 // writes act as aquire & release, so:
2479 // (2) A read cannot let unrelated NON-volatile memory refs that
2480 // happen after the read float up to before the read. It's OK for
2481 // non-volatile memory refs that happen before the volatile read to
2482 // float down below it.
2483 // (3) Similar a volatile write cannot let unrelated NON-volatile
2484 // memory refs that happen BEFORE the write float down to after the
2485 // write. It's OK for non-volatile memory refs that happen after the
2486 // volatile write to float up before it.
2487 //
2488 // We only put in barriers around volatile refs (they are expensive),
2489 // not _between_ memory refs (that would require us to track the
2490 // flavor of the previous memory refs). Requirements (2) and (3)
2491 // require some barriers before volatile stores and after volatile
2492 // loads. These nearly cover requirement (1) but miss the
2493 // volatile-store-volatile-load case. This final case is placed after
2494 // volatile-stores although it could just as well go before
2495 // volatile-loads.
2496
2497 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2498 // Helper function to insert a is-volatile test and memory barrier
2499 if(!os::is_MP()) return; // Not needed on single CPU
2500 __ membar(order_constraint);
2501 }
2502
2503 void TemplateTable::resolve_cache_and_index(int byte_no,
2504 Register Rcache,
2505 Register index,
2506 size_t index_size) {
2507 const Register temp = rbx;
2508 assert_different_registers(Rcache, index, temp);
2509
2510 Label resolved;
2511
2512 Bytecodes::Code code = bytecode();
2513 switch (code) {
2514 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2515 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2516 }
2517
2518 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2519 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2520 __ cmpl(temp, code); // have we resolved this bytecode?
2521 __ jcc(Assembler::equal, resolved);
2522
2523 // resolve first time through
2524 address entry;
2525 switch (code) {
2526 case Bytecodes::_getstatic : // fall through
2527 case Bytecodes::_putstatic : // fall through
2528 case Bytecodes::_getfield : // fall through
2529 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2530 case Bytecodes::_invokevirtual : // fall through
2531 case Bytecodes::_invokespecial : // fall through
2532 case Bytecodes::_invokestatic : // fall through
2533 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2534 case Bytecodes::_invokehandle : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle); break;
2535 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2536 default:
2537 fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(code)));
2538 break;
2539 }
2540 __ movl(temp, code);
2541 __ call_VM(noreg, entry, temp);
2542 // Update registers with resolved info
2543 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2544 __ bind(resolved);
2545 }
2546
2547 // The cache and index registers must be set before call
2548 void TemplateTable::load_field_cp_cache_entry(Register obj,
2549 Register cache,
2550 Register index,
2551 Register off,
2552 Register flags,
2553 bool is_static = false) {
2554 assert_different_registers(cache, index, flags, off);
2555
2556 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2557 // Field offset
2558 __ movptr(off, Address(cache, index, Address::times_ptr,
2559 in_bytes(cp_base_offset +
2560 ConstantPoolCacheEntry::f2_offset())));
2561 // Flags
2562 __ movl(flags, Address(cache, index, Address::times_ptr,
2563 in_bytes(cp_base_offset +
2564 ConstantPoolCacheEntry::flags_offset())));
2565
2566 // klass overwrite register
2567 if (is_static) {
2568 __ movptr(obj, Address(cache, index, Address::times_ptr,
2569 in_bytes(cp_base_offset +
2570 ConstantPoolCacheEntry::f1_offset())));
2571 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2572 __ movptr(obj, Address(obj, mirror_offset));
2573 }
2574 }
2575
2576 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2577 Register method,
2578 Register itable_index,
2579 Register flags,
2580 bool is_invokevirtual,
2581 bool is_invokevfinal, /*unused*/
2582 bool is_invokedynamic) {
2583 // setup registers
2584 const Register cache = rcx;
2585 const Register index = rdx;
2586 assert_different_registers(method, flags);
2587 assert_different_registers(method, cache, index);
2588 assert_different_registers(itable_index, flags);
2589 assert_different_registers(itable_index, cache, index);
2590 // determine constant pool cache field offsets
2591 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2592 const int method_offset = in_bytes(
2593 ConstantPoolCache::base_offset() +
2594 ((byte_no == f2_byte)
2595 ? ConstantPoolCacheEntry::f2_offset()
2596 : ConstantPoolCacheEntry::f1_offset()));
2597 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2598 ConstantPoolCacheEntry::flags_offset());
2599 // access constant pool cache fields
2600 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2601 ConstantPoolCacheEntry::f2_offset());
2602
2603 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2604 resolve_cache_and_index(byte_no, cache, index, index_size);
2605 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2606
2607 if (itable_index != noreg) {
2608 // pick up itable or appendix index from f2 also:
2609 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2610 }
2611 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2612 }
2613
2614 // The registers cache and index expected to be set before call.
2615 // Correct values of the cache and index registers are preserved.
2616 void TemplateTable::jvmti_post_field_access(Register cache,
2617 Register index,
2618 bool is_static,
2619 bool has_tos) {
2620 if (JvmtiExport::can_post_field_access()) {
2621 // Check to see if a field access watch has been set before we take
2622 // the time to call into the VM.
2623 Label L1;
2624 assert_different_registers(cache, index, rax);
2625 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2626 __ testl(rax,rax);
2627 __ jcc(Assembler::zero, L1);
2628
2629 // cache entry pointer
2630 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2631 __ shll(index, LogBytesPerWord);
2632 __ addptr(cache, index);
2633 if (is_static) {
2634 __ xorptr(rax, rax); // NULL object reference
2635 } else {
2636 __ pop(atos); // Get the object
2637 __ verify_oop(rax);
2638 __ push(atos); // Restore stack state
2639 }
2640 // rax,: object pointer or NULL
2641 // cache: cache entry pointer
2642 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2643 rax, cache);
2644 __ get_cache_and_index_at_bcp(cache, index, 1);
2645 __ bind(L1);
2646 }
2647 }
2648
2649 void TemplateTable::pop_and_check_object(Register r) {
2650 __ pop_ptr(r);
2651 __ null_check(r); // for field access must check obj.
2652 __ verify_oop(r);
2653 }
2654
2655 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2656 transition(vtos, vtos);
2657
2658 const Register cache = rcx;
2659 const Register index = rdx;
2660 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2661 const Register off = rbx;
2662 const Register flags = rax;
2663 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2664
2665 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2666 jvmti_post_field_access(cache, index, is_static, false);
2667 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2668
2669 if (!is_static) pop_and_check_object(obj);
2670
2671 const Address field(obj, off, Address::times_1, 0*wordSize);
2672 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2673
2674 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2675
2676 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2677 // Make sure we don't need to mask edx after the above shift
2678 assert(btos == 0, "change code, btos != 0");
2679
2680 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2681
2682 __ jcc(Assembler::notZero, notByte);
2683 // btos
2684 __ load_signed_byte(rax, field);
2685 __ push(btos);
2686 // Rewrite bytecode to be faster
2687 if (!is_static && rc == may_rewrite) {
2688 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2689 }
2690 __ jmp(Done);
2691
2692 __ bind(notByte);
2693 __ cmpl(flags, atos);
2694 __ jcc(Assembler::notEqual, notObj);
2695 // atos
2696 __ load_heap_oop(rax, field);
2697 __ push(atos);
2698 if (!is_static && rc == may_rewrite) {
2699 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2700 }
2701 __ jmp(Done);
2702
2703 __ bind(notObj);
2704 __ cmpl(flags, itos);
2705 __ jcc(Assembler::notEqual, notInt);
2706 // itos
2707 __ movl(rax, field);
2708 __ push(itos);
2709 // Rewrite bytecode to be faster
2710 if (!is_static && rc == may_rewrite) {
2711 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2712 }
2713 __ jmp(Done);
2714
2715 __ bind(notInt);
2716 __ cmpl(flags, ctos);
2717 __ jcc(Assembler::notEqual, notChar);
2718 // ctos
2719 __ load_unsigned_short(rax, field);
2720 __ push(ctos);
2721 // Rewrite bytecode to be faster
2722 if (!is_static && rc == may_rewrite) {
2723 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2724 }
2725 __ jmp(Done);
2726
2727 __ bind(notChar);
2728 __ cmpl(flags, stos);
2729 __ jcc(Assembler::notEqual, notShort);
2730 // stos
2731 __ load_signed_short(rax, field);
2732 __ push(stos);
2733 // Rewrite bytecode to be faster
2734 if (!is_static && rc == may_rewrite) {
2735 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2736 }
2737 __ jmp(Done);
2738
2739 __ bind(notShort);
2740 __ cmpl(flags, ltos);
2741 __ jcc(Assembler::notEqual, notLong);
2742 // ltos
2743
2744 #ifndef _LP64
2745 // Generate code as if volatile. There just aren't enough registers to
2746 // save that information and this code is faster than the test.
2747 __ fild_d(field); // Must load atomically
2748 __ subptr(rsp,2*wordSize); // Make space for store
2749 __ fistp_d(Address(rsp,0));
2750 __ pop(rax);
2751 __ pop(rdx);
2752 #else
2753 __ movq(rax, field);
2754 #endif
2755
2756 __ push(ltos);
2757 // Rewrite bytecode to be faster
2758 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2759 __ jmp(Done);
2760
2761 __ bind(notLong);
2762 __ cmpl(flags, ftos);
2763 __ jcc(Assembler::notEqual, notFloat);
2764 // ftos
2765
2766 LP64_ONLY(__ movflt(xmm0, field));
2767 NOT_LP64(__ fld_s(field));
2768 __ push(ftos);
2769 // Rewrite bytecode to be faster
2770 if (!is_static && rc == may_rewrite) {
2771 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2772 }
2773 __ jmp(Done);
2774
2775 __ bind(notFloat);
2776 #ifdef ASSERT
2777 __ cmpl(flags, dtos);
2778 __ jcc(Assembler::notEqual, notDouble);
2779 #endif
2780 // dtos
2781 LP64_ONLY(__ movdbl(xmm0, field));
2782 NOT_LP64(__ fld_d(field));
2783 __ push(dtos);
2784 // Rewrite bytecode to be faster
2785 if (!is_static && rc == may_rewrite) {
2786 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2787 }
2788 #ifdef ASSERT
2789 __ jmp(Done);
2790
2791
2792 __ bind(notDouble);
2793 __ stop("Bad state");
2794 #endif
2795
2796 __ bind(Done);
2797 // [jk] not needed currently
2798 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2799 // Assembler::LoadStore));
2800 }
2801
2802 void TemplateTable::getfield(int byte_no) {
2803 getfield_or_static(byte_no, false);
2804 }
2805
2806 void TemplateTable::nofast_getfield(int byte_no) {
2807 getfield_or_static(byte_no, false, may_not_rewrite);
2808 }
2809
2810 void TemplateTable::getstatic(int byte_no) {
2811 getfield_or_static(byte_no, true);
2812 }
2813
2814
2815 // The registers cache and index expected to be set before call.
2816 // The function may destroy various registers, just not the cache and index registers.
2817 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2818
2819 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
2820 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2821 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2822 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
2823
2824 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2825
2826 if (JvmtiExport::can_post_field_modification()) {
2827 // Check to see if a field modification watch has been set before
2828 // we take the time to call into the VM.
2829 Label L1;
2830 assert_different_registers(cache, index, rax);
2831 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2832 __ testl(rax, rax);
2833 __ jcc(Assembler::zero, L1);
2834
2835 __ get_cache_and_index_at_bcp(robj, RDX, 1);
2836
2837
2838 if (is_static) {
2839 // Life is simple. Null out the object pointer.
2840 __ xorl(RBX, RBX);
2841
2842 } else {
2843 // Life is harder. The stack holds the value on top, followed by
2844 // the object. We don't know the size of the value, though; it
2845 // could be one or two words depending on its type. As a result,
2846 // we must find the type to determine where the object is.
2847 #ifndef _LP64
2848 Label two_word, valsize_known;
2849 #endif
2850 __ movl(RCX, Address(robj, RDX,
2851 Address::times_ptr,
2852 in_bytes(cp_base_offset +
2853 ConstantPoolCacheEntry::flags_offset())));
2854 NOT_LP64(__ mov(rbx, rsp));
2855 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
2856
2857 // Make sure we don't need to mask rcx after the above shift
2858 ConstantPoolCacheEntry::verify_tos_state_shift();
2859 #ifdef _LP64
2860 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2861 __ cmpl(c_rarg3, ltos);
2862 __ cmovptr(Assembler::equal,
2863 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2864 __ cmpl(c_rarg3, dtos);
2865 __ cmovptr(Assembler::equal,
2866 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2867 #else
2868 __ cmpl(rcx, ltos);
2869 __ jccb(Assembler::equal, two_word);
2870 __ cmpl(rcx, dtos);
2871 __ jccb(Assembler::equal, two_word);
2872 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2873 __ jmpb(valsize_known);
2874
2875 __ bind(two_word);
2876 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2877
2878 __ bind(valsize_known);
2879 // setup object pointer
2880 __ movptr(rbx, Address(rbx, 0));
2881 #endif
2882 }
2883 // cache entry pointer
2884 __ addptr(robj, in_bytes(cp_base_offset));
2885 __ shll(RDX, LogBytesPerWord);
2886 __ addptr(robj, RDX);
2887 // object (tos)
2888 __ mov(RCX, rsp);
2889 // c_rarg1: object pointer set up above (NULL if static)
2890 // c_rarg2: cache entry pointer
2891 // c_rarg3: jvalue object on the stack
2892 __ call_VM(noreg,
2893 CAST_FROM_FN_PTR(address,
2894 InterpreterRuntime::post_field_modification),
2895 RBX, robj, RCX);
2896 __ get_cache_and_index_at_bcp(cache, index, 1);
2897 __ bind(L1);
2898 }
2899 }
2900
2901 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2902 transition(vtos, vtos);
2903
2904 const Register cache = rcx;
2905 const Register index = rdx;
2906 const Register obj = rcx;
2907 const Register off = rbx;
2908 const Register flags = rax;
2909 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2910
2911 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2912 jvmti_post_field_mod(cache, index, is_static);
2913 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2914
2915 // [jk] not needed currently
2916 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2917 // Assembler::StoreStore));
2918
2919 Label notVolatile, Done;
2920 __ movl(rdx, flags);
2921 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2922 __ andl(rdx, 0x1);
2923
2924 // field addresses
2925 const Address field(obj, off, Address::times_1, 0*wordSize);
2926 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
2927
2928 Label notByte, notInt, notShort, notChar,
2929 notLong, notFloat, notObj, notDouble;
2930
2931 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2932
2933 assert(btos == 0, "change code, btos != 0");
2934 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2935 __ jcc(Assembler::notZero, notByte);
2936
2937 // btos
2938 {
2939 __ pop(btos);
2940 if (!is_static) pop_and_check_object(obj);
2941 __ movb(field, rax);
2942 if (!is_static && rc == may_rewrite) {
2943 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
2944 }
2945 __ jmp(Done);
2946 }
2947
2948 __ bind(notByte);
2949 __ cmpl(flags, atos);
2950 __ jcc(Assembler::notEqual, notObj);
2951
2952 // atos
2953 {
2954 __ pop(atos);
2955 if (!is_static) pop_and_check_object(obj);
2956 // Store into the field
2957 do_oop_store(_masm, field, rax, _bs->kind(), false);
2958 if (!is_static && rc == may_rewrite) {
2959 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
2960 }
2961 __ jmp(Done);
2962 }
2963
2964 __ bind(notObj);
2965 __ cmpl(flags, itos);
2966 __ jcc(Assembler::notEqual, notInt);
2967
2968 // itos
2969 {
2970 __ pop(itos);
2971 if (!is_static) pop_and_check_object(obj);
2972 __ movl(field, rax);
2973 if (!is_static && rc == may_rewrite) {
2974 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
2975 }
2976 __ jmp(Done);
2977 }
2978
2979 __ bind(notInt);
2980 __ cmpl(flags, ctos);
2981 __ jcc(Assembler::notEqual, notChar);
2982
2983 // ctos
2984 {
2985 __ pop(ctos);
2986 if (!is_static) pop_and_check_object(obj);
2987 __ movw(field, rax);
2988 if (!is_static && rc == may_rewrite) {
2989 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
2990 }
2991 __ jmp(Done);
2992 }
2993
2994 __ bind(notChar);
2995 __ cmpl(flags, stos);
2996 __ jcc(Assembler::notEqual, notShort);
2997
2998 // stos
2999 {
3000 __ pop(stos);
3001 if (!is_static) pop_and_check_object(obj);
3002 __ movw(field, rax);
3003 if (!is_static && rc == may_rewrite) {
3004 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3005 }
3006 __ jmp(Done);
3007 }
3008
3009 __ bind(notShort);
3010 __ cmpl(flags, ltos);
3011 __ jcc(Assembler::notEqual, notLong);
3012
3013 // ltos
3014 #ifdef _LP64
3015 {
3016 __ pop(ltos);
3017 if (!is_static) pop_and_check_object(obj);
3018 __ movq(field, rax);
3019 if (!is_static && rc == may_rewrite) {
3020 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3021 }
3022 __ jmp(Done);
3023 }
3024 #else
3025 {
3026 Label notVolatileLong;
3027 __ testl(rdx, rdx);
3028 __ jcc(Assembler::zero, notVolatileLong);
3029
3030 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3031 if (!is_static) pop_and_check_object(obj);
3032
3033 // Replace with real volatile test
3034 __ push(rdx);
3035 __ push(rax); // Must update atomically with FIST
3036 __ fild_d(Address(rsp,0)); // So load into FPU register
3037 __ fistp_d(field); // and put into memory atomically
3038 __ addptr(rsp, 2*wordSize);
3039 // volatile_barrier();
3040 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3041 Assembler::StoreStore));
3042 // Don't rewrite volatile version
3043 __ jmp(notVolatile);
3044
3045 __ bind(notVolatileLong);
3046
3047 __ pop(ltos); // overwrites rdx
3048 if (!is_static) pop_and_check_object(obj);
3049 __ movptr(hi, rdx);
3050 __ movptr(field, rax);
3051 // Don't rewrite to _fast_lputfield for potential volatile case.
3052 __ jmp(notVolatile);
3053 }
3054 #endif // _LP64
3055
3056 __ bind(notLong);
3057 __ cmpl(flags, ftos);
3058 __ jcc(Assembler::notEqual, notFloat);
3059
3060 // ftos
3061 {
3062 __ pop(ftos);
3063 if (!is_static) pop_and_check_object(obj);
3064 NOT_LP64( __ fstp_s(field);)
3065 LP64_ONLY( __ movflt(field, xmm0);)
3066 if (!is_static && rc == may_rewrite) {
3067 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3068 }
3069 __ jmp(Done);
3070 }
3071
3072 __ bind(notFloat);
3073 #ifdef ASSERT
3074 __ cmpl(flags, dtos);
3075 __ jcc(Assembler::notEqual, notDouble);
3076 #endif
3077
3078 // dtos
3079 {
3080 __ pop(dtos);
3081 if (!is_static) pop_and_check_object(obj);
3082 NOT_LP64( __ fstp_d(field);)
3083 LP64_ONLY( __ movdbl(field, xmm0);)
3084 if (!is_static && rc == may_rewrite) {
3085 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3086 }
3087 }
3088
3089 #ifdef ASSERT
3090 __ jmp(Done);
3091
3092 __ bind(notDouble);
3093 __ stop("Bad state");
3094 #endif
3095
3096 __ bind(Done);
3097
3098 // Check for volatile store
3099 __ testl(rdx, rdx);
3100 __ jcc(Assembler::zero, notVolatile);
3101 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3102 Assembler::StoreStore));
3103 __ bind(notVolatile);
3104 }
3105
3106 void TemplateTable::putfield(int byte_no) {
3107 putfield_or_static(byte_no, false);
3108 }
3109
3110 void TemplateTable::nofast_putfield(int byte_no) {
3111 putfield_or_static(byte_no, false, may_not_rewrite);
3112 }
3113
3114 void TemplateTable::putstatic(int byte_no) {
3115 putfield_or_static(byte_no, true);
3116 }
3117
3118 void TemplateTable::jvmti_post_fast_field_mod() {
3119
3120 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3121
3122 if (JvmtiExport::can_post_field_modification()) {
3123 // Check to see if a field modification watch has been set before
3124 // we take the time to call into the VM.
3125 Label L2;
3126 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3127 __ testl(scratch, scratch);
3128 __ jcc(Assembler::zero, L2);
3129 __ pop_ptr(rbx); // copy the object pointer from tos
3130 __ verify_oop(rbx);
3131 __ push_ptr(rbx); // put the object pointer back on tos
3132 // Save tos values before call_VM() clobbers them. Since we have
3133 // to do it for every data type, we use the saved values as the
3134 // jvalue object.
3135 switch (bytecode()) { // load values into the jvalue object
3136 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3137 case Bytecodes::_fast_bputfield: // fall through
3138 case Bytecodes::_fast_sputfield: // fall through
3139 case Bytecodes::_fast_cputfield: // fall through
3140 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3141 case Bytecodes::_fast_dputfield: __ push_d(); break;
3142 case Bytecodes::_fast_fputfield: __ push_f(); break;
3143 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3144
3145 default:
3146 ShouldNotReachHere();
3147 }
3148 __ mov(scratch, rsp); // points to jvalue on the stack
3149 // access constant pool cache entry
3150 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3151 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3152 __ verify_oop(rbx);
3153 // rbx: object pointer copied above
3154 // c_rarg2: cache entry pointer
3155 // c_rarg3: jvalue object on the stack
3156 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3157 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3158
3159 switch (bytecode()) { // restore tos values
3160 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3161 case Bytecodes::_fast_bputfield: // fall through
3162 case Bytecodes::_fast_sputfield: // fall through
3163 case Bytecodes::_fast_cputfield: // fall through
3164 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3165 case Bytecodes::_fast_dputfield: __ pop_d(); break;
3166 case Bytecodes::_fast_fputfield: __ pop_f(); break;
3167 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3168 }
3169 __ bind(L2);
3170 }
3171 }
3172
3173 void TemplateTable::fast_storefield(TosState state) {
3174 transition(state, vtos);
3175
3176 ByteSize base = ConstantPoolCache::base_offset();
3177
3178 jvmti_post_fast_field_mod();
3179
3180 // access constant pool cache
3181 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3182
3183 // test for volatile with rdx but rdx is tos register for lputfield.
3184 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3185 in_bytes(base +
3186 ConstantPoolCacheEntry::flags_offset())));
3187
3188 // replace index with field offset from cache entry
3189 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3190 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3191
3192 // [jk] not needed currently
3193 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3194 // Assembler::StoreStore));
3195
3196 Label notVolatile;
3197 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3198 __ andl(rdx, 0x1);
3199
3200 // Get object from stack
3201 pop_and_check_object(rcx);
3202
3203 // field address
3204 const Address field(rcx, rbx, Address::times_1);
3205
3206 // access field
3207 switch (bytecode()) {
3208 case Bytecodes::_fast_aputfield:
3209 do_oop_store(_masm, field, rax, _bs->kind(), false);
3210 break;
3211 case Bytecodes::_fast_lputfield:
3212 #ifdef _LP64
3213 __ movq(field, rax);
3214 #else
3215 __ stop("should not be rewritten");
3216 #endif
3217 break;
3218 case Bytecodes::_fast_iputfield:
3219 __ movl(field, rax);
3220 break;
3221 case Bytecodes::_fast_bputfield:
3222 __ movb(field, rax);
3223 break;
3224 case Bytecodes::_fast_sputfield:
3225 // fall through
3226 case Bytecodes::_fast_cputfield:
3227 __ movw(field, rax);
3228 break;
3229 case Bytecodes::_fast_fputfield:
3230 NOT_LP64( __ fstp_s(field); )
3231 LP64_ONLY( __ movflt(field, xmm0);)
3232 break;
3233 case Bytecodes::_fast_dputfield:
3234 NOT_LP64( __ fstp_d(field); )
3235 LP64_ONLY( __ movdbl(field, xmm0);)
3236 break;
3237 default:
3238 ShouldNotReachHere();
3239 }
3240
3241 // Check for volatile store
3242 __ testl(rdx, rdx);
3243 __ jcc(Assembler::zero, notVolatile);
3244 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3245 Assembler::StoreStore));
3246 __ bind(notVolatile);
3247 }
3248
3249 void TemplateTable::fast_accessfield(TosState state) {
3250 transition(atos, state);
3251
3252 // Do the JVMTI work here to avoid disturbing the register state below
3253 if (JvmtiExport::can_post_field_access()) {
3254 // Check to see if a field access watch has been set before we
3255 // take the time to call into the VM.
3256 Label L1;
3257 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3258 __ testl(rcx, rcx);
3259 __ jcc(Assembler::zero, L1);
3260 // access constant pool cache entry
3261 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3262 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3263 __ verify_oop(rax);
3264 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3265 LP64_ONLY(__ mov(c_rarg1, rax));
3266 // c_rarg1: object pointer copied above
3267 // c_rarg2: cache entry pointer
3268 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3269 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3270 __ pop_ptr(rax); // restore object pointer
3271 __ bind(L1);
3272 }
3273
3274 // access constant pool cache
3275 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3276 // replace index with field offset from cache entry
3277 // [jk] not needed currently
3278 // if (os::is_MP()) {
3279 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3280 // in_bytes(ConstantPoolCache::base_offset() +
3281 // ConstantPoolCacheEntry::flags_offset())));
3282 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3283 // __ andl(rdx, 0x1);
3284 // }
3285 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3286 in_bytes(ConstantPoolCache::base_offset() +
3287 ConstantPoolCacheEntry::f2_offset())));
3288
3289 // rax: object
3290 __ verify_oop(rax);
3291 __ null_check(rax);
3292 Address field(rax, rbx, Address::times_1);
3293
3294 // access field
3295 switch (bytecode()) {
3296 case Bytecodes::_fast_agetfield:
3297 __ load_heap_oop(rax, field);
3298 __ verify_oop(rax);
3299 break;
3300 case Bytecodes::_fast_lgetfield:
3301 #ifdef _LP64
3302 __ movq(rax, field);
3303 #else
3304 __ stop("should not be rewritten");
3305 #endif
3306 break;
3307 case Bytecodes::_fast_igetfield:
3308 __ movl(rax, field);
3309 break;
3310 case Bytecodes::_fast_bgetfield:
3311 __ movsbl(rax, field);
3312 break;
3313 case Bytecodes::_fast_sgetfield:
3314 __ load_signed_short(rax, field);
3315 break;
3316 case Bytecodes::_fast_cgetfield:
3317 __ load_unsigned_short(rax, field);
3318 break;
3319 case Bytecodes::_fast_fgetfield:
3320 LP64_ONLY(__ movflt(xmm0, field));
3321 NOT_LP64(__ fld_s(field));
3322 break;
3323 case Bytecodes::_fast_dgetfield:
3324 LP64_ONLY(__ movdbl(xmm0, field));
3325 NOT_LP64(__ fld_d(field));
3326 break;
3327 default:
3328 ShouldNotReachHere();
3329 }
3330 // [jk] not needed currently
3331 // if (os::is_MP()) {
3332 // Label notVolatile;
3333 // __ testl(rdx, rdx);
3334 // __ jcc(Assembler::zero, notVolatile);
3335 // __ membar(Assembler::LoadLoad);
3336 // __ bind(notVolatile);
3337 //};
3338 }
3339
3340 void TemplateTable::fast_xaccess(TosState state) {
3341 transition(vtos, state);
3342
3343 // get receiver
3344 __ movptr(rax, aaddress(0));
3345 // access constant pool cache
3346 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3347 __ movptr(rbx,
3348 Address(rcx, rdx, Address::times_ptr,
3349 in_bytes(ConstantPoolCache::base_offset() +
3350 ConstantPoolCacheEntry::f2_offset())));
3351 // make sure exception is reported in correct bcp range (getfield is
3352 // next instruction)
3353 __ increment(rbcp);
3354 __ null_check(rax);
3355 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3356 switch (state) {
3357 case itos:
3358 __ movl(rax, field);
3359 break;
3360 case atos:
3361 __ load_heap_oop(rax, field);
3362 __ verify_oop(rax);
3363 break;
3364 case ftos:
3365 LP64_ONLY(__ movflt(xmm0, field));
3366 NOT_LP64(__ fld_s(field));
3367 break;
3368 default:
3369 ShouldNotReachHere();
3370 }
3371
3372 // [jk] not needed currently
3373 // if (os::is_MP()) {
3374 // Label notVolatile;
3375 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3376 // in_bytes(ConstantPoolCache::base_offset() +
3377 // ConstantPoolCacheEntry::flags_offset())));
3378 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3379 // __ testl(rdx, 0x1);
3380 // __ jcc(Assembler::zero, notVolatile);
3381 // __ membar(Assembler::LoadLoad);
3382 // __ bind(notVolatile);
3383 // }
3384
3385 __ decrement(rbcp);
3386 }
3387
3388 //-----------------------------------------------------------------------------
3389 // Calls
3390
3391 void TemplateTable::count_calls(Register method, Register temp) {
3392 // implemented elsewhere
3393 ShouldNotReachHere();
3394 }
3395
3396 void TemplateTable::prepare_invoke(int byte_no,
3397 Register method, // linked method (or i-klass)
3398 Register index, // itable index, MethodType, etc.
3399 Register recv, // if caller wants to see it
3400 Register flags // if caller wants to test it
3401 ) {
3402 // determine flags
3403 const Bytecodes::Code code = bytecode();
3404 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3405 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3406 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3407 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3408 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3409 const bool load_receiver = (recv != noreg);
3410 const bool save_flags = (flags != noreg);
3411 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3412 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3413 assert(flags == noreg || flags == rdx, "");
3414 assert(recv == noreg || recv == rcx, "");
3415
3416 // setup registers & access constant pool cache
3417 if (recv == noreg) recv = rcx;
3418 if (flags == noreg) flags = rdx;
3419 assert_different_registers(method, index, recv, flags);
3420
3421 // save 'interpreter return address'
3422 __ save_bcp();
3423
3424 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3425
3426 // maybe push appendix to arguments (just before return address)
3427 if (is_invokedynamic || is_invokehandle) {
3428 Label L_no_push;
3429 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3430 __ jcc(Assembler::zero, L_no_push);
3431 // Push the appendix as a trailing parameter.
3432 // This must be done before we get the receiver,
3433 // since the parameter_size includes it.
3434 __ push(rbx);
3435 __ mov(rbx, index);
3436 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3437 __ load_resolved_reference_at_index(index, rbx);
3438 __ pop(rbx);
3439 __ push(index); // push appendix (MethodType, CallSite, etc.)
3440 __ bind(L_no_push);
3441 }
3442
3443 // load receiver if needed (after appendix is pushed so parameter size is correct)
3444 // Note: no return address pushed yet
3445 if (load_receiver) {
3446 __ movl(recv, flags);
3447 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3448 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3449 const int receiver_is_at_end = -1; // back off one slot to get receiver
3450 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3451 __ movptr(recv, recv_addr);
3452 __ verify_oop(recv);
3453 }
3454
3455 if (save_flags) {
3456 __ movl(rbcp, flags);
3457 }
3458
3459 // compute return type
3460 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3461 // Make sure we don't need to mask flags after the above shift
3462 ConstantPoolCacheEntry::verify_tos_state_shift();
3463 // load return address
3464 {
3465 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3466 ExternalAddress table(table_addr);
3467 LP64_ONLY(__ lea(rscratch1, table));
3468 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3469 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3470 }
3471
3472 // push return address
3473 __ push(flags);
3474
3475 // Restore flags value from the constant pool cache, and restore rsi
3476 // for later null checks. r13 is the bytecode pointer
3477 if (save_flags) {
3478 __ movl(flags, rbcp);
3479 __ restore_bcp();
3480 }
3481 }
3482
3483 void TemplateTable::invokevirtual_helper(Register index,
3484 Register recv,
3485 Register flags) {
3486 // Uses temporary registers rax, rdx
3487 assert_different_registers(index, recv, rax, rdx);
3488 assert(index == rbx, "");
3489 assert(recv == rcx, "");
3490
3491 // Test for an invoke of a final method
3492 Label notFinal;
3493 __ movl(rax, flags);
3494 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3495 __ jcc(Assembler::zero, notFinal);
3496
3497 const Register method = index; // method must be rbx
3498 assert(method == rbx,
3499 "Method* must be rbx for interpreter calling convention");
3500
3501 // do the call - the index is actually the method to call
3502 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3503
3504 // It's final, need a null check here!
3505 __ null_check(recv);
3506
3507 // profile this call
3508 __ profile_final_call(rax);
3509 __ profile_arguments_type(rax, method, rbcp, true);
3510
3511 __ jump_from_interpreted(method, rax);
3512
3513 __ bind(notFinal);
3514
3515 // get receiver klass
3516 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3517 __ load_klass(rax, recv);
3518
3519 // profile this call
3520 __ profile_virtual_call(rax, rlocals, rdx);
3521 // get target Method* & entry point
3522 __ lookup_virtual_method(rax, index, method);
3523 __ profile_arguments_type(rdx, method, rbcp, true);
3524 __ jump_from_interpreted(method, rdx);
3525 }
3526
3527 void TemplateTable::invokevirtual(int byte_no) {
3528 transition(vtos, vtos);
3529 assert(byte_no == f2_byte, "use this argument");
3530 prepare_invoke(byte_no,
3531 rbx, // method or vtable index
3532 noreg, // unused itable index
3533 rcx, rdx); // recv, flags
3534
3535 // rbx: index
3536 // rcx: receiver
3537 // rdx: flags
3538
3539 invokevirtual_helper(rbx, rcx, rdx);
3540 }
3541
3542 void TemplateTable::invokespecial(int byte_no) {
3543 transition(vtos, vtos);
3544 assert(byte_no == f1_byte, "use this argument");
3545 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3546 rcx); // get receiver also for null check
3547 __ verify_oop(rcx);
3548 __ null_check(rcx);
3549 // do the call
3550 __ profile_call(rax);
3551 __ profile_arguments_type(rax, rbx, rbcp, false);
3552 __ jump_from_interpreted(rbx, rax);
3553 }
3554
3555 void TemplateTable::invokestatic(int byte_no) {
3556 transition(vtos, vtos);
3557 assert(byte_no == f1_byte, "use this argument");
3558 prepare_invoke(byte_no, rbx); // get f1 Method*
3559 // do the call
3560 __ profile_call(rax);
3561 __ profile_arguments_type(rax, rbx, rbcp, false);
3562 __ jump_from_interpreted(rbx, rax);
3563 }
3564
3565
3566 void TemplateTable::fast_invokevfinal(int byte_no) {
3567 transition(vtos, vtos);
3568 assert(byte_no == f2_byte, "use this argument");
3569 __ stop("fast_invokevfinal not used on x86");
3570 }
3571
3572
3573 void TemplateTable::invokeinterface(int byte_no) {
3574 transition(vtos, vtos);
3575 assert(byte_no == f1_byte, "use this argument");
3576 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3577 rcx, rdx); // recv, flags
3578
3579 // rax: interface klass (from f1)
3580 // rbx: itable index (from f2)
3581 // rcx: receiver
3582 // rdx: flags
3583
3584 // Special case of invokeinterface called for virtual method of
3585 // java.lang.Object. See cpCacheOop.cpp for details.
3586 // This code isn't produced by javac, but could be produced by
3587 // another compliant java compiler.
3588 Label notMethod;
3589 __ movl(rlocals, rdx);
3590 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3591
3592 __ jcc(Assembler::zero, notMethod);
3593
3594 invokevirtual_helper(rbx, rcx, rdx);
3595 __ bind(notMethod);
3596
3597 // Get receiver klass into rdx - also a null check
3598 __ restore_locals(); // restore r14
3599 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3600 __ load_klass(rdx, rcx);
3601
3602 // profile this call
3603 __ profile_virtual_call(rdx, rbcp, rlocals);
3604
3605 Label no_such_interface, no_such_method;
3606
3607 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3608 rdx, rax, rbx,
3609 // outputs: method, scan temp. reg
3610 rbx, rbcp,
3611 no_such_interface);
3612
3613 // rbx: Method* to call
3614 // rcx: receiver
3615 // Check for abstract method error
3616 // Note: This should be done more efficiently via a throw_abstract_method_error
3617 // interpreter entry point and a conditional jump to it in case of a null
3618 // method.
3619 __ testptr(rbx, rbx);
3620 __ jcc(Assembler::zero, no_such_method);
3621
3622 __ profile_arguments_type(rdx, rbx, rbcp, true);
3623
3624 // do the call
3625 // rcx: receiver
3626 // rbx,: Method*
3627 __ jump_from_interpreted(rbx, rdx);
3628 __ should_not_reach_here();
3629
3630 // exception handling code follows...
3631 // note: must restore interpreter registers to canonical
3632 // state for exception handling to work correctly!
3633
3634 __ bind(no_such_method);
3635 // throw exception
3636 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3637 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3638 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3639 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3640 // the call_VM checks for exception, so we should never return here.
3641 __ should_not_reach_here();
3642
3643 __ bind(no_such_interface);
3644 // throw exception
3645 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3646 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3647 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3648 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3649 InterpreterRuntime::throw_IncompatibleClassChangeError));
3650 // the call_VM checks for exception, so we should never return here.
3651 __ should_not_reach_here();
3652 }
3653
3654 void TemplateTable::invokehandle(int byte_no) {
3655 transition(vtos, vtos);
3656 assert(byte_no == f1_byte, "use this argument");
3657 const Register rbx_method = rbx;
3658 const Register rax_mtype = rax;
3659 const Register rcx_recv = rcx;
3660 const Register rdx_flags = rdx;
3661
3662 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3663 __ verify_method_ptr(rbx_method);
3664 __ verify_oop(rcx_recv);
3665 __ null_check(rcx_recv);
3666
3667 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3668 // rbx: MH.invokeExact_MT method (from f2)
3669
3670 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3671
3672 // FIXME: profile the LambdaForm also
3673 __ profile_final_call(rax);
3674 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3675
3676 __ jump_from_interpreted(rbx_method, rdx);
3677 }
3678
3679 void TemplateTable::invokedynamic(int byte_no) {
3680 transition(vtos, vtos);
3681 assert(byte_no == f1_byte, "use this argument");
3682
3683 const Register rbx_method = rbx;
3684 const Register rax_callsite = rax;
3685
3686 prepare_invoke(byte_no, rbx_method, rax_callsite);
3687
3688 // rax: CallSite object (from cpool->resolved_references[f1])
3689 // rbx: MH.linkToCallSite method (from f2)
3690
3691 // Note: rax_callsite is already pushed by prepare_invoke
3692
3693 // %%% should make a type profile for any invokedynamic that takes a ref argument
3694 // profile this call
3695 __ profile_call(rbcp);
3696 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3697
3698 __ verify_oop(rax_callsite);
3699
3700 __ jump_from_interpreted(rbx_method, rdx);
3701 }
3702
3703 //-----------------------------------------------------------------------------
3704 // Allocation
3705
3706 void TemplateTable::_new() {
3707 transition(vtos, atos);
3708 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3709 Label slow_case;
3710 Label slow_case_no_pop;
3711 Label done;
3712 Label initialize_header;
3713 Label initialize_object; // including clearing the fields
3714 Label allocate_shared;
3715
3716 __ get_cpool_and_tags(rcx, rax);
3717
3718 // Make sure the class we're about to instantiate has been resolved.
3719 // This is done before loading InstanceKlass to be consistent with the order
3720 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3721 const int tags_offset = Array<u1>::base_offset_in_bytes();
3722 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3723 __ jcc(Assembler::notEqual, slow_case_no_pop);
3724
3725 // get InstanceKlass
3726 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool)));
3727 __ push(rcx); // save the contexts of klass for initializing the header
3728
3729 // make sure klass is initialized & doesn't have finalizer
3730 // make sure klass is fully initialized
3731 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3732 __ jcc(Assembler::notEqual, slow_case);
3733
3734 // get instance_size in InstanceKlass (scaled to a count of bytes)
3735 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3736 // test to see if it has a finalizer or is malformed in some way
3737 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3738 __ jcc(Assembler::notZero, slow_case);
3739
3740 //
3741 // Allocate the instance
3742 // 1) Try to allocate in the TLAB
3743 // 2) if fail and the object is large allocate in the shared Eden
3744 // 3) if the above fails (or is not applicable), go to a slow case
3745 // (creates a new TLAB, etc.)
3746
3747 const bool allow_shared_alloc =
3748 Universe::heap()->supports_inline_contig_alloc();
3749
3750 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
3751 #ifndef _LP64
3752 if (UseTLAB || allow_shared_alloc) {
3753 __ get_thread(thread);
3754 }
3755 #endif // _LP64
3756
3757 if (UseTLAB) {
3758 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3759 __ lea(rbx, Address(rax, rdx, Address::times_1));
3760 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3761 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3762 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3763 if (ZeroTLAB) {
3764 // the fields have been already cleared
3765 __ jmp(initialize_header);
3766 } else {
3767 // initialize both the header and fields
3768 __ jmp(initialize_object);
3769 }
3770 }
3771
3772 // Allocation in the shared Eden, if allowed.
3773 //
3774 // rdx: instance size in bytes
3775 if (allow_shared_alloc) {
3776 __ bind(allocate_shared);
3777
3778 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3779 ExternalAddress heap_end((address)Universe::heap()->end_addr());
3780
3781 Label retry;
3782 __ bind(retry);
3783 __ movptr(rax, heap_top);
3784 __ lea(rbx, Address(rax, rdx, Address::times_1));
3785 __ cmpptr(rbx, heap_end);
3786 __ jcc(Assembler::above, slow_case);
3787
3788 // Compare rax, with the top addr, and if still equal, store the new
3789 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3790 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3791 //
3792 // rax,: object begin
3793 // rbx,: object end
3794 // rdx: instance size in bytes
3795 __ locked_cmpxchgptr(rbx, heap_top);
3796
3797 // if someone beat us on the allocation, try again, otherwise continue
3798 __ jcc(Assembler::notEqual, retry);
3799
3800 __ incr_allocated_bytes(thread, rdx, 0);
3801 }
3802
3803 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3804 // The object is initialized before the header. If the object size is
3805 // zero, go directly to the header initialization.
3806 __ bind(initialize_object);
3807 __ decrement(rdx, sizeof(oopDesc));
3808 __ jcc(Assembler::zero, initialize_header);
3809
3810 // Initialize topmost object field, divide rdx by 8, check if odd and
3811 // test if zero.
3812 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3813 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3814
3815 // rdx must have been multiple of 8
3816 #ifdef ASSERT
3817 // make sure rdx was multiple of 8
3818 Label L;
3819 // Ignore partial flag stall after shrl() since it is debug VM
3820 __ jccb(Assembler::carryClear, L);
3821 __ stop("object size is not multiple of 2 - adjust this code");
3822 __ bind(L);
3823 // rdx must be > 0, no extra check needed here
3824 #endif
3825
3826 // initialize remaining object fields: rdx was a multiple of 8
3827 { Label loop;
3828 __ bind(loop);
3829 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3830 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3831 __ decrement(rdx);
3832 __ jcc(Assembler::notZero, loop);
3833 }
3834
3835 // initialize object header only.
3836 __ bind(initialize_header);
3837 if (UseBiasedLocking) {
3838 __ pop(rcx); // get saved klass back in the register.
3839 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
3840 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3841 } else {
3842 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3843 (intptr_t)markOopDesc::prototype()); // header
3844 __ pop(rcx); // get saved klass back in the register.
3845 }
3846 #ifdef _LP64
3847 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
3848 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
3849 #endif
3850 __ store_klass(rax, rcx); // klass
3851
3852 {
3853 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3854 // Trigger dtrace event for fastpath
3855 __ push(atos);
3856 __ call_VM_leaf(
3857 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3858 __ pop(atos);
3859 }
3860
3861 __ jmp(done);
3862 }
3863
3864 // slow case
3865 __ bind(slow_case);
3866 __ pop(rcx); // restore stack pointer to what it was when we came in.
3867 __ bind(slow_case_no_pop);
3868
3869 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
3870 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3871
3872 __ get_constant_pool(rarg1);
3873 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
3874 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
3875 __ verify_oop(rax);
3876
3877 // continue
3878 __ bind(done);
3879 }
3880
3881 void TemplateTable::newarray() {
3882 transition(itos, atos);
3883 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
3884 __ load_unsigned_byte(rarg1, at_bcp(1));
3885 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3886 rarg1, rax);
3887 }
3888
3889 void TemplateTable::anewarray() {
3890 transition(itos, atos);
3891
3892 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
3893 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3894
3895 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
3896 __ get_constant_pool(rarg1);
3897 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3898 rarg1, rarg2, rax);
3899 }
3900
3901 void TemplateTable::arraylength() {
3902 transition(atos, itos);
3903 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3904 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3905 }
3906
3907 void TemplateTable::checkcast() {
3908 transition(atos, atos);
3909 Label done, is_null, ok_is_subtype, quicked, resolved;
3910 __ testptr(rax, rax); // object is in rax
3911 __ jcc(Assembler::zero, is_null);
3912
3913 // Get cpool & tags index
3914 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3915 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3916 // See if bytecode has already been quicked
3917 __ cmpb(Address(rdx, rbx,
3918 Address::times_1,
3919 Array<u1>::base_offset_in_bytes()),
3920 JVM_CONSTANT_Class);
3921 __ jcc(Assembler::equal, quicked);
3922 __ push(atos); // save receiver for result, and for GC
3923 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3924
3925 // vm_result_2 has metadata result
3926 #ifndef _LP64
3927 // borrow rdi from locals
3928 __ get_thread(rdi);
3929 __ get_vm_result_2(rax, rdi);
3930 __ restore_locals();
3931 #else
3932 __ get_vm_result_2(rax, r15_thread);
3933 #endif
3934
3935 __ pop_ptr(rdx); // restore receiver
3936 __ jmpb(resolved);
3937
3938 // Get superklass in rax and subklass in rbx
3939 __ bind(quicked);
3940 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
3941 __ movptr(rax, Address(rcx, rbx,
3942 Address::times_ptr, sizeof(ConstantPool)));
3943
3944 __ bind(resolved);
3945 __ load_klass(rbx, rdx);
3946
3947 // Generate subtype check. Blows rcx, rdi. Object in rdx.
3948 // Superklass in rax. Subklass in rbx.
3949 __ gen_subtype_check(rbx, ok_is_subtype);
3950
3951 // Come here on failure
3952 __ push_ptr(rdx);
3953 // object is at TOS
3954 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3955
3956 // Come here on success
3957 __ bind(ok_is_subtype);
3958 __ mov(rax, rdx); // Restore object in rdx
3959
3960 // Collect counts on whether this check-cast sees NULLs a lot or not.
3961 if (ProfileInterpreter) {
3962 __ jmp(done);
3963 __ bind(is_null);
3964 __ profile_null_seen(rcx);
3965 } else {
3966 __ bind(is_null); // same as 'done'
3967 }
3968 __ bind(done);
3969 }
3970
3971 void TemplateTable::instanceof() {
3972 transition(atos, itos);
3973 Label done, is_null, ok_is_subtype, quicked, resolved;
3974 __ testptr(rax, rax);
3975 __ jcc(Assembler::zero, is_null);
3976
3977 // Get cpool & tags index
3978 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3979 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3980 // See if bytecode has already been quicked
3981 __ cmpb(Address(rdx, rbx,
3982 Address::times_1,
3983 Array<u1>::base_offset_in_bytes()),
3984 JVM_CONSTANT_Class);
3985 __ jcc(Assembler::equal, quicked);
3986
3987 __ push(atos); // save receiver for result, and for GC
3988 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
3989 // vm_result_2 has metadata result
3990
3991 #ifndef _LP64
3992 // borrow rdi from locals
3993 __ get_thread(rdi);
3994 __ get_vm_result_2(rax, rdi);
3995 __ restore_locals();
3996 #else
3997 __ get_vm_result_2(rax, r15_thread);
3998 #endif
3999
4000 __ pop_ptr(rdx); // restore receiver
4001 __ verify_oop(rdx);
4002 __ load_klass(rdx, rdx);
4003 __ jmpb(resolved);
4004
4005 // Get superklass in rax and subklass in rdx
4006 __ bind(quicked);
4007 __ load_klass(rdx, rax);
4008 __ movptr(rax, Address(rcx, rbx,
4009 Address::times_ptr, sizeof(ConstantPool)));
4010
4011 __ bind(resolved);
4012
4013 // Generate subtype check. Blows rcx, rdi
4014 // Superklass in rax. Subklass in rdx.
4015 __ gen_subtype_check(rdx, ok_is_subtype);
4016
4017 // Come here on failure
4018 __ xorl(rax, rax);
4019 __ jmpb(done);
4020 // Come here on success
4021 __ bind(ok_is_subtype);
4022 __ movl(rax, 1);
4023
4024 // Collect counts on whether this test sees NULLs a lot or not.
4025 if (ProfileInterpreter) {
4026 __ jmp(done);
4027 __ bind(is_null);
4028 __ profile_null_seen(rcx);
4029 } else {
4030 __ bind(is_null); // same as 'done'
4031 }
4032 __ bind(done);
4033 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4034 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4035 }
4036
4037
4038 //----------------------------------------------------------------------------------------------------
4039 // Breakpoints
4040 void TemplateTable::_breakpoint() {
4041 // Note: We get here even if we are single stepping..
4042 // jbug insists on setting breakpoints at every bytecode
4043 // even if we are in single step mode.
4044
4045 transition(vtos, vtos);
4046
4047 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4048
4049 // get the unpatched byte code
4050 __ get_method(rarg);
4051 __ call_VM(noreg,
4052 CAST_FROM_FN_PTR(address,
4053 InterpreterRuntime::get_original_bytecode_at),
4054 rarg, rbcp);
4055 __ mov(rbx, rax); // why?
4056
4057 // post the breakpoint event
4058 __ get_method(rarg);
4059 __ call_VM(noreg,
4060 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4061 rarg, rbcp);
4062
4063 // complete the execution of original bytecode
4064 __ dispatch_only_normal(vtos);
4065 }
4066
4067 //-----------------------------------------------------------------------------
4068 // Exceptions
4069
4070 void TemplateTable::athrow() {
4071 transition(atos, vtos);
4072 __ null_check(rax);
4073 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4074 }
4075
4076 //-----------------------------------------------------------------------------
4077 // Synchronization
4078 //
4079 // Note: monitorenter & exit are symmetric routines; which is reflected
4080 // in the assembly code structure as well
4081 //
4082 // Stack layout:
4083 //
4084 // [expressions ] <--- rsp = expression stack top
4085 // ..
4086 // [expressions ]
4087 // [monitor entry] <--- monitor block top = expression stack bot
4088 // ..
4089 // [monitor entry]
4090 // [frame data ] <--- monitor block bot
4091 // ...
4092 // [saved rbp ] <--- rbp
4093 void TemplateTable::monitorenter() {
4094 transition(atos, vtos);
4095
4096 // check for NULL object
4097 __ null_check(rax);
4098
4099 const Address monitor_block_top(
4100 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4101 const Address monitor_block_bot(
4102 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4103 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4104
4105 Label allocated;
4106
4107 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4108 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4109 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4110
4111 // initialize entry pointer
4112 __ xorl(rmon, rmon); // points to free slot or NULL
4113
4114 // find a free slot in the monitor block (result in rmon)
4115 {
4116 Label entry, loop, exit;
4117 __ movptr(rtop, monitor_block_top); // points to current entry,
4118 // starting with top-most entry
4119 __ lea(rbot, monitor_block_bot); // points to word before bottom
4120 // of monitor block
4121 __ jmpb(entry);
4122
4123 __ bind(loop);
4124 // check if current entry is used
4125 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4126 // if not used then remember entry in rmon
4127 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4128 // check if current entry is for same object
4129 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4130 // if same object then stop searching
4131 __ jccb(Assembler::equal, exit);
4132 // otherwise advance to next entry
4133 __ addptr(rtop, entry_size);
4134 __ bind(entry);
4135 // check if bottom reached
4136 __ cmpptr(rtop, rbot);
4137 // if not at bottom then check this entry
4138 __ jcc(Assembler::notEqual, loop);
4139 __ bind(exit);
4140 }
4141
4142 __ testptr(rmon, rmon); // check if a slot has been found
4143 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4144
4145 // allocate one if there's no free slot
4146 {
4147 Label entry, loop;
4148 // 1. compute new pointers // rsp: old expression stack top
4149 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4150 __ subptr(rsp, entry_size); // move expression stack top
4151 __ subptr(rmon, entry_size); // move expression stack bottom
4152 __ mov(rtop, rsp); // set start value for copy loop
4153 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4154 __ jmp(entry);
4155 // 2. move expression stack contents
4156 __ bind(loop);
4157 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4158 // word from old location
4159 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4160 __ addptr(rtop, wordSize); // advance to next word
4161 __ bind(entry);
4162 __ cmpptr(rtop, rmon); // check if bottom reached
4163 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4164 // copy next word
4165 }
4166
4167 // call run-time routine
4168 // rmon: points to monitor entry
4169 __ bind(allocated);
4170
4171 // Increment bcp to point to the next bytecode, so exception
4172 // handling for async. exceptions work correctly.
4173 // The object has already been poped from the stack, so the
4174 // expression stack looks correct.
4175 __ increment(rbcp);
4176
4177 // store object
4178 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4179 __ lock_object(rmon);
4180
4181 // check to make sure this monitor doesn't cause stack overflow after locking
4182 __ save_bcp(); // in case of exception
4183 __ generate_stack_overflow_check(0);
4184
4185 // The bcp has already been incremented. Just need to dispatch to
4186 // next instruction.
4187 __ dispatch_next(vtos);
4188 }
4189
4190 void TemplateTable::monitorexit() {
4191 transition(atos, vtos);
4192
4193 // check for NULL object
4194 __ null_check(rax);
4195
4196 const Address monitor_block_top(
4197 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4198 const Address monitor_block_bot(
4199 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4200 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4201
4202 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4203 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4204
4205 Label found;
4206
4207 // find matching slot
4208 {
4209 Label entry, loop;
4210 __ movptr(rtop, monitor_block_top); // points to current entry,
4211 // starting with top-most entry
4212 __ lea(rbot, monitor_block_bot); // points to word before bottom
4213 // of monitor block
4214 __ jmpb(entry);
4215
4216 __ bind(loop);
4217 // check if current entry is for same object
4218 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4219 // if same object then stop searching
4220 __ jcc(Assembler::equal, found);
4221 // otherwise advance to next entry
4222 __ addptr(rtop, entry_size);
4223 __ bind(entry);
4224 // check if bottom reached
4225 __ cmpptr(rtop, rbot);
4226 // if not at bottom then check this entry
4227 __ jcc(Assembler::notEqual, loop);
4228 }
4229
4230 // error handling. Unlocking was not block-structured
4231 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4232 InterpreterRuntime::throw_illegal_monitor_state_exception));
4233 __ should_not_reach_here();
4234
4235 // call run-time routine
4236 __ bind(found);
4237 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4238 __ unlock_object(rtop);
4239 __ pop_ptr(rax); // discard object
4240 }
4241
4242 // Wide instructions
4243 void TemplateTable::wide() {
4244 transition(vtos, vtos);
4245 __ load_unsigned_byte(rbx, at_bcp(1));
4246 ExternalAddress wtable((address)Interpreter::_wentry_point);
4247 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4248 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4249 }
4250
4251 // Multi arrays
4252 void TemplateTable::multianewarray() {
4253 transition(vtos, atos);
4254
4255 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4256 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4257 // last dim is on top of stack; we want address of first one:
4258 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4259 // the latter wordSize to point to the beginning of the array.
4260 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4261 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4262 __ load_unsigned_byte(rbx, at_bcp(3));
4263 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4264 }
4265 #endif /* !CC_INTERP */
4266