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 #define __ _masm->
42
43 // Global Register Names
44 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
45 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
46
47 // Platform-dependent initialization
48 void TemplateTable::pd_initialize() {
49 // No x86 specific initialization
50 }
51
52 // Address Computation: local variables
53 static inline Address iaddress(int n) {
54 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
55 }
56
57 static inline Address laddress(int n) {
58 return iaddress(n + 1);
59 }
60
61 #ifndef _LP64
62 static inline Address haddress(int n) {
63 return iaddress(n + 0);
64 }
65 #endif
66
67 static inline Address faddress(int n) {
68 return iaddress(n);
69 }
70
71 static inline Address daddress(int n) {
72 return laddress(n);
73 }
74
75 static inline Address aaddress(int n) {
76 return iaddress(n);
77 }
78
79 static inline Address iaddress(Register r) {
80 return Address(rlocals, r, Address::times_ptr);
81 }
82
83 static inline Address laddress(Register r) {
84 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
85 }
86
87 #ifndef _LP64
88 static inline Address haddress(Register r) {
89 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
90 }
91 #endif
92
93 static inline Address faddress(Register r) {
94 return iaddress(r);
95 }
96
97 static inline Address daddress(Register r) {
98 return laddress(r);
99 }
100
101 static inline Address aaddress(Register r) {
102 return iaddress(r);
103 }
104
105
106 // expression stack
107 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
108 // data beyond the rsp which is potentially unsafe in an MT environment;
109 // an interrupt may overwrite that data.)
110 static inline Address at_rsp () {
111 return Address(rsp, 0);
112 }
113
114 // At top of Java expression stack which may be different than esp(). It
115 // isn't for category 1 objects.
116 static inline Address at_tos () {
117 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
118 }
119
120 static inline Address at_tos_p1() {
121 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
122 }
123
124 static inline Address at_tos_p2() {
125 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
126 }
127
128 // Condition conversion
129 static Assembler::Condition j_not(TemplateTable::Condition cc) {
130 switch (cc) {
131 case TemplateTable::equal : return Assembler::notEqual;
132 case TemplateTable::not_equal : return Assembler::equal;
133 case TemplateTable::less : return Assembler::greaterEqual;
134 case TemplateTable::less_equal : return Assembler::greater;
135 case TemplateTable::greater : return Assembler::lessEqual;
136 case TemplateTable::greater_equal: return Assembler::less;
137 }
138 ShouldNotReachHere();
139 return Assembler::zero;
140 }
141
142
143
144 // Miscelaneous helper routines
145 // Store an oop (or NULL) at the address described by obj.
146 // If val == noreg this means store a NULL
147
148
149 static void do_oop_store(InterpreterMacroAssembler* _masm,
150 Address obj,
151 Register val,
152 BarrierSet::Name barrier,
153 bool precise) {
154 assert(val == noreg || val == rax, "parameter is just for looks");
155 switch (barrier) {
156 #if INCLUDE_ALL_GCS
157 case BarrierSet::G1SATBCTLogging:
158 {
159 // flatten object address if needed
160 // We do it regardless of precise because we need the registers
161 if (obj.index() == noreg && obj.disp() == 0) {
162 if (obj.base() != rdx) {
163 __ movptr(rdx, obj.base());
164 }
165 } else {
166 __ lea(rdx, obj);
167 }
168
169 Register rtmp = LP64_ONLY(r8) NOT_LP64(rsi);
170 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
171
172 NOT_LP64(__ get_thread(rcx));
173 NOT_LP64(__ save_bcp());
174
175 __ g1_write_barrier_pre(rdx /* obj */,
176 rbx /* pre_val */,
177 rthread /* thread */,
178 rtmp /* tmp */,
179 val != noreg /* tosca_live */,
180 false /* expand_call */);
181 if (val == noreg) {
182 __ store_heap_oop_null(Address(rdx, 0));
183 } else {
184 // G1 barrier needs uncompressed oop for region cross check.
185 Register new_val = val;
186 if (UseCompressedOops) {
187 new_val = rbx;
188 __ movptr(new_val, val);
189 }
190 __ store_heap_oop(Address(rdx, 0), val);
191 __ g1_write_barrier_post(rdx /* store_adr */,
192 new_val /* new_val */,
193 rthread /* thread */,
194 rtmp /* tmp */,
195 rbx /* tmp2 */);
196 }
197 NOT_LP64( __ restore_bcp());
198 }
199 break;
200 #endif // INCLUDE_ALL_GCS
201 case BarrierSet::CardTableForRS:
202 case BarrierSet::CardTableExtension:
203 {
204 if (val == noreg) {
205 __ store_heap_oop_null(obj);
206 } else {
207 __ store_heap_oop(obj, val);
208 // flatten object address if needed
209 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
210 __ store_check(obj.base());
211 } else {
212 __ lea(rdx, obj);
213 __ store_check(rdx);
214 }
215 }
216 }
217 break;
218 case BarrierSet::ModRef:
219 if (val == noreg) {
220 __ store_heap_oop_null(obj);
221 } else {
222 __ store_heap_oop(obj, val);
223 }
224 break;
225 default :
226 ShouldNotReachHere();
227
228 }
229 }
230
231 Address TemplateTable::at_bcp(int offset) {
232 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
233 return Address(rbcp, offset);
234 }
235
236
237 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
238 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
239 int byte_no) {
240 if (!RewriteBytecodes) return;
241 Label L_patch_done;
242
243 switch (bc) {
244 case Bytecodes::_fast_aputfield:
245 case Bytecodes::_fast_bputfield:
246 case Bytecodes::_fast_cputfield:
247 case Bytecodes::_fast_dputfield:
248 case Bytecodes::_fast_fputfield:
249 case Bytecodes::_fast_iputfield:
250 case Bytecodes::_fast_lputfield:
251 case Bytecodes::_fast_sputfield:
252 {
253 // We skip bytecode quickening for putfield instructions when
254 // the put_code written to the constant pool cache is zero.
255 // This is required so that every execution of this instruction
256 // calls out to InterpreterRuntime::resolve_get_put to do
257 // additional, required work.
258 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
259 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
260 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
261 __ movl(bc_reg, bc);
262 __ cmpl(temp_reg, (int) 0);
263 __ jcc(Assembler::zero, L_patch_done); // don't patch
264 }
265 break;
266 default:
267 assert(byte_no == -1, "sanity");
268 // the pair bytecodes have already done the load.
269 if (load_bc_into_bc_reg) {
270 __ movl(bc_reg, bc);
271 }
272 }
273
274 if (JvmtiExport::can_post_breakpoint()) {
275 Label L_fast_patch;
276 // if a breakpoint is present we can't rewrite the stream directly
277 __ movzbl(temp_reg, at_bcp(0));
278 __ cmpl(temp_reg, Bytecodes::_breakpoint);
279 __ jcc(Assembler::notEqual, L_fast_patch);
280 __ get_method(temp_reg);
281 // Let breakpoint table handling rewrite to quicker bytecode
282 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
283 #ifndef ASSERT
284 __ jmpb(L_patch_done);
285 #else
286 __ jmp(L_patch_done);
287 #endif
288 __ bind(L_fast_patch);
289 }
290
291 #ifdef ASSERT
292 Label L_okay;
293 __ load_unsigned_byte(temp_reg, at_bcp(0));
294 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
295 __ jcc(Assembler::equal, L_okay);
296 __ cmpl(temp_reg, bc_reg);
297 __ jcc(Assembler::equal, L_okay);
298 __ stop("patching the wrong bytecode");
299 __ bind(L_okay);
300 #endif
301
302 // patch bytecode
303 __ movb(at_bcp(0), bc_reg);
304 __ bind(L_patch_done);
305 }
306 // Individual instructions
307
308
309 void TemplateTable::nop() {
310 transition(vtos, vtos);
311 // nothing to do
312 }
313
314 void TemplateTable::shouldnotreachhere() {
315 transition(vtos, vtos);
316 __ stop("shouldnotreachhere bytecode");
317 }
318
319 void TemplateTable::aconst_null() {
320 transition(vtos, atos);
321 __ xorl(rax, rax);
322 }
323
324 void TemplateTable::iconst(int value) {
325 transition(vtos, itos);
326 if (value == 0) {
327 __ xorl(rax, rax);
328 } else {
329 __ movl(rax, value);
330 }
331 }
332
333 void TemplateTable::lconst(int value) {
334 transition(vtos, ltos);
335 if (value == 0) {
336 __ xorl(rax, rax);
337 } else {
338 __ movl(rax, value);
339 }
340 #ifndef _LP64
341 assert(value >= 0, "check this code");
342 __ xorptr(rdx, rdx);
343 #endif
344 }
345
346
347
348 void TemplateTable::fconst(int value) {
349 transition(vtos, ftos);
350 if (UseSSE >= 1) {
351 static float one = 1.0f, two = 2.0f;
352 switch (value) {
353 case 0:
354 __ xorps(xmm0, xmm0);
355 break;
356 case 1:
357 __ movflt(xmm0, ExternalAddress((address) &one));
358 break;
359 case 2:
360 __ movflt(xmm0, ExternalAddress((address) &two));
361 break;
362 default:
363 ShouldNotReachHere();
364 break;
365 }
366 } else {
367 #ifdef _LP64
368 ShouldNotReachHere();
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 // _LP64
376 }
377 }
378
379 void TemplateTable::dconst(int value) {
380 transition(vtos, dtos);
381 if (UseSSE >= 2) {
382 static double one = 1.0;
383 switch (value) {
384 case 0:
385 __ xorpd(xmm0, xmm0);
386 break;
387 case 1:
388 __ movdbl(xmm0, ExternalAddress((address) &one));
389 break;
390 default:
391 ShouldNotReachHere();
392 break;
393 }
394 } else {
395 #ifdef _LP64
396 ShouldNotReachHere();
397 #else
398 if (value == 0) { __ fldz();
399 } else if (value == 1) { __ fld1();
400 } else { ShouldNotReachHere();
401 }
402 #endif
403 }
404 }
405
406 void TemplateTable::bipush() {
407 transition(vtos, itos);
408 __ load_signed_byte(rax, at_bcp(1));
409 }
410
411 void TemplateTable::sipush() {
412 transition(vtos, itos);
413 __ load_unsigned_short(rax, at_bcp(1));
414 __ bswapl(rax);
415 __ sarl(rax, 16);
416 }
417
418 void TemplateTable::ldc(bool wide) {
419 transition(vtos, vtos);
420 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
421 Label call_ldc, notFloat, notClass, Done;
422
423 if (wide) {
424 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
425 } else {
426 __ load_unsigned_byte(rbx, at_bcp(1));
427 }
428
429 __ get_cpool_and_tags(rcx, rax);
430 const int base_offset = ConstantPool::header_size() * wordSize;
431 const int tags_offset = Array<u1>::base_offset_in_bytes();
432
433 // get type
434 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
435
436 // unresolved class - get the resolved class
437 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
438 __ jccb(Assembler::equal, call_ldc);
439
440 // unresolved class in error state - call into runtime to throw the error
441 // from the first resolution attempt
442 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
443 __ jccb(Assembler::equal, call_ldc);
444
445 // resolved class - need to call vm to get java mirror of the class
446 __ cmpl(rdx, JVM_CONSTANT_Class);
447 __ jcc(Assembler::notEqual, notClass);
448
449 __ bind(call_ldc);
450
451 __ movl(rarg, wide);
452 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
453
454 __ push(atos);
455 __ jmp(Done);
456
457 __ bind(notClass);
458 __ cmpl(rdx, JVM_CONSTANT_Float);
459 __ jccb(Assembler::notEqual, notFloat);
460
461 // ftos
462 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
463 __ push(ftos);
464 __ jmp(Done);
465
466 __ bind(notFloat);
467 #ifdef ASSERT
468 {
469 Label L;
470 __ cmpl(rdx, JVM_CONSTANT_Integer);
471 __ jcc(Assembler::equal, L);
472 // String and Object are rewritten to fast_aldc
473 __ stop("unexpected tag type in ldc");
474 __ bind(L);
475 }
476 #endif
477 // itos JVM_CONSTANT_Integer only
478 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
479 __ push(itos);
480 __ bind(Done);
481 }
482
483 // Fast path for caching oop constants.
484 void TemplateTable::fast_aldc(bool wide) {
485 transition(vtos, atos);
486
487 Register result = rax;
488 Register tmp = rdx;
489 int index_size = wide ? sizeof(u2) : sizeof(u1);
490
491 Label resolved;
492
493 // We are resolved if the resolved reference cache entry contains a
494 // non-null object (String, MethodType, etc.)
495 assert_different_registers(result, tmp);
496 __ get_cache_index_at_bcp(tmp, 1, index_size);
497 __ load_resolved_reference_at_index(result, tmp);
498 __ testl(result, result);
499 __ jcc(Assembler::notZero, resolved);
500
501 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
502
503 // first time invocation - must resolve first
504 __ movl(tmp, (int)bytecode());
505 __ call_VM(result, entry, tmp);
506
507 __ bind(resolved);
508
509 if (VerifyOops) {
510 __ verify_oop(result);
511 }
512 }
513
514 void TemplateTable::ldc2_w() {
515 transition(vtos, vtos);
516 Label Long, Done;
517 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
518
519 __ get_cpool_and_tags(rcx, rax);
520 const int base_offset = ConstantPool::header_size() * wordSize;
521 const int tags_offset = Array<u1>::base_offset_in_bytes();
522
523 // get type
524 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
525 JVM_CONSTANT_Double);
526 __ jccb(Assembler::notEqual, Long);
527
528 // dtos
529 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
530 __ push(dtos);
531
532 __ jmpb(Done);
533 __ bind(Long);
534
535 // ltos
536 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
537 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
538 __ push(ltos);
539
540 __ bind(Done);
541 }
542
543 void TemplateTable::locals_index(Register reg, int offset) {
544 __ load_unsigned_byte(reg, at_bcp(offset));
545 __ negptr(reg);
546 }
547
548 void TemplateTable::iload() {
549 iload_internal();
550 }
551
552 void TemplateTable::nofast_iload() {
553 iload_internal(may_not_rewrite);
554 }
555
556 void TemplateTable::iload_internal(RewriteControl rc) {
557 transition(vtos, itos);
558 if (RewriteFrequentPairs && rc == may_rewrite) {
559 Label rewrite, done;
560 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
561 LP64_ONLY(assert(rbx != bc, "register damaged"));
562
563 // get next byte
564 __ load_unsigned_byte(rbx,
565 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
566 // if _iload, wait to rewrite to iload2. We only want to rewrite the
567 // last two iloads in a pair. Comparing against fast_iload means that
568 // the next bytecode is neither an iload or a caload, and therefore
569 // an iload pair.
570 __ cmpl(rbx, Bytecodes::_iload);
571 __ jcc(Assembler::equal, done);
572
573 __ cmpl(rbx, Bytecodes::_fast_iload);
574 __ movl(bc, Bytecodes::_fast_iload2);
575
576 __ jccb(Assembler::equal, rewrite);
577
578 // if _caload, rewrite to fast_icaload
579 __ cmpl(rbx, Bytecodes::_caload);
580 __ movl(bc, Bytecodes::_fast_icaload);
581 __ jccb(Assembler::equal, rewrite);
582
583 // rewrite so iload doesn't check again.
584 __ movl(bc, Bytecodes::_fast_iload);
585
586 // rewrite
587 // bc: fast bytecode
588 __ bind(rewrite);
589 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
590 __ bind(done);
591 }
592
593 // Get the local value into tos
594 locals_index(rbx);
595 __ movl(rax, iaddress(rbx));
596 }
597
598 void TemplateTable::fast_iload2() {
599 transition(vtos, itos);
600 locals_index(rbx);
601 __ movl(rax, iaddress(rbx));
602 __ push(itos);
603 locals_index(rbx, 3);
604 __ movl(rax, iaddress(rbx));
605 }
606
607 void TemplateTable::fast_iload() {
608 transition(vtos, itos);
609 locals_index(rbx);
610 __ movl(rax, iaddress(rbx));
611 }
612
613 void TemplateTable::lload() {
614 transition(vtos, ltos);
615 locals_index(rbx);
616 __ movptr(rax, laddress(rbx));
617 NOT_LP64(__ movl(rdx, haddress(rbx)));
618 }
619
620 void TemplateTable::fload() {
621 transition(vtos, ftos);
622 locals_index(rbx);
623 __ load_float(faddress(rbx));
624 }
625
626 void TemplateTable::dload() {
627 transition(vtos, dtos);
628 locals_index(rbx);
629 __ load_double(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 __ load_float(faddress(rbx));
662 }
663
664 void TemplateTable::wide_dload() {
665 transition(vtos, dtos);
666 locals_index_wide(rbx);
667 __ load_double(daddress(rbx));
668 }
669
670 void TemplateTable::wide_aload() {
671 transition(vtos, atos);
672 locals_index_wide(rbx);
673 __ movptr(rax, aaddress(rbx));
674 }
675
676 void TemplateTable::index_check(Register array, Register index) {
677 // Pop ptr into array
678 __ pop_ptr(array);
679 index_check_without_pop(array, index);
680 }
681
682 void TemplateTable::index_check_without_pop(Register array, Register index) {
683 // destroys rbx
684 // check array
685 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
686 // sign extend index for use by indexed load
687 __ movl2ptr(index, index);
688 // check index
689 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
690 if (index != rbx) {
691 // ??? convention: move aberrant index into rbx for exception message
692 assert(rbx != array, "different registers");
693 __ movl(rbx, index);
694 }
695 __ jump_cc(Assembler::aboveEqual,
696 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
697 }
698
699
700 void TemplateTable::iaload() {
701 transition(itos, itos);
702 // rax: index
703 // rdx: array
704 index_check(rdx, rax); // kills rbx
705 __ movl(rax, Address(rdx, rax,
706 Address::times_4,
707 arrayOopDesc::base_offset_in_bytes(T_INT)));
708 }
709
710 void TemplateTable::laload() {
711 transition(itos, ltos);
712 // rax: index
713 // rdx: array
714 index_check(rdx, rax); // kills rbx
715 NOT_LP64(__ mov(rbx, rax));
716 // rbx,: index
717 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
718 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
719 }
720
721
722
723 void TemplateTable::faload() {
724 transition(itos, ftos);
725 // rax: index
726 // rdx: array
727 index_check(rdx, rax); // kills rbx
728 __ load_float(Address(rdx, rax,
729 Address::times_4,
730 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
731 }
732
733 void TemplateTable::daload() {
734 transition(itos, dtos);
735 // rax: index
736 // rdx: array
737 index_check(rdx, rax); // kills rbx
738 __ load_double(Address(rdx, rax,
739 Address::times_8,
740 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
741 }
742
743 void TemplateTable::aaload() {
744 transition(itos, atos);
745 // rax: index
746 // rdx: array
747 index_check(rdx, rax); // kills rbx
748 __ load_heap_oop(rax, Address(rdx, rax,
749 UseCompressedOops ? Address::times_4 : Address::times_ptr,
750 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
751 }
752
753 void TemplateTable::baload() {
754 transition(itos, itos);
755 // rax: index
756 // rdx: array
757 index_check(rdx, rax); // kills rbx
758 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
759 }
760
761 void TemplateTable::caload() {
762 transition(itos, itos);
763 // rax: index
764 // rdx: array
765 index_check(rdx, rax); // kills rbx
766 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
767 }
768
769 // iload followed by caload frequent pair
770 void TemplateTable::fast_icaload() {
771 transition(vtos, itos);
772 // load index out of locals
773 locals_index(rbx);
774 __ movl(rax, iaddress(rbx));
775
776 // rax: index
777 // rdx: array
778 index_check(rdx, rax); // kills rbx
779 __ load_unsigned_short(rax,
780 Address(rdx, rax,
781 Address::times_2,
782 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
783 }
784
785
786 void TemplateTable::saload() {
787 transition(itos, itos);
788 // rax: index
789 // rdx: array
790 index_check(rdx, rax); // kills rbx
791 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
792 }
793
794 void TemplateTable::iload(int n) {
795 transition(vtos, itos);
796 __ movl(rax, iaddress(n));
797 }
798
799 void TemplateTable::lload(int n) {
800 transition(vtos, ltos);
801 __ movptr(rax, laddress(n));
802 NOT_LP64(__ movptr(rdx, haddress(n)));
803 }
804
805 void TemplateTable::fload(int n) {
806 transition(vtos, ftos);
807 __ load_float(faddress(n));
808 }
809
810 void TemplateTable::dload(int n) {
811 transition(vtos, dtos);
812 __ load_double(daddress(n));
813 }
814
815 void TemplateTable::aload(int n) {
816 transition(vtos, atos);
817 __ movptr(rax, aaddress(n));
818 }
819
820 void TemplateTable::aload_0() {
821 aload_0_internal();
822 }
823
824 void TemplateTable::nofast_aload_0() {
825 aload_0_internal(may_not_rewrite);
826 }
827
828 void TemplateTable::aload_0_internal(RewriteControl rc) {
829 transition(vtos, atos);
830 // According to bytecode histograms, the pairs:
831 //
832 // _aload_0, _fast_igetfield
833 // _aload_0, _fast_agetfield
834 // _aload_0, _fast_fgetfield
835 //
836 // occur frequently. If RewriteFrequentPairs is set, the (slow)
837 // _aload_0 bytecode checks if the next bytecode is either
838 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
839 // rewrites the current bytecode into a pair bytecode; otherwise it
840 // rewrites the current bytecode into _fast_aload_0 that doesn't do
841 // the pair check anymore.
842 //
843 // Note: If the next bytecode is _getfield, the rewrite must be
844 // delayed, otherwise we may miss an opportunity for a pair.
845 //
846 // Also rewrite frequent pairs
847 // aload_0, aload_1
848 // aload_0, iload_1
849 // These bytecodes with a small amount of code are most profitable
850 // to rewrite
851 if (RewriteFrequentPairs && rc == may_rewrite) {
852 Label rewrite, done;
853
854 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
855 LP64_ONLY(assert(rbx != bc, "register damaged"));
856
857 // get next byte
858 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
859
860 // do actual aload_0
861 aload(0);
862
863 // if _getfield then wait with rewrite
864 __ cmpl(rbx, Bytecodes::_getfield);
865 __ jcc(Assembler::equal, done);
866
867 // if _igetfield then reqrite to _fast_iaccess_0
868 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
869 __ cmpl(rbx, Bytecodes::_fast_igetfield);
870 __ movl(bc, Bytecodes::_fast_iaccess_0);
871 __ jccb(Assembler::equal, rewrite);
872
873 // if _agetfield then reqrite to _fast_aaccess_0
874 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
875 __ cmpl(rbx, Bytecodes::_fast_agetfield);
876 __ movl(bc, Bytecodes::_fast_aaccess_0);
877 __ jccb(Assembler::equal, rewrite);
878
879 // if _fgetfield then reqrite to _fast_faccess_0
880 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
881 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
882 __ movl(bc, Bytecodes::_fast_faccess_0);
883 __ jccb(Assembler::equal, rewrite);
884
885 // else rewrite to _fast_aload0
886 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
887 __ movl(bc, Bytecodes::_fast_aload_0);
888
889 // rewrite
890 // bc: fast bytecode
891 __ bind(rewrite);
892 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
893
894 __ bind(done);
895 } else {
896 aload(0);
897 }
898 }
899
900 void TemplateTable::istore() {
901 transition(itos, vtos);
902 locals_index(rbx);
903 __ movl(iaddress(rbx), rax);
904 }
905
906
907 void TemplateTable::lstore() {
908 transition(ltos, vtos);
909 locals_index(rbx);
910 __ movptr(laddress(rbx), rax);
911 NOT_LP64(__ movptr(haddress(rbx), rdx));
912 }
913
914 void TemplateTable::fstore() {
915 transition(ftos, vtos);
916 locals_index(rbx);
917 __ store_float(faddress(rbx));
918 }
919
920 void TemplateTable::dstore() {
921 transition(dtos, vtos);
922 locals_index(rbx);
923 __ store_double(daddress(rbx));
924 }
925
926 void TemplateTable::astore() {
927 transition(vtos, vtos);
928 __ pop_ptr(rax);
929 locals_index(rbx);
930 __ movptr(aaddress(rbx), rax);
931 }
932
933 void TemplateTable::wide_istore() {
934 transition(vtos, vtos);
935 __ pop_i();
936 locals_index_wide(rbx);
937 __ movl(iaddress(rbx), rax);
938 }
939
940 void TemplateTable::wide_lstore() {
941 transition(vtos, vtos);
942 NOT_LP64(__ pop_l(rax, rdx));
943 LP64_ONLY(__ pop_l());
944 locals_index_wide(rbx);
945 __ movptr(laddress(rbx), rax);
946 NOT_LP64(__ movl(haddress(rbx), rdx));
947 }
948
949 void TemplateTable::wide_fstore() {
950 #ifdef _LP64
951 transition(vtos, vtos);
952 __ pop_f(xmm0);
953 locals_index_wide(rbx);
954 __ movflt(faddress(rbx), xmm0);
955 #else
956 wide_istore();
957 #endif
958 }
959
960 void TemplateTable::wide_dstore() {
961 #ifdef _LP64
962 transition(vtos, vtos);
963 __ pop_d(xmm0);
964 locals_index_wide(rbx);
965 __ movdbl(daddress(rbx), xmm0);
966 #else
967 wide_lstore();
968 #endif
969 }
970
971 void TemplateTable::wide_astore() {
972 transition(vtos, vtos);
973 __ pop_ptr(rax);
974 locals_index_wide(rbx);
975 __ movptr(aaddress(rbx), rax);
976 }
977
978 void TemplateTable::iastore() {
979 transition(itos, vtos);
980 __ pop_i(rbx);
981 // rax: value
982 // rbx: index
983 // rdx: array
984 index_check(rdx, rbx); // prefer index in rbx
985 __ movl(Address(rdx, rbx,
986 Address::times_4,
987 arrayOopDesc::base_offset_in_bytes(T_INT)),
988 rax);
989 }
990
991 void TemplateTable::lastore() {
992 transition(ltos, vtos);
993 __ pop_i(rbx);
994 // rax,: low(value)
995 // rcx: array
996 // rdx: high(value)
997 index_check(rcx, rbx); // prefer index in rbx,
998 // rbx,: index
999 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1000 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1001 }
1002
1003
1004 void TemplateTable::fastore() {
1005 transition(ftos, vtos);
1006 __ pop_i(rbx);
1007 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1008 // rbx: index
1009 // rdx: array
1010 index_check(rdx, rbx); // prefer index in rbx
1011 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1012 }
1013
1014 void TemplateTable::dastore() {
1015 transition(dtos, vtos);
1016 __ pop_i(rbx);
1017 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1018 // rbx: index
1019 // rdx: array
1020 index_check(rdx, rbx); // prefer index in rbx
1021 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1022 }
1023
1024 void TemplateTable::aastore() {
1025 Label is_null, ok_is_subtype, done;
1026 transition(vtos, vtos);
1027 // stack: ..., array, index, value
1028 __ movptr(rax, at_tos()); // value
1029 __ movl(rcx, at_tos_p1()); // index
1030 __ movptr(rdx, at_tos_p2()); // array
1031
1032 Address element_address(rdx, rcx,
1033 UseCompressedOops? Address::times_4 : Address::times_ptr,
1034 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1035
1036 index_check_without_pop(rdx, rcx); // kills rbx
1037 __ testptr(rax, rax);
1038 __ jcc(Assembler::zero, is_null);
1039
1040 // Move subklass into rbx
1041 __ load_klass(rbx, rax);
1042 // Move superklass into rax
1043 __ load_klass(rax, rdx);
1044 __ movptr(rax, Address(rax,
1045 ObjArrayKlass::element_klass_offset()));
1046 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1047 __ lea(rdx, element_address);
1048
1049 // Generate subtype check. Blows rcx, rdi
1050 // Superklass in rax. Subklass in rbx.
1051 __ gen_subtype_check(rbx, ok_is_subtype);
1052
1053 // Come here on failure
1054 // object is at TOS
1055 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1056
1057 // Come here on success
1058 __ bind(ok_is_subtype);
1059
1060 // Get the value we will store
1061 __ movptr(rax, at_tos());
1062 // Now store using the appropriate barrier
1063 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1064 __ jmp(done);
1065
1066 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1067 __ bind(is_null);
1068 __ profile_null_seen(rbx);
1069
1070 // Store a NULL
1071 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1072
1073 // Pop stack arguments
1074 __ bind(done);
1075 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1076 }
1077
1078 void TemplateTable::bastore() {
1079 transition(itos, vtos);
1080 __ pop_i(rbx);
1081 // rax: value
1082 // rbx: index
1083 // rdx: array
1084 index_check(rdx, rbx); // prefer index in rbx
1085 __ movb(Address(rdx, rbx,
1086 Address::times_1,
1087 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1088 rax);
1089 }
1090
1091 void TemplateTable::castore() {
1092 transition(itos, vtos);
1093 __ pop_i(rbx);
1094 // rax: value
1095 // rbx: index
1096 // rdx: array
1097 index_check(rdx, rbx); // prefer index in rbx
1098 __ movw(Address(rdx, rbx,
1099 Address::times_2,
1100 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1101 rax);
1102 }
1103
1104
1105 void TemplateTable::sastore() {
1106 castore();
1107 }
1108
1109 void TemplateTable::istore(int n) {
1110 transition(itos, vtos);
1111 __ movl(iaddress(n), rax);
1112 }
1113
1114 void TemplateTable::lstore(int n) {
1115 transition(ltos, vtos);
1116 __ movptr(laddress(n), rax);
1117 NOT_LP64(__ movptr(haddress(n), rdx));
1118 }
1119
1120 void TemplateTable::fstore(int n) {
1121 transition(ftos, vtos);
1122 __ store_float(faddress(n));
1123 }
1124
1125 void TemplateTable::dstore(int n) {
1126 transition(dtos, vtos);
1127 __ store_double(daddress(n));
1128 }
1129
1130
1131 void TemplateTable::astore(int n) {
1132 transition(vtos, vtos);
1133 __ pop_ptr(rax);
1134 __ movptr(aaddress(n), rax);
1135 }
1136
1137 void TemplateTable::pop() {
1138 transition(vtos, vtos);
1139 __ addptr(rsp, Interpreter::stackElementSize);
1140 }
1141
1142 void TemplateTable::pop2() {
1143 transition(vtos, vtos);
1144 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1145 }
1146
1147
1148 void TemplateTable::dup() {
1149 transition(vtos, vtos);
1150 __ load_ptr(0, rax);
1151 __ push_ptr(rax);
1152 // stack: ..., a, a
1153 }
1154
1155 void TemplateTable::dup_x1() {
1156 transition(vtos, vtos);
1157 // stack: ..., a, b
1158 __ load_ptr( 0, rax); // load b
1159 __ load_ptr( 1, rcx); // load a
1160 __ store_ptr(1, rax); // store b
1161 __ store_ptr(0, rcx); // store a
1162 __ push_ptr(rax); // push b
1163 // stack: ..., b, a, b
1164 }
1165
1166 void TemplateTable::dup_x2() {
1167 transition(vtos, vtos);
1168 // stack: ..., a, b, c
1169 __ load_ptr( 0, rax); // load c
1170 __ load_ptr( 2, rcx); // load a
1171 __ store_ptr(2, rax); // store c in a
1172 __ push_ptr(rax); // push c
1173 // stack: ..., c, b, c, c
1174 __ load_ptr( 2, rax); // load b
1175 __ store_ptr(2, rcx); // store a in b
1176 // stack: ..., c, a, c, c
1177 __ store_ptr(1, rax); // store b in c
1178 // stack: ..., c, a, b, c
1179 }
1180
1181 void TemplateTable::dup2() {
1182 transition(vtos, vtos);
1183 // stack: ..., a, b
1184 __ load_ptr(1, rax); // load a
1185 __ push_ptr(rax); // push a
1186 __ load_ptr(1, rax); // load b
1187 __ push_ptr(rax); // push b
1188 // stack: ..., a, b, a, b
1189 }
1190
1191
1192 void TemplateTable::dup2_x1() {
1193 transition(vtos, vtos);
1194 // stack: ..., a, b, c
1195 __ load_ptr( 0, rcx); // load c
1196 __ load_ptr( 1, rax); // load b
1197 __ push_ptr(rax); // push b
1198 __ push_ptr(rcx); // push c
1199 // stack: ..., a, b, c, b, c
1200 __ store_ptr(3, rcx); // store c in b
1201 // stack: ..., a, c, c, b, c
1202 __ load_ptr( 4, rcx); // load a
1203 __ store_ptr(2, rcx); // store a in 2nd c
1204 // stack: ..., a, c, a, b, c
1205 __ store_ptr(4, rax); // store b in a
1206 // stack: ..., b, c, a, b, c
1207 }
1208
1209 void TemplateTable::dup2_x2() {
1210 transition(vtos, vtos);
1211 // stack: ..., a, b, c, d
1212 __ load_ptr( 0, rcx); // load d
1213 __ load_ptr( 1, rax); // load c
1214 __ push_ptr(rax); // push c
1215 __ push_ptr(rcx); // push d
1216 // stack: ..., a, b, c, d, c, d
1217 __ load_ptr( 4, rax); // load b
1218 __ store_ptr(2, rax); // store b in d
1219 __ store_ptr(4, rcx); // store d in b
1220 // stack: ..., a, d, c, b, c, d
1221 __ load_ptr( 5, rcx); // load a
1222 __ load_ptr( 3, rax); // load c
1223 __ store_ptr(3, rcx); // store a in c
1224 __ store_ptr(5, rax); // store c in a
1225 // stack: ..., c, d, a, b, c, d
1226 }
1227
1228 void TemplateTable::swap() {
1229 transition(vtos, vtos);
1230 // stack: ..., a, b
1231 __ load_ptr( 1, rcx); // load a
1232 __ load_ptr( 0, rax); // load b
1233 __ store_ptr(0, rcx); // store a in b
1234 __ store_ptr(1, rax); // store b in a
1235 // stack: ..., b, a
1236 }
1237
1238 void TemplateTable::iop2(Operation op) {
1239 transition(itos, itos);
1240 switch (op) {
1241 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1242 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1243 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1244 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1245 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1246 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1247 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1248 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1249 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1250 default : ShouldNotReachHere();
1251 }
1252 }
1253
1254 void TemplateTable::lop2(Operation op) {
1255 transition(ltos, ltos);
1256 #ifdef _LP64
1257 switch (op) {
1258 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1259 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1260 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1261 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1262 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1263 default : ShouldNotReachHere();
1264 }
1265 #else
1266 __ pop_l(rbx, rcx);
1267 switch (op) {
1268 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1269 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1270 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1271 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1272 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1273 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1274 default : ShouldNotReachHere();
1275 }
1276 #endif
1277 }
1278
1279 void TemplateTable::idiv() {
1280 transition(itos, itos);
1281 __ movl(rcx, rax);
1282 __ pop_i(rax);
1283 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1284 // they are not equal, one could do a normal division (no correction
1285 // needed), which may speed up this implementation for the common case.
1286 // (see also JVM spec., p.243 & p.271)
1287 __ corrected_idivl(rcx);
1288 }
1289
1290 void TemplateTable::irem() {
1291 transition(itos, itos);
1292 __ movl(rcx, rax);
1293 __ pop_i(rax);
1294 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1295 // they are not equal, one could do a normal division (no correction
1296 // needed), which may speed up this implementation for the common case.
1297 // (see also JVM spec., p.243 & p.271)
1298 __ corrected_idivl(rcx);
1299 __ movl(rax, rdx);
1300 }
1301
1302 void TemplateTable::lmul() {
1303 transition(ltos, ltos);
1304 #ifdef _LP64
1305 __ pop_l(rdx);
1306 __ imulq(rax, rdx);
1307 #else
1308 __ pop_l(rbx, rcx);
1309 __ push(rcx); __ push(rbx);
1310 __ push(rdx); __ push(rax);
1311 __ lmul(2 * wordSize, 0);
1312 __ addptr(rsp, 4 * wordSize); // take off temporaries
1313 #endif
1314 }
1315
1316 void TemplateTable::ldiv() {
1317 transition(ltos, ltos);
1318 #ifdef _LP64
1319 __ mov(rcx, rax);
1320 __ pop_l(rax);
1321 // generate explicit div0 check
1322 __ testq(rcx, rcx);
1323 __ jump_cc(Assembler::zero,
1324 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1325 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1326 // they are not equal, one could do a normal division (no correction
1327 // needed), which may speed up this implementation for the common case.
1328 // (see also JVM spec., p.243 & p.271)
1329 __ corrected_idivq(rcx); // kills rbx
1330 #else
1331 __ pop_l(rbx, rcx);
1332 __ push(rcx); __ push(rbx);
1333 __ push(rdx); __ push(rax);
1334 // check if y = 0
1335 __ orl(rax, rdx);
1336 __ jump_cc(Assembler::zero,
1337 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1338 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1339 __ addptr(rsp, 4 * wordSize); // take off temporaries
1340 #endif
1341 }
1342
1343 void TemplateTable::lrem() {
1344 transition(ltos, ltos);
1345 #ifdef _LP64
1346 __ mov(rcx, rax);
1347 __ pop_l(rax);
1348 __ testq(rcx, rcx);
1349 __ jump_cc(Assembler::zero,
1350 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1351 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1352 // they are not equal, one could do a normal division (no correction
1353 // needed), which may speed up this implementation for the common case.
1354 // (see also JVM spec., p.243 & p.271)
1355 __ corrected_idivq(rcx); // kills rbx
1356 __ mov(rax, rdx);
1357 #else
1358 __ pop_l(rbx, rcx);
1359 __ push(rcx); __ push(rbx);
1360 __ push(rdx); __ push(rax);
1361 // check if y = 0
1362 __ orl(rax, rdx);
1363 __ jump_cc(Assembler::zero,
1364 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1365 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1366 __ addptr(rsp, 4 * wordSize);
1367 #endif
1368 }
1369
1370 void TemplateTable::lshl() {
1371 transition(itos, ltos);
1372 __ movl(rcx, rax); // get shift count
1373 #ifdef _LP64
1374 __ pop_l(rax); // get shift value
1375 __ shlq(rax);
1376 #else
1377 __ pop_l(rax, rdx); // get shift value
1378 __ lshl(rdx, rax);
1379 #endif
1380 }
1381
1382 void TemplateTable::lshr() {
1383 #ifdef _LP64
1384 transition(itos, ltos);
1385 __ movl(rcx, rax); // get shift count
1386 __ pop_l(rax); // get shift value
1387 __ sarq(rax);
1388 #else
1389 transition(itos, ltos);
1390 __ mov(rcx, rax); // get shift count
1391 __ pop_l(rax, rdx); // get shift value
1392 __ lshr(rdx, rax, true);
1393 #endif
1394 }
1395
1396 void TemplateTable::lushr() {
1397 transition(itos, ltos);
1398 #ifdef _LP64
1399 __ movl(rcx, rax); // get shift count
1400 __ pop_l(rax); // get shift value
1401 __ shrq(rax);
1402 #else
1403 __ mov(rcx, rax); // get shift count
1404 __ pop_l(rax, rdx); // get shift value
1405 __ lshr(rdx, rax);
1406 #endif
1407 }
1408
1409 void TemplateTable::fop2(Operation op) {
1410 transition(ftos, ftos);
1411
1412 if (UseSSE >= 1) {
1413 switch (op) {
1414 case add:
1415 __ addss(xmm0, at_rsp());
1416 __ addptr(rsp, Interpreter::stackElementSize);
1417 break;
1418 case sub:
1419 __ movflt(xmm1, xmm0);
1420 __ pop_f(xmm0);
1421 __ subss(xmm0, xmm1);
1422 break;
1423 case mul:
1424 __ mulss(xmm0, at_rsp());
1425 __ addptr(rsp, Interpreter::stackElementSize);
1426 break;
1427 case div:
1428 __ movflt(xmm1, xmm0);
1429 __ pop_f(xmm0);
1430 __ divss(xmm0, xmm1);
1431 break;
1432 case rem:
1433 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1434 // modulo operation. The frem method calls the function
1435 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1436 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1437 // (signalling or quiet) is returned.
1438 //
1439 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1440 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1441 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1442 // The fprem instruction used on x86_32 is functionally equivalent to
1443 // SharedRuntime::frem in that it returns a NaN.
1444 #ifdef _LP64
1445 __ movflt(xmm1, xmm0);
1446 __ pop_f(xmm0);
1447 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1448 #else
1449 __ push_f(xmm0);
1450 __ pop_f();
1451 __ fld_s(at_rsp());
1452 __ fremr(rax);
1453 __ f2ieee();
1454 __ pop(rax); // pop second operand off the stack
1455 __ push_f();
1456 __ pop_f(xmm0);
1457 #endif
1458 break;
1459 default:
1460 ShouldNotReachHere();
1461 break;
1462 }
1463 } else {
1464 #ifdef _LP64
1465 ShouldNotReachHere();
1466 #else
1467 switch (op) {
1468 case add: __ fadd_s (at_rsp()); break;
1469 case sub: __ fsubr_s(at_rsp()); break;
1470 case mul: __ fmul_s (at_rsp()); break;
1471 case div: __ fdivr_s(at_rsp()); break;
1472 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1473 default : ShouldNotReachHere();
1474 }
1475 __ f2ieee();
1476 __ pop(rax); // pop second operand off the stack
1477 #endif // _LP64
1478 }
1479 }
1480
1481 void TemplateTable::dop2(Operation op) {
1482 transition(dtos, dtos);
1483 if (UseSSE >= 2) {
1484 switch (op) {
1485 case add:
1486 __ addsd(xmm0, at_rsp());
1487 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1488 break;
1489 case sub:
1490 __ movdbl(xmm1, xmm0);
1491 __ pop_d(xmm0);
1492 __ subsd(xmm0, xmm1);
1493 break;
1494 case mul:
1495 __ mulsd(xmm0, at_rsp());
1496 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1497 break;
1498 case div:
1499 __ movdbl(xmm1, xmm0);
1500 __ pop_d(xmm0);
1501 __ divsd(xmm0, xmm1);
1502 break;
1503 case rem:
1504 // Similar to fop2(), the modulo operation is performed using the
1505 // SharedRuntime::drem method (on x86_64 platforms) or using the
1506 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1507 #ifdef _LP64
1508 __ movdbl(xmm1, xmm0);
1509 __ pop_d(xmm0);
1510 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1511 #else
1512 __ push_d(xmm0);
1513 __ pop_d();
1514 __ fld_d(at_rsp());
1515 __ fremr(rax);
1516 __ d2ieee();
1517 __ pop(rax);
1518 __ pop(rdx);
1519 __ push_d();
1520 __ pop_d(xmm0);
1521 #endif
1522 break;
1523 default:
1524 ShouldNotReachHere();
1525 break;
1526 }
1527 } else {
1528 #ifdef _LP64
1529 ShouldNotReachHere();
1530 #else
1531 switch (op) {
1532 case add: __ fadd_d (at_rsp()); break;
1533 case sub: __ fsubr_d(at_rsp()); break;
1534 case mul: {
1535 Label L_strict;
1536 Label L_join;
1537 const Address access_flags (rcx, Method::access_flags_offset());
1538 __ get_method(rcx);
1539 __ movl(rcx, access_flags);
1540 __ testl(rcx, JVM_ACC_STRICT);
1541 __ jccb(Assembler::notZero, L_strict);
1542 __ fmul_d (at_rsp());
1543 __ jmpb(L_join);
1544 __ bind(L_strict);
1545 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1546 __ fmulp();
1547 __ fmul_d (at_rsp());
1548 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1549 __ fmulp();
1550 __ bind(L_join);
1551 break;
1552 }
1553 case div: {
1554 Label L_strict;
1555 Label L_join;
1556 const Address access_flags (rcx, Method::access_flags_offset());
1557 __ get_method(rcx);
1558 __ movl(rcx, access_flags);
1559 __ testl(rcx, JVM_ACC_STRICT);
1560 __ jccb(Assembler::notZero, L_strict);
1561 __ fdivr_d(at_rsp());
1562 __ jmp(L_join);
1563 __ bind(L_strict);
1564 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1565 __ fmul_d (at_rsp());
1566 __ fdivrp();
1567 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1568 __ fmulp();
1569 __ bind(L_join);
1570 break;
1571 }
1572 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1573 default : ShouldNotReachHere();
1574 }
1575 __ d2ieee();
1576 // Pop double precision number from rsp.
1577 __ pop(rax);
1578 __ pop(rdx);
1579 #endif
1580 }
1581 }
1582
1583 void TemplateTable::ineg() {
1584 transition(itos, itos);
1585 __ negl(rax);
1586 }
1587
1588 void TemplateTable::lneg() {
1589 transition(ltos, ltos);
1590 LP64_ONLY(__ negq(rax));
1591 NOT_LP64(__ lneg(rdx, rax));
1592 }
1593
1594 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1595 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1596 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1597 // of 128-bits operands for SSE instructions.
1598 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1599 // Store the value to a 128-bits operand.
1600 operand[0] = lo;
1601 operand[1] = hi;
1602 return operand;
1603 }
1604
1605 // Buffer for 128-bits masks used by SSE instructions.
1606 static jlong float_signflip_pool[2*2];
1607 static jlong double_signflip_pool[2*2];
1608
1609 void TemplateTable::fneg() {
1610 transition(ftos, ftos);
1611 if (UseSSE >= 1) {
1612 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1613 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1614 } else {
1615 LP64_ONLY(ShouldNotReachHere());
1616 NOT_LP64(__ fchs());
1617 }
1618 }
1619
1620 void TemplateTable::dneg() {
1621 transition(dtos, dtos);
1622 if (UseSSE >= 2) {
1623 static jlong *double_signflip =
1624 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1625 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1626 } else {
1627 #ifdef _LP64
1628 ShouldNotReachHere();
1629 #else
1630 __ fchs();
1631 #endif
1632 }
1633 }
1634
1635 void TemplateTable::iinc() {
1636 transition(vtos, vtos);
1637 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1638 locals_index(rbx);
1639 __ addl(iaddress(rbx), rdx);
1640 }
1641
1642 void TemplateTable::wide_iinc() {
1643 transition(vtos, vtos);
1644 __ movl(rdx, at_bcp(4)); // get constant
1645 locals_index_wide(rbx);
1646 __ bswapl(rdx); // swap bytes & sign-extend constant
1647 __ sarl(rdx, 16);
1648 __ addl(iaddress(rbx), rdx);
1649 // Note: should probably use only one movl to get both
1650 // the index and the constant -> fix this
1651 }
1652
1653 void TemplateTable::convert() {
1654 #ifdef _LP64
1655 // Checking
1656 #ifdef ASSERT
1657 {
1658 TosState tos_in = ilgl;
1659 TosState tos_out = ilgl;
1660 switch (bytecode()) {
1661 case Bytecodes::_i2l: // fall through
1662 case Bytecodes::_i2f: // fall through
1663 case Bytecodes::_i2d: // fall through
1664 case Bytecodes::_i2b: // fall through
1665 case Bytecodes::_i2c: // fall through
1666 case Bytecodes::_i2s: tos_in = itos; break;
1667 case Bytecodes::_l2i: // fall through
1668 case Bytecodes::_l2f: // fall through
1669 case Bytecodes::_l2d: tos_in = ltos; break;
1670 case Bytecodes::_f2i: // fall through
1671 case Bytecodes::_f2l: // fall through
1672 case Bytecodes::_f2d: tos_in = ftos; break;
1673 case Bytecodes::_d2i: // fall through
1674 case Bytecodes::_d2l: // fall through
1675 case Bytecodes::_d2f: tos_in = dtos; break;
1676 default : ShouldNotReachHere();
1677 }
1678 switch (bytecode()) {
1679 case Bytecodes::_l2i: // fall through
1680 case Bytecodes::_f2i: // fall through
1681 case Bytecodes::_d2i: // fall through
1682 case Bytecodes::_i2b: // fall through
1683 case Bytecodes::_i2c: // fall through
1684 case Bytecodes::_i2s: tos_out = itos; break;
1685 case Bytecodes::_i2l: // fall through
1686 case Bytecodes::_f2l: // fall through
1687 case Bytecodes::_d2l: tos_out = ltos; break;
1688 case Bytecodes::_i2f: // fall through
1689 case Bytecodes::_l2f: // fall through
1690 case Bytecodes::_d2f: tos_out = ftos; break;
1691 case Bytecodes::_i2d: // fall through
1692 case Bytecodes::_l2d: // fall through
1693 case Bytecodes::_f2d: tos_out = dtos; break;
1694 default : ShouldNotReachHere();
1695 }
1696 transition(tos_in, tos_out);
1697 }
1698 #endif // ASSERT
1699
1700 static const int64_t is_nan = 0x8000000000000000L;
1701
1702 // Conversion
1703 switch (bytecode()) {
1704 case Bytecodes::_i2l:
1705 __ movslq(rax, rax);
1706 break;
1707 case Bytecodes::_i2f:
1708 __ cvtsi2ssl(xmm0, rax);
1709 break;
1710 case Bytecodes::_i2d:
1711 __ cvtsi2sdl(xmm0, rax);
1712 break;
1713 case Bytecodes::_i2b:
1714 __ movsbl(rax, rax);
1715 break;
1716 case Bytecodes::_i2c:
1717 __ movzwl(rax, rax);
1718 break;
1719 case Bytecodes::_i2s:
1720 __ movswl(rax, rax);
1721 break;
1722 case Bytecodes::_l2i:
1723 __ movl(rax, rax);
1724 break;
1725 case Bytecodes::_l2f:
1726 __ cvtsi2ssq(xmm0, rax);
1727 break;
1728 case Bytecodes::_l2d:
1729 __ cvtsi2sdq(xmm0, rax);
1730 break;
1731 case Bytecodes::_f2i:
1732 {
1733 Label L;
1734 __ cvttss2sil(rax, xmm0);
1735 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1736 __ jcc(Assembler::notEqual, L);
1737 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1738 __ bind(L);
1739 }
1740 break;
1741 case Bytecodes::_f2l:
1742 {
1743 Label L;
1744 __ cvttss2siq(rax, xmm0);
1745 // NaN or overflow/underflow?
1746 __ cmp64(rax, ExternalAddress((address) &is_nan));
1747 __ jcc(Assembler::notEqual, L);
1748 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1749 __ bind(L);
1750 }
1751 break;
1752 case Bytecodes::_f2d:
1753 __ cvtss2sd(xmm0, xmm0);
1754 break;
1755 case Bytecodes::_d2i:
1756 {
1757 Label L;
1758 __ cvttsd2sil(rax, xmm0);
1759 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1760 __ jcc(Assembler::notEqual, L);
1761 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1762 __ bind(L);
1763 }
1764 break;
1765 case Bytecodes::_d2l:
1766 {
1767 Label L;
1768 __ cvttsd2siq(rax, xmm0);
1769 // NaN or overflow/underflow?
1770 __ cmp64(rax, ExternalAddress((address) &is_nan));
1771 __ jcc(Assembler::notEqual, L);
1772 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1773 __ bind(L);
1774 }
1775 break;
1776 case Bytecodes::_d2f:
1777 __ cvtsd2ss(xmm0, xmm0);
1778 break;
1779 default:
1780 ShouldNotReachHere();
1781 }
1782 #else
1783 // Checking
1784 #ifdef ASSERT
1785 { TosState tos_in = ilgl;
1786 TosState tos_out = ilgl;
1787 switch (bytecode()) {
1788 case Bytecodes::_i2l: // fall through
1789 case Bytecodes::_i2f: // fall through
1790 case Bytecodes::_i2d: // fall through
1791 case Bytecodes::_i2b: // fall through
1792 case Bytecodes::_i2c: // fall through
1793 case Bytecodes::_i2s: tos_in = itos; break;
1794 case Bytecodes::_l2i: // fall through
1795 case Bytecodes::_l2f: // fall through
1796 case Bytecodes::_l2d: tos_in = ltos; break;
1797 case Bytecodes::_f2i: // fall through
1798 case Bytecodes::_f2l: // fall through
1799 case Bytecodes::_f2d: tos_in = ftos; break;
1800 case Bytecodes::_d2i: // fall through
1801 case Bytecodes::_d2l: // fall through
1802 case Bytecodes::_d2f: tos_in = dtos; break;
1803 default : ShouldNotReachHere();
1804 }
1805 switch (bytecode()) {
1806 case Bytecodes::_l2i: // fall through
1807 case Bytecodes::_f2i: // fall through
1808 case Bytecodes::_d2i: // fall through
1809 case Bytecodes::_i2b: // fall through
1810 case Bytecodes::_i2c: // fall through
1811 case Bytecodes::_i2s: tos_out = itos; break;
1812 case Bytecodes::_i2l: // fall through
1813 case Bytecodes::_f2l: // fall through
1814 case Bytecodes::_d2l: tos_out = ltos; break;
1815 case Bytecodes::_i2f: // fall through
1816 case Bytecodes::_l2f: // fall through
1817 case Bytecodes::_d2f: tos_out = ftos; break;
1818 case Bytecodes::_i2d: // fall through
1819 case Bytecodes::_l2d: // fall through
1820 case Bytecodes::_f2d: tos_out = dtos; break;
1821 default : ShouldNotReachHere();
1822 }
1823 transition(tos_in, tos_out);
1824 }
1825 #endif // ASSERT
1826
1827 // Conversion
1828 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1829 switch (bytecode()) {
1830 case Bytecodes::_i2l:
1831 __ extend_sign(rdx, rax);
1832 break;
1833 case Bytecodes::_i2f:
1834 if (UseSSE >= 1) {
1835 __ cvtsi2ssl(xmm0, rax);
1836 } else {
1837 __ push(rax); // store int on tos
1838 __ fild_s(at_rsp()); // load int to ST0
1839 __ f2ieee(); // truncate to float size
1840 __ pop(rcx); // adjust rsp
1841 }
1842 break;
1843 case Bytecodes::_i2d:
1844 if (UseSSE >= 2) {
1845 __ cvtsi2sdl(xmm0, rax);
1846 } else {
1847 __ push(rax); // add one slot for d2ieee()
1848 __ push(rax); // store int on tos
1849 __ fild_s(at_rsp()); // load int to ST0
1850 __ d2ieee(); // truncate to double size
1851 __ pop(rcx); // adjust rsp
1852 __ pop(rcx);
1853 }
1854 break;
1855 case Bytecodes::_i2b:
1856 __ shll(rax, 24); // truncate upper 24 bits
1857 __ sarl(rax, 24); // and sign-extend byte
1858 LP64_ONLY(__ movsbl(rax, rax));
1859 break;
1860 case Bytecodes::_i2c:
1861 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1862 LP64_ONLY(__ movzwl(rax, rax));
1863 break;
1864 case Bytecodes::_i2s:
1865 __ shll(rax, 16); // truncate upper 16 bits
1866 __ sarl(rax, 16); // and sign-extend short
1867 LP64_ONLY(__ movswl(rax, rax));
1868 break;
1869 case Bytecodes::_l2i:
1870 /* nothing to do */
1871 break;
1872 case Bytecodes::_l2f:
1873 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1874 // 64-bit long values to floats. On 32-bit platforms it is not possible
1875 // to use that instruction with 64-bit operands, therefore the FPU is
1876 // used to perform the conversion.
1877 __ push(rdx); // store long on tos
1878 __ push(rax);
1879 __ fild_d(at_rsp()); // load long to ST0
1880 __ f2ieee(); // truncate to float size
1881 __ pop(rcx); // adjust rsp
1882 __ pop(rcx);
1883 if (UseSSE >= 1) {
1884 __ push_f();
1885 __ pop_f(xmm0);
1886 }
1887 break;
1888 case Bytecodes::_l2d:
1889 // On 32-bit platforms the FPU is used for conversion because on
1890 // 32-bit platforms it is not not possible to use the cvtsi2sdq
1891 // instruction with 64-bit operands.
1892 __ push(rdx); // store long on tos
1893 __ push(rax);
1894 __ fild_d(at_rsp()); // load long to ST0
1895 __ d2ieee(); // truncate to double size
1896 __ pop(rcx); // adjust rsp
1897 __ pop(rcx);
1898 if (UseSSE >= 2) {
1899 __ push_d();
1900 __ pop_d(xmm0);
1901 }
1902 break;
1903 case Bytecodes::_f2i:
1904 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1905 // as it returns 0 for any NaN.
1906 if (UseSSE >= 1) {
1907 __ push_f(xmm0);
1908 } else {
1909 __ push(rcx); // reserve space for argument
1910 __ fstp_s(at_rsp()); // pass float argument on stack
1911 }
1912 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1913 break;
1914 case Bytecodes::_f2l:
1915 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
1916 // as it returns 0 for any NaN.
1917 if (UseSSE >= 1) {
1918 __ push_f(xmm0);
1919 } else {
1920 __ push(rcx); // reserve space for argument
1921 __ fstp_s(at_rsp()); // pass float argument on stack
1922 }
1923 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1924 break;
1925 case Bytecodes::_f2d:
1926 if (UseSSE < 1) {
1927 /* nothing to do */
1928 } else if (UseSSE == 1) {
1929 __ push_f(xmm0);
1930 __ pop_f();
1931 } else { // UseSSE >= 2
1932 __ cvtss2sd(xmm0, xmm0);
1933 }
1934 break;
1935 case Bytecodes::_d2i:
1936 if (UseSSE >= 2) {
1937 __ push_d(xmm0);
1938 } else {
1939 __ push(rcx); // reserve space for argument
1940 __ push(rcx);
1941 __ fstp_d(at_rsp()); // pass double argument on stack
1942 }
1943 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1944 break;
1945 case Bytecodes::_d2l:
1946 if (UseSSE >= 2) {
1947 __ push_d(xmm0);
1948 } else {
1949 __ push(rcx); // reserve space for argument
1950 __ push(rcx);
1951 __ fstp_d(at_rsp()); // pass double argument on stack
1952 }
1953 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1954 break;
1955 case Bytecodes::_d2f:
1956 if (UseSSE <= 1) {
1957 __ push(rcx); // reserve space for f2ieee()
1958 __ f2ieee(); // truncate to float size
1959 __ pop(rcx); // adjust rsp
1960 if (UseSSE == 1) {
1961 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
1962 // the conversion is performed using the FPU in this case.
1963 __ push_f();
1964 __ pop_f(xmm0);
1965 }
1966 } else { // UseSSE >= 2
1967 __ cvtsd2ss(xmm0, xmm0);
1968 }
1969 break;
1970 default :
1971 ShouldNotReachHere();
1972 }
1973 #endif
1974 }
1975
1976 void TemplateTable::lcmp() {
1977 transition(ltos, itos);
1978 #ifdef _LP64
1979 Label done;
1980 __ pop_l(rdx);
1981 __ cmpq(rdx, rax);
1982 __ movl(rax, -1);
1983 __ jccb(Assembler::less, done);
1984 __ setb(Assembler::notEqual, rax);
1985 __ movzbl(rax, rax);
1986 __ bind(done);
1987 #else
1988
1989 // y = rdx:rax
1990 __ pop_l(rbx, rcx); // get x = rcx:rbx
1991 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1992 __ mov(rax, rcx);
1993 #endif
1994 }
1995
1996 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1997 if ((is_float && UseSSE >= 1) ||
1998 (!is_float && UseSSE >= 2)) {
1999 Label done;
2000 if (is_float) {
2001 // XXX get rid of pop here, use ... reg, mem32
2002 __ pop_f(xmm1);
2003 __ ucomiss(xmm1, xmm0);
2004 } else {
2005 // XXX get rid of pop here, use ... reg, mem64
2006 __ pop_d(xmm1);
2007 __ ucomisd(xmm1, xmm0);
2008 }
2009 if (unordered_result < 0) {
2010 __ movl(rax, -1);
2011 __ jccb(Assembler::parity, done);
2012 __ jccb(Assembler::below, done);
2013 __ setb(Assembler::notEqual, rdx);
2014 __ movzbl(rax, rdx);
2015 } else {
2016 __ movl(rax, 1);
2017 __ jccb(Assembler::parity, done);
2018 __ jccb(Assembler::above, done);
2019 __ movl(rax, 0);
2020 __ jccb(Assembler::equal, done);
2021 __ decrementl(rax);
2022 }
2023 __ bind(done);
2024 } else {
2025 #ifdef _LP64
2026 ShouldNotReachHere();
2027 #else
2028 if (is_float) {
2029 __ fld_s(at_rsp());
2030 } else {
2031 __ fld_d(at_rsp());
2032 __ pop(rdx);
2033 }
2034 __ pop(rcx);
2035 __ fcmp2int(rax, unordered_result < 0);
2036 #endif // _LP64
2037 }
2038 }
2039
2040 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2041 __ get_method(rcx); // rcx holds method
2042 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2043 // holds bumped taken count
2044
2045 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2046 InvocationCounter::counter_offset();
2047 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2048 InvocationCounter::counter_offset();
2049
2050 // Load up edx with the branch displacement
2051 if (is_wide) {
2052 __ movl(rdx, at_bcp(1));
2053 } else {
2054 __ load_signed_short(rdx, at_bcp(1));
2055 }
2056 __ bswapl(rdx);
2057
2058 if (!is_wide) {
2059 __ sarl(rdx, 16);
2060 }
2061 LP64_ONLY(__ movl2ptr(rdx, rdx));
2062
2063 // Handle all the JSR stuff here, then exit.
2064 // It's much shorter and cleaner than intermingling with the non-JSR
2065 // normal-branch stuff occurring below.
2066 if (is_jsr) {
2067 // Pre-load the next target bytecode into rbx
2068 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2069
2070 // compute return address as bci in rax
2071 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2072 in_bytes(ConstMethod::codes_offset())));
2073 __ subptr(rax, Address(rcx, Method::const_offset()));
2074 // Adjust the bcp in r13 by the displacement in rdx
2075 __ addptr(rbcp, rdx);
2076 // jsr returns atos that is not an oop
2077 __ push_i(rax);
2078 __ dispatch_only(vtos);
2079 return;
2080 }
2081
2082 // Normal (non-jsr) branch handling
2083
2084 // Adjust the bcp in r13 by the displacement in rdx
2085 __ addptr(rbcp, rdx);
2086
2087 assert(UseLoopCounter || !UseOnStackReplacement,
2088 "on-stack-replacement requires loop counters");
2089 Label backedge_counter_overflow;
2090 Label profile_method;
2091 Label dispatch;
2092 if (UseLoopCounter) {
2093 // increment backedge counter for backward branches
2094 // rax: MDO
2095 // rbx: MDO bumped taken-count
2096 // rcx: method
2097 // rdx: target offset
2098 // r13: target bcp
2099 // r14: locals pointer
2100 __ testl(rdx, rdx); // check if forward or backward branch
2101 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2102
2103 // check if MethodCounters exists
2104 Label has_counters;
2105 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2106 __ testptr(rax, rax);
2107 __ jcc(Assembler::notZero, has_counters);
2108 __ push(rdx);
2109 __ push(rcx);
2110 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2111 rcx);
2112 __ pop(rcx);
2113 __ pop(rdx);
2114 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2115 __ testptr(rax, rax);
2116 __ jcc(Assembler::zero, dispatch);
2117 __ bind(has_counters);
2118
2119 if (TieredCompilation) {
2120 Label no_mdo;
2121 int increment = InvocationCounter::count_increment;
2122 if (ProfileInterpreter) {
2123 // Are we profiling?
2124 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2125 __ testptr(rbx, rbx);
2126 __ jccb(Assembler::zero, no_mdo);
2127 // Increment the MDO backedge counter
2128 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2129 in_bytes(InvocationCounter::counter_offset()));
2130 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2131 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2132 rax, false, Assembler::zero, &backedge_counter_overflow);
2133 __ jmp(dispatch);
2134 }
2135 __ bind(no_mdo);
2136 // Increment backedge counter in MethodCounters*
2137 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2138 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2139 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2140 rax, false, Assembler::zero, &backedge_counter_overflow);
2141 } else { // not TieredCompilation
2142 // increment counter
2143 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2144 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2145 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2146 __ movl(Address(rcx, be_offset), rax); // store counter
2147
2148 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2149
2150 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2151 __ addl(rax, Address(rcx, be_offset)); // add both counters
2152
2153 if (ProfileInterpreter) {
2154 // Test to see if we should create a method data oop
2155 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2156 __ jcc(Assembler::less, dispatch);
2157
2158 // if no method data exists, go to profile method
2159 __ test_method_data_pointer(rax, profile_method);
2160
2161 if (UseOnStackReplacement) {
2162 // check for overflow against rbx which is the MDO taken count
2163 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2164 __ jcc(Assembler::below, dispatch);
2165
2166 // When ProfileInterpreter is on, the backedge_count comes
2167 // from the MethodData*, which value does not get reset on
2168 // the call to frequency_counter_overflow(). To avoid
2169 // excessive calls to the overflow routine while the method is
2170 // being compiled, add a second test to make sure the overflow
2171 // function is called only once every overflow_frequency.
2172 const int overflow_frequency = 1024;
2173 __ andl(rbx, overflow_frequency - 1);
2174 __ jcc(Assembler::zero, backedge_counter_overflow);
2175
2176 }
2177 } else {
2178 if (UseOnStackReplacement) {
2179 // check for overflow against rax, which is the sum of the
2180 // counters
2181 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2182 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2183
2184 }
2185 }
2186 }
2187 __ bind(dispatch);
2188 }
2189
2190 // Pre-load the next target bytecode into rbx
2191 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2192
2193 // continue with the bytecode @ target
2194 // rax: return bci for jsr's, unused otherwise
2195 // rbx: target bytecode
2196 // r13: target bcp
2197 __ dispatch_only(vtos);
2198
2199 if (UseLoopCounter) {
2200 if (ProfileInterpreter) {
2201 // Out-of-line code to allocate method data oop.
2202 __ bind(profile_method);
2203 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2204 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2205 __ set_method_data_pointer_for_bcp();
2206 __ jmp(dispatch);
2207 }
2208
2209 if (UseOnStackReplacement) {
2210 // invocation counter overflow
2211 __ bind(backedge_counter_overflow);
2212 __ negptr(rdx);
2213 __ addptr(rdx, rbcp); // branch bcp
2214 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2215 __ call_VM(noreg,
2216 CAST_FROM_FN_PTR(address,
2217 InterpreterRuntime::frequency_counter_overflow),
2218 rdx);
2219 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2220
2221 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2222 // rbx: target bytecode
2223 // rdx: scratch
2224 // r14: locals pointer
2225 // r13: bcp
2226 __ testptr(rax, rax); // test result
2227 __ jcc(Assembler::zero, dispatch); // no osr if null
2228 // nmethod may have been invalidated (VM may block upon call_VM return)
2229 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2230 __ jcc(Assembler::notEqual, dispatch);
2231
2232 // We have the address of an on stack replacement routine in rax
2233 // We need to prepare to execute the OSR method. First we must
2234 // migrate the locals and monitors off of the stack.
2235
2236 LP64_ONLY(__ mov(r13, rax)); // save the nmethod
2237 NOT_LP64(__ mov(rbx, rax)); // save the nmethod
2238 NOT_LP64(__ get_thread(rcx));
2239
2240 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2241
2242 // rax is OSR buffer, move it to expected parameter location
2243 LP64_ONLY(__ mov(j_rarg0, rax));
2244 NOT_LP64(__ mov(rcx, rax));
2245 // We use j_rarg definitions here so that registers don't conflict as parameter
2246 // registers change across platforms as we are in the midst of a calling
2247 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2248
2249 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2250 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2251
2252
2253 // pop the interpreter frame
2254 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2255 __ leave(); // remove frame anchor
2256 __ pop(retaddr); // get return address
2257 __ mov(rsp, sender_sp); // set sp to sender sp
2258 // Ensure compiled code always sees stack at proper alignment
2259 __ andptr(rsp, -(StackAlignmentInBytes));
2260
2261 // unlike x86 we need no specialized return from compiled code
2262 // to the interpreter or the call stub.
2263
2264 // push the return address
2265 __ push(retaddr);
2266
2267 // and begin the OSR nmethod
2268 LP64_ONLY(__ jmp(Address(r13, nmethod::osr_entry_point_offset())));
2269 NOT_LP64(__ jmp(Address(rbx, nmethod::osr_entry_point_offset())));
2270 }
2271 }
2272 }
2273
2274 void TemplateTable::if_0cmp(Condition cc) {
2275 transition(itos, vtos);
2276 // assume branch is more often taken than not (loops use backward branches)
2277 Label not_taken;
2278 __ testl(rax, rax);
2279 __ jcc(j_not(cc), not_taken);
2280 branch(false, false);
2281 __ bind(not_taken);
2282 __ profile_not_taken_branch(rax);
2283 }
2284
2285 void TemplateTable::if_icmp(Condition cc) {
2286 transition(itos, vtos);
2287 // assume branch is more often taken than not (loops use backward branches)
2288 Label not_taken;
2289 __ pop_i(rdx);
2290 __ cmpl(rdx, rax);
2291 __ jcc(j_not(cc), not_taken);
2292 branch(false, false);
2293 __ bind(not_taken);
2294 __ profile_not_taken_branch(rax);
2295 }
2296
2297 void TemplateTable::if_nullcmp(Condition cc) {
2298 transition(atos, vtos);
2299 // assume branch is more often taken than not (loops use backward branches)
2300 Label not_taken;
2301 __ testptr(rax, rax);
2302 __ jcc(j_not(cc), not_taken);
2303 branch(false, false);
2304 __ bind(not_taken);
2305 __ profile_not_taken_branch(rax);
2306 }
2307
2308 void TemplateTable::if_acmp(Condition cc) {
2309 transition(atos, vtos);
2310 // assume branch is more often taken than not (loops use backward branches)
2311 Label not_taken;
2312 __ pop_ptr(rdx);
2313 __ cmpptr(rdx, rax);
2314 __ jcc(j_not(cc), not_taken);
2315 branch(false, false);
2316 __ bind(not_taken);
2317 __ profile_not_taken_branch(rax);
2318 }
2319
2320 void TemplateTable::ret() {
2321 transition(vtos, vtos);
2322 locals_index(rbx);
2323 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2324 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2325 __ profile_ret(rbx, rcx);
2326 __ get_method(rax);
2327 __ movptr(rbcp, Address(rax, Method::const_offset()));
2328 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2329 ConstMethod::codes_offset()));
2330 __ dispatch_next(vtos);
2331 }
2332
2333 void TemplateTable::wide_ret() {
2334 transition(vtos, vtos);
2335 locals_index_wide(rbx);
2336 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2337 __ profile_ret(rbx, rcx);
2338 __ get_method(rax);
2339 __ movptr(rbcp, Address(rax, Method::const_offset()));
2340 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2341 __ dispatch_next(vtos);
2342 }
2343
2344 void TemplateTable::tableswitch() {
2345 Label default_case, continue_execution;
2346 transition(itos, vtos);
2347
2348 // align r13/rsi
2349 __ lea(rbx, at_bcp(BytesPerInt));
2350 __ andptr(rbx, -BytesPerInt);
2351 // load lo & hi
2352 __ movl(rcx, Address(rbx, BytesPerInt));
2353 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2354 __ bswapl(rcx);
2355 __ bswapl(rdx);
2356 // check against lo & hi
2357 __ cmpl(rax, rcx);
2358 __ jcc(Assembler::less, default_case);
2359 __ cmpl(rax, rdx);
2360 __ jcc(Assembler::greater, default_case);
2361 // lookup dispatch offset
2362 __ subl(rax, rcx);
2363 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2364 __ profile_switch_case(rax, rbx, rcx);
2365 // continue execution
2366 __ bind(continue_execution);
2367 __ bswapl(rdx);
2368 LP64_ONLY(__ movl2ptr(rdx, rdx));
2369 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2370 __ addptr(rbcp, rdx);
2371 __ dispatch_only(vtos);
2372 // handle default
2373 __ bind(default_case);
2374 __ profile_switch_default(rax);
2375 __ movl(rdx, Address(rbx, 0));
2376 __ jmp(continue_execution);
2377 }
2378
2379 void TemplateTable::lookupswitch() {
2380 transition(itos, itos);
2381 __ stop("lookupswitch bytecode should have been rewritten");
2382 }
2383
2384 void TemplateTable::fast_linearswitch() {
2385 transition(itos, vtos);
2386 Label loop_entry, loop, found, continue_execution;
2387 // bswap rax so we can avoid bswapping the table entries
2388 __ bswapl(rax);
2389 // align r13
2390 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2391 // this instruction (change offsets
2392 // below)
2393 __ andptr(rbx, -BytesPerInt);
2394 // set counter
2395 __ movl(rcx, Address(rbx, BytesPerInt));
2396 __ bswapl(rcx);
2397 __ jmpb(loop_entry);
2398 // table search
2399 __ bind(loop);
2400 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2401 __ jcc(Assembler::equal, found);
2402 __ bind(loop_entry);
2403 __ decrementl(rcx);
2404 __ jcc(Assembler::greaterEqual, loop);
2405 // default case
2406 __ profile_switch_default(rax);
2407 __ movl(rdx, Address(rbx, 0));
2408 __ jmp(continue_execution);
2409 // entry found -> get offset
2410 __ bind(found);
2411 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2412 __ profile_switch_case(rcx, rax, rbx);
2413 // continue execution
2414 __ bind(continue_execution);
2415 __ bswapl(rdx);
2416 __ movl2ptr(rdx, rdx);
2417 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2418 __ addptr(rbcp, rdx);
2419 __ dispatch_only(vtos);
2420 }
2421
2422 void TemplateTable::fast_binaryswitch() {
2423 transition(itos, vtos);
2424 // Implementation using the following core algorithm:
2425 //
2426 // int binary_search(int key, LookupswitchPair* array, int n) {
2427 // // Binary search according to "Methodik des Programmierens" by
2428 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2429 // int i = 0;
2430 // int j = n;
2431 // while (i+1 < j) {
2432 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2433 // // with Q: for all i: 0 <= i < n: key < a[i]
2434 // // where a stands for the array and assuming that the (inexisting)
2435 // // element a[n] is infinitely big.
2436 // int h = (i + j) >> 1;
2437 // // i < h < j
2438 // if (key < array[h].fast_match()) {
2439 // j = h;
2440 // } else {
2441 // i = h;
2442 // }
2443 // }
2444 // // R: a[i] <= key < a[i+1] or Q
2445 // // (i.e., if key is within array, i is the correct index)
2446 // return i;
2447 // }
2448
2449 // Register allocation
2450 const Register key = rax; // already set (tosca)
2451 const Register array = rbx;
2452 const Register i = rcx;
2453 const Register j = rdx;
2454 const Register h = rdi;
2455 const Register temp = rsi;
2456
2457 // Find array start
2458 NOT_LP64(__ save_bcp());
2459
2460 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2461 // get rid of this
2462 // instruction (change
2463 // offsets below)
2464 __ andptr(array, -BytesPerInt);
2465
2466 // Initialize i & j
2467 __ xorl(i, i); // i = 0;
2468 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2469
2470 // Convert j into native byteordering
2471 __ bswapl(j);
2472
2473 // And start
2474 Label entry;
2475 __ jmp(entry);
2476
2477 // binary search loop
2478 {
2479 Label loop;
2480 __ bind(loop);
2481 // int h = (i + j) >> 1;
2482 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2483 __ sarl(h, 1); // h = (i + j) >> 1;
2484 // if (key < array[h].fast_match()) {
2485 // j = h;
2486 // } else {
2487 // i = h;
2488 // }
2489 // Convert array[h].match to native byte-ordering before compare
2490 __ movl(temp, Address(array, h, Address::times_8));
2491 __ bswapl(temp);
2492 __ cmpl(key, temp);
2493 // j = h if (key < array[h].fast_match())
2494 __ cmov32(Assembler::less, j, h);
2495 // i = h if (key >= array[h].fast_match())
2496 __ cmov32(Assembler::greaterEqual, i, h);
2497 // while (i+1 < j)
2498 __ bind(entry);
2499 __ leal(h, Address(i, 1)); // i+1
2500 __ cmpl(h, j); // i+1 < j
2501 __ jcc(Assembler::less, loop);
2502 }
2503
2504 // end of binary search, result index is i (must check again!)
2505 Label default_case;
2506 // Convert array[i].match to native byte-ordering before compare
2507 __ movl(temp, Address(array, i, Address::times_8));
2508 __ bswapl(temp);
2509 __ cmpl(key, temp);
2510 __ jcc(Assembler::notEqual, default_case);
2511
2512 // entry found -> j = offset
2513 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2514 __ profile_switch_case(i, key, array);
2515 __ bswapl(j);
2516 LP64_ONLY(__ movslq(j, j));
2517
2518 NOT_LP64(__ restore_bcp());
2519 NOT_LP64(__ restore_locals()); // restore rdi
2520
2521 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2522 __ addptr(rbcp, j);
2523 __ dispatch_only(vtos);
2524
2525 // default case -> j = default offset
2526 __ bind(default_case);
2527 __ profile_switch_default(i);
2528 __ movl(j, Address(array, -2 * BytesPerInt));
2529 __ bswapl(j);
2530 LP64_ONLY(__ movslq(j, j));
2531
2532 NOT_LP64(__ restore_bcp());
2533 NOT_LP64(__ restore_locals());
2534
2535 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2536 __ addptr(rbcp, j);
2537 __ dispatch_only(vtos);
2538 }
2539
2540 void TemplateTable::_return(TosState state) {
2541 transition(state, state);
2542
2543 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2544
2545 assert(_desc->calls_vm(),
2546 "inconsistent calls_vm information"); // call in remove_activation
2547
2548 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2549 assert(state == vtos, "only valid state");
2550 __ movptr(robj, aaddress(0));
2551 __ load_klass(rdi, robj);
2552 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2553 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2554 Label skip_register_finalizer;
2555 __ jcc(Assembler::zero, skip_register_finalizer);
2556
2557 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2558
2559 __ bind(skip_register_finalizer);
2560 }
2561
2562 __ remove_activation(state, rbcp);
2563 __ jmp(rbcp);
2564 }
2565
2566 // ----------------------------------------------------------------------------
2567 // Volatile variables demand their effects be made known to all CPU's
2568 // in order. Store buffers on most chips allow reads & writes to
2569 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2570 // without some kind of memory barrier (i.e., it's not sufficient that
2571 // the interpreter does not reorder volatile references, the hardware
2572 // also must not reorder them).
2573 //
2574 // According to the new Java Memory Model (JMM):
2575 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2576 // writes act as aquire & release, so:
2577 // (2) A read cannot let unrelated NON-volatile memory refs that
2578 // happen after the read float up to before the read. It's OK for
2579 // non-volatile memory refs that happen before the volatile read to
2580 // float down below it.
2581 // (3) Similar a volatile write cannot let unrelated NON-volatile
2582 // memory refs that happen BEFORE the write float down to after the
2583 // write. It's OK for non-volatile memory refs that happen after the
2584 // volatile write to float up before it.
2585 //
2586 // We only put in barriers around volatile refs (they are expensive),
2587 // not _between_ memory refs (that would require us to track the
2588 // flavor of the previous memory refs). Requirements (2) and (3)
2589 // require some barriers before volatile stores and after volatile
2590 // loads. These nearly cover requirement (1) but miss the
2591 // volatile-store-volatile-load case. This final case is placed after
2592 // volatile-stores although it could just as well go before
2593 // volatile-loads.
2594
2595 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2596 // Helper function to insert a is-volatile test and memory barrier
2597 if(!os::is_MP()) return; // Not needed on single CPU
2598 __ membar(order_constraint);
2599 }
2600
2601 void TemplateTable::resolve_cache_and_index(int byte_no,
2602 Register Rcache,
2603 Register index,
2604 size_t index_size) {
2605 const Register temp = rbx;
2606 assert_different_registers(Rcache, index, temp);
2607
2608 Label resolved;
2609
2610 Bytecodes::Code code = bytecode();
2611 switch (code) {
2612 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2613 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2614 }
2615
2616 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2617 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2618 __ cmpl(temp, code); // have we resolved this bytecode?
2619 __ jcc(Assembler::equal, resolved);
2620
2621 // resolve first time through
2622 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2623 __ movl(temp, code);
2624 __ call_VM(noreg, entry, temp);
2625 // Update registers with resolved info
2626 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2627 __ bind(resolved);
2628 }
2629
2630 // The cache and index registers must be set before call
2631 void TemplateTable::load_field_cp_cache_entry(Register obj,
2632 Register cache,
2633 Register index,
2634 Register off,
2635 Register flags,
2636 bool is_static = false) {
2637 assert_different_registers(cache, index, flags, off);
2638
2639 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2640 // Field offset
2641 __ movptr(off, Address(cache, index, Address::times_ptr,
2642 in_bytes(cp_base_offset +
2643 ConstantPoolCacheEntry::f2_offset())));
2644 // Flags
2645 __ movl(flags, Address(cache, index, Address::times_ptr,
2646 in_bytes(cp_base_offset +
2647 ConstantPoolCacheEntry::flags_offset())));
2648
2649 // klass overwrite register
2650 if (is_static) {
2651 __ movptr(obj, Address(cache, index, Address::times_ptr,
2652 in_bytes(cp_base_offset +
2653 ConstantPoolCacheEntry::f1_offset())));
2654 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2655 __ movptr(obj, Address(obj, mirror_offset));
2656 }
2657 }
2658
2659 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2660 Register method,
2661 Register itable_index,
2662 Register flags,
2663 bool is_invokevirtual,
2664 bool is_invokevfinal, /*unused*/
2665 bool is_invokedynamic) {
2666 // setup registers
2667 const Register cache = rcx;
2668 const Register index = rdx;
2669 assert_different_registers(method, flags);
2670 assert_different_registers(method, cache, index);
2671 assert_different_registers(itable_index, flags);
2672 assert_different_registers(itable_index, cache, index);
2673 // determine constant pool cache field offsets
2674 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2675 const int method_offset = in_bytes(
2676 ConstantPoolCache::base_offset() +
2677 ((byte_no == f2_byte)
2678 ? ConstantPoolCacheEntry::f2_offset()
2679 : ConstantPoolCacheEntry::f1_offset()));
2680 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2681 ConstantPoolCacheEntry::flags_offset());
2682 // access constant pool cache fields
2683 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2684 ConstantPoolCacheEntry::f2_offset());
2685
2686 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2687 resolve_cache_and_index(byte_no, cache, index, index_size);
2688 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2689
2690 if (itable_index != noreg) {
2691 // pick up itable or appendix index from f2 also:
2692 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2693 }
2694 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2695 }
2696
2697 // The registers cache and index expected to be set before call.
2698 // Correct values of the cache and index registers are preserved.
2699 void TemplateTable::jvmti_post_field_access(Register cache,
2700 Register index,
2701 bool is_static,
2702 bool has_tos) {
2703 if (JvmtiExport::can_post_field_access()) {
2704 // Check to see if a field access watch has been set before we take
2705 // the time to call into the VM.
2706 Label L1;
2707 assert_different_registers(cache, index, rax);
2708 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2709 __ testl(rax,rax);
2710 __ jcc(Assembler::zero, L1);
2711
2712 // cache entry pointer
2713 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2714 __ shll(index, LogBytesPerWord);
2715 __ addptr(cache, index);
2716 if (is_static) {
2717 __ xorptr(rax, rax); // NULL object reference
2718 } else {
2719 __ pop(atos); // Get the object
2720 __ verify_oop(rax);
2721 __ push(atos); // Restore stack state
2722 }
2723 // rax,: object pointer or NULL
2724 // cache: cache entry pointer
2725 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2726 rax, cache);
2727 __ get_cache_and_index_at_bcp(cache, index, 1);
2728 __ bind(L1);
2729 }
2730 }
2731
2732 void TemplateTable::pop_and_check_object(Register r) {
2733 __ pop_ptr(r);
2734 __ null_check(r); // for field access must check obj.
2735 __ verify_oop(r);
2736 }
2737
2738 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2739 transition(vtos, vtos);
2740
2741 const Register cache = rcx;
2742 const Register index = rdx;
2743 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2744 const Register off = rbx;
2745 const Register flags = rax;
2746 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2747
2748 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2749 jvmti_post_field_access(cache, index, is_static, false);
2750 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2751
2752 if (!is_static) pop_and_check_object(obj);
2753
2754 const Address field(obj, off, Address::times_1, 0*wordSize);
2755 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2756
2757 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2758
2759 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2760 // Make sure we don't need to mask edx after the above shift
2761 assert(btos == 0, "change code, btos != 0");
2762
2763 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2764
2765 __ jcc(Assembler::notZero, notByte);
2766 // btos
2767 __ load_signed_byte(rax, field);
2768 __ push(btos);
2769 // Rewrite bytecode to be faster
2770 if (!is_static && rc == may_rewrite) {
2771 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2772 }
2773 __ jmp(Done);
2774
2775 __ bind(notByte);
2776 __ cmpl(flags, atos);
2777 __ jcc(Assembler::notEqual, notObj);
2778 // atos
2779 __ load_heap_oop(rax, field);
2780 __ push(atos);
2781 if (!is_static && rc == may_rewrite) {
2782 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2783 }
2784 __ jmp(Done);
2785
2786 __ bind(notObj);
2787 __ cmpl(flags, itos);
2788 __ jcc(Assembler::notEqual, notInt);
2789 // itos
2790 __ movl(rax, field);
2791 __ push(itos);
2792 // Rewrite bytecode to be faster
2793 if (!is_static && rc == may_rewrite) {
2794 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2795 }
2796 __ jmp(Done);
2797
2798 __ bind(notInt);
2799 __ cmpl(flags, ctos);
2800 __ jcc(Assembler::notEqual, notChar);
2801 // ctos
2802 __ load_unsigned_short(rax, field);
2803 __ push(ctos);
2804 // Rewrite bytecode to be faster
2805 if (!is_static && rc == may_rewrite) {
2806 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2807 }
2808 __ jmp(Done);
2809
2810 __ bind(notChar);
2811 __ cmpl(flags, stos);
2812 __ jcc(Assembler::notEqual, notShort);
2813 // stos
2814 __ load_signed_short(rax, field);
2815 __ push(stos);
2816 // Rewrite bytecode to be faster
2817 if (!is_static && rc == may_rewrite) {
2818 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2819 }
2820 __ jmp(Done);
2821
2822 __ bind(notShort);
2823 __ cmpl(flags, ltos);
2824 __ jcc(Assembler::notEqual, notLong);
2825 // ltos
2826
2827 #ifndef _LP64
2828 // Generate code as if volatile. There just aren't enough registers to
2829 // save that information and this code is faster than the test.
2830 __ fild_d(field); // Must load atomically
2831 __ subptr(rsp,2*wordSize); // Make space for store
2832 __ fistp_d(Address(rsp,0));
2833 __ pop(rax);
2834 __ pop(rdx);
2835 #else
2836 __ movq(rax, field);
2837 #endif
2838
2839 __ push(ltos);
2840 // Rewrite bytecode to be faster
2841 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2842 __ jmp(Done);
2843
2844 __ bind(notLong);
2845 __ cmpl(flags, ftos);
2846 __ jcc(Assembler::notEqual, notFloat);
2847 // ftos
2848
2849 __ load_float(field);
2850 __ push(ftos);
2851 // Rewrite bytecode to be faster
2852 if (!is_static && rc == may_rewrite) {
2853 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2854 }
2855 __ jmp(Done);
2856
2857 __ bind(notFloat);
2858 #ifdef ASSERT
2859 __ cmpl(flags, dtos);
2860 __ jcc(Assembler::notEqual, notDouble);
2861 #endif
2862 // dtos
2863 __ load_double(field);
2864 __ push(dtos);
2865 // Rewrite bytecode to be faster
2866 if (!is_static && rc == may_rewrite) {
2867 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2868 }
2869 #ifdef ASSERT
2870 __ jmp(Done);
2871
2872
2873 __ bind(notDouble);
2874 __ stop("Bad state");
2875 #endif
2876
2877 __ bind(Done);
2878 // [jk] not needed currently
2879 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2880 // Assembler::LoadStore));
2881 }
2882
2883 void TemplateTable::getfield(int byte_no) {
2884 getfield_or_static(byte_no, false);
2885 }
2886
2887 void TemplateTable::nofast_getfield(int byte_no) {
2888 getfield_or_static(byte_no, false, may_not_rewrite);
2889 }
2890
2891 void TemplateTable::getstatic(int byte_no) {
2892 getfield_or_static(byte_no, true);
2893 }
2894
2895
2896 // The registers cache and index expected to be set before call.
2897 // The function may destroy various registers, just not the cache and index registers.
2898 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2899
2900 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
2901 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2902 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2903 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
2904
2905 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2906
2907 if (JvmtiExport::can_post_field_modification()) {
2908 // Check to see if a field modification watch has been set before
2909 // we take the time to call into the VM.
2910 Label L1;
2911 assert_different_registers(cache, index, rax);
2912 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2913 __ testl(rax, rax);
2914 __ jcc(Assembler::zero, L1);
2915
2916 __ get_cache_and_index_at_bcp(robj, RDX, 1);
2917
2918
2919 if (is_static) {
2920 // Life is simple. Null out the object pointer.
2921 __ xorl(RBX, RBX);
2922
2923 } else {
2924 // Life is harder. The stack holds the value on top, followed by
2925 // the object. We don't know the size of the value, though; it
2926 // could be one or two words depending on its type. As a result,
2927 // we must find the type to determine where the object is.
2928 #ifndef _LP64
2929 Label two_word, valsize_known;
2930 #endif
2931 __ movl(RCX, Address(robj, RDX,
2932 Address::times_ptr,
2933 in_bytes(cp_base_offset +
2934 ConstantPoolCacheEntry::flags_offset())));
2935 NOT_LP64(__ mov(rbx, rsp));
2936 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
2937
2938 // Make sure we don't need to mask rcx after the above shift
2939 ConstantPoolCacheEntry::verify_tos_state_shift();
2940 #ifdef _LP64
2941 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2942 __ cmpl(c_rarg3, ltos);
2943 __ cmovptr(Assembler::equal,
2944 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2945 __ cmpl(c_rarg3, dtos);
2946 __ cmovptr(Assembler::equal,
2947 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2948 #else
2949 __ cmpl(rcx, ltos);
2950 __ jccb(Assembler::equal, two_word);
2951 __ cmpl(rcx, dtos);
2952 __ jccb(Assembler::equal, two_word);
2953 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2954 __ jmpb(valsize_known);
2955
2956 __ bind(two_word);
2957 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2958
2959 __ bind(valsize_known);
2960 // setup object pointer
2961 __ movptr(rbx, Address(rbx, 0));
2962 #endif
2963 }
2964 // cache entry pointer
2965 __ addptr(robj, in_bytes(cp_base_offset));
2966 __ shll(RDX, LogBytesPerWord);
2967 __ addptr(robj, RDX);
2968 // object (tos)
2969 __ mov(RCX, rsp);
2970 // c_rarg1: object pointer set up above (NULL if static)
2971 // c_rarg2: cache entry pointer
2972 // c_rarg3: jvalue object on the stack
2973 __ call_VM(noreg,
2974 CAST_FROM_FN_PTR(address,
2975 InterpreterRuntime::post_field_modification),
2976 RBX, robj, RCX);
2977 __ get_cache_and_index_at_bcp(cache, index, 1);
2978 __ bind(L1);
2979 }
2980 }
2981
2982 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2983 transition(vtos, vtos);
2984
2985 const Register cache = rcx;
2986 const Register index = rdx;
2987 const Register obj = rcx;
2988 const Register off = rbx;
2989 const Register flags = rax;
2990 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2991
2992 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2993 jvmti_post_field_mod(cache, index, is_static);
2994 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2995
2996 // [jk] not needed currently
2997 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2998 // Assembler::StoreStore));
2999
3000 Label notVolatile, Done;
3001 __ movl(rdx, flags);
3002 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3003 __ andl(rdx, 0x1);
3004
3005 // field addresses
3006 const Address field(obj, off, Address::times_1, 0*wordSize);
3007 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3008
3009 Label notByte, notInt, notShort, notChar,
3010 notLong, notFloat, notObj, notDouble;
3011
3012 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3013
3014 assert(btos == 0, "change code, btos != 0");
3015 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3016 __ jcc(Assembler::notZero, notByte);
3017
3018 // btos
3019 {
3020 __ pop(btos);
3021 if (!is_static) pop_and_check_object(obj);
3022 __ movb(field, rax);
3023 if (!is_static && rc == may_rewrite) {
3024 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3025 }
3026 __ jmp(Done);
3027 }
3028
3029 __ bind(notByte);
3030 __ cmpl(flags, atos);
3031 __ jcc(Assembler::notEqual, notObj);
3032
3033 // atos
3034 {
3035 __ pop(atos);
3036 if (!is_static) pop_and_check_object(obj);
3037 // Store into the field
3038 do_oop_store(_masm, field, rax, _bs->kind(), false);
3039 if (!is_static && rc == may_rewrite) {
3040 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3041 }
3042 __ jmp(Done);
3043 }
3044
3045 __ bind(notObj);
3046 __ cmpl(flags, itos);
3047 __ jcc(Assembler::notEqual, notInt);
3048
3049 // itos
3050 {
3051 __ pop(itos);
3052 if (!is_static) pop_and_check_object(obj);
3053 __ movl(field, rax);
3054 if (!is_static && rc == may_rewrite) {
3055 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3056 }
3057 __ jmp(Done);
3058 }
3059
3060 __ bind(notInt);
3061 __ cmpl(flags, ctos);
3062 __ jcc(Assembler::notEqual, notChar);
3063
3064 // ctos
3065 {
3066 __ pop(ctos);
3067 if (!is_static) pop_and_check_object(obj);
3068 __ movw(field, rax);
3069 if (!is_static && rc == may_rewrite) {
3070 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3071 }
3072 __ jmp(Done);
3073 }
3074
3075 __ bind(notChar);
3076 __ cmpl(flags, stos);
3077 __ jcc(Assembler::notEqual, notShort);
3078
3079 // stos
3080 {
3081 __ pop(stos);
3082 if (!is_static) pop_and_check_object(obj);
3083 __ movw(field, rax);
3084 if (!is_static && rc == may_rewrite) {
3085 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3086 }
3087 __ jmp(Done);
3088 }
3089
3090 __ bind(notShort);
3091 __ cmpl(flags, ltos);
3092 __ jcc(Assembler::notEqual, notLong);
3093
3094 // ltos
3095 #ifdef _LP64
3096 {
3097 __ pop(ltos);
3098 if (!is_static) pop_and_check_object(obj);
3099 __ movq(field, rax);
3100 if (!is_static && rc == may_rewrite) {
3101 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3102 }
3103 __ jmp(Done);
3104 }
3105 #else
3106 {
3107 Label notVolatileLong;
3108 __ testl(rdx, rdx);
3109 __ jcc(Assembler::zero, notVolatileLong);
3110
3111 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3112 if (!is_static) pop_and_check_object(obj);
3113
3114 // Replace with real volatile test
3115 __ push(rdx);
3116 __ push(rax); // Must update atomically with FIST
3117 __ fild_d(Address(rsp,0)); // So load into FPU register
3118 __ fistp_d(field); // and put into memory atomically
3119 __ addptr(rsp, 2*wordSize);
3120 // volatile_barrier();
3121 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3122 Assembler::StoreStore));
3123 // Don't rewrite volatile version
3124 __ jmp(notVolatile);
3125
3126 __ bind(notVolatileLong);
3127
3128 __ pop(ltos); // overwrites rdx
3129 if (!is_static) pop_and_check_object(obj);
3130 __ movptr(hi, rdx);
3131 __ movptr(field, rax);
3132 // Don't rewrite to _fast_lputfield for potential volatile case.
3133 __ jmp(notVolatile);
3134 }
3135 #endif // _LP64
3136
3137 __ bind(notLong);
3138 __ cmpl(flags, ftos);
3139 __ jcc(Assembler::notEqual, notFloat);
3140
3141 // ftos
3142 {
3143 __ pop(ftos);
3144 if (!is_static) pop_and_check_object(obj);
3145 __ store_float(field);
3146 if (!is_static && rc == may_rewrite) {
3147 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3148 }
3149 __ jmp(Done);
3150 }
3151
3152 __ bind(notFloat);
3153 #ifdef ASSERT
3154 __ cmpl(flags, dtos);
3155 __ jcc(Assembler::notEqual, notDouble);
3156 #endif
3157
3158 // dtos
3159 {
3160 __ pop(dtos);
3161 if (!is_static) pop_and_check_object(obj);
3162 __ store_double(field);
3163 if (!is_static && rc == may_rewrite) {
3164 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3165 }
3166 }
3167
3168 #ifdef ASSERT
3169 __ jmp(Done);
3170
3171 __ bind(notDouble);
3172 __ stop("Bad state");
3173 #endif
3174
3175 __ bind(Done);
3176
3177 // Check for volatile store
3178 __ testl(rdx, rdx);
3179 __ jcc(Assembler::zero, notVolatile);
3180 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3181 Assembler::StoreStore));
3182 __ bind(notVolatile);
3183 }
3184
3185 void TemplateTable::putfield(int byte_no) {
3186 putfield_or_static(byte_no, false);
3187 }
3188
3189 void TemplateTable::nofast_putfield(int byte_no) {
3190 putfield_or_static(byte_no, false, may_not_rewrite);
3191 }
3192
3193 void TemplateTable::putstatic(int byte_no) {
3194 putfield_or_static(byte_no, true);
3195 }
3196
3197 void TemplateTable::jvmti_post_fast_field_mod() {
3198
3199 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3200
3201 if (JvmtiExport::can_post_field_modification()) {
3202 // Check to see if a field modification watch has been set before
3203 // we take the time to call into the VM.
3204 Label L2;
3205 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3206 __ testl(scratch, scratch);
3207 __ jcc(Assembler::zero, L2);
3208 __ pop_ptr(rbx); // copy the object pointer from tos
3209 __ verify_oop(rbx);
3210 __ push_ptr(rbx); // put the object pointer back on tos
3211 // Save tos values before call_VM() clobbers them. Since we have
3212 // to do it for every data type, we use the saved values as the
3213 // jvalue object.
3214 switch (bytecode()) { // load values into the jvalue object
3215 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3216 case Bytecodes::_fast_bputfield: // fall through
3217 case Bytecodes::_fast_sputfield: // fall through
3218 case Bytecodes::_fast_cputfield: // fall through
3219 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3220 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3221 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3222 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3223
3224 default:
3225 ShouldNotReachHere();
3226 }
3227 __ mov(scratch, rsp); // points to jvalue on the stack
3228 // access constant pool cache entry
3229 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3230 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3231 __ verify_oop(rbx);
3232 // rbx: object pointer copied above
3233 // c_rarg2: cache entry pointer
3234 // c_rarg3: jvalue object on the stack
3235 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3236 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3237
3238 switch (bytecode()) { // restore tos values
3239 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3240 case Bytecodes::_fast_bputfield: // fall through
3241 case Bytecodes::_fast_sputfield: // fall through
3242 case Bytecodes::_fast_cputfield: // fall through
3243 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3244 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3245 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3246 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3247 }
3248 __ bind(L2);
3249 }
3250 }
3251
3252 void TemplateTable::fast_storefield(TosState state) {
3253 transition(state, vtos);
3254
3255 ByteSize base = ConstantPoolCache::base_offset();
3256
3257 jvmti_post_fast_field_mod();
3258
3259 // access constant pool cache
3260 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3261
3262 // test for volatile with rdx but rdx is tos register for lputfield.
3263 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3264 in_bytes(base +
3265 ConstantPoolCacheEntry::flags_offset())));
3266
3267 // replace index with field offset from cache entry
3268 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3269 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3270
3271 // [jk] not needed currently
3272 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3273 // Assembler::StoreStore));
3274
3275 Label notVolatile;
3276 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3277 __ andl(rdx, 0x1);
3278
3279 // Get object from stack
3280 pop_and_check_object(rcx);
3281
3282 // field address
3283 const Address field(rcx, rbx, Address::times_1);
3284
3285 // access field
3286 switch (bytecode()) {
3287 case Bytecodes::_fast_aputfield:
3288 do_oop_store(_masm, field, rax, _bs->kind(), false);
3289 break;
3290 case Bytecodes::_fast_lputfield:
3291 #ifdef _LP64
3292 __ movq(field, rax);
3293 #else
3294 __ stop("should not be rewritten");
3295 #endif
3296 break;
3297 case Bytecodes::_fast_iputfield:
3298 __ movl(field, rax);
3299 break;
3300 case Bytecodes::_fast_bputfield:
3301 __ movb(field, rax);
3302 break;
3303 case Bytecodes::_fast_sputfield:
3304 // fall through
3305 case Bytecodes::_fast_cputfield:
3306 __ movw(field, rax);
3307 break;
3308 case Bytecodes::_fast_fputfield:
3309 __ store_float(field);
3310 break;
3311 case Bytecodes::_fast_dputfield:
3312 __ store_double(field);
3313 break;
3314 default:
3315 ShouldNotReachHere();
3316 }
3317
3318 // Check for volatile store
3319 __ testl(rdx, rdx);
3320 __ jcc(Assembler::zero, notVolatile);
3321 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3322 Assembler::StoreStore));
3323 __ bind(notVolatile);
3324 }
3325
3326 void TemplateTable::fast_accessfield(TosState state) {
3327 transition(atos, state);
3328
3329 // Do the JVMTI work here to avoid disturbing the register state below
3330 if (JvmtiExport::can_post_field_access()) {
3331 // Check to see if a field access watch has been set before we
3332 // take the time to call into the VM.
3333 Label L1;
3334 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3335 __ testl(rcx, rcx);
3336 __ jcc(Assembler::zero, L1);
3337 // access constant pool cache entry
3338 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3339 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3340 __ verify_oop(rax);
3341 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3342 LP64_ONLY(__ mov(c_rarg1, rax));
3343 // c_rarg1: object pointer copied above
3344 // c_rarg2: cache entry pointer
3345 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3346 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3347 __ pop_ptr(rax); // restore object pointer
3348 __ bind(L1);
3349 }
3350
3351 // access constant pool cache
3352 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3353 // replace index with field offset from cache entry
3354 // [jk] not needed currently
3355 // if (os::is_MP()) {
3356 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3357 // in_bytes(ConstantPoolCache::base_offset() +
3358 // ConstantPoolCacheEntry::flags_offset())));
3359 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3360 // __ andl(rdx, 0x1);
3361 // }
3362 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3363 in_bytes(ConstantPoolCache::base_offset() +
3364 ConstantPoolCacheEntry::f2_offset())));
3365
3366 // rax: object
3367 __ verify_oop(rax);
3368 __ null_check(rax);
3369 Address field(rax, rbx, Address::times_1);
3370
3371 // access field
3372 switch (bytecode()) {
3373 case Bytecodes::_fast_agetfield:
3374 __ load_heap_oop(rax, field);
3375 __ verify_oop(rax);
3376 break;
3377 case Bytecodes::_fast_lgetfield:
3378 #ifdef _LP64
3379 __ movq(rax, field);
3380 #else
3381 __ stop("should not be rewritten");
3382 #endif
3383 break;
3384 case Bytecodes::_fast_igetfield:
3385 __ movl(rax, field);
3386 break;
3387 case Bytecodes::_fast_bgetfield:
3388 __ movsbl(rax, field);
3389 break;
3390 case Bytecodes::_fast_sgetfield:
3391 __ load_signed_short(rax, field);
3392 break;
3393 case Bytecodes::_fast_cgetfield:
3394 __ load_unsigned_short(rax, field);
3395 break;
3396 case Bytecodes::_fast_fgetfield:
3397 __ load_float(field);
3398 break;
3399 case Bytecodes::_fast_dgetfield:
3400 __ load_double(field);
3401 break;
3402 default:
3403 ShouldNotReachHere();
3404 }
3405 // [jk] not needed currently
3406 // if (os::is_MP()) {
3407 // Label notVolatile;
3408 // __ testl(rdx, rdx);
3409 // __ jcc(Assembler::zero, notVolatile);
3410 // __ membar(Assembler::LoadLoad);
3411 // __ bind(notVolatile);
3412 //};
3413 }
3414
3415 void TemplateTable::fast_xaccess(TosState state) {
3416 transition(vtos, state);
3417
3418 // get receiver
3419 __ movptr(rax, aaddress(0));
3420 // access constant pool cache
3421 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3422 __ movptr(rbx,
3423 Address(rcx, rdx, Address::times_ptr,
3424 in_bytes(ConstantPoolCache::base_offset() +
3425 ConstantPoolCacheEntry::f2_offset())));
3426 // make sure exception is reported in correct bcp range (getfield is
3427 // next instruction)
3428 __ increment(rbcp);
3429 __ null_check(rax);
3430 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3431 switch (state) {
3432 case itos:
3433 __ movl(rax, field);
3434 break;
3435 case atos:
3436 __ load_heap_oop(rax, field);
3437 __ verify_oop(rax);
3438 break;
3439 case ftos:
3440 __ load_float(field);
3441 break;
3442 default:
3443 ShouldNotReachHere();
3444 }
3445
3446 // [jk] not needed currently
3447 // if (os::is_MP()) {
3448 // Label notVolatile;
3449 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3450 // in_bytes(ConstantPoolCache::base_offset() +
3451 // ConstantPoolCacheEntry::flags_offset())));
3452 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3453 // __ testl(rdx, 0x1);
3454 // __ jcc(Assembler::zero, notVolatile);
3455 // __ membar(Assembler::LoadLoad);
3456 // __ bind(notVolatile);
3457 // }
3458
3459 __ decrement(rbcp);
3460 }
3461
3462 //-----------------------------------------------------------------------------
3463 // Calls
3464
3465 void TemplateTable::count_calls(Register method, Register temp) {
3466 // implemented elsewhere
3467 ShouldNotReachHere();
3468 }
3469
3470 void TemplateTable::prepare_invoke(int byte_no,
3471 Register method, // linked method (or i-klass)
3472 Register index, // itable index, MethodType, etc.
3473 Register recv, // if caller wants to see it
3474 Register flags // if caller wants to test it
3475 ) {
3476 // determine flags
3477 const Bytecodes::Code code = bytecode();
3478 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3479 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3480 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3481 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3482 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3483 const bool load_receiver = (recv != noreg);
3484 const bool save_flags = (flags != noreg);
3485 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3486 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3487 assert(flags == noreg || flags == rdx, "");
3488 assert(recv == noreg || recv == rcx, "");
3489
3490 // setup registers & access constant pool cache
3491 if (recv == noreg) recv = rcx;
3492 if (flags == noreg) flags = rdx;
3493 assert_different_registers(method, index, recv, flags);
3494
3495 // save 'interpreter return address'
3496 __ save_bcp();
3497
3498 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3499
3500 // maybe push appendix to arguments (just before return address)
3501 if (is_invokedynamic || is_invokehandle) {
3502 Label L_no_push;
3503 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3504 __ jcc(Assembler::zero, L_no_push);
3505 // Push the appendix as a trailing parameter.
3506 // This must be done before we get the receiver,
3507 // since the parameter_size includes it.
3508 __ push(rbx);
3509 __ mov(rbx, index);
3510 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3511 __ load_resolved_reference_at_index(index, rbx);
3512 __ pop(rbx);
3513 __ push(index); // push appendix (MethodType, CallSite, etc.)
3514 __ bind(L_no_push);
3515 }
3516
3517 // load receiver if needed (after appendix is pushed so parameter size is correct)
3518 // Note: no return address pushed yet
3519 if (load_receiver) {
3520 __ movl(recv, flags);
3521 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3522 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3523 const int receiver_is_at_end = -1; // back off one slot to get receiver
3524 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3525 __ movptr(recv, recv_addr);
3526 __ verify_oop(recv);
3527 }
3528
3529 if (save_flags) {
3530 __ movl(rbcp, flags);
3531 }
3532
3533 // compute return type
3534 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3535 // Make sure we don't need to mask flags after the above shift
3536 ConstantPoolCacheEntry::verify_tos_state_shift();
3537 // load return address
3538 {
3539 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3540 ExternalAddress table(table_addr);
3541 LP64_ONLY(__ lea(rscratch1, table));
3542 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3543 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3544 }
3545
3546 // push return address
3547 __ push(flags);
3548
3549 // Restore flags value from the constant pool cache, and restore rsi
3550 // for later null checks. r13 is the bytecode pointer
3551 if (save_flags) {
3552 __ movl(flags, rbcp);
3553 __ restore_bcp();
3554 }
3555 }
3556
3557 void TemplateTable::invokevirtual_helper(Register index,
3558 Register recv,
3559 Register flags) {
3560 // Uses temporary registers rax, rdx
3561 assert_different_registers(index, recv, rax, rdx);
3562 assert(index == rbx, "");
3563 assert(recv == rcx, "");
3564
3565 // Test for an invoke of a final method
3566 Label notFinal;
3567 __ movl(rax, flags);
3568 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3569 __ jcc(Assembler::zero, notFinal);
3570
3571 const Register method = index; // method must be rbx
3572 assert(method == rbx,
3573 "Method* must be rbx for interpreter calling convention");
3574
3575 // do the call - the index is actually the method to call
3576 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3577
3578 // It's final, need a null check here!
3579 __ null_check(recv);
3580
3581 // profile this call
3582 __ profile_final_call(rax);
3583 __ profile_arguments_type(rax, method, rbcp, true);
3584
3585 __ jump_from_interpreted(method, rax);
3586
3587 __ bind(notFinal);
3588
3589 // get receiver klass
3590 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3591 __ load_klass(rax, recv);
3592
3593 // profile this call
3594 __ profile_virtual_call(rax, rlocals, rdx);
3595 // get target Method* & entry point
3596 __ lookup_virtual_method(rax, index, method);
3597 __ profile_called_method(method, rdx, rbcp);
3598
3599 __ profile_arguments_type(rdx, method, rbcp, true);
3600 __ jump_from_interpreted(method, rdx);
3601 }
3602
3603 void TemplateTable::invokevirtual(int byte_no) {
3604 transition(vtos, vtos);
3605 assert(byte_no == f2_byte, "use this argument");
3606 prepare_invoke(byte_no,
3607 rbx, // method or vtable index
3608 noreg, // unused itable index
3609 rcx, rdx); // recv, flags
3610
3611 // rbx: index
3612 // rcx: receiver
3613 // rdx: flags
3614
3615 invokevirtual_helper(rbx, rcx, rdx);
3616 }
3617
3618 void TemplateTable::invokespecial(int byte_no) {
3619 transition(vtos, vtos);
3620 assert(byte_no == f1_byte, "use this argument");
3621 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3622 rcx); // get receiver also for null check
3623 __ verify_oop(rcx);
3624 __ null_check(rcx);
3625 // do the call
3626 __ profile_call(rax);
3627 __ profile_arguments_type(rax, rbx, rbcp, false);
3628 __ jump_from_interpreted(rbx, rax);
3629 }
3630
3631 void TemplateTable::invokestatic(int byte_no) {
3632 transition(vtos, vtos);
3633 assert(byte_no == f1_byte, "use this argument");
3634 prepare_invoke(byte_no, rbx); // get f1 Method*
3635 // do the call
3636 __ profile_call(rax);
3637 __ profile_arguments_type(rax, rbx, rbcp, false);
3638 __ jump_from_interpreted(rbx, rax);
3639 }
3640
3641
3642 void TemplateTable::fast_invokevfinal(int byte_no) {
3643 transition(vtos, vtos);
3644 assert(byte_no == f2_byte, "use this argument");
3645 __ stop("fast_invokevfinal not used on x86");
3646 }
3647
3648
3649 void TemplateTable::invokeinterface(int byte_no) {
3650 transition(vtos, vtos);
3651 assert(byte_no == f1_byte, "use this argument");
3652 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3653 rcx, rdx); // recv, flags
3654
3655 // rax: interface klass (from f1)
3656 // rbx: itable index (from f2)
3657 // rcx: receiver
3658 // rdx: flags
3659
3660 // Special case of invokeinterface called for virtual method of
3661 // java.lang.Object. See cpCacheOop.cpp for details.
3662 // This code isn't produced by javac, but could be produced by
3663 // another compliant java compiler.
3664 Label notMethod;
3665 __ movl(rlocals, rdx);
3666 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3667
3668 __ jcc(Assembler::zero, notMethod);
3669
3670 invokevirtual_helper(rbx, rcx, rdx);
3671 __ bind(notMethod);
3672
3673 // Get receiver klass into rdx - also a null check
3674 __ restore_locals(); // restore r14
3675 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3676 __ load_klass(rdx, rcx);
3677
3678 // profile this call
3679 __ profile_virtual_call(rdx, rbcp, rlocals);
3680
3681 Label no_such_interface, no_such_method;
3682
3683 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3684 rdx, rax, rbx,
3685 // outputs: method, scan temp. reg
3686 rbx, rbcp,
3687 no_such_interface);
3688
3689 // rbx: Method* to call
3690 // rcx: receiver
3691 // Check for abstract method error
3692 // Note: This should be done more efficiently via a throw_abstract_method_error
3693 // interpreter entry point and a conditional jump to it in case of a null
3694 // method.
3695 __ testptr(rbx, rbx);
3696 __ jcc(Assembler::zero, no_such_method);
3697
3698 __ profile_called_method(rbx, rbcp, rdx);
3699 __ profile_arguments_type(rdx, rbx, rbcp, true);
3700
3701 // do the call
3702 // rcx: receiver
3703 // rbx,: Method*
3704 __ jump_from_interpreted(rbx, rdx);
3705 __ should_not_reach_here();
3706
3707 // exception handling code follows...
3708 // note: must restore interpreter registers to canonical
3709 // state for exception handling to work correctly!
3710
3711 __ bind(no_such_method);
3712 // throw exception
3713 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3714 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3715 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3716 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3717 // the call_VM checks for exception, so we should never return here.
3718 __ should_not_reach_here();
3719
3720 __ bind(no_such_interface);
3721 // throw exception
3722 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3723 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3724 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3725 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3726 InterpreterRuntime::throw_IncompatibleClassChangeError));
3727 // the call_VM checks for exception, so we should never return here.
3728 __ should_not_reach_here();
3729 }
3730
3731 void TemplateTable::invokehandle(int byte_no) {
3732 transition(vtos, vtos);
3733 assert(byte_no == f1_byte, "use this argument");
3734 const Register rbx_method = rbx;
3735 const Register rax_mtype = rax;
3736 const Register rcx_recv = rcx;
3737 const Register rdx_flags = rdx;
3738
3739 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3740 __ verify_method_ptr(rbx_method);
3741 __ verify_oop(rcx_recv);
3742 __ null_check(rcx_recv);
3743
3744 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3745 // rbx: MH.invokeExact_MT method (from f2)
3746
3747 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3748
3749 // FIXME: profile the LambdaForm also
3750 __ profile_final_call(rax);
3751 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3752
3753 __ jump_from_interpreted(rbx_method, rdx);
3754 }
3755
3756 void TemplateTable::invokedynamic(int byte_no) {
3757 transition(vtos, vtos);
3758 assert(byte_no == f1_byte, "use this argument");
3759
3760 const Register rbx_method = rbx;
3761 const Register rax_callsite = rax;
3762
3763 prepare_invoke(byte_no, rbx_method, rax_callsite);
3764
3765 // rax: CallSite object (from cpool->resolved_references[f1])
3766 // rbx: MH.linkToCallSite method (from f2)
3767
3768 // Note: rax_callsite is already pushed by prepare_invoke
3769
3770 // %%% should make a type profile for any invokedynamic that takes a ref argument
3771 // profile this call
3772 __ profile_call(rbcp);
3773 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3774
3775 __ verify_oop(rax_callsite);
3776
3777 __ jump_from_interpreted(rbx_method, rdx);
3778 }
3779
3780 //-----------------------------------------------------------------------------
3781 // Allocation
3782
3783 void TemplateTable::_new() {
3784 transition(vtos, atos);
3785 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3786 Label slow_case;
3787 Label slow_case_no_pop;
3788 Label done;
3789 Label initialize_header;
3790 Label initialize_object; // including clearing the fields
3791 Label allocate_shared;
3792
3793 __ get_cpool_and_tags(rcx, rax);
3794
3795 // Make sure the class we're about to instantiate has been resolved.
3796 // This is done before loading InstanceKlass to be consistent with the order
3797 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3798 const int tags_offset = Array<u1>::base_offset_in_bytes();
3799 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3800 __ jcc(Assembler::notEqual, slow_case_no_pop);
3801
3802 // get InstanceKlass
3803 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool)));
3804 __ push(rcx); // save the contexts of klass for initializing the header
3805
3806 // make sure klass is initialized & doesn't have finalizer
3807 // make sure klass is fully initialized
3808 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3809 __ jcc(Assembler::notEqual, slow_case);
3810
3811 // get instance_size in InstanceKlass (scaled to a count of bytes)
3812 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3813 // test to see if it has a finalizer or is malformed in some way
3814 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3815 __ jcc(Assembler::notZero, slow_case);
3816
3817 //
3818 // Allocate the instance
3819 // 1) Try to allocate in the TLAB
3820 // 2) if fail and the object is large allocate in the shared Eden
3821 // 3) if the above fails (or is not applicable), go to a slow case
3822 // (creates a new TLAB, etc.)
3823
3824 const bool allow_shared_alloc =
3825 Universe::heap()->supports_inline_contig_alloc();
3826
3827 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
3828 #ifndef _LP64
3829 if (UseTLAB || allow_shared_alloc) {
3830 __ get_thread(thread);
3831 }
3832 #endif // _LP64
3833
3834 if (UseTLAB) {
3835 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3836 __ lea(rbx, Address(rax, rdx, Address::times_1));
3837 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3838 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3839 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3840 if (ZeroTLAB) {
3841 // the fields have been already cleared
3842 __ jmp(initialize_header);
3843 } else {
3844 // initialize both the header and fields
3845 __ jmp(initialize_object);
3846 }
3847 }
3848
3849 // Allocation in the shared Eden, if allowed.
3850 //
3851 // rdx: instance size in bytes
3852 if (allow_shared_alloc) {
3853 __ bind(allocate_shared);
3854
3855 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3856 ExternalAddress heap_end((address)Universe::heap()->end_addr());
3857
3858 Label retry;
3859 __ bind(retry);
3860 __ movptr(rax, heap_top);
3861 __ lea(rbx, Address(rax, rdx, Address::times_1));
3862 __ cmpptr(rbx, heap_end);
3863 __ jcc(Assembler::above, slow_case);
3864
3865 // Compare rax, with the top addr, and if still equal, store the new
3866 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3867 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3868 //
3869 // rax,: object begin
3870 // rbx,: object end
3871 // rdx: instance size in bytes
3872 __ locked_cmpxchgptr(rbx, heap_top);
3873
3874 // if someone beat us on the allocation, try again, otherwise continue
3875 __ jcc(Assembler::notEqual, retry);
3876
3877 __ incr_allocated_bytes(thread, rdx, 0);
3878 }
3879
3880 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3881 // The object is initialized before the header. If the object size is
3882 // zero, go directly to the header initialization.
3883 __ bind(initialize_object);
3884 __ decrement(rdx, sizeof(oopDesc));
3885 __ jcc(Assembler::zero, initialize_header);
3886
3887 // Initialize topmost object field, divide rdx by 8, check if odd and
3888 // test if zero.
3889 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3890 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3891
3892 // rdx must have been multiple of 8
3893 #ifdef ASSERT
3894 // make sure rdx was multiple of 8
3895 Label L;
3896 // Ignore partial flag stall after shrl() since it is debug VM
3897 __ jccb(Assembler::carryClear, L);
3898 __ stop("object size is not multiple of 2 - adjust this code");
3899 __ bind(L);
3900 // rdx must be > 0, no extra check needed here
3901 #endif
3902
3903 // initialize remaining object fields: rdx was a multiple of 8
3904 { Label loop;
3905 __ bind(loop);
3906 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3907 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3908 __ decrement(rdx);
3909 __ jcc(Assembler::notZero, loop);
3910 }
3911
3912 // initialize object header only.
3913 __ bind(initialize_header);
3914 if (UseBiasedLocking) {
3915 __ pop(rcx); // get saved klass back in the register.
3916 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
3917 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3918 } else {
3919 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3920 (intptr_t)markOopDesc::prototype()); // header
3921 __ pop(rcx); // get saved klass back in the register.
3922 }
3923 #ifdef _LP64
3924 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
3925 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
3926 #endif
3927 __ store_klass(rax, rcx); // klass
3928
3929 {
3930 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3931 // Trigger dtrace event for fastpath
3932 __ push(atos);
3933 __ call_VM_leaf(
3934 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3935 __ pop(atos);
3936 }
3937
3938 __ jmp(done);
3939 }
3940
3941 // slow case
3942 __ bind(slow_case);
3943 __ pop(rcx); // restore stack pointer to what it was when we came in.
3944 __ bind(slow_case_no_pop);
3945
3946 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
3947 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3948
3949 __ get_constant_pool(rarg1);
3950 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
3951 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
3952 __ verify_oop(rax);
3953
3954 // continue
3955 __ bind(done);
3956 }
3957
3958 void TemplateTable::newarray() {
3959 transition(itos, atos);
3960 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
3961 __ load_unsigned_byte(rarg1, at_bcp(1));
3962 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3963 rarg1, rax);
3964 }
3965
3966 void TemplateTable::anewarray() {
3967 transition(itos, atos);
3968
3969 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
3970 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3971
3972 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
3973 __ get_constant_pool(rarg1);
3974 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3975 rarg1, rarg2, rax);
3976 }
3977
3978 void TemplateTable::arraylength() {
3979 transition(atos, itos);
3980 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3981 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3982 }
3983
3984 void TemplateTable::checkcast() {
3985 transition(atos, atos);
3986 Label done, is_null, ok_is_subtype, quicked, resolved;
3987 __ testptr(rax, rax); // object is in rax
3988 __ jcc(Assembler::zero, is_null);
3989
3990 // Get cpool & tags index
3991 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3992 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3993 // See if bytecode has already been quicked
3994 __ cmpb(Address(rdx, rbx,
3995 Address::times_1,
3996 Array<u1>::base_offset_in_bytes()),
3997 JVM_CONSTANT_Class);
3998 __ jcc(Assembler::equal, quicked);
3999 __ push(atos); // save receiver for result, and for GC
4000 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4001
4002 // vm_result_2 has metadata result
4003 #ifndef _LP64
4004 // borrow rdi from locals
4005 __ get_thread(rdi);
4006 __ get_vm_result_2(rax, rdi);
4007 __ restore_locals();
4008 #else
4009 __ get_vm_result_2(rax, r15_thread);
4010 #endif
4011
4012 __ pop_ptr(rdx); // restore receiver
4013 __ jmpb(resolved);
4014
4015 // Get superklass in rax and subklass in rbx
4016 __ bind(quicked);
4017 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4018 __ movptr(rax, Address(rcx, rbx,
4019 Address::times_ptr, sizeof(ConstantPool)));
4020
4021 __ bind(resolved);
4022 __ load_klass(rbx, rdx);
4023
4024 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4025 // Superklass in rax. Subklass in rbx.
4026 __ gen_subtype_check(rbx, ok_is_subtype);
4027
4028 // Come here on failure
4029 __ push_ptr(rdx);
4030 // object is at TOS
4031 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4032
4033 // Come here on success
4034 __ bind(ok_is_subtype);
4035 __ mov(rax, rdx); // Restore object in rdx
4036
4037 // Collect counts on whether this check-cast sees NULLs a lot or not.
4038 if (ProfileInterpreter) {
4039 __ jmp(done);
4040 __ bind(is_null);
4041 __ profile_null_seen(rcx);
4042 } else {
4043 __ bind(is_null); // same as 'done'
4044 }
4045 __ bind(done);
4046 }
4047
4048 void TemplateTable::instanceof() {
4049 transition(atos, itos);
4050 Label done, is_null, ok_is_subtype, quicked, resolved;
4051 __ testptr(rax, rax);
4052 __ jcc(Assembler::zero, is_null);
4053
4054 // Get cpool & tags index
4055 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4056 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4057 // See if bytecode has already been quicked
4058 __ cmpb(Address(rdx, rbx,
4059 Address::times_1,
4060 Array<u1>::base_offset_in_bytes()),
4061 JVM_CONSTANT_Class);
4062 __ jcc(Assembler::equal, quicked);
4063
4064 __ push(atos); // save receiver for result, and for GC
4065 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4066 // vm_result_2 has metadata result
4067
4068 #ifndef _LP64
4069 // borrow rdi from locals
4070 __ get_thread(rdi);
4071 __ get_vm_result_2(rax, rdi);
4072 __ restore_locals();
4073 #else
4074 __ get_vm_result_2(rax, r15_thread);
4075 #endif
4076
4077 __ pop_ptr(rdx); // restore receiver
4078 __ verify_oop(rdx);
4079 __ load_klass(rdx, rdx);
4080 __ jmpb(resolved);
4081
4082 // Get superklass in rax and subklass in rdx
4083 __ bind(quicked);
4084 __ load_klass(rdx, rax);
4085 __ movptr(rax, Address(rcx, rbx,
4086 Address::times_ptr, sizeof(ConstantPool)));
4087
4088 __ bind(resolved);
4089
4090 // Generate subtype check. Blows rcx, rdi
4091 // Superklass in rax. Subklass in rdx.
4092 __ gen_subtype_check(rdx, ok_is_subtype);
4093
4094 // Come here on failure
4095 __ xorl(rax, rax);
4096 __ jmpb(done);
4097 // Come here on success
4098 __ bind(ok_is_subtype);
4099 __ movl(rax, 1);
4100
4101 // Collect counts on whether this test sees NULLs a lot or not.
4102 if (ProfileInterpreter) {
4103 __ jmp(done);
4104 __ bind(is_null);
4105 __ profile_null_seen(rcx);
4106 } else {
4107 __ bind(is_null); // same as 'done'
4108 }
4109 __ bind(done);
4110 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4111 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4112 }
4113
4114
4115 //----------------------------------------------------------------------------------------------------
4116 // Breakpoints
4117 void TemplateTable::_breakpoint() {
4118 // Note: We get here even if we are single stepping..
4119 // jbug insists on setting breakpoints at every bytecode
4120 // even if we are in single step mode.
4121
4122 transition(vtos, vtos);
4123
4124 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4125
4126 // get the unpatched byte code
4127 __ get_method(rarg);
4128 __ call_VM(noreg,
4129 CAST_FROM_FN_PTR(address,
4130 InterpreterRuntime::get_original_bytecode_at),
4131 rarg, rbcp);
4132 __ mov(rbx, rax); // why?
4133
4134 // post the breakpoint event
4135 __ get_method(rarg);
4136 __ call_VM(noreg,
4137 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4138 rarg, rbcp);
4139
4140 // complete the execution of original bytecode
4141 __ dispatch_only_normal(vtos);
4142 }
4143
4144 //-----------------------------------------------------------------------------
4145 // Exceptions
4146
4147 void TemplateTable::athrow() {
4148 transition(atos, vtos);
4149 __ null_check(rax);
4150 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4151 }
4152
4153 //-----------------------------------------------------------------------------
4154 // Synchronization
4155 //
4156 // Note: monitorenter & exit are symmetric routines; which is reflected
4157 // in the assembly code structure as well
4158 //
4159 // Stack layout:
4160 //
4161 // [expressions ] <--- rsp = expression stack top
4162 // ..
4163 // [expressions ]
4164 // [monitor entry] <--- monitor block top = expression stack bot
4165 // ..
4166 // [monitor entry]
4167 // [frame data ] <--- monitor block bot
4168 // ...
4169 // [saved rbp ] <--- rbp
4170 void TemplateTable::monitorenter() {
4171 transition(atos, vtos);
4172
4173 // check for NULL object
4174 __ null_check(rax);
4175
4176 const Address monitor_block_top(
4177 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4178 const Address monitor_block_bot(
4179 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4180 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4181
4182 Label allocated;
4183
4184 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4185 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4186 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4187
4188 // initialize entry pointer
4189 __ xorl(rmon, rmon); // points to free slot or NULL
4190
4191 // find a free slot in the monitor block (result in rmon)
4192 {
4193 Label entry, loop, exit;
4194 __ movptr(rtop, monitor_block_top); // points to current entry,
4195 // starting with top-most entry
4196 __ lea(rbot, monitor_block_bot); // points to word before bottom
4197 // of monitor block
4198 __ jmpb(entry);
4199
4200 __ bind(loop);
4201 // check if current entry is used
4202 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4203 // if not used then remember entry in rmon
4204 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4205 // check if current entry is for same object
4206 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4207 // if same object then stop searching
4208 __ jccb(Assembler::equal, exit);
4209 // otherwise advance to next entry
4210 __ addptr(rtop, entry_size);
4211 __ bind(entry);
4212 // check if bottom reached
4213 __ cmpptr(rtop, rbot);
4214 // if not at bottom then check this entry
4215 __ jcc(Assembler::notEqual, loop);
4216 __ bind(exit);
4217 }
4218
4219 __ testptr(rmon, rmon); // check if a slot has been found
4220 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4221
4222 // allocate one if there's no free slot
4223 {
4224 Label entry, loop;
4225 // 1. compute new pointers // rsp: old expression stack top
4226 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4227 __ subptr(rsp, entry_size); // move expression stack top
4228 __ subptr(rmon, entry_size); // move expression stack bottom
4229 __ mov(rtop, rsp); // set start value for copy loop
4230 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4231 __ jmp(entry);
4232 // 2. move expression stack contents
4233 __ bind(loop);
4234 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4235 // word from old location
4236 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4237 __ addptr(rtop, wordSize); // advance to next word
4238 __ bind(entry);
4239 __ cmpptr(rtop, rmon); // check if bottom reached
4240 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4241 // copy next word
4242 }
4243
4244 // call run-time routine
4245 // rmon: points to monitor entry
4246 __ bind(allocated);
4247
4248 // Increment bcp to point to the next bytecode, so exception
4249 // handling for async. exceptions work correctly.
4250 // The object has already been poped from the stack, so the
4251 // expression stack looks correct.
4252 __ increment(rbcp);
4253
4254 // store object
4255 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4256 __ lock_object(rmon);
4257
4258 // check to make sure this monitor doesn't cause stack overflow after locking
4259 __ save_bcp(); // in case of exception
4260 __ generate_stack_overflow_check(0);
4261
4262 // The bcp has already been incremented. Just need to dispatch to
4263 // next instruction.
4264 __ dispatch_next(vtos);
4265 }
4266
4267 void TemplateTable::monitorexit() {
4268 transition(atos, vtos);
4269
4270 // check for NULL object
4271 __ null_check(rax);
4272
4273 const Address monitor_block_top(
4274 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4275 const Address monitor_block_bot(
4276 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4277 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4278
4279 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4280 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4281
4282 Label found;
4283
4284 // find matching slot
4285 {
4286 Label entry, loop;
4287 __ movptr(rtop, monitor_block_top); // points to current entry,
4288 // starting with top-most entry
4289 __ lea(rbot, monitor_block_bot); // points to word before bottom
4290 // of monitor block
4291 __ jmpb(entry);
4292
4293 __ bind(loop);
4294 // check if current entry is for same object
4295 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4296 // if same object then stop searching
4297 __ jcc(Assembler::equal, found);
4298 // otherwise advance to next entry
4299 __ addptr(rtop, entry_size);
4300 __ bind(entry);
4301 // check if bottom reached
4302 __ cmpptr(rtop, rbot);
4303 // if not at bottom then check this entry
4304 __ jcc(Assembler::notEqual, loop);
4305 }
4306
4307 // error handling. Unlocking was not block-structured
4308 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4309 InterpreterRuntime::throw_illegal_monitor_state_exception));
4310 __ should_not_reach_here();
4311
4312 // call run-time routine
4313 __ bind(found);
4314 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4315 __ unlock_object(rtop);
4316 __ pop_ptr(rax); // discard object
4317 }
4318
4319 // Wide instructions
4320 void TemplateTable::wide() {
4321 transition(vtos, vtos);
4322 __ load_unsigned_byte(rbx, at_bcp(1));
4323 ExternalAddress wtable((address)Interpreter::_wentry_point);
4324 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4325 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4326 }
4327
4328 // Multi arrays
4329 void TemplateTable::multianewarray() {
4330 transition(vtos, atos);
4331
4332 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4333 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4334 // last dim is on top of stack; we want address of first one:
4335 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4336 // the latter wordSize to point to the beginning of the array.
4337 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4338 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4339 __ load_unsigned_byte(rbx, at_bcp(3));
4340 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4341 }