1 /*
2 * Copyright (c) 1997, 2017, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.inline.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #define __ _masm->
42
43 // Global Register Names
44 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
45 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
46
47 // Platform-dependent initialization
48 void TemplateTable::pd_initialize() {
49 // No x86 specific initialization
50 }
51
52 // Address Computation: local variables
53 static inline Address iaddress(int n) {
54 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
55 }
56
57 static inline Address laddress(int n) {
58 return iaddress(n + 1);
59 }
60
61 #ifndef _LP64
62 static inline Address haddress(int n) {
63 return iaddress(n + 0);
64 }
65 #endif
66
67 static inline Address faddress(int n) {
68 return iaddress(n);
69 }
70
71 static inline Address daddress(int n) {
72 return laddress(n);
73 }
74
75 static inline Address aaddress(int n) {
76 return iaddress(n);
77 }
78
79 static inline Address iaddress(Register r) {
80 return Address(rlocals, r, Address::times_ptr);
81 }
82
83 static inline Address laddress(Register r) {
84 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
85 }
86
87 #ifndef _LP64
88 static inline Address haddress(Register r) {
89 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
90 }
91 #endif
92
93 static inline Address faddress(Register r) {
94 return iaddress(r);
95 }
96
97 static inline Address daddress(Register r) {
98 return laddress(r);
99 }
100
101 static inline Address aaddress(Register r) {
102 return iaddress(r);
103 }
104
105
106 // expression stack
107 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
108 // data beyond the rsp which is potentially unsafe in an MT environment;
109 // an interrupt may overwrite that data.)
110 static inline Address at_rsp () {
111 return Address(rsp, 0);
112 }
113
114 // At top of Java expression stack which may be different than esp(). It
115 // isn't for category 1 objects.
116 static inline Address at_tos () {
117 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
118 }
119
120 static inline Address at_tos_p1() {
121 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
122 }
123
124 static inline Address at_tos_p2() {
125 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
126 }
127
128 // Condition conversion
129 static Assembler::Condition j_not(TemplateTable::Condition cc) {
130 switch (cc) {
131 case TemplateTable::equal : return Assembler::notEqual;
132 case TemplateTable::not_equal : return Assembler::equal;
133 case TemplateTable::less : return Assembler::greaterEqual;
134 case TemplateTable::less_equal : return Assembler::greater;
135 case TemplateTable::greater : return Assembler::lessEqual;
136 case TemplateTable::greater_equal: return Assembler::less;
137 }
138 ShouldNotReachHere();
139 return Assembler::zero;
140 }
141
142
143
144 // Miscelaneous helper routines
145 // Store an oop (or NULL) at the address described by obj.
146 // If val == noreg this means store a NULL
147
148
149 static void do_oop_store(InterpreterMacroAssembler* _masm,
150 Address obj,
151 Register val,
152 BarrierSet::Name barrier,
153 bool precise) {
154 assert(val == noreg || val == rax, "parameter is just for looks");
155 switch (barrier) {
156 #if INCLUDE_ALL_GCS
157 case BarrierSet::G1SATBCTLogging:
158 {
159 // flatten object address if needed
160 // We do it regardless of precise because we need the registers
161 if (obj.index() == noreg && obj.disp() == 0) {
162 if (obj.base() != rdx) {
163 __ movptr(rdx, obj.base());
164 }
165 } else {
166 __ lea(rdx, obj);
167 }
168
169 Register rtmp = LP64_ONLY(r8) NOT_LP64(rsi);
170 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
171
172 NOT_LP64(__ get_thread(rcx));
173 NOT_LP64(__ save_bcp());
174
175 __ g1_write_barrier_pre(rdx /* obj */,
176 rbx /* pre_val */,
177 rthread /* thread */,
178 rtmp /* tmp */,
179 val != noreg /* tosca_live */,
180 false /* expand_call */);
181 if (val == noreg) {
182 __ store_heap_oop_null(Address(rdx, 0));
183 } else {
184 // G1 barrier needs uncompressed oop for region cross check.
185 Register new_val = val;
186 if (UseCompressedOops) {
187 new_val = rbx;
188 __ movptr(new_val, val);
189 }
190 __ store_heap_oop(Address(rdx, 0), val);
191 __ g1_write_barrier_post(rdx /* store_adr */,
192 new_val /* new_val */,
193 rthread /* thread */,
194 rtmp /* tmp */,
195 rbx /* tmp2 */);
196 }
197 NOT_LP64( __ restore_bcp());
198 }
199 break;
200 #endif // INCLUDE_ALL_GCS
201 case BarrierSet::CardTableForRS:
202 case BarrierSet::CardTableExtension:
203 {
204 if (val == noreg) {
205 __ store_heap_oop_null(obj);
206 } else {
207 __ store_heap_oop(obj, val);
208 // flatten object address if needed
209 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
210 __ store_check(obj.base());
211 } else {
212 __ lea(rdx, obj);
213 __ store_check(rdx);
214 }
215 }
216 }
217 break;
218 case BarrierSet::ModRef:
219 if (val == noreg) {
220 __ store_heap_oop_null(obj);
221 } else {
222 __ store_heap_oop(obj, val);
223 }
224 break;
225 default :
226 ShouldNotReachHere();
227
228 }
229 }
230
231 Address TemplateTable::at_bcp(int offset) {
232 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
233 return Address(rbcp, offset);
234 }
235
236
237 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
238 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
239 int byte_no) {
240 if (!RewriteBytecodes) return;
241 Label L_patch_done;
242
243 switch (bc) {
244 case Bytecodes::_fast_aputfield:
245 case Bytecodes::_fast_bputfield:
246 case Bytecodes::_fast_zputfield:
247 case Bytecodes::_fast_cputfield:
248 case Bytecodes::_fast_dputfield:
249 case Bytecodes::_fast_fputfield:
250 case Bytecodes::_fast_iputfield:
251 case Bytecodes::_fast_lputfield:
252 case Bytecodes::_fast_sputfield:
253 {
254 // We skip bytecode quickening for putfield instructions when
255 // the put_code written to the constant pool cache is zero.
256 // This is required so that every execution of this instruction
257 // calls out to InterpreterRuntime::resolve_get_put to do
258 // additional, required work.
259 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
260 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
261 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
262 __ movl(bc_reg, bc);
263 __ cmpl(temp_reg, (int) 0);
264 __ jcc(Assembler::zero, L_patch_done); // don't patch
265 }
266 break;
267 default:
268 assert(byte_no == -1, "sanity");
269 // the pair bytecodes have already done the load.
270 if (load_bc_into_bc_reg) {
271 __ movl(bc_reg, bc);
272 }
273 }
274
275 if (JvmtiExport::can_post_breakpoint()) {
276 Label L_fast_patch;
277 // if a breakpoint is present we can't rewrite the stream directly
278 __ movzbl(temp_reg, at_bcp(0));
279 __ cmpl(temp_reg, Bytecodes::_breakpoint);
280 __ jcc(Assembler::notEqual, L_fast_patch);
281 __ get_method(temp_reg);
282 // Let breakpoint table handling rewrite to quicker bytecode
283 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
284 #ifndef ASSERT
285 __ jmpb(L_patch_done);
286 #else
287 __ jmp(L_patch_done);
288 #endif
289 __ bind(L_fast_patch);
290 }
291
292 #ifdef ASSERT
293 Label L_okay;
294 __ load_unsigned_byte(temp_reg, at_bcp(0));
295 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
296 __ jcc(Assembler::equal, L_okay);
297 __ cmpl(temp_reg, bc_reg);
298 __ jcc(Assembler::equal, L_okay);
299 __ stop("patching the wrong bytecode");
300 __ bind(L_okay);
301 #endif
302
303 // patch bytecode
304 __ movb(at_bcp(0), bc_reg);
305 __ bind(L_patch_done);
306 }
307 // Individual instructions
308
309
310 void TemplateTable::nop() {
311 transition(vtos, vtos);
312 // nothing to do
313 }
314
315 void TemplateTable::shouldnotreachhere() {
316 transition(vtos, vtos);
317 __ stop("shouldnotreachhere bytecode");
318 }
319
320 void TemplateTable::aconst_null() {
321 transition(vtos, atos);
322 __ xorl(rax, rax);
323 }
324
325 void TemplateTable::iconst(int value) {
326 transition(vtos, itos);
327 if (value == 0) {
328 __ xorl(rax, rax);
329 } else {
330 __ movl(rax, value);
331 }
332 }
333
334 void TemplateTable::lconst(int value) {
335 transition(vtos, ltos);
336 if (value == 0) {
337 __ xorl(rax, rax);
338 } else {
339 __ movl(rax, value);
340 }
341 #ifndef _LP64
342 assert(value >= 0, "check this code");
343 __ xorptr(rdx, rdx);
344 #endif
345 }
346
347
348
349 void TemplateTable::fconst(int value) {
350 transition(vtos, ftos);
351 if (UseSSE >= 1) {
352 static float one = 1.0f, two = 2.0f;
353 switch (value) {
354 case 0:
355 __ xorps(xmm0, xmm0);
356 break;
357 case 1:
358 __ movflt(xmm0, ExternalAddress((address) &one));
359 break;
360 case 2:
361 __ movflt(xmm0, ExternalAddress((address) &two));
362 break;
363 default:
364 ShouldNotReachHere();
365 break;
366 }
367 } else {
368 #ifdef _LP64
369 ShouldNotReachHere();
370 #else
371 if (value == 0) { __ fldz();
372 } else if (value == 1) { __ fld1();
373 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
374 } else { ShouldNotReachHere();
375 }
376 #endif // _LP64
377 }
378 }
379
380 void TemplateTable::dconst(int value) {
381 transition(vtos, dtos);
382 if (UseSSE >= 2) {
383 static double one = 1.0;
384 switch (value) {
385 case 0:
386 __ xorpd(xmm0, xmm0);
387 break;
388 case 1:
389 __ movdbl(xmm0, ExternalAddress((address) &one));
390 break;
391 default:
392 ShouldNotReachHere();
393 break;
394 }
395 } else {
396 #ifdef _LP64
397 ShouldNotReachHere();
398 #else
399 if (value == 0) { __ fldz();
400 } else if (value == 1) { __ fld1();
401 } else { ShouldNotReachHere();
402 }
403 #endif
404 }
405 }
406
407 void TemplateTable::bipush() {
408 transition(vtos, itos);
409 __ load_signed_byte(rax, at_bcp(1));
410 }
411
412 void TemplateTable::sipush() {
413 transition(vtos, itos);
414 __ load_unsigned_short(rax, at_bcp(1));
415 __ bswapl(rax);
416 __ sarl(rax, 16);
417 }
418
419 void TemplateTable::ldc(bool wide) {
420 transition(vtos, vtos);
421 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
422 Label call_ldc, notFloat, notClass, notInt, Done;
423
424 if (wide) {
425 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
426 } else {
427 __ load_unsigned_byte(rbx, at_bcp(1));
428 }
429
430 __ get_cpool_and_tags(rcx, rax);
431 const int base_offset = ConstantPool::header_size() * wordSize;
432 const int tags_offset = Array<u1>::base_offset_in_bytes();
433
434 // get type
435 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
436
437 // unresolved class - get the resolved class
438 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
439 __ jccb(Assembler::equal, call_ldc);
440
441 // unresolved class in error state - call into runtime to throw the error
442 // from the first resolution attempt
443 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
444 __ jccb(Assembler::equal, call_ldc);
445
446 // resolved class - need to call vm to get java mirror of the class
447 __ cmpl(rdx, JVM_CONSTANT_Class);
448 __ jcc(Assembler::notEqual, notClass);
449
450 __ bind(call_ldc);
451
452 __ movl(rarg, wide);
453 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
454
455 __ push(atos);
456 __ jmp(Done);
457
458 __ bind(notClass);
459 __ cmpl(rdx, JVM_CONSTANT_Float);
460 __ jccb(Assembler::notEqual, notFloat);
461
462 // ftos
463 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
464 __ push(ftos);
465 __ jmp(Done);
466
467 __ bind(notFloat);
468 __ cmpl(rdx, JVM_CONSTANT_Integer);
469 __ jccb(Assembler::notEqual, notInt);
470
471 // itos
472 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
473 __ push(itos);
474 __ jmp(Done);
475
476 // assume the tag is for condy; if not, the VM runtime will tell us
477 __ bind(notInt);
478 condy_helper(Done);
479
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 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
490 int index_size = wide ? sizeof(u2) : sizeof(u1);
491
492 Label resolved;
493
494 // We are resolved if the resolved reference cache entry contains a
495 // non-null object (String, MethodType, etc.)
496 assert_different_registers(result, tmp);
497 __ get_cache_index_at_bcp(tmp, 1, index_size);
498 __ load_resolved_reference_at_index(result, tmp);
499 __ testptr(result, result);
500 __ jcc(Assembler::notZero, resolved);
501
502 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
503
504 // first time invocation - must resolve first
505 __ movl(rarg, (int)bytecode());
506 __ call_VM(result, entry, rarg);
507 __ bind(resolved);
508
509 { // Check for the null sentinel.
510 // If we just called the VM, that already did the mapping for us,
511 // but it's harmless to retry.
512 Label notNull;
513 ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
514 __ movptr(tmp, null_sentinel);
515 __ cmpptr(tmp, result);
516 __ jccb(Assembler::notEqual, notNull);
517 __ xorptr(result, result); // NULL object reference
518 __ bind(notNull);
519 }
520
521 if (VerifyOops) {
522 __ verify_oop(result);
523 }
524 }
525
526 void TemplateTable::ldc2_w() {
527 transition(vtos, vtos);
528 Label notDouble, notLong, Done;
529 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
530
531 __ get_cpool_and_tags(rcx, rax);
532 const int base_offset = ConstantPool::header_size() * wordSize;
533 const int tags_offset = Array<u1>::base_offset_in_bytes();
534
535 // get type
536 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
537 __ cmpl(rdx, JVM_CONSTANT_Double);
538 __ jccb(Assembler::notEqual, notDouble);
539
540 // dtos
541 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
542 __ push(dtos);
543
544 __ jmp(Done);
545 __ bind(notDouble);
546 __ cmpl(rdx, JVM_CONSTANT_Long);
547 __ jccb(Assembler::notEqual, notLong);
548
549 // ltos
550 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
551 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
552 __ push(ltos);
553 __ jmp(Done);
554
555 __ bind(notLong);
556 condy_helper(Done);
557
558 __ bind(Done);
559 }
560
561 void TemplateTable::condy_helper(Label& Done) {
562 const Register obj = rax;
563 const Register off = rbx;
564 const Register flags = rcx;
565 const Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
566 __ movl(rarg, (int)bytecode());
567 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
568 #ifndef _LP64
569 // borrow rdi from locals
570 __ get_thread(rdi);
571 __ get_vm_result_2(flags, rdi);
572 __ restore_locals();
573 #else
574 __ get_vm_result_2(flags, r15_thread);
575 #endif
576 // VMr = obj = base address to find primitive value to push
577 // VMr2 = flags = (tos, off) using format of CPCE::_flags
578 __ movl(off, flags);
579 __ andl(off, ConstantPoolCacheEntry::field_index_mask);
580 const Address field(obj, off, Address::times_1, 0*wordSize);
581
582 // What sort of thing are we loading?
583 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
584 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
585
586 switch (bytecode()) {
587 case Bytecodes::_ldc:
588 case Bytecodes::_ldc_w:
589 {
590 // tos in (itos, ftos, stos, btos, ctos, ztos)
591 Label notInt, notFloat, notShort, notByte, notChar, notBool;
592 __ cmpl(flags, itos);
593 __ jcc(Assembler::notEqual, notInt);
594 // itos
595 __ movl(rax, field);
596 __ push(itos);
597 __ jmp(Done);
598
599 __ bind(notInt);
600 __ cmpl(flags, ftos);
601 __ jcc(Assembler::notEqual, notFloat);
602 // ftos
603 __ load_float(field);
604 __ push(ftos);
605 __ jmp(Done);
606
607 __ bind(notFloat);
608 __ cmpl(flags, stos);
609 __ jcc(Assembler::notEqual, notShort);
610 // stos
611 __ load_signed_short(rax, field);
612 __ push(stos);
613 __ jmp(Done);
614
615 __ bind(notShort);
616 __ cmpl(flags, btos);
617 __ jcc(Assembler::notEqual, notByte);
618 // btos
619 __ load_signed_byte(rax, field);
620 __ push(btos);
621 __ jmp(Done);
622
623 __ bind(notByte);
624 __ cmpl(flags, ctos);
625 __ jcc(Assembler::notEqual, notChar);
626 // ctos
627 __ load_unsigned_short(rax, field);
628 __ push(ctos);
629 __ jmp(Done);
630
631 __ bind(notChar);
632 __ cmpl(flags, ztos);
633 __ jcc(Assembler::notEqual, notBool);
634 // ztos
635 __ load_signed_byte(rax, field);
636 __ push(ztos);
637 __ jmp(Done);
638
639 __ bind(notBool);
640 break;
641 }
642
643 case Bytecodes::_ldc2_w:
644 {
645 Label notLong, notDouble;
646 __ cmpl(flags, ltos);
647 __ jcc(Assembler::notEqual, notLong);
648 // ltos
649 __ movptr(rax, field);
650 NOT_LP64(__ movptr(rdx, field.plus_disp(4)));
651 __ push(ltos);
652 __ jmp(Done);
653
654 __ bind(notLong);
655 __ cmpl(flags, dtos);
656 __ jcc(Assembler::notEqual, notDouble);
657 // dtos
658 __ load_double(field);
659 __ push(dtos);
660 __ jmp(Done);
661
662 __ bind(notDouble);
663 break;
664 }
665
666 default:
667 ShouldNotReachHere();
668 }
669
670 __ stop("bad ldc/condy");
671 }
672
673 void TemplateTable::locals_index(Register reg, int offset) {
674 __ load_unsigned_byte(reg, at_bcp(offset));
675 __ negptr(reg);
676 }
677
678 void TemplateTable::iload() {
679 iload_internal();
680 }
681
682 void TemplateTable::nofast_iload() {
683 iload_internal(may_not_rewrite);
684 }
685
686 void TemplateTable::iload_internal(RewriteControl rc) {
687 transition(vtos, itos);
688 if (RewriteFrequentPairs && rc == may_rewrite) {
689 Label rewrite, done;
690 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
691 LP64_ONLY(assert(rbx != bc, "register damaged"));
692
693 // get next byte
694 __ load_unsigned_byte(rbx,
695 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
696 // if _iload, wait to rewrite to iload2. We only want to rewrite the
697 // last two iloads in a pair. Comparing against fast_iload means that
698 // the next bytecode is neither an iload or a caload, and therefore
699 // an iload pair.
700 __ cmpl(rbx, Bytecodes::_iload);
701 __ jcc(Assembler::equal, done);
702
703 __ cmpl(rbx, Bytecodes::_fast_iload);
704 __ movl(bc, Bytecodes::_fast_iload2);
705
706 __ jccb(Assembler::equal, rewrite);
707
708 // if _caload, rewrite to fast_icaload
709 __ cmpl(rbx, Bytecodes::_caload);
710 __ movl(bc, Bytecodes::_fast_icaload);
711 __ jccb(Assembler::equal, rewrite);
712
713 // rewrite so iload doesn't check again.
714 __ movl(bc, Bytecodes::_fast_iload);
715
716 // rewrite
717 // bc: fast bytecode
718 __ bind(rewrite);
719 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
720 __ bind(done);
721 }
722
723 // Get the local value into tos
724 locals_index(rbx);
725 __ movl(rax, iaddress(rbx));
726 }
727
728 void TemplateTable::fast_iload2() {
729 transition(vtos, itos);
730 locals_index(rbx);
731 __ movl(rax, iaddress(rbx));
732 __ push(itos);
733 locals_index(rbx, 3);
734 __ movl(rax, iaddress(rbx));
735 }
736
737 void TemplateTable::fast_iload() {
738 transition(vtos, itos);
739 locals_index(rbx);
740 __ movl(rax, iaddress(rbx));
741 }
742
743 void TemplateTable::lload() {
744 transition(vtos, ltos);
745 locals_index(rbx);
746 __ movptr(rax, laddress(rbx));
747 NOT_LP64(__ movl(rdx, haddress(rbx)));
748 }
749
750 void TemplateTable::fload() {
751 transition(vtos, ftos);
752 locals_index(rbx);
753 __ load_float(faddress(rbx));
754 }
755
756 void TemplateTable::dload() {
757 transition(vtos, dtos);
758 locals_index(rbx);
759 __ load_double(daddress(rbx));
760 }
761
762 void TemplateTable::aload() {
763 transition(vtos, atos);
764 locals_index(rbx);
765 __ movptr(rax, aaddress(rbx));
766 }
767
768 void TemplateTable::locals_index_wide(Register reg) {
769 __ load_unsigned_short(reg, at_bcp(2));
770 __ bswapl(reg);
771 __ shrl(reg, 16);
772 __ negptr(reg);
773 }
774
775 void TemplateTable::wide_iload() {
776 transition(vtos, itos);
777 locals_index_wide(rbx);
778 __ movl(rax, iaddress(rbx));
779 }
780
781 void TemplateTable::wide_lload() {
782 transition(vtos, ltos);
783 locals_index_wide(rbx);
784 __ movptr(rax, laddress(rbx));
785 NOT_LP64(__ movl(rdx, haddress(rbx)));
786 }
787
788 void TemplateTable::wide_fload() {
789 transition(vtos, ftos);
790 locals_index_wide(rbx);
791 __ load_float(faddress(rbx));
792 }
793
794 void TemplateTable::wide_dload() {
795 transition(vtos, dtos);
796 locals_index_wide(rbx);
797 __ load_double(daddress(rbx));
798 }
799
800 void TemplateTable::wide_aload() {
801 transition(vtos, atos);
802 locals_index_wide(rbx);
803 __ movptr(rax, aaddress(rbx));
804 }
805
806 void TemplateTable::index_check(Register array, Register index) {
807 // Pop ptr into array
808 __ pop_ptr(array);
809 index_check_without_pop(array, index);
810 }
811
812 void TemplateTable::index_check_without_pop(Register array, Register index) {
813 // destroys rbx
814 // check array
815 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
816 // sign extend index for use by indexed load
817 __ movl2ptr(index, index);
818 // check index
819 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
820 if (index != rbx) {
821 // ??? convention: move aberrant index into rbx for exception message
822 assert(rbx != array, "different registers");
823 __ movl(rbx, index);
824 }
825 __ jump_cc(Assembler::aboveEqual,
826 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
827 }
828
829
830 void TemplateTable::iaload() {
831 transition(itos, itos);
832 // rax: index
833 // rdx: array
834 index_check(rdx, rax); // kills rbx
835 __ movl(rax, Address(rdx, rax,
836 Address::times_4,
837 arrayOopDesc::base_offset_in_bytes(T_INT)));
838 }
839
840 void TemplateTable::laload() {
841 transition(itos, ltos);
842 // rax: index
843 // rdx: array
844 index_check(rdx, rax); // kills rbx
845 NOT_LP64(__ mov(rbx, rax));
846 // rbx,: index
847 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
848 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
849 }
850
851
852
853 void TemplateTable::faload() {
854 transition(itos, ftos);
855 // rax: index
856 // rdx: array
857 index_check(rdx, rax); // kills rbx
858 __ load_float(Address(rdx, rax,
859 Address::times_4,
860 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
861 }
862
863 void TemplateTable::daload() {
864 transition(itos, dtos);
865 // rax: index
866 // rdx: array
867 index_check(rdx, rax); // kills rbx
868 __ load_double(Address(rdx, rax,
869 Address::times_8,
870 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
871 }
872
873 void TemplateTable::aaload() {
874 transition(itos, atos);
875 // rax: index
876 // rdx: array
877 index_check(rdx, rax); // kills rbx
878 __ load_heap_oop(rax, Address(rdx, rax,
879 UseCompressedOops ? Address::times_4 : Address::times_ptr,
880 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
881 }
882
883 void TemplateTable::baload() {
884 transition(itos, itos);
885 // rax: index
886 // rdx: array
887 index_check(rdx, rax); // kills rbx
888 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
889 }
890
891 void TemplateTable::caload() {
892 transition(itos, itos);
893 // rax: index
894 // rdx: array
895 index_check(rdx, rax); // kills rbx
896 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
897 }
898
899 // iload followed by caload frequent pair
900 void TemplateTable::fast_icaload() {
901 transition(vtos, itos);
902 // load index out of locals
903 locals_index(rbx);
904 __ movl(rax, iaddress(rbx));
905
906 // rax: index
907 // rdx: array
908 index_check(rdx, rax); // kills rbx
909 __ load_unsigned_short(rax,
910 Address(rdx, rax,
911 Address::times_2,
912 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
913 }
914
915
916 void TemplateTable::saload() {
917 transition(itos, itos);
918 // rax: index
919 // rdx: array
920 index_check(rdx, rax); // kills rbx
921 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
922 }
923
924 void TemplateTable::iload(int n) {
925 transition(vtos, itos);
926 __ movl(rax, iaddress(n));
927 }
928
929 void TemplateTable::lload(int n) {
930 transition(vtos, ltos);
931 __ movptr(rax, laddress(n));
932 NOT_LP64(__ movptr(rdx, haddress(n)));
933 }
934
935 void TemplateTable::fload(int n) {
936 transition(vtos, ftos);
937 __ load_float(faddress(n));
938 }
939
940 void TemplateTable::dload(int n) {
941 transition(vtos, dtos);
942 __ load_double(daddress(n));
943 }
944
945 void TemplateTable::aload(int n) {
946 transition(vtos, atos);
947 __ movptr(rax, aaddress(n));
948 }
949
950 void TemplateTable::aload_0() {
951 aload_0_internal();
952 }
953
954 void TemplateTable::nofast_aload_0() {
955 aload_0_internal(may_not_rewrite);
956 }
957
958 void TemplateTable::aload_0_internal(RewriteControl rc) {
959 transition(vtos, atos);
960 // According to bytecode histograms, the pairs:
961 //
962 // _aload_0, _fast_igetfield
963 // _aload_0, _fast_agetfield
964 // _aload_0, _fast_fgetfield
965 //
966 // occur frequently. If RewriteFrequentPairs is set, the (slow)
967 // _aload_0 bytecode checks if the next bytecode is either
968 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
969 // rewrites the current bytecode into a pair bytecode; otherwise it
970 // rewrites the current bytecode into _fast_aload_0 that doesn't do
971 // the pair check anymore.
972 //
973 // Note: If the next bytecode is _getfield, the rewrite must be
974 // delayed, otherwise we may miss an opportunity for a pair.
975 //
976 // Also rewrite frequent pairs
977 // aload_0, aload_1
978 // aload_0, iload_1
979 // These bytecodes with a small amount of code are most profitable
980 // to rewrite
981 if (RewriteFrequentPairs && rc == may_rewrite) {
982 Label rewrite, done;
983
984 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
985 LP64_ONLY(assert(rbx != bc, "register damaged"));
986
987 // get next byte
988 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
989
990 // if _getfield then wait with rewrite
991 __ cmpl(rbx, Bytecodes::_getfield);
992 __ jcc(Assembler::equal, done);
993
994 // if _igetfield then rewrite to _fast_iaccess_0
995 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
996 __ cmpl(rbx, Bytecodes::_fast_igetfield);
997 __ movl(bc, Bytecodes::_fast_iaccess_0);
998 __ jccb(Assembler::equal, rewrite);
999
1000 // if _agetfield then rewrite to _fast_aaccess_0
1001 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
1002 __ cmpl(rbx, Bytecodes::_fast_agetfield);
1003 __ movl(bc, Bytecodes::_fast_aaccess_0);
1004 __ jccb(Assembler::equal, rewrite);
1005
1006 // if _fgetfield then rewrite to _fast_faccess_0
1007 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
1008 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
1009 __ movl(bc, Bytecodes::_fast_faccess_0);
1010 __ jccb(Assembler::equal, rewrite);
1011
1012 // else rewrite to _fast_aload0
1013 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
1014 __ movl(bc, Bytecodes::_fast_aload_0);
1015
1016 // rewrite
1017 // bc: fast bytecode
1018 __ bind(rewrite);
1019 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
1020
1021 __ bind(done);
1022 }
1023
1024 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
1025 aload(0);
1026 }
1027
1028 void TemplateTable::istore() {
1029 transition(itos, vtos);
1030 locals_index(rbx);
1031 __ movl(iaddress(rbx), rax);
1032 }
1033
1034
1035 void TemplateTable::lstore() {
1036 transition(ltos, vtos);
1037 locals_index(rbx);
1038 __ movptr(laddress(rbx), rax);
1039 NOT_LP64(__ movptr(haddress(rbx), rdx));
1040 }
1041
1042 void TemplateTable::fstore() {
1043 transition(ftos, vtos);
1044 locals_index(rbx);
1045 __ store_float(faddress(rbx));
1046 }
1047
1048 void TemplateTable::dstore() {
1049 transition(dtos, vtos);
1050 locals_index(rbx);
1051 __ store_double(daddress(rbx));
1052 }
1053
1054 void TemplateTable::astore() {
1055 transition(vtos, vtos);
1056 __ pop_ptr(rax);
1057 locals_index(rbx);
1058 __ movptr(aaddress(rbx), rax);
1059 }
1060
1061 void TemplateTable::wide_istore() {
1062 transition(vtos, vtos);
1063 __ pop_i();
1064 locals_index_wide(rbx);
1065 __ movl(iaddress(rbx), rax);
1066 }
1067
1068 void TemplateTable::wide_lstore() {
1069 transition(vtos, vtos);
1070 NOT_LP64(__ pop_l(rax, rdx));
1071 LP64_ONLY(__ pop_l());
1072 locals_index_wide(rbx);
1073 __ movptr(laddress(rbx), rax);
1074 NOT_LP64(__ movl(haddress(rbx), rdx));
1075 }
1076
1077 void TemplateTable::wide_fstore() {
1078 #ifdef _LP64
1079 transition(vtos, vtos);
1080 __ pop_f(xmm0);
1081 locals_index_wide(rbx);
1082 __ movflt(faddress(rbx), xmm0);
1083 #else
1084 wide_istore();
1085 #endif
1086 }
1087
1088 void TemplateTable::wide_dstore() {
1089 #ifdef _LP64
1090 transition(vtos, vtos);
1091 __ pop_d(xmm0);
1092 locals_index_wide(rbx);
1093 __ movdbl(daddress(rbx), xmm0);
1094 #else
1095 wide_lstore();
1096 #endif
1097 }
1098
1099 void TemplateTable::wide_astore() {
1100 transition(vtos, vtos);
1101 __ pop_ptr(rax);
1102 locals_index_wide(rbx);
1103 __ movptr(aaddress(rbx), rax);
1104 }
1105
1106 void TemplateTable::iastore() {
1107 transition(itos, vtos);
1108 __ pop_i(rbx);
1109 // rax: value
1110 // rbx: index
1111 // rdx: array
1112 index_check(rdx, rbx); // prefer index in rbx
1113 __ movl(Address(rdx, rbx,
1114 Address::times_4,
1115 arrayOopDesc::base_offset_in_bytes(T_INT)),
1116 rax);
1117 }
1118
1119 void TemplateTable::lastore() {
1120 transition(ltos, vtos);
1121 __ pop_i(rbx);
1122 // rax,: low(value)
1123 // rcx: array
1124 // rdx: high(value)
1125 index_check(rcx, rbx); // prefer index in rbx,
1126 // rbx,: index
1127 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1128 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1129 }
1130
1131
1132 void TemplateTable::fastore() {
1133 transition(ftos, vtos);
1134 __ pop_i(rbx);
1135 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1136 // rbx: index
1137 // rdx: array
1138 index_check(rdx, rbx); // prefer index in rbx
1139 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1140 }
1141
1142 void TemplateTable::dastore() {
1143 transition(dtos, vtos);
1144 __ pop_i(rbx);
1145 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1146 // rbx: index
1147 // rdx: array
1148 index_check(rdx, rbx); // prefer index in rbx
1149 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1150 }
1151
1152 void TemplateTable::aastore() {
1153 Label is_null, ok_is_subtype, done;
1154 transition(vtos, vtos);
1155 // stack: ..., array, index, value
1156 __ movptr(rax, at_tos()); // value
1157 __ movl(rcx, at_tos_p1()); // index
1158 __ movptr(rdx, at_tos_p2()); // array
1159
1160 Address element_address(rdx, rcx,
1161 UseCompressedOops? Address::times_4 : Address::times_ptr,
1162 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1163
1164 index_check_without_pop(rdx, rcx); // kills rbx
1165 __ testptr(rax, rax);
1166 __ jcc(Assembler::zero, is_null);
1167
1168 // Move subklass into rbx
1169 __ load_klass(rbx, rax);
1170 // Move superklass into rax
1171 __ load_klass(rax, rdx);
1172 __ movptr(rax, Address(rax,
1173 ObjArrayKlass::element_klass_offset()));
1174 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1175 __ lea(rdx, element_address);
1176
1177 // Generate subtype check. Blows rcx, rdi
1178 // Superklass in rax. Subklass in rbx.
1179 __ gen_subtype_check(rbx, ok_is_subtype);
1180
1181 // Come here on failure
1182 // object is at TOS
1183 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1184
1185 // Come here on success
1186 __ bind(ok_is_subtype);
1187
1188 // Get the value we will store
1189 __ movptr(rax, at_tos());
1190 // Now store using the appropriate barrier
1191 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1192 __ jmp(done);
1193
1194 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1195 __ bind(is_null);
1196 __ profile_null_seen(rbx);
1197
1198 // Store a NULL
1199 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1200
1201 // Pop stack arguments
1202 __ bind(done);
1203 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1204 }
1205
1206 void TemplateTable::bastore() {
1207 transition(itos, vtos);
1208 __ pop_i(rbx);
1209 // rax: value
1210 // rbx: index
1211 // rdx: array
1212 index_check(rdx, rbx); // prefer index in rbx
1213 // Need to check whether array is boolean or byte
1214 // since both types share the bastore bytecode.
1215 __ load_klass(rcx, rdx);
1216 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1217 int diffbit = Klass::layout_helper_boolean_diffbit();
1218 __ testl(rcx, diffbit);
1219 Label L_skip;
1220 __ jccb(Assembler::zero, L_skip);
1221 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1222 __ bind(L_skip);
1223 __ movb(Address(rdx, rbx,
1224 Address::times_1,
1225 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1226 rax);
1227 }
1228
1229 void TemplateTable::castore() {
1230 transition(itos, vtos);
1231 __ pop_i(rbx);
1232 // rax: value
1233 // rbx: index
1234 // rdx: array
1235 index_check(rdx, rbx); // prefer index in rbx
1236 __ movw(Address(rdx, rbx,
1237 Address::times_2,
1238 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1239 rax);
1240 }
1241
1242
1243 void TemplateTable::sastore() {
1244 castore();
1245 }
1246
1247 void TemplateTable::istore(int n) {
1248 transition(itos, vtos);
1249 __ movl(iaddress(n), rax);
1250 }
1251
1252 void TemplateTable::lstore(int n) {
1253 transition(ltos, vtos);
1254 __ movptr(laddress(n), rax);
1255 NOT_LP64(__ movptr(haddress(n), rdx));
1256 }
1257
1258 void TemplateTable::fstore(int n) {
1259 transition(ftos, vtos);
1260 __ store_float(faddress(n));
1261 }
1262
1263 void TemplateTable::dstore(int n) {
1264 transition(dtos, vtos);
1265 __ store_double(daddress(n));
1266 }
1267
1268
1269 void TemplateTable::astore(int n) {
1270 transition(vtos, vtos);
1271 __ pop_ptr(rax);
1272 __ movptr(aaddress(n), rax);
1273 }
1274
1275 void TemplateTable::pop() {
1276 transition(vtos, vtos);
1277 __ addptr(rsp, Interpreter::stackElementSize);
1278 }
1279
1280 void TemplateTable::pop2() {
1281 transition(vtos, vtos);
1282 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1283 }
1284
1285
1286 void TemplateTable::dup() {
1287 transition(vtos, vtos);
1288 __ load_ptr(0, rax);
1289 __ push_ptr(rax);
1290 // stack: ..., a, a
1291 }
1292
1293 void TemplateTable::dup_x1() {
1294 transition(vtos, vtos);
1295 // stack: ..., a, b
1296 __ load_ptr( 0, rax); // load b
1297 __ load_ptr( 1, rcx); // load a
1298 __ store_ptr(1, rax); // store b
1299 __ store_ptr(0, rcx); // store a
1300 __ push_ptr(rax); // push b
1301 // stack: ..., b, a, b
1302 }
1303
1304 void TemplateTable::dup_x2() {
1305 transition(vtos, vtos);
1306 // stack: ..., a, b, c
1307 __ load_ptr( 0, rax); // load c
1308 __ load_ptr( 2, rcx); // load a
1309 __ store_ptr(2, rax); // store c in a
1310 __ push_ptr(rax); // push c
1311 // stack: ..., c, b, c, c
1312 __ load_ptr( 2, rax); // load b
1313 __ store_ptr(2, rcx); // store a in b
1314 // stack: ..., c, a, c, c
1315 __ store_ptr(1, rax); // store b in c
1316 // stack: ..., c, a, b, c
1317 }
1318
1319 void TemplateTable::dup2() {
1320 transition(vtos, vtos);
1321 // stack: ..., a, b
1322 __ load_ptr(1, rax); // load a
1323 __ push_ptr(rax); // push a
1324 __ load_ptr(1, rax); // load b
1325 __ push_ptr(rax); // push b
1326 // stack: ..., a, b, a, b
1327 }
1328
1329
1330 void TemplateTable::dup2_x1() {
1331 transition(vtos, vtos);
1332 // stack: ..., a, b, c
1333 __ load_ptr( 0, rcx); // load c
1334 __ load_ptr( 1, rax); // load b
1335 __ push_ptr(rax); // push b
1336 __ push_ptr(rcx); // push c
1337 // stack: ..., a, b, c, b, c
1338 __ store_ptr(3, rcx); // store c in b
1339 // stack: ..., a, c, c, b, c
1340 __ load_ptr( 4, rcx); // load a
1341 __ store_ptr(2, rcx); // store a in 2nd c
1342 // stack: ..., a, c, a, b, c
1343 __ store_ptr(4, rax); // store b in a
1344 // stack: ..., b, c, a, b, c
1345 }
1346
1347 void TemplateTable::dup2_x2() {
1348 transition(vtos, vtos);
1349 // stack: ..., a, b, c, d
1350 __ load_ptr( 0, rcx); // load d
1351 __ load_ptr( 1, rax); // load c
1352 __ push_ptr(rax); // push c
1353 __ push_ptr(rcx); // push d
1354 // stack: ..., a, b, c, d, c, d
1355 __ load_ptr( 4, rax); // load b
1356 __ store_ptr(2, rax); // store b in d
1357 __ store_ptr(4, rcx); // store d in b
1358 // stack: ..., a, d, c, b, c, d
1359 __ load_ptr( 5, rcx); // load a
1360 __ load_ptr( 3, rax); // load c
1361 __ store_ptr(3, rcx); // store a in c
1362 __ store_ptr(5, rax); // store c in a
1363 // stack: ..., c, d, a, b, c, d
1364 }
1365
1366 void TemplateTable::swap() {
1367 transition(vtos, vtos);
1368 // stack: ..., a, b
1369 __ load_ptr( 1, rcx); // load a
1370 __ load_ptr( 0, rax); // load b
1371 __ store_ptr(0, rcx); // store a in b
1372 __ store_ptr(1, rax); // store b in a
1373 // stack: ..., b, a
1374 }
1375
1376 void TemplateTable::iop2(Operation op) {
1377 transition(itos, itos);
1378 switch (op) {
1379 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1380 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1381 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1382 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1383 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1384 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1385 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1386 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1387 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1388 default : ShouldNotReachHere();
1389 }
1390 }
1391
1392 void TemplateTable::lop2(Operation op) {
1393 transition(ltos, ltos);
1394 #ifdef _LP64
1395 switch (op) {
1396 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1397 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1398 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1399 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1400 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1401 default : ShouldNotReachHere();
1402 }
1403 #else
1404 __ pop_l(rbx, rcx);
1405 switch (op) {
1406 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1407 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1408 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1409 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1410 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1411 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1412 default : ShouldNotReachHere();
1413 }
1414 #endif
1415 }
1416
1417 void TemplateTable::idiv() {
1418 transition(itos, itos);
1419 __ movl(rcx, rax);
1420 __ pop_i(rax);
1421 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1422 // they are not equal, one could do a normal division (no correction
1423 // needed), which may speed up this implementation for the common case.
1424 // (see also JVM spec., p.243 & p.271)
1425 __ corrected_idivl(rcx);
1426 }
1427
1428 void TemplateTable::irem() {
1429 transition(itos, itos);
1430 __ movl(rcx, rax);
1431 __ pop_i(rax);
1432 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1433 // they are not equal, one could do a normal division (no correction
1434 // needed), which may speed up this implementation for the common case.
1435 // (see also JVM spec., p.243 & p.271)
1436 __ corrected_idivl(rcx);
1437 __ movl(rax, rdx);
1438 }
1439
1440 void TemplateTable::lmul() {
1441 transition(ltos, ltos);
1442 #ifdef _LP64
1443 __ pop_l(rdx);
1444 __ imulq(rax, rdx);
1445 #else
1446 __ pop_l(rbx, rcx);
1447 __ push(rcx); __ push(rbx);
1448 __ push(rdx); __ push(rax);
1449 __ lmul(2 * wordSize, 0);
1450 __ addptr(rsp, 4 * wordSize); // take off temporaries
1451 #endif
1452 }
1453
1454 void TemplateTable::ldiv() {
1455 transition(ltos, ltos);
1456 #ifdef _LP64
1457 __ mov(rcx, rax);
1458 __ pop_l(rax);
1459 // generate explicit div0 check
1460 __ testq(rcx, rcx);
1461 __ jump_cc(Assembler::zero,
1462 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1463 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1464 // they are not equal, one could do a normal division (no correction
1465 // needed), which may speed up this implementation for the common case.
1466 // (see also JVM spec., p.243 & p.271)
1467 __ corrected_idivq(rcx); // kills rbx
1468 #else
1469 __ pop_l(rbx, rcx);
1470 __ push(rcx); __ push(rbx);
1471 __ push(rdx); __ push(rax);
1472 // check if y = 0
1473 __ orl(rax, rdx);
1474 __ jump_cc(Assembler::zero,
1475 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1476 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1477 __ addptr(rsp, 4 * wordSize); // take off temporaries
1478 #endif
1479 }
1480
1481 void TemplateTable::lrem() {
1482 transition(ltos, ltos);
1483 #ifdef _LP64
1484 __ mov(rcx, rax);
1485 __ pop_l(rax);
1486 __ testq(rcx, rcx);
1487 __ jump_cc(Assembler::zero,
1488 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1489 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1490 // they are not equal, one could do a normal division (no correction
1491 // needed), which may speed up this implementation for the common case.
1492 // (see also JVM spec., p.243 & p.271)
1493 __ corrected_idivq(rcx); // kills rbx
1494 __ mov(rax, rdx);
1495 #else
1496 __ pop_l(rbx, rcx);
1497 __ push(rcx); __ push(rbx);
1498 __ push(rdx); __ push(rax);
1499 // check if y = 0
1500 __ orl(rax, rdx);
1501 __ jump_cc(Assembler::zero,
1502 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1503 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1504 __ addptr(rsp, 4 * wordSize);
1505 #endif
1506 }
1507
1508 void TemplateTable::lshl() {
1509 transition(itos, ltos);
1510 __ movl(rcx, rax); // get shift count
1511 #ifdef _LP64
1512 __ pop_l(rax); // get shift value
1513 __ shlq(rax);
1514 #else
1515 __ pop_l(rax, rdx); // get shift value
1516 __ lshl(rdx, rax);
1517 #endif
1518 }
1519
1520 void TemplateTable::lshr() {
1521 #ifdef _LP64
1522 transition(itos, ltos);
1523 __ movl(rcx, rax); // get shift count
1524 __ pop_l(rax); // get shift value
1525 __ sarq(rax);
1526 #else
1527 transition(itos, ltos);
1528 __ mov(rcx, rax); // get shift count
1529 __ pop_l(rax, rdx); // get shift value
1530 __ lshr(rdx, rax, true);
1531 #endif
1532 }
1533
1534 void TemplateTable::lushr() {
1535 transition(itos, ltos);
1536 #ifdef _LP64
1537 __ movl(rcx, rax); // get shift count
1538 __ pop_l(rax); // get shift value
1539 __ shrq(rax);
1540 #else
1541 __ mov(rcx, rax); // get shift count
1542 __ pop_l(rax, rdx); // get shift value
1543 __ lshr(rdx, rax);
1544 #endif
1545 }
1546
1547 void TemplateTable::fop2(Operation op) {
1548 transition(ftos, ftos);
1549
1550 if (UseSSE >= 1) {
1551 switch (op) {
1552 case add:
1553 __ addss(xmm0, at_rsp());
1554 __ addptr(rsp, Interpreter::stackElementSize);
1555 break;
1556 case sub:
1557 __ movflt(xmm1, xmm0);
1558 __ pop_f(xmm0);
1559 __ subss(xmm0, xmm1);
1560 break;
1561 case mul:
1562 __ mulss(xmm0, at_rsp());
1563 __ addptr(rsp, Interpreter::stackElementSize);
1564 break;
1565 case div:
1566 __ movflt(xmm1, xmm0);
1567 __ pop_f(xmm0);
1568 __ divss(xmm0, xmm1);
1569 break;
1570 case rem:
1571 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1572 // modulo operation. The frem method calls the function
1573 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1574 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1575 // (signalling or quiet) is returned.
1576 //
1577 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1578 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1579 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1580 // The fprem instruction used on x86_32 is functionally equivalent to
1581 // SharedRuntime::frem in that it returns a NaN.
1582 #ifdef _LP64
1583 __ movflt(xmm1, xmm0);
1584 __ pop_f(xmm0);
1585 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1586 #else
1587 __ push_f(xmm0);
1588 __ pop_f();
1589 __ fld_s(at_rsp());
1590 __ fremr(rax);
1591 __ f2ieee();
1592 __ pop(rax); // pop second operand off the stack
1593 __ push_f();
1594 __ pop_f(xmm0);
1595 #endif
1596 break;
1597 default:
1598 ShouldNotReachHere();
1599 break;
1600 }
1601 } else {
1602 #ifdef _LP64
1603 ShouldNotReachHere();
1604 #else
1605 switch (op) {
1606 case add: __ fadd_s (at_rsp()); break;
1607 case sub: __ fsubr_s(at_rsp()); break;
1608 case mul: __ fmul_s (at_rsp()); break;
1609 case div: __ fdivr_s(at_rsp()); break;
1610 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1611 default : ShouldNotReachHere();
1612 }
1613 __ f2ieee();
1614 __ pop(rax); // pop second operand off the stack
1615 #endif // _LP64
1616 }
1617 }
1618
1619 void TemplateTable::dop2(Operation op) {
1620 transition(dtos, dtos);
1621 if (UseSSE >= 2) {
1622 switch (op) {
1623 case add:
1624 __ addsd(xmm0, at_rsp());
1625 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1626 break;
1627 case sub:
1628 __ movdbl(xmm1, xmm0);
1629 __ pop_d(xmm0);
1630 __ subsd(xmm0, xmm1);
1631 break;
1632 case mul:
1633 __ mulsd(xmm0, at_rsp());
1634 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1635 break;
1636 case div:
1637 __ movdbl(xmm1, xmm0);
1638 __ pop_d(xmm0);
1639 __ divsd(xmm0, xmm1);
1640 break;
1641 case rem:
1642 // Similar to fop2(), the modulo operation is performed using the
1643 // SharedRuntime::drem method (on x86_64 platforms) or using the
1644 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1645 #ifdef _LP64
1646 __ movdbl(xmm1, xmm0);
1647 __ pop_d(xmm0);
1648 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1649 #else
1650 __ push_d(xmm0);
1651 __ pop_d();
1652 __ fld_d(at_rsp());
1653 __ fremr(rax);
1654 __ d2ieee();
1655 __ pop(rax);
1656 __ pop(rdx);
1657 __ push_d();
1658 __ pop_d(xmm0);
1659 #endif
1660 break;
1661 default:
1662 ShouldNotReachHere();
1663 break;
1664 }
1665 } else {
1666 #ifdef _LP64
1667 ShouldNotReachHere();
1668 #else
1669 switch (op) {
1670 case add: __ fadd_d (at_rsp()); break;
1671 case sub: __ fsubr_d(at_rsp()); break;
1672 case mul: {
1673 Label L_strict;
1674 Label L_join;
1675 const Address access_flags (rcx, Method::access_flags_offset());
1676 __ get_method(rcx);
1677 __ movl(rcx, access_flags);
1678 __ testl(rcx, JVM_ACC_STRICT);
1679 __ jccb(Assembler::notZero, L_strict);
1680 __ fmul_d (at_rsp());
1681 __ jmpb(L_join);
1682 __ bind(L_strict);
1683 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1684 __ fmulp();
1685 __ fmul_d (at_rsp());
1686 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1687 __ fmulp();
1688 __ bind(L_join);
1689 break;
1690 }
1691 case div: {
1692 Label L_strict;
1693 Label L_join;
1694 const Address access_flags (rcx, Method::access_flags_offset());
1695 __ get_method(rcx);
1696 __ movl(rcx, access_flags);
1697 __ testl(rcx, JVM_ACC_STRICT);
1698 __ jccb(Assembler::notZero, L_strict);
1699 __ fdivr_d(at_rsp());
1700 __ jmp(L_join);
1701 __ bind(L_strict);
1702 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1703 __ fmul_d (at_rsp());
1704 __ fdivrp();
1705 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1706 __ fmulp();
1707 __ bind(L_join);
1708 break;
1709 }
1710 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1711 default : ShouldNotReachHere();
1712 }
1713 __ d2ieee();
1714 // Pop double precision number from rsp.
1715 __ pop(rax);
1716 __ pop(rdx);
1717 #endif
1718 }
1719 }
1720
1721 void TemplateTable::ineg() {
1722 transition(itos, itos);
1723 __ negl(rax);
1724 }
1725
1726 void TemplateTable::lneg() {
1727 transition(ltos, ltos);
1728 LP64_ONLY(__ negq(rax));
1729 NOT_LP64(__ lneg(rdx, rax));
1730 }
1731
1732 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1733 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1734 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1735 // of 128-bits operands for SSE instructions.
1736 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1737 // Store the value to a 128-bits operand.
1738 operand[0] = lo;
1739 operand[1] = hi;
1740 return operand;
1741 }
1742
1743 // Buffer for 128-bits masks used by SSE instructions.
1744 static jlong float_signflip_pool[2*2];
1745 static jlong double_signflip_pool[2*2];
1746
1747 void TemplateTable::fneg() {
1748 transition(ftos, ftos);
1749 if (UseSSE >= 1) {
1750 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1751 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1752 } else {
1753 LP64_ONLY(ShouldNotReachHere());
1754 NOT_LP64(__ fchs());
1755 }
1756 }
1757
1758 void TemplateTable::dneg() {
1759 transition(dtos, dtos);
1760 if (UseSSE >= 2) {
1761 static jlong *double_signflip =
1762 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1763 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1764 } else {
1765 #ifdef _LP64
1766 ShouldNotReachHere();
1767 #else
1768 __ fchs();
1769 #endif
1770 }
1771 }
1772
1773 void TemplateTable::iinc() {
1774 transition(vtos, vtos);
1775 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1776 locals_index(rbx);
1777 __ addl(iaddress(rbx), rdx);
1778 }
1779
1780 void TemplateTable::wide_iinc() {
1781 transition(vtos, vtos);
1782 __ movl(rdx, at_bcp(4)); // get constant
1783 locals_index_wide(rbx);
1784 __ bswapl(rdx); // swap bytes & sign-extend constant
1785 __ sarl(rdx, 16);
1786 __ addl(iaddress(rbx), rdx);
1787 // Note: should probably use only one movl to get both
1788 // the index and the constant -> fix this
1789 }
1790
1791 void TemplateTable::convert() {
1792 #ifdef _LP64
1793 // Checking
1794 #ifdef ASSERT
1795 {
1796 TosState tos_in = ilgl;
1797 TosState tos_out = ilgl;
1798 switch (bytecode()) {
1799 case Bytecodes::_i2l: // fall through
1800 case Bytecodes::_i2f: // fall through
1801 case Bytecodes::_i2d: // fall through
1802 case Bytecodes::_i2b: // fall through
1803 case Bytecodes::_i2c: // fall through
1804 case Bytecodes::_i2s: tos_in = itos; break;
1805 case Bytecodes::_l2i: // fall through
1806 case Bytecodes::_l2f: // fall through
1807 case Bytecodes::_l2d: tos_in = ltos; break;
1808 case Bytecodes::_f2i: // fall through
1809 case Bytecodes::_f2l: // fall through
1810 case Bytecodes::_f2d: tos_in = ftos; break;
1811 case Bytecodes::_d2i: // fall through
1812 case Bytecodes::_d2l: // fall through
1813 case Bytecodes::_d2f: tos_in = dtos; break;
1814 default : ShouldNotReachHere();
1815 }
1816 switch (bytecode()) {
1817 case Bytecodes::_l2i: // fall through
1818 case Bytecodes::_f2i: // fall through
1819 case Bytecodes::_d2i: // fall through
1820 case Bytecodes::_i2b: // fall through
1821 case Bytecodes::_i2c: // fall through
1822 case Bytecodes::_i2s: tos_out = itos; break;
1823 case Bytecodes::_i2l: // fall through
1824 case Bytecodes::_f2l: // fall through
1825 case Bytecodes::_d2l: tos_out = ltos; break;
1826 case Bytecodes::_i2f: // fall through
1827 case Bytecodes::_l2f: // fall through
1828 case Bytecodes::_d2f: tos_out = ftos; break;
1829 case Bytecodes::_i2d: // fall through
1830 case Bytecodes::_l2d: // fall through
1831 case Bytecodes::_f2d: tos_out = dtos; break;
1832 default : ShouldNotReachHere();
1833 }
1834 transition(tos_in, tos_out);
1835 }
1836 #endif // ASSERT
1837
1838 static const int64_t is_nan = 0x8000000000000000L;
1839
1840 // Conversion
1841 switch (bytecode()) {
1842 case Bytecodes::_i2l:
1843 __ movslq(rax, rax);
1844 break;
1845 case Bytecodes::_i2f:
1846 __ cvtsi2ssl(xmm0, rax);
1847 break;
1848 case Bytecodes::_i2d:
1849 __ cvtsi2sdl(xmm0, rax);
1850 break;
1851 case Bytecodes::_i2b:
1852 __ movsbl(rax, rax);
1853 break;
1854 case Bytecodes::_i2c:
1855 __ movzwl(rax, rax);
1856 break;
1857 case Bytecodes::_i2s:
1858 __ movswl(rax, rax);
1859 break;
1860 case Bytecodes::_l2i:
1861 __ movl(rax, rax);
1862 break;
1863 case Bytecodes::_l2f:
1864 __ cvtsi2ssq(xmm0, rax);
1865 break;
1866 case Bytecodes::_l2d:
1867 __ cvtsi2sdq(xmm0, rax);
1868 break;
1869 case Bytecodes::_f2i:
1870 {
1871 Label L;
1872 __ cvttss2sil(rax, xmm0);
1873 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1874 __ jcc(Assembler::notEqual, L);
1875 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1876 __ bind(L);
1877 }
1878 break;
1879 case Bytecodes::_f2l:
1880 {
1881 Label L;
1882 __ cvttss2siq(rax, xmm0);
1883 // NaN or overflow/underflow?
1884 __ cmp64(rax, ExternalAddress((address) &is_nan));
1885 __ jcc(Assembler::notEqual, L);
1886 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1887 __ bind(L);
1888 }
1889 break;
1890 case Bytecodes::_f2d:
1891 __ cvtss2sd(xmm0, xmm0);
1892 break;
1893 case Bytecodes::_d2i:
1894 {
1895 Label L;
1896 __ cvttsd2sil(rax, xmm0);
1897 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1898 __ jcc(Assembler::notEqual, L);
1899 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1900 __ bind(L);
1901 }
1902 break;
1903 case Bytecodes::_d2l:
1904 {
1905 Label L;
1906 __ cvttsd2siq(rax, xmm0);
1907 // NaN or overflow/underflow?
1908 __ cmp64(rax, ExternalAddress((address) &is_nan));
1909 __ jcc(Assembler::notEqual, L);
1910 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1911 __ bind(L);
1912 }
1913 break;
1914 case Bytecodes::_d2f:
1915 __ cvtsd2ss(xmm0, xmm0);
1916 break;
1917 default:
1918 ShouldNotReachHere();
1919 }
1920 #else
1921 // Checking
1922 #ifdef ASSERT
1923 { TosState tos_in = ilgl;
1924 TosState tos_out = ilgl;
1925 switch (bytecode()) {
1926 case Bytecodes::_i2l: // fall through
1927 case Bytecodes::_i2f: // fall through
1928 case Bytecodes::_i2d: // fall through
1929 case Bytecodes::_i2b: // fall through
1930 case Bytecodes::_i2c: // fall through
1931 case Bytecodes::_i2s: tos_in = itos; break;
1932 case Bytecodes::_l2i: // fall through
1933 case Bytecodes::_l2f: // fall through
1934 case Bytecodes::_l2d: tos_in = ltos; break;
1935 case Bytecodes::_f2i: // fall through
1936 case Bytecodes::_f2l: // fall through
1937 case Bytecodes::_f2d: tos_in = ftos; break;
1938 case Bytecodes::_d2i: // fall through
1939 case Bytecodes::_d2l: // fall through
1940 case Bytecodes::_d2f: tos_in = dtos; break;
1941 default : ShouldNotReachHere();
1942 }
1943 switch (bytecode()) {
1944 case Bytecodes::_l2i: // fall through
1945 case Bytecodes::_f2i: // fall through
1946 case Bytecodes::_d2i: // fall through
1947 case Bytecodes::_i2b: // fall through
1948 case Bytecodes::_i2c: // fall through
1949 case Bytecodes::_i2s: tos_out = itos; break;
1950 case Bytecodes::_i2l: // fall through
1951 case Bytecodes::_f2l: // fall through
1952 case Bytecodes::_d2l: tos_out = ltos; break;
1953 case Bytecodes::_i2f: // fall through
1954 case Bytecodes::_l2f: // fall through
1955 case Bytecodes::_d2f: tos_out = ftos; break;
1956 case Bytecodes::_i2d: // fall through
1957 case Bytecodes::_l2d: // fall through
1958 case Bytecodes::_f2d: tos_out = dtos; break;
1959 default : ShouldNotReachHere();
1960 }
1961 transition(tos_in, tos_out);
1962 }
1963 #endif // ASSERT
1964
1965 // Conversion
1966 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1967 switch (bytecode()) {
1968 case Bytecodes::_i2l:
1969 __ extend_sign(rdx, rax);
1970 break;
1971 case Bytecodes::_i2f:
1972 if (UseSSE >= 1) {
1973 __ cvtsi2ssl(xmm0, rax);
1974 } else {
1975 __ push(rax); // store int on tos
1976 __ fild_s(at_rsp()); // load int to ST0
1977 __ f2ieee(); // truncate to float size
1978 __ pop(rcx); // adjust rsp
1979 }
1980 break;
1981 case Bytecodes::_i2d:
1982 if (UseSSE >= 2) {
1983 __ cvtsi2sdl(xmm0, rax);
1984 } else {
1985 __ push(rax); // add one slot for d2ieee()
1986 __ push(rax); // store int on tos
1987 __ fild_s(at_rsp()); // load int to ST0
1988 __ d2ieee(); // truncate to double size
1989 __ pop(rcx); // adjust rsp
1990 __ pop(rcx);
1991 }
1992 break;
1993 case Bytecodes::_i2b:
1994 __ shll(rax, 24); // truncate upper 24 bits
1995 __ sarl(rax, 24); // and sign-extend byte
1996 LP64_ONLY(__ movsbl(rax, rax));
1997 break;
1998 case Bytecodes::_i2c:
1999 __ andl(rax, 0xFFFF); // truncate upper 16 bits
2000 LP64_ONLY(__ movzwl(rax, rax));
2001 break;
2002 case Bytecodes::_i2s:
2003 __ shll(rax, 16); // truncate upper 16 bits
2004 __ sarl(rax, 16); // and sign-extend short
2005 LP64_ONLY(__ movswl(rax, rax));
2006 break;
2007 case Bytecodes::_l2i:
2008 /* nothing to do */
2009 break;
2010 case Bytecodes::_l2f:
2011 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
2012 // 64-bit long values to floats. On 32-bit platforms it is not possible
2013 // to use that instruction with 64-bit operands, therefore the FPU is
2014 // used to perform the conversion.
2015 __ push(rdx); // store long on tos
2016 __ push(rax);
2017 __ fild_d(at_rsp()); // load long to ST0
2018 __ f2ieee(); // truncate to float size
2019 __ pop(rcx); // adjust rsp
2020 __ pop(rcx);
2021 if (UseSSE >= 1) {
2022 __ push_f();
2023 __ pop_f(xmm0);
2024 }
2025 break;
2026 case Bytecodes::_l2d:
2027 // On 32-bit platforms the FPU is used for conversion because on
2028 // 32-bit platforms it is not not possible to use the cvtsi2sdq
2029 // instruction with 64-bit operands.
2030 __ push(rdx); // store long on tos
2031 __ push(rax);
2032 __ fild_d(at_rsp()); // load long to ST0
2033 __ d2ieee(); // truncate to double size
2034 __ pop(rcx); // adjust rsp
2035 __ pop(rcx);
2036 if (UseSSE >= 2) {
2037 __ push_d();
2038 __ pop_d(xmm0);
2039 }
2040 break;
2041 case Bytecodes::_f2i:
2042 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
2043 // as it returns 0 for any NaN.
2044 if (UseSSE >= 1) {
2045 __ push_f(xmm0);
2046 } else {
2047 __ push(rcx); // reserve space for argument
2048 __ fstp_s(at_rsp()); // pass float argument on stack
2049 }
2050 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
2051 break;
2052 case Bytecodes::_f2l:
2053 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
2054 // as it returns 0 for any NaN.
2055 if (UseSSE >= 1) {
2056 __ push_f(xmm0);
2057 } else {
2058 __ push(rcx); // reserve space for argument
2059 __ fstp_s(at_rsp()); // pass float argument on stack
2060 }
2061 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2062 break;
2063 case Bytecodes::_f2d:
2064 if (UseSSE < 1) {
2065 /* nothing to do */
2066 } else if (UseSSE == 1) {
2067 __ push_f(xmm0);
2068 __ pop_f();
2069 } else { // UseSSE >= 2
2070 __ cvtss2sd(xmm0, xmm0);
2071 }
2072 break;
2073 case Bytecodes::_d2i:
2074 if (UseSSE >= 2) {
2075 __ push_d(xmm0);
2076 } else {
2077 __ push(rcx); // reserve space for argument
2078 __ push(rcx);
2079 __ fstp_d(at_rsp()); // pass double argument on stack
2080 }
2081 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2082 break;
2083 case Bytecodes::_d2l:
2084 if (UseSSE >= 2) {
2085 __ push_d(xmm0);
2086 } else {
2087 __ push(rcx); // reserve space for argument
2088 __ push(rcx);
2089 __ fstp_d(at_rsp()); // pass double argument on stack
2090 }
2091 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2092 break;
2093 case Bytecodes::_d2f:
2094 if (UseSSE <= 1) {
2095 __ push(rcx); // reserve space for f2ieee()
2096 __ f2ieee(); // truncate to float size
2097 __ pop(rcx); // adjust rsp
2098 if (UseSSE == 1) {
2099 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2100 // the conversion is performed using the FPU in this case.
2101 __ push_f();
2102 __ pop_f(xmm0);
2103 }
2104 } else { // UseSSE >= 2
2105 __ cvtsd2ss(xmm0, xmm0);
2106 }
2107 break;
2108 default :
2109 ShouldNotReachHere();
2110 }
2111 #endif
2112 }
2113
2114 void TemplateTable::lcmp() {
2115 transition(ltos, itos);
2116 #ifdef _LP64
2117 Label done;
2118 __ pop_l(rdx);
2119 __ cmpq(rdx, rax);
2120 __ movl(rax, -1);
2121 __ jccb(Assembler::less, done);
2122 __ setb(Assembler::notEqual, rax);
2123 __ movzbl(rax, rax);
2124 __ bind(done);
2125 #else
2126
2127 // y = rdx:rax
2128 __ pop_l(rbx, rcx); // get x = rcx:rbx
2129 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2130 __ mov(rax, rcx);
2131 #endif
2132 }
2133
2134 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2135 if ((is_float && UseSSE >= 1) ||
2136 (!is_float && UseSSE >= 2)) {
2137 Label done;
2138 if (is_float) {
2139 // XXX get rid of pop here, use ... reg, mem32
2140 __ pop_f(xmm1);
2141 __ ucomiss(xmm1, xmm0);
2142 } else {
2143 // XXX get rid of pop here, use ... reg, mem64
2144 __ pop_d(xmm1);
2145 __ ucomisd(xmm1, xmm0);
2146 }
2147 if (unordered_result < 0) {
2148 __ movl(rax, -1);
2149 __ jccb(Assembler::parity, done);
2150 __ jccb(Assembler::below, done);
2151 __ setb(Assembler::notEqual, rdx);
2152 __ movzbl(rax, rdx);
2153 } else {
2154 __ movl(rax, 1);
2155 __ jccb(Assembler::parity, done);
2156 __ jccb(Assembler::above, done);
2157 __ movl(rax, 0);
2158 __ jccb(Assembler::equal, done);
2159 __ decrementl(rax);
2160 }
2161 __ bind(done);
2162 } else {
2163 #ifdef _LP64
2164 ShouldNotReachHere();
2165 #else
2166 if (is_float) {
2167 __ fld_s(at_rsp());
2168 } else {
2169 __ fld_d(at_rsp());
2170 __ pop(rdx);
2171 }
2172 __ pop(rcx);
2173 __ fcmp2int(rax, unordered_result < 0);
2174 #endif // _LP64
2175 }
2176 }
2177
2178 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2179 __ get_method(rcx); // rcx holds method
2180 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2181 // holds bumped taken count
2182
2183 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2184 InvocationCounter::counter_offset();
2185 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2186 InvocationCounter::counter_offset();
2187
2188 // Load up edx with the branch displacement
2189 if (is_wide) {
2190 __ movl(rdx, at_bcp(1));
2191 } else {
2192 __ load_signed_short(rdx, at_bcp(1));
2193 }
2194 __ bswapl(rdx);
2195
2196 if (!is_wide) {
2197 __ sarl(rdx, 16);
2198 }
2199 LP64_ONLY(__ movl2ptr(rdx, rdx));
2200
2201 // Handle all the JSR stuff here, then exit.
2202 // It's much shorter and cleaner than intermingling with the non-JSR
2203 // normal-branch stuff occurring below.
2204 if (is_jsr) {
2205 // Pre-load the next target bytecode into rbx
2206 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2207
2208 // compute return address as bci in rax
2209 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2210 in_bytes(ConstMethod::codes_offset())));
2211 __ subptr(rax, Address(rcx, Method::const_offset()));
2212 // Adjust the bcp in r13 by the displacement in rdx
2213 __ addptr(rbcp, rdx);
2214 // jsr returns atos that is not an oop
2215 __ push_i(rax);
2216 __ dispatch_only(vtos, true);
2217 return;
2218 }
2219
2220 // Normal (non-jsr) branch handling
2221
2222 // Adjust the bcp in r13 by the displacement in rdx
2223 __ addptr(rbcp, rdx);
2224
2225 assert(UseLoopCounter || !UseOnStackReplacement,
2226 "on-stack-replacement requires loop counters");
2227 Label backedge_counter_overflow;
2228 Label profile_method;
2229 Label dispatch;
2230 if (UseLoopCounter) {
2231 // increment backedge counter for backward branches
2232 // rax: MDO
2233 // rbx: MDO bumped taken-count
2234 // rcx: method
2235 // rdx: target offset
2236 // r13: target bcp
2237 // r14: locals pointer
2238 __ testl(rdx, rdx); // check if forward or backward branch
2239 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2240
2241 // check if MethodCounters exists
2242 Label has_counters;
2243 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2244 __ testptr(rax, rax);
2245 __ jcc(Assembler::notZero, has_counters);
2246 __ push(rdx);
2247 __ push(rcx);
2248 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2249 rcx);
2250 __ pop(rcx);
2251 __ pop(rdx);
2252 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2253 __ testptr(rax, rax);
2254 __ jcc(Assembler::zero, dispatch);
2255 __ bind(has_counters);
2256
2257 if (TieredCompilation) {
2258 Label no_mdo;
2259 int increment = InvocationCounter::count_increment;
2260 if (ProfileInterpreter) {
2261 // Are we profiling?
2262 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2263 __ testptr(rbx, rbx);
2264 __ jccb(Assembler::zero, no_mdo);
2265 // Increment the MDO backedge counter
2266 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2267 in_bytes(InvocationCounter::counter_offset()));
2268 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2269 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2270 rax, false, Assembler::zero, &backedge_counter_overflow);
2271 __ jmp(dispatch);
2272 }
2273 __ bind(no_mdo);
2274 // Increment backedge counter in MethodCounters*
2275 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2276 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2277 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2278 rax, false, Assembler::zero, &backedge_counter_overflow);
2279 } else { // not TieredCompilation
2280 // increment counter
2281 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2282 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2283 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2284 __ movl(Address(rcx, be_offset), rax); // store counter
2285
2286 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2287
2288 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2289 __ addl(rax, Address(rcx, be_offset)); // add both counters
2290
2291 if (ProfileInterpreter) {
2292 // Test to see if we should create a method data oop
2293 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2294 __ jcc(Assembler::less, dispatch);
2295
2296 // if no method data exists, go to profile method
2297 __ test_method_data_pointer(rax, profile_method);
2298
2299 if (UseOnStackReplacement) {
2300 // check for overflow against rbx which is the MDO taken count
2301 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2302 __ jcc(Assembler::below, dispatch);
2303
2304 // When ProfileInterpreter is on, the backedge_count comes
2305 // from the MethodData*, which value does not get reset on
2306 // the call to frequency_counter_overflow(). To avoid
2307 // excessive calls to the overflow routine while the method is
2308 // being compiled, add a second test to make sure the overflow
2309 // function is called only once every overflow_frequency.
2310 const int overflow_frequency = 1024;
2311 __ andl(rbx, overflow_frequency - 1);
2312 __ jcc(Assembler::zero, backedge_counter_overflow);
2313
2314 }
2315 } else {
2316 if (UseOnStackReplacement) {
2317 // check for overflow against rax, which is the sum of the
2318 // counters
2319 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2320 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2321
2322 }
2323 }
2324 }
2325 __ bind(dispatch);
2326 }
2327
2328 // Pre-load the next target bytecode into rbx
2329 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2330
2331 // continue with the bytecode @ target
2332 // rax: return bci for jsr's, unused otherwise
2333 // rbx: target bytecode
2334 // r13: target bcp
2335 __ dispatch_only(vtos, true);
2336
2337 if (UseLoopCounter) {
2338 if (ProfileInterpreter) {
2339 // Out-of-line code to allocate method data oop.
2340 __ bind(profile_method);
2341 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2342 __ set_method_data_pointer_for_bcp();
2343 __ jmp(dispatch);
2344 }
2345
2346 if (UseOnStackReplacement) {
2347 // invocation counter overflow
2348 __ bind(backedge_counter_overflow);
2349 __ negptr(rdx);
2350 __ addptr(rdx, rbcp); // branch bcp
2351 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2352 __ call_VM(noreg,
2353 CAST_FROM_FN_PTR(address,
2354 InterpreterRuntime::frequency_counter_overflow),
2355 rdx);
2356
2357 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2358 // rdx: scratch
2359 // r14: locals pointer
2360 // r13: bcp
2361 __ testptr(rax, rax); // test result
2362 __ jcc(Assembler::zero, dispatch); // no osr if null
2363 // nmethod may have been invalidated (VM may block upon call_VM return)
2364 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2365 __ jcc(Assembler::notEqual, dispatch);
2366
2367 // We have the address of an on stack replacement routine in rax.
2368 // In preparation of invoking it, first we must migrate the locals
2369 // and monitors from off the interpreter frame on the stack.
2370 // Ensure to save the osr nmethod over the migration call,
2371 // it will be preserved in rbx.
2372 __ mov(rbx, rax);
2373
2374 NOT_LP64(__ get_thread(rcx));
2375
2376 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2377
2378 // rax is OSR buffer, move it to expected parameter location
2379 LP64_ONLY(__ mov(j_rarg0, rax));
2380 NOT_LP64(__ mov(rcx, rax));
2381 // We use j_rarg definitions here so that registers don't conflict as parameter
2382 // registers change across platforms as we are in the midst of a calling
2383 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2384
2385 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2386 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2387
2388 // pop the interpreter frame
2389 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2390 __ leave(); // remove frame anchor
2391 __ pop(retaddr); // get return address
2392 __ mov(rsp, sender_sp); // set sp to sender sp
2393 // Ensure compiled code always sees stack at proper alignment
2394 __ andptr(rsp, -(StackAlignmentInBytes));
2395
2396 // unlike x86 we need no specialized return from compiled code
2397 // to the interpreter or the call stub.
2398
2399 // push the return address
2400 __ push(retaddr);
2401
2402 // and begin the OSR nmethod
2403 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2404 }
2405 }
2406 }
2407
2408 void TemplateTable::if_0cmp(Condition cc) {
2409 transition(itos, vtos);
2410 // assume branch is more often taken than not (loops use backward branches)
2411 Label not_taken;
2412 __ testl(rax, rax);
2413 __ jcc(j_not(cc), not_taken);
2414 branch(false, false);
2415 __ bind(not_taken);
2416 __ profile_not_taken_branch(rax);
2417 }
2418
2419 void TemplateTable::if_icmp(Condition cc) {
2420 transition(itos, vtos);
2421 // assume branch is more often taken than not (loops use backward branches)
2422 Label not_taken;
2423 __ pop_i(rdx);
2424 __ cmpl(rdx, rax);
2425 __ jcc(j_not(cc), not_taken);
2426 branch(false, false);
2427 __ bind(not_taken);
2428 __ profile_not_taken_branch(rax);
2429 }
2430
2431 void TemplateTable::if_nullcmp(Condition cc) {
2432 transition(atos, vtos);
2433 // assume branch is more often taken than not (loops use backward branches)
2434 Label not_taken;
2435 __ testptr(rax, rax);
2436 __ jcc(j_not(cc), not_taken);
2437 branch(false, false);
2438 __ bind(not_taken);
2439 __ profile_not_taken_branch(rax);
2440 }
2441
2442 void TemplateTable::if_acmp(Condition cc) {
2443 transition(atos, vtos);
2444 // assume branch is more often taken than not (loops use backward branches)
2445 Label not_taken, is_null;
2446 __ pop_ptr(rdx);
2447
2448 if (UseNewAcmp) {
2449 // TODO what about interpreter buffered values?
2450 const int mask = Universe::oop_metadata_odd_mask();
2451 __ testptr(rdx, rdx);
2452 __ jcc(Assembler::zero, is_null);
2453 __ movl(rbx, Address(rdx, oopDesc::klass_offset_in_bytes()));
2454 __ andptr(rbx, mask);
2455 // Check if a shift is required for perturbation to affect aligned bits
2456 if (mask == KlassPtrEvenOddMask && ObjectAlignmentInBytes <= KlassAlignmentInBytes) {
2457 assert((mask >> LogKlassAlignmentInBytes) == 1, "invalid shift");
2458 __ shrptr(rbx, LogKlassAlignmentInBytes);
2459 } else {
2460 assert(mask < ObjectAlignmentInBytes, "invalid mask");
2461 }
2462 __ orptr(rdx, rbx);
2463 __ bind(is_null);
2464 }
2465
2466 __ cmpoop(rdx, rax);
2467 __ jcc(j_not(cc), not_taken);
2468 branch(false, false);
2469 __ bind(not_taken);
2470 __ profile_not_taken_branch(rax);
2471 }
2472
2473 void TemplateTable::ret() {
2474 transition(vtos, vtos);
2475 locals_index(rbx);
2476 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2477 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2478 __ profile_ret(rbx, rcx);
2479 __ get_method(rax);
2480 __ movptr(rbcp, Address(rax, Method::const_offset()));
2481 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2482 ConstMethod::codes_offset()));
2483 __ dispatch_next(vtos, 0, true);
2484 }
2485
2486 void TemplateTable::wide_ret() {
2487 transition(vtos, vtos);
2488 locals_index_wide(rbx);
2489 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2490 __ profile_ret(rbx, rcx);
2491 __ get_method(rax);
2492 __ movptr(rbcp, Address(rax, Method::const_offset()));
2493 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2494 __ dispatch_next(vtos, 0, true);
2495 }
2496
2497 void TemplateTable::tableswitch() {
2498 Label default_case, continue_execution;
2499 transition(itos, vtos);
2500
2501 // align r13/rsi
2502 __ lea(rbx, at_bcp(BytesPerInt));
2503 __ andptr(rbx, -BytesPerInt);
2504 // load lo & hi
2505 __ movl(rcx, Address(rbx, BytesPerInt));
2506 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2507 __ bswapl(rcx);
2508 __ bswapl(rdx);
2509 // check against lo & hi
2510 __ cmpl(rax, rcx);
2511 __ jcc(Assembler::less, default_case);
2512 __ cmpl(rax, rdx);
2513 __ jcc(Assembler::greater, default_case);
2514 // lookup dispatch offset
2515 __ subl(rax, rcx);
2516 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2517 __ profile_switch_case(rax, rbx, rcx);
2518 // continue execution
2519 __ bind(continue_execution);
2520 __ bswapl(rdx);
2521 LP64_ONLY(__ movl2ptr(rdx, rdx));
2522 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2523 __ addptr(rbcp, rdx);
2524 __ dispatch_only(vtos, true);
2525 // handle default
2526 __ bind(default_case);
2527 __ profile_switch_default(rax);
2528 __ movl(rdx, Address(rbx, 0));
2529 __ jmp(continue_execution);
2530 }
2531
2532 void TemplateTable::lookupswitch() {
2533 transition(itos, itos);
2534 __ stop("lookupswitch bytecode should have been rewritten");
2535 }
2536
2537 void TemplateTable::fast_linearswitch() {
2538 transition(itos, vtos);
2539 Label loop_entry, loop, found, continue_execution;
2540 // bswap rax so we can avoid bswapping the table entries
2541 __ bswapl(rax);
2542 // align r13
2543 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2544 // this instruction (change offsets
2545 // below)
2546 __ andptr(rbx, -BytesPerInt);
2547 // set counter
2548 __ movl(rcx, Address(rbx, BytesPerInt));
2549 __ bswapl(rcx);
2550 __ jmpb(loop_entry);
2551 // table search
2552 __ bind(loop);
2553 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2554 __ jcc(Assembler::equal, found);
2555 __ bind(loop_entry);
2556 __ decrementl(rcx);
2557 __ jcc(Assembler::greaterEqual, loop);
2558 // default case
2559 __ profile_switch_default(rax);
2560 __ movl(rdx, Address(rbx, 0));
2561 __ jmp(continue_execution);
2562 // entry found -> get offset
2563 __ bind(found);
2564 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2565 __ profile_switch_case(rcx, rax, rbx);
2566 // continue execution
2567 __ bind(continue_execution);
2568 __ bswapl(rdx);
2569 __ movl2ptr(rdx, rdx);
2570 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2571 __ addptr(rbcp, rdx);
2572 __ dispatch_only(vtos, true);
2573 }
2574
2575 void TemplateTable::fast_binaryswitch() {
2576 transition(itos, vtos);
2577 // Implementation using the following core algorithm:
2578 //
2579 // int binary_search(int key, LookupswitchPair* array, int n) {
2580 // // Binary search according to "Methodik des Programmierens" by
2581 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2582 // int i = 0;
2583 // int j = n;
2584 // while (i+1 < j) {
2585 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2586 // // with Q: for all i: 0 <= i < n: key < a[i]
2587 // // where a stands for the array and assuming that the (inexisting)
2588 // // element a[n] is infinitely big.
2589 // int h = (i + j) >> 1;
2590 // // i < h < j
2591 // if (key < array[h].fast_match()) {
2592 // j = h;
2593 // } else {
2594 // i = h;
2595 // }
2596 // }
2597 // // R: a[i] <= key < a[i+1] or Q
2598 // // (i.e., if key is within array, i is the correct index)
2599 // return i;
2600 // }
2601
2602 // Register allocation
2603 const Register key = rax; // already set (tosca)
2604 const Register array = rbx;
2605 const Register i = rcx;
2606 const Register j = rdx;
2607 const Register h = rdi;
2608 const Register temp = rsi;
2609
2610 // Find array start
2611 NOT_LP64(__ save_bcp());
2612
2613 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2614 // get rid of this
2615 // instruction (change
2616 // offsets below)
2617 __ andptr(array, -BytesPerInt);
2618
2619 // Initialize i & j
2620 __ xorl(i, i); // i = 0;
2621 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2622
2623 // Convert j into native byteordering
2624 __ bswapl(j);
2625
2626 // And start
2627 Label entry;
2628 __ jmp(entry);
2629
2630 // binary search loop
2631 {
2632 Label loop;
2633 __ bind(loop);
2634 // int h = (i + j) >> 1;
2635 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2636 __ sarl(h, 1); // h = (i + j) >> 1;
2637 // if (key < array[h].fast_match()) {
2638 // j = h;
2639 // } else {
2640 // i = h;
2641 // }
2642 // Convert array[h].match to native byte-ordering before compare
2643 __ movl(temp, Address(array, h, Address::times_8));
2644 __ bswapl(temp);
2645 __ cmpl(key, temp);
2646 // j = h if (key < array[h].fast_match())
2647 __ cmov32(Assembler::less, j, h);
2648 // i = h if (key >= array[h].fast_match())
2649 __ cmov32(Assembler::greaterEqual, i, h);
2650 // while (i+1 < j)
2651 __ bind(entry);
2652 __ leal(h, Address(i, 1)); // i+1
2653 __ cmpl(h, j); // i+1 < j
2654 __ jcc(Assembler::less, loop);
2655 }
2656
2657 // end of binary search, result index is i (must check again!)
2658 Label default_case;
2659 // Convert array[i].match to native byte-ordering before compare
2660 __ movl(temp, Address(array, i, Address::times_8));
2661 __ bswapl(temp);
2662 __ cmpl(key, temp);
2663 __ jcc(Assembler::notEqual, default_case);
2664
2665 // entry found -> j = offset
2666 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2667 __ profile_switch_case(i, key, array);
2668 __ bswapl(j);
2669 LP64_ONLY(__ movslq(j, j));
2670
2671 NOT_LP64(__ restore_bcp());
2672 NOT_LP64(__ restore_locals()); // restore rdi
2673
2674 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2675 __ addptr(rbcp, j);
2676 __ dispatch_only(vtos, true);
2677
2678 // default case -> j = default offset
2679 __ bind(default_case);
2680 __ profile_switch_default(i);
2681 __ movl(j, Address(array, -2 * BytesPerInt));
2682 __ bswapl(j);
2683 LP64_ONLY(__ movslq(j, j));
2684
2685 NOT_LP64(__ restore_bcp());
2686 NOT_LP64(__ restore_locals());
2687
2688 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2689 __ addptr(rbcp, j);
2690 __ dispatch_only(vtos, true);
2691 }
2692
2693 void TemplateTable::_return(TosState state) {
2694 transition(state, state);
2695
2696 assert(_desc->calls_vm(),
2697 "inconsistent calls_vm information"); // call in remove_activation
2698
2699 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2700 assert(state == vtos, "only valid state");
2701 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2702 __ movptr(robj, aaddress(0));
2703 __ load_klass(rdi, robj);
2704 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2705 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2706 Label skip_register_finalizer;
2707 __ jcc(Assembler::zero, skip_register_finalizer);
2708
2709 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2710
2711 __ bind(skip_register_finalizer);
2712 }
2713
2714 #ifdef _LP64
2715 if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2716 Label no_safepoint;
2717 NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2718 __ testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2719 __ jcc(Assembler::zero, no_safepoint);
2720 __ push(state);
2721 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2722 InterpreterRuntime::at_safepoint));
2723 __ pop(state);
2724 __ bind(no_safepoint);
2725 }
2726 #endif
2727
2728 // Narrow result if state is itos but result type is smaller.
2729 // Need to narrow in the return bytecode rather than in generate_return_entry
2730 // since compiled code callers expect the result to already be narrowed.
2731 if (state == itos) {
2732 __ narrow(rax);
2733 }
2734 __ remove_activation(state, rbcp);
2735
2736 __ jmp(rbcp);
2737 }
2738
2739 // ----------------------------------------------------------------------------
2740 // Volatile variables demand their effects be made known to all CPU's
2741 // in order. Store buffers on most chips allow reads & writes to
2742 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2743 // without some kind of memory barrier (i.e., it's not sufficient that
2744 // the interpreter does not reorder volatile references, the hardware
2745 // also must not reorder them).
2746 //
2747 // According to the new Java Memory Model (JMM):
2748 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2749 // writes act as aquire & release, so:
2750 // (2) A read cannot let unrelated NON-volatile memory refs that
2751 // happen after the read float up to before the read. It's OK for
2752 // non-volatile memory refs that happen before the volatile read to
2753 // float down below it.
2754 // (3) Similar a volatile write cannot let unrelated NON-volatile
2755 // memory refs that happen BEFORE the write float down to after the
2756 // write. It's OK for non-volatile memory refs that happen after the
2757 // volatile write to float up before it.
2758 //
2759 // We only put in barriers around volatile refs (they are expensive),
2760 // not _between_ memory refs (that would require us to track the
2761 // flavor of the previous memory refs). Requirements (2) and (3)
2762 // require some barriers before volatile stores and after volatile
2763 // loads. These nearly cover requirement (1) but miss the
2764 // volatile-store-volatile-load case. This final case is placed after
2765 // volatile-stores although it could just as well go before
2766 // volatile-loads.
2767
2768 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2769 // Helper function to insert a is-volatile test and memory barrier
2770 if(!os::is_MP()) return; // Not needed on single CPU
2771 __ membar(order_constraint);
2772 }
2773
2774 void TemplateTable::resolve_cache_and_index(int byte_no,
2775 Register Rcache,
2776 Register index,
2777 size_t index_size) {
2778 const Register temp = rbx;
2779 assert_different_registers(Rcache, index, temp);
2780
2781 Label resolved;
2782
2783 Bytecodes::Code code = bytecode();
2784 switch (code) {
2785 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2786 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2787 default: break;
2788 }
2789
2790 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2791 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2792 __ cmpl(temp, code); // have we resolved this bytecode?
2793 __ jcc(Assembler::equal, resolved);
2794
2795 // resolve first time through
2796 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2797 __ movl(temp, code);
2798 __ call_VM(noreg, entry, temp);
2799 // Update registers with resolved info
2800 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2801 __ bind(resolved);
2802 }
2803
2804 // The cache and index registers must be set before call
2805 void TemplateTable::load_field_cp_cache_entry(Register obj,
2806 Register cache,
2807 Register index,
2808 Register off,
2809 Register flags,
2810 bool is_static = false) {
2811 assert_different_registers(cache, index, flags, off);
2812
2813 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2814 // Field offset
2815 __ movptr(off, Address(cache, index, Address::times_ptr,
2816 in_bytes(cp_base_offset +
2817 ConstantPoolCacheEntry::f2_offset())));
2818 // Flags
2819 __ movl(flags, Address(cache, index, Address::times_ptr,
2820 in_bytes(cp_base_offset +
2821 ConstantPoolCacheEntry::flags_offset())));
2822
2823 // klass overwrite register
2824 if (is_static) {
2825 __ movptr(obj, Address(cache, index, Address::times_ptr,
2826 in_bytes(cp_base_offset +
2827 ConstantPoolCacheEntry::f1_offset())));
2828 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2829 __ movptr(obj, Address(obj, mirror_offset));
2830 __ resolve_oop_handle(obj);
2831 }
2832 }
2833
2834 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2835 Register method,
2836 Register itable_index,
2837 Register flags,
2838 bool is_invokevirtual,
2839 bool is_invokevfinal, /*unused*/
2840 bool is_invokedynamic) {
2841 // setup registers
2842 const Register cache = rcx;
2843 const Register index = rdx;
2844 assert_different_registers(method, flags);
2845 assert_different_registers(method, cache, index);
2846 assert_different_registers(itable_index, flags);
2847 assert_different_registers(itable_index, cache, index);
2848 // determine constant pool cache field offsets
2849 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2850 const int method_offset = in_bytes(
2851 ConstantPoolCache::base_offset() +
2852 ((byte_no == f2_byte)
2853 ? ConstantPoolCacheEntry::f2_offset()
2854 : ConstantPoolCacheEntry::f1_offset()));
2855 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2856 ConstantPoolCacheEntry::flags_offset());
2857 // access constant pool cache fields
2858 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2859 ConstantPoolCacheEntry::f2_offset());
2860
2861 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2862 resolve_cache_and_index(byte_no, cache, index, index_size);
2863 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2864
2865 if (itable_index != noreg) {
2866 // pick up itable or appendix index from f2 also:
2867 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2868 }
2869 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2870 }
2871
2872 // The registers cache and index expected to be set before call.
2873 // Correct values of the cache and index registers are preserved.
2874 void TemplateTable::jvmti_post_field_access(Register cache,
2875 Register index,
2876 bool is_static,
2877 bool has_tos) {
2878 if (JvmtiExport::can_post_field_access()) {
2879 // Check to see if a field access watch has been set before we take
2880 // the time to call into the VM.
2881 Label L1;
2882 assert_different_registers(cache, index, rax);
2883 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2884 __ testl(rax,rax);
2885 __ jcc(Assembler::zero, L1);
2886
2887 // cache entry pointer
2888 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2889 __ shll(index, LogBytesPerWord);
2890 __ addptr(cache, index);
2891 if (is_static) {
2892 __ xorptr(rax, rax); // NULL object reference
2893 } else {
2894 __ pop(atos); // Get the object
2895 __ verify_oop(rax);
2896 __ push(atos); // Restore stack state
2897 }
2898 // rax,: object pointer or NULL
2899 // cache: cache entry pointer
2900 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2901 rax, cache);
2902 __ get_cache_and_index_at_bcp(cache, index, 1);
2903 __ bind(L1);
2904 }
2905 }
2906
2907 void TemplateTable::pop_and_check_object(Register r) {
2908 __ pop_ptr(r);
2909 __ null_check(r); // for field access must check obj.
2910 __ verify_oop(r);
2911 }
2912
2913 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2914 transition(vtos, vtos);
2915
2916 const Register cache = rcx;
2917 const Register index = rdx;
2918 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2919 const Register off = rbx;
2920 const Register flags = rax;
2921 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2922
2923 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2924 jvmti_post_field_access(cache, index, is_static, false);
2925 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2926
2927 if (!is_static) pop_and_check_object(obj);
2928
2929 const Address field(obj, off, Address::times_1, 0*wordSize);
2930 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2931
2932 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2933
2934 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2935 // Make sure we don't need to mask edx after the above shift
2936 assert(btos == 0, "change code, btos != 0");
2937
2938 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2939
2940 __ jcc(Assembler::notZero, notByte);
2941 // btos
2942 __ load_signed_byte(rax, field);
2943 __ push(btos);
2944 // Rewrite bytecode to be faster
2945 if (!is_static && rc == may_rewrite) {
2946 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2947 }
2948 __ jmp(Done);
2949
2950 __ bind(notByte);
2951 __ cmpl(flags, ztos);
2952 __ jcc(Assembler::notEqual, notBool);
2953
2954 // ztos (same code as btos)
2955 __ load_signed_byte(rax, field);
2956 __ push(ztos);
2957 // Rewrite bytecode to be faster
2958 if (!is_static && rc == may_rewrite) {
2959 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2960 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2961 }
2962 __ jmp(Done);
2963
2964 __ bind(notBool);
2965 __ cmpl(flags, atos);
2966 __ jcc(Assembler::notEqual, notObj);
2967 // atos
2968 __ load_heap_oop(rax, field);
2969 __ push(atos);
2970 if (!is_static && rc == may_rewrite) {
2971 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2972 }
2973 __ jmp(Done);
2974
2975 __ bind(notObj);
2976 __ cmpl(flags, itos);
2977 __ jcc(Assembler::notEqual, notInt);
2978 // itos
2979 __ movl(rax, field);
2980 __ push(itos);
2981 // Rewrite bytecode to be faster
2982 if (!is_static && rc == may_rewrite) {
2983 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2984 }
2985 __ jmp(Done);
2986
2987 __ bind(notInt);
2988 __ cmpl(flags, ctos);
2989 __ jcc(Assembler::notEqual, notChar);
2990 // ctos
2991 __ load_unsigned_short(rax, field);
2992 __ push(ctos);
2993 // Rewrite bytecode to be faster
2994 if (!is_static && rc == may_rewrite) {
2995 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2996 }
2997 __ jmp(Done);
2998
2999 __ bind(notChar);
3000 __ cmpl(flags, stos);
3001 __ jcc(Assembler::notEqual, notShort);
3002 // stos
3003 __ load_signed_short(rax, field);
3004 __ push(stos);
3005 // Rewrite bytecode to be faster
3006 if (!is_static && rc == may_rewrite) {
3007 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
3008 }
3009 __ jmp(Done);
3010
3011 __ bind(notShort);
3012 __ cmpl(flags, ltos);
3013 __ jcc(Assembler::notEqual, notLong);
3014 // ltos
3015
3016 #ifndef _LP64
3017 // Generate code as if volatile. There just aren't enough registers to
3018 // save that information and this code is faster than the test.
3019 __ fild_d(field); // Must load atomically
3020 __ subptr(rsp,2*wordSize); // Make space for store
3021 __ fistp_d(Address(rsp,0));
3022 __ pop(rax);
3023 __ pop(rdx);
3024 #else
3025 __ movq(rax, field);
3026 #endif
3027
3028 __ push(ltos);
3029 // Rewrite bytecode to be faster
3030 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
3031 __ jmp(Done);
3032
3033 __ bind(notLong);
3034 __ cmpl(flags, ftos);
3035 __ jcc(Assembler::notEqual, notFloat);
3036 // ftos
3037
3038 __ load_float(field);
3039 __ push(ftos);
3040 // Rewrite bytecode to be faster
3041 if (!is_static && rc == may_rewrite) {
3042 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
3043 }
3044 __ jmp(Done);
3045
3046 __ bind(notFloat);
3047 #ifdef ASSERT
3048 __ cmpl(flags, dtos);
3049 __ jcc(Assembler::notEqual, notDouble);
3050 #endif
3051 // dtos
3052 __ load_double(field);
3053 __ push(dtos);
3054 // Rewrite bytecode to be faster
3055 if (!is_static && rc == may_rewrite) {
3056 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
3057 }
3058 #ifdef ASSERT
3059 __ jmp(Done);
3060
3061
3062 __ bind(notDouble);
3063 __ stop("Bad state");
3064 #endif
3065
3066 __ bind(Done);
3067 // [jk] not needed currently
3068 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3069 // Assembler::LoadStore));
3070 }
3071
3072 void TemplateTable::getfield(int byte_no) {
3073 getfield_or_static(byte_no, false);
3074 }
3075
3076 void TemplateTable::nofast_getfield(int byte_no) {
3077 getfield_or_static(byte_no, false, may_not_rewrite);
3078 }
3079
3080 void TemplateTable::getstatic(int byte_no) {
3081 getfield_or_static(byte_no, true);
3082 }
3083
3084
3085 // The registers cache and index expected to be set before call.
3086 // The function may destroy various registers, just not the cache and index registers.
3087 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3088
3089 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
3090 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
3091 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3092 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
3093
3094 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3095
3096 if (JvmtiExport::can_post_field_modification()) {
3097 // Check to see if a field modification watch has been set before
3098 // we take the time to call into the VM.
3099 Label L1;
3100 assert_different_registers(cache, index, rax);
3101 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3102 __ testl(rax, rax);
3103 __ jcc(Assembler::zero, L1);
3104
3105 __ get_cache_and_index_at_bcp(robj, RDX, 1);
3106
3107
3108 if (is_static) {
3109 // Life is simple. Null out the object pointer.
3110 __ xorl(RBX, RBX);
3111
3112 } else {
3113 // Life is harder. The stack holds the value on top, followed by
3114 // the object. We don't know the size of the value, though; it
3115 // could be one or two words depending on its type. As a result,
3116 // we must find the type to determine where the object is.
3117 #ifndef _LP64
3118 Label two_word, valsize_known;
3119 #endif
3120 __ movl(RCX, Address(robj, RDX,
3121 Address::times_ptr,
3122 in_bytes(cp_base_offset +
3123 ConstantPoolCacheEntry::flags_offset())));
3124 NOT_LP64(__ mov(rbx, rsp));
3125 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3126
3127 // Make sure we don't need to mask rcx after the above shift
3128 ConstantPoolCacheEntry::verify_tos_state_shift();
3129 #ifdef _LP64
3130 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
3131 __ cmpl(c_rarg3, ltos);
3132 __ cmovptr(Assembler::equal,
3133 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3134 __ cmpl(c_rarg3, dtos);
3135 __ cmovptr(Assembler::equal,
3136 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3137 #else
3138 __ cmpl(rcx, ltos);
3139 __ jccb(Assembler::equal, two_word);
3140 __ cmpl(rcx, dtos);
3141 __ jccb(Assembler::equal, two_word);
3142 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3143 __ jmpb(valsize_known);
3144
3145 __ bind(two_word);
3146 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3147
3148 __ bind(valsize_known);
3149 // setup object pointer
3150 __ movptr(rbx, Address(rbx, 0));
3151 #endif
3152 }
3153 // cache entry pointer
3154 __ addptr(robj, in_bytes(cp_base_offset));
3155 __ shll(RDX, LogBytesPerWord);
3156 __ addptr(robj, RDX);
3157 // object (tos)
3158 __ mov(RCX, rsp);
3159 // c_rarg1: object pointer set up above (NULL if static)
3160 // c_rarg2: cache entry pointer
3161 // c_rarg3: jvalue object on the stack
3162 __ call_VM(noreg,
3163 CAST_FROM_FN_PTR(address,
3164 InterpreterRuntime::post_field_modification),
3165 RBX, robj, RCX);
3166 __ get_cache_and_index_at_bcp(cache, index, 1);
3167 __ bind(L1);
3168 }
3169 }
3170
3171 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3172 transition(vtos, vtos);
3173
3174 const Register cache = rcx;
3175 const Register index = rdx;
3176 const Register obj = rcx;
3177 const Register off = rbx;
3178 const Register flags = rax;
3179 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3180
3181 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3182 jvmti_post_field_mod(cache, index, is_static);
3183 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3184
3185 // [jk] not needed currently
3186 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3187 // Assembler::StoreStore));
3188
3189 Label notVolatile, Done;
3190 __ movl(rdx, flags);
3191 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3192 __ andl(rdx, 0x1);
3193
3194 // field addresses
3195 const Address field(obj, off, Address::times_1, 0*wordSize);
3196 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3197
3198 Label notByte, notBool, notInt, notShort, notChar,
3199 notLong, notFloat, notObj, notDouble;
3200
3201 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3202
3203 assert(btos == 0, "change code, btos != 0");
3204 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3205 __ jcc(Assembler::notZero, notByte);
3206
3207 // btos
3208 {
3209 __ pop(btos);
3210 if (!is_static) pop_and_check_object(obj);
3211 __ movb(field, rax);
3212 if (!is_static && rc == may_rewrite) {
3213 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3214 }
3215 __ jmp(Done);
3216 }
3217
3218 __ bind(notByte);
3219 __ cmpl(flags, ztos);
3220 __ jcc(Assembler::notEqual, notBool);
3221
3222 // ztos
3223 {
3224 __ pop(ztos);
3225 if (!is_static) pop_and_check_object(obj);
3226 __ andl(rax, 0x1);
3227 __ movb(field, rax);
3228 if (!is_static && rc == may_rewrite) {
3229 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3230 }
3231 __ jmp(Done);
3232 }
3233
3234 __ bind(notBool);
3235 __ cmpl(flags, atos);
3236 __ jcc(Assembler::notEqual, notObj);
3237
3238 // atos
3239 {
3240 __ pop(atos);
3241 if (!is_static) pop_and_check_object(obj);
3242 // Store into the field
3243 do_oop_store(_masm, field, rax, _bs->kind(), false);
3244 if (!is_static && rc == may_rewrite) {
3245 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3246 }
3247 __ jmp(Done);
3248 }
3249
3250 __ bind(notObj);
3251 __ cmpl(flags, itos);
3252 __ jcc(Assembler::notEqual, notInt);
3253
3254 // itos
3255 {
3256 __ pop(itos);
3257 if (!is_static) pop_and_check_object(obj);
3258 __ movl(field, rax);
3259 if (!is_static && rc == may_rewrite) {
3260 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3261 }
3262 __ jmp(Done);
3263 }
3264
3265 __ bind(notInt);
3266 __ cmpl(flags, ctos);
3267 __ jcc(Assembler::notEqual, notChar);
3268
3269 // ctos
3270 {
3271 __ pop(ctos);
3272 if (!is_static) pop_and_check_object(obj);
3273 __ movw(field, rax);
3274 if (!is_static && rc == may_rewrite) {
3275 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3276 }
3277 __ jmp(Done);
3278 }
3279
3280 __ bind(notChar);
3281 __ cmpl(flags, stos);
3282 __ jcc(Assembler::notEqual, notShort);
3283
3284 // stos
3285 {
3286 __ pop(stos);
3287 if (!is_static) pop_and_check_object(obj);
3288 __ movw(field, rax);
3289 if (!is_static && rc == may_rewrite) {
3290 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3291 }
3292 __ jmp(Done);
3293 }
3294
3295 __ bind(notShort);
3296 __ cmpl(flags, ltos);
3297 __ jcc(Assembler::notEqual, notLong);
3298
3299 // ltos
3300 #ifdef _LP64
3301 {
3302 __ pop(ltos);
3303 if (!is_static) pop_and_check_object(obj);
3304 __ movq(field, rax);
3305 if (!is_static && rc == may_rewrite) {
3306 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3307 }
3308 __ jmp(Done);
3309 }
3310 #else
3311 {
3312 Label notVolatileLong;
3313 __ testl(rdx, rdx);
3314 __ jcc(Assembler::zero, notVolatileLong);
3315
3316 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3317 if (!is_static) pop_and_check_object(obj);
3318
3319 // Replace with real volatile test
3320 __ push(rdx);
3321 __ push(rax); // Must update atomically with FIST
3322 __ fild_d(Address(rsp,0)); // So load into FPU register
3323 __ fistp_d(field); // and put into memory atomically
3324 __ addptr(rsp, 2*wordSize);
3325 // volatile_barrier();
3326 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3327 Assembler::StoreStore));
3328 // Don't rewrite volatile version
3329 __ jmp(notVolatile);
3330
3331 __ bind(notVolatileLong);
3332
3333 __ pop(ltos); // overwrites rdx
3334 if (!is_static) pop_and_check_object(obj);
3335 __ movptr(hi, rdx);
3336 __ movptr(field, rax);
3337 // Don't rewrite to _fast_lputfield for potential volatile case.
3338 __ jmp(notVolatile);
3339 }
3340 #endif // _LP64
3341
3342 __ bind(notLong);
3343 __ cmpl(flags, ftos);
3344 __ jcc(Assembler::notEqual, notFloat);
3345
3346 // ftos
3347 {
3348 __ pop(ftos);
3349 if (!is_static) pop_and_check_object(obj);
3350 __ store_float(field);
3351 if (!is_static && rc == may_rewrite) {
3352 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3353 }
3354 __ jmp(Done);
3355 }
3356
3357 __ bind(notFloat);
3358 #ifdef ASSERT
3359 __ cmpl(flags, dtos);
3360 __ jcc(Assembler::notEqual, notDouble);
3361 #endif
3362
3363 // dtos
3364 {
3365 __ pop(dtos);
3366 if (!is_static) pop_and_check_object(obj);
3367 __ store_double(field);
3368 if (!is_static && rc == may_rewrite) {
3369 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3370 }
3371 }
3372
3373 #ifdef ASSERT
3374 __ jmp(Done);
3375
3376 __ bind(notDouble);
3377 __ stop("Bad state");
3378 #endif
3379
3380 __ bind(Done);
3381
3382 // Check for volatile store
3383 __ testl(rdx, rdx);
3384 __ jcc(Assembler::zero, notVolatile);
3385 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3386 Assembler::StoreStore));
3387 __ bind(notVolatile);
3388 }
3389
3390 void TemplateTable::putfield(int byte_no) {
3391 putfield_or_static(byte_no, false);
3392 }
3393
3394 void TemplateTable::nofast_putfield(int byte_no) {
3395 putfield_or_static(byte_no, false, may_not_rewrite);
3396 }
3397
3398 void TemplateTable::putstatic(int byte_no) {
3399 putfield_or_static(byte_no, true);
3400 }
3401
3402 void TemplateTable::jvmti_post_fast_field_mod() {
3403
3404 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3405
3406 if (JvmtiExport::can_post_field_modification()) {
3407 // Check to see if a field modification watch has been set before
3408 // we take the time to call into the VM.
3409 Label L2;
3410 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3411 __ testl(scratch, scratch);
3412 __ jcc(Assembler::zero, L2);
3413 __ pop_ptr(rbx); // copy the object pointer from tos
3414 __ verify_oop(rbx);
3415 __ push_ptr(rbx); // put the object pointer back on tos
3416 // Save tos values before call_VM() clobbers them. Since we have
3417 // to do it for every data type, we use the saved values as the
3418 // jvalue object.
3419 switch (bytecode()) { // load values into the jvalue object
3420 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3421 case Bytecodes::_fast_bputfield: // fall through
3422 case Bytecodes::_fast_zputfield: // fall through
3423 case Bytecodes::_fast_sputfield: // fall through
3424 case Bytecodes::_fast_cputfield: // fall through
3425 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3426 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3427 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3428 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3429
3430 default:
3431 ShouldNotReachHere();
3432 }
3433 __ mov(scratch, rsp); // points to jvalue on the stack
3434 // access constant pool cache entry
3435 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3436 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3437 __ verify_oop(rbx);
3438 // rbx: object pointer copied above
3439 // c_rarg2: cache entry pointer
3440 // c_rarg3: jvalue object on the stack
3441 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3442 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3443
3444 switch (bytecode()) { // restore tos values
3445 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3446 case Bytecodes::_fast_bputfield: // fall through
3447 case Bytecodes::_fast_zputfield: // fall through
3448 case Bytecodes::_fast_sputfield: // fall through
3449 case Bytecodes::_fast_cputfield: // fall through
3450 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3451 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3452 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3453 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3454 default: break;
3455 }
3456 __ bind(L2);
3457 }
3458 }
3459
3460 void TemplateTable::fast_storefield(TosState state) {
3461 transition(state, vtos);
3462
3463 ByteSize base = ConstantPoolCache::base_offset();
3464
3465 jvmti_post_fast_field_mod();
3466
3467 // access constant pool cache
3468 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3469
3470 // test for volatile with rdx but rdx is tos register for lputfield.
3471 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3472 in_bytes(base +
3473 ConstantPoolCacheEntry::flags_offset())));
3474
3475 // replace index with field offset from cache entry
3476 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3477 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3478
3479 // [jk] not needed currently
3480 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3481 // Assembler::StoreStore));
3482
3483 Label notVolatile;
3484 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3485 __ andl(rdx, 0x1);
3486
3487 // Get object from stack
3488 pop_and_check_object(rcx);
3489
3490 // field address
3491 const Address field(rcx, rbx, Address::times_1);
3492
3493 // access field
3494 switch (bytecode()) {
3495 case Bytecodes::_fast_aputfield:
3496 do_oop_store(_masm, field, rax, _bs->kind(), false);
3497 break;
3498 case Bytecodes::_fast_lputfield:
3499 #ifdef _LP64
3500 __ movq(field, rax);
3501 #else
3502 __ stop("should not be rewritten");
3503 #endif
3504 break;
3505 case Bytecodes::_fast_iputfield:
3506 __ movl(field, rax);
3507 break;
3508 case Bytecodes::_fast_zputfield:
3509 __ andl(rax, 0x1); // boolean is true if LSB is 1
3510 // fall through to bputfield
3511 case Bytecodes::_fast_bputfield:
3512 __ movb(field, rax);
3513 break;
3514 case Bytecodes::_fast_sputfield:
3515 // fall through
3516 case Bytecodes::_fast_cputfield:
3517 __ movw(field, rax);
3518 break;
3519 case Bytecodes::_fast_fputfield:
3520 __ store_float(field);
3521 break;
3522 case Bytecodes::_fast_dputfield:
3523 __ store_double(field);
3524 break;
3525 default:
3526 ShouldNotReachHere();
3527 }
3528
3529 // Check for volatile store
3530 __ testl(rdx, rdx);
3531 __ jcc(Assembler::zero, notVolatile);
3532 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3533 Assembler::StoreStore));
3534 __ bind(notVolatile);
3535 }
3536
3537 void TemplateTable::fast_accessfield(TosState state) {
3538 transition(atos, state);
3539
3540 // Do the JVMTI work here to avoid disturbing the register state below
3541 if (JvmtiExport::can_post_field_access()) {
3542 // Check to see if a field access watch has been set before we
3543 // take the time to call into the VM.
3544 Label L1;
3545 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3546 __ testl(rcx, rcx);
3547 __ jcc(Assembler::zero, L1);
3548 // access constant pool cache entry
3549 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3550 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3551 __ verify_oop(rax);
3552 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3553 LP64_ONLY(__ mov(c_rarg1, rax));
3554 // c_rarg1: object pointer copied above
3555 // c_rarg2: cache entry pointer
3556 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3557 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3558 __ pop_ptr(rax); // restore object pointer
3559 __ bind(L1);
3560 }
3561
3562 // access constant pool cache
3563 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3564 // replace index with field offset from cache entry
3565 // [jk] not needed currently
3566 // if (os::is_MP()) {
3567 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3568 // in_bytes(ConstantPoolCache::base_offset() +
3569 // ConstantPoolCacheEntry::flags_offset())));
3570 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3571 // __ andl(rdx, 0x1);
3572 // }
3573 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3574 in_bytes(ConstantPoolCache::base_offset() +
3575 ConstantPoolCacheEntry::f2_offset())));
3576
3577 // rax: object
3578 __ verify_oop(rax);
3579 __ null_check(rax);
3580 Address field(rax, rbx, Address::times_1);
3581
3582 // access field
3583 switch (bytecode()) {
3584 case Bytecodes::_fast_agetfield:
3585 __ load_heap_oop(rax, field);
3586 __ verify_oop(rax);
3587 break;
3588 case Bytecodes::_fast_lgetfield:
3589 #ifdef _LP64
3590 __ movq(rax, field);
3591 #else
3592 __ stop("should not be rewritten");
3593 #endif
3594 break;
3595 case Bytecodes::_fast_igetfield:
3596 __ movl(rax, field);
3597 break;
3598 case Bytecodes::_fast_bgetfield:
3599 __ movsbl(rax, field);
3600 break;
3601 case Bytecodes::_fast_sgetfield:
3602 __ load_signed_short(rax, field);
3603 break;
3604 case Bytecodes::_fast_cgetfield:
3605 __ load_unsigned_short(rax, field);
3606 break;
3607 case Bytecodes::_fast_fgetfield:
3608 __ load_float(field);
3609 break;
3610 case Bytecodes::_fast_dgetfield:
3611 __ load_double(field);
3612 break;
3613 default:
3614 ShouldNotReachHere();
3615 }
3616 // [jk] not needed currently
3617 // if (os::is_MP()) {
3618 // Label notVolatile;
3619 // __ testl(rdx, rdx);
3620 // __ jcc(Assembler::zero, notVolatile);
3621 // __ membar(Assembler::LoadLoad);
3622 // __ bind(notVolatile);
3623 //};
3624 }
3625
3626 void TemplateTable::fast_xaccess(TosState state) {
3627 transition(vtos, state);
3628
3629 // get receiver
3630 __ movptr(rax, aaddress(0));
3631 // access constant pool cache
3632 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3633 __ movptr(rbx,
3634 Address(rcx, rdx, Address::times_ptr,
3635 in_bytes(ConstantPoolCache::base_offset() +
3636 ConstantPoolCacheEntry::f2_offset())));
3637 // make sure exception is reported in correct bcp range (getfield is
3638 // next instruction)
3639 __ increment(rbcp);
3640 __ null_check(rax);
3641 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3642 switch (state) {
3643 case itos:
3644 __ movl(rax, field);
3645 break;
3646 case atos:
3647 __ load_heap_oop(rax, field);
3648 __ verify_oop(rax);
3649 break;
3650 case ftos:
3651 __ load_float(field);
3652 break;
3653 default:
3654 ShouldNotReachHere();
3655 }
3656
3657 // [jk] not needed currently
3658 // if (os::is_MP()) {
3659 // Label notVolatile;
3660 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3661 // in_bytes(ConstantPoolCache::base_offset() +
3662 // ConstantPoolCacheEntry::flags_offset())));
3663 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3664 // __ testl(rdx, 0x1);
3665 // __ jcc(Assembler::zero, notVolatile);
3666 // __ membar(Assembler::LoadLoad);
3667 // __ bind(notVolatile);
3668 // }
3669
3670 __ decrement(rbcp);
3671 }
3672
3673 //-----------------------------------------------------------------------------
3674 // Calls
3675
3676 void TemplateTable::count_calls(Register method, Register temp) {
3677 // implemented elsewhere
3678 ShouldNotReachHere();
3679 }
3680
3681 void TemplateTable::prepare_invoke(int byte_no,
3682 Register method, // linked method (or i-klass)
3683 Register index, // itable index, MethodType, etc.
3684 Register recv, // if caller wants to see it
3685 Register flags // if caller wants to test it
3686 ) {
3687 // determine flags
3688 const Bytecodes::Code code = bytecode();
3689 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3690 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3691 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3692 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3693 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3694 const bool load_receiver = (recv != noreg);
3695 const bool save_flags = (flags != noreg);
3696 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3697 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3698 assert(flags == noreg || flags == rdx, "");
3699 assert(recv == noreg || recv == rcx, "");
3700
3701 // setup registers & access constant pool cache
3702 if (recv == noreg) recv = rcx;
3703 if (flags == noreg) flags = rdx;
3704 assert_different_registers(method, index, recv, flags);
3705
3706 // save 'interpreter return address'
3707 __ save_bcp();
3708
3709 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3710
3711 // maybe push appendix to arguments (just before return address)
3712 if (is_invokedynamic || is_invokehandle) {
3713 Label L_no_push;
3714 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3715 __ jcc(Assembler::zero, L_no_push);
3716 // Push the appendix as a trailing parameter.
3717 // This must be done before we get the receiver,
3718 // since the parameter_size includes it.
3719 __ push(rbx);
3720 __ mov(rbx, index);
3721 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3722 __ load_resolved_reference_at_index(index, rbx);
3723 __ pop(rbx);
3724 __ push(index); // push appendix (MethodType, CallSite, etc.)
3725 __ bind(L_no_push);
3726 }
3727
3728 // load receiver if needed (after appendix is pushed so parameter size is correct)
3729 // Note: no return address pushed yet
3730 if (load_receiver) {
3731 __ movl(recv, flags);
3732 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3733 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3734 const int receiver_is_at_end = -1; // back off one slot to get receiver
3735 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3736 __ movptr(recv, recv_addr);
3737 __ verify_oop(recv);
3738 }
3739
3740 if (save_flags) {
3741 __ movl(rbcp, flags);
3742 }
3743
3744 // compute return type
3745 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3746 // Make sure we don't need to mask flags after the above shift
3747 ConstantPoolCacheEntry::verify_tos_state_shift();
3748 // load return address
3749 {
3750 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3751 ExternalAddress table(table_addr);
3752 LP64_ONLY(__ lea(rscratch1, table));
3753 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3754 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3755 }
3756
3757 // push return address
3758 __ push(flags);
3759
3760 // Restore flags value from the constant pool cache, and restore rsi
3761 // for later null checks. r13 is the bytecode pointer
3762 if (save_flags) {
3763 __ movl(flags, rbcp);
3764 __ restore_bcp();
3765 }
3766 }
3767
3768 void TemplateTable::invokevirtual_helper(Register index,
3769 Register recv,
3770 Register flags) {
3771 // Uses temporary registers rax, rdx
3772 assert_different_registers(index, recv, rax, rdx);
3773 assert(index == rbx, "");
3774 assert(recv == rcx, "");
3775
3776 // Test for an invoke of a final method
3777 Label notFinal;
3778 __ movl(rax, flags);
3779 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3780 __ jcc(Assembler::zero, notFinal);
3781
3782 const Register method = index; // method must be rbx
3783 assert(method == rbx,
3784 "Method* must be rbx for interpreter calling convention");
3785
3786 // do the call - the index is actually the method to call
3787 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3788
3789 // It's final, need a null check here!
3790 __ null_check(recv);
3791
3792 // profile this call
3793 __ profile_final_call(rax);
3794 __ profile_arguments_type(rax, method, rbcp, true);
3795
3796 __ jump_from_interpreted(method, rax);
3797
3798 __ bind(notFinal);
3799
3800 // get receiver klass
3801 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3802 __ load_klass(rax, recv);
3803
3804 // profile this call
3805 __ profile_virtual_call(rax, rlocals, rdx);
3806 // get target Method* & entry point
3807 __ lookup_virtual_method(rax, index, method);
3808 __ profile_called_method(method, rdx, rbcp);
3809
3810 __ profile_arguments_type(rdx, method, rbcp, true);
3811 __ jump_from_interpreted(method, rdx);
3812 }
3813
3814 void TemplateTable::invokevirtual(int byte_no) {
3815 transition(vtos, vtos);
3816 assert(byte_no == f2_byte, "use this argument");
3817 prepare_invoke(byte_no,
3818 rbx, // method or vtable index
3819 noreg, // unused itable index
3820 rcx, rdx); // recv, flags
3821
3822 // rbx: index
3823 // rcx: receiver
3824 // rdx: flags
3825
3826 invokevirtual_helper(rbx, rcx, rdx);
3827 }
3828
3829 void TemplateTable::invokespecial(int byte_no) {
3830 transition(vtos, vtos);
3831 assert(byte_no == f1_byte, "use this argument");
3832 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3833 rcx); // get receiver also for null check
3834 __ verify_oop(rcx);
3835 __ null_check(rcx);
3836 // do the call
3837 __ profile_call(rax);
3838 __ profile_arguments_type(rax, rbx, rbcp, false);
3839 __ jump_from_interpreted(rbx, rax);
3840 }
3841
3842 void TemplateTable::invokestatic(int byte_no) {
3843 transition(vtos, vtos);
3844 assert(byte_no == f1_byte, "use this argument");
3845 prepare_invoke(byte_no, rbx); // get f1 Method*
3846 // do the call
3847 __ profile_call(rax);
3848 __ profile_arguments_type(rax, rbx, rbcp, false);
3849 __ jump_from_interpreted(rbx, rax);
3850 }
3851
3852
3853 void TemplateTable::fast_invokevfinal(int byte_no) {
3854 transition(vtos, vtos);
3855 assert(byte_no == f2_byte, "use this argument");
3856 __ stop("fast_invokevfinal not used on x86");
3857 }
3858
3859
3860 void TemplateTable::invokeinterface(int byte_no) {
3861 transition(vtos, vtos);
3862 assert(byte_no == f1_byte, "use this argument");
3863 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 Method*
3864 rcx, rdx); // recv, flags
3865
3866 // rax: reference klass (from f1)
3867 // rbx: method (from f2)
3868 // rcx: receiver
3869 // rdx: flags
3870
3871 // Special case of invokeinterface called for virtual method of
3872 // java.lang.Object. See cpCacheOop.cpp for details.
3873 // This code isn't produced by javac, but could be produced by
3874 // another compliant java compiler.
3875 Label notMethod;
3876 __ movl(rlocals, rdx);
3877 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3878
3879 __ jcc(Assembler::zero, notMethod);
3880
3881 invokevirtual_helper(rbx, rcx, rdx);
3882 __ bind(notMethod);
3883
3884 // Get receiver klass into rdx - also a null check
3885 __ restore_locals(); // restore r14
3886 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3887 __ load_klass(rdx, rcx);
3888
3889 Label no_such_interface, no_such_method;
3890
3891 // Receiver subtype check against REFC.
3892 // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
3893 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3894 rdx, rax, noreg,
3895 // outputs: scan temp. reg, scan temp. reg
3896 rbcp, rlocals,
3897 no_such_interface,
3898 /*return_method=*/false);
3899
3900 // profile this call
3901 __ restore_bcp(); // rbcp was destroyed by receiver type check
3902 __ profile_virtual_call(rdx, rbcp, rlocals);
3903
3904 // Get declaring interface class from method, and itable index
3905 __ movptr(rax, Address(rbx, Method::const_offset()));
3906 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
3907 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
3908 __ movl(rbx, Address(rbx, Method::itable_index_offset()));
3909 __ subl(rbx, Method::itable_index_max);
3910 __ negl(rbx);
3911
3912 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3913 rdx, rax, rbx,
3914 // outputs: method, scan temp. reg
3915 rbx, rbcp,
3916 no_such_interface);
3917
3918 // rbx: Method* to call
3919 // rcx: receiver
3920 // Check for abstract method error
3921 // Note: This should be done more efficiently via a throw_abstract_method_error
3922 // interpreter entry point and a conditional jump to it in case of a null
3923 // method.
3924 __ testptr(rbx, rbx);
3925 __ jcc(Assembler::zero, no_such_method);
3926
3927 __ profile_called_method(rbx, rbcp, rdx);
3928 __ profile_arguments_type(rdx, rbx, rbcp, true);
3929
3930 // do the call
3931 // rcx: receiver
3932 // rbx,: Method*
3933 __ jump_from_interpreted(rbx, rdx);
3934 __ should_not_reach_here();
3935
3936 // exception handling code follows...
3937 // note: must restore interpreter registers to canonical
3938 // state for exception handling to work correctly!
3939
3940 __ bind(no_such_method);
3941 // throw exception
3942 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3943 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3944 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3945 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3946 // the call_VM checks for exception, so we should never return here.
3947 __ should_not_reach_here();
3948
3949 __ bind(no_such_interface);
3950 // throw exception
3951 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3952 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3953 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3954 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3955 InterpreterRuntime::throw_IncompatibleClassChangeError));
3956 // the call_VM checks for exception, so we should never return here.
3957 __ should_not_reach_here();
3958 }
3959
3960 void TemplateTable::invokehandle(int byte_no) {
3961 transition(vtos, vtos);
3962 assert(byte_no == f1_byte, "use this argument");
3963 const Register rbx_method = rbx;
3964 const Register rax_mtype = rax;
3965 const Register rcx_recv = rcx;
3966 const Register rdx_flags = rdx;
3967
3968 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3969 __ verify_method_ptr(rbx_method);
3970 __ verify_oop(rcx_recv);
3971 __ null_check(rcx_recv);
3972
3973 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3974 // rbx: MH.invokeExact_MT method (from f2)
3975
3976 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3977
3978 // FIXME: profile the LambdaForm also
3979 __ profile_final_call(rax);
3980 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3981
3982 __ jump_from_interpreted(rbx_method, rdx);
3983 }
3984
3985 void TemplateTable::invokedynamic(int byte_no) {
3986 transition(vtos, vtos);
3987 assert(byte_no == f1_byte, "use this argument");
3988
3989 const Register rbx_method = rbx;
3990 const Register rax_callsite = rax;
3991
3992 prepare_invoke(byte_no, rbx_method, rax_callsite);
3993
3994 // rax: CallSite object (from cpool->resolved_references[f1])
3995 // rbx: MH.linkToCallSite method (from f2)
3996
3997 // Note: rax_callsite is already pushed by prepare_invoke
3998
3999 // %%% should make a type profile for any invokedynamic that takes a ref argument
4000 // profile this call
4001 __ profile_call(rbcp);
4002 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
4003
4004 __ verify_oop(rax_callsite);
4005
4006 __ jump_from_interpreted(rbx_method, rdx);
4007 }
4008
4009 //-----------------------------------------------------------------------------
4010 // Allocation
4011
4012 void TemplateTable::_new() {
4013 transition(vtos, atos);
4014 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
4015 Label slow_case;
4016 Label slow_case_no_pop;
4017 Label done;
4018 Label initialize_header;
4019 Label initialize_object; // including clearing the fields
4020
4021 __ get_cpool_and_tags(rcx, rax);
4022
4023 // Make sure the class we're about to instantiate has been resolved.
4024 // This is done before loading InstanceKlass to be consistent with the order
4025 // how Constant Pool is updated (see ConstantPool::klass_at_put)
4026 const int tags_offset = Array<u1>::base_offset_in_bytes();
4027 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
4028 __ jcc(Assembler::notEqual, slow_case_no_pop);
4029
4030 // get InstanceKlass
4031 __ load_resolved_klass_at_index(rcx, rdx, rcx);
4032 __ push(rcx); // save the contexts of klass for initializing the header
4033
4034 // make sure klass is initialized & doesn't have finalizer
4035 // make sure klass is fully initialized
4036 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4037 __ jcc(Assembler::notEqual, slow_case);
4038
4039 // get instance_size in InstanceKlass (scaled to a count of bytes)
4040 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
4041 // test to see if it has a finalizer or is malformed in some way
4042 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
4043 __ jcc(Assembler::notZero, slow_case);
4044
4045 // Allocate the instance:
4046 // If TLAB is enabled:
4047 // Try to allocate in the TLAB.
4048 // If fails, go to the slow path.
4049 // Else If inline contiguous allocations are enabled:
4050 // Try to allocate in eden.
4051 // If fails due to heap end, go to slow path.
4052 //
4053 // If TLAB is enabled OR inline contiguous is enabled:
4054 // Initialize the allocation.
4055 // Exit.
4056 //
4057 // Go to slow path.
4058
4059 const bool allow_shared_alloc =
4060 Universe::heap()->supports_inline_contig_alloc();
4061
4062 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4063 #ifndef _LP64
4064 if (UseTLAB || allow_shared_alloc) {
4065 __ get_thread(thread);
4066 }
4067 #endif // _LP64
4068
4069 if (UseTLAB) {
4070 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
4071 __ lea(rbx, Address(rax, rdx, Address::times_1));
4072 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
4073 __ jcc(Assembler::above, slow_case);
4074 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
4075 if (ZeroTLAB) {
4076 // the fields have been already cleared
4077 __ jmp(initialize_header);
4078 } else {
4079 // initialize both the header and fields
4080 __ jmp(initialize_object);
4081 }
4082 } else {
4083 // Allocation in the shared Eden, if allowed.
4084 //
4085 // rdx: instance size in bytes
4086 if (allow_shared_alloc) {
4087 ExternalAddress heap_top((address)Universe::heap()->top_addr());
4088 ExternalAddress heap_end((address)Universe::heap()->end_addr());
4089
4090 Label retry;
4091 __ bind(retry);
4092 __ movptr(rax, heap_top);
4093 __ lea(rbx, Address(rax, rdx, Address::times_1));
4094 __ cmpptr(rbx, heap_end);
4095 __ jcc(Assembler::above, slow_case);
4096
4097 // Compare rax, with the top addr, and if still equal, store the new
4098 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
4099 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
4100 //
4101 // rax,: object begin
4102 // rbx,: object end
4103 // rdx: instance size in bytes
4104 __ locked_cmpxchgptr(rbx, heap_top);
4105
4106 // if someone beat us on the allocation, try again, otherwise continue
4107 __ jcc(Assembler::notEqual, retry);
4108
4109 __ incr_allocated_bytes(thread, rdx, 0);
4110 }
4111 }
4112
4113 // If UseTLAB or allow_shared_alloc are true, the object is created above and
4114 // there is an initialize need. Otherwise, skip and go to the slow path.
4115 if (UseTLAB || allow_shared_alloc) {
4116 // The object is initialized before the header. If the object size is
4117 // zero, go directly to the header initialization.
4118 __ bind(initialize_object);
4119 __ decrement(rdx, sizeof(oopDesc));
4120 __ jcc(Assembler::zero, initialize_header);
4121
4122 // Initialize topmost object field, divide rdx by 8, check if odd and
4123 // test if zero.
4124 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
4125 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4126
4127 // rdx must have been multiple of 8
4128 #ifdef ASSERT
4129 // make sure rdx was multiple of 8
4130 Label L;
4131 // Ignore partial flag stall after shrl() since it is debug VM
4132 __ jccb(Assembler::carryClear, L);
4133 __ stop("object size is not multiple of 2 - adjust this code");
4134 __ bind(L);
4135 // rdx must be > 0, no extra check needed here
4136 #endif
4137
4138 // initialize remaining object fields: rdx was a multiple of 8
4139 { Label loop;
4140 __ bind(loop);
4141 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4142 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4143 __ decrement(rdx);
4144 __ jcc(Assembler::notZero, loop);
4145 }
4146
4147 // initialize object header only.
4148 __ bind(initialize_header);
4149 if (UseBiasedLocking) {
4150 __ pop(rcx); // get saved klass back in the register.
4151 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4152 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4153 } else {
4154 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4155 (intptr_t)markOopDesc::prototype()); // header
4156 __ pop(rcx); // get saved klass back in the register.
4157 }
4158 #ifdef _LP64
4159 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4160 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
4161 #endif
4162 __ store_klass(rax, rcx); // klass
4163
4164 {
4165 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4166 // Trigger dtrace event for fastpath
4167 __ push(atos);
4168 __ call_VM_leaf(
4169 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4170 __ pop(atos);
4171 }
4172
4173 __ jmp(done);
4174 }
4175
4176 // slow case
4177 __ bind(slow_case);
4178 __ pop(rcx); // restore stack pointer to what it was when we came in.
4179 __ bind(slow_case_no_pop);
4180
4181 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4182 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4183
4184 __ get_constant_pool(rarg1);
4185 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4186 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4187 __ verify_oop(rax);
4188
4189 // continue
4190 __ bind(done);
4191 }
4192
4193 void TemplateTable::newarray() {
4194 transition(itos, atos);
4195 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4196 __ load_unsigned_byte(rarg1, at_bcp(1));
4197 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4198 rarg1, rax);
4199 }
4200
4201 void TemplateTable::anewarray() {
4202 transition(itos, atos);
4203
4204 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4205 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4206
4207 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4208 __ get_constant_pool(rarg1);
4209 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4210 rarg1, rarg2, rax);
4211 }
4212
4213 void TemplateTable::arraylength() {
4214 transition(atos, itos);
4215 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4216 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4217 }
4218
4219 void TemplateTable::checkcast() {
4220 transition(atos, atos);
4221 Label done, is_null, ok_is_subtype, quicked, resolved;
4222 __ testptr(rax, rax); // object is in rax
4223 __ jcc(Assembler::zero, is_null);
4224
4225 // Get cpool & tags index
4226 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4227 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4228 // See if bytecode has already been quicked
4229 __ cmpb(Address(rdx, rbx,
4230 Address::times_1,
4231 Array<u1>::base_offset_in_bytes()),
4232 JVM_CONSTANT_Class);
4233 __ jcc(Assembler::equal, quicked);
4234 __ push(atos); // save receiver for result, and for GC
4235 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4236
4237 // vm_result_2 has metadata result
4238 #ifndef _LP64
4239 // borrow rdi from locals
4240 __ get_thread(rdi);
4241 __ get_vm_result_2(rax, rdi);
4242 __ restore_locals();
4243 #else
4244 __ get_vm_result_2(rax, r15_thread);
4245 #endif
4246
4247 __ pop_ptr(rdx); // restore receiver
4248 __ jmpb(resolved);
4249
4250 // Get superklass in rax and subklass in rbx
4251 __ bind(quicked);
4252 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4253 __ load_resolved_klass_at_index(rcx, rbx, rax);
4254
4255 __ bind(resolved);
4256 __ load_klass(rbx, rdx);
4257
4258 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4259 // Superklass in rax. Subklass in rbx.
4260 __ gen_subtype_check(rbx, ok_is_subtype);
4261
4262 // Come here on failure
4263 __ push_ptr(rdx);
4264 // object is at TOS
4265 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4266
4267 // Come here on success
4268 __ bind(ok_is_subtype);
4269 __ mov(rax, rdx); // Restore object in rdx
4270
4271 // Collect counts on whether this check-cast sees NULLs a lot or not.
4272 if (ProfileInterpreter) {
4273 __ jmp(done);
4274 __ bind(is_null);
4275 __ profile_null_seen(rcx);
4276 } else {
4277 __ bind(is_null); // same as 'done'
4278 }
4279 __ bind(done);
4280 }
4281
4282 void TemplateTable::instanceof() {
4283 transition(atos, itos);
4284 Label done, is_null, ok_is_subtype, quicked, resolved;
4285 __ testptr(rax, rax);
4286 __ jcc(Assembler::zero, is_null);
4287
4288 // Get cpool & tags index
4289 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4290 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4291 // See if bytecode has already been quicked
4292 __ cmpb(Address(rdx, rbx,
4293 Address::times_1,
4294 Array<u1>::base_offset_in_bytes()),
4295 JVM_CONSTANT_Class);
4296 __ jcc(Assembler::equal, quicked);
4297
4298 __ push(atos); // save receiver for result, and for GC
4299 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4300 // vm_result_2 has metadata result
4301
4302 #ifndef _LP64
4303 // borrow rdi from locals
4304 __ get_thread(rdi);
4305 __ get_vm_result_2(rax, rdi);
4306 __ restore_locals();
4307 #else
4308 __ get_vm_result_2(rax, r15_thread);
4309 #endif
4310
4311 __ pop_ptr(rdx); // restore receiver
4312 __ verify_oop(rdx);
4313 __ load_klass(rdx, rdx);
4314 __ jmpb(resolved);
4315
4316 // Get superklass in rax and subklass in rdx
4317 __ bind(quicked);
4318 __ load_klass(rdx, rax);
4319 __ load_resolved_klass_at_index(rcx, rbx, rax);
4320
4321 __ bind(resolved);
4322
4323 // Generate subtype check. Blows rcx, rdi
4324 // Superklass in rax. Subklass in rdx.
4325 __ gen_subtype_check(rdx, ok_is_subtype);
4326
4327 // Come here on failure
4328 __ xorl(rax, rax);
4329 __ jmpb(done);
4330 // Come here on success
4331 __ bind(ok_is_subtype);
4332 __ movl(rax, 1);
4333
4334 // Collect counts on whether this test sees NULLs a lot or not.
4335 if (ProfileInterpreter) {
4336 __ jmp(done);
4337 __ bind(is_null);
4338 __ profile_null_seen(rcx);
4339 } else {
4340 __ bind(is_null); // same as 'done'
4341 }
4342 __ bind(done);
4343 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4344 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4345 }
4346
4347
4348 //----------------------------------------------------------------------------------------------------
4349 // Breakpoints
4350 void TemplateTable::_breakpoint() {
4351 // Note: We get here even if we are single stepping..
4352 // jbug insists on setting breakpoints at every bytecode
4353 // even if we are in single step mode.
4354
4355 transition(vtos, vtos);
4356
4357 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4358
4359 // get the unpatched byte code
4360 __ get_method(rarg);
4361 __ call_VM(noreg,
4362 CAST_FROM_FN_PTR(address,
4363 InterpreterRuntime::get_original_bytecode_at),
4364 rarg, rbcp);
4365 __ mov(rbx, rax); // why?
4366
4367 // post the breakpoint event
4368 __ get_method(rarg);
4369 __ call_VM(noreg,
4370 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4371 rarg, rbcp);
4372
4373 // complete the execution of original bytecode
4374 __ dispatch_only_normal(vtos);
4375 }
4376
4377 //-----------------------------------------------------------------------------
4378 // Exceptions
4379
4380 void TemplateTable::athrow() {
4381 transition(atos, vtos);
4382 __ null_check(rax);
4383 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4384 }
4385
4386 //-----------------------------------------------------------------------------
4387 // Synchronization
4388 //
4389 // Note: monitorenter & exit are symmetric routines; which is reflected
4390 // in the assembly code structure as well
4391 //
4392 // Stack layout:
4393 //
4394 // [expressions ] <--- rsp = expression stack top
4395 // ..
4396 // [expressions ]
4397 // [monitor entry] <--- monitor block top = expression stack bot
4398 // ..
4399 // [monitor entry]
4400 // [frame data ] <--- monitor block bot
4401 // ...
4402 // [saved rbp ] <--- rbp
4403 void TemplateTable::monitorenter() {
4404 transition(atos, vtos);
4405
4406 // check for NULL object
4407 __ null_check(rax);
4408
4409 const Address monitor_block_top(
4410 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4411 const Address monitor_block_bot(
4412 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4413 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4414
4415 Label allocated;
4416
4417 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4418 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4419 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4420
4421 // initialize entry pointer
4422 __ xorl(rmon, rmon); // points to free slot or NULL
4423
4424 // find a free slot in the monitor block (result in rmon)
4425 {
4426 Label entry, loop, exit;
4427 __ movptr(rtop, monitor_block_top); // points to current entry,
4428 // starting with top-most entry
4429 __ lea(rbot, monitor_block_bot); // points to word before bottom
4430 // of monitor block
4431 __ jmpb(entry);
4432
4433 __ bind(loop);
4434 // check if current entry is used
4435 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4436 // if not used then remember entry in rmon
4437 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4438 // check if current entry is for same object
4439 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4440 // if same object then stop searching
4441 __ jccb(Assembler::equal, exit);
4442 // otherwise advance to next entry
4443 __ addptr(rtop, entry_size);
4444 __ bind(entry);
4445 // check if bottom reached
4446 __ cmpptr(rtop, rbot);
4447 // if not at bottom then check this entry
4448 __ jcc(Assembler::notEqual, loop);
4449 __ bind(exit);
4450 }
4451
4452 __ testptr(rmon, rmon); // check if a slot has been found
4453 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4454
4455 // allocate one if there's no free slot
4456 {
4457 Label entry, loop;
4458 // 1. compute new pointers // rsp: old expression stack top
4459 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4460 __ subptr(rsp, entry_size); // move expression stack top
4461 __ subptr(rmon, entry_size); // move expression stack bottom
4462 __ mov(rtop, rsp); // set start value for copy loop
4463 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4464 __ jmp(entry);
4465 // 2. move expression stack contents
4466 __ bind(loop);
4467 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4468 // word from old location
4469 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4470 __ addptr(rtop, wordSize); // advance to next word
4471 __ bind(entry);
4472 __ cmpptr(rtop, rmon); // check if bottom reached
4473 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4474 // copy next word
4475 }
4476
4477 // call run-time routine
4478 // rmon: points to monitor entry
4479 __ bind(allocated);
4480
4481 // Increment bcp to point to the next bytecode, so exception
4482 // handling for async. exceptions work correctly.
4483 // The object has already been poped from the stack, so the
4484 // expression stack looks correct.
4485 __ increment(rbcp);
4486
4487 // store object
4488 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4489 __ lock_object(rmon);
4490
4491 // check to make sure this monitor doesn't cause stack overflow after locking
4492 __ save_bcp(); // in case of exception
4493 __ generate_stack_overflow_check(0);
4494
4495 // The bcp has already been incremented. Just need to dispatch to
4496 // next instruction.
4497 __ dispatch_next(vtos);
4498 }
4499
4500 void TemplateTable::monitorexit() {
4501 transition(atos, vtos);
4502
4503 // check for NULL object
4504 __ null_check(rax);
4505
4506 const Address monitor_block_top(
4507 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4508 const Address monitor_block_bot(
4509 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4510 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4511
4512 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4513 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4514
4515 Label found;
4516
4517 // find matching slot
4518 {
4519 Label entry, loop;
4520 __ movptr(rtop, monitor_block_top); // points to current entry,
4521 // starting with top-most entry
4522 __ lea(rbot, monitor_block_bot); // points to word before bottom
4523 // of monitor block
4524 __ jmpb(entry);
4525
4526 __ bind(loop);
4527 // check if current entry is for same object
4528 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4529 // if same object then stop searching
4530 __ jcc(Assembler::equal, found);
4531 // otherwise advance to next entry
4532 __ addptr(rtop, entry_size);
4533 __ bind(entry);
4534 // check if bottom reached
4535 __ cmpptr(rtop, rbot);
4536 // if not at bottom then check this entry
4537 __ jcc(Assembler::notEqual, loop);
4538 }
4539
4540 // error handling. Unlocking was not block-structured
4541 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4542 InterpreterRuntime::throw_illegal_monitor_state_exception));
4543 __ should_not_reach_here();
4544
4545 // call run-time routine
4546 __ bind(found);
4547 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4548 __ unlock_object(rtop);
4549 __ pop_ptr(rax); // discard object
4550 }
4551
4552 // Wide instructions
4553 void TemplateTable::wide() {
4554 transition(vtos, vtos);
4555 __ load_unsigned_byte(rbx, at_bcp(1));
4556 ExternalAddress wtable((address)Interpreter::_wentry_point);
4557 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4558 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4559 }
4560
4561 // Multi arrays
4562 void TemplateTable::multianewarray() {
4563 transition(vtos, atos);
4564
4565 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4566 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4567 // last dim is on top of stack; we want address of first one:
4568 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4569 // the latter wordSize to point to the beginning of the array.
4570 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4571 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4572 __ load_unsigned_byte(rbx, at_bcp(3));
4573 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4574 }