1 /*
2 * Copyright (c) 1997, 2016, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.inline.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #define __ _masm->
42
43 // Global Register Names
44 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
45 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
46
47 // Platform-dependent initialization
48 void TemplateTable::pd_initialize() {
49 // No x86 specific initialization
50 }
51
52 // Address Computation: local variables
53 static inline Address iaddress(int n) {
54 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
55 }
56
57 static inline Address laddress(int n) {
58 return iaddress(n + 1);
59 }
60
61 #ifndef _LP64
62 static inline Address haddress(int n) {
63 return iaddress(n + 0);
64 }
65 #endif
66
67 static inline Address faddress(int n) {
68 return iaddress(n);
69 }
70
71 static inline Address daddress(int n) {
72 return laddress(n);
73 }
74
75 static inline Address aaddress(int n) {
76 return iaddress(n);
77 }
78
79 static inline Address iaddress(Register r) {
80 return Address(rlocals, r, Address::times_ptr);
81 }
82
83 static inline Address laddress(Register r) {
84 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
85 }
86
87 #ifndef _LP64
88 static inline Address haddress(Register r) {
89 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
90 }
91 #endif
92
93 static inline Address faddress(Register r) {
94 return iaddress(r);
95 }
96
97 static inline Address daddress(Register r) {
98 return laddress(r);
99 }
100
101 static inline Address aaddress(Register r) {
102 return iaddress(r);
103 }
104
105
106 // expression stack
107 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
108 // data beyond the rsp which is potentially unsafe in an MT environment;
109 // an interrupt may overwrite that data.)
110 static inline Address at_rsp () {
111 return Address(rsp, 0);
112 }
113
114 // At top of Java expression stack which may be different than esp(). It
115 // isn't for category 1 objects.
116 static inline Address at_tos () {
117 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
118 }
119
120 static inline Address at_tos_p1() {
121 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
122 }
123
124 static inline Address at_tos_p2() {
125 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
126 }
127
128 // Condition conversion
129 static Assembler::Condition j_not(TemplateTable::Condition cc) {
130 switch (cc) {
131 case TemplateTable::equal : return Assembler::notEqual;
132 case TemplateTable::not_equal : return Assembler::equal;
133 case TemplateTable::less : return Assembler::greaterEqual;
134 case TemplateTable::less_equal : return Assembler::greater;
135 case TemplateTable::greater : return Assembler::lessEqual;
136 case TemplateTable::greater_equal: return Assembler::less;
137 }
138 ShouldNotReachHere();
139 return Assembler::zero;
140 }
141
142
143
144 // Miscelaneous helper routines
145 // Store an oop (or NULL) at the address described by obj.
146 // If val == noreg this means store a NULL
147
148
149 static void do_oop_store(InterpreterMacroAssembler* _masm,
150 Address obj,
151 Register val,
152 BarrierSet::Name barrier,
153 bool precise) {
154 assert(val == noreg || val == rax, "parameter is just for looks");
155 switch (barrier) {
156 #if INCLUDE_ALL_GCS
157 case BarrierSet::G1SATBCTLogging:
158 case BarrierSet::ShenandoahBarrierSet:
159 {
160 // flatten object address if needed
161 // We do it regardless of precise because we need the registers
162 if (obj.index() == noreg && obj.disp() == 0) {
163 if (obj.base() != rdx) {
164 __ movptr(rdx, obj.base());
165 }
166 } else {
167 __ lea(rdx, obj);
168 }
169
170 Register rtmp = LP64_ONLY(r8) NOT_LP64(rsi);
171 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
172
173 NOT_LP64(__ get_thread(rcx));
174 NOT_LP64(__ save_bcp());
175
176 __ g1_write_barrier_pre(rdx /* obj */,
177 rbx /* pre_val */,
178 rthread /* thread */,
179 rtmp /* tmp */,
180 val != noreg /* tosca_live */,
181 false /* expand_call */);
182 if (val == noreg) {
183 __ store_heap_oop_null(Address(rdx, 0));
184 } else {
185 // G1 barrier needs uncompressed oop for region cross check.
186 Register new_val = val;
187 if (UseCompressedOops) {
188 new_val = rbx;
189 __ movptr(new_val, val);
190 }
191 // For Shenandoah, make sure we only store refs into to-space.
192 oopDesc::bs()->interpreter_read_barrier(_masm, val);
193
194 __ store_heap_oop(Address(rdx, 0), val);
195 __ g1_write_barrier_post(rdx /* store_adr */,
196 new_val /* new_val */,
197 rthread /* thread */,
198 rtmp /* tmp */,
199 rbx /* tmp2 */);
200 }
201 NOT_LP64( __ restore_bcp());
202 }
203 break;
204 #endif // INCLUDE_ALL_GCS
205 case BarrierSet::CardTableForRS:
206 case BarrierSet::CardTableExtension:
207 {
208 if (val == noreg) {
209 __ store_heap_oop_null(obj);
210 } else {
211 __ store_heap_oop(obj, val);
212 // flatten object address if needed
213 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
214 __ store_check(obj.base());
215 } else {
216 __ lea(rdx, obj);
217 __ store_check(rdx);
218 }
219 }
220 }
221 break;
222 case BarrierSet::ModRef:
223 if (val == noreg) {
224 __ store_heap_oop_null(obj);
225 } else {
226 __ store_heap_oop(obj, val);
227 }
228 break;
229 default :
230 ShouldNotReachHere();
231
232 }
233 }
234
235 Address TemplateTable::at_bcp(int offset) {
236 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
237 return Address(rbcp, offset);
238 }
239
240
241 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
242 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
243 int byte_no) {
244 if (!RewriteBytecodes) return;
245 Label L_patch_done;
246
247 switch (bc) {
248 case Bytecodes::_fast_aputfield:
249 case Bytecodes::_fast_bputfield:
250 case Bytecodes::_fast_zputfield:
251 case Bytecodes::_fast_cputfield:
252 case Bytecodes::_fast_dputfield:
253 case Bytecodes::_fast_fputfield:
254 case Bytecodes::_fast_iputfield:
255 case Bytecodes::_fast_lputfield:
256 case Bytecodes::_fast_sputfield:
257 {
258 // We skip bytecode quickening for putfield instructions when
259 // the put_code written to the constant pool cache is zero.
260 // This is required so that every execution of this instruction
261 // calls out to InterpreterRuntime::resolve_get_put to do
262 // additional, required work.
263 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
264 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
265 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
266 __ movl(bc_reg, bc);
267 __ cmpl(temp_reg, (int) 0);
268 __ jcc(Assembler::zero, L_patch_done); // don't patch
269 }
270 break;
271 default:
272 assert(byte_no == -1, "sanity");
273 // the pair bytecodes have already done the load.
274 if (load_bc_into_bc_reg) {
275 __ movl(bc_reg, bc);
276 }
277 }
278
279 if (JvmtiExport::can_post_breakpoint()) {
280 Label L_fast_patch;
281 // if a breakpoint is present we can't rewrite the stream directly
282 __ movzbl(temp_reg, at_bcp(0));
283 __ cmpl(temp_reg, Bytecodes::_breakpoint);
284 __ jcc(Assembler::notEqual, L_fast_patch);
285 __ get_method(temp_reg);
286 // Let breakpoint table handling rewrite to quicker bytecode
287 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
288 #ifndef ASSERT
289 __ jmpb(L_patch_done);
290 #else
291 __ jmp(L_patch_done);
292 #endif
293 __ bind(L_fast_patch);
294 }
295
296 #ifdef ASSERT
297 Label L_okay;
298 __ load_unsigned_byte(temp_reg, at_bcp(0));
299 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
300 __ jcc(Assembler::equal, L_okay);
301 __ cmpl(temp_reg, bc_reg);
302 __ jcc(Assembler::equal, L_okay);
303 __ stop("patching the wrong bytecode");
304 __ bind(L_okay);
305 #endif
306
307 // patch bytecode
308 __ movb(at_bcp(0), bc_reg);
309 __ bind(L_patch_done);
310 }
311 // Individual instructions
312
313
314 void TemplateTable::nop() {
315 transition(vtos, vtos);
316 // nothing to do
317 }
318
319 void TemplateTable::shouldnotreachhere() {
320 transition(vtos, vtos);
321 __ stop("shouldnotreachhere bytecode");
322 }
323
324 void TemplateTable::aconst_null() {
325 transition(vtos, atos);
326 __ xorl(rax, rax);
327 }
328
329 void TemplateTable::iconst(int value) {
330 transition(vtos, itos);
331 if (value == 0) {
332 __ xorl(rax, rax);
333 } else {
334 __ movl(rax, value);
335 }
336 }
337
338 void TemplateTable::lconst(int value) {
339 transition(vtos, ltos);
340 if (value == 0) {
341 __ xorl(rax, rax);
342 } else {
343 __ movl(rax, value);
344 }
345 #ifndef _LP64
346 assert(value >= 0, "check this code");
347 __ xorptr(rdx, rdx);
348 #endif
349 }
350
351
352
353 void TemplateTable::fconst(int value) {
354 transition(vtos, ftos);
355 if (UseSSE >= 1) {
356 static float one = 1.0f, two = 2.0f;
357 switch (value) {
358 case 0:
359 __ xorps(xmm0, xmm0);
360 break;
361 case 1:
362 __ movflt(xmm0, ExternalAddress((address) &one));
363 break;
364 case 2:
365 __ movflt(xmm0, ExternalAddress((address) &two));
366 break;
367 default:
368 ShouldNotReachHere();
369 break;
370 }
371 } else {
372 #ifdef _LP64
373 ShouldNotReachHere();
374 #else
375 if (value == 0) { __ fldz();
376 } else if (value == 1) { __ fld1();
377 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
378 } else { ShouldNotReachHere();
379 }
380 #endif // _LP64
381 }
382 }
383
384 void TemplateTable::dconst(int value) {
385 transition(vtos, dtos);
386 if (UseSSE >= 2) {
387 static double one = 1.0;
388 switch (value) {
389 case 0:
390 __ xorpd(xmm0, xmm0);
391 break;
392 case 1:
393 __ movdbl(xmm0, ExternalAddress((address) &one));
394 break;
395 default:
396 ShouldNotReachHere();
397 break;
398 }
399 } else {
400 #ifdef _LP64
401 ShouldNotReachHere();
402 #else
403 if (value == 0) { __ fldz();
404 } else if (value == 1) { __ fld1();
405 } else { ShouldNotReachHere();
406 }
407 #endif
408 }
409 }
410
411 void TemplateTable::bipush() {
412 transition(vtos, itos);
413 __ load_signed_byte(rax, at_bcp(1));
414 }
415
416 void TemplateTable::sipush() {
417 transition(vtos, itos);
418 __ load_unsigned_short(rax, at_bcp(1));
419 __ bswapl(rax);
420 __ sarl(rax, 16);
421 }
422
423 void TemplateTable::ldc(bool wide) {
424 transition(vtos, vtos);
425 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
426 Label call_ldc, notFloat, notClass, Done;
427
428 if (wide) {
429 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
430 } else {
431 __ load_unsigned_byte(rbx, at_bcp(1));
432 }
433
434 __ get_cpool_and_tags(rcx, rax);
435 const int base_offset = ConstantPool::header_size() * wordSize;
436 const int tags_offset = Array<u1>::base_offset_in_bytes();
437
438 // get type
439 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
440
441 // unresolved class - get the resolved class
442 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
443 __ jccb(Assembler::equal, call_ldc);
444
445 // unresolved class in error state - call into runtime to throw the error
446 // from the first resolution attempt
447 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
448 __ jccb(Assembler::equal, call_ldc);
449
450 // resolved class - need to call vm to get java mirror of the class
451 __ cmpl(rdx, JVM_CONSTANT_Class);
452 __ jcc(Assembler::notEqual, notClass);
453
454 __ bind(call_ldc);
455
456 __ movl(rarg, wide);
457 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
458
459 __ push(atos);
460 __ jmp(Done);
461
462 __ bind(notClass);
463 __ cmpl(rdx, JVM_CONSTANT_Float);
464 __ jccb(Assembler::notEqual, notFloat);
465
466 // ftos
467 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
468 __ push(ftos);
469 __ jmp(Done);
470
471 __ bind(notFloat);
472 #ifdef ASSERT
473 {
474 Label L;
475 __ cmpl(rdx, JVM_CONSTANT_Integer);
476 __ jcc(Assembler::equal, L);
477 // String and Object are rewritten to fast_aldc
478 __ stop("unexpected tag type in ldc");
479 __ bind(L);
480 }
481 #endif
482 // itos JVM_CONSTANT_Integer only
483 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
484 __ push(itos);
485 __ bind(Done);
486 }
487
488 // Fast path for caching oop constants.
489 void TemplateTable::fast_aldc(bool wide) {
490 transition(vtos, atos);
491
492 Register result = rax;
493 Register tmp = rdx;
494 int index_size = wide ? sizeof(u2) : sizeof(u1);
495
496 Label resolved;
497
498 // We are resolved if the resolved reference cache entry contains a
499 // non-null object (String, MethodType, etc.)
500 assert_different_registers(result, tmp);
501 __ get_cache_index_at_bcp(tmp, 1, index_size);
502 __ load_resolved_reference_at_index(result, tmp);
503 __ testl(result, result);
504 __ jcc(Assembler::notZero, resolved);
505
506 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
507
508 // first time invocation - must resolve first
509 __ movl(tmp, (int)bytecode());
510 __ call_VM(result, entry, tmp);
511
512 __ bind(resolved);
513
514 if (VerifyOops) {
515 __ verify_oop(result);
516 }
517 }
518
519 void TemplateTable::ldc2_w() {
520 transition(vtos, vtos);
521 Label Long, Done;
522 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
523
524 __ get_cpool_and_tags(rcx, rax);
525 const int base_offset = ConstantPool::header_size() * wordSize;
526 const int tags_offset = Array<u1>::base_offset_in_bytes();
527
528 // get type
529 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
530 JVM_CONSTANT_Double);
531 __ jccb(Assembler::notEqual, Long);
532
533 // dtos
534 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
535 __ push(dtos);
536
537 __ jmpb(Done);
538 __ bind(Long);
539
540 // ltos
541 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
542 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
543 __ push(ltos);
544
545 __ bind(Done);
546 }
547
548 void TemplateTable::locals_index(Register reg, int offset) {
549 __ load_unsigned_byte(reg, at_bcp(offset));
550 __ negptr(reg);
551 }
552
553 void TemplateTable::iload() {
554 iload_internal();
555 }
556
557 void TemplateTable::nofast_iload() {
558 iload_internal(may_not_rewrite);
559 }
560
561 void TemplateTable::iload_internal(RewriteControl rc) {
562 transition(vtos, itos);
563 if (RewriteFrequentPairs && rc == may_rewrite) {
564 Label rewrite, done;
565 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
566 LP64_ONLY(assert(rbx != bc, "register damaged"));
567
568 // get next byte
569 __ load_unsigned_byte(rbx,
570 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
571 // if _iload, wait to rewrite to iload2. We only want to rewrite the
572 // last two iloads in a pair. Comparing against fast_iload means that
573 // the next bytecode is neither an iload or a caload, and therefore
574 // an iload pair.
575 __ cmpl(rbx, Bytecodes::_iload);
576 __ jcc(Assembler::equal, done);
577
578 __ cmpl(rbx, Bytecodes::_fast_iload);
579 __ movl(bc, Bytecodes::_fast_iload2);
580
581 __ jccb(Assembler::equal, rewrite);
582
583 // if _caload, rewrite to fast_icaload
584 __ cmpl(rbx, Bytecodes::_caload);
585 __ movl(bc, Bytecodes::_fast_icaload);
586 __ jccb(Assembler::equal, rewrite);
587
588 // rewrite so iload doesn't check again.
589 __ movl(bc, Bytecodes::_fast_iload);
590
591 // rewrite
592 // bc: fast bytecode
593 __ bind(rewrite);
594 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
595 __ bind(done);
596 }
597
598 // Get the local value into tos
599 locals_index(rbx);
600 __ movl(rax, iaddress(rbx));
601 }
602
603 void TemplateTable::fast_iload2() {
604 transition(vtos, itos);
605 locals_index(rbx);
606 __ movl(rax, iaddress(rbx));
607 __ push(itos);
608 locals_index(rbx, 3);
609 __ movl(rax, iaddress(rbx));
610 }
611
612 void TemplateTable::fast_iload() {
613 transition(vtos, itos);
614 locals_index(rbx);
615 __ movl(rax, iaddress(rbx));
616 }
617
618 void TemplateTable::lload() {
619 transition(vtos, ltos);
620 locals_index(rbx);
621 __ movptr(rax, laddress(rbx));
622 NOT_LP64(__ movl(rdx, haddress(rbx)));
623 }
624
625 void TemplateTable::fload() {
626 transition(vtos, ftos);
627 locals_index(rbx);
628 __ load_float(faddress(rbx));
629 }
630
631 void TemplateTable::dload() {
632 transition(vtos, dtos);
633 locals_index(rbx);
634 __ load_double(daddress(rbx));
635 }
636
637 void TemplateTable::aload() {
638 transition(vtos, atos);
639 locals_index(rbx);
640 __ movptr(rax, aaddress(rbx));
641 }
642
643 void TemplateTable::locals_index_wide(Register reg) {
644 __ load_unsigned_short(reg, at_bcp(2));
645 __ bswapl(reg);
646 __ shrl(reg, 16);
647 __ negptr(reg);
648 }
649
650 void TemplateTable::wide_iload() {
651 transition(vtos, itos);
652 locals_index_wide(rbx);
653 __ movl(rax, iaddress(rbx));
654 }
655
656 void TemplateTable::wide_lload() {
657 transition(vtos, ltos);
658 locals_index_wide(rbx);
659 __ movptr(rax, laddress(rbx));
660 NOT_LP64(__ movl(rdx, haddress(rbx)));
661 }
662
663 void TemplateTable::wide_fload() {
664 transition(vtos, ftos);
665 locals_index_wide(rbx);
666 __ load_float(faddress(rbx));
667 }
668
669 void TemplateTable::wide_dload() {
670 transition(vtos, dtos);
671 locals_index_wide(rbx);
672 __ load_double(daddress(rbx));
673 }
674
675 void TemplateTable::wide_aload() {
676 transition(vtos, atos);
677 locals_index_wide(rbx);
678 __ movptr(rax, aaddress(rbx));
679 }
680
681 void TemplateTable::index_check(Register array, Register index) {
682 // Pop ptr into array
683 __ pop_ptr(array);
684 index_check_without_pop(array, index);
685 }
686
687 void TemplateTable::index_check_without_pop(Register array, Register index) {
688 // destroys rbx
689 // check array
690
691 if (ShenandoahVerifyReadsToFromSpace) {
692 oopDesc::bs()->interpreter_read_barrier(_masm, array);
693 }
694
695 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
696 // sign extend index for use by indexed load
697 __ movl2ptr(index, index);
698 // check index
699 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
700 if (index != rbx) {
701 // ??? convention: move aberrant index into rbx for exception message
702 assert(rbx != array, "different registers");
703 __ movl(rbx, index);
704 }
705 __ jump_cc(Assembler::aboveEqual,
706 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
707 }
708
709
710 void TemplateTable::iaload() {
711 transition(itos, itos);
712 // rax: index
713 // rdx: array
714 index_check(rdx, rax); // kills rbx
715 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
716 __ movl(rax, Address(rdx, rax,
717 Address::times_4,
718 arrayOopDesc::base_offset_in_bytes(T_INT)));
719 }
720
721 void TemplateTable::laload() {
722 transition(itos, ltos);
723 // rax: index
724 // rdx: array
725 index_check(rdx, rax); // kills rbx
726 NOT_LP64(__ mov(rbx, rax));
727 // rbx,: index
728 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
729 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
730 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
731 }
732
733
734
735 void TemplateTable::faload() {
736 transition(itos, ftos);
737 // rax: index
738 // rdx: array
739 index_check(rdx, rax); // kills rbx
740 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
741 __ load_float(Address(rdx, rax,
742 Address::times_4,
743 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
744 }
745
746 void TemplateTable::daload() {
747 transition(itos, dtos);
748 // rax: index
749 // rdx: array
750 index_check(rdx, rax); // kills rbx
751 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
752 __ load_double(Address(rdx, rax,
753 Address::times_8,
754 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
755 }
756
757 void TemplateTable::aaload() {
758 transition(itos, atos);
759 // rax: index
760 // rdx: array
761 index_check(rdx, rax); // kills rbx
762 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
763 __ load_heap_oop(rax, Address(rdx, rax,
764 UseCompressedOops ? Address::times_4 : Address::times_ptr,
765 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
766 }
767
768 void TemplateTable::baload() {
769 transition(itos, itos);
770 // rax: index
771 // rdx: array
772 index_check(rdx, rax); // kills rbx
773 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
774 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
775 }
776
777 void TemplateTable::caload() {
778 transition(itos, itos);
779 // rax: index
780 // rdx: array
781 index_check(rdx, rax); // kills rbx
782 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
783 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
784 }
785
786 // iload followed by caload frequent pair
787 void TemplateTable::fast_icaload() {
788 transition(vtos, itos);
789 // load index out of locals
790 locals_index(rbx);
791 __ movl(rax, iaddress(rbx));
792
793 // rax: index
794 // rdx: array
795 index_check(rdx, rax); // kills rbx
796 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
797 __ load_unsigned_short(rax,
798 Address(rdx, rax,
799 Address::times_2,
800 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
801 }
802
803
804 void TemplateTable::saload() {
805 transition(itos, itos);
806 // rax: index
807 // rdx: array
808 index_check(rdx, rax); // kills rbx
809 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
810 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
811 }
812
813 void TemplateTable::iload(int n) {
814 transition(vtos, itos);
815 __ movl(rax, iaddress(n));
816 }
817
818 void TemplateTable::lload(int n) {
819 transition(vtos, ltos);
820 __ movptr(rax, laddress(n));
821 NOT_LP64(__ movptr(rdx, haddress(n)));
822 }
823
824 void TemplateTable::fload(int n) {
825 transition(vtos, ftos);
826 __ load_float(faddress(n));
827 }
828
829 void TemplateTable::dload(int n) {
830 transition(vtos, dtos);
831 __ load_double(daddress(n));
832 }
833
834 void TemplateTable::aload(int n) {
835 transition(vtos, atos);
836 __ movptr(rax, aaddress(n));
837 }
838
839 void TemplateTable::aload_0() {
840 aload_0_internal();
841 }
842
843 void TemplateTable::nofast_aload_0() {
844 aload_0_internal(may_not_rewrite);
845 }
846
847 void TemplateTable::aload_0_internal(RewriteControl rc) {
848 transition(vtos, atos);
849 // According to bytecode histograms, the pairs:
850 //
851 // _aload_0, _fast_igetfield
852 // _aload_0, _fast_agetfield
853 // _aload_0, _fast_fgetfield
854 //
855 // occur frequently. If RewriteFrequentPairs is set, the (slow)
856 // _aload_0 bytecode checks if the next bytecode is either
857 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
858 // rewrites the current bytecode into a pair bytecode; otherwise it
859 // rewrites the current bytecode into _fast_aload_0 that doesn't do
860 // the pair check anymore.
861 //
862 // Note: If the next bytecode is _getfield, the rewrite must be
863 // delayed, otherwise we may miss an opportunity for a pair.
864 //
865 // Also rewrite frequent pairs
866 // aload_0, aload_1
867 // aload_0, iload_1
868 // These bytecodes with a small amount of code are most profitable
869 // to rewrite
870 if (RewriteFrequentPairs && rc == may_rewrite) {
871 Label rewrite, done;
872
873 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
874 LP64_ONLY(assert(rbx != bc, "register damaged"));
875
876 // get next byte
877 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
878
879 // if _getfield then wait with rewrite
880 __ cmpl(rbx, Bytecodes::_getfield);
881 __ jcc(Assembler::equal, done);
882
883 // if _igetfield then rewrite to _fast_iaccess_0
884 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
885 __ cmpl(rbx, Bytecodes::_fast_igetfield);
886 __ movl(bc, Bytecodes::_fast_iaccess_0);
887 __ jccb(Assembler::equal, rewrite);
888
889 // if _agetfield then rewrite to _fast_aaccess_0
890 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
891 __ cmpl(rbx, Bytecodes::_fast_agetfield);
892 __ movl(bc, Bytecodes::_fast_aaccess_0);
893 __ jccb(Assembler::equal, rewrite);
894
895 // if _fgetfield then rewrite to _fast_faccess_0
896 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
897 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
898 __ movl(bc, Bytecodes::_fast_faccess_0);
899 __ jccb(Assembler::equal, rewrite);
900
901 // else rewrite to _fast_aload0
902 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
903 __ movl(bc, Bytecodes::_fast_aload_0);
904
905 // rewrite
906 // bc: fast bytecode
907 __ bind(rewrite);
908 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
909
910 __ bind(done);
911 }
912
913 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
914 aload(0);
915 }
916
917 void TemplateTable::istore() {
918 transition(itos, vtos);
919 locals_index(rbx);
920 __ movl(iaddress(rbx), rax);
921 }
922
923
924 void TemplateTable::lstore() {
925 transition(ltos, vtos);
926 locals_index(rbx);
927 __ movptr(laddress(rbx), rax);
928 NOT_LP64(__ movptr(haddress(rbx), rdx));
929 }
930
931 void TemplateTable::fstore() {
932 transition(ftos, vtos);
933 locals_index(rbx);
934 __ store_float(faddress(rbx));
935 }
936
937 void TemplateTable::dstore() {
938 transition(dtos, vtos);
939 locals_index(rbx);
940 __ store_double(daddress(rbx));
941 }
942
943 void TemplateTable::astore() {
944 transition(vtos, vtos);
945 __ pop_ptr(rax);
946 locals_index(rbx);
947 __ movptr(aaddress(rbx), rax);
948 }
949
950 void TemplateTable::wide_istore() {
951 transition(vtos, vtos);
952 __ pop_i();
953 locals_index_wide(rbx);
954 __ movl(iaddress(rbx), rax);
955 }
956
957 void TemplateTable::wide_lstore() {
958 transition(vtos, vtos);
959 NOT_LP64(__ pop_l(rax, rdx));
960 LP64_ONLY(__ pop_l());
961 locals_index_wide(rbx);
962 __ movptr(laddress(rbx), rax);
963 NOT_LP64(__ movl(haddress(rbx), rdx));
964 }
965
966 void TemplateTable::wide_fstore() {
967 #ifdef _LP64
968 transition(vtos, vtos);
969 __ pop_f(xmm0);
970 locals_index_wide(rbx);
971 __ movflt(faddress(rbx), xmm0);
972 #else
973 wide_istore();
974 #endif
975 }
976
977 void TemplateTable::wide_dstore() {
978 #ifdef _LP64
979 transition(vtos, vtos);
980 __ pop_d(xmm0);
981 locals_index_wide(rbx);
982 __ movdbl(daddress(rbx), xmm0);
983 #else
984 wide_lstore();
985 #endif
986 }
987
988 void TemplateTable::wide_astore() {
989 transition(vtos, vtos);
990 __ pop_ptr(rax);
991 locals_index_wide(rbx);
992 __ movptr(aaddress(rbx), rax);
993 }
994
995 void TemplateTable::iastore() {
996 transition(itos, vtos);
997 __ pop_i(rbx);
998 // rax: value
999 // rbx: index
1000 // rdx: array
1001 index_check(rdx, rbx); // prefer index in rbx
1002 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1003 __ movl(Address(rdx, rbx,
1004 Address::times_4,
1005 arrayOopDesc::base_offset_in_bytes(T_INT)),
1006 rax);
1007 }
1008
1009 void TemplateTable::lastore() {
1010 transition(ltos, vtos);
1011 __ pop_i(rbx);
1012 // rax,: low(value)
1013 // rcx: array
1014 // rdx: high(value)
1015 index_check(rcx, rbx); // prefer index in rbx,
1016 // rbx,: index
1017 oopDesc::bs()->interpreter_write_barrier(_masm, rcx);
1018 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1019 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1020 }
1021
1022
1023 void TemplateTable::fastore() {
1024 transition(ftos, vtos);
1025 __ pop_i(rbx);
1026 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1027 // rbx: index
1028 // rdx: array
1029 index_check(rdx, rbx); // prefer index in rbx
1030 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1031 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1032 }
1033
1034 void TemplateTable::dastore() {
1035 transition(dtos, vtos);
1036 __ pop_i(rbx);
1037 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1038 // rbx: index
1039 // rdx: array
1040 index_check(rdx, rbx); // prefer index in rbx
1041 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1042 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1043 }
1044
1045 void TemplateTable::aastore() {
1046 Label is_null, ok_is_subtype, done;
1047 transition(vtos, vtos);
1048 // stack: ..., array, index, value
1049 __ movptr(rax, at_tos()); // value
1050 __ movl(rcx, at_tos_p1()); // index
1051 __ movptr(rdx, at_tos_p2()); // array
1052
1053 Address element_address(rdx, rcx,
1054 UseCompressedOops? Address::times_4 : Address::times_ptr,
1055 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1056
1057 index_check_without_pop(rdx, rcx); // kills rbx
1058 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1059 __ testptr(rax, rax);
1060 __ jcc(Assembler::zero, is_null);
1061
1062 // Move subklass into rbx
1063 __ load_klass(rbx, rax);
1064 // Move superklass into rax
1065 __ load_klass(rax, rdx);
1066 __ movptr(rax, Address(rax,
1067 ObjArrayKlass::element_klass_offset()));
1068 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1069 __ lea(rdx, element_address);
1070
1071 // Generate subtype check. Blows rcx, rdi
1072 // Superklass in rax. Subklass in rbx.
1073 __ gen_subtype_check(rbx, ok_is_subtype);
1074
1075 // Come here on failure
1076 // object is at TOS
1077 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1078
1079 // Come here on success
1080 __ bind(ok_is_subtype);
1081
1082 // Get the value we will store
1083 __ movptr(rax, at_tos());
1084 // Now store using the appropriate barrier
1085 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1086 __ jmp(done);
1087
1088 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1089 __ bind(is_null);
1090 __ profile_null_seen(rbx);
1091
1092 // Store a NULL
1093 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1094
1095 // Pop stack arguments
1096 __ bind(done);
1097 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1098 }
1099
1100 void TemplateTable::bastore() {
1101 transition(itos, vtos);
1102 __ pop_i(rbx);
1103 // rax: value
1104 // rbx: index
1105 // rdx: array
1106 index_check(rdx, rbx); // prefer index in rbx
1107 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1108 // Need to check whether array is boolean or byte
1109 // since both types share the bastore bytecode.
1110 __ load_klass(rcx, rdx);
1111 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1112 int diffbit = Klass::layout_helper_boolean_diffbit();
1113 __ testl(rcx, diffbit);
1114 Label L_skip;
1115 __ jccb(Assembler::zero, L_skip);
1116 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1117 __ bind(L_skip);
1118 __ movb(Address(rdx, rbx,
1119 Address::times_1,
1120 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1121 rax);
1122 }
1123
1124 void TemplateTable::castore() {
1125 transition(itos, vtos);
1126 __ pop_i(rbx);
1127 // rax: value
1128 // rbx: index
1129 // rdx: array
1130 index_check(rdx, rbx); // prefer index in rbx
1131 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1132 __ movw(Address(rdx, rbx,
1133 Address::times_2,
1134 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1135 rax);
1136 }
1137
1138
1139 void TemplateTable::sastore() {
1140 castore();
1141 }
1142
1143 void TemplateTable::istore(int n) {
1144 transition(itos, vtos);
1145 __ movl(iaddress(n), rax);
1146 }
1147
1148 void TemplateTable::lstore(int n) {
1149 transition(ltos, vtos);
1150 __ movptr(laddress(n), rax);
1151 NOT_LP64(__ movptr(haddress(n), rdx));
1152 }
1153
1154 void TemplateTable::fstore(int n) {
1155 transition(ftos, vtos);
1156 __ store_float(faddress(n));
1157 }
1158
1159 void TemplateTable::dstore(int n) {
1160 transition(dtos, vtos);
1161 __ store_double(daddress(n));
1162 }
1163
1164
1165 void TemplateTable::astore(int n) {
1166 transition(vtos, vtos);
1167 __ pop_ptr(rax);
1168 __ movptr(aaddress(n), rax);
1169 }
1170
1171 void TemplateTable::pop() {
1172 transition(vtos, vtos);
1173 __ addptr(rsp, Interpreter::stackElementSize);
1174 }
1175
1176 void TemplateTable::pop2() {
1177 transition(vtos, vtos);
1178 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1179 }
1180
1181
1182 void TemplateTable::dup() {
1183 transition(vtos, vtos);
1184 __ load_ptr(0, rax);
1185 __ push_ptr(rax);
1186 // stack: ..., a, a
1187 }
1188
1189 void TemplateTable::dup_x1() {
1190 transition(vtos, vtos);
1191 // stack: ..., a, b
1192 __ load_ptr( 0, rax); // load b
1193 __ load_ptr( 1, rcx); // load a
1194 __ store_ptr(1, rax); // store b
1195 __ store_ptr(0, rcx); // store a
1196 __ push_ptr(rax); // push b
1197 // stack: ..., b, a, b
1198 }
1199
1200 void TemplateTable::dup_x2() {
1201 transition(vtos, vtos);
1202 // stack: ..., a, b, c
1203 __ load_ptr( 0, rax); // load c
1204 __ load_ptr( 2, rcx); // load a
1205 __ store_ptr(2, rax); // store c in a
1206 __ push_ptr(rax); // push c
1207 // stack: ..., c, b, c, c
1208 __ load_ptr( 2, rax); // load b
1209 __ store_ptr(2, rcx); // store a in b
1210 // stack: ..., c, a, c, c
1211 __ store_ptr(1, rax); // store b in c
1212 // stack: ..., c, a, b, c
1213 }
1214
1215 void TemplateTable::dup2() {
1216 transition(vtos, vtos);
1217 // stack: ..., a, b
1218 __ load_ptr(1, rax); // load a
1219 __ push_ptr(rax); // push a
1220 __ load_ptr(1, rax); // load b
1221 __ push_ptr(rax); // push b
1222 // stack: ..., a, b, a, b
1223 }
1224
1225
1226 void TemplateTable::dup2_x1() {
1227 transition(vtos, vtos);
1228 // stack: ..., a, b, c
1229 __ load_ptr( 0, rcx); // load c
1230 __ load_ptr( 1, rax); // load b
1231 __ push_ptr(rax); // push b
1232 __ push_ptr(rcx); // push c
1233 // stack: ..., a, b, c, b, c
1234 __ store_ptr(3, rcx); // store c in b
1235 // stack: ..., a, c, c, b, c
1236 __ load_ptr( 4, rcx); // load a
1237 __ store_ptr(2, rcx); // store a in 2nd c
1238 // stack: ..., a, c, a, b, c
1239 __ store_ptr(4, rax); // store b in a
1240 // stack: ..., b, c, a, b, c
1241 }
1242
1243 void TemplateTable::dup2_x2() {
1244 transition(vtos, vtos);
1245 // stack: ..., a, b, c, d
1246 __ load_ptr( 0, rcx); // load d
1247 __ load_ptr( 1, rax); // load c
1248 __ push_ptr(rax); // push c
1249 __ push_ptr(rcx); // push d
1250 // stack: ..., a, b, c, d, c, d
1251 __ load_ptr( 4, rax); // load b
1252 __ store_ptr(2, rax); // store b in d
1253 __ store_ptr(4, rcx); // store d in b
1254 // stack: ..., a, d, c, b, c, d
1255 __ load_ptr( 5, rcx); // load a
1256 __ load_ptr( 3, rax); // load c
1257 __ store_ptr(3, rcx); // store a in c
1258 __ store_ptr(5, rax); // store c in a
1259 // stack: ..., c, d, a, b, c, d
1260 }
1261
1262 void TemplateTable::swap() {
1263 transition(vtos, vtos);
1264 // stack: ..., a, b
1265 __ load_ptr( 1, rcx); // load a
1266 __ load_ptr( 0, rax); // load b
1267 __ store_ptr(0, rcx); // store a in b
1268 __ store_ptr(1, rax); // store b in a
1269 // stack: ..., b, a
1270 }
1271
1272 void TemplateTable::iop2(Operation op) {
1273 transition(itos, itos);
1274 switch (op) {
1275 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1276 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1277 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1278 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1279 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1280 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1281 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1282 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1283 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1284 default : ShouldNotReachHere();
1285 }
1286 }
1287
1288 void TemplateTable::lop2(Operation op) {
1289 transition(ltos, ltos);
1290 #ifdef _LP64
1291 switch (op) {
1292 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1293 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1294 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1295 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1296 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1297 default : ShouldNotReachHere();
1298 }
1299 #else
1300 __ pop_l(rbx, rcx);
1301 switch (op) {
1302 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1303 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1304 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1305 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1306 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1307 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1308 default : ShouldNotReachHere();
1309 }
1310 #endif
1311 }
1312
1313 void TemplateTable::idiv() {
1314 transition(itos, itos);
1315 __ movl(rcx, rax);
1316 __ pop_i(rax);
1317 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1318 // they are not equal, one could do a normal division (no correction
1319 // needed), which may speed up this implementation for the common case.
1320 // (see also JVM spec., p.243 & p.271)
1321 __ corrected_idivl(rcx);
1322 }
1323
1324 void TemplateTable::irem() {
1325 transition(itos, itos);
1326 __ movl(rcx, rax);
1327 __ pop_i(rax);
1328 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1329 // they are not equal, one could do a normal division (no correction
1330 // needed), which may speed up this implementation for the common case.
1331 // (see also JVM spec., p.243 & p.271)
1332 __ corrected_idivl(rcx);
1333 __ movl(rax, rdx);
1334 }
1335
1336 void TemplateTable::lmul() {
1337 transition(ltos, ltos);
1338 #ifdef _LP64
1339 __ pop_l(rdx);
1340 __ imulq(rax, rdx);
1341 #else
1342 __ pop_l(rbx, rcx);
1343 __ push(rcx); __ push(rbx);
1344 __ push(rdx); __ push(rax);
1345 __ lmul(2 * wordSize, 0);
1346 __ addptr(rsp, 4 * wordSize); // take off temporaries
1347 #endif
1348 }
1349
1350 void TemplateTable::ldiv() {
1351 transition(ltos, ltos);
1352 #ifdef _LP64
1353 __ mov(rcx, rax);
1354 __ pop_l(rax);
1355 // generate explicit div0 check
1356 __ testq(rcx, rcx);
1357 __ jump_cc(Assembler::zero,
1358 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1359 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1360 // they are not equal, one could do a normal division (no correction
1361 // needed), which may speed up this implementation for the common case.
1362 // (see also JVM spec., p.243 & p.271)
1363 __ corrected_idivq(rcx); // kills rbx
1364 #else
1365 __ pop_l(rbx, rcx);
1366 __ push(rcx); __ push(rbx);
1367 __ push(rdx); __ push(rax);
1368 // check if y = 0
1369 __ orl(rax, rdx);
1370 __ jump_cc(Assembler::zero,
1371 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1372 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1373 __ addptr(rsp, 4 * wordSize); // take off temporaries
1374 #endif
1375 }
1376
1377 void TemplateTable::lrem() {
1378 transition(ltos, ltos);
1379 #ifdef _LP64
1380 __ mov(rcx, rax);
1381 __ pop_l(rax);
1382 __ testq(rcx, rcx);
1383 __ jump_cc(Assembler::zero,
1384 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1385 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1386 // they are not equal, one could do a normal division (no correction
1387 // needed), which may speed up this implementation for the common case.
1388 // (see also JVM spec., p.243 & p.271)
1389 __ corrected_idivq(rcx); // kills rbx
1390 __ mov(rax, rdx);
1391 #else
1392 __ pop_l(rbx, rcx);
1393 __ push(rcx); __ push(rbx);
1394 __ push(rdx); __ push(rax);
1395 // check if y = 0
1396 __ orl(rax, rdx);
1397 __ jump_cc(Assembler::zero,
1398 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1399 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1400 __ addptr(rsp, 4 * wordSize);
1401 #endif
1402 }
1403
1404 void TemplateTable::lshl() {
1405 transition(itos, ltos);
1406 __ movl(rcx, rax); // get shift count
1407 #ifdef _LP64
1408 __ pop_l(rax); // get shift value
1409 __ shlq(rax);
1410 #else
1411 __ pop_l(rax, rdx); // get shift value
1412 __ lshl(rdx, rax);
1413 #endif
1414 }
1415
1416 void TemplateTable::lshr() {
1417 #ifdef _LP64
1418 transition(itos, ltos);
1419 __ movl(rcx, rax); // get shift count
1420 __ pop_l(rax); // get shift value
1421 __ sarq(rax);
1422 #else
1423 transition(itos, ltos);
1424 __ mov(rcx, rax); // get shift count
1425 __ pop_l(rax, rdx); // get shift value
1426 __ lshr(rdx, rax, true);
1427 #endif
1428 }
1429
1430 void TemplateTable::lushr() {
1431 transition(itos, ltos);
1432 #ifdef _LP64
1433 __ movl(rcx, rax); // get shift count
1434 __ pop_l(rax); // get shift value
1435 __ shrq(rax);
1436 #else
1437 __ mov(rcx, rax); // get shift count
1438 __ pop_l(rax, rdx); // get shift value
1439 __ lshr(rdx, rax);
1440 #endif
1441 }
1442
1443 void TemplateTable::fop2(Operation op) {
1444 transition(ftos, ftos);
1445
1446 if (UseSSE >= 1) {
1447 switch (op) {
1448 case add:
1449 __ addss(xmm0, at_rsp());
1450 __ addptr(rsp, Interpreter::stackElementSize);
1451 break;
1452 case sub:
1453 __ movflt(xmm1, xmm0);
1454 __ pop_f(xmm0);
1455 __ subss(xmm0, xmm1);
1456 break;
1457 case mul:
1458 __ mulss(xmm0, at_rsp());
1459 __ addptr(rsp, Interpreter::stackElementSize);
1460 break;
1461 case div:
1462 __ movflt(xmm1, xmm0);
1463 __ pop_f(xmm0);
1464 __ divss(xmm0, xmm1);
1465 break;
1466 case rem:
1467 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1468 // modulo operation. The frem method calls the function
1469 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1470 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1471 // (signalling or quiet) is returned.
1472 //
1473 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1474 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1475 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1476 // The fprem instruction used on x86_32 is functionally equivalent to
1477 // SharedRuntime::frem in that it returns a NaN.
1478 #ifdef _LP64
1479 __ movflt(xmm1, xmm0);
1480 __ pop_f(xmm0);
1481 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1482 #else
1483 __ push_f(xmm0);
1484 __ pop_f();
1485 __ fld_s(at_rsp());
1486 __ fremr(rax);
1487 __ f2ieee();
1488 __ pop(rax); // pop second operand off the stack
1489 __ push_f();
1490 __ pop_f(xmm0);
1491 #endif
1492 break;
1493 default:
1494 ShouldNotReachHere();
1495 break;
1496 }
1497 } else {
1498 #ifdef _LP64
1499 ShouldNotReachHere();
1500 #else
1501 switch (op) {
1502 case add: __ fadd_s (at_rsp()); break;
1503 case sub: __ fsubr_s(at_rsp()); break;
1504 case mul: __ fmul_s (at_rsp()); break;
1505 case div: __ fdivr_s(at_rsp()); break;
1506 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1507 default : ShouldNotReachHere();
1508 }
1509 __ f2ieee();
1510 __ pop(rax); // pop second operand off the stack
1511 #endif // _LP64
1512 }
1513 }
1514
1515 void TemplateTable::dop2(Operation op) {
1516 transition(dtos, dtos);
1517 if (UseSSE >= 2) {
1518 switch (op) {
1519 case add:
1520 __ addsd(xmm0, at_rsp());
1521 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1522 break;
1523 case sub:
1524 __ movdbl(xmm1, xmm0);
1525 __ pop_d(xmm0);
1526 __ subsd(xmm0, xmm1);
1527 break;
1528 case mul:
1529 __ mulsd(xmm0, at_rsp());
1530 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1531 break;
1532 case div:
1533 __ movdbl(xmm1, xmm0);
1534 __ pop_d(xmm0);
1535 __ divsd(xmm0, xmm1);
1536 break;
1537 case rem:
1538 // Similar to fop2(), the modulo operation is performed using the
1539 // SharedRuntime::drem method (on x86_64 platforms) or using the
1540 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1541 #ifdef _LP64
1542 __ movdbl(xmm1, xmm0);
1543 __ pop_d(xmm0);
1544 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1545 #else
1546 __ push_d(xmm0);
1547 __ pop_d();
1548 __ fld_d(at_rsp());
1549 __ fremr(rax);
1550 __ d2ieee();
1551 __ pop(rax);
1552 __ pop(rdx);
1553 __ push_d();
1554 __ pop_d(xmm0);
1555 #endif
1556 break;
1557 default:
1558 ShouldNotReachHere();
1559 break;
1560 }
1561 } else {
1562 #ifdef _LP64
1563 ShouldNotReachHere();
1564 #else
1565 switch (op) {
1566 case add: __ fadd_d (at_rsp()); break;
1567 case sub: __ fsubr_d(at_rsp()); break;
1568 case mul: {
1569 Label L_strict;
1570 Label L_join;
1571 const Address access_flags (rcx, Method::access_flags_offset());
1572 __ get_method(rcx);
1573 __ movl(rcx, access_flags);
1574 __ testl(rcx, JVM_ACC_STRICT);
1575 __ jccb(Assembler::notZero, L_strict);
1576 __ fmul_d (at_rsp());
1577 __ jmpb(L_join);
1578 __ bind(L_strict);
1579 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1580 __ fmulp();
1581 __ fmul_d (at_rsp());
1582 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1583 __ fmulp();
1584 __ bind(L_join);
1585 break;
1586 }
1587 case div: {
1588 Label L_strict;
1589 Label L_join;
1590 const Address access_flags (rcx, Method::access_flags_offset());
1591 __ get_method(rcx);
1592 __ movl(rcx, access_flags);
1593 __ testl(rcx, JVM_ACC_STRICT);
1594 __ jccb(Assembler::notZero, L_strict);
1595 __ fdivr_d(at_rsp());
1596 __ jmp(L_join);
1597 __ bind(L_strict);
1598 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1599 __ fmul_d (at_rsp());
1600 __ fdivrp();
1601 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1602 __ fmulp();
1603 __ bind(L_join);
1604 break;
1605 }
1606 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1607 default : ShouldNotReachHere();
1608 }
1609 __ d2ieee();
1610 // Pop double precision number from rsp.
1611 __ pop(rax);
1612 __ pop(rdx);
1613 #endif
1614 }
1615 }
1616
1617 void TemplateTable::ineg() {
1618 transition(itos, itos);
1619 __ negl(rax);
1620 }
1621
1622 void TemplateTable::lneg() {
1623 transition(ltos, ltos);
1624 LP64_ONLY(__ negq(rax));
1625 NOT_LP64(__ lneg(rdx, rax));
1626 }
1627
1628 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1629 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1630 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1631 // of 128-bits operands for SSE instructions.
1632 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1633 // Store the value to a 128-bits operand.
1634 operand[0] = lo;
1635 operand[1] = hi;
1636 return operand;
1637 }
1638
1639 // Buffer for 128-bits masks used by SSE instructions.
1640 static jlong float_signflip_pool[2*2];
1641 static jlong double_signflip_pool[2*2];
1642
1643 void TemplateTable::fneg() {
1644 transition(ftos, ftos);
1645 if (UseSSE >= 1) {
1646 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1647 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1648 } else {
1649 LP64_ONLY(ShouldNotReachHere());
1650 NOT_LP64(__ fchs());
1651 }
1652 }
1653
1654 void TemplateTable::dneg() {
1655 transition(dtos, dtos);
1656 if (UseSSE >= 2) {
1657 static jlong *double_signflip =
1658 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1659 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1660 } else {
1661 #ifdef _LP64
1662 ShouldNotReachHere();
1663 #else
1664 __ fchs();
1665 #endif
1666 }
1667 }
1668
1669 void TemplateTable::iinc() {
1670 transition(vtos, vtos);
1671 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1672 locals_index(rbx);
1673 __ addl(iaddress(rbx), rdx);
1674 }
1675
1676 void TemplateTable::wide_iinc() {
1677 transition(vtos, vtos);
1678 __ movl(rdx, at_bcp(4)); // get constant
1679 locals_index_wide(rbx);
1680 __ bswapl(rdx); // swap bytes & sign-extend constant
1681 __ sarl(rdx, 16);
1682 __ addl(iaddress(rbx), rdx);
1683 // Note: should probably use only one movl to get both
1684 // the index and the constant -> fix this
1685 }
1686
1687 void TemplateTable::convert() {
1688 #ifdef _LP64
1689 // Checking
1690 #ifdef ASSERT
1691 {
1692 TosState tos_in = ilgl;
1693 TosState tos_out = ilgl;
1694 switch (bytecode()) {
1695 case Bytecodes::_i2l: // fall through
1696 case Bytecodes::_i2f: // fall through
1697 case Bytecodes::_i2d: // fall through
1698 case Bytecodes::_i2b: // fall through
1699 case Bytecodes::_i2c: // fall through
1700 case Bytecodes::_i2s: tos_in = itos; break;
1701 case Bytecodes::_l2i: // fall through
1702 case Bytecodes::_l2f: // fall through
1703 case Bytecodes::_l2d: tos_in = ltos; break;
1704 case Bytecodes::_f2i: // fall through
1705 case Bytecodes::_f2l: // fall through
1706 case Bytecodes::_f2d: tos_in = ftos; break;
1707 case Bytecodes::_d2i: // fall through
1708 case Bytecodes::_d2l: // fall through
1709 case Bytecodes::_d2f: tos_in = dtos; break;
1710 default : ShouldNotReachHere();
1711 }
1712 switch (bytecode()) {
1713 case Bytecodes::_l2i: // fall through
1714 case Bytecodes::_f2i: // fall through
1715 case Bytecodes::_d2i: // fall through
1716 case Bytecodes::_i2b: // fall through
1717 case Bytecodes::_i2c: // fall through
1718 case Bytecodes::_i2s: tos_out = itos; break;
1719 case Bytecodes::_i2l: // fall through
1720 case Bytecodes::_f2l: // fall through
1721 case Bytecodes::_d2l: tos_out = ltos; break;
1722 case Bytecodes::_i2f: // fall through
1723 case Bytecodes::_l2f: // fall through
1724 case Bytecodes::_d2f: tos_out = ftos; break;
1725 case Bytecodes::_i2d: // fall through
1726 case Bytecodes::_l2d: // fall through
1727 case Bytecodes::_f2d: tos_out = dtos; break;
1728 default : ShouldNotReachHere();
1729 }
1730 transition(tos_in, tos_out);
1731 }
1732 #endif // ASSERT
1733
1734 static const int64_t is_nan = 0x8000000000000000L;
1735
1736 // Conversion
1737 switch (bytecode()) {
1738 case Bytecodes::_i2l:
1739 __ movslq(rax, rax);
1740 break;
1741 case Bytecodes::_i2f:
1742 __ cvtsi2ssl(xmm0, rax);
1743 break;
1744 case Bytecodes::_i2d:
1745 __ cvtsi2sdl(xmm0, rax);
1746 break;
1747 case Bytecodes::_i2b:
1748 __ movsbl(rax, rax);
1749 break;
1750 case Bytecodes::_i2c:
1751 __ movzwl(rax, rax);
1752 break;
1753 case Bytecodes::_i2s:
1754 __ movswl(rax, rax);
1755 break;
1756 case Bytecodes::_l2i:
1757 __ movl(rax, rax);
1758 break;
1759 case Bytecodes::_l2f:
1760 __ cvtsi2ssq(xmm0, rax);
1761 break;
1762 case Bytecodes::_l2d:
1763 __ cvtsi2sdq(xmm0, rax);
1764 break;
1765 case Bytecodes::_f2i:
1766 {
1767 Label L;
1768 __ cvttss2sil(rax, xmm0);
1769 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1770 __ jcc(Assembler::notEqual, L);
1771 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1772 __ bind(L);
1773 }
1774 break;
1775 case Bytecodes::_f2l:
1776 {
1777 Label L;
1778 __ cvttss2siq(rax, xmm0);
1779 // NaN or overflow/underflow?
1780 __ cmp64(rax, ExternalAddress((address) &is_nan));
1781 __ jcc(Assembler::notEqual, L);
1782 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1783 __ bind(L);
1784 }
1785 break;
1786 case Bytecodes::_f2d:
1787 __ cvtss2sd(xmm0, xmm0);
1788 break;
1789 case Bytecodes::_d2i:
1790 {
1791 Label L;
1792 __ cvttsd2sil(rax, xmm0);
1793 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1794 __ jcc(Assembler::notEqual, L);
1795 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1796 __ bind(L);
1797 }
1798 break;
1799 case Bytecodes::_d2l:
1800 {
1801 Label L;
1802 __ cvttsd2siq(rax, xmm0);
1803 // NaN or overflow/underflow?
1804 __ cmp64(rax, ExternalAddress((address) &is_nan));
1805 __ jcc(Assembler::notEqual, L);
1806 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1807 __ bind(L);
1808 }
1809 break;
1810 case Bytecodes::_d2f:
1811 __ cvtsd2ss(xmm0, xmm0);
1812 break;
1813 default:
1814 ShouldNotReachHere();
1815 }
1816 #else
1817 // Checking
1818 #ifdef ASSERT
1819 { TosState tos_in = ilgl;
1820 TosState tos_out = ilgl;
1821 switch (bytecode()) {
1822 case Bytecodes::_i2l: // fall through
1823 case Bytecodes::_i2f: // fall through
1824 case Bytecodes::_i2d: // fall through
1825 case Bytecodes::_i2b: // fall through
1826 case Bytecodes::_i2c: // fall through
1827 case Bytecodes::_i2s: tos_in = itos; break;
1828 case Bytecodes::_l2i: // fall through
1829 case Bytecodes::_l2f: // fall through
1830 case Bytecodes::_l2d: tos_in = ltos; break;
1831 case Bytecodes::_f2i: // fall through
1832 case Bytecodes::_f2l: // fall through
1833 case Bytecodes::_f2d: tos_in = ftos; break;
1834 case Bytecodes::_d2i: // fall through
1835 case Bytecodes::_d2l: // fall through
1836 case Bytecodes::_d2f: tos_in = dtos; break;
1837 default : ShouldNotReachHere();
1838 }
1839 switch (bytecode()) {
1840 case Bytecodes::_l2i: // fall through
1841 case Bytecodes::_f2i: // fall through
1842 case Bytecodes::_d2i: // fall through
1843 case Bytecodes::_i2b: // fall through
1844 case Bytecodes::_i2c: // fall through
1845 case Bytecodes::_i2s: tos_out = itos; break;
1846 case Bytecodes::_i2l: // fall through
1847 case Bytecodes::_f2l: // fall through
1848 case Bytecodes::_d2l: tos_out = ltos; break;
1849 case Bytecodes::_i2f: // fall through
1850 case Bytecodes::_l2f: // fall through
1851 case Bytecodes::_d2f: tos_out = ftos; break;
1852 case Bytecodes::_i2d: // fall through
1853 case Bytecodes::_l2d: // fall through
1854 case Bytecodes::_f2d: tos_out = dtos; break;
1855 default : ShouldNotReachHere();
1856 }
1857 transition(tos_in, tos_out);
1858 }
1859 #endif // ASSERT
1860
1861 // Conversion
1862 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1863 switch (bytecode()) {
1864 case Bytecodes::_i2l:
1865 __ extend_sign(rdx, rax);
1866 break;
1867 case Bytecodes::_i2f:
1868 if (UseSSE >= 1) {
1869 __ cvtsi2ssl(xmm0, rax);
1870 } else {
1871 __ push(rax); // store int on tos
1872 __ fild_s(at_rsp()); // load int to ST0
1873 __ f2ieee(); // truncate to float size
1874 __ pop(rcx); // adjust rsp
1875 }
1876 break;
1877 case Bytecodes::_i2d:
1878 if (UseSSE >= 2) {
1879 __ cvtsi2sdl(xmm0, rax);
1880 } else {
1881 __ push(rax); // add one slot for d2ieee()
1882 __ push(rax); // store int on tos
1883 __ fild_s(at_rsp()); // load int to ST0
1884 __ d2ieee(); // truncate to double size
1885 __ pop(rcx); // adjust rsp
1886 __ pop(rcx);
1887 }
1888 break;
1889 case Bytecodes::_i2b:
1890 __ shll(rax, 24); // truncate upper 24 bits
1891 __ sarl(rax, 24); // and sign-extend byte
1892 LP64_ONLY(__ movsbl(rax, rax));
1893 break;
1894 case Bytecodes::_i2c:
1895 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1896 LP64_ONLY(__ movzwl(rax, rax));
1897 break;
1898 case Bytecodes::_i2s:
1899 __ shll(rax, 16); // truncate upper 16 bits
1900 __ sarl(rax, 16); // and sign-extend short
1901 LP64_ONLY(__ movswl(rax, rax));
1902 break;
1903 case Bytecodes::_l2i:
1904 /* nothing to do */
1905 break;
1906 case Bytecodes::_l2f:
1907 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1908 // 64-bit long values to floats. On 32-bit platforms it is not possible
1909 // to use that instruction with 64-bit operands, therefore the FPU is
1910 // used to perform the conversion.
1911 __ push(rdx); // store long on tos
1912 __ push(rax);
1913 __ fild_d(at_rsp()); // load long to ST0
1914 __ f2ieee(); // truncate to float size
1915 __ pop(rcx); // adjust rsp
1916 __ pop(rcx);
1917 if (UseSSE >= 1) {
1918 __ push_f();
1919 __ pop_f(xmm0);
1920 }
1921 break;
1922 case Bytecodes::_l2d:
1923 // On 32-bit platforms the FPU is used for conversion because on
1924 // 32-bit platforms it is not not possible to use the cvtsi2sdq
1925 // instruction with 64-bit operands.
1926 __ push(rdx); // store long on tos
1927 __ push(rax);
1928 __ fild_d(at_rsp()); // load long to ST0
1929 __ d2ieee(); // truncate to double size
1930 __ pop(rcx); // adjust rsp
1931 __ pop(rcx);
1932 if (UseSSE >= 2) {
1933 __ push_d();
1934 __ pop_d(xmm0);
1935 }
1936 break;
1937 case Bytecodes::_f2i:
1938 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1939 // as it returns 0 for any NaN.
1940 if (UseSSE >= 1) {
1941 __ push_f(xmm0);
1942 } else {
1943 __ push(rcx); // reserve space for argument
1944 __ fstp_s(at_rsp()); // pass float argument on stack
1945 }
1946 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1947 break;
1948 case Bytecodes::_f2l:
1949 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
1950 // as it returns 0 for any NaN.
1951 if (UseSSE >= 1) {
1952 __ push_f(xmm0);
1953 } else {
1954 __ push(rcx); // reserve space for argument
1955 __ fstp_s(at_rsp()); // pass float argument on stack
1956 }
1957 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1958 break;
1959 case Bytecodes::_f2d:
1960 if (UseSSE < 1) {
1961 /* nothing to do */
1962 } else if (UseSSE == 1) {
1963 __ push_f(xmm0);
1964 __ pop_f();
1965 } else { // UseSSE >= 2
1966 __ cvtss2sd(xmm0, xmm0);
1967 }
1968 break;
1969 case Bytecodes::_d2i:
1970 if (UseSSE >= 2) {
1971 __ push_d(xmm0);
1972 } else {
1973 __ push(rcx); // reserve space for argument
1974 __ push(rcx);
1975 __ fstp_d(at_rsp()); // pass double argument on stack
1976 }
1977 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1978 break;
1979 case Bytecodes::_d2l:
1980 if (UseSSE >= 2) {
1981 __ push_d(xmm0);
1982 } else {
1983 __ push(rcx); // reserve space for argument
1984 __ push(rcx);
1985 __ fstp_d(at_rsp()); // pass double argument on stack
1986 }
1987 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1988 break;
1989 case Bytecodes::_d2f:
1990 if (UseSSE <= 1) {
1991 __ push(rcx); // reserve space for f2ieee()
1992 __ f2ieee(); // truncate to float size
1993 __ pop(rcx); // adjust rsp
1994 if (UseSSE == 1) {
1995 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
1996 // the conversion is performed using the FPU in this case.
1997 __ push_f();
1998 __ pop_f(xmm0);
1999 }
2000 } else { // UseSSE >= 2
2001 __ cvtsd2ss(xmm0, xmm0);
2002 }
2003 break;
2004 default :
2005 ShouldNotReachHere();
2006 }
2007 #endif
2008 }
2009
2010 void TemplateTable::lcmp() {
2011 transition(ltos, itos);
2012 #ifdef _LP64
2013 Label done;
2014 __ pop_l(rdx);
2015 __ cmpq(rdx, rax);
2016 __ movl(rax, -1);
2017 __ jccb(Assembler::less, done);
2018 __ setb(Assembler::notEqual, rax);
2019 __ movzbl(rax, rax);
2020 __ bind(done);
2021 #else
2022
2023 // y = rdx:rax
2024 __ pop_l(rbx, rcx); // get x = rcx:rbx
2025 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2026 __ mov(rax, rcx);
2027 #endif
2028 }
2029
2030 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2031 if ((is_float && UseSSE >= 1) ||
2032 (!is_float && UseSSE >= 2)) {
2033 Label done;
2034 if (is_float) {
2035 // XXX get rid of pop here, use ... reg, mem32
2036 __ pop_f(xmm1);
2037 __ ucomiss(xmm1, xmm0);
2038 } else {
2039 // XXX get rid of pop here, use ... reg, mem64
2040 __ pop_d(xmm1);
2041 __ ucomisd(xmm1, xmm0);
2042 }
2043 if (unordered_result < 0) {
2044 __ movl(rax, -1);
2045 __ jccb(Assembler::parity, done);
2046 __ jccb(Assembler::below, done);
2047 __ setb(Assembler::notEqual, rdx);
2048 __ movzbl(rax, rdx);
2049 } else {
2050 __ movl(rax, 1);
2051 __ jccb(Assembler::parity, done);
2052 __ jccb(Assembler::above, done);
2053 __ movl(rax, 0);
2054 __ jccb(Assembler::equal, done);
2055 __ decrementl(rax);
2056 }
2057 __ bind(done);
2058 } else {
2059 #ifdef _LP64
2060 ShouldNotReachHere();
2061 #else
2062 if (is_float) {
2063 __ fld_s(at_rsp());
2064 } else {
2065 __ fld_d(at_rsp());
2066 __ pop(rdx);
2067 }
2068 __ pop(rcx);
2069 __ fcmp2int(rax, unordered_result < 0);
2070 #endif // _LP64
2071 }
2072 }
2073
2074 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2075 __ get_method(rcx); // rcx holds method
2076 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2077 // holds bumped taken count
2078
2079 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2080 InvocationCounter::counter_offset();
2081 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2082 InvocationCounter::counter_offset();
2083
2084 // Load up edx with the branch displacement
2085 if (is_wide) {
2086 __ movl(rdx, at_bcp(1));
2087 } else {
2088 __ load_signed_short(rdx, at_bcp(1));
2089 }
2090 __ bswapl(rdx);
2091
2092 if (!is_wide) {
2093 __ sarl(rdx, 16);
2094 }
2095 LP64_ONLY(__ movl2ptr(rdx, rdx));
2096
2097 // Handle all the JSR stuff here, then exit.
2098 // It's much shorter and cleaner than intermingling with the non-JSR
2099 // normal-branch stuff occurring below.
2100 if (is_jsr) {
2101 // Pre-load the next target bytecode into rbx
2102 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2103
2104 // compute return address as bci in rax
2105 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2106 in_bytes(ConstMethod::codes_offset())));
2107 __ subptr(rax, Address(rcx, Method::const_offset()));
2108 // Adjust the bcp in r13 by the displacement in rdx
2109 __ addptr(rbcp, rdx);
2110 // jsr returns atos that is not an oop
2111 __ push_i(rax);
2112 __ dispatch_only(vtos);
2113 return;
2114 }
2115
2116 // Normal (non-jsr) branch handling
2117
2118 // Adjust the bcp in r13 by the displacement in rdx
2119 __ addptr(rbcp, rdx);
2120
2121 assert(UseLoopCounter || !UseOnStackReplacement,
2122 "on-stack-replacement requires loop counters");
2123 Label backedge_counter_overflow;
2124 Label profile_method;
2125 Label dispatch;
2126 if (UseLoopCounter) {
2127 // increment backedge counter for backward branches
2128 // rax: MDO
2129 // rbx: MDO bumped taken-count
2130 // rcx: method
2131 // rdx: target offset
2132 // r13: target bcp
2133 // r14: locals pointer
2134 __ testl(rdx, rdx); // check if forward or backward branch
2135 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2136
2137 // check if MethodCounters exists
2138 Label has_counters;
2139 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2140 __ testptr(rax, rax);
2141 __ jcc(Assembler::notZero, has_counters);
2142 __ push(rdx);
2143 __ push(rcx);
2144 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2145 rcx);
2146 __ pop(rcx);
2147 __ pop(rdx);
2148 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2149 __ testptr(rax, rax);
2150 __ jcc(Assembler::zero, dispatch);
2151 __ bind(has_counters);
2152
2153 if (TieredCompilation) {
2154 Label no_mdo;
2155 int increment = InvocationCounter::count_increment;
2156 if (ProfileInterpreter) {
2157 // Are we profiling?
2158 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2159 __ testptr(rbx, rbx);
2160 __ jccb(Assembler::zero, no_mdo);
2161 // Increment the MDO backedge counter
2162 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2163 in_bytes(InvocationCounter::counter_offset()));
2164 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2165 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2166 rax, false, Assembler::zero, &backedge_counter_overflow);
2167 __ jmp(dispatch);
2168 }
2169 __ bind(no_mdo);
2170 // Increment backedge counter in MethodCounters*
2171 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2172 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2173 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2174 rax, false, Assembler::zero, &backedge_counter_overflow);
2175 } else { // not TieredCompilation
2176 // increment counter
2177 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2178 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2179 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2180 __ movl(Address(rcx, be_offset), rax); // store counter
2181
2182 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2183
2184 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2185 __ addl(rax, Address(rcx, be_offset)); // add both counters
2186
2187 if (ProfileInterpreter) {
2188 // Test to see if we should create a method data oop
2189 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2190 __ jcc(Assembler::less, dispatch);
2191
2192 // if no method data exists, go to profile method
2193 __ test_method_data_pointer(rax, profile_method);
2194
2195 if (UseOnStackReplacement) {
2196 // check for overflow against rbx which is the MDO taken count
2197 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2198 __ jcc(Assembler::below, dispatch);
2199
2200 // When ProfileInterpreter is on, the backedge_count comes
2201 // from the MethodData*, which value does not get reset on
2202 // the call to frequency_counter_overflow(). To avoid
2203 // excessive calls to the overflow routine while the method is
2204 // being compiled, add a second test to make sure the overflow
2205 // function is called only once every overflow_frequency.
2206 const int overflow_frequency = 1024;
2207 __ andl(rbx, overflow_frequency - 1);
2208 __ jcc(Assembler::zero, backedge_counter_overflow);
2209
2210 }
2211 } else {
2212 if (UseOnStackReplacement) {
2213 // check for overflow against rax, which is the sum of the
2214 // counters
2215 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2216 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2217
2218 }
2219 }
2220 }
2221 __ bind(dispatch);
2222 }
2223
2224 // Pre-load the next target bytecode into rbx
2225 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2226
2227 // continue with the bytecode @ target
2228 // rax: return bci for jsr's, unused otherwise
2229 // rbx: target bytecode
2230 // r13: target bcp
2231 __ dispatch_only(vtos);
2232
2233 if (UseLoopCounter) {
2234 if (ProfileInterpreter) {
2235 // Out-of-line code to allocate method data oop.
2236 __ bind(profile_method);
2237 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2238 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2239 __ set_method_data_pointer_for_bcp();
2240 __ jmp(dispatch);
2241 }
2242
2243 if (UseOnStackReplacement) {
2244 // invocation counter overflow
2245 __ bind(backedge_counter_overflow);
2246 __ negptr(rdx);
2247 __ addptr(rdx, rbcp); // branch bcp
2248 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2249 __ call_VM(noreg,
2250 CAST_FROM_FN_PTR(address,
2251 InterpreterRuntime::frequency_counter_overflow),
2252 rdx);
2253 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2254
2255 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2256 // rbx: target bytecode
2257 // rdx: scratch
2258 // r14: locals pointer
2259 // r13: bcp
2260 __ testptr(rax, rax); // test result
2261 __ jcc(Assembler::zero, dispatch); // no osr if null
2262 // nmethod may have been invalidated (VM may block upon call_VM return)
2263 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2264 __ jcc(Assembler::notEqual, dispatch);
2265
2266 // We have the address of an on stack replacement routine in rax
2267 // We need to prepare to execute the OSR method. First we must
2268 // migrate the locals and monitors off of the stack.
2269
2270 LP64_ONLY(__ mov(r13, rax)); // save the nmethod
2271 NOT_LP64(__ mov(rbx, rax)); // save the nmethod
2272 NOT_LP64(__ get_thread(rcx));
2273
2274 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2275
2276 // rax is OSR buffer, move it to expected parameter location
2277 LP64_ONLY(__ mov(j_rarg0, rax));
2278 NOT_LP64(__ mov(rcx, rax));
2279 // We use j_rarg definitions here so that registers don't conflict as parameter
2280 // registers change across platforms as we are in the midst of a calling
2281 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2282
2283 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2284 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2285
2286
2287 // pop the interpreter frame
2288 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2289 __ leave(); // remove frame anchor
2290 __ pop(retaddr); // get return address
2291 __ mov(rsp, sender_sp); // set sp to sender sp
2292 // Ensure compiled code always sees stack at proper alignment
2293 __ andptr(rsp, -(StackAlignmentInBytes));
2294
2295 // unlike x86 we need no specialized return from compiled code
2296 // to the interpreter or the call stub.
2297
2298 // push the return address
2299 __ push(retaddr);
2300
2301 // and begin the OSR nmethod
2302 LP64_ONLY(__ jmp(Address(r13, nmethod::osr_entry_point_offset())));
2303 NOT_LP64(__ jmp(Address(rbx, nmethod::osr_entry_point_offset())));
2304 }
2305 }
2306 }
2307
2308 void TemplateTable::if_0cmp(Condition cc) {
2309 transition(itos, vtos);
2310 // assume branch is more often taken than not (loops use backward branches)
2311 Label not_taken;
2312 __ testl(rax, rax);
2313 __ jcc(j_not(cc), not_taken);
2314 branch(false, false);
2315 __ bind(not_taken);
2316 __ profile_not_taken_branch(rax);
2317 }
2318
2319 void TemplateTable::if_icmp(Condition cc) {
2320 transition(itos, vtos);
2321 // assume branch is more often taken than not (loops use backward branches)
2322 Label not_taken;
2323 __ pop_i(rdx);
2324 __ cmpl(rdx, rax);
2325 __ jcc(j_not(cc), not_taken);
2326 branch(false, false);
2327 __ bind(not_taken);
2328 __ profile_not_taken_branch(rax);
2329 }
2330
2331 void TemplateTable::if_nullcmp(Condition cc) {
2332 transition(atos, vtos);
2333 // assume branch is more often taken than not (loops use backward branches)
2334 Label not_taken;
2335 __ testptr(rax, rax);
2336 __ jcc(j_not(cc), not_taken);
2337 branch(false, false);
2338 __ bind(not_taken);
2339 __ profile_not_taken_branch(rax);
2340 }
2341
2342 void TemplateTable::if_acmp(Condition cc) {
2343 transition(atos, vtos);
2344 // assume branch is more often taken than not (loops use backward branches)
2345 Label not_taken;
2346 __ pop_ptr(rdx);
2347 __ cmpptr(rdx, rax);
2348 oopDesc::bs()->asm_acmp_barrier(_masm, rdx, rax);
2349 __ jcc(j_not(cc), not_taken);
2350 branch(false, false);
2351 __ bind(not_taken);
2352 __ profile_not_taken_branch(rax);
2353 }
2354
2355 void TemplateTable::ret() {
2356 transition(vtos, vtos);
2357 locals_index(rbx);
2358 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2359 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2360 __ profile_ret(rbx, rcx);
2361 __ get_method(rax);
2362 __ movptr(rbcp, Address(rax, Method::const_offset()));
2363 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2364 ConstMethod::codes_offset()));
2365 __ dispatch_next(vtos);
2366 }
2367
2368 void TemplateTable::wide_ret() {
2369 transition(vtos, vtos);
2370 locals_index_wide(rbx);
2371 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2372 __ profile_ret(rbx, rcx);
2373 __ get_method(rax);
2374 __ movptr(rbcp, Address(rax, Method::const_offset()));
2375 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2376 __ dispatch_next(vtos);
2377 }
2378
2379 void TemplateTable::tableswitch() {
2380 Label default_case, continue_execution;
2381 transition(itos, vtos);
2382
2383 // align r13/rsi
2384 __ lea(rbx, at_bcp(BytesPerInt));
2385 __ andptr(rbx, -BytesPerInt);
2386 // load lo & hi
2387 __ movl(rcx, Address(rbx, BytesPerInt));
2388 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2389 __ bswapl(rcx);
2390 __ bswapl(rdx);
2391 // check against lo & hi
2392 __ cmpl(rax, rcx);
2393 __ jcc(Assembler::less, default_case);
2394 __ cmpl(rax, rdx);
2395 __ jcc(Assembler::greater, default_case);
2396 // lookup dispatch offset
2397 __ subl(rax, rcx);
2398 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2399 __ profile_switch_case(rax, rbx, rcx);
2400 // continue execution
2401 __ bind(continue_execution);
2402 __ bswapl(rdx);
2403 LP64_ONLY(__ movl2ptr(rdx, rdx));
2404 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2405 __ addptr(rbcp, rdx);
2406 __ dispatch_only(vtos);
2407 // handle default
2408 __ bind(default_case);
2409 __ profile_switch_default(rax);
2410 __ movl(rdx, Address(rbx, 0));
2411 __ jmp(continue_execution);
2412 }
2413
2414 void TemplateTable::lookupswitch() {
2415 transition(itos, itos);
2416 __ stop("lookupswitch bytecode should have been rewritten");
2417 }
2418
2419 void TemplateTable::fast_linearswitch() {
2420 transition(itos, vtos);
2421 Label loop_entry, loop, found, continue_execution;
2422 // bswap rax so we can avoid bswapping the table entries
2423 __ bswapl(rax);
2424 // align r13
2425 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2426 // this instruction (change offsets
2427 // below)
2428 __ andptr(rbx, -BytesPerInt);
2429 // set counter
2430 __ movl(rcx, Address(rbx, BytesPerInt));
2431 __ bswapl(rcx);
2432 __ jmpb(loop_entry);
2433 // table search
2434 __ bind(loop);
2435 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2436 __ jcc(Assembler::equal, found);
2437 __ bind(loop_entry);
2438 __ decrementl(rcx);
2439 __ jcc(Assembler::greaterEqual, loop);
2440 // default case
2441 __ profile_switch_default(rax);
2442 __ movl(rdx, Address(rbx, 0));
2443 __ jmp(continue_execution);
2444 // entry found -> get offset
2445 __ bind(found);
2446 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2447 __ profile_switch_case(rcx, rax, rbx);
2448 // continue execution
2449 __ bind(continue_execution);
2450 __ bswapl(rdx);
2451 __ movl2ptr(rdx, rdx);
2452 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2453 __ addptr(rbcp, rdx);
2454 __ dispatch_only(vtos);
2455 }
2456
2457 void TemplateTable::fast_binaryswitch() {
2458 transition(itos, vtos);
2459 // Implementation using the following core algorithm:
2460 //
2461 // int binary_search(int key, LookupswitchPair* array, int n) {
2462 // // Binary search according to "Methodik des Programmierens" by
2463 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2464 // int i = 0;
2465 // int j = n;
2466 // while (i+1 < j) {
2467 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2468 // // with Q: for all i: 0 <= i < n: key < a[i]
2469 // // where a stands for the array and assuming that the (inexisting)
2470 // // element a[n] is infinitely big.
2471 // int h = (i + j) >> 1;
2472 // // i < h < j
2473 // if (key < array[h].fast_match()) {
2474 // j = h;
2475 // } else {
2476 // i = h;
2477 // }
2478 // }
2479 // // R: a[i] <= key < a[i+1] or Q
2480 // // (i.e., if key is within array, i is the correct index)
2481 // return i;
2482 // }
2483
2484 // Register allocation
2485 const Register key = rax; // already set (tosca)
2486 const Register array = rbx;
2487 const Register i = rcx;
2488 const Register j = rdx;
2489 const Register h = rdi;
2490 const Register temp = rsi;
2491
2492 // Find array start
2493 NOT_LP64(__ save_bcp());
2494
2495 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2496 // get rid of this
2497 // instruction (change
2498 // offsets below)
2499 __ andptr(array, -BytesPerInt);
2500
2501 // Initialize i & j
2502 __ xorl(i, i); // i = 0;
2503 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2504
2505 // Convert j into native byteordering
2506 __ bswapl(j);
2507
2508 // And start
2509 Label entry;
2510 __ jmp(entry);
2511
2512 // binary search loop
2513 {
2514 Label loop;
2515 __ bind(loop);
2516 // int h = (i + j) >> 1;
2517 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2518 __ sarl(h, 1); // h = (i + j) >> 1;
2519 // if (key < array[h].fast_match()) {
2520 // j = h;
2521 // } else {
2522 // i = h;
2523 // }
2524 // Convert array[h].match to native byte-ordering before compare
2525 __ movl(temp, Address(array, h, Address::times_8));
2526 __ bswapl(temp);
2527 __ cmpl(key, temp);
2528 // j = h if (key < array[h].fast_match())
2529 __ cmov32(Assembler::less, j, h);
2530 // i = h if (key >= array[h].fast_match())
2531 __ cmov32(Assembler::greaterEqual, i, h);
2532 // while (i+1 < j)
2533 __ bind(entry);
2534 __ leal(h, Address(i, 1)); // i+1
2535 __ cmpl(h, j); // i+1 < j
2536 __ jcc(Assembler::less, loop);
2537 }
2538
2539 // end of binary search, result index is i (must check again!)
2540 Label default_case;
2541 // Convert array[i].match to native byte-ordering before compare
2542 __ movl(temp, Address(array, i, Address::times_8));
2543 __ bswapl(temp);
2544 __ cmpl(key, temp);
2545 __ jcc(Assembler::notEqual, default_case);
2546
2547 // entry found -> j = offset
2548 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2549 __ profile_switch_case(i, key, array);
2550 __ bswapl(j);
2551 LP64_ONLY(__ movslq(j, j));
2552
2553 NOT_LP64(__ restore_bcp());
2554 NOT_LP64(__ restore_locals()); // restore rdi
2555
2556 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2557 __ addptr(rbcp, j);
2558 __ dispatch_only(vtos);
2559
2560 // default case -> j = default offset
2561 __ bind(default_case);
2562 __ profile_switch_default(i);
2563 __ movl(j, Address(array, -2 * BytesPerInt));
2564 __ bswapl(j);
2565 LP64_ONLY(__ movslq(j, j));
2566
2567 NOT_LP64(__ restore_bcp());
2568 NOT_LP64(__ restore_locals());
2569
2570 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2571 __ addptr(rbcp, j);
2572 __ dispatch_only(vtos);
2573 }
2574
2575 void TemplateTable::_return(TosState state) {
2576 transition(state, state);
2577
2578 assert(_desc->calls_vm(),
2579 "inconsistent calls_vm information"); // call in remove_activation
2580
2581 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2582 assert(state == vtos, "only valid state");
2583 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2584 __ movptr(robj, aaddress(0));
2585 __ load_klass(rdi, robj);
2586 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2587 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2588 Label skip_register_finalizer;
2589 __ jcc(Assembler::zero, skip_register_finalizer);
2590
2591 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2592
2593 __ bind(skip_register_finalizer);
2594 }
2595
2596 // Narrow result if state is itos but result type is smaller.
2597 // Need to narrow in the return bytecode rather than in generate_return_entry
2598 // since compiled code callers expect the result to already be narrowed.
2599 if (state == itos) {
2600 __ narrow(rax);
2601 }
2602 __ remove_activation(state, rbcp);
2603
2604 __ jmp(rbcp);
2605 }
2606
2607 // ----------------------------------------------------------------------------
2608 // Volatile variables demand their effects be made known to all CPU's
2609 // in order. Store buffers on most chips allow reads & writes to
2610 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2611 // without some kind of memory barrier (i.e., it's not sufficient that
2612 // the interpreter does not reorder volatile references, the hardware
2613 // also must not reorder them).
2614 //
2615 // According to the new Java Memory Model (JMM):
2616 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2617 // writes act as aquire & release, so:
2618 // (2) A read cannot let unrelated NON-volatile memory refs that
2619 // happen after the read float up to before the read. It's OK for
2620 // non-volatile memory refs that happen before the volatile read to
2621 // float down below it.
2622 // (3) Similar a volatile write cannot let unrelated NON-volatile
2623 // memory refs that happen BEFORE the write float down to after the
2624 // write. It's OK for non-volatile memory refs that happen after the
2625 // volatile write to float up before it.
2626 //
2627 // We only put in barriers around volatile refs (they are expensive),
2628 // not _between_ memory refs (that would require us to track the
2629 // flavor of the previous memory refs). Requirements (2) and (3)
2630 // require some barriers before volatile stores and after volatile
2631 // loads. These nearly cover requirement (1) but miss the
2632 // volatile-store-volatile-load case. This final case is placed after
2633 // volatile-stores although it could just as well go before
2634 // volatile-loads.
2635
2636 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2637 // Helper function to insert a is-volatile test and memory barrier
2638 if(!os::is_MP()) return; // Not needed on single CPU
2639 __ membar(order_constraint);
2640 }
2641
2642 void TemplateTable::resolve_cache_and_index(int byte_no,
2643 Register Rcache,
2644 Register index,
2645 size_t index_size) {
2646 const Register temp = rbx;
2647 assert_different_registers(Rcache, index, temp);
2648
2649 Label resolved;
2650
2651 Bytecodes::Code code = bytecode();
2652 switch (code) {
2653 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2654 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2655 }
2656
2657 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2658 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2659 __ cmpl(temp, code); // have we resolved this bytecode?
2660 __ jcc(Assembler::equal, resolved);
2661
2662 // resolve first time through
2663 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2664 __ movl(temp, code);
2665 __ call_VM(noreg, entry, temp);
2666 // Update registers with resolved info
2667 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2668 __ bind(resolved);
2669 }
2670
2671 // The cache and index registers must be set before call
2672 void TemplateTable::load_field_cp_cache_entry(Register obj,
2673 Register cache,
2674 Register index,
2675 Register off,
2676 Register flags,
2677 bool is_static = false) {
2678 assert_different_registers(cache, index, flags, off);
2679
2680 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2681 // Field offset
2682 __ movptr(off, Address(cache, index, Address::times_ptr,
2683 in_bytes(cp_base_offset +
2684 ConstantPoolCacheEntry::f2_offset())));
2685 // Flags
2686 __ movl(flags, Address(cache, index, Address::times_ptr,
2687 in_bytes(cp_base_offset +
2688 ConstantPoolCacheEntry::flags_offset())));
2689
2690 // klass overwrite register
2691 if (is_static) {
2692 __ movptr(obj, Address(cache, index, Address::times_ptr,
2693 in_bytes(cp_base_offset +
2694 ConstantPoolCacheEntry::f1_offset())));
2695 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2696 __ movptr(obj, Address(obj, mirror_offset));
2697 }
2698 }
2699
2700 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2701 Register method,
2702 Register itable_index,
2703 Register flags,
2704 bool is_invokevirtual,
2705 bool is_invokevfinal, /*unused*/
2706 bool is_invokedynamic) {
2707 // setup registers
2708 const Register cache = rcx;
2709 const Register index = rdx;
2710 assert_different_registers(method, flags);
2711 assert_different_registers(method, cache, index);
2712 assert_different_registers(itable_index, flags);
2713 assert_different_registers(itable_index, cache, index);
2714 // determine constant pool cache field offsets
2715 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2716 const int method_offset = in_bytes(
2717 ConstantPoolCache::base_offset() +
2718 ((byte_no == f2_byte)
2719 ? ConstantPoolCacheEntry::f2_offset()
2720 : ConstantPoolCacheEntry::f1_offset()));
2721 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2722 ConstantPoolCacheEntry::flags_offset());
2723 // access constant pool cache fields
2724 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2725 ConstantPoolCacheEntry::f2_offset());
2726
2727 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2728 resolve_cache_and_index(byte_no, cache, index, index_size);
2729 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2730
2731 if (itable_index != noreg) {
2732 // pick up itable or appendix index from f2 also:
2733 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2734 }
2735 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2736 }
2737
2738 // The registers cache and index expected to be set before call.
2739 // Correct values of the cache and index registers are preserved.
2740 void TemplateTable::jvmti_post_field_access(Register cache,
2741 Register index,
2742 bool is_static,
2743 bool has_tos) {
2744 if (JvmtiExport::can_post_field_access()) {
2745 // Check to see if a field access watch has been set before we take
2746 // the time to call into the VM.
2747 Label L1;
2748 assert_different_registers(cache, index, rax);
2749 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2750 __ testl(rax,rax);
2751 __ jcc(Assembler::zero, L1);
2752
2753 // cache entry pointer
2754 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2755 __ shll(index, LogBytesPerWord);
2756 __ addptr(cache, index);
2757 if (is_static) {
2758 __ xorptr(rax, rax); // NULL object reference
2759 } else {
2760 __ pop(atos); // Get the object
2761 __ verify_oop(rax);
2762 __ push(atos); // Restore stack state
2763 }
2764 // rax,: object pointer or NULL
2765 // cache: cache entry pointer
2766 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2767 rax, cache);
2768 __ get_cache_and_index_at_bcp(cache, index, 1);
2769 __ bind(L1);
2770 }
2771 }
2772
2773 void TemplateTable::pop_and_check_object(Register r) {
2774 __ pop_ptr(r);
2775 __ null_check(r); // for field access must check obj.
2776 __ verify_oop(r);
2777 }
2778
2779 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2780 transition(vtos, vtos);
2781
2782 const Register cache = rcx;
2783 const Register index = rdx;
2784 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2785 const Register off = rbx;
2786 const Register flags = rax;
2787 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2788
2789 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2790 jvmti_post_field_access(cache, index, is_static, false);
2791 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2792
2793 if (!is_static) pop_and_check_object(obj);
2794 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, obj);
2795
2796 const Address field(obj, off, Address::times_1, 0*wordSize);
2797 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2798
2799 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2800
2801 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2802 // Make sure we don't need to mask edx after the above shift
2803 assert(btos == 0, "change code, btos != 0");
2804
2805 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2806
2807 __ jcc(Assembler::notZero, notByte);
2808 // btos
2809 __ load_signed_byte(rax, field);
2810 __ push(btos);
2811 // Rewrite bytecode to be faster
2812 if (!is_static && rc == may_rewrite) {
2813 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2814 }
2815 __ jmp(Done);
2816
2817 __ bind(notByte);
2818 __ cmpl(flags, ztos);
2819 __ jcc(Assembler::notEqual, notBool);
2820
2821 // ztos (same code as btos)
2822 __ load_signed_byte(rax, field);
2823 __ push(ztos);
2824 // Rewrite bytecode to be faster
2825 if (!is_static && rc == may_rewrite) {
2826 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2827 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2828 }
2829 __ jmp(Done);
2830
2831 __ bind(notBool);
2832 __ cmpl(flags, atos);
2833 __ jcc(Assembler::notEqual, notObj);
2834 // atos
2835 __ load_heap_oop(rax, field);
2836 __ push(atos);
2837 if (!is_static && rc == may_rewrite) {
2838 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2839 }
2840 __ jmp(Done);
2841
2842 __ bind(notObj);
2843 __ cmpl(flags, itos);
2844 __ jcc(Assembler::notEqual, notInt);
2845 // itos
2846 __ movl(rax, field);
2847 __ push(itos);
2848 // Rewrite bytecode to be faster
2849 if (!is_static && rc == may_rewrite) {
2850 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2851 }
2852 __ jmp(Done);
2853
2854 __ bind(notInt);
2855 __ cmpl(flags, ctos);
2856 __ jcc(Assembler::notEqual, notChar);
2857 // ctos
2858 __ load_unsigned_short(rax, field);
2859 __ push(ctos);
2860 // Rewrite bytecode to be faster
2861 if (!is_static && rc == may_rewrite) {
2862 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2863 }
2864 __ jmp(Done);
2865
2866 __ bind(notChar);
2867 __ cmpl(flags, stos);
2868 __ jcc(Assembler::notEqual, notShort);
2869 // stos
2870 __ load_signed_short(rax, field);
2871 __ push(stos);
2872 // Rewrite bytecode to be faster
2873 if (!is_static && rc == may_rewrite) {
2874 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2875 }
2876 __ jmp(Done);
2877
2878 __ bind(notShort);
2879 __ cmpl(flags, ltos);
2880 __ jcc(Assembler::notEqual, notLong);
2881 // ltos
2882
2883 #ifndef _LP64
2884 // Generate code as if volatile. There just aren't enough registers to
2885 // save that information and this code is faster than the test.
2886 __ fild_d(field); // Must load atomically
2887 __ subptr(rsp,2*wordSize); // Make space for store
2888 __ fistp_d(Address(rsp,0));
2889 __ pop(rax);
2890 __ pop(rdx);
2891 #else
2892 __ movq(rax, field);
2893 #endif
2894
2895 __ push(ltos);
2896 // Rewrite bytecode to be faster
2897 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2898 __ jmp(Done);
2899
2900 __ bind(notLong);
2901 __ cmpl(flags, ftos);
2902 __ jcc(Assembler::notEqual, notFloat);
2903 // ftos
2904
2905 __ load_float(field);
2906 __ push(ftos);
2907 // Rewrite bytecode to be faster
2908 if (!is_static && rc == may_rewrite) {
2909 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2910 }
2911 __ jmp(Done);
2912
2913 __ bind(notFloat);
2914 #ifdef ASSERT
2915 __ cmpl(flags, dtos);
2916 __ jcc(Assembler::notEqual, notDouble);
2917 #endif
2918 // dtos
2919 __ load_double(field);
2920 __ push(dtos);
2921 // Rewrite bytecode to be faster
2922 if (!is_static && rc == may_rewrite) {
2923 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2924 }
2925 #ifdef ASSERT
2926 __ jmp(Done);
2927
2928
2929 __ bind(notDouble);
2930 __ stop("Bad state");
2931 #endif
2932
2933 __ bind(Done);
2934 // [jk] not needed currently
2935 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2936 // Assembler::LoadStore));
2937 }
2938
2939 void TemplateTable::getfield(int byte_no) {
2940 getfield_or_static(byte_no, false);
2941 }
2942
2943 void TemplateTable::nofast_getfield(int byte_no) {
2944 getfield_or_static(byte_no, false, may_not_rewrite);
2945 }
2946
2947 void TemplateTable::getstatic(int byte_no) {
2948 getfield_or_static(byte_no, true);
2949 }
2950
2951
2952 // The registers cache and index expected to be set before call.
2953 // The function may destroy various registers, just not the cache and index registers.
2954 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2955
2956 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
2957 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2958 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2959 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
2960
2961 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2962
2963 if (JvmtiExport::can_post_field_modification()) {
2964 // Check to see if a field modification watch has been set before
2965 // we take the time to call into the VM.
2966 Label L1;
2967 assert_different_registers(cache, index, rax);
2968 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2969 __ testl(rax, rax);
2970 __ jcc(Assembler::zero, L1);
2971
2972 __ get_cache_and_index_at_bcp(robj, RDX, 1);
2973
2974
2975 if (is_static) {
2976 // Life is simple. Null out the object pointer.
2977 __ xorl(RBX, RBX);
2978
2979 } else {
2980 // Life is harder. The stack holds the value on top, followed by
2981 // the object. We don't know the size of the value, though; it
2982 // could be one or two words depending on its type. As a result,
2983 // we must find the type to determine where the object is.
2984 #ifndef _LP64
2985 Label two_word, valsize_known;
2986 #endif
2987 __ movl(RCX, Address(robj, RDX,
2988 Address::times_ptr,
2989 in_bytes(cp_base_offset +
2990 ConstantPoolCacheEntry::flags_offset())));
2991 NOT_LP64(__ mov(rbx, rsp));
2992 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
2993
2994 // Make sure we don't need to mask rcx after the above shift
2995 ConstantPoolCacheEntry::verify_tos_state_shift();
2996 #ifdef _LP64
2997 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2998 __ cmpl(c_rarg3, ltos);
2999 __ cmovptr(Assembler::equal,
3000 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3001 __ cmpl(c_rarg3, dtos);
3002 __ cmovptr(Assembler::equal,
3003 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3004 #else
3005 __ cmpl(rcx, ltos);
3006 __ jccb(Assembler::equal, two_word);
3007 __ cmpl(rcx, dtos);
3008 __ jccb(Assembler::equal, two_word);
3009 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3010 __ jmpb(valsize_known);
3011
3012 __ bind(two_word);
3013 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3014
3015 __ bind(valsize_known);
3016 // setup object pointer
3017 __ movptr(rbx, Address(rbx, 0));
3018 #endif
3019 }
3020 // cache entry pointer
3021 __ addptr(robj, in_bytes(cp_base_offset));
3022 __ shll(RDX, LogBytesPerWord);
3023 __ addptr(robj, RDX);
3024 // object (tos)
3025 __ mov(RCX, rsp);
3026 // c_rarg1: object pointer set up above (NULL if static)
3027 // c_rarg2: cache entry pointer
3028 // c_rarg3: jvalue object on the stack
3029 __ call_VM(noreg,
3030 CAST_FROM_FN_PTR(address,
3031 InterpreterRuntime::post_field_modification),
3032 RBX, robj, RCX);
3033 __ get_cache_and_index_at_bcp(cache, index, 1);
3034 __ bind(L1);
3035 }
3036 }
3037
3038 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3039 transition(vtos, vtos);
3040
3041 const Register cache = rcx;
3042 const Register index = rdx;
3043 const Register obj = rcx;
3044 const Register off = rbx;
3045 const Register flags = rax;
3046 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3047
3048 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3049 jvmti_post_field_mod(cache, index, is_static);
3050 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3051
3052 // [jk] not needed currently
3053 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3054 // Assembler::StoreStore));
3055
3056 Label notVolatile, Done;
3057 __ movl(rdx, flags);
3058 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3059 __ andl(rdx, 0x1);
3060
3061 // field addresses
3062 const Address field(obj, off, Address::times_1, 0*wordSize);
3063 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3064
3065 Label notByte, notBool, notInt, notShort, notChar,
3066 notLong, notFloat, notObj, notDouble;
3067
3068 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3069
3070 assert(btos == 0, "change code, btos != 0");
3071 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3072 __ jcc(Assembler::notZero, notByte);
3073
3074 // btos
3075 {
3076 __ pop(btos);
3077 if (!is_static) pop_and_check_object(obj);
3078 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3079 __ movb(field, rax);
3080 if (!is_static && rc == may_rewrite) {
3081 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3082 }
3083 __ jmp(Done);
3084 }
3085
3086 __ bind(notByte);
3087 __ cmpl(flags, ztos);
3088 __ jcc(Assembler::notEqual, notBool);
3089
3090 // ztos
3091 {
3092 __ pop(ztos);
3093 if (!is_static) pop_and_check_object(obj);
3094 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3095 __ andl(rax, 0x1);
3096 __ movb(field, rax);
3097 if (!is_static && rc == may_rewrite) {
3098 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3099 }
3100 __ jmp(Done);
3101 }
3102
3103 __ bind(notBool);
3104 __ cmpl(flags, atos);
3105 __ jcc(Assembler::notEqual, notObj);
3106
3107 // atos
3108 {
3109 __ pop(atos);
3110 if (!is_static) pop_and_check_object(obj);
3111 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3112 // Store into the field
3113 do_oop_store(_masm, field, rax, _bs->kind(), false);
3114 if (!is_static && rc == may_rewrite) {
3115 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3116 }
3117 __ jmp(Done);
3118 }
3119
3120 __ bind(notObj);
3121 __ cmpl(flags, itos);
3122 __ jcc(Assembler::notEqual, notInt);
3123
3124 // itos
3125 {
3126 __ pop(itos);
3127 if (!is_static) pop_and_check_object(obj);
3128 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3129 __ movl(field, rax);
3130 if (!is_static && rc == may_rewrite) {
3131 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3132 }
3133 __ jmp(Done);
3134 }
3135
3136 __ bind(notInt);
3137 __ cmpl(flags, ctos);
3138 __ jcc(Assembler::notEqual, notChar);
3139
3140 // ctos
3141 {
3142 __ pop(ctos);
3143 if (!is_static) pop_and_check_object(obj);
3144 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3145 __ movw(field, rax);
3146 if (!is_static && rc == may_rewrite) {
3147 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3148 }
3149 __ jmp(Done);
3150 }
3151
3152 __ bind(notChar);
3153 __ cmpl(flags, stos);
3154 __ jcc(Assembler::notEqual, notShort);
3155
3156 // stos
3157 {
3158 __ pop(stos);
3159 if (!is_static) pop_and_check_object(obj);
3160 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3161 __ movw(field, rax);
3162 if (!is_static && rc == may_rewrite) {
3163 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3164 }
3165 __ jmp(Done);
3166 }
3167
3168 __ bind(notShort);
3169 __ cmpl(flags, ltos);
3170 __ jcc(Assembler::notEqual, notLong);
3171
3172 // ltos
3173 #ifdef _LP64
3174 {
3175 __ pop(ltos);
3176 if (!is_static) pop_and_check_object(obj);
3177 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3178 __ movq(field, rax);
3179 if (!is_static && rc == may_rewrite) {
3180 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3181 }
3182 __ jmp(Done);
3183 }
3184 #else
3185 {
3186 Label notVolatileLong;
3187 __ testl(rdx, rdx);
3188 __ jcc(Assembler::zero, notVolatileLong);
3189
3190 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3191 if (!is_static) pop_and_check_object(obj);
3192
3193 // Replace with real volatile test
3194 __ push(rdx);
3195 __ push(rax); // Must update atomically with FIST
3196 __ fild_d(Address(rsp,0)); // So load into FPU register
3197 __ fistp_d(field); // and put into memory atomically
3198 __ addptr(rsp, 2*wordSize);
3199 // volatile_barrier();
3200 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3201 Assembler::StoreStore));
3202 // Don't rewrite volatile version
3203 __ jmp(notVolatile);
3204
3205 __ bind(notVolatileLong);
3206
3207 __ pop(ltos); // overwrites rdx
3208 if (!is_static) pop_and_check_object(obj);
3209 __ movptr(hi, rdx);
3210 __ movptr(field, rax);
3211 // Don't rewrite to _fast_lputfield for potential volatile case.
3212 __ jmp(notVolatile);
3213 }
3214 #endif // _LP64
3215
3216 __ bind(notLong);
3217 __ cmpl(flags, ftos);
3218 __ jcc(Assembler::notEqual, notFloat);
3219
3220 // ftos
3221 {
3222 __ pop(ftos);
3223 if (!is_static) pop_and_check_object(obj);
3224 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3225 __ store_float(field);
3226 if (!is_static && rc == may_rewrite) {
3227 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3228 }
3229 __ jmp(Done);
3230 }
3231
3232 __ bind(notFloat);
3233 #ifdef ASSERT
3234 __ cmpl(flags, dtos);
3235 __ jcc(Assembler::notEqual, notDouble);
3236 #endif
3237
3238 // dtos
3239 {
3240 __ pop(dtos);
3241 if (!is_static) pop_and_check_object(obj);
3242 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3243 __ store_double(field);
3244 if (!is_static && rc == may_rewrite) {
3245 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3246 }
3247 }
3248
3249 #ifdef ASSERT
3250 __ jmp(Done);
3251
3252 __ bind(notDouble);
3253 __ stop("Bad state");
3254 #endif
3255
3256 __ bind(Done);
3257
3258 // Check for volatile store
3259 __ testl(rdx, rdx);
3260 __ jcc(Assembler::zero, notVolatile);
3261 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3262 Assembler::StoreStore));
3263 __ bind(notVolatile);
3264 }
3265
3266 void TemplateTable::putfield(int byte_no) {
3267 putfield_or_static(byte_no, false);
3268 }
3269
3270 void TemplateTable::nofast_putfield(int byte_no) {
3271 putfield_or_static(byte_no, false, may_not_rewrite);
3272 }
3273
3274 void TemplateTable::putstatic(int byte_no) {
3275 putfield_or_static(byte_no, true);
3276 }
3277
3278 void TemplateTable::jvmti_post_fast_field_mod() {
3279
3280 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3281
3282 if (JvmtiExport::can_post_field_modification()) {
3283 // Check to see if a field modification watch has been set before
3284 // we take the time to call into the VM.
3285 Label L2;
3286 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3287 __ testl(scratch, scratch);
3288 __ jcc(Assembler::zero, L2);
3289 __ pop_ptr(rbx); // copy the object pointer from tos
3290 __ verify_oop(rbx);
3291 __ push_ptr(rbx); // put the object pointer back on tos
3292 // Save tos values before call_VM() clobbers them. Since we have
3293 // to do it for every data type, we use the saved values as the
3294 // jvalue object.
3295 switch (bytecode()) { // load values into the jvalue object
3296 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3297 case Bytecodes::_fast_bputfield: // fall through
3298 case Bytecodes::_fast_zputfield: // fall through
3299 case Bytecodes::_fast_sputfield: // fall through
3300 case Bytecodes::_fast_cputfield: // fall through
3301 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3302 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3303 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3304 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3305
3306 default:
3307 ShouldNotReachHere();
3308 }
3309 __ mov(scratch, rsp); // points to jvalue on the stack
3310 // access constant pool cache entry
3311 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3312 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3313 __ verify_oop(rbx);
3314 // rbx: object pointer copied above
3315 // c_rarg2: cache entry pointer
3316 // c_rarg3: jvalue object on the stack
3317 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3318 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3319
3320 switch (bytecode()) { // restore tos values
3321 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3322 case Bytecodes::_fast_bputfield: // fall through
3323 case Bytecodes::_fast_zputfield: // fall through
3324 case Bytecodes::_fast_sputfield: // fall through
3325 case Bytecodes::_fast_cputfield: // fall through
3326 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3327 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3328 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3329 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3330 }
3331 __ bind(L2);
3332 }
3333 }
3334
3335 void TemplateTable::fast_storefield(TosState state) {
3336 transition(state, vtos);
3337
3338 ByteSize base = ConstantPoolCache::base_offset();
3339
3340 jvmti_post_fast_field_mod();
3341
3342 // access constant pool cache
3343 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3344
3345 // test for volatile with rdx but rdx is tos register for lputfield.
3346 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3347 in_bytes(base +
3348 ConstantPoolCacheEntry::flags_offset())));
3349
3350 // replace index with field offset from cache entry
3351 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3352 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3353
3354 // [jk] not needed currently
3355 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3356 // Assembler::StoreStore));
3357
3358 Label notVolatile;
3359 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3360 __ andl(rdx, 0x1);
3361
3362 // Get object from stack
3363 pop_and_check_object(rcx);
3364 oopDesc::bs()->interpreter_write_barrier(_masm, rcx);
3365
3366 // field address
3367 const Address field(rcx, rbx, Address::times_1);
3368
3369 // access field
3370 switch (bytecode()) {
3371 case Bytecodes::_fast_aputfield:
3372 do_oop_store(_masm, field, rax, _bs->kind(), false);
3373 break;
3374 case Bytecodes::_fast_lputfield:
3375 #ifdef _LP64
3376 __ movq(field, rax);
3377 #else
3378 __ stop("should not be rewritten");
3379 #endif
3380 break;
3381 case Bytecodes::_fast_iputfield:
3382 __ movl(field, rax);
3383 break;
3384 case Bytecodes::_fast_zputfield:
3385 __ andl(rax, 0x1); // boolean is true if LSB is 1
3386 // fall through to bputfield
3387 case Bytecodes::_fast_bputfield:
3388 __ movb(field, rax);
3389 break;
3390 case Bytecodes::_fast_sputfield:
3391 // fall through
3392 case Bytecodes::_fast_cputfield:
3393 __ movw(field, rax);
3394 break;
3395 case Bytecodes::_fast_fputfield:
3396 __ store_float(field);
3397 break;
3398 case Bytecodes::_fast_dputfield:
3399 __ store_double(field);
3400 break;
3401 default:
3402 ShouldNotReachHere();
3403 }
3404
3405 // Check for volatile store
3406 __ testl(rdx, rdx);
3407 __ jcc(Assembler::zero, notVolatile);
3408 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3409 Assembler::StoreStore));
3410 __ bind(notVolatile);
3411 }
3412
3413 void TemplateTable::fast_accessfield(TosState state) {
3414 transition(atos, state);
3415
3416 // Do the JVMTI work here to avoid disturbing the register state below
3417 if (JvmtiExport::can_post_field_access()) {
3418 // Check to see if a field access watch has been set before we
3419 // take the time to call into the VM.
3420 Label L1;
3421 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3422 __ testl(rcx, rcx);
3423 __ jcc(Assembler::zero, L1);
3424 // access constant pool cache entry
3425 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3426 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3427 __ verify_oop(rax);
3428 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3429 LP64_ONLY(__ mov(c_rarg1, rax));
3430 // c_rarg1: object pointer copied above
3431 // c_rarg2: cache entry pointer
3432 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3433 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3434 __ pop_ptr(rax); // restore object pointer
3435 __ bind(L1);
3436 }
3437
3438 // access constant pool cache
3439 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3440 // replace index with field offset from cache entry
3441 // [jk] not needed currently
3442 // if (os::is_MP()) {
3443 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3444 // in_bytes(ConstantPoolCache::base_offset() +
3445 // ConstantPoolCacheEntry::flags_offset())));
3446 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3447 // __ andl(rdx, 0x1);
3448 // }
3449 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3450 in_bytes(ConstantPoolCache::base_offset() +
3451 ConstantPoolCacheEntry::f2_offset())));
3452
3453 // rax: object
3454 __ verify_oop(rax);
3455 __ null_check(rax);
3456 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rax);
3457 Address field(rax, rbx, Address::times_1);
3458
3459 // access field
3460 switch (bytecode()) {
3461 case Bytecodes::_fast_agetfield:
3462 __ load_heap_oop(rax, field);
3463 __ verify_oop(rax);
3464 break;
3465 case Bytecodes::_fast_lgetfield:
3466 #ifdef _LP64
3467 __ movq(rax, field);
3468 #else
3469 __ stop("should not be rewritten");
3470 #endif
3471 break;
3472 case Bytecodes::_fast_igetfield:
3473 __ movl(rax, field);
3474 break;
3475 case Bytecodes::_fast_bgetfield:
3476 __ movsbl(rax, field);
3477 break;
3478 case Bytecodes::_fast_sgetfield:
3479 __ load_signed_short(rax, field);
3480 break;
3481 case Bytecodes::_fast_cgetfield:
3482 __ load_unsigned_short(rax, field);
3483 break;
3484 case Bytecodes::_fast_fgetfield:
3485 __ load_float(field);
3486 break;
3487 case Bytecodes::_fast_dgetfield:
3488 __ load_double(field);
3489 break;
3490 default:
3491 ShouldNotReachHere();
3492 }
3493 // [jk] not needed currently
3494 // if (os::is_MP()) {
3495 // Label notVolatile;
3496 // __ testl(rdx, rdx);
3497 // __ jcc(Assembler::zero, notVolatile);
3498 // __ membar(Assembler::LoadLoad);
3499 // __ bind(notVolatile);
3500 //};
3501 }
3502
3503 void TemplateTable::fast_xaccess(TosState state) {
3504 transition(vtos, state);
3505
3506 // get receiver
3507 __ movptr(rax, aaddress(0));
3508 // access constant pool cache
3509 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3510 __ movptr(rbx,
3511 Address(rcx, rdx, Address::times_ptr,
3512 in_bytes(ConstantPoolCache::base_offset() +
3513 ConstantPoolCacheEntry::f2_offset())));
3514 // make sure exception is reported in correct bcp range (getfield is
3515 // next instruction)
3516 __ increment(rbcp);
3517 __ null_check(rax);
3518 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rax);
3519 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3520 switch (state) {
3521 case itos:
3522 __ movl(rax, field);
3523 break;
3524 case atos:
3525 __ load_heap_oop(rax, field);
3526 __ verify_oop(rax);
3527 break;
3528 case ftos:
3529 __ load_float(field);
3530 break;
3531 default:
3532 ShouldNotReachHere();
3533 }
3534
3535 // [jk] not needed currently
3536 // if (os::is_MP()) {
3537 // Label notVolatile;
3538 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3539 // in_bytes(ConstantPoolCache::base_offset() +
3540 // ConstantPoolCacheEntry::flags_offset())));
3541 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3542 // __ testl(rdx, 0x1);
3543 // __ jcc(Assembler::zero, notVolatile);
3544 // __ membar(Assembler::LoadLoad);
3545 // __ bind(notVolatile);
3546 // }
3547
3548 __ decrement(rbcp);
3549 }
3550
3551 //-----------------------------------------------------------------------------
3552 // Calls
3553
3554 void TemplateTable::count_calls(Register method, Register temp) {
3555 // implemented elsewhere
3556 ShouldNotReachHere();
3557 }
3558
3559 void TemplateTable::prepare_invoke(int byte_no,
3560 Register method, // linked method (or i-klass)
3561 Register index, // itable index, MethodType, etc.
3562 Register recv, // if caller wants to see it
3563 Register flags // if caller wants to test it
3564 ) {
3565 // determine flags
3566 const Bytecodes::Code code = bytecode();
3567 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3568 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3569 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3570 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3571 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3572 const bool load_receiver = (recv != noreg);
3573 const bool save_flags = (flags != noreg);
3574 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3575 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3576 assert(flags == noreg || flags == rdx, "");
3577 assert(recv == noreg || recv == rcx, "");
3578
3579 // setup registers & access constant pool cache
3580 if (recv == noreg) recv = rcx;
3581 if (flags == noreg) flags = rdx;
3582 assert_different_registers(method, index, recv, flags);
3583
3584 // save 'interpreter return address'
3585 __ save_bcp();
3586
3587 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3588
3589 // maybe push appendix to arguments (just before return address)
3590 if (is_invokedynamic || is_invokehandle) {
3591 Label L_no_push;
3592 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3593 __ jcc(Assembler::zero, L_no_push);
3594 // Push the appendix as a trailing parameter.
3595 // This must be done before we get the receiver,
3596 // since the parameter_size includes it.
3597 __ push(rbx);
3598 __ mov(rbx, index);
3599 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3600 __ load_resolved_reference_at_index(index, rbx);
3601 __ pop(rbx);
3602 __ push(index); // push appendix (MethodType, CallSite, etc.)
3603 __ bind(L_no_push);
3604 }
3605
3606 // load receiver if needed (after appendix is pushed so parameter size is correct)
3607 // Note: no return address pushed yet
3608 if (load_receiver) {
3609 __ movl(recv, flags);
3610 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3611 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3612 const int receiver_is_at_end = -1; // back off one slot to get receiver
3613 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3614 __ movptr(recv, recv_addr);
3615 __ verify_oop(recv);
3616 }
3617
3618 if (save_flags) {
3619 __ movl(rbcp, flags);
3620 }
3621
3622 // compute return type
3623 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3624 // Make sure we don't need to mask flags after the above shift
3625 ConstantPoolCacheEntry::verify_tos_state_shift();
3626 // load return address
3627 {
3628 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3629 ExternalAddress table(table_addr);
3630 LP64_ONLY(__ lea(rscratch1, table));
3631 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3632 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3633 }
3634
3635 // push return address
3636 __ push(flags);
3637
3638 // Restore flags value from the constant pool cache, and restore rsi
3639 // for later null checks. r13 is the bytecode pointer
3640 if (save_flags) {
3641 __ movl(flags, rbcp);
3642 __ restore_bcp();
3643 }
3644 }
3645
3646 void TemplateTable::invokevirtual_helper(Register index,
3647 Register recv,
3648 Register flags) {
3649 // Uses temporary registers rax, rdx
3650 assert_different_registers(index, recv, rax, rdx);
3651 assert(index == rbx, "");
3652 assert(recv == rcx, "");
3653
3654 // Test for an invoke of a final method
3655 Label notFinal;
3656 __ movl(rax, flags);
3657 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3658 __ jcc(Assembler::zero, notFinal);
3659
3660 const Register method = index; // method must be rbx
3661 assert(method == rbx,
3662 "Method* must be rbx for interpreter calling convention");
3663
3664 // do the call - the index is actually the method to call
3665 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3666
3667 // It's final, need a null check here!
3668 __ null_check(recv);
3669
3670 // profile this call
3671 __ profile_final_call(rax);
3672 __ profile_arguments_type(rax, method, rbcp, true);
3673
3674 __ jump_from_interpreted(method, rax);
3675
3676 __ bind(notFinal);
3677
3678 // get receiver klass
3679 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3680 __ load_klass(rax, recv);
3681
3682 // profile this call
3683 __ profile_virtual_call(rax, rlocals, rdx);
3684 // get target Method* & entry point
3685 __ lookup_virtual_method(rax, index, method);
3686 __ profile_called_method(method, rdx, rbcp);
3687
3688 __ profile_arguments_type(rdx, method, rbcp, true);
3689 __ jump_from_interpreted(method, rdx);
3690 }
3691
3692 void TemplateTable::invokevirtual(int byte_no) {
3693 transition(vtos, vtos);
3694 assert(byte_no == f2_byte, "use this argument");
3695 prepare_invoke(byte_no,
3696 rbx, // method or vtable index
3697 noreg, // unused itable index
3698 rcx, rdx); // recv, flags
3699
3700 // rbx: index
3701 // rcx: receiver
3702 // rdx: flags
3703
3704 invokevirtual_helper(rbx, rcx, rdx);
3705 }
3706
3707 void TemplateTable::invokespecial(int byte_no) {
3708 transition(vtos, vtos);
3709 assert(byte_no == f1_byte, "use this argument");
3710 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3711 rcx); // get receiver also for null check
3712 __ verify_oop(rcx);
3713 __ null_check(rcx);
3714 // do the call
3715 __ profile_call(rax);
3716 __ profile_arguments_type(rax, rbx, rbcp, false);
3717 __ jump_from_interpreted(rbx, rax);
3718 }
3719
3720 void TemplateTable::invokestatic(int byte_no) {
3721 transition(vtos, vtos);
3722 assert(byte_no == f1_byte, "use this argument");
3723 prepare_invoke(byte_no, rbx); // get f1 Method*
3724 // do the call
3725 __ profile_call(rax);
3726 __ profile_arguments_type(rax, rbx, rbcp, false);
3727 __ jump_from_interpreted(rbx, rax);
3728 }
3729
3730
3731 void TemplateTable::fast_invokevfinal(int byte_no) {
3732 transition(vtos, vtos);
3733 assert(byte_no == f2_byte, "use this argument");
3734 __ stop("fast_invokevfinal not used on x86");
3735 }
3736
3737
3738 void TemplateTable::invokeinterface(int byte_no) {
3739 transition(vtos, vtos);
3740 assert(byte_no == f1_byte, "use this argument");
3741 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3742 rcx, rdx); // recv, flags
3743
3744 // rax: interface klass (from f1)
3745 // rbx: itable index (from f2)
3746 // rcx: receiver
3747 // rdx: flags
3748
3749 // Special case of invokeinterface called for virtual method of
3750 // java.lang.Object. See cpCacheOop.cpp for details.
3751 // This code isn't produced by javac, but could be produced by
3752 // another compliant java compiler.
3753 Label notMethod;
3754 __ movl(rlocals, rdx);
3755 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3756
3757 __ jcc(Assembler::zero, notMethod);
3758
3759 invokevirtual_helper(rbx, rcx, rdx);
3760 __ bind(notMethod);
3761
3762 // Get receiver klass into rdx - also a null check
3763 __ restore_locals(); // restore r14
3764 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3765 __ load_klass(rdx, rcx);
3766
3767 // profile this call
3768 __ profile_virtual_call(rdx, rbcp, rlocals);
3769
3770 Label no_such_interface, no_such_method;
3771
3772 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3773 rdx, rax, rbx,
3774 // outputs: method, scan temp. reg
3775 rbx, rbcp,
3776 no_such_interface);
3777
3778 // rbx: Method* to call
3779 // rcx: receiver
3780 // Check for abstract method error
3781 // Note: This should be done more efficiently via a throw_abstract_method_error
3782 // interpreter entry point and a conditional jump to it in case of a null
3783 // method.
3784 __ testptr(rbx, rbx);
3785 __ jcc(Assembler::zero, no_such_method);
3786
3787 __ profile_called_method(rbx, rbcp, rdx);
3788 __ profile_arguments_type(rdx, rbx, rbcp, true);
3789
3790 // do the call
3791 // rcx: receiver
3792 // rbx,: Method*
3793 __ jump_from_interpreted(rbx, rdx);
3794 __ should_not_reach_here();
3795
3796 // exception handling code follows...
3797 // note: must restore interpreter registers to canonical
3798 // state for exception handling to work correctly!
3799
3800 __ bind(no_such_method);
3801 // throw exception
3802 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3803 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3804 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3805 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3806 // the call_VM checks for exception, so we should never return here.
3807 __ should_not_reach_here();
3808
3809 __ bind(no_such_interface);
3810 // throw exception
3811 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3812 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3813 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3814 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3815 InterpreterRuntime::throw_IncompatibleClassChangeError));
3816 // the call_VM checks for exception, so we should never return here.
3817 __ should_not_reach_here();
3818 }
3819
3820 void TemplateTable::invokehandle(int byte_no) {
3821 transition(vtos, vtos);
3822 assert(byte_no == f1_byte, "use this argument");
3823 const Register rbx_method = rbx;
3824 const Register rax_mtype = rax;
3825 const Register rcx_recv = rcx;
3826 const Register rdx_flags = rdx;
3827
3828 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3829 __ verify_method_ptr(rbx_method);
3830 __ verify_oop(rcx_recv);
3831 __ null_check(rcx_recv);
3832
3833 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3834 // rbx: MH.invokeExact_MT method (from f2)
3835
3836 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3837
3838 // FIXME: profile the LambdaForm also
3839 __ profile_final_call(rax);
3840 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3841
3842 __ jump_from_interpreted(rbx_method, rdx);
3843 }
3844
3845 void TemplateTable::invokedynamic(int byte_no) {
3846 transition(vtos, vtos);
3847 assert(byte_no == f1_byte, "use this argument");
3848
3849 const Register rbx_method = rbx;
3850 const Register rax_callsite = rax;
3851
3852 prepare_invoke(byte_no, rbx_method, rax_callsite);
3853
3854 // rax: CallSite object (from cpool->resolved_references[f1])
3855 // rbx: MH.linkToCallSite method (from f2)
3856
3857 // Note: rax_callsite is already pushed by prepare_invoke
3858
3859 // %%% should make a type profile for any invokedynamic that takes a ref argument
3860 // profile this call
3861 __ profile_call(rbcp);
3862 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3863
3864 __ verify_oop(rax_callsite);
3865
3866 __ jump_from_interpreted(rbx_method, rdx);
3867 }
3868
3869 //-----------------------------------------------------------------------------
3870 // Allocation
3871
3872 void TemplateTable::_new() {
3873 transition(vtos, atos);
3874 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3875 Label slow_case;
3876 Label slow_case_no_pop;
3877 Label done;
3878 Label initialize_header;
3879 Label initialize_object; // including clearing the fields
3880 Label allocate_shared;
3881
3882 __ get_cpool_and_tags(rcx, rax);
3883
3884 // Make sure the class we're about to instantiate has been resolved.
3885 // This is done before loading InstanceKlass to be consistent with the order
3886 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3887 const int tags_offset = Array<u1>::base_offset_in_bytes();
3888 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3889 __ jcc(Assembler::notEqual, slow_case_no_pop);
3890
3891 // get InstanceKlass
3892 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool)));
3893 __ push(rcx); // save the contexts of klass for initializing the header
3894
3895 // make sure klass is initialized & doesn't have finalizer
3896 // make sure klass is fully initialized
3897 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3898 __ jcc(Assembler::notEqual, slow_case);
3899
3900 // get instance_size in InstanceKlass (scaled to a count of bytes)
3901 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3902 // test to see if it has a finalizer or is malformed in some way
3903 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3904 __ jcc(Assembler::notZero, slow_case);
3905
3906 //
3907 // Allocate the instance
3908 // 1) Try to allocate in the TLAB
3909 // 2) if fail and the object is large allocate in the shared Eden
3910 // 3) if the above fails (or is not applicable), go to a slow case
3911 // (creates a new TLAB, etc.)
3912
3913 const bool allow_shared_alloc =
3914 Universe::heap()->supports_inline_contig_alloc();
3915
3916 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
3917 #ifndef _LP64
3918 if (UseTLAB || allow_shared_alloc) {
3919 __ get_thread(thread);
3920 }
3921 #endif // _LP64
3922
3923 if (UseTLAB) {
3924 uint oop_extra_words = Universe::heap()->oop_extra_words();
3925 if (oop_extra_words > 0) {
3926 __ addq(rdx, oop_extra_words * HeapWordSize);
3927 }
3928
3929 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3930 __ lea(rbx, Address(rax, rdx, Address::times_1));
3931 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3932 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3933 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3934 Universe::heap()->compile_prepare_oop(_masm, rax);
3935 if (ZeroTLAB) {
3936 // the fields have been already cleared
3937 __ jmp(initialize_header);
3938 } else {
3939 // initialize both the header and fields
3940 __ jmp(initialize_object);
3941 }
3942 }
3943
3944 // Allocation in the shared Eden, if allowed.
3945 //
3946 // rdx: instance size in bytes
3947 if (allow_shared_alloc) {
3948 __ bind(allocate_shared);
3949
3950 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3951 ExternalAddress heap_end((address)Universe::heap()->end_addr());
3952
3953 Label retry;
3954 __ bind(retry);
3955 __ movptr(rax, heap_top);
3956 __ lea(rbx, Address(rax, rdx, Address::times_1));
3957 __ cmpptr(rbx, heap_end);
3958 __ jcc(Assembler::above, slow_case);
3959
3960 // Compare rax, with the top addr, and if still equal, store the new
3961 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3962 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3963 //
3964 // rax,: object begin
3965 // rbx,: object end
3966 // rdx: instance size in bytes
3967 __ locked_cmpxchgptr(rbx, heap_top);
3968
3969 // if someone beat us on the allocation, try again, otherwise continue
3970 __ jcc(Assembler::notEqual, retry);
3971
3972 __ incr_allocated_bytes(thread, rdx, 0);
3973 }
3974
3975 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3976 // The object is initialized before the header. If the object size is
3977 // zero, go directly to the header initialization.
3978 __ bind(initialize_object);
3979 __ decrement(rdx, sizeof(oopDesc));
3980 __ jcc(Assembler::zero, initialize_header);
3981
3982 // Initialize topmost object field, divide rdx by 8, check if odd and
3983 // test if zero.
3984 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3985 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3986
3987 // rdx must have been multiple of 8
3988 #ifdef ASSERT
3989 // make sure rdx was multiple of 8
3990 Label L;
3991 // Ignore partial flag stall after shrl() since it is debug VM
3992 __ jccb(Assembler::carryClear, L);
3993 __ stop("object size is not multiple of 2 - adjust this code");
3994 __ bind(L);
3995 // rdx must be > 0, no extra check needed here
3996 #endif
3997
3998 // initialize remaining object fields: rdx was a multiple of 8
3999 { Label loop;
4000 __ bind(loop);
4001 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4002 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4003 __ decrement(rdx);
4004 __ jcc(Assembler::notZero, loop);
4005 }
4006
4007 // initialize object header only.
4008 __ bind(initialize_header);
4009 if (UseBiasedLocking) {
4010 __ pop(rcx); // get saved klass back in the register.
4011 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4012 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4013 } else {
4014 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4015 (intptr_t)markOopDesc::prototype()); // header
4016 __ pop(rcx); // get saved klass back in the register.
4017 }
4018 #ifdef _LP64
4019 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4020 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
4021 #endif
4022 __ store_klass(rax, rcx); // klass
4023
4024 {
4025 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4026 // Trigger dtrace event for fastpath
4027 __ push(atos);
4028 __ call_VM_leaf(
4029 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4030 __ pop(atos);
4031 }
4032
4033 __ jmp(done);
4034 }
4035
4036 // slow case
4037 __ bind(slow_case);
4038 __ pop(rcx); // restore stack pointer to what it was when we came in.
4039 __ bind(slow_case_no_pop);
4040
4041 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4042 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4043
4044 __ get_constant_pool(rarg1);
4045 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4046 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4047 __ verify_oop(rax);
4048
4049 // continue
4050 __ bind(done);
4051 }
4052
4053 void TemplateTable::newarray() {
4054 transition(itos, atos);
4055 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4056 __ load_unsigned_byte(rarg1, at_bcp(1));
4057 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4058 rarg1, rax);
4059 }
4060
4061 void TemplateTable::anewarray() {
4062 transition(itos, atos);
4063
4064 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4065 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4066
4067 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4068 __ get_constant_pool(rarg1);
4069 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4070 rarg1, rarg2, rax);
4071 }
4072
4073 void TemplateTable::arraylength() {
4074 transition(atos, itos);
4075 if (ShenandoahVerifyReadsToFromSpace) {
4076 oopDesc::bs()->interpreter_read_barrier(_masm, rax);
4077 }
4078 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4079 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4080 }
4081
4082 void TemplateTable::checkcast() {
4083 transition(atos, atos);
4084 Label done, is_null, ok_is_subtype, quicked, resolved;
4085 __ testptr(rax, rax); // object is in rax
4086 __ jcc(Assembler::zero, is_null);
4087
4088 // Get cpool & tags index
4089 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4090 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4091 // See if bytecode has already been quicked
4092 __ cmpb(Address(rdx, rbx,
4093 Address::times_1,
4094 Array<u1>::base_offset_in_bytes()),
4095 JVM_CONSTANT_Class);
4096 __ jcc(Assembler::equal, quicked);
4097 __ push(atos); // save receiver for result, and for GC
4098 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4099
4100 // vm_result_2 has metadata result
4101 #ifndef _LP64
4102 // borrow rdi from locals
4103 __ get_thread(rdi);
4104 __ get_vm_result_2(rax, rdi);
4105 __ restore_locals();
4106 #else
4107 __ get_vm_result_2(rax, r15_thread);
4108 #endif
4109
4110 __ pop_ptr(rdx); // restore receiver
4111 __ jmpb(resolved);
4112
4113 // Get superklass in rax and subklass in rbx
4114 __ bind(quicked);
4115 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4116 __ movptr(rax, Address(rcx, rbx,
4117 Address::times_ptr, sizeof(ConstantPool)));
4118
4119 __ bind(resolved);
4120 __ load_klass(rbx, rdx);
4121
4122 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4123 // Superklass in rax. Subklass in rbx.
4124 __ gen_subtype_check(rbx, ok_is_subtype);
4125
4126 // Come here on failure
4127 __ push_ptr(rdx);
4128 // object is at TOS
4129 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4130
4131 // Come here on success
4132 __ bind(ok_is_subtype);
4133 __ mov(rax, rdx); // Restore object in rdx
4134
4135 // Collect counts on whether this check-cast sees NULLs a lot or not.
4136 if (ProfileInterpreter) {
4137 __ jmp(done);
4138 __ bind(is_null);
4139 __ profile_null_seen(rcx);
4140 } else {
4141 __ bind(is_null); // same as 'done'
4142 }
4143 __ bind(done);
4144 }
4145
4146 void TemplateTable::instanceof() {
4147 transition(atos, itos);
4148 Label done, is_null, ok_is_subtype, quicked, resolved;
4149 __ testptr(rax, rax);
4150 __ jcc(Assembler::zero, is_null);
4151
4152 // Get cpool & tags index
4153 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4154 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4155 // See if bytecode has already been quicked
4156 __ cmpb(Address(rdx, rbx,
4157 Address::times_1,
4158 Array<u1>::base_offset_in_bytes()),
4159 JVM_CONSTANT_Class);
4160 __ jcc(Assembler::equal, quicked);
4161
4162 __ push(atos); // save receiver for result, and for GC
4163 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4164 // vm_result_2 has metadata result
4165
4166 #ifndef _LP64
4167 // borrow rdi from locals
4168 __ get_thread(rdi);
4169 __ get_vm_result_2(rax, rdi);
4170 __ restore_locals();
4171 #else
4172 __ get_vm_result_2(rax, r15_thread);
4173 #endif
4174
4175 __ pop_ptr(rdx); // restore receiver
4176 __ verify_oop(rdx);
4177 __ load_klass(rdx, rdx);
4178 if (ShenandoahVerifyReadsToFromSpace) {
4179 __ jmp(resolved);
4180 } else {
4181 __ jmpb(resolved);
4182 }
4183
4184 // Get superklass in rax and subklass in rdx
4185 __ bind(quicked);
4186 __ load_klass(rdx, rax);
4187 __ movptr(rax, Address(rcx, rbx,
4188 Address::times_ptr, sizeof(ConstantPool)));
4189
4190 __ bind(resolved);
4191
4192 // Generate subtype check. Blows rcx, rdi
4193 // Superklass in rax. Subklass in rdx.
4194 __ gen_subtype_check(rdx, ok_is_subtype);
4195
4196 // Come here on failure
4197 __ xorl(rax, rax);
4198 __ jmpb(done);
4199 // Come here on success
4200 __ bind(ok_is_subtype);
4201 __ movl(rax, 1);
4202
4203 // Collect counts on whether this test sees NULLs a lot or not.
4204 if (ProfileInterpreter) {
4205 __ jmp(done);
4206 __ bind(is_null);
4207 __ profile_null_seen(rcx);
4208 } else {
4209 __ bind(is_null); // same as 'done'
4210 }
4211 __ bind(done);
4212 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4213 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4214 }
4215
4216
4217 //----------------------------------------------------------------------------------------------------
4218 // Breakpoints
4219 void TemplateTable::_breakpoint() {
4220 // Note: We get here even if we are single stepping..
4221 // jbug insists on setting breakpoints at every bytecode
4222 // even if we are in single step mode.
4223
4224 transition(vtos, vtos);
4225
4226 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4227
4228 // get the unpatched byte code
4229 __ get_method(rarg);
4230 __ call_VM(noreg,
4231 CAST_FROM_FN_PTR(address,
4232 InterpreterRuntime::get_original_bytecode_at),
4233 rarg, rbcp);
4234 __ mov(rbx, rax); // why?
4235
4236 // post the breakpoint event
4237 __ get_method(rarg);
4238 __ call_VM(noreg,
4239 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4240 rarg, rbcp);
4241
4242 // complete the execution of original bytecode
4243 __ dispatch_only_normal(vtos);
4244 }
4245
4246 //-----------------------------------------------------------------------------
4247 // Exceptions
4248
4249 void TemplateTable::athrow() {
4250 transition(atos, vtos);
4251 __ null_check(rax);
4252 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4253 }
4254
4255 //-----------------------------------------------------------------------------
4256 // Synchronization
4257 //
4258 // Note: monitorenter & exit are symmetric routines; which is reflected
4259 // in the assembly code structure as well
4260 //
4261 // Stack layout:
4262 //
4263 // [expressions ] <--- rsp = expression stack top
4264 // ..
4265 // [expressions ]
4266 // [monitor entry] <--- monitor block top = expression stack bot
4267 // ..
4268 // [monitor entry]
4269 // [frame data ] <--- monitor block bot
4270 // ...
4271 // [saved rbp ] <--- rbp
4272 void TemplateTable::monitorenter() {
4273 transition(atos, vtos);
4274
4275 // check for NULL object
4276 __ null_check(rax);
4277
4278 oopDesc::bs()->interpreter_write_barrier(_masm, rax);
4279
4280 const Address monitor_block_top(
4281 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4282 const Address monitor_block_bot(
4283 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4284 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4285
4286 Label allocated;
4287
4288 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4289 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4290 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4291
4292 // initialize entry pointer
4293 __ xorl(rmon, rmon); // points to free slot or NULL
4294
4295 // find a free slot in the monitor block (result in rmon)
4296 {
4297 Label entry, loop, exit;
4298 __ movptr(rtop, monitor_block_top); // points to current entry,
4299 // starting with top-most entry
4300 __ lea(rbot, monitor_block_bot); // points to word before bottom
4301 // of monitor block
4302 __ jmpb_if_possible(entry);
4303
4304 __ bind(loop);
4305 // check if current entry is used
4306 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4307 // if not used then remember entry in rmon
4308 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4309 // check if current entry is for same object
4310 __ shenandoah_lock_check(rtop); // Invariant
4311 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4312 // if same object then stop searching
4313 __ jccb(Assembler::equal, exit);
4314 // otherwise advance to next entry
4315 __ addptr(rtop, entry_size);
4316 __ bind(entry);
4317 // check if bottom reached
4318 __ cmpptr(rtop, rbot);
4319 // if not at bottom then check this entry
4320 __ jcc(Assembler::notEqual, loop);
4321 __ bind(exit);
4322 }
4323
4324 __ testptr(rmon, rmon); // check if a slot has been found
4325 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4326
4327 // allocate one if there's no free slot
4328 {
4329 Label entry, loop;
4330 // 1. compute new pointers // rsp: old expression stack top
4331 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4332 __ subptr(rsp, entry_size); // move expression stack top
4333 __ subptr(rmon, entry_size); // move expression stack bottom
4334 __ mov(rtop, rsp); // set start value for copy loop
4335 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4336 __ jmp(entry);
4337 // 2. move expression stack contents
4338 __ bind(loop);
4339 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4340 // word from old location
4341 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4342 __ addptr(rtop, wordSize); // advance to next word
4343 __ bind(entry);
4344 __ cmpptr(rtop, rmon); // check if bottom reached
4345 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4346 // copy next word
4347 }
4348
4349 // call run-time routine
4350 // rmon: points to monitor entry
4351 __ bind(allocated);
4352
4353 // Increment bcp to point to the next bytecode, so exception
4354 // handling for async. exceptions work correctly.
4355 // The object has already been poped from the stack, so the
4356 // expression stack looks correct.
4357 __ increment(rbcp);
4358
4359 // store object
4360 __ shenandoah_store_addr_check(rax); // Invariant
4361 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4362 __ lock_object(rmon);
4363
4364 // check to make sure this monitor doesn't cause stack overflow after locking
4365 __ save_bcp(); // in case of exception
4366 __ generate_stack_overflow_check(0);
4367
4368 // The bcp has already been incremented. Just need to dispatch to
4369 // next instruction.
4370 __ dispatch_next(vtos);
4371 }
4372
4373 void TemplateTable::monitorexit() {
4374 transition(atos, vtos);
4375
4376 // check for NULL object
4377 __ null_check(rax);
4378
4379 oopDesc::bs()->interpreter_write_barrier(_masm, rax);
4380
4381 const Address monitor_block_top(
4382 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4383 const Address monitor_block_bot(
4384 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4385 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4386
4387 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4388 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4389
4390 Label found;
4391
4392 // find matching slot
4393 {
4394 Label entry, loop;
4395 __ movptr(rtop, monitor_block_top); // points to current entry,
4396 // starting with top-most entry
4397 __ lea(rbot, monitor_block_bot); // points to word before bottom
4398 // of monitor block
4399 __ jmpb_if_possible(entry);
4400
4401 __ bind(loop);
4402 // check if current entry is for same object
4403 __ shenandoah_lock_check(rtop); // Invariant
4404 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4405 // if same object then stop searching
4406 __ jcc(Assembler::equal, found);
4407 // otherwise advance to next entry
4408 __ addptr(rtop, entry_size);
4409 __ bind(entry);
4410 // check if bottom reached
4411 __ cmpptr(rtop, rbot);
4412 // if not at bottom then check this entry
4413 __ jcc(Assembler::notEqual, loop);
4414 }
4415
4416 // error handling. Unlocking was not block-structured
4417 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4418 InterpreterRuntime::throw_illegal_monitor_state_exception));
4419 __ should_not_reach_here();
4420
4421 // call run-time routine
4422 __ bind(found);
4423 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4424 __ unlock_object(rtop);
4425 __ pop_ptr(rax); // discard object
4426 }
4427
4428 // Wide instructions
4429 void TemplateTable::wide() {
4430 transition(vtos, vtos);
4431 __ load_unsigned_byte(rbx, at_bcp(1));
4432 ExternalAddress wtable((address)Interpreter::_wentry_point);
4433 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4434 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4435 }
4436
4437 // Multi arrays
4438 void TemplateTable::multianewarray() {
4439 transition(vtos, atos);
4440
4441 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4442 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4443 // last dim is on top of stack; we want address of first one:
4444 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4445 // the latter wordSize to point to the beginning of the array.
4446 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4447 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4448 __ load_unsigned_byte(rbx, at_bcp(3));
4449 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4450 }