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 // For Shenandoah, make sure we only store refs into to-space.
186 oopDesc::bs()->interpreter_read_barrier(_masm, val);
187
188 // G1 barrier needs uncompressed oop for region cross check.
189 Register new_val = val;
190 if (UseCompressedOops) {
191 new_val = rbx;
192 __ movptr(new_val, val);
193 }
194
195 __ store_heap_oop(Address(rdx, 0), val);
196 if (UseG1GC) {
197 __ g1_write_barrier_post(rdx /* store_adr */,
198 new_val /* new_val */,
199 rthread /* thread */,
200 rtmp /* tmp */,
201 rbx /* tmp2 */);
202 } else {
203 assert(UseShenandoahGC, "sanity");
204 __ shenandoah_write_barrier_post(rdx /* store_adr */,
205 new_val /* new_val */,
206 rthread /* thread */,
207 rtmp /* tmp */,
208 rbx /* tmp2 */);
209 }
210 }
211 NOT_LP64( __ restore_bcp());
212 }
213 break;
214 #endif // INCLUDE_ALL_GCS
215 case BarrierSet::CardTableForRS:
216 case BarrierSet::CardTableExtension:
217 {
218 if (val == noreg) {
219 __ store_heap_oop_null(obj);
220 } else {
221 __ store_heap_oop(obj, val);
222 // flatten object address if needed
223 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
224 __ store_check(obj.base());
225 } else {
226 __ lea(rdx, obj);
227 __ store_check(rdx);
228 }
229 }
230 }
231 break;
232 case BarrierSet::ModRef:
233 if (val == noreg) {
234 __ store_heap_oop_null(obj);
235 } else {
236 __ store_heap_oop(obj, val);
237 }
238 break;
239 default :
240 ShouldNotReachHere();
241
242 }
243 }
244
245 Address TemplateTable::at_bcp(int offset) {
246 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
247 return Address(rbcp, offset);
248 }
249
250
251 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
252 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
253 int byte_no) {
254 if (!RewriteBytecodes) return;
255 Label L_patch_done;
256
257 switch (bc) {
258 case Bytecodes::_fast_aputfield:
259 case Bytecodes::_fast_bputfield:
260 case Bytecodes::_fast_zputfield:
261 case Bytecodes::_fast_cputfield:
262 case Bytecodes::_fast_dputfield:
263 case Bytecodes::_fast_fputfield:
264 case Bytecodes::_fast_iputfield:
265 case Bytecodes::_fast_lputfield:
266 case Bytecodes::_fast_sputfield:
267 {
268 // We skip bytecode quickening for putfield instructions when
269 // the put_code written to the constant pool cache is zero.
270 // This is required so that every execution of this instruction
271 // calls out to InterpreterRuntime::resolve_get_put to do
272 // additional, required work.
273 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
274 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
275 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
276 __ movl(bc_reg, bc);
277 __ cmpl(temp_reg, (int) 0);
278 __ jcc(Assembler::zero, L_patch_done); // don't patch
279 }
280 break;
281 default:
282 assert(byte_no == -1, "sanity");
283 // the pair bytecodes have already done the load.
284 if (load_bc_into_bc_reg) {
285 __ movl(bc_reg, bc);
286 }
287 }
288
289 if (JvmtiExport::can_post_breakpoint()) {
290 Label L_fast_patch;
291 // if a breakpoint is present we can't rewrite the stream directly
292 __ movzbl(temp_reg, at_bcp(0));
293 __ cmpl(temp_reg, Bytecodes::_breakpoint);
294 __ jcc(Assembler::notEqual, L_fast_patch);
295 __ get_method(temp_reg);
296 // Let breakpoint table handling rewrite to quicker bytecode
297 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
298 #ifndef ASSERT
299 __ jmpb(L_patch_done);
300 #else
301 __ jmp(L_patch_done);
302 #endif
303 __ bind(L_fast_patch);
304 }
305
306 #ifdef ASSERT
307 Label L_okay;
308 __ load_unsigned_byte(temp_reg, at_bcp(0));
309 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
310 __ jcc(Assembler::equal, L_okay);
311 __ cmpl(temp_reg, bc_reg);
312 __ jcc(Assembler::equal, L_okay);
313 __ stop("patching the wrong bytecode");
314 __ bind(L_okay);
315 #endif
316
317 // patch bytecode
318 __ movb(at_bcp(0), bc_reg);
319 __ bind(L_patch_done);
320 }
321 // Individual instructions
322
323
324 void TemplateTable::nop() {
325 transition(vtos, vtos);
326 // nothing to do
327 }
328
329 void TemplateTable::shouldnotreachhere() {
330 transition(vtos, vtos);
331 __ stop("shouldnotreachhere bytecode");
332 }
333
334 void TemplateTable::aconst_null() {
335 transition(vtos, atos);
336 __ xorl(rax, rax);
337 }
338
339 void TemplateTable::iconst(int value) {
340 transition(vtos, itos);
341 if (value == 0) {
342 __ xorl(rax, rax);
343 } else {
344 __ movl(rax, value);
345 }
346 }
347
348 void TemplateTable::lconst(int value) {
349 transition(vtos, ltos);
350 if (value == 0) {
351 __ xorl(rax, rax);
352 } else {
353 __ movl(rax, value);
354 }
355 #ifndef _LP64
356 assert(value >= 0, "check this code");
357 __ xorptr(rdx, rdx);
358 #endif
359 }
360
361
362
363 void TemplateTable::fconst(int value) {
364 transition(vtos, ftos);
365 if (UseSSE >= 1) {
366 static float one = 1.0f, two = 2.0f;
367 switch (value) {
368 case 0:
369 __ xorps(xmm0, xmm0);
370 break;
371 case 1:
372 __ movflt(xmm0, ExternalAddress((address) &one));
373 break;
374 case 2:
375 __ movflt(xmm0, ExternalAddress((address) &two));
376 break;
377 default:
378 ShouldNotReachHere();
379 break;
380 }
381 } else {
382 #ifdef _LP64
383 ShouldNotReachHere();
384 #else
385 if (value == 0) { __ fldz();
386 } else if (value == 1) { __ fld1();
387 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
388 } else { ShouldNotReachHere();
389 }
390 #endif // _LP64
391 }
392 }
393
394 void TemplateTable::dconst(int value) {
395 transition(vtos, dtos);
396 if (UseSSE >= 2) {
397 static double one = 1.0;
398 switch (value) {
399 case 0:
400 __ xorpd(xmm0, xmm0);
401 break;
402 case 1:
403 __ movdbl(xmm0, ExternalAddress((address) &one));
404 break;
405 default:
406 ShouldNotReachHere();
407 break;
408 }
409 } else {
410 #ifdef _LP64
411 ShouldNotReachHere();
412 #else
413 if (value == 0) { __ fldz();
414 } else if (value == 1) { __ fld1();
415 } else { ShouldNotReachHere();
416 }
417 #endif
418 }
419 }
420
421 void TemplateTable::bipush() {
422 transition(vtos, itos);
423 __ load_signed_byte(rax, at_bcp(1));
424 }
425
426 void TemplateTable::sipush() {
427 transition(vtos, itos);
428 __ load_unsigned_short(rax, at_bcp(1));
429 __ bswapl(rax);
430 __ sarl(rax, 16);
431 }
432
433 void TemplateTable::ldc(bool wide) {
434 transition(vtos, vtos);
435 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
436 Label call_ldc, notFloat, notClass, Done;
437
438 if (wide) {
439 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
440 } else {
441 __ load_unsigned_byte(rbx, at_bcp(1));
442 }
443
444 __ get_cpool_and_tags(rcx, rax);
445 const int base_offset = ConstantPool::header_size() * wordSize;
446 const int tags_offset = Array<u1>::base_offset_in_bytes();
447
448 // get type
449 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
450
451 // unresolved class - get the resolved class
452 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
453 __ jccb(Assembler::equal, call_ldc);
454
455 // unresolved class in error state - call into runtime to throw the error
456 // from the first resolution attempt
457 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
458 __ jccb(Assembler::equal, call_ldc);
459
460 // resolved class - need to call vm to get java mirror of the class
461 __ cmpl(rdx, JVM_CONSTANT_Class);
462 __ jcc(Assembler::notEqual, notClass);
463
464 __ bind(call_ldc);
465
466 __ movl(rarg, wide);
467 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
468
469 __ push(atos);
470 __ jmp(Done);
471
472 __ bind(notClass);
473 __ cmpl(rdx, JVM_CONSTANT_Float);
474 __ jccb(Assembler::notEqual, notFloat);
475
476 // ftos
477 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
478 __ push(ftos);
479 __ jmp(Done);
480
481 __ bind(notFloat);
482 #ifdef ASSERT
483 {
484 Label L;
485 __ cmpl(rdx, JVM_CONSTANT_Integer);
486 __ jcc(Assembler::equal, L);
487 // String and Object are rewritten to fast_aldc
488 __ stop("unexpected tag type in ldc");
489 __ bind(L);
490 }
491 #endif
492 // itos JVM_CONSTANT_Integer only
493 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
494 __ push(itos);
495 __ bind(Done);
496 }
497
498 // Fast path for caching oop constants.
499 void TemplateTable::fast_aldc(bool wide) {
500 transition(vtos, atos);
501
502 Register result = rax;
503 Register tmp = rdx;
504 int index_size = wide ? sizeof(u2) : sizeof(u1);
505
506 Label resolved;
507
508 // We are resolved if the resolved reference cache entry contains a
509 // non-null object (String, MethodType, etc.)
510 assert_different_registers(result, tmp);
511 __ get_cache_index_at_bcp(tmp, 1, index_size);
512 __ load_resolved_reference_at_index(result, tmp);
513 __ testl(result, result);
514 __ jcc(Assembler::notZero, resolved);
515
516 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
517
518 // first time invocation - must resolve first
519 __ movl(tmp, (int)bytecode());
520 __ call_VM(result, entry, tmp);
521
522 __ bind(resolved);
523
524 if (VerifyOops) {
525 __ verify_oop(result);
526 }
527 }
528
529 void TemplateTable::ldc2_w() {
530 transition(vtos, vtos);
531 Label Long, Done;
532 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
533
534 __ get_cpool_and_tags(rcx, rax);
535 const int base_offset = ConstantPool::header_size() * wordSize;
536 const int tags_offset = Array<u1>::base_offset_in_bytes();
537
538 // get type
539 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
540 JVM_CONSTANT_Double);
541 __ jccb(Assembler::notEqual, Long);
542
543 // dtos
544 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
545 __ push(dtos);
546
547 __ jmpb(Done);
548 __ bind(Long);
549
550 // ltos
551 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
552 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
553 __ push(ltos);
554
555 __ bind(Done);
556 }
557
558 void TemplateTable::locals_index(Register reg, int offset) {
559 __ load_unsigned_byte(reg, at_bcp(offset));
560 __ negptr(reg);
561 }
562
563 void TemplateTable::iload() {
564 iload_internal();
565 }
566
567 void TemplateTable::nofast_iload() {
568 iload_internal(may_not_rewrite);
569 }
570
571 void TemplateTable::iload_internal(RewriteControl rc) {
572 transition(vtos, itos);
573 if (RewriteFrequentPairs && rc == may_rewrite) {
574 Label rewrite, done;
575 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
576 LP64_ONLY(assert(rbx != bc, "register damaged"));
577
578 // get next byte
579 __ load_unsigned_byte(rbx,
580 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
581 // if _iload, wait to rewrite to iload2. We only want to rewrite the
582 // last two iloads in a pair. Comparing against fast_iload means that
583 // the next bytecode is neither an iload or a caload, and therefore
584 // an iload pair.
585 __ cmpl(rbx, Bytecodes::_iload);
586 __ jcc(Assembler::equal, done);
587
588 __ cmpl(rbx, Bytecodes::_fast_iload);
589 __ movl(bc, Bytecodes::_fast_iload2);
590
591 __ jccb(Assembler::equal, rewrite);
592
593 // if _caload, rewrite to fast_icaload
594 __ cmpl(rbx, Bytecodes::_caload);
595 __ movl(bc, Bytecodes::_fast_icaload);
596 __ jccb(Assembler::equal, rewrite);
597
598 // rewrite so iload doesn't check again.
599 __ movl(bc, Bytecodes::_fast_iload);
600
601 // rewrite
602 // bc: fast bytecode
603 __ bind(rewrite);
604 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
605 __ bind(done);
606 }
607
608 // Get the local value into tos
609 locals_index(rbx);
610 __ movl(rax, iaddress(rbx));
611 }
612
613 void TemplateTable::fast_iload2() {
614 transition(vtos, itos);
615 locals_index(rbx);
616 __ movl(rax, iaddress(rbx));
617 __ push(itos);
618 locals_index(rbx, 3);
619 __ movl(rax, iaddress(rbx));
620 }
621
622 void TemplateTable::fast_iload() {
623 transition(vtos, itos);
624 locals_index(rbx);
625 __ movl(rax, iaddress(rbx));
626 }
627
628 void TemplateTable::lload() {
629 transition(vtos, ltos);
630 locals_index(rbx);
631 __ movptr(rax, laddress(rbx));
632 NOT_LP64(__ movl(rdx, haddress(rbx)));
633 }
634
635 void TemplateTable::fload() {
636 transition(vtos, ftos);
637 locals_index(rbx);
638 __ load_float(faddress(rbx));
639 }
640
641 void TemplateTable::dload() {
642 transition(vtos, dtos);
643 locals_index(rbx);
644 __ load_double(daddress(rbx));
645 }
646
647 void TemplateTable::aload() {
648 transition(vtos, atos);
649 locals_index(rbx);
650 __ movptr(rax, aaddress(rbx));
651 }
652
653 void TemplateTable::locals_index_wide(Register reg) {
654 __ load_unsigned_short(reg, at_bcp(2));
655 __ bswapl(reg);
656 __ shrl(reg, 16);
657 __ negptr(reg);
658 }
659
660 void TemplateTable::wide_iload() {
661 transition(vtos, itos);
662 locals_index_wide(rbx);
663 __ movl(rax, iaddress(rbx));
664 }
665
666 void TemplateTable::wide_lload() {
667 transition(vtos, ltos);
668 locals_index_wide(rbx);
669 __ movptr(rax, laddress(rbx));
670 NOT_LP64(__ movl(rdx, haddress(rbx)));
671 }
672
673 void TemplateTable::wide_fload() {
674 transition(vtos, ftos);
675 locals_index_wide(rbx);
676 __ load_float(faddress(rbx));
677 }
678
679 void TemplateTable::wide_dload() {
680 transition(vtos, dtos);
681 locals_index_wide(rbx);
682 __ load_double(daddress(rbx));
683 }
684
685 void TemplateTable::wide_aload() {
686 transition(vtos, atos);
687 locals_index_wide(rbx);
688 __ movptr(rax, aaddress(rbx));
689 }
690
691 void TemplateTable::index_check(Register array, Register index) {
692 // Pop ptr into array
693 __ pop_ptr(array);
694 index_check_without_pop(array, index);
695 }
696
697 void TemplateTable::index_check_without_pop(Register array, Register index) {
698 // destroys rbx
699 // check array
700
701 if (ShenandoahVerifyReadsToFromSpace) {
702 oopDesc::bs()->interpreter_read_barrier(_masm, array);
703 }
704
705 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
706 // sign extend index for use by indexed load
707 __ movl2ptr(index, index);
708 // check index
709 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
710 if (index != rbx) {
711 // ??? convention: move aberrant index into rbx for exception message
712 assert(rbx != array, "different registers");
713 __ movl(rbx, index);
714 }
715 __ jump_cc(Assembler::aboveEqual,
716 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
717 }
718
719
720 void TemplateTable::iaload() {
721 transition(itos, itos);
722 // rax: index
723 // rdx: array
724 index_check(rdx, rax); // kills rbx
725 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
726 __ movl(rax, Address(rdx, rax,
727 Address::times_4,
728 arrayOopDesc::base_offset_in_bytes(T_INT)));
729 }
730
731 void TemplateTable::laload() {
732 transition(itos, ltos);
733 // rax: index
734 // rdx: array
735 index_check(rdx, rax); // kills rbx
736 NOT_LP64(__ mov(rbx, rax));
737 // rbx,: index
738 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
739 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
740 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
741 }
742
743
744
745 void TemplateTable::faload() {
746 transition(itos, ftos);
747 // rax: index
748 // rdx: array
749 index_check(rdx, rax); // kills rbx
750 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
751 __ load_float(Address(rdx, rax,
752 Address::times_4,
753 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
754 }
755
756 void TemplateTable::daload() {
757 transition(itos, dtos);
758 // rax: index
759 // rdx: array
760 index_check(rdx, rax); // kills rbx
761 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
762 __ load_double(Address(rdx, rax,
763 Address::times_8,
764 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
765 }
766
767 void TemplateTable::aaload() {
768 transition(itos, atos);
769 // rax: index
770 // rdx: array
771 index_check(rdx, rax); // kills rbx
772 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
773 __ load_heap_oop(rax, Address(rdx, rax,
774 UseCompressedOops ? Address::times_4 : Address::times_ptr,
775 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
776 }
777
778 void TemplateTable::baload() {
779 transition(itos, itos);
780 // rax: index
781 // rdx: array
782 index_check(rdx, rax); // kills rbx
783 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
784 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
785 }
786
787 void TemplateTable::caload() {
788 transition(itos, itos);
789 // rax: index
790 // rdx: array
791 index_check(rdx, rax); // kills rbx
792 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
793 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
794 }
795
796 // iload followed by caload frequent pair
797 void TemplateTable::fast_icaload() {
798 transition(vtos, itos);
799 // load index out of locals
800 locals_index(rbx);
801 __ movl(rax, iaddress(rbx));
802
803 // rax: index
804 // rdx: array
805 index_check(rdx, rax); // kills rbx
806 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
807 __ load_unsigned_short(rax,
808 Address(rdx, rax,
809 Address::times_2,
810 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
811 }
812
813
814 void TemplateTable::saload() {
815 transition(itos, itos);
816 // rax: index
817 // rdx: array
818 index_check(rdx, rax); // kills rbx
819 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rdx);
820 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
821 }
822
823 void TemplateTable::iload(int n) {
824 transition(vtos, itos);
825 __ movl(rax, iaddress(n));
826 }
827
828 void TemplateTable::lload(int n) {
829 transition(vtos, ltos);
830 __ movptr(rax, laddress(n));
831 NOT_LP64(__ movptr(rdx, haddress(n)));
832 }
833
834 void TemplateTable::fload(int n) {
835 transition(vtos, ftos);
836 __ load_float(faddress(n));
837 }
838
839 void TemplateTable::dload(int n) {
840 transition(vtos, dtos);
841 __ load_double(daddress(n));
842 }
843
844 void TemplateTable::aload(int n) {
845 transition(vtos, atos);
846 __ movptr(rax, aaddress(n));
847 }
848
849 void TemplateTable::aload_0() {
850 aload_0_internal();
851 }
852
853 void TemplateTable::nofast_aload_0() {
854 aload_0_internal(may_not_rewrite);
855 }
856
857 void TemplateTable::aload_0_internal(RewriteControl rc) {
858 transition(vtos, atos);
859 // According to bytecode histograms, the pairs:
860 //
861 // _aload_0, _fast_igetfield
862 // _aload_0, _fast_agetfield
863 // _aload_0, _fast_fgetfield
864 //
865 // occur frequently. If RewriteFrequentPairs is set, the (slow)
866 // _aload_0 bytecode checks if the next bytecode is either
867 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
868 // rewrites the current bytecode into a pair bytecode; otherwise it
869 // rewrites the current bytecode into _fast_aload_0 that doesn't do
870 // the pair check anymore.
871 //
872 // Note: If the next bytecode is _getfield, the rewrite must be
873 // delayed, otherwise we may miss an opportunity for a pair.
874 //
875 // Also rewrite frequent pairs
876 // aload_0, aload_1
877 // aload_0, iload_1
878 // These bytecodes with a small amount of code are most profitable
879 // to rewrite
880 if (RewriteFrequentPairs && rc == may_rewrite) {
881 Label rewrite, done;
882
883 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
884 LP64_ONLY(assert(rbx != bc, "register damaged"));
885
886 // get next byte
887 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
888
889 // if _getfield then wait with rewrite
890 __ cmpl(rbx, Bytecodes::_getfield);
891 __ jcc(Assembler::equal, done);
892
893 // if _igetfield then rewrite to _fast_iaccess_0
894 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
895 __ cmpl(rbx, Bytecodes::_fast_igetfield);
896 __ movl(bc, Bytecodes::_fast_iaccess_0);
897 __ jccb(Assembler::equal, rewrite);
898
899 // if _agetfield then rewrite to _fast_aaccess_0
900 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
901 __ cmpl(rbx, Bytecodes::_fast_agetfield);
902 __ movl(bc, Bytecodes::_fast_aaccess_0);
903 __ jccb(Assembler::equal, rewrite);
904
905 // if _fgetfield then rewrite to _fast_faccess_0
906 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
907 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
908 __ movl(bc, Bytecodes::_fast_faccess_0);
909 __ jccb(Assembler::equal, rewrite);
910
911 // else rewrite to _fast_aload0
912 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
913 __ movl(bc, Bytecodes::_fast_aload_0);
914
915 // rewrite
916 // bc: fast bytecode
917 __ bind(rewrite);
918 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
919
920 __ bind(done);
921 }
922
923 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
924 aload(0);
925 }
926
927 void TemplateTable::istore() {
928 transition(itos, vtos);
929 locals_index(rbx);
930 __ movl(iaddress(rbx), rax);
931 }
932
933
934 void TemplateTable::lstore() {
935 transition(ltos, vtos);
936 locals_index(rbx);
937 __ movptr(laddress(rbx), rax);
938 NOT_LP64(__ movptr(haddress(rbx), rdx));
939 }
940
941 void TemplateTable::fstore() {
942 transition(ftos, vtos);
943 locals_index(rbx);
944 __ store_float(faddress(rbx));
945 }
946
947 void TemplateTable::dstore() {
948 transition(dtos, vtos);
949 locals_index(rbx);
950 __ store_double(daddress(rbx));
951 }
952
953 void TemplateTable::astore() {
954 transition(vtos, vtos);
955 __ pop_ptr(rax);
956 locals_index(rbx);
957 __ movptr(aaddress(rbx), rax);
958 }
959
960 void TemplateTable::wide_istore() {
961 transition(vtos, vtos);
962 __ pop_i();
963 locals_index_wide(rbx);
964 __ movl(iaddress(rbx), rax);
965 }
966
967 void TemplateTable::wide_lstore() {
968 transition(vtos, vtos);
969 NOT_LP64(__ pop_l(rax, rdx));
970 LP64_ONLY(__ pop_l());
971 locals_index_wide(rbx);
972 __ movptr(laddress(rbx), rax);
973 NOT_LP64(__ movl(haddress(rbx), rdx));
974 }
975
976 void TemplateTable::wide_fstore() {
977 #ifdef _LP64
978 transition(vtos, vtos);
979 __ pop_f(xmm0);
980 locals_index_wide(rbx);
981 __ movflt(faddress(rbx), xmm0);
982 #else
983 wide_istore();
984 #endif
985 }
986
987 void TemplateTable::wide_dstore() {
988 #ifdef _LP64
989 transition(vtos, vtos);
990 __ pop_d(xmm0);
991 locals_index_wide(rbx);
992 __ movdbl(daddress(rbx), xmm0);
993 #else
994 wide_lstore();
995 #endif
996 }
997
998 void TemplateTable::wide_astore() {
999 transition(vtos, vtos);
1000 __ pop_ptr(rax);
1001 locals_index_wide(rbx);
1002 __ movptr(aaddress(rbx), rax);
1003 }
1004
1005 void TemplateTable::iastore() {
1006 transition(itos, vtos);
1007 __ pop_i(rbx);
1008 // rax: value
1009 // rbx: index
1010 // rdx: array
1011 index_check(rdx, rbx); // prefer index in rbx
1012 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1013 __ movl(Address(rdx, rbx,
1014 Address::times_4,
1015 arrayOopDesc::base_offset_in_bytes(T_INT)),
1016 rax);
1017 }
1018
1019 void TemplateTable::lastore() {
1020 transition(ltos, vtos);
1021 __ pop_i(rbx);
1022 // rax,: low(value)
1023 // rcx: array
1024 // rdx: high(value)
1025 index_check(rcx, rbx); // prefer index in rbx,
1026 // rbx,: index
1027 oopDesc::bs()->interpreter_write_barrier(_masm, rcx);
1028 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1029 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1030 }
1031
1032
1033 void TemplateTable::fastore() {
1034 transition(ftos, vtos);
1035 __ pop_i(rbx);
1036 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1037 // rbx: index
1038 // rdx: array
1039 index_check(rdx, rbx); // prefer index in rbx
1040 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1041 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1042 }
1043
1044 void TemplateTable::dastore() {
1045 transition(dtos, vtos);
1046 __ pop_i(rbx);
1047 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1048 // rbx: index
1049 // rdx: array
1050 index_check(rdx, rbx); // prefer index in rbx
1051 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1052 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1053 }
1054
1055 void TemplateTable::aastore() {
1056 Label is_null, ok_is_subtype, done;
1057 transition(vtos, vtos);
1058 // stack: ..., array, index, value
1059 __ movptr(rax, at_tos()); // value
1060 __ movl(rcx, at_tos_p1()); // index
1061 __ movptr(rdx, at_tos_p2()); // array
1062
1063 Address element_address(rdx, rcx,
1064 UseCompressedOops? Address::times_4 : Address::times_ptr,
1065 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1066
1067 index_check_without_pop(rdx, rcx); // kills rbx
1068 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1069 __ testptr(rax, rax);
1070 __ jcc(Assembler::zero, is_null);
1071
1072 // Move subklass into rbx
1073 __ load_klass(rbx, rax);
1074 // Move superklass into rax
1075 __ load_klass(rax, rdx);
1076 __ movptr(rax, Address(rax,
1077 ObjArrayKlass::element_klass_offset()));
1078 // Compress array + index*oopSize + 12 into a single register. Frees rcx.
1079 __ lea(rdx, element_address);
1080
1081 // Generate subtype check. Blows rcx, rdi
1082 // Superklass in rax. Subklass in rbx.
1083 __ gen_subtype_check(rbx, ok_is_subtype);
1084
1085 // Come here on failure
1086 // object is at TOS
1087 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1088
1089 // Come here on success
1090 __ bind(ok_is_subtype);
1091
1092 // Get the value we will store
1093 __ movptr(rax, at_tos());
1094 // Now store using the appropriate barrier
1095 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1096 __ jmp(done);
1097
1098 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1099 __ bind(is_null);
1100 __ profile_null_seen(rbx);
1101
1102 // Store a NULL
1103 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
1104
1105 // Pop stack arguments
1106 __ bind(done);
1107 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1108 }
1109
1110 void TemplateTable::bastore() {
1111 transition(itos, vtos);
1112 __ pop_i(rbx);
1113 // rax: value
1114 // rbx: index
1115 // rdx: array
1116 index_check(rdx, rbx); // prefer index in rbx
1117 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1118 // Need to check whether array is boolean or byte
1119 // since both types share the bastore bytecode.
1120 __ load_klass(rcx, rdx);
1121 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1122 int diffbit = Klass::layout_helper_boolean_diffbit();
1123 __ testl(rcx, diffbit);
1124 Label L_skip;
1125 __ jccb(Assembler::zero, L_skip);
1126 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1127 __ bind(L_skip);
1128 __ movb(Address(rdx, rbx,
1129 Address::times_1,
1130 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1131 rax);
1132 }
1133
1134 void TemplateTable::castore() {
1135 transition(itos, vtos);
1136 __ pop_i(rbx);
1137 // rax: value
1138 // rbx: index
1139 // rdx: array
1140 index_check(rdx, rbx); // prefer index in rbx
1141 oopDesc::bs()->interpreter_write_barrier(_masm, rdx);
1142 __ movw(Address(rdx, rbx,
1143 Address::times_2,
1144 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1145 rax);
1146 }
1147
1148
1149 void TemplateTable::sastore() {
1150 castore();
1151 }
1152
1153 void TemplateTable::istore(int n) {
1154 transition(itos, vtos);
1155 __ movl(iaddress(n), rax);
1156 }
1157
1158 void TemplateTable::lstore(int n) {
1159 transition(ltos, vtos);
1160 __ movptr(laddress(n), rax);
1161 NOT_LP64(__ movptr(haddress(n), rdx));
1162 }
1163
1164 void TemplateTable::fstore(int n) {
1165 transition(ftos, vtos);
1166 __ store_float(faddress(n));
1167 }
1168
1169 void TemplateTable::dstore(int n) {
1170 transition(dtos, vtos);
1171 __ store_double(daddress(n));
1172 }
1173
1174
1175 void TemplateTable::astore(int n) {
1176 transition(vtos, vtos);
1177 __ pop_ptr(rax);
1178 __ movptr(aaddress(n), rax);
1179 }
1180
1181 void TemplateTable::pop() {
1182 transition(vtos, vtos);
1183 __ addptr(rsp, Interpreter::stackElementSize);
1184 }
1185
1186 void TemplateTable::pop2() {
1187 transition(vtos, vtos);
1188 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1189 }
1190
1191
1192 void TemplateTable::dup() {
1193 transition(vtos, vtos);
1194 __ load_ptr(0, rax);
1195 __ push_ptr(rax);
1196 // stack: ..., a, a
1197 }
1198
1199 void TemplateTable::dup_x1() {
1200 transition(vtos, vtos);
1201 // stack: ..., a, b
1202 __ load_ptr( 0, rax); // load b
1203 __ load_ptr( 1, rcx); // load a
1204 __ store_ptr(1, rax); // store b
1205 __ store_ptr(0, rcx); // store a
1206 __ push_ptr(rax); // push b
1207 // stack: ..., b, a, b
1208 }
1209
1210 void TemplateTable::dup_x2() {
1211 transition(vtos, vtos);
1212 // stack: ..., a, b, c
1213 __ load_ptr( 0, rax); // load c
1214 __ load_ptr( 2, rcx); // load a
1215 __ store_ptr(2, rax); // store c in a
1216 __ push_ptr(rax); // push c
1217 // stack: ..., c, b, c, c
1218 __ load_ptr( 2, rax); // load b
1219 __ store_ptr(2, rcx); // store a in b
1220 // stack: ..., c, a, c, c
1221 __ store_ptr(1, rax); // store b in c
1222 // stack: ..., c, a, b, c
1223 }
1224
1225 void TemplateTable::dup2() {
1226 transition(vtos, vtos);
1227 // stack: ..., a, b
1228 __ load_ptr(1, rax); // load a
1229 __ push_ptr(rax); // push a
1230 __ load_ptr(1, rax); // load b
1231 __ push_ptr(rax); // push b
1232 // stack: ..., a, b, a, b
1233 }
1234
1235
1236 void TemplateTable::dup2_x1() {
1237 transition(vtos, vtos);
1238 // stack: ..., a, b, c
1239 __ load_ptr( 0, rcx); // load c
1240 __ load_ptr( 1, rax); // load b
1241 __ push_ptr(rax); // push b
1242 __ push_ptr(rcx); // push c
1243 // stack: ..., a, b, c, b, c
1244 __ store_ptr(3, rcx); // store c in b
1245 // stack: ..., a, c, c, b, c
1246 __ load_ptr( 4, rcx); // load a
1247 __ store_ptr(2, rcx); // store a in 2nd c
1248 // stack: ..., a, c, a, b, c
1249 __ store_ptr(4, rax); // store b in a
1250 // stack: ..., b, c, a, b, c
1251 }
1252
1253 void TemplateTable::dup2_x2() {
1254 transition(vtos, vtos);
1255 // stack: ..., a, b, c, d
1256 __ load_ptr( 0, rcx); // load d
1257 __ load_ptr( 1, rax); // load c
1258 __ push_ptr(rax); // push c
1259 __ push_ptr(rcx); // push d
1260 // stack: ..., a, b, c, d, c, d
1261 __ load_ptr( 4, rax); // load b
1262 __ store_ptr(2, rax); // store b in d
1263 __ store_ptr(4, rcx); // store d in b
1264 // stack: ..., a, d, c, b, c, d
1265 __ load_ptr( 5, rcx); // load a
1266 __ load_ptr( 3, rax); // load c
1267 __ store_ptr(3, rcx); // store a in c
1268 __ store_ptr(5, rax); // store c in a
1269 // stack: ..., c, d, a, b, c, d
1270 }
1271
1272 void TemplateTable::swap() {
1273 transition(vtos, vtos);
1274 // stack: ..., a, b
1275 __ load_ptr( 1, rcx); // load a
1276 __ load_ptr( 0, rax); // load b
1277 __ store_ptr(0, rcx); // store a in b
1278 __ store_ptr(1, rax); // store b in a
1279 // stack: ..., b, a
1280 }
1281
1282 void TemplateTable::iop2(Operation op) {
1283 transition(itos, itos);
1284 switch (op) {
1285 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1286 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1287 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1288 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1289 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1290 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1291 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1292 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1293 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1294 default : ShouldNotReachHere();
1295 }
1296 }
1297
1298 void TemplateTable::lop2(Operation op) {
1299 transition(ltos, ltos);
1300 #ifdef _LP64
1301 switch (op) {
1302 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1303 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1304 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1305 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1306 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1307 default : ShouldNotReachHere();
1308 }
1309 #else
1310 __ pop_l(rbx, rcx);
1311 switch (op) {
1312 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1313 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1314 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1315 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1316 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1317 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1318 default : ShouldNotReachHere();
1319 }
1320 #endif
1321 }
1322
1323 void TemplateTable::idiv() {
1324 transition(itos, itos);
1325 __ movl(rcx, rax);
1326 __ pop_i(rax);
1327 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1328 // they are not equal, one could do a normal division (no correction
1329 // needed), which may speed up this implementation for the common case.
1330 // (see also JVM spec., p.243 & p.271)
1331 __ corrected_idivl(rcx);
1332 }
1333
1334 void TemplateTable::irem() {
1335 transition(itos, itos);
1336 __ movl(rcx, rax);
1337 __ pop_i(rax);
1338 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1339 // they are not equal, one could do a normal division (no correction
1340 // needed), which may speed up this implementation for the common case.
1341 // (see also JVM spec., p.243 & p.271)
1342 __ corrected_idivl(rcx);
1343 __ movl(rax, rdx);
1344 }
1345
1346 void TemplateTable::lmul() {
1347 transition(ltos, ltos);
1348 #ifdef _LP64
1349 __ pop_l(rdx);
1350 __ imulq(rax, rdx);
1351 #else
1352 __ pop_l(rbx, rcx);
1353 __ push(rcx); __ push(rbx);
1354 __ push(rdx); __ push(rax);
1355 __ lmul(2 * wordSize, 0);
1356 __ addptr(rsp, 4 * wordSize); // take off temporaries
1357 #endif
1358 }
1359
1360 void TemplateTable::ldiv() {
1361 transition(ltos, ltos);
1362 #ifdef _LP64
1363 __ mov(rcx, rax);
1364 __ pop_l(rax);
1365 // generate explicit div0 check
1366 __ testq(rcx, rcx);
1367 __ jump_cc(Assembler::zero,
1368 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1369 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1370 // they are not equal, one could do a normal division (no correction
1371 // needed), which may speed up this implementation for the common case.
1372 // (see also JVM spec., p.243 & p.271)
1373 __ corrected_idivq(rcx); // kills rbx
1374 #else
1375 __ pop_l(rbx, rcx);
1376 __ push(rcx); __ push(rbx);
1377 __ push(rdx); __ push(rax);
1378 // check if y = 0
1379 __ orl(rax, rdx);
1380 __ jump_cc(Assembler::zero,
1381 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1382 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1383 __ addptr(rsp, 4 * wordSize); // take off temporaries
1384 #endif
1385 }
1386
1387 void TemplateTable::lrem() {
1388 transition(ltos, ltos);
1389 #ifdef _LP64
1390 __ mov(rcx, rax);
1391 __ pop_l(rax);
1392 __ testq(rcx, rcx);
1393 __ jump_cc(Assembler::zero,
1394 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1395 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1396 // they are not equal, one could do a normal division (no correction
1397 // needed), which may speed up this implementation for the common case.
1398 // (see also JVM spec., p.243 & p.271)
1399 __ corrected_idivq(rcx); // kills rbx
1400 __ mov(rax, rdx);
1401 #else
1402 __ pop_l(rbx, rcx);
1403 __ push(rcx); __ push(rbx);
1404 __ push(rdx); __ push(rax);
1405 // check if y = 0
1406 __ orl(rax, rdx);
1407 __ jump_cc(Assembler::zero,
1408 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1409 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1410 __ addptr(rsp, 4 * wordSize);
1411 #endif
1412 }
1413
1414 void TemplateTable::lshl() {
1415 transition(itos, ltos);
1416 __ movl(rcx, rax); // get shift count
1417 #ifdef _LP64
1418 __ pop_l(rax); // get shift value
1419 __ shlq(rax);
1420 #else
1421 __ pop_l(rax, rdx); // get shift value
1422 __ lshl(rdx, rax);
1423 #endif
1424 }
1425
1426 void TemplateTable::lshr() {
1427 #ifdef _LP64
1428 transition(itos, ltos);
1429 __ movl(rcx, rax); // get shift count
1430 __ pop_l(rax); // get shift value
1431 __ sarq(rax);
1432 #else
1433 transition(itos, ltos);
1434 __ mov(rcx, rax); // get shift count
1435 __ pop_l(rax, rdx); // get shift value
1436 __ lshr(rdx, rax, true);
1437 #endif
1438 }
1439
1440 void TemplateTable::lushr() {
1441 transition(itos, ltos);
1442 #ifdef _LP64
1443 __ movl(rcx, rax); // get shift count
1444 __ pop_l(rax); // get shift value
1445 __ shrq(rax);
1446 #else
1447 __ mov(rcx, rax); // get shift count
1448 __ pop_l(rax, rdx); // get shift value
1449 __ lshr(rdx, rax);
1450 #endif
1451 }
1452
1453 void TemplateTable::fop2(Operation op) {
1454 transition(ftos, ftos);
1455
1456 if (UseSSE >= 1) {
1457 switch (op) {
1458 case add:
1459 __ addss(xmm0, at_rsp());
1460 __ addptr(rsp, Interpreter::stackElementSize);
1461 break;
1462 case sub:
1463 __ movflt(xmm1, xmm0);
1464 __ pop_f(xmm0);
1465 __ subss(xmm0, xmm1);
1466 break;
1467 case mul:
1468 __ mulss(xmm0, at_rsp());
1469 __ addptr(rsp, Interpreter::stackElementSize);
1470 break;
1471 case div:
1472 __ movflt(xmm1, xmm0);
1473 __ pop_f(xmm0);
1474 __ divss(xmm0, xmm1);
1475 break;
1476 case rem:
1477 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1478 // modulo operation. The frem method calls the function
1479 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1480 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1481 // (signalling or quiet) is returned.
1482 //
1483 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1484 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1485 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1486 // The fprem instruction used on x86_32 is functionally equivalent to
1487 // SharedRuntime::frem in that it returns a NaN.
1488 #ifdef _LP64
1489 __ movflt(xmm1, xmm0);
1490 __ pop_f(xmm0);
1491 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1492 #else
1493 __ push_f(xmm0);
1494 __ pop_f();
1495 __ fld_s(at_rsp());
1496 __ fremr(rax);
1497 __ f2ieee();
1498 __ pop(rax); // pop second operand off the stack
1499 __ push_f();
1500 __ pop_f(xmm0);
1501 #endif
1502 break;
1503 default:
1504 ShouldNotReachHere();
1505 break;
1506 }
1507 } else {
1508 #ifdef _LP64
1509 ShouldNotReachHere();
1510 #else
1511 switch (op) {
1512 case add: __ fadd_s (at_rsp()); break;
1513 case sub: __ fsubr_s(at_rsp()); break;
1514 case mul: __ fmul_s (at_rsp()); break;
1515 case div: __ fdivr_s(at_rsp()); break;
1516 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1517 default : ShouldNotReachHere();
1518 }
1519 __ f2ieee();
1520 __ pop(rax); // pop second operand off the stack
1521 #endif // _LP64
1522 }
1523 }
1524
1525 void TemplateTable::dop2(Operation op) {
1526 transition(dtos, dtos);
1527 if (UseSSE >= 2) {
1528 switch (op) {
1529 case add:
1530 __ addsd(xmm0, at_rsp());
1531 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1532 break;
1533 case sub:
1534 __ movdbl(xmm1, xmm0);
1535 __ pop_d(xmm0);
1536 __ subsd(xmm0, xmm1);
1537 break;
1538 case mul:
1539 __ mulsd(xmm0, at_rsp());
1540 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1541 break;
1542 case div:
1543 __ movdbl(xmm1, xmm0);
1544 __ pop_d(xmm0);
1545 __ divsd(xmm0, xmm1);
1546 break;
1547 case rem:
1548 // Similar to fop2(), the modulo operation is performed using the
1549 // SharedRuntime::drem method (on x86_64 platforms) or using the
1550 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1551 #ifdef _LP64
1552 __ movdbl(xmm1, xmm0);
1553 __ pop_d(xmm0);
1554 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1555 #else
1556 __ push_d(xmm0);
1557 __ pop_d();
1558 __ fld_d(at_rsp());
1559 __ fremr(rax);
1560 __ d2ieee();
1561 __ pop(rax);
1562 __ pop(rdx);
1563 __ push_d();
1564 __ pop_d(xmm0);
1565 #endif
1566 break;
1567 default:
1568 ShouldNotReachHere();
1569 break;
1570 }
1571 } else {
1572 #ifdef _LP64
1573 ShouldNotReachHere();
1574 #else
1575 switch (op) {
1576 case add: __ fadd_d (at_rsp()); break;
1577 case sub: __ fsubr_d(at_rsp()); break;
1578 case mul: {
1579 Label L_strict;
1580 Label L_join;
1581 const Address access_flags (rcx, Method::access_flags_offset());
1582 __ get_method(rcx);
1583 __ movl(rcx, access_flags);
1584 __ testl(rcx, JVM_ACC_STRICT);
1585 __ jccb(Assembler::notZero, L_strict);
1586 __ fmul_d (at_rsp());
1587 __ jmpb(L_join);
1588 __ bind(L_strict);
1589 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1590 __ fmulp();
1591 __ fmul_d (at_rsp());
1592 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1593 __ fmulp();
1594 __ bind(L_join);
1595 break;
1596 }
1597 case div: {
1598 Label L_strict;
1599 Label L_join;
1600 const Address access_flags (rcx, Method::access_flags_offset());
1601 __ get_method(rcx);
1602 __ movl(rcx, access_flags);
1603 __ testl(rcx, JVM_ACC_STRICT);
1604 __ jccb(Assembler::notZero, L_strict);
1605 __ fdivr_d(at_rsp());
1606 __ jmp(L_join);
1607 __ bind(L_strict);
1608 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1609 __ fmul_d (at_rsp());
1610 __ fdivrp();
1611 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1612 __ fmulp();
1613 __ bind(L_join);
1614 break;
1615 }
1616 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1617 default : ShouldNotReachHere();
1618 }
1619 __ d2ieee();
1620 // Pop double precision number from rsp.
1621 __ pop(rax);
1622 __ pop(rdx);
1623 #endif
1624 }
1625 }
1626
1627 void TemplateTable::ineg() {
1628 transition(itos, itos);
1629 __ negl(rax);
1630 }
1631
1632 void TemplateTable::lneg() {
1633 transition(ltos, ltos);
1634 LP64_ONLY(__ negq(rax));
1635 NOT_LP64(__ lneg(rdx, rax));
1636 }
1637
1638 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1639 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1640 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1641 // of 128-bits operands for SSE instructions.
1642 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1643 // Store the value to a 128-bits operand.
1644 operand[0] = lo;
1645 operand[1] = hi;
1646 return operand;
1647 }
1648
1649 // Buffer for 128-bits masks used by SSE instructions.
1650 static jlong float_signflip_pool[2*2];
1651 static jlong double_signflip_pool[2*2];
1652
1653 void TemplateTable::fneg() {
1654 transition(ftos, ftos);
1655 if (UseSSE >= 1) {
1656 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1657 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1658 } else {
1659 LP64_ONLY(ShouldNotReachHere());
1660 NOT_LP64(__ fchs());
1661 }
1662 }
1663
1664 void TemplateTable::dneg() {
1665 transition(dtos, dtos);
1666 if (UseSSE >= 2) {
1667 static jlong *double_signflip =
1668 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1669 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1670 } else {
1671 #ifdef _LP64
1672 ShouldNotReachHere();
1673 #else
1674 __ fchs();
1675 #endif
1676 }
1677 }
1678
1679 void TemplateTable::iinc() {
1680 transition(vtos, vtos);
1681 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1682 locals_index(rbx);
1683 __ addl(iaddress(rbx), rdx);
1684 }
1685
1686 void TemplateTable::wide_iinc() {
1687 transition(vtos, vtos);
1688 __ movl(rdx, at_bcp(4)); // get constant
1689 locals_index_wide(rbx);
1690 __ bswapl(rdx); // swap bytes & sign-extend constant
1691 __ sarl(rdx, 16);
1692 __ addl(iaddress(rbx), rdx);
1693 // Note: should probably use only one movl to get both
1694 // the index and the constant -> fix this
1695 }
1696
1697 void TemplateTable::convert() {
1698 #ifdef _LP64
1699 // Checking
1700 #ifdef ASSERT
1701 {
1702 TosState tos_in = ilgl;
1703 TosState tos_out = ilgl;
1704 switch (bytecode()) {
1705 case Bytecodes::_i2l: // fall through
1706 case Bytecodes::_i2f: // fall through
1707 case Bytecodes::_i2d: // fall through
1708 case Bytecodes::_i2b: // fall through
1709 case Bytecodes::_i2c: // fall through
1710 case Bytecodes::_i2s: tos_in = itos; break;
1711 case Bytecodes::_l2i: // fall through
1712 case Bytecodes::_l2f: // fall through
1713 case Bytecodes::_l2d: tos_in = ltos; break;
1714 case Bytecodes::_f2i: // fall through
1715 case Bytecodes::_f2l: // fall through
1716 case Bytecodes::_f2d: tos_in = ftos; break;
1717 case Bytecodes::_d2i: // fall through
1718 case Bytecodes::_d2l: // fall through
1719 case Bytecodes::_d2f: tos_in = dtos; break;
1720 default : ShouldNotReachHere();
1721 }
1722 switch (bytecode()) {
1723 case Bytecodes::_l2i: // fall through
1724 case Bytecodes::_f2i: // fall through
1725 case Bytecodes::_d2i: // fall through
1726 case Bytecodes::_i2b: // fall through
1727 case Bytecodes::_i2c: // fall through
1728 case Bytecodes::_i2s: tos_out = itos; break;
1729 case Bytecodes::_i2l: // fall through
1730 case Bytecodes::_f2l: // fall through
1731 case Bytecodes::_d2l: tos_out = ltos; break;
1732 case Bytecodes::_i2f: // fall through
1733 case Bytecodes::_l2f: // fall through
1734 case Bytecodes::_d2f: tos_out = ftos; break;
1735 case Bytecodes::_i2d: // fall through
1736 case Bytecodes::_l2d: // fall through
1737 case Bytecodes::_f2d: tos_out = dtos; break;
1738 default : ShouldNotReachHere();
1739 }
1740 transition(tos_in, tos_out);
1741 }
1742 #endif // ASSERT
1743
1744 static const int64_t is_nan = 0x8000000000000000L;
1745
1746 // Conversion
1747 switch (bytecode()) {
1748 case Bytecodes::_i2l:
1749 __ movslq(rax, rax);
1750 break;
1751 case Bytecodes::_i2f:
1752 __ cvtsi2ssl(xmm0, rax);
1753 break;
1754 case Bytecodes::_i2d:
1755 __ cvtsi2sdl(xmm0, rax);
1756 break;
1757 case Bytecodes::_i2b:
1758 __ movsbl(rax, rax);
1759 break;
1760 case Bytecodes::_i2c:
1761 __ movzwl(rax, rax);
1762 break;
1763 case Bytecodes::_i2s:
1764 __ movswl(rax, rax);
1765 break;
1766 case Bytecodes::_l2i:
1767 __ movl(rax, rax);
1768 break;
1769 case Bytecodes::_l2f:
1770 __ cvtsi2ssq(xmm0, rax);
1771 break;
1772 case Bytecodes::_l2d:
1773 __ cvtsi2sdq(xmm0, rax);
1774 break;
1775 case Bytecodes::_f2i:
1776 {
1777 Label L;
1778 __ cvttss2sil(rax, xmm0);
1779 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1780 __ jcc(Assembler::notEqual, L);
1781 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1782 __ bind(L);
1783 }
1784 break;
1785 case Bytecodes::_f2l:
1786 {
1787 Label L;
1788 __ cvttss2siq(rax, xmm0);
1789 // NaN or overflow/underflow?
1790 __ cmp64(rax, ExternalAddress((address) &is_nan));
1791 __ jcc(Assembler::notEqual, L);
1792 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1793 __ bind(L);
1794 }
1795 break;
1796 case Bytecodes::_f2d:
1797 __ cvtss2sd(xmm0, xmm0);
1798 break;
1799 case Bytecodes::_d2i:
1800 {
1801 Label L;
1802 __ cvttsd2sil(rax, xmm0);
1803 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1804 __ jcc(Assembler::notEqual, L);
1805 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1806 __ bind(L);
1807 }
1808 break;
1809 case Bytecodes::_d2l:
1810 {
1811 Label L;
1812 __ cvttsd2siq(rax, xmm0);
1813 // NaN or overflow/underflow?
1814 __ cmp64(rax, ExternalAddress((address) &is_nan));
1815 __ jcc(Assembler::notEqual, L);
1816 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1817 __ bind(L);
1818 }
1819 break;
1820 case Bytecodes::_d2f:
1821 __ cvtsd2ss(xmm0, xmm0);
1822 break;
1823 default:
1824 ShouldNotReachHere();
1825 }
1826 #else
1827 // Checking
1828 #ifdef ASSERT
1829 { TosState tos_in = ilgl;
1830 TosState tos_out = ilgl;
1831 switch (bytecode()) {
1832 case Bytecodes::_i2l: // fall through
1833 case Bytecodes::_i2f: // fall through
1834 case Bytecodes::_i2d: // fall through
1835 case Bytecodes::_i2b: // fall through
1836 case Bytecodes::_i2c: // fall through
1837 case Bytecodes::_i2s: tos_in = itos; break;
1838 case Bytecodes::_l2i: // fall through
1839 case Bytecodes::_l2f: // fall through
1840 case Bytecodes::_l2d: tos_in = ltos; break;
1841 case Bytecodes::_f2i: // fall through
1842 case Bytecodes::_f2l: // fall through
1843 case Bytecodes::_f2d: tos_in = ftos; break;
1844 case Bytecodes::_d2i: // fall through
1845 case Bytecodes::_d2l: // fall through
1846 case Bytecodes::_d2f: tos_in = dtos; break;
1847 default : ShouldNotReachHere();
1848 }
1849 switch (bytecode()) {
1850 case Bytecodes::_l2i: // fall through
1851 case Bytecodes::_f2i: // fall through
1852 case Bytecodes::_d2i: // fall through
1853 case Bytecodes::_i2b: // fall through
1854 case Bytecodes::_i2c: // fall through
1855 case Bytecodes::_i2s: tos_out = itos; break;
1856 case Bytecodes::_i2l: // fall through
1857 case Bytecodes::_f2l: // fall through
1858 case Bytecodes::_d2l: tos_out = ltos; break;
1859 case Bytecodes::_i2f: // fall through
1860 case Bytecodes::_l2f: // fall through
1861 case Bytecodes::_d2f: tos_out = ftos; break;
1862 case Bytecodes::_i2d: // fall through
1863 case Bytecodes::_l2d: // fall through
1864 case Bytecodes::_f2d: tos_out = dtos; break;
1865 default : ShouldNotReachHere();
1866 }
1867 transition(tos_in, tos_out);
1868 }
1869 #endif // ASSERT
1870
1871 // Conversion
1872 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1873 switch (bytecode()) {
1874 case Bytecodes::_i2l:
1875 __ extend_sign(rdx, rax);
1876 break;
1877 case Bytecodes::_i2f:
1878 if (UseSSE >= 1) {
1879 __ cvtsi2ssl(xmm0, rax);
1880 } else {
1881 __ push(rax); // store int on tos
1882 __ fild_s(at_rsp()); // load int to ST0
1883 __ f2ieee(); // truncate to float size
1884 __ pop(rcx); // adjust rsp
1885 }
1886 break;
1887 case Bytecodes::_i2d:
1888 if (UseSSE >= 2) {
1889 __ cvtsi2sdl(xmm0, rax);
1890 } else {
1891 __ push(rax); // add one slot for d2ieee()
1892 __ push(rax); // store int on tos
1893 __ fild_s(at_rsp()); // load int to ST0
1894 __ d2ieee(); // truncate to double size
1895 __ pop(rcx); // adjust rsp
1896 __ pop(rcx);
1897 }
1898 break;
1899 case Bytecodes::_i2b:
1900 __ shll(rax, 24); // truncate upper 24 bits
1901 __ sarl(rax, 24); // and sign-extend byte
1902 LP64_ONLY(__ movsbl(rax, rax));
1903 break;
1904 case Bytecodes::_i2c:
1905 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1906 LP64_ONLY(__ movzwl(rax, rax));
1907 break;
1908 case Bytecodes::_i2s:
1909 __ shll(rax, 16); // truncate upper 16 bits
1910 __ sarl(rax, 16); // and sign-extend short
1911 LP64_ONLY(__ movswl(rax, rax));
1912 break;
1913 case Bytecodes::_l2i:
1914 /* nothing to do */
1915 break;
1916 case Bytecodes::_l2f:
1917 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1918 // 64-bit long values to floats. On 32-bit platforms it is not possible
1919 // to use that instruction with 64-bit operands, therefore the FPU is
1920 // used to perform the conversion.
1921 __ push(rdx); // store long on tos
1922 __ push(rax);
1923 __ fild_d(at_rsp()); // load long to ST0
1924 __ f2ieee(); // truncate to float size
1925 __ pop(rcx); // adjust rsp
1926 __ pop(rcx);
1927 if (UseSSE >= 1) {
1928 __ push_f();
1929 __ pop_f(xmm0);
1930 }
1931 break;
1932 case Bytecodes::_l2d:
1933 // On 32-bit platforms the FPU is used for conversion because on
1934 // 32-bit platforms it is not not possible to use the cvtsi2sdq
1935 // instruction with 64-bit operands.
1936 __ push(rdx); // store long on tos
1937 __ push(rax);
1938 __ fild_d(at_rsp()); // load long to ST0
1939 __ d2ieee(); // truncate to double size
1940 __ pop(rcx); // adjust rsp
1941 __ pop(rcx);
1942 if (UseSSE >= 2) {
1943 __ push_d();
1944 __ pop_d(xmm0);
1945 }
1946 break;
1947 case Bytecodes::_f2i:
1948 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1949 // as it returns 0 for any NaN.
1950 if (UseSSE >= 1) {
1951 __ push_f(xmm0);
1952 } else {
1953 __ push(rcx); // reserve space for argument
1954 __ fstp_s(at_rsp()); // pass float argument on stack
1955 }
1956 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1957 break;
1958 case Bytecodes::_f2l:
1959 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
1960 // as it returns 0 for any NaN.
1961 if (UseSSE >= 1) {
1962 __ push_f(xmm0);
1963 } else {
1964 __ push(rcx); // reserve space for argument
1965 __ fstp_s(at_rsp()); // pass float argument on stack
1966 }
1967 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1968 break;
1969 case Bytecodes::_f2d:
1970 if (UseSSE < 1) {
1971 /* nothing to do */
1972 } else if (UseSSE == 1) {
1973 __ push_f(xmm0);
1974 __ pop_f();
1975 } else { // UseSSE >= 2
1976 __ cvtss2sd(xmm0, xmm0);
1977 }
1978 break;
1979 case Bytecodes::_d2i:
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::d2i), 2);
1988 break;
1989 case Bytecodes::_d2l:
1990 if (UseSSE >= 2) {
1991 __ push_d(xmm0);
1992 } else {
1993 __ push(rcx); // reserve space for argument
1994 __ push(rcx);
1995 __ fstp_d(at_rsp()); // pass double argument on stack
1996 }
1997 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1998 break;
1999 case Bytecodes::_d2f:
2000 if (UseSSE <= 1) {
2001 __ push(rcx); // reserve space for f2ieee()
2002 __ f2ieee(); // truncate to float size
2003 __ pop(rcx); // adjust rsp
2004 if (UseSSE == 1) {
2005 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2006 // the conversion is performed using the FPU in this case.
2007 __ push_f();
2008 __ pop_f(xmm0);
2009 }
2010 } else { // UseSSE >= 2
2011 __ cvtsd2ss(xmm0, xmm0);
2012 }
2013 break;
2014 default :
2015 ShouldNotReachHere();
2016 }
2017 #endif
2018 }
2019
2020 void TemplateTable::lcmp() {
2021 transition(ltos, itos);
2022 #ifdef _LP64
2023 Label done;
2024 __ pop_l(rdx);
2025 __ cmpq(rdx, rax);
2026 __ movl(rax, -1);
2027 __ jccb(Assembler::less, done);
2028 __ setb(Assembler::notEqual, rax);
2029 __ movzbl(rax, rax);
2030 __ bind(done);
2031 #else
2032
2033 // y = rdx:rax
2034 __ pop_l(rbx, rcx); // get x = rcx:rbx
2035 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2036 __ mov(rax, rcx);
2037 #endif
2038 }
2039
2040 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2041 if ((is_float && UseSSE >= 1) ||
2042 (!is_float && UseSSE >= 2)) {
2043 Label done;
2044 if (is_float) {
2045 // XXX get rid of pop here, use ... reg, mem32
2046 __ pop_f(xmm1);
2047 __ ucomiss(xmm1, xmm0);
2048 } else {
2049 // XXX get rid of pop here, use ... reg, mem64
2050 __ pop_d(xmm1);
2051 __ ucomisd(xmm1, xmm0);
2052 }
2053 if (unordered_result < 0) {
2054 __ movl(rax, -1);
2055 __ jccb(Assembler::parity, done);
2056 __ jccb(Assembler::below, done);
2057 __ setb(Assembler::notEqual, rdx);
2058 __ movzbl(rax, rdx);
2059 } else {
2060 __ movl(rax, 1);
2061 __ jccb(Assembler::parity, done);
2062 __ jccb(Assembler::above, done);
2063 __ movl(rax, 0);
2064 __ jccb(Assembler::equal, done);
2065 __ decrementl(rax);
2066 }
2067 __ bind(done);
2068 } else {
2069 #ifdef _LP64
2070 ShouldNotReachHere();
2071 #else
2072 if (is_float) {
2073 __ fld_s(at_rsp());
2074 } else {
2075 __ fld_d(at_rsp());
2076 __ pop(rdx);
2077 }
2078 __ pop(rcx);
2079 __ fcmp2int(rax, unordered_result < 0);
2080 #endif // _LP64
2081 }
2082 }
2083
2084 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2085 __ get_method(rcx); // rcx holds method
2086 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2087 // holds bumped taken count
2088
2089 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2090 InvocationCounter::counter_offset();
2091 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2092 InvocationCounter::counter_offset();
2093
2094 // Load up edx with the branch displacement
2095 if (is_wide) {
2096 __ movl(rdx, at_bcp(1));
2097 } else {
2098 __ load_signed_short(rdx, at_bcp(1));
2099 }
2100 __ bswapl(rdx);
2101
2102 if (!is_wide) {
2103 __ sarl(rdx, 16);
2104 }
2105 LP64_ONLY(__ movl2ptr(rdx, rdx));
2106
2107 // Handle all the JSR stuff here, then exit.
2108 // It's much shorter and cleaner than intermingling with the non-JSR
2109 // normal-branch stuff occurring below.
2110 if (is_jsr) {
2111 // Pre-load the next target bytecode into rbx
2112 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2113
2114 // compute return address as bci in rax
2115 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2116 in_bytes(ConstMethod::codes_offset())));
2117 __ subptr(rax, Address(rcx, Method::const_offset()));
2118 // Adjust the bcp in r13 by the displacement in rdx
2119 __ addptr(rbcp, rdx);
2120 // jsr returns atos that is not an oop
2121 __ push_i(rax);
2122 __ dispatch_only(vtos);
2123 return;
2124 }
2125
2126 // Normal (non-jsr) branch handling
2127
2128 // Adjust the bcp in r13 by the displacement in rdx
2129 __ addptr(rbcp, rdx);
2130
2131 assert(UseLoopCounter || !UseOnStackReplacement,
2132 "on-stack-replacement requires loop counters");
2133 Label backedge_counter_overflow;
2134 Label profile_method;
2135 Label dispatch;
2136 if (UseLoopCounter) {
2137 // increment backedge counter for backward branches
2138 // rax: MDO
2139 // rbx: MDO bumped taken-count
2140 // rcx: method
2141 // rdx: target offset
2142 // r13: target bcp
2143 // r14: locals pointer
2144 __ testl(rdx, rdx); // check if forward or backward branch
2145 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2146
2147 // check if MethodCounters exists
2148 Label has_counters;
2149 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2150 __ testptr(rax, rax);
2151 __ jcc(Assembler::notZero, has_counters);
2152 __ push(rdx);
2153 __ push(rcx);
2154 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2155 rcx);
2156 __ pop(rcx);
2157 __ pop(rdx);
2158 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2159 __ testptr(rax, rax);
2160 __ jcc(Assembler::zero, dispatch);
2161 __ bind(has_counters);
2162
2163 if (TieredCompilation) {
2164 Label no_mdo;
2165 int increment = InvocationCounter::count_increment;
2166 if (ProfileInterpreter) {
2167 // Are we profiling?
2168 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2169 __ testptr(rbx, rbx);
2170 __ jccb(Assembler::zero, no_mdo);
2171 // Increment the MDO backedge counter
2172 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2173 in_bytes(InvocationCounter::counter_offset()));
2174 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2175 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2176 rax, false, Assembler::zero, &backedge_counter_overflow);
2177 __ jmp(dispatch);
2178 }
2179 __ bind(no_mdo);
2180 // Increment backedge counter in MethodCounters*
2181 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2182 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2183 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2184 rax, false, Assembler::zero, &backedge_counter_overflow);
2185 } else { // not TieredCompilation
2186 // increment counter
2187 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2188 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2189 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2190 __ movl(Address(rcx, be_offset), rax); // store counter
2191
2192 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2193
2194 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2195 __ addl(rax, Address(rcx, be_offset)); // add both counters
2196
2197 if (ProfileInterpreter) {
2198 // Test to see if we should create a method data oop
2199 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2200 __ jcc(Assembler::less, dispatch);
2201
2202 // if no method data exists, go to profile method
2203 __ test_method_data_pointer(rax, profile_method);
2204
2205 if (UseOnStackReplacement) {
2206 // check for overflow against rbx which is the MDO taken count
2207 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2208 __ jcc(Assembler::below, dispatch);
2209
2210 // When ProfileInterpreter is on, the backedge_count comes
2211 // from the MethodData*, which value does not get reset on
2212 // the call to frequency_counter_overflow(). To avoid
2213 // excessive calls to the overflow routine while the method is
2214 // being compiled, add a second test to make sure the overflow
2215 // function is called only once every overflow_frequency.
2216 const int overflow_frequency = 1024;
2217 __ andl(rbx, overflow_frequency - 1);
2218 __ jcc(Assembler::zero, backedge_counter_overflow);
2219
2220 }
2221 } else {
2222 if (UseOnStackReplacement) {
2223 // check for overflow against rax, which is the sum of the
2224 // counters
2225 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2226 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2227
2228 }
2229 }
2230 }
2231 __ bind(dispatch);
2232 }
2233
2234 // Pre-load the next target bytecode into rbx
2235 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2236
2237 // continue with the bytecode @ target
2238 // rax: return bci for jsr's, unused otherwise
2239 // rbx: target bytecode
2240 // r13: target bcp
2241 __ dispatch_only(vtos);
2242
2243 if (UseLoopCounter) {
2244 if (ProfileInterpreter) {
2245 // Out-of-line code to allocate method data oop.
2246 __ bind(profile_method);
2247 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2248 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2249 __ set_method_data_pointer_for_bcp();
2250 __ jmp(dispatch);
2251 }
2252
2253 if (UseOnStackReplacement) {
2254 // invocation counter overflow
2255 __ bind(backedge_counter_overflow);
2256 __ negptr(rdx);
2257 __ addptr(rdx, rbcp); // branch bcp
2258 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2259 __ call_VM(noreg,
2260 CAST_FROM_FN_PTR(address,
2261 InterpreterRuntime::frequency_counter_overflow),
2262 rdx);
2263 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2264
2265 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2266 // rbx: target bytecode
2267 // rdx: scratch
2268 // r14: locals pointer
2269 // r13: bcp
2270 __ testptr(rax, rax); // test result
2271 __ jcc(Assembler::zero, dispatch); // no osr if null
2272 // nmethod may have been invalidated (VM may block upon call_VM return)
2273 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2274 __ jcc(Assembler::notEqual, dispatch);
2275
2276 // We have the address of an on stack replacement routine in rax
2277 // We need to prepare to execute the OSR method. First we must
2278 // migrate the locals and monitors off of the stack.
2279
2280 LP64_ONLY(__ mov(r13, rax)); // save the nmethod
2281 NOT_LP64(__ mov(rbx, rax)); // save the nmethod
2282 NOT_LP64(__ get_thread(rcx));
2283
2284 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2285
2286 // rax is OSR buffer, move it to expected parameter location
2287 LP64_ONLY(__ mov(j_rarg0, rax));
2288 NOT_LP64(__ mov(rcx, rax));
2289 // We use j_rarg definitions here so that registers don't conflict as parameter
2290 // registers change across platforms as we are in the midst of a calling
2291 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2292
2293 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2294 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2295
2296
2297 // pop the interpreter frame
2298 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2299 __ leave(); // remove frame anchor
2300 __ pop(retaddr); // get return address
2301 __ mov(rsp, sender_sp); // set sp to sender sp
2302 // Ensure compiled code always sees stack at proper alignment
2303 __ andptr(rsp, -(StackAlignmentInBytes));
2304
2305 // unlike x86 we need no specialized return from compiled code
2306 // to the interpreter or the call stub.
2307
2308 // push the return address
2309 __ push(retaddr);
2310
2311 // and begin the OSR nmethod
2312 LP64_ONLY(__ jmp(Address(r13, nmethod::osr_entry_point_offset())));
2313 NOT_LP64(__ jmp(Address(rbx, nmethod::osr_entry_point_offset())));
2314 }
2315 }
2316 }
2317
2318 void TemplateTable::if_0cmp(Condition cc) {
2319 transition(itos, vtos);
2320 // assume branch is more often taken than not (loops use backward branches)
2321 Label not_taken;
2322 __ testl(rax, rax);
2323 __ jcc(j_not(cc), not_taken);
2324 branch(false, false);
2325 __ bind(not_taken);
2326 __ profile_not_taken_branch(rax);
2327 }
2328
2329 void TemplateTable::if_icmp(Condition cc) {
2330 transition(itos, vtos);
2331 // assume branch is more often taken than not (loops use backward branches)
2332 Label not_taken;
2333 __ pop_i(rdx);
2334 __ cmpl(rdx, rax);
2335 __ jcc(j_not(cc), not_taken);
2336 branch(false, false);
2337 __ bind(not_taken);
2338 __ profile_not_taken_branch(rax);
2339 }
2340
2341 void TemplateTable::if_nullcmp(Condition cc) {
2342 transition(atos, vtos);
2343 // assume branch is more often taken than not (loops use backward branches)
2344 Label not_taken;
2345 __ testptr(rax, rax);
2346 __ jcc(j_not(cc), not_taken);
2347 branch(false, false);
2348 __ bind(not_taken);
2349 __ profile_not_taken_branch(rax);
2350 }
2351
2352 void TemplateTable::if_acmp(Condition cc) {
2353 transition(atos, vtos);
2354 // assume branch is more often taken than not (loops use backward branches)
2355 Label not_taken;
2356 __ pop_ptr(rdx);
2357 __ cmpptr(rdx, rax);
2358 oopDesc::bs()->asm_acmp_barrier(_masm, rdx, rax);
2359 __ jcc(j_not(cc), not_taken);
2360 branch(false, false);
2361 __ bind(not_taken);
2362 __ profile_not_taken_branch(rax);
2363 }
2364
2365 void TemplateTable::ret() {
2366 transition(vtos, vtos);
2367 locals_index(rbx);
2368 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2369 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2370 __ profile_ret(rbx, rcx);
2371 __ get_method(rax);
2372 __ movptr(rbcp, Address(rax, Method::const_offset()));
2373 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2374 ConstMethod::codes_offset()));
2375 __ dispatch_next(vtos);
2376 }
2377
2378 void TemplateTable::wide_ret() {
2379 transition(vtos, vtos);
2380 locals_index_wide(rbx);
2381 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2382 __ profile_ret(rbx, rcx);
2383 __ get_method(rax);
2384 __ movptr(rbcp, Address(rax, Method::const_offset()));
2385 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2386 __ dispatch_next(vtos);
2387 }
2388
2389 void TemplateTable::tableswitch() {
2390 Label default_case, continue_execution;
2391 transition(itos, vtos);
2392
2393 // align r13/rsi
2394 __ lea(rbx, at_bcp(BytesPerInt));
2395 __ andptr(rbx, -BytesPerInt);
2396 // load lo & hi
2397 __ movl(rcx, Address(rbx, BytesPerInt));
2398 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2399 __ bswapl(rcx);
2400 __ bswapl(rdx);
2401 // check against lo & hi
2402 __ cmpl(rax, rcx);
2403 __ jcc(Assembler::less, default_case);
2404 __ cmpl(rax, rdx);
2405 __ jcc(Assembler::greater, default_case);
2406 // lookup dispatch offset
2407 __ subl(rax, rcx);
2408 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2409 __ profile_switch_case(rax, rbx, rcx);
2410 // continue execution
2411 __ bind(continue_execution);
2412 __ bswapl(rdx);
2413 LP64_ONLY(__ movl2ptr(rdx, rdx));
2414 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2415 __ addptr(rbcp, rdx);
2416 __ dispatch_only(vtos);
2417 // handle default
2418 __ bind(default_case);
2419 __ profile_switch_default(rax);
2420 __ movl(rdx, Address(rbx, 0));
2421 __ jmp(continue_execution);
2422 }
2423
2424 void TemplateTable::lookupswitch() {
2425 transition(itos, itos);
2426 __ stop("lookupswitch bytecode should have been rewritten");
2427 }
2428
2429 void TemplateTable::fast_linearswitch() {
2430 transition(itos, vtos);
2431 Label loop_entry, loop, found, continue_execution;
2432 // bswap rax so we can avoid bswapping the table entries
2433 __ bswapl(rax);
2434 // align r13
2435 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2436 // this instruction (change offsets
2437 // below)
2438 __ andptr(rbx, -BytesPerInt);
2439 // set counter
2440 __ movl(rcx, Address(rbx, BytesPerInt));
2441 __ bswapl(rcx);
2442 __ jmpb(loop_entry);
2443 // table search
2444 __ bind(loop);
2445 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2446 __ jcc(Assembler::equal, found);
2447 __ bind(loop_entry);
2448 __ decrementl(rcx);
2449 __ jcc(Assembler::greaterEqual, loop);
2450 // default case
2451 __ profile_switch_default(rax);
2452 __ movl(rdx, Address(rbx, 0));
2453 __ jmp(continue_execution);
2454 // entry found -> get offset
2455 __ bind(found);
2456 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2457 __ profile_switch_case(rcx, rax, rbx);
2458 // continue execution
2459 __ bind(continue_execution);
2460 __ bswapl(rdx);
2461 __ movl2ptr(rdx, rdx);
2462 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2463 __ addptr(rbcp, rdx);
2464 __ dispatch_only(vtos);
2465 }
2466
2467 void TemplateTable::fast_binaryswitch() {
2468 transition(itos, vtos);
2469 // Implementation using the following core algorithm:
2470 //
2471 // int binary_search(int key, LookupswitchPair* array, int n) {
2472 // // Binary search according to "Methodik des Programmierens" by
2473 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2474 // int i = 0;
2475 // int j = n;
2476 // while (i+1 < j) {
2477 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2478 // // with Q: for all i: 0 <= i < n: key < a[i]
2479 // // where a stands for the array and assuming that the (inexisting)
2480 // // element a[n] is infinitely big.
2481 // int h = (i + j) >> 1;
2482 // // i < h < j
2483 // if (key < array[h].fast_match()) {
2484 // j = h;
2485 // } else {
2486 // i = h;
2487 // }
2488 // }
2489 // // R: a[i] <= key < a[i+1] or Q
2490 // // (i.e., if key is within array, i is the correct index)
2491 // return i;
2492 // }
2493
2494 // Register allocation
2495 const Register key = rax; // already set (tosca)
2496 const Register array = rbx;
2497 const Register i = rcx;
2498 const Register j = rdx;
2499 const Register h = rdi;
2500 const Register temp = rsi;
2501
2502 // Find array start
2503 NOT_LP64(__ save_bcp());
2504
2505 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2506 // get rid of this
2507 // instruction (change
2508 // offsets below)
2509 __ andptr(array, -BytesPerInt);
2510
2511 // Initialize i & j
2512 __ xorl(i, i); // i = 0;
2513 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2514
2515 // Convert j into native byteordering
2516 __ bswapl(j);
2517
2518 // And start
2519 Label entry;
2520 __ jmp(entry);
2521
2522 // binary search loop
2523 {
2524 Label loop;
2525 __ bind(loop);
2526 // int h = (i + j) >> 1;
2527 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2528 __ sarl(h, 1); // h = (i + j) >> 1;
2529 // if (key < array[h].fast_match()) {
2530 // j = h;
2531 // } else {
2532 // i = h;
2533 // }
2534 // Convert array[h].match to native byte-ordering before compare
2535 __ movl(temp, Address(array, h, Address::times_8));
2536 __ bswapl(temp);
2537 __ cmpl(key, temp);
2538 // j = h if (key < array[h].fast_match())
2539 __ cmov32(Assembler::less, j, h);
2540 // i = h if (key >= array[h].fast_match())
2541 __ cmov32(Assembler::greaterEqual, i, h);
2542 // while (i+1 < j)
2543 __ bind(entry);
2544 __ leal(h, Address(i, 1)); // i+1
2545 __ cmpl(h, j); // i+1 < j
2546 __ jcc(Assembler::less, loop);
2547 }
2548
2549 // end of binary search, result index is i (must check again!)
2550 Label default_case;
2551 // Convert array[i].match to native byte-ordering before compare
2552 __ movl(temp, Address(array, i, Address::times_8));
2553 __ bswapl(temp);
2554 __ cmpl(key, temp);
2555 __ jcc(Assembler::notEqual, default_case);
2556
2557 // entry found -> j = offset
2558 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2559 __ profile_switch_case(i, key, array);
2560 __ bswapl(j);
2561 LP64_ONLY(__ movslq(j, j));
2562
2563 NOT_LP64(__ restore_bcp());
2564 NOT_LP64(__ restore_locals()); // restore rdi
2565
2566 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2567 __ addptr(rbcp, j);
2568 __ dispatch_only(vtos);
2569
2570 // default case -> j = default offset
2571 __ bind(default_case);
2572 __ profile_switch_default(i);
2573 __ movl(j, Address(array, -2 * BytesPerInt));
2574 __ bswapl(j);
2575 LP64_ONLY(__ movslq(j, j));
2576
2577 NOT_LP64(__ restore_bcp());
2578 NOT_LP64(__ restore_locals());
2579
2580 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2581 __ addptr(rbcp, j);
2582 __ dispatch_only(vtos);
2583 }
2584
2585 void TemplateTable::_return(TosState state) {
2586 transition(state, state);
2587
2588 assert(_desc->calls_vm(),
2589 "inconsistent calls_vm information"); // call in remove_activation
2590
2591 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2592 assert(state == vtos, "only valid state");
2593 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2594 __ movptr(robj, aaddress(0));
2595 __ load_klass(rdi, robj);
2596 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2597 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2598 Label skip_register_finalizer;
2599 __ jcc(Assembler::zero, skip_register_finalizer);
2600
2601 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2602
2603 __ bind(skip_register_finalizer);
2604 }
2605
2606 // Narrow result if state is itos but result type is smaller.
2607 // Need to narrow in the return bytecode rather than in generate_return_entry
2608 // since compiled code callers expect the result to already be narrowed.
2609 if (state == itos) {
2610 __ narrow(rax);
2611 }
2612 __ remove_activation(state, rbcp);
2613
2614 __ jmp(rbcp);
2615 }
2616
2617 // ----------------------------------------------------------------------------
2618 // Volatile variables demand their effects be made known to all CPU's
2619 // in order. Store buffers on most chips allow reads & writes to
2620 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2621 // without some kind of memory barrier (i.e., it's not sufficient that
2622 // the interpreter does not reorder volatile references, the hardware
2623 // also must not reorder them).
2624 //
2625 // According to the new Java Memory Model (JMM):
2626 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2627 // writes act as aquire & release, so:
2628 // (2) A read cannot let unrelated NON-volatile memory refs that
2629 // happen after the read float up to before the read. It's OK for
2630 // non-volatile memory refs that happen before the volatile read to
2631 // float down below it.
2632 // (3) Similar a volatile write cannot let unrelated NON-volatile
2633 // memory refs that happen BEFORE the write float down to after the
2634 // write. It's OK for non-volatile memory refs that happen after the
2635 // volatile write to float up before it.
2636 //
2637 // We only put in barriers around volatile refs (they are expensive),
2638 // not _between_ memory refs (that would require us to track the
2639 // flavor of the previous memory refs). Requirements (2) and (3)
2640 // require some barriers before volatile stores and after volatile
2641 // loads. These nearly cover requirement (1) but miss the
2642 // volatile-store-volatile-load case. This final case is placed after
2643 // volatile-stores although it could just as well go before
2644 // volatile-loads.
2645
2646 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2647 // Helper function to insert a is-volatile test and memory barrier
2648 if(!os::is_MP()) return; // Not needed on single CPU
2649 __ membar(order_constraint);
2650 }
2651
2652 void TemplateTable::resolve_cache_and_index(int byte_no,
2653 Register Rcache,
2654 Register index,
2655 size_t index_size) {
2656 const Register temp = rbx;
2657 assert_different_registers(Rcache, index, temp);
2658
2659 Label resolved;
2660
2661 Bytecodes::Code code = bytecode();
2662 switch (code) {
2663 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2664 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2665 }
2666
2667 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2668 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2669 __ cmpl(temp, code); // have we resolved this bytecode?
2670 __ jcc(Assembler::equal, resolved);
2671
2672 // resolve first time through
2673 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2674 __ movl(temp, code);
2675 __ call_VM(noreg, entry, temp);
2676 // Update registers with resolved info
2677 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2678 __ bind(resolved);
2679 }
2680
2681 // The cache and index registers must be set before call
2682 void TemplateTable::load_field_cp_cache_entry(Register obj,
2683 Register cache,
2684 Register index,
2685 Register off,
2686 Register flags,
2687 bool is_static = false) {
2688 assert_different_registers(cache, index, flags, off);
2689
2690 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2691 // Field offset
2692 __ movptr(off, Address(cache, index, Address::times_ptr,
2693 in_bytes(cp_base_offset +
2694 ConstantPoolCacheEntry::f2_offset())));
2695 // Flags
2696 __ movl(flags, Address(cache, index, Address::times_ptr,
2697 in_bytes(cp_base_offset +
2698 ConstantPoolCacheEntry::flags_offset())));
2699
2700 // klass overwrite register
2701 if (is_static) {
2702 __ movptr(obj, Address(cache, index, Address::times_ptr,
2703 in_bytes(cp_base_offset +
2704 ConstantPoolCacheEntry::f1_offset())));
2705 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2706 __ movptr(obj, Address(obj, mirror_offset));
2707 }
2708 }
2709
2710 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2711 Register method,
2712 Register itable_index,
2713 Register flags,
2714 bool is_invokevirtual,
2715 bool is_invokevfinal, /*unused*/
2716 bool is_invokedynamic) {
2717 // setup registers
2718 const Register cache = rcx;
2719 const Register index = rdx;
2720 assert_different_registers(method, flags);
2721 assert_different_registers(method, cache, index);
2722 assert_different_registers(itable_index, flags);
2723 assert_different_registers(itable_index, cache, index);
2724 // determine constant pool cache field offsets
2725 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2726 const int method_offset = in_bytes(
2727 ConstantPoolCache::base_offset() +
2728 ((byte_no == f2_byte)
2729 ? ConstantPoolCacheEntry::f2_offset()
2730 : ConstantPoolCacheEntry::f1_offset()));
2731 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2732 ConstantPoolCacheEntry::flags_offset());
2733 // access constant pool cache fields
2734 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2735 ConstantPoolCacheEntry::f2_offset());
2736
2737 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2738 resolve_cache_and_index(byte_no, cache, index, index_size);
2739 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2740
2741 if (itable_index != noreg) {
2742 // pick up itable or appendix index from f2 also:
2743 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2744 }
2745 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2746 }
2747
2748 // The registers cache and index expected to be set before call.
2749 // Correct values of the cache and index registers are preserved.
2750 void TemplateTable::jvmti_post_field_access(Register cache,
2751 Register index,
2752 bool is_static,
2753 bool has_tos) {
2754 if (JvmtiExport::can_post_field_access()) {
2755 // Check to see if a field access watch has been set before we take
2756 // the time to call into the VM.
2757 Label L1;
2758 assert_different_registers(cache, index, rax);
2759 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2760 __ testl(rax,rax);
2761 __ jcc(Assembler::zero, L1);
2762
2763 // cache entry pointer
2764 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2765 __ shll(index, LogBytesPerWord);
2766 __ addptr(cache, index);
2767 if (is_static) {
2768 __ xorptr(rax, rax); // NULL object reference
2769 } else {
2770 __ pop(atos); // Get the object
2771 __ verify_oop(rax);
2772 __ push(atos); // Restore stack state
2773 }
2774 // rax,: object pointer or NULL
2775 // cache: cache entry pointer
2776 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2777 rax, cache);
2778 __ get_cache_and_index_at_bcp(cache, index, 1);
2779 __ bind(L1);
2780 }
2781 }
2782
2783 void TemplateTable::pop_and_check_object(Register r) {
2784 __ pop_ptr(r);
2785 __ null_check(r); // for field access must check obj.
2786 __ verify_oop(r);
2787 }
2788
2789 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2790 transition(vtos, vtos);
2791
2792 const Register cache = rcx;
2793 const Register index = rdx;
2794 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2795 const Register off = rbx;
2796 const Register flags = rax;
2797 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2798
2799 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2800 jvmti_post_field_access(cache, index, is_static, false);
2801 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2802
2803 if (!is_static) pop_and_check_object(obj);
2804 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, obj);
2805
2806 const Address field(obj, off, Address::times_1, 0*wordSize);
2807 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2808
2809 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2810
2811 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2812 // Make sure we don't need to mask edx after the above shift
2813 assert(btos == 0, "change code, btos != 0");
2814
2815 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2816
2817 __ jcc(Assembler::notZero, notByte);
2818 // btos
2819 __ load_signed_byte(rax, field);
2820 __ push(btos);
2821 // Rewrite bytecode to be faster
2822 if (!is_static && rc == may_rewrite) {
2823 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2824 }
2825 __ jmp(Done);
2826
2827 __ bind(notByte);
2828 __ cmpl(flags, ztos);
2829 __ jcc(Assembler::notEqual, notBool);
2830
2831 // ztos (same code as btos)
2832 __ load_signed_byte(rax, field);
2833 __ push(ztos);
2834 // Rewrite bytecode to be faster
2835 if (!is_static && rc == may_rewrite) {
2836 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2837 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2838 }
2839 __ jmp(Done);
2840
2841 __ bind(notBool);
2842 __ cmpl(flags, atos);
2843 __ jcc(Assembler::notEqual, notObj);
2844 // atos
2845 __ load_heap_oop(rax, field);
2846 __ push(atos);
2847 if (!is_static && rc == may_rewrite) {
2848 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2849 }
2850 __ jmp(Done);
2851
2852 __ bind(notObj);
2853 __ cmpl(flags, itos);
2854 __ jcc(Assembler::notEqual, notInt);
2855 // itos
2856 __ movl(rax, field);
2857 __ push(itos);
2858 // Rewrite bytecode to be faster
2859 if (!is_static && rc == may_rewrite) {
2860 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2861 }
2862 __ jmp(Done);
2863
2864 __ bind(notInt);
2865 __ cmpl(flags, ctos);
2866 __ jcc(Assembler::notEqual, notChar);
2867 // ctos
2868 __ load_unsigned_short(rax, field);
2869 __ push(ctos);
2870 // Rewrite bytecode to be faster
2871 if (!is_static && rc == may_rewrite) {
2872 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2873 }
2874 __ jmp(Done);
2875
2876 __ bind(notChar);
2877 __ cmpl(flags, stos);
2878 __ jcc(Assembler::notEqual, notShort);
2879 // stos
2880 __ load_signed_short(rax, field);
2881 __ push(stos);
2882 // Rewrite bytecode to be faster
2883 if (!is_static && rc == may_rewrite) {
2884 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2885 }
2886 __ jmp(Done);
2887
2888 __ bind(notShort);
2889 __ cmpl(flags, ltos);
2890 __ jcc(Assembler::notEqual, notLong);
2891 // ltos
2892
2893 #ifndef _LP64
2894 // Generate code as if volatile. There just aren't enough registers to
2895 // save that information and this code is faster than the test.
2896 __ fild_d(field); // Must load atomically
2897 __ subptr(rsp,2*wordSize); // Make space for store
2898 __ fistp_d(Address(rsp,0));
2899 __ pop(rax);
2900 __ pop(rdx);
2901 #else
2902 __ movq(rax, field);
2903 #endif
2904
2905 __ push(ltos);
2906 // Rewrite bytecode to be faster
2907 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2908 __ jmp(Done);
2909
2910 __ bind(notLong);
2911 __ cmpl(flags, ftos);
2912 __ jcc(Assembler::notEqual, notFloat);
2913 // ftos
2914
2915 __ load_float(field);
2916 __ push(ftos);
2917 // Rewrite bytecode to be faster
2918 if (!is_static && rc == may_rewrite) {
2919 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2920 }
2921 __ jmp(Done);
2922
2923 __ bind(notFloat);
2924 #ifdef ASSERT
2925 __ cmpl(flags, dtos);
2926 __ jcc(Assembler::notEqual, notDouble);
2927 #endif
2928 // dtos
2929 __ load_double(field);
2930 __ push(dtos);
2931 // Rewrite bytecode to be faster
2932 if (!is_static && rc == may_rewrite) {
2933 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2934 }
2935 #ifdef ASSERT
2936 __ jmp(Done);
2937
2938
2939 __ bind(notDouble);
2940 __ stop("Bad state");
2941 #endif
2942
2943 __ bind(Done);
2944 // [jk] not needed currently
2945 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2946 // Assembler::LoadStore));
2947 }
2948
2949 void TemplateTable::getfield(int byte_no) {
2950 getfield_or_static(byte_no, false);
2951 }
2952
2953 void TemplateTable::nofast_getfield(int byte_no) {
2954 getfield_or_static(byte_no, false, may_not_rewrite);
2955 }
2956
2957 void TemplateTable::getstatic(int byte_no) {
2958 getfield_or_static(byte_no, true);
2959 }
2960
2961
2962 // The registers cache and index expected to be set before call.
2963 // The function may destroy various registers, just not the cache and index registers.
2964 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2965
2966 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
2967 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2968 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2969 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
2970
2971 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2972
2973 if (JvmtiExport::can_post_field_modification()) {
2974 // Check to see if a field modification watch has been set before
2975 // we take the time to call into the VM.
2976 Label L1;
2977 assert_different_registers(cache, index, rax);
2978 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2979 __ testl(rax, rax);
2980 __ jcc(Assembler::zero, L1);
2981
2982 __ get_cache_and_index_at_bcp(robj, RDX, 1);
2983
2984
2985 if (is_static) {
2986 // Life is simple. Null out the object pointer.
2987 __ xorl(RBX, RBX);
2988
2989 } else {
2990 // Life is harder. The stack holds the value on top, followed by
2991 // the object. We don't know the size of the value, though; it
2992 // could be one or two words depending on its type. As a result,
2993 // we must find the type to determine where the object is.
2994 #ifndef _LP64
2995 Label two_word, valsize_known;
2996 #endif
2997 __ movl(RCX, Address(robj, RDX,
2998 Address::times_ptr,
2999 in_bytes(cp_base_offset +
3000 ConstantPoolCacheEntry::flags_offset())));
3001 NOT_LP64(__ mov(rbx, rsp));
3002 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3003
3004 // Make sure we don't need to mask rcx after the above shift
3005 ConstantPoolCacheEntry::verify_tos_state_shift();
3006 #ifdef _LP64
3007 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
3008 __ cmpl(c_rarg3, ltos);
3009 __ cmovptr(Assembler::equal,
3010 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3011 __ cmpl(c_rarg3, dtos);
3012 __ cmovptr(Assembler::equal,
3013 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3014 #else
3015 __ cmpl(rcx, ltos);
3016 __ jccb(Assembler::equal, two_word);
3017 __ cmpl(rcx, dtos);
3018 __ jccb(Assembler::equal, two_word);
3019 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3020 __ jmpb(valsize_known);
3021
3022 __ bind(two_word);
3023 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3024
3025 __ bind(valsize_known);
3026 // setup object pointer
3027 __ movptr(rbx, Address(rbx, 0));
3028 #endif
3029 }
3030 // cache entry pointer
3031 __ addptr(robj, in_bytes(cp_base_offset));
3032 __ shll(RDX, LogBytesPerWord);
3033 __ addptr(robj, RDX);
3034 // object (tos)
3035 __ mov(RCX, rsp);
3036 // c_rarg1: object pointer set up above (NULL if static)
3037 // c_rarg2: cache entry pointer
3038 // c_rarg3: jvalue object on the stack
3039 __ call_VM(noreg,
3040 CAST_FROM_FN_PTR(address,
3041 InterpreterRuntime::post_field_modification),
3042 RBX, robj, RCX);
3043 __ get_cache_and_index_at_bcp(cache, index, 1);
3044 __ bind(L1);
3045 }
3046 }
3047
3048 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3049 transition(vtos, vtos);
3050
3051 const Register cache = rcx;
3052 const Register index = rdx;
3053 const Register obj = rcx;
3054 const Register off = rbx;
3055 const Register flags = rax;
3056 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3057
3058 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3059 jvmti_post_field_mod(cache, index, is_static);
3060 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3061
3062 // [jk] not needed currently
3063 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3064 // Assembler::StoreStore));
3065
3066 Label notVolatile, Done;
3067 __ movl(rdx, flags);
3068 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3069 __ andl(rdx, 0x1);
3070
3071 // field addresses
3072 const Address field(obj, off, Address::times_1, 0*wordSize);
3073 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3074
3075 Label notByte, notBool, notInt, notShort, notChar,
3076 notLong, notFloat, notObj, notDouble;
3077
3078 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3079
3080 assert(btos == 0, "change code, btos != 0");
3081 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3082 __ jcc(Assembler::notZero, notByte);
3083
3084 // btos
3085 {
3086 __ pop(btos);
3087 if (!is_static) pop_and_check_object(obj);
3088 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3089 __ movb(field, rax);
3090 if (!is_static && rc == may_rewrite) {
3091 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3092 }
3093 __ jmp(Done);
3094 }
3095
3096 __ bind(notByte);
3097 __ cmpl(flags, ztos);
3098 __ jcc(Assembler::notEqual, notBool);
3099
3100 // ztos
3101 {
3102 __ pop(ztos);
3103 if (!is_static) pop_and_check_object(obj);
3104 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3105 __ andl(rax, 0x1);
3106 __ movb(field, rax);
3107 if (!is_static && rc == may_rewrite) {
3108 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3109 }
3110 __ jmp(Done);
3111 }
3112
3113 __ bind(notBool);
3114 __ cmpl(flags, atos);
3115 __ jcc(Assembler::notEqual, notObj);
3116
3117 // atos
3118 {
3119 __ pop(atos);
3120 if (!is_static) pop_and_check_object(obj);
3121 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3122 // Store into the field
3123 do_oop_store(_masm, field, rax, _bs->kind(), false);
3124 if (!is_static && rc == may_rewrite) {
3125 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3126 }
3127 __ jmp(Done);
3128 }
3129
3130 __ bind(notObj);
3131 __ cmpl(flags, itos);
3132 __ jcc(Assembler::notEqual, notInt);
3133
3134 // itos
3135 {
3136 __ pop(itos);
3137 if (!is_static) pop_and_check_object(obj);
3138 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3139 __ movl(field, rax);
3140 if (!is_static && rc == may_rewrite) {
3141 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3142 }
3143 __ jmp(Done);
3144 }
3145
3146 __ bind(notInt);
3147 __ cmpl(flags, ctos);
3148 __ jcc(Assembler::notEqual, notChar);
3149
3150 // ctos
3151 {
3152 __ pop(ctos);
3153 if (!is_static) pop_and_check_object(obj);
3154 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3155 __ movw(field, rax);
3156 if (!is_static && rc == may_rewrite) {
3157 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3158 }
3159 __ jmp(Done);
3160 }
3161
3162 __ bind(notChar);
3163 __ cmpl(flags, stos);
3164 __ jcc(Assembler::notEqual, notShort);
3165
3166 // stos
3167 {
3168 __ pop(stos);
3169 if (!is_static) pop_and_check_object(obj);
3170 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3171 __ movw(field, rax);
3172 if (!is_static && rc == may_rewrite) {
3173 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3174 }
3175 __ jmp(Done);
3176 }
3177
3178 __ bind(notShort);
3179 __ cmpl(flags, ltos);
3180 __ jcc(Assembler::notEqual, notLong);
3181
3182 // ltos
3183 #ifdef _LP64
3184 {
3185 __ pop(ltos);
3186 if (!is_static) pop_and_check_object(obj);
3187 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3188 __ movq(field, rax);
3189 if (!is_static && rc == may_rewrite) {
3190 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3191 }
3192 __ jmp(Done);
3193 }
3194 #else
3195 {
3196 Label notVolatileLong;
3197 __ testl(rdx, rdx);
3198 __ jcc(Assembler::zero, notVolatileLong);
3199
3200 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3201 if (!is_static) pop_and_check_object(obj);
3202
3203 // Replace with real volatile test
3204 __ push(rdx);
3205 __ push(rax); // Must update atomically with FIST
3206 __ fild_d(Address(rsp,0)); // So load into FPU register
3207 __ fistp_d(field); // and put into memory atomically
3208 __ addptr(rsp, 2*wordSize);
3209 // volatile_barrier();
3210 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3211 Assembler::StoreStore));
3212 // Don't rewrite volatile version
3213 __ jmp(notVolatile);
3214
3215 __ bind(notVolatileLong);
3216
3217 __ pop(ltos); // overwrites rdx
3218 if (!is_static) pop_and_check_object(obj);
3219 __ movptr(hi, rdx);
3220 __ movptr(field, rax);
3221 // Don't rewrite to _fast_lputfield for potential volatile case.
3222 __ jmp(notVolatile);
3223 }
3224 #endif // _LP64
3225
3226 __ bind(notLong);
3227 __ cmpl(flags, ftos);
3228 __ jcc(Assembler::notEqual, notFloat);
3229
3230 // ftos
3231 {
3232 __ pop(ftos);
3233 if (!is_static) pop_and_check_object(obj);
3234 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3235 __ store_float(field);
3236 if (!is_static && rc == may_rewrite) {
3237 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3238 }
3239 __ jmp(Done);
3240 }
3241
3242 __ bind(notFloat);
3243 #ifdef ASSERT
3244 __ cmpl(flags, dtos);
3245 __ jcc(Assembler::notEqual, notDouble);
3246 #endif
3247
3248 // dtos
3249 {
3250 __ pop(dtos);
3251 if (!is_static) pop_and_check_object(obj);
3252 oopDesc::bs()->interpreter_write_barrier(_masm, obj);
3253 __ store_double(field);
3254 if (!is_static && rc == may_rewrite) {
3255 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3256 }
3257 }
3258
3259 #ifdef ASSERT
3260 __ jmp(Done);
3261
3262 __ bind(notDouble);
3263 __ stop("Bad state");
3264 #endif
3265
3266 __ bind(Done);
3267
3268 // Check for volatile store
3269 __ testl(rdx, rdx);
3270 __ jcc(Assembler::zero, notVolatile);
3271 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3272 Assembler::StoreStore));
3273 __ bind(notVolatile);
3274 }
3275
3276 void TemplateTable::putfield(int byte_no) {
3277 putfield_or_static(byte_no, false);
3278 }
3279
3280 void TemplateTable::nofast_putfield(int byte_no) {
3281 putfield_or_static(byte_no, false, may_not_rewrite);
3282 }
3283
3284 void TemplateTable::putstatic(int byte_no) {
3285 putfield_or_static(byte_no, true);
3286 }
3287
3288 void TemplateTable::jvmti_post_fast_field_mod() {
3289
3290 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3291
3292 if (JvmtiExport::can_post_field_modification()) {
3293 // Check to see if a field modification watch has been set before
3294 // we take the time to call into the VM.
3295 Label L2;
3296 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3297 __ testl(scratch, scratch);
3298 __ jcc(Assembler::zero, L2);
3299 __ pop_ptr(rbx); // copy the object pointer from tos
3300 __ verify_oop(rbx);
3301 __ push_ptr(rbx); // put the object pointer back on tos
3302 // Save tos values before call_VM() clobbers them. Since we have
3303 // to do it for every data type, we use the saved values as the
3304 // jvalue object.
3305 switch (bytecode()) { // load values into the jvalue object
3306 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3307 case Bytecodes::_fast_bputfield: // fall through
3308 case Bytecodes::_fast_zputfield: // fall through
3309 case Bytecodes::_fast_sputfield: // fall through
3310 case Bytecodes::_fast_cputfield: // fall through
3311 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3312 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3313 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3314 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3315
3316 default:
3317 ShouldNotReachHere();
3318 }
3319 __ mov(scratch, rsp); // points to jvalue on the stack
3320 // access constant pool cache entry
3321 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3322 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3323 __ verify_oop(rbx);
3324 // rbx: object pointer copied above
3325 // c_rarg2: cache entry pointer
3326 // c_rarg3: jvalue object on the stack
3327 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3328 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3329
3330 switch (bytecode()) { // restore tos values
3331 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3332 case Bytecodes::_fast_bputfield: // fall through
3333 case Bytecodes::_fast_zputfield: // fall through
3334 case Bytecodes::_fast_sputfield: // fall through
3335 case Bytecodes::_fast_cputfield: // fall through
3336 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3337 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3338 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3339 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3340 }
3341 __ bind(L2);
3342 }
3343 }
3344
3345 void TemplateTable::fast_storefield(TosState state) {
3346 transition(state, vtos);
3347
3348 ByteSize base = ConstantPoolCache::base_offset();
3349
3350 jvmti_post_fast_field_mod();
3351
3352 // access constant pool cache
3353 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3354
3355 // test for volatile with rdx but rdx is tos register for lputfield.
3356 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3357 in_bytes(base +
3358 ConstantPoolCacheEntry::flags_offset())));
3359
3360 // replace index with field offset from cache entry
3361 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3362 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3363
3364 // [jk] not needed currently
3365 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3366 // Assembler::StoreStore));
3367
3368 Label notVolatile;
3369 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3370 __ andl(rdx, 0x1);
3371
3372 // Get object from stack
3373 pop_and_check_object(rcx);
3374 oopDesc::bs()->interpreter_write_barrier(_masm, rcx);
3375
3376 // field address
3377 const Address field(rcx, rbx, Address::times_1);
3378
3379 // access field
3380 switch (bytecode()) {
3381 case Bytecodes::_fast_aputfield:
3382 do_oop_store(_masm, field, rax, _bs->kind(), false);
3383 break;
3384 case Bytecodes::_fast_lputfield:
3385 #ifdef _LP64
3386 __ movq(field, rax);
3387 #else
3388 __ stop("should not be rewritten");
3389 #endif
3390 break;
3391 case Bytecodes::_fast_iputfield:
3392 __ movl(field, rax);
3393 break;
3394 case Bytecodes::_fast_zputfield:
3395 __ andl(rax, 0x1); // boolean is true if LSB is 1
3396 // fall through to bputfield
3397 case Bytecodes::_fast_bputfield:
3398 __ movb(field, rax);
3399 break;
3400 case Bytecodes::_fast_sputfield:
3401 // fall through
3402 case Bytecodes::_fast_cputfield:
3403 __ movw(field, rax);
3404 break;
3405 case Bytecodes::_fast_fputfield:
3406 __ store_float(field);
3407 break;
3408 case Bytecodes::_fast_dputfield:
3409 __ store_double(field);
3410 break;
3411 default:
3412 ShouldNotReachHere();
3413 }
3414
3415 // Check for volatile store
3416 __ testl(rdx, rdx);
3417 __ jcc(Assembler::zero, notVolatile);
3418 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3419 Assembler::StoreStore));
3420 __ bind(notVolatile);
3421 }
3422
3423 void TemplateTable::fast_accessfield(TosState state) {
3424 transition(atos, state);
3425
3426 // Do the JVMTI work here to avoid disturbing the register state below
3427 if (JvmtiExport::can_post_field_access()) {
3428 // Check to see if a field access watch has been set before we
3429 // take the time to call into the VM.
3430 Label L1;
3431 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3432 __ testl(rcx, rcx);
3433 __ jcc(Assembler::zero, L1);
3434 // access constant pool cache entry
3435 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3436 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3437 __ verify_oop(rax);
3438 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3439 LP64_ONLY(__ mov(c_rarg1, rax));
3440 // c_rarg1: object pointer copied above
3441 // c_rarg2: cache entry pointer
3442 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3443 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3444 __ pop_ptr(rax); // restore object pointer
3445 __ bind(L1);
3446 }
3447
3448 // access constant pool cache
3449 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3450 // replace index with field offset from cache entry
3451 // [jk] not needed currently
3452 // if (os::is_MP()) {
3453 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3454 // in_bytes(ConstantPoolCache::base_offset() +
3455 // ConstantPoolCacheEntry::flags_offset())));
3456 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3457 // __ andl(rdx, 0x1);
3458 // }
3459 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3460 in_bytes(ConstantPoolCache::base_offset() +
3461 ConstantPoolCacheEntry::f2_offset())));
3462
3463 // rax: object
3464 __ verify_oop(rax);
3465 __ null_check(rax);
3466 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rax);
3467 Address field(rax, rbx, Address::times_1);
3468
3469 // access field
3470 switch (bytecode()) {
3471 case Bytecodes::_fast_agetfield:
3472 __ load_heap_oop(rax, field);
3473 __ verify_oop(rax);
3474 break;
3475 case Bytecodes::_fast_lgetfield:
3476 #ifdef _LP64
3477 __ movq(rax, field);
3478 #else
3479 __ stop("should not be rewritten");
3480 #endif
3481 break;
3482 case Bytecodes::_fast_igetfield:
3483 __ movl(rax, field);
3484 break;
3485 case Bytecodes::_fast_bgetfield:
3486 __ movsbl(rax, field);
3487 break;
3488 case Bytecodes::_fast_sgetfield:
3489 __ load_signed_short(rax, field);
3490 break;
3491 case Bytecodes::_fast_cgetfield:
3492 __ load_unsigned_short(rax, field);
3493 break;
3494 case Bytecodes::_fast_fgetfield:
3495 __ load_float(field);
3496 break;
3497 case Bytecodes::_fast_dgetfield:
3498 __ load_double(field);
3499 break;
3500 default:
3501 ShouldNotReachHere();
3502 }
3503 // [jk] not needed currently
3504 // if (os::is_MP()) {
3505 // Label notVolatile;
3506 // __ testl(rdx, rdx);
3507 // __ jcc(Assembler::zero, notVolatile);
3508 // __ membar(Assembler::LoadLoad);
3509 // __ bind(notVolatile);
3510 //};
3511 }
3512
3513 void TemplateTable::fast_xaccess(TosState state) {
3514 transition(vtos, state);
3515
3516 // get receiver
3517 __ movptr(rax, aaddress(0));
3518 // access constant pool cache
3519 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3520 __ movptr(rbx,
3521 Address(rcx, rdx, Address::times_ptr,
3522 in_bytes(ConstantPoolCache::base_offset() +
3523 ConstantPoolCacheEntry::f2_offset())));
3524 // make sure exception is reported in correct bcp range (getfield is
3525 // next instruction)
3526 __ increment(rbcp);
3527 __ null_check(rax);
3528 oopDesc::bs()->interpreter_read_barrier_not_null(_masm, rax);
3529 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3530 switch (state) {
3531 case itos:
3532 __ movl(rax, field);
3533 break;
3534 case atos:
3535 __ load_heap_oop(rax, field);
3536 __ verify_oop(rax);
3537 break;
3538 case ftos:
3539 __ load_float(field);
3540 break;
3541 default:
3542 ShouldNotReachHere();
3543 }
3544
3545 // [jk] not needed currently
3546 // if (os::is_MP()) {
3547 // Label notVolatile;
3548 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3549 // in_bytes(ConstantPoolCache::base_offset() +
3550 // ConstantPoolCacheEntry::flags_offset())));
3551 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3552 // __ testl(rdx, 0x1);
3553 // __ jcc(Assembler::zero, notVolatile);
3554 // __ membar(Assembler::LoadLoad);
3555 // __ bind(notVolatile);
3556 // }
3557
3558 __ decrement(rbcp);
3559 }
3560
3561 //-----------------------------------------------------------------------------
3562 // Calls
3563
3564 void TemplateTable::count_calls(Register method, Register temp) {
3565 // implemented elsewhere
3566 ShouldNotReachHere();
3567 }
3568
3569 void TemplateTable::prepare_invoke(int byte_no,
3570 Register method, // linked method (or i-klass)
3571 Register index, // itable index, MethodType, etc.
3572 Register recv, // if caller wants to see it
3573 Register flags // if caller wants to test it
3574 ) {
3575 // determine flags
3576 const Bytecodes::Code code = bytecode();
3577 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3578 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3579 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3580 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3581 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3582 const bool load_receiver = (recv != noreg);
3583 const bool save_flags = (flags != noreg);
3584 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3585 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3586 assert(flags == noreg || flags == rdx, "");
3587 assert(recv == noreg || recv == rcx, "");
3588
3589 // setup registers & access constant pool cache
3590 if (recv == noreg) recv = rcx;
3591 if (flags == noreg) flags = rdx;
3592 assert_different_registers(method, index, recv, flags);
3593
3594 // save 'interpreter return address'
3595 __ save_bcp();
3596
3597 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3598
3599 // maybe push appendix to arguments (just before return address)
3600 if (is_invokedynamic || is_invokehandle) {
3601 Label L_no_push;
3602 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3603 __ jcc(Assembler::zero, L_no_push);
3604 // Push the appendix as a trailing parameter.
3605 // This must be done before we get the receiver,
3606 // since the parameter_size includes it.
3607 __ push(rbx);
3608 __ mov(rbx, index);
3609 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3610 __ load_resolved_reference_at_index(index, rbx);
3611 __ pop(rbx);
3612 __ push(index); // push appendix (MethodType, CallSite, etc.)
3613 __ bind(L_no_push);
3614 }
3615
3616 // load receiver if needed (after appendix is pushed so parameter size is correct)
3617 // Note: no return address pushed yet
3618 if (load_receiver) {
3619 __ movl(recv, flags);
3620 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3621 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3622 const int receiver_is_at_end = -1; // back off one slot to get receiver
3623 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3624 __ movptr(recv, recv_addr);
3625 __ verify_oop(recv);
3626 }
3627
3628 if (save_flags) {
3629 __ movl(rbcp, flags);
3630 }
3631
3632 // compute return type
3633 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3634 // Make sure we don't need to mask flags after the above shift
3635 ConstantPoolCacheEntry::verify_tos_state_shift();
3636 // load return address
3637 {
3638 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3639 ExternalAddress table(table_addr);
3640 LP64_ONLY(__ lea(rscratch1, table));
3641 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3642 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3643 }
3644
3645 // push return address
3646 __ push(flags);
3647
3648 // Restore flags value from the constant pool cache, and restore rsi
3649 // for later null checks. r13 is the bytecode pointer
3650 if (save_flags) {
3651 __ movl(flags, rbcp);
3652 __ restore_bcp();
3653 }
3654 }
3655
3656 void TemplateTable::invokevirtual_helper(Register index,
3657 Register recv,
3658 Register flags) {
3659 // Uses temporary registers rax, rdx
3660 assert_different_registers(index, recv, rax, rdx);
3661 assert(index == rbx, "");
3662 assert(recv == rcx, "");
3663
3664 // Test for an invoke of a final method
3665 Label notFinal;
3666 __ movl(rax, flags);
3667 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3668 __ jcc(Assembler::zero, notFinal);
3669
3670 const Register method = index; // method must be rbx
3671 assert(method == rbx,
3672 "Method* must be rbx for interpreter calling convention");
3673
3674 // do the call - the index is actually the method to call
3675 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3676
3677 // It's final, need a null check here!
3678 __ null_check(recv);
3679
3680 // profile this call
3681 __ profile_final_call(rax);
3682 __ profile_arguments_type(rax, method, rbcp, true);
3683
3684 __ jump_from_interpreted(method, rax);
3685
3686 __ bind(notFinal);
3687
3688 // get receiver klass
3689 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3690 __ load_klass(rax, recv);
3691
3692 // profile this call
3693 __ profile_virtual_call(rax, rlocals, rdx);
3694 // get target Method* & entry point
3695 __ lookup_virtual_method(rax, index, method);
3696 __ profile_called_method(method, rdx, rbcp);
3697
3698 __ profile_arguments_type(rdx, method, rbcp, true);
3699 __ jump_from_interpreted(method, rdx);
3700 }
3701
3702 void TemplateTable::invokevirtual(int byte_no) {
3703 transition(vtos, vtos);
3704 assert(byte_no == f2_byte, "use this argument");
3705 prepare_invoke(byte_no,
3706 rbx, // method or vtable index
3707 noreg, // unused itable index
3708 rcx, rdx); // recv, flags
3709
3710 // rbx: index
3711 // rcx: receiver
3712 // rdx: flags
3713
3714 invokevirtual_helper(rbx, rcx, rdx);
3715 }
3716
3717 void TemplateTable::invokespecial(int byte_no) {
3718 transition(vtos, vtos);
3719 assert(byte_no == f1_byte, "use this argument");
3720 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3721 rcx); // get receiver also for null check
3722 __ verify_oop(rcx);
3723 __ null_check(rcx);
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 void TemplateTable::invokestatic(int byte_no) {
3731 transition(vtos, vtos);
3732 assert(byte_no == f1_byte, "use this argument");
3733 prepare_invoke(byte_no, rbx); // get f1 Method*
3734 // do the call
3735 __ profile_call(rax);
3736 __ profile_arguments_type(rax, rbx, rbcp, false);
3737 __ jump_from_interpreted(rbx, rax);
3738 }
3739
3740
3741 void TemplateTable::fast_invokevfinal(int byte_no) {
3742 transition(vtos, vtos);
3743 assert(byte_no == f2_byte, "use this argument");
3744 __ stop("fast_invokevfinal not used on x86");
3745 }
3746
3747
3748 void TemplateTable::invokeinterface(int byte_no) {
3749 transition(vtos, vtos);
3750 assert(byte_no == f1_byte, "use this argument");
3751 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3752 rcx, rdx); // recv, flags
3753
3754 // rax: interface klass (from f1)
3755 // rbx: itable index (from f2)
3756 // rcx: receiver
3757 // rdx: flags
3758
3759 // Special case of invokeinterface called for virtual method of
3760 // java.lang.Object. See cpCacheOop.cpp for details.
3761 // This code isn't produced by javac, but could be produced by
3762 // another compliant java compiler.
3763 Label notMethod;
3764 __ movl(rlocals, rdx);
3765 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3766
3767 __ jcc(Assembler::zero, notMethod);
3768
3769 invokevirtual_helper(rbx, rcx, rdx);
3770 __ bind(notMethod);
3771
3772 // Get receiver klass into rdx - also a null check
3773 __ restore_locals(); // restore r14
3774 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3775 __ load_klass(rdx, rcx);
3776
3777 // profile this call
3778 __ profile_virtual_call(rdx, rbcp, rlocals);
3779
3780 Label no_such_interface, no_such_method;
3781
3782 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3783 rdx, rax, rbx,
3784 // outputs: method, scan temp. reg
3785 rbx, rbcp,
3786 no_such_interface);
3787
3788 // rbx: Method* to call
3789 // rcx: receiver
3790 // Check for abstract method error
3791 // Note: This should be done more efficiently via a throw_abstract_method_error
3792 // interpreter entry point and a conditional jump to it in case of a null
3793 // method.
3794 __ testptr(rbx, rbx);
3795 __ jcc(Assembler::zero, no_such_method);
3796
3797 __ profile_called_method(rbx, rbcp, rdx);
3798 __ profile_arguments_type(rdx, rbx, rbcp, true);
3799
3800 // do the call
3801 // rcx: receiver
3802 // rbx,: Method*
3803 __ jump_from_interpreted(rbx, rdx);
3804 __ should_not_reach_here();
3805
3806 // exception handling code follows...
3807 // note: must restore interpreter registers to canonical
3808 // state for exception handling to work correctly!
3809
3810 __ bind(no_such_method);
3811 // throw exception
3812 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3813 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3814 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3815 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3816 // the call_VM checks for exception, so we should never return here.
3817 __ should_not_reach_here();
3818
3819 __ bind(no_such_interface);
3820 // throw exception
3821 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3822 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3823 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3824 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3825 InterpreterRuntime::throw_IncompatibleClassChangeError));
3826 // the call_VM checks for exception, so we should never return here.
3827 __ should_not_reach_here();
3828 }
3829
3830 void TemplateTable::invokehandle(int byte_no) {
3831 transition(vtos, vtos);
3832 assert(byte_no == f1_byte, "use this argument");
3833 const Register rbx_method = rbx;
3834 const Register rax_mtype = rax;
3835 const Register rcx_recv = rcx;
3836 const Register rdx_flags = rdx;
3837
3838 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3839 __ verify_method_ptr(rbx_method);
3840 __ verify_oop(rcx_recv);
3841 __ null_check(rcx_recv);
3842
3843 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3844 // rbx: MH.invokeExact_MT method (from f2)
3845
3846 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3847
3848 // FIXME: profile the LambdaForm also
3849 __ profile_final_call(rax);
3850 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3851
3852 __ jump_from_interpreted(rbx_method, rdx);
3853 }
3854
3855 void TemplateTable::invokedynamic(int byte_no) {
3856 transition(vtos, vtos);
3857 assert(byte_no == f1_byte, "use this argument");
3858
3859 const Register rbx_method = rbx;
3860 const Register rax_callsite = rax;
3861
3862 prepare_invoke(byte_no, rbx_method, rax_callsite);
3863
3864 // rax: CallSite object (from cpool->resolved_references[f1])
3865 // rbx: MH.linkToCallSite method (from f2)
3866
3867 // Note: rax_callsite is already pushed by prepare_invoke
3868
3869 // %%% should make a type profile for any invokedynamic that takes a ref argument
3870 // profile this call
3871 __ profile_call(rbcp);
3872 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3873
3874 __ verify_oop(rax_callsite);
3875
3876 __ jump_from_interpreted(rbx_method, rdx);
3877 }
3878
3879 //-----------------------------------------------------------------------------
3880 // Allocation
3881
3882 void TemplateTable::_new() {
3883 transition(vtos, atos);
3884 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3885 Label slow_case;
3886 Label slow_case_no_pop;
3887 Label done;
3888 Label initialize_header;
3889 Label initialize_object; // including clearing the fields
3890 Label allocate_shared;
3891
3892 __ get_cpool_and_tags(rcx, rax);
3893
3894 // Make sure the class we're about to instantiate has been resolved.
3895 // This is done before loading InstanceKlass to be consistent with the order
3896 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3897 const int tags_offset = Array<u1>::base_offset_in_bytes();
3898 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3899 __ jcc(Assembler::notEqual, slow_case_no_pop);
3900
3901 // get InstanceKlass
3902 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool)));
3903 __ push(rcx); // save the contexts of klass for initializing the header
3904
3905 // make sure klass is initialized & doesn't have finalizer
3906 // make sure klass is fully initialized
3907 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3908 __ jcc(Assembler::notEqual, slow_case);
3909
3910 // get instance_size in InstanceKlass (scaled to a count of bytes)
3911 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3912 // test to see if it has a finalizer or is malformed in some way
3913 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3914 __ jcc(Assembler::notZero, slow_case);
3915
3916 //
3917 // Allocate the instance
3918 // 1) Try to allocate in the TLAB
3919 // 2) if fail and the object is large allocate in the shared Eden
3920 // 3) if the above fails (or is not applicable), go to a slow case
3921 // (creates a new TLAB, etc.)
3922
3923 const bool allow_shared_alloc =
3924 Universe::heap()->supports_inline_contig_alloc();
3925
3926 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
3927 #ifndef _LP64
3928 if (UseTLAB || allow_shared_alloc) {
3929 __ get_thread(thread);
3930 }
3931 #endif // _LP64
3932
3933 if (UseTLAB) {
3934 uint oop_extra_words = Universe::heap()->oop_extra_words();
3935 if (oop_extra_words > 0) {
3936 __ addq(rdx, oop_extra_words * HeapWordSize);
3937 }
3938
3939 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3940 __ lea(rbx, Address(rax, rdx, Address::times_1));
3941 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3942 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3943 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3944 Universe::heap()->compile_prepare_oop(_masm, rax);
3945 if (ZeroTLAB) {
3946 // the fields have been already cleared
3947 __ jmp(initialize_header);
3948 } else {
3949 // initialize both the header and fields
3950 __ jmp(initialize_object);
3951 }
3952 }
3953
3954 // Allocation in the shared Eden, if allowed.
3955 //
3956 // rdx: instance size in bytes
3957 if (allow_shared_alloc) {
3958 __ bind(allocate_shared);
3959
3960 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3961 ExternalAddress heap_end((address)Universe::heap()->end_addr());
3962
3963 Label retry;
3964 __ bind(retry);
3965 __ movptr(rax, heap_top);
3966 __ lea(rbx, Address(rax, rdx, Address::times_1));
3967 __ cmpptr(rbx, heap_end);
3968 __ jcc(Assembler::above, slow_case);
3969
3970 // Compare rax, with the top addr, and if still equal, store the new
3971 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3972 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3973 //
3974 // rax,: object begin
3975 // rbx,: object end
3976 // rdx: instance size in bytes
3977 __ locked_cmpxchgptr(rbx, heap_top);
3978
3979 // if someone beat us on the allocation, try again, otherwise continue
3980 __ jcc(Assembler::notEqual, retry);
3981
3982 __ incr_allocated_bytes(thread, rdx, 0);
3983 }
3984
3985 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3986 // The object is initialized before the header. If the object size is
3987 // zero, go directly to the header initialization.
3988 __ bind(initialize_object);
3989 __ decrement(rdx, sizeof(oopDesc));
3990 __ jcc(Assembler::zero, initialize_header);
3991
3992 // Initialize topmost object field, divide rdx by 8, check if odd and
3993 // test if zero.
3994 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3995 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3996
3997 // rdx must have been multiple of 8
3998 #ifdef ASSERT
3999 // make sure rdx was multiple of 8
4000 Label L;
4001 // Ignore partial flag stall after shrl() since it is debug VM
4002 __ jccb(Assembler::carryClear, L);
4003 __ stop("object size is not multiple of 2 - adjust this code");
4004 __ bind(L);
4005 // rdx must be > 0, no extra check needed here
4006 #endif
4007
4008 // initialize remaining object fields: rdx was a multiple of 8
4009 { Label loop;
4010 __ bind(loop);
4011 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4012 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4013 __ decrement(rdx);
4014 __ jcc(Assembler::notZero, loop);
4015 }
4016
4017 // initialize object header only.
4018 __ bind(initialize_header);
4019 if (UseBiasedLocking) {
4020 __ pop(rcx); // get saved klass back in the register.
4021 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4022 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4023 } else {
4024 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4025 (intptr_t)markOopDesc::prototype()); // header
4026 __ pop(rcx); // get saved klass back in the register.
4027 }
4028 #ifdef _LP64
4029 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4030 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
4031 #endif
4032 __ store_klass(rax, rcx); // klass
4033
4034 {
4035 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4036 // Trigger dtrace event for fastpath
4037 __ push(atos);
4038 __ call_VM_leaf(
4039 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4040 __ pop(atos);
4041 }
4042
4043 __ jmp(done);
4044 }
4045
4046 // slow case
4047 __ bind(slow_case);
4048 __ pop(rcx); // restore stack pointer to what it was when we came in.
4049 __ bind(slow_case_no_pop);
4050
4051 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4052 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4053
4054 __ get_constant_pool(rarg1);
4055 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4056 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4057 __ verify_oop(rax);
4058
4059 // continue
4060 __ bind(done);
4061 }
4062
4063 void TemplateTable::newarray() {
4064 transition(itos, atos);
4065 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4066 __ load_unsigned_byte(rarg1, at_bcp(1));
4067 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4068 rarg1, rax);
4069 }
4070
4071 void TemplateTable::anewarray() {
4072 transition(itos, atos);
4073
4074 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4075 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4076
4077 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4078 __ get_constant_pool(rarg1);
4079 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4080 rarg1, rarg2, rax);
4081 }
4082
4083 void TemplateTable::arraylength() {
4084 transition(atos, itos);
4085 if (ShenandoahVerifyReadsToFromSpace) {
4086 oopDesc::bs()->interpreter_read_barrier(_masm, rax);
4087 }
4088 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4089 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4090 }
4091
4092 void TemplateTable::checkcast() {
4093 transition(atos, atos);
4094 Label done, is_null, ok_is_subtype, quicked, resolved;
4095 __ testptr(rax, rax); // object is in rax
4096 __ jcc(Assembler::zero, is_null);
4097
4098 // Get cpool & tags index
4099 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4100 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4101 // See if bytecode has already been quicked
4102 __ cmpb(Address(rdx, rbx,
4103 Address::times_1,
4104 Array<u1>::base_offset_in_bytes()),
4105 JVM_CONSTANT_Class);
4106 __ jcc(Assembler::equal, quicked);
4107 __ push(atos); // save receiver for result, and for GC
4108 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4109
4110 // vm_result_2 has metadata result
4111 #ifndef _LP64
4112 // borrow rdi from locals
4113 __ get_thread(rdi);
4114 __ get_vm_result_2(rax, rdi);
4115 __ restore_locals();
4116 #else
4117 __ get_vm_result_2(rax, r15_thread);
4118 #endif
4119
4120 __ pop_ptr(rdx); // restore receiver
4121 __ jmpb(resolved);
4122
4123 // Get superklass in rax and subklass in rbx
4124 __ bind(quicked);
4125 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4126 __ movptr(rax, Address(rcx, rbx,
4127 Address::times_ptr, sizeof(ConstantPool)));
4128
4129 __ bind(resolved);
4130 __ load_klass(rbx, rdx);
4131
4132 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4133 // Superklass in rax. Subklass in rbx.
4134 __ gen_subtype_check(rbx, ok_is_subtype);
4135
4136 // Come here on failure
4137 __ push_ptr(rdx);
4138 // object is at TOS
4139 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4140
4141 // Come here on success
4142 __ bind(ok_is_subtype);
4143 __ mov(rax, rdx); // Restore object in rdx
4144
4145 // Collect counts on whether this check-cast sees NULLs a lot or not.
4146 if (ProfileInterpreter) {
4147 __ jmp(done);
4148 __ bind(is_null);
4149 __ profile_null_seen(rcx);
4150 } else {
4151 __ bind(is_null); // same as 'done'
4152 }
4153 __ bind(done);
4154 }
4155
4156 void TemplateTable::instanceof() {
4157 transition(atos, itos);
4158 Label done, is_null, ok_is_subtype, quicked, resolved;
4159 __ testptr(rax, rax);
4160 __ jcc(Assembler::zero, is_null);
4161
4162 // Get cpool & tags index
4163 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4164 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4165 // See if bytecode has already been quicked
4166 __ cmpb(Address(rdx, rbx,
4167 Address::times_1,
4168 Array<u1>::base_offset_in_bytes()),
4169 JVM_CONSTANT_Class);
4170 __ jcc(Assembler::equal, quicked);
4171
4172 __ push(atos); // save receiver for result, and for GC
4173 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4174 // vm_result_2 has metadata result
4175
4176 #ifndef _LP64
4177 // borrow rdi from locals
4178 __ get_thread(rdi);
4179 __ get_vm_result_2(rax, rdi);
4180 __ restore_locals();
4181 #else
4182 __ get_vm_result_2(rax, r15_thread);
4183 #endif
4184
4185 __ pop_ptr(rdx); // restore receiver
4186 __ verify_oop(rdx);
4187 __ load_klass(rdx, rdx);
4188 if (ShenandoahVerifyReadsToFromSpace) {
4189 __ jmp(resolved);
4190 } else {
4191 __ jmpb(resolved);
4192 }
4193
4194 // Get superklass in rax and subklass in rdx
4195 __ bind(quicked);
4196 __ load_klass(rdx, rax);
4197 __ movptr(rax, Address(rcx, rbx,
4198 Address::times_ptr, sizeof(ConstantPool)));
4199
4200 __ bind(resolved);
4201
4202 // Generate subtype check. Blows rcx, rdi
4203 // Superklass in rax. Subklass in rdx.
4204 __ gen_subtype_check(rdx, ok_is_subtype);
4205
4206 // Come here on failure
4207 __ xorl(rax, rax);
4208 __ jmpb(done);
4209 // Come here on success
4210 __ bind(ok_is_subtype);
4211 __ movl(rax, 1);
4212
4213 // Collect counts on whether this test sees NULLs a lot or not.
4214 if (ProfileInterpreter) {
4215 __ jmp(done);
4216 __ bind(is_null);
4217 __ profile_null_seen(rcx);
4218 } else {
4219 __ bind(is_null); // same as 'done'
4220 }
4221 __ bind(done);
4222 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4223 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4224 }
4225
4226
4227 //----------------------------------------------------------------------------------------------------
4228 // Breakpoints
4229 void TemplateTable::_breakpoint() {
4230 // Note: We get here even if we are single stepping..
4231 // jbug insists on setting breakpoints at every bytecode
4232 // even if we are in single step mode.
4233
4234 transition(vtos, vtos);
4235
4236 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4237
4238 // get the unpatched byte code
4239 __ get_method(rarg);
4240 __ call_VM(noreg,
4241 CAST_FROM_FN_PTR(address,
4242 InterpreterRuntime::get_original_bytecode_at),
4243 rarg, rbcp);
4244 __ mov(rbx, rax); // why?
4245
4246 // post the breakpoint event
4247 __ get_method(rarg);
4248 __ call_VM(noreg,
4249 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4250 rarg, rbcp);
4251
4252 // complete the execution of original bytecode
4253 __ dispatch_only_normal(vtos);
4254 }
4255
4256 //-----------------------------------------------------------------------------
4257 // Exceptions
4258
4259 void TemplateTable::athrow() {
4260 transition(atos, vtos);
4261 __ null_check(rax);
4262 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4263 }
4264
4265 //-----------------------------------------------------------------------------
4266 // Synchronization
4267 //
4268 // Note: monitorenter & exit are symmetric routines; which is reflected
4269 // in the assembly code structure as well
4270 //
4271 // Stack layout:
4272 //
4273 // [expressions ] <--- rsp = expression stack top
4274 // ..
4275 // [expressions ]
4276 // [monitor entry] <--- monitor block top = expression stack bot
4277 // ..
4278 // [monitor entry]
4279 // [frame data ] <--- monitor block bot
4280 // ...
4281 // [saved rbp ] <--- rbp
4282 void TemplateTable::monitorenter() {
4283 transition(atos, vtos);
4284
4285 // check for NULL object
4286 __ null_check(rax);
4287
4288 oopDesc::bs()->interpreter_write_barrier(_masm, rax);
4289
4290 const Address monitor_block_top(
4291 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4292 const Address monitor_block_bot(
4293 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4294 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4295
4296 Label allocated;
4297
4298 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4299 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4300 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4301
4302 // initialize entry pointer
4303 __ xorl(rmon, rmon); // points to free slot or NULL
4304
4305 // find a free slot in the monitor block (result in rmon)
4306 {
4307 Label entry, loop, exit;
4308 __ movptr(rtop, monitor_block_top); // points to current entry,
4309 // starting with top-most entry
4310 __ lea(rbot, monitor_block_bot); // points to word before bottom
4311 // of monitor block
4312 __ jmpb_if_possible(entry);
4313
4314 __ bind(loop);
4315 // check if current entry is used
4316 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4317 // if not used then remember entry in rmon
4318 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4319 // check if current entry is for same object
4320 __ shenandoah_lock_check(rtop); // Invariant
4321 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4322 // if same object then stop searching
4323 __ jccb(Assembler::equal, exit);
4324 // otherwise advance to next entry
4325 __ addptr(rtop, entry_size);
4326 __ bind(entry);
4327 // check if bottom reached
4328 __ cmpptr(rtop, rbot);
4329 // if not at bottom then check this entry
4330 __ jcc(Assembler::notEqual, loop);
4331 __ bind(exit);
4332 }
4333
4334 __ testptr(rmon, rmon); // check if a slot has been found
4335 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4336
4337 // allocate one if there's no free slot
4338 {
4339 Label entry, loop;
4340 // 1. compute new pointers // rsp: old expression stack top
4341 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4342 __ subptr(rsp, entry_size); // move expression stack top
4343 __ subptr(rmon, entry_size); // move expression stack bottom
4344 __ mov(rtop, rsp); // set start value for copy loop
4345 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4346 __ jmp(entry);
4347 // 2. move expression stack contents
4348 __ bind(loop);
4349 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4350 // word from old location
4351 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4352 __ addptr(rtop, wordSize); // advance to next word
4353 __ bind(entry);
4354 __ cmpptr(rtop, rmon); // check if bottom reached
4355 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4356 // copy next word
4357 }
4358
4359 // call run-time routine
4360 // rmon: points to monitor entry
4361 __ bind(allocated);
4362
4363 // Increment bcp to point to the next bytecode, so exception
4364 // handling for async. exceptions work correctly.
4365 // The object has already been poped from the stack, so the
4366 // expression stack looks correct.
4367 __ increment(rbcp);
4368
4369 // store object
4370 __ shenandoah_store_addr_check(rax); // Invariant
4371 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4372 __ lock_object(rmon);
4373
4374 // check to make sure this monitor doesn't cause stack overflow after locking
4375 __ save_bcp(); // in case of exception
4376 __ generate_stack_overflow_check(0);
4377
4378 // The bcp has already been incremented. Just need to dispatch to
4379 // next instruction.
4380 __ dispatch_next(vtos);
4381 }
4382
4383 void TemplateTable::monitorexit() {
4384 transition(atos, vtos);
4385
4386 // check for NULL object
4387 __ null_check(rax);
4388
4389 oopDesc::bs()->interpreter_write_barrier(_masm, rax);
4390
4391 const Address monitor_block_top(
4392 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4393 const Address monitor_block_bot(
4394 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4395 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4396
4397 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4398 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4399
4400 Label found;
4401
4402 // find matching slot
4403 {
4404 Label entry, loop;
4405 __ movptr(rtop, monitor_block_top); // points to current entry,
4406 // starting with top-most entry
4407 __ lea(rbot, monitor_block_bot); // points to word before bottom
4408 // of monitor block
4409 __ jmpb_if_possible(entry);
4410
4411 __ bind(loop);
4412 // check if current entry is for same object
4413 __ shenandoah_lock_check(rtop); // Invariant
4414 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4415 // if same object then stop searching
4416 __ jcc(Assembler::equal, found);
4417 // otherwise advance to next entry
4418 __ addptr(rtop, entry_size);
4419 __ bind(entry);
4420 // check if bottom reached
4421 __ cmpptr(rtop, rbot);
4422 // if not at bottom then check this entry
4423 __ jcc(Assembler::notEqual, loop);
4424 }
4425
4426 // error handling. Unlocking was not block-structured
4427 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4428 InterpreterRuntime::throw_illegal_monitor_state_exception));
4429 __ should_not_reach_here();
4430
4431 // call run-time routine
4432 __ bind(found);
4433 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4434 __ unlock_object(rtop);
4435 __ pop_ptr(rax); // discard object
4436 }
4437
4438 // Wide instructions
4439 void TemplateTable::wide() {
4440 transition(vtos, vtos);
4441 __ load_unsigned_byte(rbx, at_bcp(1));
4442 ExternalAddress wtable((address)Interpreter::_wentry_point);
4443 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4444 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4445 }
4446
4447 // Multi arrays
4448 void TemplateTable::multianewarray() {
4449 transition(vtos, atos);
4450
4451 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4452 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4453 // last dim is on top of stack; we want address of first one:
4454 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4455 // the latter wordSize to point to the beginning of the array.
4456 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4457 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4458 __ load_unsigned_byte(rbx, at_bcp(3));
4459 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4460 }