New src/cpu/x86/vm/templateTable_x86.cpp (original) (raw)
1 /*
2 * Copyright (c) 1997, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 /
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/interp_masm.hpp"
30 #include "interpreter/templateTable.hpp"
31 #include "memory/universe.inline.hpp"
32 #include "oops/methodData.hpp"
33 #include "oops/objArrayKlass.hpp"
34 #include "oops/oop.inline.hpp"
35 #include "prims/methodHandles.hpp"
36 #include "runtime/sharedRuntime.hpp"
37 #include "runtime/stubRoutines.hpp"
38 #include "runtime/synchronizer.hpp"
39 #include "utilities/macros.hpp"
40
41 #ifndef CC_INTERP
42
43 #define __ _masm->
44
45 // Global Register Names
46 Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
47 Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
48
49 // Platform-dependent initialization
50 void TemplateTable::pd_initialize() {
51 // No x86 specific initialization
52 }
53
54 // Address Computation: local variables
55 static inline Address iaddress(int n) {
56 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
57 }
58
59 static inline Address laddress(int n) {
60 return iaddress(n + 1);
61 }
62
63 #ifndef _LP64
64 static inline Address haddress(int n) {
65 return iaddress(n + 0);
66 }
67 #endif
68
69 static inline Address faddress(int n) {
70 return iaddress(n);
71 }
72
73 static inline Address daddress(int n) {
74 return laddress(n);
75 }
76
77 static inline Address aaddress(int n) {
78 return iaddress(n);
79 }
80
81 static inline Address iaddress(Register r) {
82 return Address(rlocals, r, Address::times_ptr);
83 }
84
85 static inline Address laddress(Register r) {
86 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
87 }
88
89 #ifndef _LP64
90 static inline Address haddress(Register r) {
91 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
92 }
93 #endif
94
95 static inline Address faddress(Register r) {
96 return iaddress(r);
97 }
98
99 static inline Address daddress(Register r) {
100 return laddress(r);
101 }
102
103 static inline Address aaddress(Register r) {
104 return iaddress(r);
105 }
106
107
108 // expression stack
109 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
110 // data beyond the rsp which is potentially unsafe in an MT environment;
111 // an interrupt may overwrite that data.)
112 static inline Address at_rsp () {
113 return Address(rsp, 0);
114 }
115
116 // At top of Java expression stack which may be different than esp(). It
117 // isn't for category 1 objects.
118 static inline Address at_tos () {
119 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
120 }
121
122 static inline Address at_tos_p1() {
123 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
124 }
125
126 static inline Address at_tos_p2() {
127 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
128 }
129
130 // Condition conversion
131 static Assembler::Condition j_not(TemplateTable::Condition cc) {
132 switch (cc) {
133 case TemplateTable::equal : return Assembler::notEqual;
134 case TemplateTable::not_equal : return Assembler::equal;
135 case TemplateTable::less : return Assembler::greaterEqual;
136 case TemplateTable::less_equal : return Assembler::greater;
137 case TemplateTable::greater : return Assembler::lessEqual;
138 case TemplateTable::greater_equal: return Assembler::less;
139 }
140 ShouldNotReachHere();
141 return Assembler::zero;
142 }
143
144
145
146 // Miscelaneous helper routines
147 // Store an oop (or NULL) at the address described by obj.
148 // If val == noreg this means store a NULL
149
150
151 static void do_oop_store(InterpreterMacroAssembler masm,
152 Address obj,
153 Register val,
154 BarrierSet::Name barrier,
155 bool precise) {
156 assert(val == noreg || val == rax, "parameter is just for looks");
157 switch (barrier) {
158 #if INCLUDE_ALL_GCS
159 case BarrierSet::G1SATBCTLogging:
160 {
161 // flatten object address if needed
162 // We do it regardless of precise because we need the registers
163 if (obj.index() == noreg && obj.disp() == 0) {
164 if (obj.base() != rdx) {
165 __ movptr(rdx, obj.base());
166 }
167 } else {
168 __ lea(rdx, obj);
169 }
170
171 Register rtmp = LP64_ONLY(r8) NOT_LP64(rsi);
172 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
173
174 NOT_LP64(_ get_thread(rcx));
175 NOT_LP64(__ save_bcp());
176
177 __ g1_write_barrier_pre(rdx /* obj /,
178 rbx / pre_val /,
179 rthread / thread /,
180 rtmp / tmp /,
181 val != noreg / tosca_live /,
182 false / expand_call /);
183 if (val == noreg) {
184 __ store_heap_oop_null(Address(rdx, 0));
185 } else {
186 // G1 barrier needs uncompressed oop for region cross check.
187 Register new_val = val;
188 if (UseCompressedOops) {
189 new_val = rbx;
190 __ movptr(new_val, val);
191 }
192 __ store_heap_oop(Address(rdx, 0), val);
193 __ g1_write_barrier_post(rdx / store_adr /,
194 new_val / new_val /,
195 rthread / thread /,
196 rtmp / tmp /,
197 rbx / tmp2 /);
198 }
199 NOT_LP64( __ restore_bcp());
200 }
201 break;
202 #endif // INCLUDE_ALL_GCS
203 case BarrierSet::CardTableModRef:
204 case BarrierSet::CardTableExtension:
205 {
206 if (val == noreg) {
207 __ store_heap_oop_null(obj);
208 } else {
209 __ store_heap_oop(obj, val);
210 // flatten object address if needed
211 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
212 __ store_check(obj.base());
213 } else {
214 __ lea(rdx, obj);
215 __ store_check(rdx);
216 }
217 }
218 }
219 break;
220 case BarrierSet::ModRef:
221 if (val == noreg) {
222 __ store_heap_oop_null(obj);
223 } else {
224 __ store_heap_oop(obj, val);
225 }
226 break;
227 default :
228 ShouldNotReachHere();
229
230 }
231 }
232
233 Address TemplateTable::at_bcp(int offset) {
234 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
235 return Address(rbcp, offset);
236 }
237
238
239 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
240 Register temp_reg, bool load_bc_into_bc_reg/=true*/,
241 int byte_no) {
242 if (!RewriteBytecodes) return;
243 Label L_patch_done;
244
245 switch (bc) {
246 case Bytecodes::_fast_aputfield:
247 case Bytecodes::_fast_bputfield:
248 case Bytecodes::_fast_cputfield:
249 case Bytecodes::_fast_dputfield:
250 case Bytecodes::_fast_fputfield:
251 case Bytecodes::_fast_iputfield:
252 case Bytecodes::_fast_lputfield:
253 case Bytecodes::_fast_sputfield:
254 {
255 // We skip bytecode quickening for putfield instructions when
256 // the put_code written to the constant pool cache is zero.
257 // This is required so that every execution of this instruction
258 // calls out to InterpreterRuntime::resolve_get_put to do
259 // additional, required work.
260 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
261 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
262 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
263 __ movl(bc_reg, bc);
264 __ cmpl(temp_reg, (int) 0);
265 __ jcc(Assembler::zero, L_patch_done); // don't patch
266 }
267 break;
268 default:
269 assert(byte_no == -1, "sanity");
270 // the pair bytecodes have already done the load.
271 if (load_bc_into_bc_reg) {
272 __ movl(bc_reg, bc);
273 }
274 }
275
276 if (JvmtiExport::can_post_breakpoint()) {
277 Label L_fast_patch;
278 // if a breakpoint is present we can't rewrite the stream directly
279 __ movzbl(temp_reg, at_bcp(0));
280 __ cmpl(temp_reg, Bytecodes::_breakpoint);
281 __ jcc(Assembler::notEqual, L_fast_patch);
282 __ get_method(temp_reg);
283 // Let breakpoint table handling rewrite to quicker bytecode
284 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
285 #ifndef ASSERT
286 __ jmpb(L_patch_done);
287 #else
288 __ jmp(L_patch_done);
289 #endif
290 __ bind(L_fast_patch);
291 }
292
293 #ifdef ASSERT
294 Label L_okay;
295 __ load_unsigned_byte(temp_reg, at_bcp(0));
296 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
297 __ jcc(Assembler::equal, L_okay);
298 __ cmpl(temp_reg, bc_reg);
299 __ jcc(Assembler::equal, L_okay);
300 __ stop("patching the wrong bytecode");
301 __ bind(L_okay);
302 #endif
303
304 // patch bytecode
305 __ movb(at_bcp(0), bc_reg);
306 __ bind(L_patch_done);
307 }
308 // Individual instructions
309
310
311 void TemplateTable::nop() {
312 transition(vtos, vtos);
313 // nothing to do
314 }
315
316 void TemplateTable::shouldnotreachhere() {
317 transition(vtos, vtos);
318 __ stop("shouldnotreachhere bytecode");
319 }
320
321 void TemplateTable::aconst_null() {
322 transition(vtos, atos);
323 __ xorl(rax, rax);
324 }
325
326 void TemplateTable::iconst(int value) {
327 transition(vtos, itos);
328 if (value == 0) {
329 __ xorl(rax, rax);
330 } else {
331 __ movl(rax, value);
332 }
333 }
334
335 void TemplateTable::lconst(int value) {
336 transition(vtos, ltos);
337 if (value == 0) {
338 __ xorl(rax, rax);
339 } else {
340 __ movl(rax, value);
341 }
342 #ifndef _LP64
343 assert(value >= 0, "check this code");
344 __ xorptr(rdx, rdx);
345 #endif
346 }
347
348
349
350 void TemplateTable::fconst(int value) {
351 transition(vtos, ftos);
352 if (UseSSE >= 1) {
353 static float one = 1.0f, two = 2.0f;
354 switch (value) {
355 case 0:
356 __ xorps(xmm0, xmm0);
357 break;
358 case 1:
359 __ movflt(xmm0, ExternalAddress((address) &one));
360 break;
361 case 2:
362 __ movflt(xmm0, ExternalAddress((address) &two));
363 break;
364 default:
365 ShouldNotReachHere();
366 break;
367 }
368 } else {
369 #ifdef _LP64
370 ShouldNotReachHere();
371 #else
372 if (value == 0) { __ fldz();
373 } else if (value == 1) { __ fld1();
374 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
375 } else { ShouldNotReachHere();
376 }
377 #endif // LP64
378 }
379 }
380
381 void TemplateTable::dconst(int value) {
382 transition(vtos, dtos);
383 if (UseSSE >= 2) {
384 static double one = 1.0;
385 switch (value) {
386 case 0:
387 __ xorpd(xmm0, xmm0);
388 break;
389 case 1:
390 __ movdbl(xmm0, ExternalAddress((address) &one));
391 break;
392 default:
393 ShouldNotReachHere();
394 break;
395 }
396 } else {
397 #ifdef LP64
398 ShouldNotReachHere();
399 #else
400 if (value == 0) { __ fldz();
401 } else if (value == 1) { __ fld1();
402 } else { ShouldNotReachHere();
403 }
404 #endif
405 }
406 }
407
408 void TemplateTable::bipush() {
409 transition(vtos, itos);
410 __ load_signed_byte(rax, at_bcp(1));
411 }
412
413 void TemplateTable::sipush() {
414 transition(vtos, itos);
415 __ load_unsigned_short(rax, at_bcp(1));
416 __ bswapl(rax);
417 __ sarl(rax, 16);
418 }
419
420 void TemplateTable::ldc(bool wide) {
421 transition(vtos, vtos);
422 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
423 Label call_ldc, notFloat, notClass, Done;
424
425 if (wide) {
426 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
427 } else {
428 __ load_unsigned_byte(rbx, at_bcp(1));
429 }
430
431 __ get_cpool_and_tags(rcx, rax);
432 const int base_offset = ConstantPool::header_size() * wordSize;
433 const int tags_offset = Array::base_offset_in_bytes();
434
435 // get type
436 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
437
438 // unresolved class - get the resolved class
439 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
440 __ jccb(Assembler::equal, call_ldc);
441
442 // unresolved class in error state - call into runtime to throw the error
443 // from the first resolution attempt
444 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
445 __ jccb(Assembler::equal, call_ldc);
446
447 // resolved class - need to call vm to get java mirror of the class
448 __ cmpl(rdx, JVM_CONSTANT_Class);
449 __ jcc(Assembler::notEqual, notClass);
450
451 __ bind(call_ldc);
452
453 __ movl(rarg, wide);
454 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
455
456 __ push(atos);
457 __ jmp(Done);
458
459 __ bind(notClass);
460 __ cmpl(rdx, JVM_CONSTANT_Float);
461 __ jccb(Assembler::notEqual, notFloat);
462
463 // ftos
464 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
465 __ push(ftos);
466 __ jmp(Done);
467
468 __ bind(notFloat);
469 #ifdef ASSERT
470 {
471 Label L;
472 __ cmpl(rdx, JVM_CONSTANT_Integer);
473 __ jcc(Assembler::equal, L);
474 // String and Object are rewritten to fast_aldc
475 __ stop("unexpected tag type in ldc");
476 __ bind(L);
477 }
478 #endif
479 // itos JVM_CONSTANT_Integer only
480 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
481 __ push(itos);
482 __ bind(Done);
483 }
484
485 // Fast path for caching oop constants.
486 void TemplateTable::fast_aldc(bool wide) {
487 transition(vtos, atos);
488
489 Register result = rax;
490 Register tmp = rdx;
491 int index_size = wide ? sizeof(u2) : sizeof(u1);
492
493 Label resolved;
494
495 // We are resolved if the resolved reference cache entry contains a
496 // non-null object (String, MethodType, etc.)
497 assert_different_registers(result, tmp);
498 __ get_cache_index_at_bcp(tmp, 1, index_size);
499 __ load_resolved_reference_at_index(result, tmp);
500 __ testl(result, result);
501 __ jcc(Assembler::notZero, resolved);
502
503 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
504
505 // first time invocation - must resolve first
506 __ movl(tmp, (int)bytecode());
507 __ call_VM(result, entry, tmp);
508
509 __ bind(resolved);
510
511 if (VerifyOops) {
512 __ verify_oop(result);
513 }
514 }
515
516 void TemplateTable::ldc2_w() {
517 transition(vtos, vtos);
518 Label Long, Done;
519 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
520
521 __ get_cpool_and_tags(rcx, rax);
522 const int base_offset = ConstantPool::header_size() * wordSize;
523 const int tags_offset = Array::base_offset_in_bytes();
524
525 // get type
526 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset),
527 JVM_CONSTANT_Double);
528 __ jccb(Assembler::notEqual, Long);
529
530 // dtos
531 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
532 __ push(dtos);
533
534 __ jmpb(Done);
535 __ bind(Long);
536
537 // ltos
538 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
539 NOT_LP64( movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
540 __ push(ltos);
541
542 __ bind(Done);
543 }
544
545 void TemplateTable::locals_index(Register reg, int offset) {
546 __ load_unsigned_byte(reg, at_bcp(offset));
547 __ negptr(reg);
548 }
549
550 void TemplateTable::iload() {
551 iload_internal();
552 }
553
554 void TemplateTable::nofast_iload() {
555 iload_internal(may_not_rewrite);
556 }
557
558 void TemplateTable::iload_internal(RewriteControl rc) {
559 transition(vtos, itos);
560 if (RewriteFrequentPairs && rc == may_rewrite) {
561 Label rewrite, done;
562 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
563 LP64_ONLY(assert(rbx != bc, "register damaged"));
564
565 // get next byte
566 __ load_unsigned_byte(rbx,
567 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
568 // if iload, wait to rewrite to iload2. We only want to rewrite the
569 // last two iloads in a pair. Comparing against fast_iload means that
570 // the next bytecode is neither an iload or a caload, and therefore
571 // an iload pair.
572 __ cmpl(rbx, Bytecodes::iload);
573 __ jcc(Assembler::equal, done);
574
575 __ cmpl(rbx, Bytecodes::fast_iload);
576 __ movl(bc, Bytecodes::fast_iload2);
577
578 __ jccb(Assembler::equal, rewrite);
579
580 // if caload, rewrite to fast_icaload
581 __ cmpl(rbx, Bytecodes::caload);
582 __ movl(bc, Bytecodes::fast_icaload);
583 __ jccb(Assembler::equal, rewrite);
584
585 // rewrite so iload doesn't check again.
586 __ movl(bc, Bytecodes::fast_iload);
587
588 // rewrite
589 // bc: fast bytecode
590 __ bind(rewrite);
591 patch_bytecode(Bytecodes::iload, bc, rbx, false);
592 __ bind(done);
593 }
594
595 // Get the local value into tos
596 locals_index(rbx);
597 __ movl(rax, iaddress(rbx));
598 }
599
600 void TemplateTable::fast_iload2() {
601 transition(vtos, itos);
602 locals_index(rbx);
603 __ movl(rax, iaddress(rbx));
604 __ push(itos);
605 locals_index(rbx, 3);
606 __ movl(rax, iaddress(rbx));
607 }
608
609 void TemplateTable::fast_iload() {
610 transition(vtos, itos);
611 locals_index(rbx);
612 __ movl(rax, iaddress(rbx));
613 }
614
615 void TemplateTable::lload() {
616 transition(vtos, ltos);
617 locals_index(rbx);
618 __ movptr(rax, laddress(rbx));
619 NOT_LP64( movl(rdx, haddress(rbx)));
620 }
621
622 void TemplateTable::fload() {
623 transition(vtos, ftos);
624 locals_index(rbx);
625 __ load_float(faddress(rbx));
626 }
627
628 void TemplateTable::dload() {
629 transition(vtos, dtos);
630 locals_index(rbx);
631 __ load_double(daddress(rbx));
632 }
633
634 void TemplateTable::aload() {
635 transition(vtos, atos);
636 locals_index(rbx);
637 __ movptr(rax, aaddress(rbx));
638 }
639
640 void TemplateTable::locals_index_wide(Register reg) {
641 __ load_unsigned_short(reg, at_bcp(2));
642 __ bswapl(reg);
643 __ shrl(reg, 16);
644 __ negptr(reg);
645 }
646
647 void TemplateTable::wide_iload() {
648 transition(vtos, itos);
649 locals_index_wide(rbx);
650 __ movl(rax, iaddress(rbx));
651 }
652
653 void TemplateTable::wide_lload() {
654 transition(vtos, ltos);
655 locals_index_wide(rbx);
656 __ movptr(rax, laddress(rbx));
657 NOT_LP64( movl(rdx, haddress(rbx)));
658 }
659
660 void TemplateTable::wide_fload() {
661 transition(vtos, ftos);
662 locals_index_wide(rbx);
663 __ load_float(faddress(rbx));
664 }
665
666 void TemplateTable::wide_dload() {
667 transition(vtos, dtos);
668 locals_index_wide(rbx);
669 __ load_double(daddress(rbx));
670 }
671
672 void TemplateTable::wide_aload() {
673 transition(vtos, atos);
674 locals_index_wide(rbx);
675 __ movptr(rax, aaddress(rbx));
676 }
677
678 void TemplateTable::index_check(Register array, Register index) {
679 // Pop ptr into array
680 __ pop_ptr(array);
681 index_check_without_pop(array, index);
682 }
683
684 void TemplateTable::index_check_without_pop(Register array, Register index) {
685 // destroys rbx
686 // check array
687 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
688 // sign extend index for use by indexed load
689 __ movl2ptr(index, index);
690 // check index
691 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
692 if (index != rbx) {
693 // ??? convention: move aberrant index into rbx for exception message
694 assert(rbx != array, "different registers");
695 __ movl(rbx, index);
696 }
697 __ jump_cc(Assembler::aboveEqual,
698 ExternalAddress(Interpreter::throw_ArrayIndexOutOfBoundsException_entry));
699 }
700
701
702 void TemplateTable::iaload() {
703 transition(itos, itos);
704 // rax: index
705 // rdx: array
706 index_check(rdx, rax); // kills rbx
707 __ movl(rax, Address(rdx, rax,
708 Address::times_4,
709 arrayOopDesc::base_offset_in_bytes(T_INT)));
710 }
711
712 void TemplateTable::laload() {
713 transition(itos, ltos);
714 // rax: index
715 // rdx: array
716 index_check(rdx, rax); // kills rbx
717 NOT_LP64( mov(rbx, rax));
718 // rbx,: index
719 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
720 NOT_LP64( movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
721 }
722
723
724
725 void TemplateTable::faload() {
726 transition(itos, ftos);
727 // rax: index
728 // rdx: array
729 index_check(rdx, rax); // kills rbx
730 __ load_float(Address(rdx, rax,
731 Address::times_4,
732 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
733 }
734
735 void TemplateTable::daload() {
736 transition(itos, dtos);
737 // rax: index
738 // rdx: array
739 index_check(rdx, rax); // kills rbx
740 __ load_double(Address(rdx, rax,
741 Address::times_8,
742 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
743 }
744
745 void TemplateTable::aaload() {
746 transition(itos, atos);
747 // rax: index
748 // rdx: array
749 index_check(rdx, rax); // kills rbx
750 __ load_heap_oop(rax, Address(rdx, rax,
751 UseCompressedOops ? Address::times_4 : Address::times_ptr,
752 arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
753 }
754
755 void TemplateTable::baload() {
756 transition(itos, itos);
757 // rax: index
758 // rdx: array
759 index_check(rdx, rax); // kills rbx
760 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
761 }
762
763 void TemplateTable::caload() {
764 transition(itos, itos);
765 // rax: index
766 // rdx: array
767 index_check(rdx, rax); // kills rbx
768 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
769 }
770
771 // iload followed by caload frequent pair
772 void TemplateTable::fast_icaload() {
773 transition(vtos, itos);
774 // load index out of locals
775 locals_index(rbx);
776 __ movl(rax, iaddress(rbx));
777
778 // rax: index
779 // rdx: array
780 index_check(rdx, rax); // kills rbx
781 __ load_unsigned_short(rax,
782 Address(rdx, rax,
783 Address::times_2,
784 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
785 }
786
787
788 void TemplateTable::saload() {
789 transition(itos, itos);
790 // rax: index
791 // rdx: array
792 index_check(rdx, rax); // kills rbx
793 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
794 }
795
796 void TemplateTable::iload(int n) {
797 transition(vtos, itos);
798 __ movl(rax, iaddress(n));
799 }
800
801 void TemplateTable::lload(int n) {
802 transition(vtos, ltos);
803 __ movptr(rax, laddress(n));
804 NOT_LP64( movptr(rdx, haddress(n)));
805 }
806
807 void TemplateTable::fload(int n) {
808 transition(vtos, ftos);
809 __ load_float(faddress(n));
810 }
811
812 void TemplateTable::dload(int n) {
813 transition(vtos, dtos);
814 __ load_double(daddress(n));
815 }
816
817 void TemplateTable::aload(int n) {
818 transition(vtos, atos);
819 __ movptr(rax, aaddress(n));
820 }
821
822 void TemplateTable::aload_0() {
823 aload_0_internal();
824 }
825
826 void TemplateTable::nofast_aload_0() {
827 aload_0_internal(may_not_rewrite);
828 }
829
830 void TemplateTable::aload_0_internal(RewriteControl rc) {
831 transition(vtos, atos);
832 // According to bytecode histograms, the pairs:
833 //
834 // _aload_0, _fast_igetfield
835 // _aload_0, _fast_agetfield
836 // _aload_0, _fast_fgetfield
837 //
838 // occur frequently. If RewriteFrequentPairs is set, the (slow)
839 // _aload_0 bytecode checks if the next bytecode is either
840 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
841 // rewrites the current bytecode into a pair bytecode; otherwise it
842 // rewrites the current bytecode into _fast_aload_0 that doesn't do
843 // the pair check anymore.
844 //
845 // Note: If the next bytecode is _getfield, the rewrite must be
846 // delayed, otherwise we may miss an opportunity for a pair.
847 //
848 // Also rewrite frequent pairs
849 // aload_0, aload_1
850 // aload_0, iload_1
851 // These bytecodes with a small amount of code are most profitable
852 // to rewrite
853 if (RewriteFrequentPairs && rc == may_rewrite) {
854 Label rewrite, done;
855
856 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
857 LP64_ONLY(assert(rbx != bc, "register damaged"));
858
859 // get next byte
860 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
861
862 // do actual aload_0
863 aload(0);
864
865 // if _getfield then wait with rewrite
866 __ cmpl(rbx, Bytecodes::_getfield);
867 __ jcc(Assembler::equal, done);
868
869 // if _igetfield then reqrite to _fast_iaccess_0
870 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
871 __ cmpl(rbx, Bytecodes::_fast_igetfield);
872 __ movl(bc, Bytecodes::_fast_iaccess_0);
873 __ jccb(Assembler::equal, rewrite);
874
875 // if _agetfield then reqrite to _fast_aaccess_0
876 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
877 __ cmpl(rbx, Bytecodes::_fast_agetfield);
878 __ movl(bc, Bytecodes::_fast_aaccess_0);
879 __ jccb(Assembler::equal, rewrite);
880
881 // if _fgetfield then reqrite to _fast_faccess_0
882 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::aload_0, "fix bytecode definition");
883 __ cmpl(rbx, Bytecodes::fast_fgetfield);
884 __ movl(bc, Bytecodes::fast_faccess_0);
885 __ jccb(Assembler::equal, rewrite);
886
887 // else rewrite to fast_aload0
888 assert(Bytecodes::java_code(Bytecodes::fast_aload_0) == Bytecodes::aload_0, "fix bytecode definition");
889 __ movl(bc, Bytecodes::fast_aload_0);
890
891 // rewrite
892 // bc: fast bytecode
893 __ bind(rewrite);
894 patch_bytecode(Bytecodes::aload_0, bc, rbx, false);
895
896 __ bind(done);
897 } else {
898 aload(0);
899 }
900 }
901
902 void TemplateTable::istore() {
903 transition(itos, vtos);
904 locals_index(rbx);
905 __ movl(iaddress(rbx), rax);
906 }
907
908
909 void TemplateTable::lstore() {
910 transition(ltos, vtos);
911 locals_index(rbx);
912 __ movptr(laddress(rbx), rax);
913 NOT_LP64( movptr(haddress(rbx), rdx));
914 }
915
916 void TemplateTable::fstore() {
917 transition(ftos, vtos);
918 locals_index(rbx);
919 __ store_float(faddress(rbx));
920 }
921
922 void TemplateTable::dstore() {
923 transition(dtos, vtos);
924 locals_index(rbx);
925 __ store_double(daddress(rbx));
926 }
927
928 void TemplateTable::astore() {
929 transition(vtos, vtos);
930 __ pop_ptr(rax);
931 locals_index(rbx);
932 __ movptr(aaddress(rbx), rax);
933 }
934
935 void TemplateTable::wide_istore() {
936 transition(vtos, vtos);
937 __ pop_i();
938 locals_index_wide(rbx);
939 __ movl(iaddress(rbx), rax);
940 }
941
942 void TemplateTable::wide_lstore() {
943 transition(vtos, vtos);
944 NOT_LP64( pop_l(rax, rdx));
945 LP64_ONLY( pop_l());
946 locals_index_wide(rbx);
947 __ movptr(laddress(rbx), rax);
948 NOT_LP64( movl(haddress(rbx), rdx));
949 }
950
951 void TemplateTable::wide_fstore() {
952 #ifdef LP64
953 transition(vtos, vtos);
954 __ pop_f(xmm0);
955 locals_index_wide(rbx);
956 __ movflt(faddress(rbx), xmm0);
957 #else
958 wide_istore();
959 #endif
960 }
961
962 void TemplateTable::wide_dstore() {
963 #ifdef LP64
964 transition(vtos, vtos);
965 __ pop_d(xmm0);
966 locals_index_wide(rbx);
967 __ movdbl(daddress(rbx), xmm0);
968 #else
969 wide_lstore();
970 #endif
971 }
972
973 void TemplateTable::wide_astore() {
974 transition(vtos, vtos);
975 __ pop_ptr(rax);
976 locals_index_wide(rbx);
977 __ movptr(aaddress(rbx), rax);
978 }
979
980 void TemplateTable::iastore() {
981 transition(itos, vtos);
982 __ pop_i(rbx);
983 // rax: value
984 // rbx: index
985 // rdx: array
986 index_check(rdx, rbx); // prefer index in rbx
987 __ movl(Address(rdx, rbx,
988 Address::times_4,
989 arrayOopDesc::base_offset_in_bytes(T_INT)),
990 rax);
991 }
992
993 void TemplateTable::lastore() {
994 transition(ltos, vtos);
995 __ pop_i(rbx);
996 // rax,: low(value)
997 // rcx: array
998 // rdx: high(value)
999 index_check(rcx, rbx); // prefer index in rbx,
1000 // rbx,: index
1001 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1002 NOT_LP64( movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1003 }
1004
1005
1006 void TemplateTable::fastore() {
1007 transition(ftos, vtos);
1008 __ pop_i(rbx);
1009 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1010 // rbx: index
1011 // rdx: array
1012 index_check(rdx, rbx); // prefer index in rbx
1013 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1014 }
1015
1016 void TemplateTable::dastore() {
1017 transition(dtos, vtos);
1018 __ pop_i(rbx);
1019 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1020 // rbx: index
1021 // rdx: array
1022 index_check(rdx, rbx); // prefer index in rbx
1023 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1024 }
1025
1026 void TemplateTable::aastore() {
1027 Label is_null, ok_is_subtype, done;
1028 transition(vtos, vtos);
1029 // stack: ..., array, index, value
1030 __ movptr(rax, at_tos()); // value
1031 __ movl(rcx, at_tos_p1()); // index
1032 __ movptr(rdx, at_tos_p2()); // array
1033
1034 Address element_address(rdx, rcx,
1035 UseCompressedOops? Address::times_4 : Address::times_ptr,
1036 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1037
1038 index_check_without_pop(rdx, rcx); // kills rbx
1039 __ testptr(rax, rax);
1040 __ jcc(Assembler::zero, is_null);
1041
1042 // Move subklass into rbx
1043 __ load_klass(rbx, rax);
1044 // Move superklass into rax
1045 __ load_klass(rax, rdx);
1046 __ movptr(rax, Address(rax,
1047 ObjArrayKlass::element_klass_offset()));
1048 // Compress array + indexoopSize + 12 into a single register. Frees rcx.
1049 __ lea(rdx, element_address);
1050
1051 // Generate subtype check. Blows rcx, rdi
1052 // Superklass in rax. Subklass in rbx.
1053 __ gen_subtype_check(rbx, ok_is_subtype);
1054
1055 // Come here on failure
1056 // object is at TOS
1057 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1058
1059 // Come here on success
1060 __ bind(ok_is_subtype);
1061
1062 // Get the value we will store
1063 __ movptr(rax, at_tos());
1064 // Now store using the appropriate barrier
1065 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
1066 __ jmp(done);
1067
1068 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1069 __ bind(is_null);
1070 __ profile_null_seen(rbx);
1071
1072 // Store a NULL
1073 do_oop_store(masm, element_address, noreg, bs->kind(), true);
1074
1075 // Pop stack arguments
1076 __ bind(done);
1077 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1078 }
1079
1080 void TemplateTable::bastore() {
1081 transition(itos, vtos);
1082 __ pop_i(rbx);
1083 // rax: value
1084 // rbx: index
1085 // rdx: array
1086 index_check(rdx, rbx); // prefer index in rbx
1087 __ movb(Address(rdx, rbx,
1088 Address::times_1,
1089 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1090 rax);
1091 }
1092
1093 void TemplateTable::castore() {
1094 transition(itos, vtos);
1095 __ pop_i(rbx);
1096 // rax: value
1097 // rbx: index
1098 // rdx: array
1099 index_check(rdx, rbx); // prefer index in rbx
1100 __ movw(Address(rdx, rbx,
1101 Address::times_2,
1102 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1103 rax);
1104 }
1105
1106
1107 void TemplateTable::sastore() {
1108 castore();
1109 }
1110
1111 void TemplateTable::istore(int n) {
1112 transition(itos, vtos);
1113 __ movl(iaddress(n), rax);
1114 }
1115
1116 void TemplateTable::lstore(int n) {
1117 transition(ltos, vtos);
1118 __ movptr(laddress(n), rax);
1119 NOT_LP64( movptr(haddress(n), rdx));
1120 }
1121
1122 void TemplateTable::fstore(int n) {
1123 transition(ftos, vtos);
1124 __ store_float(faddress(n));
1125 }
1126
1127 void TemplateTable::dstore(int n) {
1128 transition(dtos, vtos);
1129 __ store_double(daddress(n));
1130 }
1131
1132
1133 void TemplateTable::astore(int n) {
1134 transition(vtos, vtos);
1135 __ pop_ptr(rax);
1136 __ movptr(aaddress(n), rax);
1137 }
1138
1139 void TemplateTable::pop() {
1140 transition(vtos, vtos);
1141 __ addptr(rsp, Interpreter::stackElementSize);
1142 }
1143
1144 void TemplateTable::pop2() {
1145 transition(vtos, vtos);
1146 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1147 }
1148
1149
1150 void TemplateTable::dup() {
1151 transition(vtos, vtos);
1152 __ load_ptr(0, rax);
1153 __ push_ptr(rax);
1154 // stack: ..., a, a
1155 }
1156
1157 void TemplateTable::dup_x1() {
1158 transition(vtos, vtos);
1159 // stack: ..., a, b
1160 __ load_ptr( 0, rax); // load b
1161 __ load_ptr( 1, rcx); // load a
1162 __ store_ptr(1, rax); // store b
1163 __ store_ptr(0, rcx); // store a
1164 __ push_ptr(rax); // push b
1165 // stack: ..., b, a, b
1166 }
1167
1168 void TemplateTable::dup_x2() {
1169 transition(vtos, vtos);
1170 // stack: ..., a, b, c
1171 __ load_ptr( 0, rax); // load c
1172 __ load_ptr( 2, rcx); // load a
1173 __ store_ptr(2, rax); // store c in a
1174 __ push_ptr(rax); // push c
1175 // stack: ..., c, b, c, c
1176 __ load_ptr( 2, rax); // load b
1177 __ store_ptr(2, rcx); // store a in b
1178 // stack: ..., c, a, c, c
1179 __ store_ptr(1, rax); // store b in c
1180 // stack: ..., c, a, b, c
1181 }
1182
1183 void TemplateTable::dup2() {
1184 transition(vtos, vtos);
1185 // stack: ..., a, b
1186 __ load_ptr(1, rax); // load a
1187 __ push_ptr(rax); // push a
1188 __ load_ptr(1, rax); // load b
1189 __ push_ptr(rax); // push b
1190 // stack: ..., a, b, a, b
1191 }
1192
1193
1194 void TemplateTable::dup2_x1() {
1195 transition(vtos, vtos);
1196 // stack: ..., a, b, c
1197 __ load_ptr( 0, rcx); // load c
1198 __ load_ptr( 1, rax); // load b
1199 __ push_ptr(rax); // push b
1200 __ push_ptr(rcx); // push c
1201 // stack: ..., a, b, c, b, c
1202 __ store_ptr(3, rcx); // store c in b
1203 // stack: ..., a, c, c, b, c
1204 __ load_ptr( 4, rcx); // load a
1205 __ store_ptr(2, rcx); // store a in 2nd c
1206 // stack: ..., a, c, a, b, c
1207 __ store_ptr(4, rax); // store b in a
1208 // stack: ..., b, c, a, b, c
1209 }
1210
1211 void TemplateTable::dup2_x2() {
1212 transition(vtos, vtos);
1213 // stack: ..., a, b, c, d
1214 __ load_ptr( 0, rcx); // load d
1215 __ load_ptr( 1, rax); // load c
1216 __ push_ptr(rax); // push c
1217 __ push_ptr(rcx); // push d
1218 // stack: ..., a, b, c, d, c, d
1219 __ load_ptr( 4, rax); // load b
1220 __ store_ptr(2, rax); // store b in d
1221 __ store_ptr(4, rcx); // store d in b
1222 // stack: ..., a, d, c, b, c, d
1223 __ load_ptr( 5, rcx); // load a
1224 __ load_ptr( 3, rax); // load c
1225 __ store_ptr(3, rcx); // store a in c
1226 __ store_ptr(5, rax); // store c in a
1227 // stack: ..., c, d, a, b, c, d
1228 }
1229
1230 void TemplateTable::swap() {
1231 transition(vtos, vtos);
1232 // stack: ..., a, b
1233 __ load_ptr( 1, rcx); // load a
1234 __ load_ptr( 0, rax); // load b
1235 __ store_ptr(0, rcx); // store a in b
1236 __ store_ptr(1, rax); // store b in a
1237 // stack: ..., b, a
1238 }
1239
1240 void TemplateTable::iop2(Operation op) {
1241 transition(itos, itos);
1242 switch (op) {
1243 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1244 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1245 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1246 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1247 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1248 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1249 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1250 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1251 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1252 default : ShouldNotReachHere();
1253 }
1254 }
1255
1256 void TemplateTable::lop2(Operation op) {
1257 transition(ltos, ltos);
1258 #ifdef _LP64
1259 switch (op) {
1260 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1261 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1262 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1263 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1264 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1265 default : ShouldNotReachHere();
1266 }
1267 #else
1268 __ pop_l(rbx, rcx);
1269 switch (op) {
1270 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1271 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1272 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1273 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1274 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1275 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1276 default : ShouldNotReachHere();
1277 }
1278 #endif
1279 }
1280
1281 void TemplateTable::idiv() {
1282 transition(itos, itos);
1283 __ movl(rcx, rax);
1284 __ pop_i(rax);
1285 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1286 // they are not equal, one could do a normal division (no correction
1287 // needed), which may speed up this implementation for the common case.
1288 // (see also JVM spec., p.243 & p.271)
1289 __ corrected_idivl(rcx);
1290 }
1291
1292 void TemplateTable::irem() {
1293 transition(itos, itos);
1294 __ movl(rcx, rax);
1295 __ pop_i(rax);
1296 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1297 // they are not equal, one could do a normal division (no correction
1298 // needed), which may speed up this implementation for the common case.
1299 // (see also JVM spec., p.243 & p.271)
1300 __ corrected_idivl(rcx);
1301 __ movl(rax, rdx);
1302 }
1303
1304 void TemplateTable::lmul() {
1305 transition(ltos, ltos);
1306 #ifdef _LP64
1307 __ pop_l(rdx);
1308 __ imulq(rax, rdx);
1309 #else
1310 __ pop_l(rbx, rcx);
1311 __ push(rcx); __ push(rbx);
1312 __ push(rdx); __ push(rax);
1313 __ lmul(2 * wordSize, 0);
1314 __ addptr(rsp, 4 * wordSize); // take off temporaries
1315 #endif
1316 }
1317
1318 void TemplateTable::ldiv() {
1319 transition(ltos, ltos);
1320 #ifdef _LP64
1321 __ mov(rcx, rax);
1322 __ pop_l(rax);
1323 // generate explicit div0 check
1324 __ testq(rcx, rcx);
1325 __ jump_cc(Assembler::zero,
1326 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1327 // Note: could xor rax and rcx 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_idivq(rcx); // kills rbx
1332 #else
1333 __ pop_l(rbx, rcx);
1334 __ push(rcx); __ push(rbx);
1335 __ push(rdx); __ push(rax);
1336 // check if y = 0
1337 __ orl(rax, rdx);
1338 __ jump_cc(Assembler::zero,
1339 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1340 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1341 __ addptr(rsp, 4 * wordSize); // take off temporaries
1342 #endif
1343 }
1344
1345 void TemplateTable::lrem() {
1346 transition(ltos, ltos);
1347 #ifdef _LP64
1348 __ mov(rcx, rax);
1349 __ pop_l(rax);
1350 __ testq(rcx, rcx);
1351 __ jump_cc(Assembler::zero,
1352 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1353 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1354 // they are not equal, one could do a normal division (no correction
1355 // needed), which may speed up this implementation for the common case.
1356 // (see also JVM spec., p.243 & p.271)
1357 __ corrected_idivq(rcx); // kills rbx
1358 __ mov(rax, rdx);
1359 #else
1360 __ pop_l(rbx, rcx);
1361 __ push(rcx); __ push(rbx);
1362 __ push(rdx); __ push(rax);
1363 // check if y = 0
1364 __ orl(rax, rdx);
1365 __ jump_cc(Assembler::zero,
1366 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1367 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1368 __ addptr(rsp, 4 * wordSize);
1369 #endif
1370 }
1371
1372 void TemplateTable::lshl() {
1373 transition(itos, ltos);
1374 __ movl(rcx, rax); // get shift count
1375 #ifdef _LP64
1376 __ pop_l(rax); // get shift value
1377 __ shlq(rax);
1378 #else
1379 __ pop_l(rax, rdx); // get shift value
1380 __ lshl(rdx, rax);
1381 #endif
1382 }
1383
1384 void TemplateTable::lshr() {
1385 #ifdef _LP64
1386 transition(itos, ltos);
1387 __ movl(rcx, rax); // get shift count
1388 __ pop_l(rax); // get shift value
1389 __ sarq(rax);
1390 #else
1391 transition(itos, ltos);
1392 __ mov(rcx, rax); // get shift count
1393 __ pop_l(rax, rdx); // get shift value
1394 __ lshr(rdx, rax, true);
1395 #endif
1396 }
1397
1398 void TemplateTable::lushr() {
1399 transition(itos, ltos);
1400 #ifdef _LP64
1401 __ movl(rcx, rax); // get shift count
1402 __ pop_l(rax); // get shift value
1403 __ shrq(rax);
1404 #else
1405 __ mov(rcx, rax); // get shift count
1406 __ pop_l(rax, rdx); // get shift value
1407 __ lshr(rdx, rax);
1408 #endif
1409 }
1410
1411 void TemplateTable::fop2(Operation op) {
1412 transition(ftos, ftos);
1413
1414 if (UseSSE >= 1) {
1415 switch (op) {
1416 case add:
1417 __ addss(xmm0, at_rsp());
1418 __ addptr(rsp, Interpreter::stackElementSize);
1419 break;
1420 case sub:
1421 __ movflt(xmm1, xmm0);
1422 __ pop_f(xmm0);
1423 __ subss(xmm0, xmm1);
1424 break;
1425 case mul:
1426 __ mulss(xmm0, at_rsp());
1427 __ addptr(rsp, Interpreter::stackElementSize);
1428 break;
1429 case div:
1430 __ movflt(xmm1, xmm0);
1431 __ pop_f(xmm0);
1432 __ divss(xmm0, xmm1);
1433 break;
1434 case rem:
1435 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1436 // modulo operation. The frem method calls the function
1437 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1438 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1439 // (signalling or quiet) is returned.
1440 //
1441 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1442 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1443 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1444 // The fprem instruction used on x86_32 is functionally equivalent to
1445 // SharedRuntime::frem in that it returns a NaN.
1446 #ifdef _LP64
1447 __ movflt(xmm1, xmm0);
1448 __ pop_f(xmm0);
1449 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1450 #else
1451 __ push_f(xmm0);
1452 __ pop_f();
1453 __ fld_s(at_rsp());
1454 __ fremr(rax);
1455 __ f2ieee();
1456 __ pop(rax); // pop second operand off the stack
1457 __ push_f();
1458 __ pop_f(xmm0);
1459 #endif
1460 break;
1461 default:
1462 ShouldNotReachHere();
1463 break;
1464 }
1465 } else {
1466 #ifdef LP64
1467 ShouldNotReachHere();
1468 #else
1469 switch (op) {
1470 case add: __ fadd_s (at_rsp()); break;
1471 case sub: __ fsubr_s(at_rsp()); break;
1472 case mul: __ fmul_s (at_rsp()); break;
1473 case div: __ fdivr_s(at_rsp()); break;
1474 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1475 default : ShouldNotReachHere();
1476 }
1477 __ f2ieee();
1478 __ pop(rax); // pop second operand off the stack
1479 #endif // LP64
1480 }
1481 }
1482
1483 void TemplateTable::dop2(Operation op) {
1484 transition(dtos, dtos);
1485 if (UseSSE >= 2) {
1486 switch (op) {
1487 case add:
1488 __ addsd(xmm0, at_rsp());
1489 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1490 break;
1491 case sub:
1492 __ movdbl(xmm1, xmm0);
1493 __ pop_d(xmm0);
1494 __ subsd(xmm0, xmm1);
1495 break;
1496 case mul:
1497 __ mulsd(xmm0, at_rsp());
1498 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1499 break;
1500 case div:
1501 __ movdbl(xmm1, xmm0);
1502 __ pop_d(xmm0);
1503 __ divsd(xmm0, xmm1);
1504 break;
1505 case rem:
1506 // Similar to fop2(), the modulo operation is performed using the
1507 // SharedRuntime::drem method (on x86_64 platforms) or using the
1508 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1509 #ifdef LP64
1510 __ movdbl(xmm1, xmm0);
1511 __ pop_d(xmm0);
1512 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1513 #else
1514 __ push_d(xmm0);
1515 __ pop_d();
1516 __ fld_d(at_rsp());
1517 __ fremr(rax);
1518 __ d2ieee();
1519 __ pop(rax);
1520 __ pop(rdx);
1521 __ push_d();
1522 __ pop_d(xmm0);
1523 #endif
1524 break;
1525 default:
1526 ShouldNotReachHere();
1527 break;
1528 }
1529 } else {
1530 #ifdef LP64
1531 ShouldNotReachHere();
1532 #else
1533 switch (op) {
1534 case add: __ fadd_d (at_rsp()); break;
1535 case sub: __ fsubr_d(at_rsp()); break;
1536 case mul: {
1537 Label L_strict;
1538 Label L_join;
1539 const Address access_flags (rcx, Method::access_flags_offset());
1540 __ get_method(rcx);
1541 __ movl(rcx, access_flags);
1542 __ testl(rcx, JVM_ACC_STRICT);
1543 __ jccb(Assembler::notZero, L_strict);
1544 __ fmul_d (at_rsp());
1545 __ jmpb(L_join);
1546 __ bind(L_strict);
1547 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1548 __ fmulp();
1549 __ fmul_d (at_rsp());
1550 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1551 __ fmulp();
1552 __ bind(L_join);
1553 break;
1554 }
1555 case div: {
1556 Label L_strict;
1557 Label L_join;
1558 const Address access_flags (rcx, Method::access_flags_offset());
1559 __ get_method(rcx);
1560 __ movl(rcx, access_flags);
1561 __ testl(rcx, JVM_ACC_STRICT);
1562 __ jccb(Assembler::notZero, L_strict);
1563 __ fdivr_d(at_rsp());
1564 __ jmp(L_join);
1565 __ bind(L_strict);
1566 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1567 __ fmul_d (at_rsp());
1568 __ fdivrp();
1569 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1570 __ fmulp();
1571 __ bind(L_join);
1572 break;
1573 }
1574 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1575 default : ShouldNotReachHere();
1576 }
1577 __ d2ieee();
1578 // Pop double precision number from rsp.
1579 __ pop(rax);
1580 __ pop(rdx);
1581 #endif
1582 }
1583 }
1584
1585 void TemplateTable::ineg() {
1586 transition(itos, itos);
1587 __ negl(rax);
1588 }
1589
1590 void TemplateTable::lneg() {
1591 transition(ltos, ltos);
1592 LP64_ONLY( negq(rax));
1593 NOT_LP64( lneg(rdx, rax));
1594 }
1595
1596 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1597 static jlong double_quadword(jlong adr, jlong lo, jlong hi) {
1598 // Use the expression (adr)&(0xF) to provide 128-bits aligned address
1599 // of 128-bits operands for SSE instructions.
1600 jlong operand = (jlong)(((intptr_t)adr)&((intptr_t)(0xF)));
1601 // Store the value to a 128-bits operand.
1602 operand[0] = lo;
1603 operand[1] = hi;
1604 return operand;
1605 }
1606
1607 // Buffer for 128-bits masks used by SSE instructions.
1608 static jlong float_signflip_pool[22];
1609 static jlong double_signflip_pool[22];
1610
1611 void TemplateTable::fneg() {
1612 transition(ftos, ftos);
1613 if (UseSSE >= 1) {
1614 static jlong float_signflip = double_quadword(&float_signflip_pool[1], 0x8000000080000000, 0x8000000080000000);
1615 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1616 } else {
1617 LP64_ONLY(ShouldNotReachHere());
1618 NOT_LP64(__ fchs());
1619 }
1620 }
1621
1622 void TemplateTable::dneg() {
1623 transition(dtos, dtos);
1624 if (UseSSE >= 2) {
1625 static jlong double_signflip = double_quadword(&double_signflip_pool[1], 0x8000000000000000, 0x8000000000000000);
1626 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1627 } else {
1628 #ifdef _LP64
1629 ShouldNotReachHere();
1630 #else
1631 __ fchs();
1632 #endif
1633 }
1634 }
1635
1636 void TemplateTable::iinc() {
1637 transition(vtos, vtos);
1638 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1639 locals_index(rbx);
1640 __ addl(iaddress(rbx), rdx);
1641 }
1642
1643 void TemplateTable::wide_iinc() {
1644 transition(vtos, vtos);
1645 __ movl(rdx, at_bcp(4)); // get constant
1646 locals_index_wide(rbx);
1647 __ bswapl(rdx); // swap bytes & sign-extend constant
1648 __ sarl(rdx, 16);
1649 __ addl(iaddress(rbx), rdx);
1650 // Note: should probably use only one movl to get both
1651 // the index and the constant -> fix this
1652 }
1653
1654 void TemplateTable::convert() {
1655 #ifdef _LP64
1656 // Checking
1657 #ifdef ASSERT
1658 {
1659 TosState tos_in = ilgl;
1660 TosState tos_out = ilgl;
1661 switch (bytecode()) {
1662 case Bytecodes::_i2l: // fall through
1663 case Bytecodes::_i2f: // fall through
1664 case Bytecodes::_i2d: // fall through
1665 case Bytecodes::_i2b: // fall through
1666 case Bytecodes::_i2c: // fall through
1667 case Bytecodes::_i2s: tos_in = itos; break;
1668 case Bytecodes::_l2i: // fall through
1669 case Bytecodes::_l2f: // fall through
1670 case Bytecodes::_l2d: tos_in = ltos; break;
1671 case Bytecodes::_f2i: // fall through
1672 case Bytecodes::_f2l: // fall through
1673 case Bytecodes::_f2d: tos_in = ftos; break;
1674 case Bytecodes::_d2i: // fall through
1675 case Bytecodes::_d2l: // fall through
1676 case Bytecodes::_d2f: tos_in = dtos; break;
1677 default : ShouldNotReachHere();
1678 }
1679 switch (bytecode()) {
1680 case Bytecodes::_l2i: // fall through
1681 case Bytecodes::_f2i: // fall through
1682 case Bytecodes::_d2i: // fall through
1683 case Bytecodes::_i2b: // fall through
1684 case Bytecodes::_i2c: // fall through
1685 case Bytecodes::_i2s: tos_out = itos; break;
1686 case Bytecodes::_i2l: // fall through
1687 case Bytecodes::_f2l: // fall through
1688 case Bytecodes::_d2l: tos_out = ltos; break;
1689 case Bytecodes::_i2f: // fall through
1690 case Bytecodes::_l2f: // fall through
1691 case Bytecodes::_d2f: tos_out = ftos; break;
1692 case Bytecodes::_i2d: // fall through
1693 case Bytecodes::_l2d: // fall through
1694 case Bytecodes::_f2d: tos_out = dtos; break;
1695 default : ShouldNotReachHere();
1696 }
1697 transition(tos_in, tos_out);
1698 }
1699 #endif // ASSERT
1700
1701 static const int64_t is_nan = 0x8000000000000000L;
1702
1703 // Conversion
1704 switch (bytecode()) {
1705 case Bytecodes::_i2l:
1706 __ movslq(rax, rax);
1707 break;
1708 case Bytecodes::_i2f:
1709 __ cvtsi2ssl(xmm0, rax);
1710 break;
1711 case Bytecodes::_i2d:
1712 __ cvtsi2sdl(xmm0, rax);
1713 break;
1714 case Bytecodes::_i2b:
1715 __ movsbl(rax, rax);
1716 break;
1717 case Bytecodes::_i2c:
1718 __ movzwl(rax, rax);
1719 break;
1720 case Bytecodes::_i2s:
1721 __ movswl(rax, rax);
1722 break;
1723 case Bytecodes::_l2i:
1724 __ movl(rax, rax);
1725 break;
1726 case Bytecodes::_l2f:
1727 __ cvtsi2ssq(xmm0, rax);
1728 break;
1729 case Bytecodes::_l2d:
1730 __ cvtsi2sdq(xmm0, rax);
1731 break;
1732 case Bytecodes::_f2i:
1733 {
1734 Label L;
1735 __ cvttss2sil(rax, xmm0);
1736 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1737 __ jcc(Assembler::notEqual, L);
1738 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1739 __ bind(L);
1740 }
1741 break;
1742 case Bytecodes::_f2l:
1743 {
1744 Label L;
1745 __ cvttss2siq(rax, xmm0);
1746 // NaN or overflow/underflow?
1747 __ cmp64(rax, ExternalAddress((address) &is_nan));
1748 __ jcc(Assembler::notEqual, L);
1749 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1750 __ bind(L);
1751 }
1752 break;
1753 case Bytecodes::_f2d:
1754 __ cvtss2sd(xmm0, xmm0);
1755 break;
1756 case Bytecodes::_d2i:
1757 {
1758 Label L;
1759 __ cvttsd2sil(rax, xmm0);
1760 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1761 __ jcc(Assembler::notEqual, L);
1762 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1763 __ bind(L);
1764 }
1765 break;
1766 case Bytecodes::_d2l:
1767 {
1768 Label L;
1769 __ cvttsd2siq(rax, xmm0);
1770 // NaN or overflow/underflow?
1771 __ cmp64(rax, ExternalAddress((address) &is_nan));
1772 __ jcc(Assembler::notEqual, L);
1773 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1774 __ bind(L);
1775 }
1776 break;
1777 case Bytecodes::_d2f:
1778 __ cvtsd2ss(xmm0, xmm0);
1779 break;
1780 default:
1781 ShouldNotReachHere();
1782 }
1783 #else
1784 // Checking
1785 #ifdef ASSERT
1786 { TosState tos_in = ilgl;
1787 TosState tos_out = ilgl;
1788 switch (bytecode()) {
1789 case Bytecodes::_i2l: // fall through
1790 case Bytecodes::_i2f: // fall through
1791 case Bytecodes::_i2d: // fall through
1792 case Bytecodes::_i2b: // fall through
1793 case Bytecodes::_i2c: // fall through
1794 case Bytecodes::_i2s: tos_in = itos; break;
1795 case Bytecodes::_l2i: // fall through
1796 case Bytecodes::_l2f: // fall through
1797 case Bytecodes::_l2d: tos_in = ltos; break;
1798 case Bytecodes::_f2i: // fall through
1799 case Bytecodes::_f2l: // fall through
1800 case Bytecodes::_f2d: tos_in = ftos; break;
1801 case Bytecodes::_d2i: // fall through
1802 case Bytecodes::_d2l: // fall through
1803 case Bytecodes::_d2f: tos_in = dtos; break;
1804 default : ShouldNotReachHere();
1805 }
1806 switch (bytecode()) {
1807 case Bytecodes::_l2i: // fall through
1808 case Bytecodes::_f2i: // fall through
1809 case Bytecodes::_d2i: // fall through
1810 case Bytecodes::_i2b: // fall through
1811 case Bytecodes::_i2c: // fall through
1812 case Bytecodes::_i2s: tos_out = itos; break;
1813 case Bytecodes::_i2l: // fall through
1814 case Bytecodes::_f2l: // fall through
1815 case Bytecodes::_d2l: tos_out = ltos; break;
1816 case Bytecodes::_i2f: // fall through
1817 case Bytecodes::_l2f: // fall through
1818 case Bytecodes::_d2f: tos_out = ftos; break;
1819 case Bytecodes::_i2d: // fall through
1820 case Bytecodes::_l2d: // fall through
1821 case Bytecodes::_f2d: tos_out = dtos; break;
1822 default : ShouldNotReachHere();
1823 }
1824 transition(tos_in, tos_out);
1825 }
1826 #endif // ASSERT
1827
1828 // Conversion
1829 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1830 switch (bytecode()) {
1831 case Bytecodes::i2l:
1832 __ extend_sign(rdx, rax);
1833 break;
1834 case Bytecodes::i2f:
1835 if (UseSSE >= 1) {
1836 __ cvtsi2ssl(xmm0, rax);
1837 } else {
1838 __ push(rax); // store int on tos
1839 __ fild_s(at_rsp()); // load int to ST0
1840 __ f2ieee(); // truncate to float size
1841 __ pop(rcx); // adjust rsp
1842 }
1843 break;
1844 case Bytecodes::i2d:
1845 if (UseSSE >= 2) {
1846 __ cvtsi2sdl(xmm0, rax);
1847 } else {
1848 __ push(rax); // add one slot for d2ieee()
1849 __ push(rax); // store int on tos
1850 __ fild_s(at_rsp()); // load int to ST0
1851 __ d2ieee(); // truncate to double size
1852 __ pop(rcx); // adjust rsp
1853 __ pop(rcx);
1854 }
1855 break;
1856 case Bytecodes::i2b:
1857 __ shll(rax, 24); // truncate upper 24 bits
1858 __ sarl(rax, 24); // and sign-extend byte
1859 LP64_ONLY( movsbl(rax, rax));
1860 break;
1861 case Bytecodes::i2c:
1862 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1863 LP64_ONLY( movzwl(rax, rax));
1864 break;
1865 case Bytecodes::i2s:
1866 __ shll(rax, 16); // truncate upper 16 bits
1867 __ sarl(rax, 16); // and sign-extend short
1868 LP64_ONLY( movswl(rax, rax));
1869 break;
1870 case Bytecodes::_l2i:
1871 / nothing to do /
1872 break;
1873 case Bytecodes::_l2f:
1874 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1875 // 64-bit long values to floats. On 32-bit platforms it is not possible
1876 // to use that instruction with 64-bit operands, therefore the FPU is
1877 // used to perform the conversion.
1878 __ push(rdx); // store long on tos
1879 __ push(rax);
1880 __ fild_d(at_rsp()); // load long to ST0
1881 __ f2ieee(); // truncate to float size
1882 __ pop(rcx); // adjust rsp
1883 __ pop(rcx);
1884 if (UseSSE >= 1) {
1885 __ push_f();
1886 __ pop_f(xmm0);
1887 }
1888 break;
1889 case Bytecodes::_l2d:
1890 // On 32-bit platforms the FPU is used for conversion because on
1891 // 32-bit platforms it is not not possible to use the cvtsi2sdq
1892 // instruction with 64-bit operands.
1893 __ push(rdx); // store long on tos
1894 __ push(rax);
1895 __ fild_d(at_rsp()); // load long to ST0
1896 __ d2ieee(); // truncate to double size
1897 __ pop(rcx); // adjust rsp
1898 __ pop(rcx);
1899 if (UseSSE >= 2) {
1900 __ push_d();
1901 __ pop_d(xmm0);
1902 }
1903 break;
1904 case Bytecodes::_f2i:
1905 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
1906 // as it returns 0 for any NaN.
1907 if (UseSSE >= 1) {
1908 __ push_f(xmm0);
1909 } else {
1910 __ push(rcx); // reserve space for argument
1911 __ fstp_s(at_rsp()); // pass float argument on stack
1912 }
1913 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1914 break;
1915 case Bytecodes::_f2l:
1916 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
1917 // as it returns 0 for any NaN.
1918 if (UseSSE >= 1) {
1919 __ push_f(xmm0);
1920 } else {
1921 __ push(rcx); // reserve space for argument
1922 __ fstp_s(at_rsp()); // pass float argument on stack
1923 }
1924 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1925 break;
1926 case Bytecodes::_f2d:
1927 if (UseSSE < 1) {
1928 /* nothing to do */
1929 } else if (UseSSE == 1) {
1930 __ push_f(xmm0);
1931 __ pop_f();
1932 } else { // UseSSE >= 2
1933 __ cvtss2sd(xmm0, xmm0);
1934 }
1935 break;
1936 case Bytecodes::_d2i:
1937 if (UseSSE >= 2) {
1938 __ push_d(xmm0);
1939 } else {
1940 __ push(rcx); // reserve space for argument
1941 __ push(rcx);
1942 __ fstp_d(at_rsp()); // pass double argument on stack
1943 }
1944 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1945 break;
1946 case Bytecodes::_d2l:
1947 if (UseSSE >= 2) {
1948 __ push_d(xmm0);
1949 } else {
1950 __ push(rcx); // reserve space for argument
1951 __ push(rcx);
1952 __ fstp_d(at_rsp()); // pass double argument on stack
1953 }
1954 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1955 break;
1956 case Bytecodes::_d2f:
1957 if (UseSSE <= 1) {
1958 __ push(rcx); // reserve space for f2ieee()
1959 __ f2ieee(); // truncate to float size
1960 __ pop(rcx); // adjust rsp
1961 if (UseSSE == 1) {
1962 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
1963 // the conversion is performed using the FPU in this case.
1964 __ push_f();
1965 __ pop_f(xmm0);
1966 }
1967 } else { // UseSSE >= 2
1968 __ cvtsd2ss(xmm0, xmm0);
1969 }
1970 break;
1971 default :
1972 ShouldNotReachHere();
1973 }
1974 #endif
1975 }
1976
1977 void TemplateTable::lcmp() {
1978 transition(ltos, itos);
1979 #ifdef _LP64
1980 Label done;
1981 __ pop_l(rdx);
1982 __ cmpq(rdx, rax);
1983 __ movl(rax, -1);
1984 __ jccb(Assembler::less, done);
1985 __ setb(Assembler::notEqual, rax);
1986 __ movzbl(rax, rax);
1987 __ bind(done);
1988 #else
1989
1990 // y = rdx:rax
1991 __ pop_l(rbx, rcx); // get x = rcx:rbx
1992 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1993 __ mov(rax, rcx);
1994 #endif
1995 }
1996
1997 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1998 if ((is_float && UseSSE >= 1) ||
1999 (!is_float && UseSSE >= 2)) {
2000 Label done;
2001 if (is_float) {
2002 // XXX get rid of pop here, use ... reg, mem32
2003 __ pop_f(xmm1);
2004 __ ucomiss(xmm1, xmm0);
2005 } else {
2006 // XXX get rid of pop here, use ... reg, mem64
2007 __ pop_d(xmm1);
2008 __ ucomisd(xmm1, xmm0);
2009 }
2010 if (unordered_result < 0) {
2011 __ movl(rax, -1);
2012 __ jccb(Assembler::parity, done);
2013 __ jccb(Assembler::below, done);
2014 __ setb(Assembler::notEqual, rdx);
2015 __ movzbl(rax, rdx);
2016 } else {
2017 __ movl(rax, 1);
2018 __ jccb(Assembler::parity, done);
2019 __ jccb(Assembler::above, done);
2020 __ movl(rax, 0);
2021 __ jccb(Assembler::equal, done);
2022 __ decrementl(rax);
2023 }
2024 __ bind(done);
2025 } else {
2026 #ifdef _LP64
2027 ShouldNotReachHere();
2028 #else
2029 if (is_float) {
2030 __ fld_s(at_rsp());
2031 } else {
2032 __ fld_d(at_rsp());
2033 __ pop(rdx);
2034 }
2035 __ pop(rcx);
2036 __ fcmp2int(rax, unordered_result < 0);
2037 #endif // _LP64
2038 }
2039 }
2040
2041 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2042 __ get_method(rcx); // rcx holds method
2043 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2044 // holds bumped taken count
2045
2046 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2047 InvocationCounter::counter_offset();
2048 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2049 InvocationCounter::counter_offset();
2050
2051 // Load up edx with the branch displacement
2052 if (is_wide) {
2053 __ movl(rdx, at_bcp(1));
2054 } else {
2055 __ load_signed_short(rdx, at_bcp(1));
2056 }
2057 __ bswapl(rdx);
2058
2059 if (!is_wide) {
2060 __ sarl(rdx, 16);
2061 }
2062 LP64_ONLY(__ movl2ptr(rdx, rdx));
2063
2064 // Handle all the JSR stuff here, then exit.
2065 // It's much shorter and cleaner than intermingling with the non-JSR
2066 // normal-branch stuff occurring below.
2067 if (is_jsr) {
2068 // Pre-load the next target bytecode into rbx
2069 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2070
2071 // compute return address as bci in rax
2072 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2073 in_bytes(ConstMethod::codes_offset())));
2074 __ subptr(rax, Address(rcx, Method::const_offset()));
2075 // Adjust the bcp in r13 by the displacement in rdx
2076 __ addptr(rbcp, rdx);
2077 // jsr returns atos that is not an oop
2078 __ push_i(rax);
2079 __ dispatch_only(vtos);
2080 return;
2081 }
2082
2083 // Normal (non-jsr) branch handling
2084
2085 // Adjust the bcp in r13 by the displacement in rdx
2086 __ addptr(rbcp, rdx);
2087
2088 assert(UseLoopCounter || !UseOnStackReplacement,
2089 "on-stack-replacement requires loop counters");
2090 Label backedge_counter_overflow;
2091 Label profile_method;
2092 Label dispatch;
2093 if (UseLoopCounter) {
2094 // increment backedge counter for backward branches
2095 // rax: MDO
2096 // rbx: MDO bumped taken-count
2097 // rcx: method
2098 // rdx: target offset
2099 // r13: target bcp
2100 // r14: locals pointer
2101 __ testl(rdx, rdx); // check if forward or backward branch
2102 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2103
2104 // check if MethodCounters exists
2105 Label has_counters;
2106 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2107 __ testptr(rax, rax);
2108 __ jcc(Assembler::notZero, has_counters);
2109 __ push(rdx);
2110 __ push(rcx);
2111 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2112 rcx);
2113 __ pop(rcx);
2114 __ pop(rdx);
2115 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2116 __ testptr(rax, rax);
2117 __ jcc(Assembler::zero, dispatch);
2118 __ bind(has_counters);
2119
2120 if (TieredCompilation) {
2121 Label no_mdo;
2122 int increment = InvocationCounter::count_increment;
2123 if (ProfileInterpreter) {
2124 // Are we profiling?
2125 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2126 __ testptr(rbx, rbx);
2127 __ jccb(Assembler::zero, no_mdo);
2128 // Increment the MDO backedge counter
2129 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2130 in_bytes(InvocationCounter::counter_offset()));
2131 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2132 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2133 rax, false, Assembler::zero, &backedge_counter_overflow);
2134 __ jmp(dispatch);
2135 }
2136 __ bind(no_mdo);
2137 // Increment backedge counter in MethodCounters*
2138 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2139 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2140 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2141 rax, false, Assembler::zero, &backedge_counter_overflow);
2142 } else { // not TieredCompilation
2143 // increment counter
2144 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2145 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2146 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2147 __ movl(Address(rcx, be_offset), rax); // store counter
2148
2149 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2150
2151 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2152 __ addl(rax, Address(rcx, be_offset)); // add both counters
2153
2154 if (ProfileInterpreter) {
2155 // Test to see if we should create a method data oop
2156 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2157 __ jcc(Assembler::less, dispatch);
2158
2159 // if no method data exists, go to profile method
2160 __ test_method_data_pointer(rax, profile_method);
2161
2162 if (UseOnStackReplacement) {
2163 // check for overflow against rbx which is the MDO taken count
2164 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2165 __ jcc(Assembler::below, dispatch);
2166
2167 // When ProfileInterpreter is on, the backedge_count comes
2168 // from the MethodData*, which value does not get reset on
2169 // the call to frequency_counter_overflow(). To avoid
2170 // excessive calls to the overflow routine while the method is
2171 // being compiled, add a second test to make sure the overflow
2172 // function is called only once every overflow_frequency.
2173 const int overflow_frequency = 1024;
2174 __ andl(rbx, overflow_frequency - 1);
2175 __ jcc(Assembler::zero, backedge_counter_overflow);
2176
2177 }
2178 } else {
2179 if (UseOnStackReplacement) {
2180 // check for overflow against rax, which is the sum of the
2181 // counters
2182 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2183 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2184
2185 }
2186 }
2187 }
2188 __ bind(dispatch);
2189 }
2190
2191 // Pre-load the next target bytecode into rbx
2192 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2193
2194 // continue with the bytecode @ target
2195 // rax: return bci for jsr's, unused otherwise
2196 // rbx: target bytecode
2197 // r13: target bcp
2198 __ dispatch_only(vtos);
2199
2200 if (UseLoopCounter) {
2201 if (ProfileInterpreter) {
2202 // Out-of-line code to allocate method data oop.
2203 __ bind(profile_method);
2204 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2205 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2206 __ set_method_data_pointer_for_bcp();
2207 __ jmp(dispatch);
2208 }
2209
2210 if (UseOnStackReplacement) {
2211 // invocation counter overflow
2212 __ bind(backedge_counter_overflow);
2213 __ negptr(rdx);
2214 __ addptr(rdx, rbcp); // branch bcp
2215 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2216 __ call_VM(noreg,
2217 CAST_FROM_FN_PTR(address,
2218 InterpreterRuntime::frequency_counter_overflow),
2219 rdx);
2220 __ load_unsigned_byte(rbx, Address(rbcp, 0)); // restore target bytecode
2221
2222 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2223 // rbx: target bytecode
2224 // rdx: scratch
2225 // r14: locals pointer
2226 // r13: bcp
2227 __ testptr(rax, rax); // test result
2228 __ jcc(Assembler::zero, dispatch); // no osr if null
2229 // nmethod may have been invalidated (VM may block upon call_VM return)
2230 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2231 __ jcc(Assembler::notEqual, dispatch);
2232
2233 // We have the address of an on stack replacement routine in rax
2234 // We need to prepare to execute the OSR method. First we must
2235 // migrate the locals and monitors off of the stack.
2236
2237 LP64_ONLY(__ mov(r13, rax)); // save the nmethod
2238 NOT_LP64(__ mov(rbx, rax)); // save the nmethod
2239 NOT_LP64(__ get_thread(rcx));
2240
2241 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2242
2243 // rax is OSR buffer, move it to expected parameter location
2244 LP64_ONLY(__ mov(j_rarg0, rax));
2245 NOT_LP64(__ mov(rcx, rax));
2246 // We use j_rarg definitions here so that registers don't conflict as parameter
2247 // registers change across platforms as we are in the midst of a calling
2248 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2249
2250 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2251 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2252
2253
2254 // pop the interpreter frame
2255 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2256 __ leave(); // remove frame anchor
2257 __ pop(retaddr); // get return address
2258 __ mov(rsp, sender_sp); // set sp to sender sp
2259 // Ensure compiled code always sees stack at proper alignment
2260 __ andptr(rsp, -(StackAlignmentInBytes));
2261
2262 // unlike x86 we need no specialized return from compiled code
2263 // to the interpreter or the call stub.
2264
2265 // push the return address
2266 __ push(retaddr);
2267
2268 // and begin the OSR nmethod
2269 LP64_ONLY(__ jmp(Address(r13, nmethod::osr_entry_point_offset())));
2270 NOT_LP64(__ jmp(Address(rbx, nmethod::osr_entry_point_offset())));
2271 }
2272 }
2273 }
2274
2275 void TemplateTable::if_0cmp(Condition cc) {
2276 transition(itos, vtos);
2277 // assume branch is more often taken than not (loops use backward branches)
2278 Label not_taken;
2279 __ testl(rax, rax);
2280 __ jcc(j_not(cc), not_taken);
2281 branch(false, false);
2282 __ bind(not_taken);
2283 __ profile_not_taken_branch(rax);
2284 }
2285
2286 void TemplateTable::if_icmp(Condition cc) {
2287 transition(itos, vtos);
2288 // assume branch is more often taken than not (loops use backward branches)
2289 Label not_taken;
2290 __ pop_i(rdx);
2291 __ cmpl(rdx, rax);
2292 __ jcc(j_not(cc), not_taken);
2293 branch(false, false);
2294 __ bind(not_taken);
2295 __ profile_not_taken_branch(rax);
2296 }
2297
2298 void TemplateTable::if_nullcmp(Condition cc) {
2299 transition(atos, vtos);
2300 // assume branch is more often taken than not (loops use backward branches)
2301 Label not_taken;
2302 __ testptr(rax, rax);
2303 __ jcc(j_not(cc), not_taken);
2304 branch(false, false);
2305 __ bind(not_taken);
2306 __ profile_not_taken_branch(rax);
2307 }
2308
2309 void TemplateTable::if_acmp(Condition cc) {
2310 transition(atos, vtos);
2311 // assume branch is more often taken than not (loops use backward branches)
2312 Label not_taken;
2313 __ pop_ptr(rdx);
2314 __ cmpptr(rdx, rax);
2315 __ jcc(j_not(cc), not_taken);
2316 branch(false, false);
2317 __ bind(not_taken);
2318 __ profile_not_taken_branch(rax);
2319 }
2320
2321 void TemplateTable::ret() {
2322 transition(vtos, vtos);
2323 locals_index(rbx);
2324 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2325 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2326 __ profile_ret(rbx, rcx);
2327 __ get_method(rax);
2328 __ movptr(rbcp, Address(rax, Method::const_offset()));
2329 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2330 ConstMethod::codes_offset()));
2331 __ dispatch_next(vtos);
2332 }
2333
2334 void TemplateTable::wide_ret() {
2335 transition(vtos, vtos);
2336 locals_index_wide(rbx);
2337 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2338 __ profile_ret(rbx, rcx);
2339 __ get_method(rax);
2340 __ movptr(rbcp, Address(rax, Method::const_offset()));
2341 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2342 __ dispatch_next(vtos);
2343 }
2344
2345 void TemplateTable::tableswitch() {
2346 Label default_case, continue_execution;
2347 transition(itos, vtos);
2348
2349 // align r13/rsi
2350 __ lea(rbx, at_bcp(BytesPerInt));
2351 __ andptr(rbx, -BytesPerInt);
2352 // load lo & hi
2353 __ movl(rcx, Address(rbx, BytesPerInt));
2354 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2355 __ bswapl(rcx);
2356 __ bswapl(rdx);
2357 // check against lo & hi
2358 __ cmpl(rax, rcx);
2359 __ jcc(Assembler::less, default_case);
2360 __ cmpl(rax, rdx);
2361 __ jcc(Assembler::greater, default_case);
2362 // lookup dispatch offset
2363 __ subl(rax, rcx);
2364 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2365 __ profile_switch_case(rax, rbx, rcx);
2366 // continue execution
2367 __ bind(continue_execution);
2368 __ bswapl(rdx);
2369 LP64_ONLY(__ movl2ptr(rdx, rdx));
2370 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2371 __ addptr(rbcp, rdx);
2372 __ dispatch_only(vtos);
2373 // handle default
2374 __ bind(default_case);
2375 __ profile_switch_default(rax);
2376 __ movl(rdx, Address(rbx, 0));
2377 __ jmp(continue_execution);
2378 }
2379
2380 void TemplateTable::lookupswitch() {
2381 transition(itos, itos);
2382 __ stop("lookupswitch bytecode should have been rewritten");
2383 }
2384
2385 void TemplateTable::fast_linearswitch() {
2386 transition(itos, vtos);
2387 Label loop_entry, loop, found, continue_execution;
2388 // bswap rax so we can avoid bswapping the table entries
2389 __ bswapl(rax);
2390 // align r13
2391 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2392 // this instruction (change offsets
2393 // below)
2394 __ andptr(rbx, -BytesPerInt);
2395 // set counter
2396 __ movl(rcx, Address(rbx, BytesPerInt));
2397 __ bswapl(rcx);
2398 __ jmpb(loop_entry);
2399 // table search
2400 __ bind(loop);
2401 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2402 __ jcc(Assembler::equal, found);
2403 __ bind(loop_entry);
2404 __ decrementl(rcx);
2405 __ jcc(Assembler::greaterEqual, loop);
2406 // default case
2407 __ profile_switch_default(rax);
2408 __ movl(rdx, Address(rbx, 0));
2409 __ jmp(continue_execution);
2410 // entry found -> get offset
2411 __ bind(found);
2412 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2413 __ profile_switch_case(rcx, rax, rbx);
2414 // continue execution
2415 __ bind(continue_execution);
2416 __ bswapl(rdx);
2417 __ movl2ptr(rdx, rdx);
2418 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2419 __ addptr(rbcp, rdx);
2420 __ dispatch_only(vtos);
2421 }
2422
2423 void TemplateTable::fast_binaryswitch() {
2424 transition(itos, vtos);
2425 // Implementation using the following core algorithm:
2426 //
2427 // int binary_search(int key, LookupswitchPair array, int n) {
2428 // // Binary search according to "Methodik des Programmierens" by
2429 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2430 // int i = 0;
2431 // int j = n;
2432 // while (i+1 < j) {
2433 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2434 // // with Q: for all i: 0 <= i < n: key < a[i]
2435 // // where a stands for the array and assuming that the (inexisting)
2436 // // element a[n] is infinitely big.
2437 // int h = (i + j) >> 1;
2438 // // i < h < j
2439 // if (key < array[h].fast_match()) {
2440 // j = h;
2441 // } else {
2442 // i = h;
2443 // }
2444 // }
2445 // // R: a[i] <= key < a[i+1] or Q
2446 // // (i.e., if key is within array, i is the correct index)
2447 // return i;
2448 // }
2449
2450 // Register allocation
2451 const Register key = rax; // already set (tosca)
2452 const Register array = rbx;
2453 const Register i = rcx;
2454 const Register j = rdx;
2455 const Register h = rdi;
2456 const Register temp = rsi;
2457
2458 // Find array start
2459 NOT_LP64(__ save_bcp());
2460
2461 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2462 // get rid of this
2463 // instruction (change
2464 // offsets below)
2465 __ andptr(array, -BytesPerInt);
2466
2467 // Initialize i & j
2468 __ xorl(i, i); // i = 0;
2469 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2470
2471 // Convert j into native byteordering
2472 __ bswapl(j);
2473
2474 // And start
2475 Label entry;
2476 __ jmp(entry);
2477
2478 // binary search loop
2479 {
2480 Label loop;
2481 __ bind(loop);
2482 // int h = (i + j) >> 1;
2483 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2484 __ sarl(h, 1); // h = (i + j) >> 1;
2485 // if (key < array[h].fast_match()) {
2486 // j = h;
2487 // } else {
2488 // i = h;
2489 // }
2490 // Convert array[h].match to native byte-ordering before compare
2491 __ movl(temp, Address(array, h, Address::times_8));
2492 __ bswapl(temp);
2493 __ cmpl(key, temp);
2494 // j = h if (key < array[h].fast_match())
2495 __ cmov32(Assembler::less, j, h);
2496 // i = h if (key >= array[h].fast_match())
2497 __ cmov32(Assembler::greaterEqual, i, h);
2498 // while (i+1 < j)
2499 __ bind(entry);
2500 __ leal(h, Address(i, 1)); // i+1
2501 __ cmpl(h, j); // i+1 < j
2502 __ jcc(Assembler::less, loop);
2503 }
2504
2505 // end of binary search, result index is i (must check again!)
2506 Label default_case;
2507 // Convert array[i].match to native byte-ordering before compare
2508 __ movl(temp, Address(array, i, Address::times_8));
2509 __ bswapl(temp);
2510 __ cmpl(key, temp);
2511 __ jcc(Assembler::notEqual, default_case);
2512
2513 // entry found -> j = offset
2514 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2515 __ profile_switch_case(i, key, array);
2516 __ bswapl(j);
2517 LP64_ONLY(__ movslq(j, j));
2518
2519 NOT_LP64(__ restore_bcp());
2520 NOT_LP64(__ restore_locals()); // restore rdi
2521
2522 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2523 __ addptr(rbcp, j);
2524 __ dispatch_only(vtos);
2525
2526 // default case -> j = default offset
2527 __ bind(default_case);
2528 __ profile_switch_default(i);
2529 __ movl(j, Address(array, -2 * BytesPerInt));
2530 __ bswapl(j);
2531 LP64_ONLY(__ movslq(j, j));
2532
2533 NOT_LP64(__ restore_bcp());
2534 NOT_LP64(__ restore_locals());
2535
2536 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2537 __ addptr(rbcp, j);
2538 __ dispatch_only(vtos);
2539 }
2540
2541 void TemplateTable::_return(TosState state) {
2542 transition(state, state);
2543
2544 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2545
2546 assert(_desc->calls_vm(),
2547 "inconsistent calls_vm information"); // call in remove_activation
2548
2549 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2550 assert(state == vtos, "only valid state");
2551 __ movptr(robj, aaddress(0));
2552 __ load_klass(rdi, robj);
2553 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2554 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2555 Label skip_register_finalizer;
2556 __ jcc(Assembler::zero, skip_register_finalizer);
2557
2558 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2559
2560 __ bind(skip_register_finalizer);
2561 }
2562
2563 __ remove_activation(state, rbcp);
2564 __ jmp(rbcp);
2565 }
2566
2567 // ----------------------------------------------------------------------------
2568 // Volatile variables demand their effects be made known to all CPU's
2569 // in order. Store buffers on most chips allow reads & writes to
2570 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2571 // without some kind of memory barrier (i.e., it's not sufficient that
2572 // the interpreter does not reorder volatile references, the hardware
2573 // also must not reorder them).
2574 //
2575 // According to the new Java Memory Model (JMM):
2576 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2577 // writes act as aquire & release, so:
2578 // (2) A read cannot let unrelated NON-volatile memory refs that
2579 // happen after the read float up to before the read. It's OK for
2580 // non-volatile memory refs that happen before the volatile read to
2581 // float down below it.
2582 // (3) Similar a volatile write cannot let unrelated NON-volatile
2583 // memory refs that happen BEFORE the write float down to after the
2584 // write. It's OK for non-volatile memory refs that happen after the
2585 // volatile write to float up before it.
2586 //
2587 // We only put in barriers around volatile refs (they are expensive),
2588 // not between memory refs (that would require us to track the
2589 // flavor of the previous memory refs). Requirements (2) and (3)
2590 // require some barriers before volatile stores and after volatile
2591 // loads. These nearly cover requirement (1) but miss the
2592 // volatile-store-volatile-load case. This final case is placed after
2593 // volatile-stores although it could just as well go before
2594 // volatile-loads.
2595
2596 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2597 // Helper function to insert a is-volatile test and memory barrier
2598 if(!os::is_MP()) return; // Not needed on single CPU
2599 __ membar(order_constraint);
2600 }
2601
2602 void TemplateTable::resolve_cache_and_index(int byte_no,
2603 Register Rcache,
2604 Register index,
2605 size_t index_size) {
2606 const Register temp = rbx;
2607 assert_different_registers(Rcache, index, temp);
2608
2609 Label resolved;
2610
2611 Bytecodes::Code code = bytecode();
2612 switch (code) {
2613 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2614 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2615 }
2616
2617 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2618 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2619 __ cmpl(temp, code); // have we resolved this bytecode?
2620 __ jcc(Assembler::equal, resolved);
2621
2622 // resolve first time through
2623 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2624 __ movl(temp, code);
2625 __ call_VM(noreg, entry, temp);
2626 // Update registers with resolved info
2627 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2628 __ bind(resolved);
2629 }
2630
2631 // The cache and index registers must be set before call
2632 void TemplateTable::load_field_cp_cache_entry(Register obj,
2633 Register cache,
2634 Register index,
2635 Register off,
2636 Register flags,
2637 bool is_static = false) {
2638 assert_different_registers(cache, index, flags, off);
2639
2640 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2641 // Field offset
2642 __ movptr(off, Address(cache, index, Address::times_ptr,
2643 in_bytes(cp_base_offset +
2644 ConstantPoolCacheEntry::f2_offset())));
2645 // Flags
2646 __ movl(flags, Address(cache, index, Address::times_ptr,
2647 in_bytes(cp_base_offset +
2648 ConstantPoolCacheEntry::flags_offset())));
2649
2650 // klass overwrite register
2651 if (is_static) {
2652 __ movptr(obj, Address(cache, index, Address::times_ptr,
2653 in_bytes(cp_base_offset +
2654 ConstantPoolCacheEntry::f1_offset())));
2655 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2656 __ movptr(obj, Address(obj, mirror_offset));
2657 }
2658 }
2659
2660 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2661 Register method,
2662 Register itable_index,
2663 Register flags,
2664 bool is_invokevirtual,
2665 bool is_invokevfinal, /unused/
2666 bool is_invokedynamic) {
2667 // setup registers
2668 const Register cache = rcx;
2669 const Register index = rdx;
2670 assert_different_registers(method, flags);
2671 assert_different_registers(method, cache, index);
2672 assert_different_registers(itable_index, flags);
2673 assert_different_registers(itable_index, cache, index);
2674 // determine constant pool cache field offsets
2675 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2676 const int method_offset = in_bytes(
2677 ConstantPoolCache::base_offset() +
2678 ((byte_no == f2_byte)
2679 ? ConstantPoolCacheEntry::f2_offset()
2680 : ConstantPoolCacheEntry::f1_offset()));
2681 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2682 ConstantPoolCacheEntry::flags_offset());
2683 // access constant pool cache fields
2684 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2685 ConstantPoolCacheEntry::f2_offset());
2686
2687 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2688 resolve_cache_and_index(byte_no, cache, index, index_size);
2689 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2690
2691 if (itable_index != noreg) {
2692 // pick up itable or appendix index from f2 also:
2693 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2694 }
2695 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2696 }
2697
2698 // The registers cache and index expected to be set before call.
2699 // Correct values of the cache and index registers are preserved.
2700 void TemplateTable::jvmti_post_field_access(Register cache,
2701 Register index,
2702 bool is_static,
2703 bool has_tos) {
2704 if (JvmtiExport::can_post_field_access()) {
2705 // Check to see if a field access watch has been set before we take
2706 // the time to call into the VM.
2707 Label L1;
2708 assert_different_registers(cache, index, rax);
2709 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2710 __ testl(rax,rax);
2711 __ jcc(Assembler::zero, L1);
2712
2713 // cache entry pointer
2714 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2715 __ shll(index, LogBytesPerWord);
2716 __ addptr(cache, index);
2717 if (is_static) {
2718 __ xorptr(rax, rax); // NULL object reference
2719 } else {
2720 __ pop(atos); // Get the object
2721 __ verify_oop(rax);
2722 __ push(atos); // Restore stack state
2723 }
2724 // rax,: object pointer or NULL
2725 // cache: cache entry pointer
2726 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2727 rax, cache);
2728 __ get_cache_and_index_at_bcp(cache, index, 1);
2729 __ bind(L1);
2730 }
2731 }
2732
2733 void TemplateTable::pop_and_check_object(Register r) {
2734 __ pop_ptr(r);
2735 __ null_check(r); // for field access must check obj.
2736 __ verify_oop(r);
2737 }
2738
2739 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2740 transition(vtos, vtos);
2741
2742 const Register cache = rcx;
2743 const Register index = rdx;
2744 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2745 const Register off = rbx;
2746 const Register flags = rax;
2747 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2748
2749 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2750 jvmti_post_field_access(cache, index, is_static, false);
2751 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2752
2753 if (!is_static) pop_and_check_object(obj);
2754
2755 const Address field(obj, off, Address::times_1, 0wordSize);
2756 NOT_LP64(const Address hi(obj, off, Address::times_1, 1wordSize));
2757
2758 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2759
2760 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2761 // Make sure we don't need to mask edx after the above shift
2762 assert(btos == 0, "change code, btos != 0");
2763
2764 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2765
2766 __ jcc(Assembler::notZero, notByte);
2767 // btos
2768 __ load_signed_byte(rax, field);
2769 __ push(btos);
2770 // Rewrite bytecode to be faster
2771 if (!is_static && rc == may_rewrite) {
2772 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2773 }
2774 __ jmp(Done);
2775
2776 __ bind(notByte);
2777 __ cmpl(flags, atos);
2778 __ jcc(Assembler::notEqual, notObj);
2779 // atos
2780 __ load_heap_oop(rax, field);
2781 __ push(atos);
2782 if (!is_static && rc == may_rewrite) {
2783 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2784 }
2785 __ jmp(Done);
2786
2787 __ bind(notObj);
2788 __ cmpl(flags, itos);
2789 __ jcc(Assembler::notEqual, notInt);
2790 // itos
2791 __ movl(rax, field);
2792 __ push(itos);
2793 // Rewrite bytecode to be faster
2794 if (!is_static && rc == may_rewrite) {
2795 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2796 }
2797 __ jmp(Done);
2798
2799 __ bind(notInt);
2800 __ cmpl(flags, ctos);
2801 __ jcc(Assembler::notEqual, notChar);
2802 // ctos
2803 __ load_unsigned_short(rax, field);
2804 __ push(ctos);
2805 // Rewrite bytecode to be faster
2806 if (!is_static && rc == may_rewrite) {
2807 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2808 }
2809 __ jmp(Done);
2810
2811 __ bind(notChar);
2812 __ cmpl(flags, stos);
2813 __ jcc(Assembler::notEqual, notShort);
2814 // stos
2815 __ load_signed_short(rax, field);
2816 __ push(stos);
2817 // Rewrite bytecode to be faster
2818 if (!is_static && rc == may_rewrite) {
2819 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2820 }
2821 __ jmp(Done);
2822
2823 __ bind(notShort);
2824 __ cmpl(flags, ltos);
2825 __ jcc(Assembler::notEqual, notLong);
2826 // ltos
2827
2828 #ifndef _LP64
2829 // Generate code as if volatile. There just aren't enough registers to
2830 // save that information and this code is faster than the test.
2831 __ fild_d(field); // Must load atomically
2832 __ subptr(rsp,2wordSize); // Make space for store
2833 __ fistp_d(Address(rsp,0));
2834 __ pop(rax);
2835 __ pop(rdx);
2836 #else
2837 __ movq(rax, field);
2838 #endif
2839
2840 __ push(ltos);
2841 // Rewrite bytecode to be faster
2842 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2843 __ jmp(Done);
2844
2845 __ bind(notLong);
2846 __ cmpl(flags, ftos);
2847 __ jcc(Assembler::notEqual, notFloat);
2848 // ftos
2849
2850 __ load_float(field);
2851 __ push(ftos);
2852 // Rewrite bytecode to be faster
2853 if (!is_static && rc == may_rewrite) {
2854 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2855 }
2856 __ jmp(Done);
2857
2858 __ bind(notFloat);
2859 #ifdef ASSERT
2860 __ cmpl(flags, dtos);
2861 __ jcc(Assembler::notEqual, notDouble);
2862 #endif
2863 // dtos
2864 __ load_double(field);
2865 __ push(dtos);
2866 // Rewrite bytecode to be faster
2867 if (!is_static && rc == may_rewrite) {
2868 patch_bytecode(Bytecodes::fast_dgetfield, bc, rbx);
2869 }
2870 #ifdef ASSERT
2871 __ jmp(Done);
2872
2873
2874 __ bind(notDouble);
2875 __ stop("Bad state");
2876 #endif
2877
2878 __ bind(Done);
2879 // [jk] not needed currently
2880 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
2881 // Assembler::LoadStore));
2882 }
2883
2884 void TemplateTable::getfield(int byte_no) {
2885 getfield_or_static(byte_no, false);
2886 }
2887
2888 void TemplateTable::nofast_getfield(int byte_no) {
2889 getfield_or_static(byte_no, false, may_not_rewrite);
2890 }
2891
2892 void TemplateTable::getstatic(int byte_no) {
2893 getfield_or_static(byte_no, true);
2894 }
2895
2896
2897 // The registers cache and index expected to be set before call.
2898 // The function may destroy various registers, just not the cache and index registers.
2899 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2900
2901 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
2902 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
2903 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2904 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
2905
2906 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2907
2908 if (JvmtiExport::can_post_field_modification()) {
2909 // Check to see if a field modification watch has been set before
2910 // we take the time to call into the VM.
2911 Label L1;
2912 assert_different_registers(cache, index, rax);
2913 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2914 __ testl(rax, rax);
2915 __ jcc(Assembler::zero, L1);
2916
2917 __ get_cache_and_index_at_bcp(robj, RDX, 1);
2918
2919
2920 if (is_static) {
2921 // Life is simple. Null out the object pointer.
2922 __ xorl(RBX, RBX);
2923
2924 } else {
2925 // Life is harder. The stack holds the value on top, followed by
2926 // the object. We don't know the size of the value, though; it
2927 // could be one or two words depending on its type. As a result,
2928 // we must find the type to determine where the object is.
2929 #ifndef LP64
2930 Label two_word, valsize_known;
2931 #endif
2932 __ movl(RCX, Address(robj, RDX,
2933 Address::times_ptr,
2934 in_bytes(cp_base_offset +
2935 ConstantPoolCacheEntry::flags_offset())));
2936 NOT_LP64( mov(rbx, rsp));
2937 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
2938
2939 // Make sure we don't need to mask rcx after the above shift
2940 ConstantPoolCacheEntry::verify_tos_state_shift();
2941 #ifdef _LP64
2942 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
2943 __ cmpl(c_rarg3, ltos);
2944 __ cmovptr(Assembler::equal,
2945 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
2946 __ cmpl(c_rarg3, dtos);
2947 __ cmovptr(Assembler::equal,
2948 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
2949 #else
2950 __ cmpl(rcx, ltos);
2951 __ jccb(Assembler::equal, two_word);
2952 __ cmpl(rcx, dtos);
2953 __ jccb(Assembler::equal, two_word);
2954 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2955 __ jmpb(valsize_known);
2956
2957 __ bind(two_word);
2958 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2959
2960 __ bind(valsize_known);
2961 // setup object pointer
2962 __ movptr(rbx, Address(rbx, 0));
2963 #endif
2964 }
2965 // cache entry pointer
2966 __ addptr(robj, in_bytes(cp_base_offset));
2967 __ shll(RDX, LogBytesPerWord);
2968 __ addptr(robj, RDX);
2969 // object (tos)
2970 __ mov(RCX, rsp);
2971 // c_rarg1: object pointer set up above (NULL if static)
2972 // c_rarg2: cache entry pointer
2973 // c_rarg3: jvalue object on the stack
2974 __ call_VM(noreg,
2975 CAST_FROM_FN_PTR(address,
2976 InterpreterRuntime::post_field_modification),
2977 RBX, robj, RCX);
2978 __ get_cache_and_index_at_bcp(cache, index, 1);
2979 __ bind(L1);
2980 }
2981 }
2982
2983 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2984 transition(vtos, vtos);
2985
2986 const Register cache = rcx;
2987 const Register index = rdx;
2988 const Register obj = rcx;
2989 const Register off = rbx;
2990 const Register flags = rax;
2991 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2992
2993 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2994 jvmti_post_field_mod(cache, index, is_static);
2995 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2996
2997 // [jk] not needed currently
2998 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
2999 // Assembler::StoreStore));
3000
3001 Label notVolatile, Done;
3002 __ movl(rdx, flags);
3003 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3004 __ andl(rdx, 0x1);
3005
3006 // field addresses
3007 const Address field(obj, off, Address::times_1, 0wordSize);
3008 NOT_LP64( const Address hi(obj, off, Address::times_1, 1wordSize);)
3009
3010 Label notByte, notInt, notShort, notChar,
3011 notLong, notFloat, notObj, notDouble;
3012
3013 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3014
3015 assert(btos == 0, "change code, btos != 0");
3016 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3017 __ jcc(Assembler::notZero, notByte);
3018
3019 // btos
3020 {
3021 __ pop(btos);
3022 if (!is_static) pop_and_check_object(obj);
3023 __ movb(field, rax);
3024 if (!is_static && rc == may_rewrite) {
3025 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3026 }
3027 __ jmp(Done);
3028 }
3029
3030 __ bind(notByte);
3031 __ cmpl(flags, atos);
3032 __ jcc(Assembler::notEqual, notObj);
3033
3034 // atos
3035 {
3036 __ pop(atos);
3037 if (!is_static) pop_and_check_object(obj);
3038 // Store into the field
3039 do_oop_store(_masm, field, rax, _bs->kind(), false);
3040 if (!is_static && rc == may_rewrite) {
3041 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3042 }
3043 __ jmp(Done);
3044 }
3045
3046 __ bind(notObj);
3047 __ cmpl(flags, itos);
3048 __ jcc(Assembler::notEqual, notInt);
3049
3050 // itos
3051 {
3052 __ pop(itos);
3053 if (!is_static) pop_and_check_object(obj);
3054 __ movl(field, rax);
3055 if (!is_static && rc == may_rewrite) {
3056 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3057 }
3058 __ jmp(Done);
3059 }
3060
3061 __ bind(notInt);
3062 __ cmpl(flags, ctos);
3063 __ jcc(Assembler::notEqual, notChar);
3064
3065 // ctos
3066 {
3067 __ pop(ctos);
3068 if (!is_static) pop_and_check_object(obj);
3069 __ movw(field, rax);
3070 if (!is_static && rc == may_rewrite) {
3071 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3072 }
3073 __ jmp(Done);
3074 }
3075
3076 __ bind(notChar);
3077 __ cmpl(flags, stos);
3078 __ jcc(Assembler::notEqual, notShort);
3079
3080 // stos
3081 {
3082 __ pop(stos);
3083 if (!is_static) pop_and_check_object(obj);
3084 __ movw(field, rax);
3085 if (!is_static && rc == may_rewrite) {
3086 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3087 }
3088 __ jmp(Done);
3089 }
3090
3091 __ bind(notShort);
3092 __ cmpl(flags, ltos);
3093 __ jcc(Assembler::notEqual, notLong);
3094
3095 // ltos
3096 #ifdef _LP64
3097 {
3098 __ pop(ltos);
3099 if (!is_static) pop_and_check_object(obj);
3100 __ movq(field, rax);
3101 if (!is_static && rc == may_rewrite) {
3102 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3103 }
3104 __ jmp(Done);
3105 }
3106 #else
3107 {
3108 Label notVolatileLong;
3109 __ testl(rdx, rdx);
3110 __ jcc(Assembler::zero, notVolatileLong);
3111
3112 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3113 if (!is_static) pop_and_check_object(obj);
3114
3115 // Replace with real volatile test
3116 __ push(rdx);
3117 __ push(rax); // Must update atomically with FIST
3118 __ fild_d(Address(rsp,0)); // So load into FPU register
3119 __ fistp_d(field); // and put into memory atomically
3120 __ addptr(rsp, 2wordSize);
3121 // volatile_barrier();
3122 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3123 Assembler::StoreStore));
3124 // Don't rewrite volatile version
3125 __ jmp(notVolatile);
3126
3127 __ bind(notVolatileLong);
3128
3129 __ pop(ltos); // overwrites rdx
3130 if (!is_static) pop_and_check_object(obj);
3131 __ movptr(hi, rdx);
3132 __ movptr(field, rax);
3133 // Don't rewrite to _fast_lputfield for potential volatile case.
3134 __ jmp(notVolatile);
3135 }
3136 #endif // _LP64
3137
3138 __ bind(notLong);
3139 __ cmpl(flags, ftos);
3140 __ jcc(Assembler::notEqual, notFloat);
3141
3142 // ftos
3143 {
3144 __ pop(ftos);
3145 if (!is_static) pop_and_check_object(obj);
3146 __ store_float(field);
3147 if (!is_static && rc == may_rewrite) {
3148 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3149 }
3150 __ jmp(Done);
3151 }
3152
3153 __ bind(notFloat);
3154 #ifdef ASSERT
3155 __ cmpl(flags, dtos);
3156 __ jcc(Assembler::notEqual, notDouble);
3157 #endif
3158
3159 // dtos
3160 {
3161 __ pop(dtos);
3162 if (!is_static) pop_and_check_object(obj);
3163 __ store_double(field);
3164 if (!is_static && rc == may_rewrite) {
3165 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3166 }
3167 }
3168
3169 #ifdef ASSERT
3170 __ jmp(Done);
3171
3172 __ bind(notDouble);
3173 __ stop("Bad state");
3174 #endif
3175
3176 __ bind(Done);
3177
3178 // Check for volatile store
3179 __ testl(rdx, rdx);
3180 __ jcc(Assembler::zero, notVolatile);
3181 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3182 Assembler::StoreStore));
3183 __ bind(notVolatile);
3184 }
3185
3186 void TemplateTable::putfield(int byte_no) {
3187 putfield_or_static(byte_no, false);
3188 }
3189
3190 void TemplateTable::nofast_putfield(int byte_no) {
3191 putfield_or_static(byte_no, false, may_not_rewrite);
3192 }
3193
3194 void TemplateTable::putstatic(int byte_no) {
3195 putfield_or_static(byte_no, true);
3196 }
3197
3198 void TemplateTable::jvmti_post_fast_field_mod() {
3199
3200 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3201
3202 if (JvmtiExport::can_post_field_modification()) {
3203 // Check to see if a field modification watch has been set before
3204 // we take the time to call into the VM.
3205 Label L2;
3206 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3207 __ testl(scratch, scratch);
3208 __ jcc(Assembler::zero, L2);
3209 __ pop_ptr(rbx); // copy the object pointer from tos
3210 __ verify_oop(rbx);
3211 __ push_ptr(rbx); // put the object pointer back on tos
3212 // Save tos values before call_VM() clobbers them. Since we have
3213 // to do it for every data type, we use the saved values as the
3214 // jvalue object.
3215 switch (bytecode()) { // load values into the jvalue object
3216 case Bytecodes::fast_aputfield: __ push_ptr(rax); break;
3217 case Bytecodes::fast_bputfield: // fall through
3218 case Bytecodes::fast_sputfield: // fall through
3219 case Bytecodes::fast_cputfield: // fall through
3220 case Bytecodes::fast_iputfield: __ push_i(rax); break;
3221 case Bytecodes::fast_dputfield: __ push(dtos); break;
3222 case Bytecodes::fast_fputfield: __ push(ftos); break;
3223 case Bytecodes::fast_lputfield: __ push_l(rax); break;
3224
3225 default:
3226 ShouldNotReachHere();
3227 }
3228 __ mov(scratch, rsp); // points to jvalue on the stack
3229 // access constant pool cache entry
3230 LP64_ONLY( get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3231 NOT_LP64( get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3232 __ verify_oop(rbx);
3233 // rbx: object pointer copied above
3234 // c_rarg2: cache entry pointer
3235 // c_rarg3: jvalue object on the stack
3236 LP64_ONLY( call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3237 NOT_LP64( call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3238
3239 switch (bytecode()) { // restore tos values
3240 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3241 case Bytecodes::_fast_bputfield: // fall through
3242 case Bytecodes::_fast_sputfield: // fall through
3243 case Bytecodes::_fast_cputfield: // fall through
3244 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3245 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3246 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3247 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3248 }
3249 __ bind(L2);
3250 }
3251 }
3252
3253 void TemplateTable::fast_storefield(TosState state) {
3254 transition(state, vtos);
3255
3256 ByteSize base = ConstantPoolCache::base_offset();
3257
3258 jvmti_post_fast_field_mod();
3259
3260 // access constant pool cache
3261 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3262
3263 // test for volatile with rdx but rdx is tos register for lputfield.
3264 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3265 in_bytes(base +
3266 ConstantPoolCacheEntry::flags_offset())));
3267
3268 // replace index with field offset from cache entry
3269 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3270 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3271
3272 // [jk] not needed currently
3273 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3274 // Assembler::StoreStore));
3275
3276 Label notVolatile;
3277 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3278 __ andl(rdx, 0x1);
3279
3280 // Get object from stack
3281 pop_and_check_object(rcx);
3282
3283 // field address
3284 const Address field(rcx, rbx, Address::times_1);
3285
3286 // access field
3287 switch (bytecode()) {
3288 case Bytecodes::_fast_aputfield:
3289 do_oop_store(masm, field, rax, bs->kind(), false);
3290 break;
3291 case Bytecodes::fast_lputfield:
3292 #ifdef LP64
3293 __ movq(field, rax);
3294 #else
3295 __ stop("should not be rewritten");
3296 #endif
3297 break;
3298 case Bytecodes::fast_iputfield:
3299 __ movl(field, rax);
3300 break;
3301 case Bytecodes::fast_bputfield:
3302 __ movb(field, rax);
3303 break;
3304 case Bytecodes::fast_sputfield:
3305 // fall through
3306 case Bytecodes::fast_cputfield:
3307 __ movw(field, rax);
3308 break;
3309 case Bytecodes::fast_fputfield:
3310 __ store_float(field);
3311 break;
3312 case Bytecodes::fast_dputfield:
3313 __ store_double(field);
3314 break;
3315 default:
3316 ShouldNotReachHere();
3317 }
3318
3319 // Check for volatile store
3320 __ testl(rdx, rdx);
3321 __ jcc(Assembler::zero, notVolatile);
3322 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3323 Assembler::StoreStore));
3324 __ bind(notVolatile);
3325 }
3326
3327 void TemplateTable::fast_accessfield(TosState state) {
3328 transition(atos, state);
3329
3330 // Do the JVMTI work here to avoid disturbing the register state below
3331 if (JvmtiExport::can_post_field_access()) {
3332 // Check to see if a field access watch has been set before we
3333 // take the time to call into the VM.
3334 Label L1;
3335 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3336 __ testl(rcx, rcx);
3337 __ jcc(Assembler::zero, L1);
3338 // access constant pool cache entry
3339 LP64_ONLY( get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3340 NOT_LP64( get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3341 __ verify_oop(rax);
3342 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3343 LP64_ONLY( mov(c_rarg1, rax));
3344 // c_rarg1: object pointer copied above
3345 // c_rarg2: cache entry pointer
3346 LP64_ONLY( call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3347 NOT_LP64( call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3348 __ pop_ptr(rax); // restore object pointer
3349 __ bind(L1);
3350 }
3351
3352 // access constant pool cache
3353 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3354 // replace index with field offset from cache entry
3355 // [jk] not needed currently
3356 // if (os::is_MP()) {
3357 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3358 // in_bytes(ConstantPoolCache::base_offset() +
3359 // ConstantPoolCacheEntry::flags_offset())));
3360 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3361 // __ andl(rdx, 0x1);
3362 // }
3363 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3364 in_bytes(ConstantPoolCache::base_offset() +
3365 ConstantPoolCacheEntry::f2_offset())));
3366
3367 // rax: object
3368 __ verify_oop(rax);
3369 __ null_check(rax);
3370 Address field(rax, rbx, Address::times_1);
3371
3372 // access field
3373 switch (bytecode()) {
3374 case Bytecodes::_fast_agetfield:
3375 __ load_heap_oop(rax, field);
3376 __ verify_oop(rax);
3377 break;
3378 case Bytecodes::_fast_lgetfield:
3379 #ifdef _LP64
3380 __ movq(rax, field);
3381 #else
3382 __ stop("should not be rewritten");
3383 #endif
3384 break;
3385 case Bytecodes::_fast_igetfield:
3386 __ movl(rax, field);
3387 break;
3388 case Bytecodes::_fast_bgetfield:
3389 __ movsbl(rax, field);
3390 break;
3391 case Bytecodes::_fast_sgetfield:
3392 __ load_signed_short(rax, field);
3393 break;
3394 case Bytecodes::_fast_cgetfield:
3395 __ load_unsigned_short(rax, field);
3396 break;
3397 case Bytecodes::_fast_fgetfield:
3398 __ load_float(field);
3399 break;
3400 case Bytecodes::_fast_dgetfield:
3401 __ load_double(field);
3402 break;
3403 default:
3404 ShouldNotReachHere();
3405 }
3406 // [jk] not needed currently
3407 // if (os::is_MP()) {
3408 // Label notVolatile;
3409 // __ testl(rdx, rdx);
3410 // __ jcc(Assembler::zero, notVolatile);
3411 // __ membar(Assembler::LoadLoad);
3412 // __ bind(notVolatile);
3413 //};
3414 }
3415
3416 void TemplateTable::fast_xaccess(TosState state) {
3417 transition(vtos, state);
3418
3419 // get receiver
3420 __ movptr(rax, aaddress(0));
3421 // access constant pool cache
3422 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3423 __ movptr(rbx,
3424 Address(rcx, rdx, Address::times_ptr,
3425 in_bytes(ConstantPoolCache::base_offset() +
3426 ConstantPoolCacheEntry::f2_offset())));
3427 // make sure exception is reported in correct bcp range (getfield is
3428 // next instruction)
3429 __ increment(rbcp);
3430 __ null_check(rax);
3431 const Address field = Address(rax, rbx, Address::times_1, 0wordSize);
3432 switch (state) {
3433 case itos:
3434 __ movl(rax, field);
3435 break;
3436 case atos:
3437 __ load_heap_oop(rax, field);
3438 __ verify_oop(rax);
3439 break;
3440 case ftos:
3441 __ load_float(field);
3442 break;
3443 default:
3444 ShouldNotReachHere();
3445 }
3446
3447 // [jk] not needed currently
3448 // if (os::is_MP()) {
3449 // Label notVolatile;
3450 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3451 // in_bytes(ConstantPoolCache::base_offset() +
3452 // ConstantPoolCacheEntry::flags_offset())));
3453 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3454 // __ testl(rdx, 0x1);
3455 // __ jcc(Assembler::zero, notVolatile);
3456 // __ membar(Assembler::LoadLoad);
3457 // __ bind(notVolatile);
3458 // }
3459
3460 __ decrement(rbcp);
3461 }
3462
3463 //-----------------------------------------------------------------------------
3464 // Calls
3465
3466 void TemplateTable::count_calls(Register method, Register temp) {
3467 // implemented elsewhere
3468 ShouldNotReachHere();
3469 }
3470
3471 void TemplateTable::prepare_invoke(int byte_no,
3472 Register method, // linked method (or i-klass)
3473 Register index, // itable index, MethodType, etc.
3474 Register recv, // if caller wants to see it
3475 Register flags // if caller wants to test it
3476 ) {
3477 // determine flags
3478 const Bytecodes::Code code = bytecode();
3479 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3480 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3481 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3482 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3483 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3484 const bool load_receiver = (recv != noreg);
3485 const bool save_flags = (flags != noreg);
3486 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3487 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3488 assert(flags == noreg || flags == rdx, "");
3489 assert(recv == noreg || recv == rcx, "");
3490
3491 // setup registers & access constant pool cache
3492 if (recv == noreg) recv = rcx;
3493 if (flags == noreg) flags = rdx;
3494 assert_different_registers(method, index, recv, flags);
3495
3496 // save 'interpreter return address'
3497 __ save_bcp();
3498
3499 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3500
3501 // maybe push appendix to arguments (just before return address)
3502 if (is_invokedynamic || is_invokehandle) {
3503 Label L_no_push;
3504 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3505 __ jcc(Assembler::zero, L_no_push);
3506 // Push the appendix as a trailing parameter.
3507 // This must be done before we get the receiver,
3508 // since the parameter_size includes it.
3509 __ push(rbx);
3510 __ mov(rbx, index);
3511 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3512 __ load_resolved_reference_at_index(index, rbx);
3513 __ pop(rbx);
3514 __ push(index); // push appendix (MethodType, CallSite, etc.)
3515 __ bind(L_no_push);
3516 }
3517
3518 // load receiver if needed (after appendix is pushed so parameter size is correct)
3519 // Note: no return address pushed yet
3520 if (load_receiver) {
3521 __ movl(recv, flags);
3522 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3523 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3524 const int receiver_is_at_end = -1; // back off one slot to get receiver
3525 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3526 __ movptr(recv, recv_addr);
3527 __ verify_oop(recv);
3528 }
3529
3530 if (save_flags) {
3531 __ movl(rbcp, flags);
3532 }
3533
3534 // compute return type
3535 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3536 // Make sure we don't need to mask flags after the above shift
3537 ConstantPoolCacheEntry::verify_tos_state_shift();
3538 // load return address
3539 {
3540 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3541 ExternalAddress table(table_addr);
3542 LP64_ONLY(__ lea(rscratch1, table));
3543 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3544 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3545 }
3546
3547 // push return address
3548 __ push(flags);
3549
3550 // Restore flags value from the constant pool cache, and restore rsi
3551 // for later null checks. r13 is the bytecode pointer
3552 if (save_flags) {
3553 __ movl(flags, rbcp);
3554 __ restore_bcp();
3555 }
3556 }
3557
3558 void TemplateTable::invokevirtual_helper(Register index,
3559 Register recv,
3560 Register flags) {
3561 // Uses temporary registers rax, rdx
3562 assert_different_registers(index, recv, rax, rdx);
3563 assert(index == rbx, "");
3564 assert(recv == rcx, "");
3565
3566 // Test for an invoke of a final method
3567 Label notFinal;
3568 __ movl(rax, flags);
3569 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3570 __ jcc(Assembler::zero, notFinal);
3571
3572 const Register method = index; // method must be rbx
3573 assert(method == rbx,
3574 "Method* must be rbx for interpreter calling convention");
3575
3576 // do the call - the index is actually the method to call
3577 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3578
3579 // It's final, need a null check here!
3580 __ null_check(recv);
3581
3582 // profile this call
3583 __ profile_final_call(rax);
3584 __ profile_arguments_type(rax, method, rbcp, true);
3585
3586 __ jump_from_interpreted(method, rax);
3587
3588 __ bind(notFinal);
3589
3590 // get receiver klass
3591 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3592 __ load_klass(rax, recv);
3593
3594 // profile this call
3595 __ profile_virtual_call(rax, rlocals, rdx);
3596 // get target Method* & entry point
3597 __ lookup_virtual_method(rax, index, method);
3598 #if INCLUDE_JVMCI
3599 __ profile_called_method(method, rdx, rbcp);
3600 #endif
3601
3602 __ profile_arguments_type(rdx, method, rbcp, true);
3603 __ jump_from_interpreted(method, rdx);
3604 }
3605
3606 void TemplateTable::invokevirtual(int byte_no) {
3607 transition(vtos, vtos);
3608 assert(byte_no == f2_byte, "use this argument");
3609 prepare_invoke(byte_no,
3610 rbx, // method or vtable index
3611 noreg, // unused itable index
3612 rcx, rdx); // recv, flags
3613
3614 // rbx: index
3615 // rcx: receiver
3616 // rdx: flags
3617
3618 invokevirtual_helper(rbx, rcx, rdx);
3619 }
3620
3621 void TemplateTable::invokespecial(int byte_no) {
3622 transition(vtos, vtos);
3623 assert(byte_no == f1_byte, "use this argument");
3624 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3625 rcx); // get receiver also for null check
3626 __ verify_oop(rcx);
3627 __ null_check(rcx);
3628 // do the call
3629 __ profile_call(rax);
3630 __ profile_arguments_type(rax, rbx, rbcp, false);
3631 __ jump_from_interpreted(rbx, rax);
3632 }
3633
3634 void TemplateTable::invokestatic(int byte_no) {
3635 transition(vtos, vtos);
3636 assert(byte_no == f1_byte, "use this argument");
3637 prepare_invoke(byte_no, rbx); // get f1 Method*
3638 // do the call
3639 __ profile_call(rax);
3640 __ profile_arguments_type(rax, rbx, rbcp, false);
3641 __ jump_from_interpreted(rbx, rax);
3642 }
3643
3644
3645 void TemplateTable::fast_invokevfinal(int byte_no) {
3646 transition(vtos, vtos);
3647 assert(byte_no == f2_byte, "use this argument");
3648 __ stop("fast_invokevfinal not used on x86");
3649 }
3650
3651
3652 void TemplateTable::invokeinterface(int byte_no) {
3653 transition(vtos, vtos);
3654 assert(byte_no == f1_byte, "use this argument");
3655 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3656 rcx, rdx); // recv, flags
3657
3658 // rax: interface klass (from f1)
3659 // rbx: itable index (from f2)
3660 // rcx: receiver
3661 // rdx: flags
3662
3663 // Special case of invokeinterface called for virtual method of
3664 // java.lang.Object. See cpCacheOop.cpp for details.
3665 // This code isn't produced by javac, but could be produced by
3666 // another compliant java compiler.
3667 Label notMethod;
3668 __ movl(rlocals, rdx);
3669 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3670
3671 __ jcc(Assembler::zero, notMethod);
3672
3673 invokevirtual_helper(rbx, rcx, rdx);
3674 __ bind(notMethod);
3675
3676 // Get receiver klass into rdx - also a null check
3677 __ restore_locals(); // restore r14
3678 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3679 __ load_klass(rdx, rcx);
3680
3681 // profile this call
3682 __ profile_virtual_call(rdx, rbcp, rlocals);
3683
3684 Label no_such_interface, no_such_method;
3685
3686 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3687 rdx, rax, rbx,
3688 // outputs: method, scan temp. reg
3689 rbx, rbcp,
3690 no_such_interface);
3691
3692 // rbx: Method* to call
3693 // rcx: receiver
3694 // Check for abstract method error
3695 // Note: This should be done more efficiently via a throw_abstract_method_error
3696 // interpreter entry point and a conditional jump to it in case of a null
3697 // method.
3698 __ testptr(rbx, rbx);
3699 __ jcc(Assembler::zero, no_such_method);
3700
3701 #if INCLUDE_JVMCI
3702 __ profile_called_method(rbx, rbcp, rdx);
3703 #endif
3704 __ profile_arguments_type(rdx, rbx, rbcp, true);
3705
3706 // do the call
3707 // rcx: receiver
3708 // rbx,: Method*
3709 __ jump_from_interpreted(rbx, rdx);
3710 __ should_not_reach_here();
3711
3712 // exception handling code follows...
3713 // note: must restore interpreter registers to canonical
3714 // state for exception handling to work correctly!
3715
3716 __ bind(no_such_method);
3717 // throw exception
3718 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3719 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3720 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3721 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3722 // the call_VM checks for exception, so we should never return here.
3723 __ should_not_reach_here();
3724
3725 __ bind(no_such_interface);
3726 // throw exception
3727 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3728 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3729 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3730 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3731 InterpreterRuntime::throw_IncompatibleClassChangeError));
3732 // the call_VM checks for exception, so we should never return here.
3733 __ should_not_reach_here();
3734 }
3735
3736 void TemplateTable::invokehandle(int byte_no) {
3737 transition(vtos, vtos);
3738 assert(byte_no == f1_byte, "use this argument");
3739 const Register rbx_method = rbx;
3740 const Register rax_mtype = rax;
3741 const Register rcx_recv = rcx;
3742 const Register rdx_flags = rdx;
3743
3744 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3745 __ verify_method_ptr(rbx_method);
3746 __ verify_oop(rcx_recv);
3747 __ null_check(rcx_recv);
3748
3749 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3750 // rbx: MH.invokeExact_MT method (from f2)
3751
3752 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3753
3754 // FIXME: profile the LambdaForm also
3755 __ profile_final_call(rax);
3756 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3757
3758 __ jump_from_interpreted(rbx_method, rdx);
3759 }
3760
3761 void TemplateTable::invokedynamic(int byte_no) {
3762 transition(vtos, vtos);
3763 assert(byte_no == f1_byte, "use this argument");
3764
3765 const Register rbx_method = rbx;
3766 const Register rax_callsite = rax;
3767
3768 prepare_invoke(byte_no, rbx_method, rax_callsite);
3769
3770 // rax: CallSite object (from cpool->resolved_references[f1])
3771 // rbx: MH.linkToCallSite method (from f2)
3772
3773 // Note: rax_callsite is already pushed by prepare_invoke
3774
3775 // %%% should make a type profile for any invokedynamic that takes a ref argument
3776 // profile this call
3777 __ profile_call(rbcp);
3778 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3779
3780 __ verify_oop(rax_callsite);
3781
3782 __ jump_from_interpreted(rbx_method, rdx);
3783 }
3784
3785 //-----------------------------------------------------------------------------
3786 // Allocation
3787
3788 void TemplateTable::_new() {
3789 transition(vtos, atos);
3790 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3791 Label slow_case;
3792 Label slow_case_no_pop;
3793 Label done;
3794 Label initialize_header;
3795 Label initialize_object; // including clearing the fields
3796 Label allocate_shared;
3797
3798 __ get_cpool_and_tags(rcx, rax);
3799
3800 // Make sure the class we're about to instantiate has been resolved.
3801 // This is done before loading InstanceKlass to be consistent with the order
3802 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3803 const int tags_offset = Array::base_offset_in_bytes();
3804 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3805 __ jcc(Assembler::notEqual, slow_case_no_pop);
3806
3807 // get InstanceKlass
3808 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool)));
3809 __ push(rcx); // save the contexts of klass for initializing the header
3810
3811 // make sure klass is initialized & doesn't have finalizer
3812 // make sure klass is fully initialized
3813 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3814 __ jcc(Assembler::notEqual, slow_case);
3815
3816 // get instance_size in InstanceKlass (scaled to a count of bytes)
3817 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3818 // test to see if it has a finalizer or is malformed in some way
3819 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3820 __ jcc(Assembler::notZero, slow_case);
3821
3822 //
3823 // Allocate the instance
3824 // 1) Try to allocate in the TLAB
3825 // 2) if fail and the object is large allocate in the shared Eden
3826 // 3) if the above fails (or is not applicable), go to a slow case
3827 // (creates a new TLAB, etc.)
3828
3829 const bool allow_shared_alloc =
3830 Universe::heap()->supports_inline_contig_alloc();
3831
3832 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
3833 #ifndef _LP64
3834 if (UseTLAB || allow_shared_alloc) {
3835 __ get_thread(thread);
3836 }
3837 #endif // _LP64
3838
3839 if (UseTLAB) {
3840 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3841 __ lea(rbx, Address(rax, rdx, Address::times_1));
3842 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3843 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3844 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3845 if (ZeroTLAB) {
3846 // the fields have been already cleared
3847 __ jmp(initialize_header);
3848 } else {
3849 // initialize both the header and fields
3850 __ jmp(initialize_object);
3851 }
3852 }
3853
3854 // Allocation in the shared Eden, if allowed.
3855 //
3856 // rdx: instance size in bytes
3857 if (allow_shared_alloc) {
3858 __ bind(allocate_shared);
3859
3860 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3861 ExternalAddress heap_end((address)Universe::heap()->end_addr());
3862
3863 Label retry;
3864 __ bind(retry);
3865 __ movptr(rax, heap_top);
3866 __ lea(rbx, Address(rax, rdx, Address::times_1));
3867 __ cmpptr(rbx, heap_end);
3868 __ jcc(Assembler::above, slow_case);
3869
3870 // Compare rax, with the top addr, and if still equal, store the new
3871 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3872 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3873 //
3874 // rax,: object begin
3875 // rbx,: object end
3876 // rdx: instance size in bytes
3877 __ locked_cmpxchgptr(rbx, heap_top);
3878
3879 // if someone beat us on the allocation, try again, otherwise continue
3880 __ jcc(Assembler::notEqual, retry);
3881
3882 __ incr_allocated_bytes(thread, rdx, 0);
3883 }
3884
3885 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3886 // The object is initialized before the header. If the object size is
3887 // zero, go directly to the header initialization.
3888 __ bind(initialize_object);
3889 __ decrement(rdx, sizeof(oopDesc));
3890 __ jcc(Assembler::zero, initialize_header);
3891
3892 // Initialize topmost object field, divide rdx by 8, check if odd and
3893 // test if zero.
3894 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3895 __ shrl(rdx, LogBytesPerLong); // divide by 2oopSize and set carry flag if odd
3896
3897 // rdx must have been multiple of 8
3898 #ifdef ASSERT
3899 // make sure rdx was multiple of 8
3900 Label L;
3901 // Ignore partial flag stall after shrl() since it is debug VM
3902 __ jccb(Assembler::carryClear, L);
3903 __ stop("object size is not multiple of 2 - adjust this code");
3904 __ bind(L);
3905 // rdx must be > 0, no extra check needed here
3906 #endif
3907
3908 // initialize remaining object fields: rdx was a multiple of 8
3909 { Label loop;
3910 __ bind(loop);
3911 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1oopSize), rcx);
3912 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2oopSize), rcx));
3913 __ decrement(rdx);
3914 __ jcc(Assembler::notZero, loop);
3915 }
3916
3917 // initialize object header only.
3918 __ bind(initialize_header);
3919 if (UseBiasedLocking) {
3920 __ pop(rcx); // get saved klass back in the register.
3921 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
3922 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3923 } else {
3924 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3925 (intptr_t)markOopDesc::prototype()); // header
3926 __ pop(rcx); // get saved klass back in the register.
3927 }
3928 #ifdef _LP64
3929 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
3930 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
3931 #endif
3932 __ store_klass(rax, rcx); // klass
3933
3934 {
3935 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3936 // Trigger dtrace event for fastpath
3937 __ push(atos);
3938 __ call_VM_leaf(
3939 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3940 __ pop(atos);
3941 }
3942
3943 __ jmp(done);
3944 }
3945
3946 // slow case
3947 __ bind(slow_case);
3948 __ pop(rcx); // restore stack pointer to what it was when we came in.
3949 __ bind(slow_case_no_pop);
3950
3951 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
3952 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3953
3954 __ get_constant_pool(rarg1);
3955 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
3956 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
3957 __ verify_oop(rax);
3958
3959 // continue
3960 __ bind(done);
3961 }
3962
3963 void TemplateTable::newarray() {
3964 transition(itos, atos);
3965 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
3966 __ load_unsigned_byte(rarg1, at_bcp(1));
3967 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
3968 rarg1, rax);
3969 }
3970
3971 void TemplateTable::anewarray() {
3972 transition(itos, atos);
3973
3974 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
3975 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3976
3977 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
3978 __ get_constant_pool(rarg1);
3979 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
3980 rarg1, rarg2, rax);
3981 }
3982
3983 void TemplateTable::arraylength() {
3984 transition(atos, itos);
3985 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3986 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3987 }
3988
3989 void TemplateTable::checkcast() {
3990 transition(atos, atos);
3991 Label done, is_null, ok_is_subtype, quicked, resolved;
3992 __ testptr(rax, rax); // object is in rax
3993 __ jcc(Assembler::zero, is_null);
3994
3995 // Get cpool & tags index
3996 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
3997 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
3998 // See if bytecode has already been quicked
3999 __ cmpb(Address(rdx, rbx,
4000 Address::times_1,
4001 Array::base_offset_in_bytes()),
4002 JVM_CONSTANT_Class);
4003 __ jcc(Assembler::equal, quicked);
4004 __ push(atos); // save receiver for result, and for GC
4005 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4006
4007 // vm_result_2 has metadata result
4008 #ifndef _LP64
4009 // borrow rdi from locals
4010 __ get_thread(rdi);
4011 __ get_vm_result_2(rax, rdi);
4012 __ restore_locals();
4013 #else
4014 __ get_vm_result_2(rax, r15_thread);
4015 #endif
4016
4017 __ pop_ptr(rdx); // restore receiver
4018 __ jmpb(resolved);
4019
4020 // Get superklass in rax and subklass in rbx
4021 __ bind(quicked);
4022 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4023 __ movptr(rax, Address(rcx, rbx,
4024 Address::times_ptr, sizeof(ConstantPool)));
4025
4026 __ bind(resolved);
4027 __ load_klass(rbx, rdx);
4028
4029 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4030 // Superklass in rax. Subklass in rbx.
4031 __ gen_subtype_check(rbx, ok_is_subtype);
4032
4033 // Come here on failure
4034 __ push_ptr(rdx);
4035 // object is at TOS
4036 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4037
4038 // Come here on success
4039 __ bind(ok_is_subtype);
4040 __ mov(rax, rdx); // Restore object in rdx
4041
4042 // Collect counts on whether this check-cast sees NULLs a lot or not.
4043 if (ProfileInterpreter) {
4044 __ jmp(done);
4045 __ bind(is_null);
4046 __ profile_null_seen(rcx);
4047 } else {
4048 __ bind(is_null); // same as 'done'
4049 }
4050 __ bind(done);
4051 }
4052
4053 void TemplateTable::instanceof() {
4054 transition(atos, itos);
4055 Label done, is_null, ok_is_subtype, quicked, resolved;
4056 __ testptr(rax, rax);
4057 __ jcc(Assembler::zero, is_null);
4058
4059 // Get cpool & tags index
4060 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4061 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4062 // See if bytecode has already been quicked
4063 __ cmpb(Address(rdx, rbx,
4064 Address::times_1,
4065 Array::base_offset_in_bytes()),
4066 JVM_CONSTANT_Class);
4067 __ jcc(Assembler::equal, quicked);
4068
4069 __ push(atos); // save receiver for result, and for GC
4070 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4071 // vm_result_2 has metadata result
4072
4073 #ifndef _LP64
4074 // borrow rdi from locals
4075 __ get_thread(rdi);
4076 __ get_vm_result_2(rax, rdi);
4077 __ restore_locals();
4078 #else
4079 __ get_vm_result_2(rax, r15_thread);
4080 #endif
4081
4082 __ pop_ptr(rdx); // restore receiver
4083 __ verify_oop(rdx);
4084 __ load_klass(rdx, rdx);
4085 __ jmpb(resolved);
4086
4087 // Get superklass in rax and subklass in rdx
4088 __ bind(quicked);
4089 __ load_klass(rdx, rax);
4090 __ movptr(rax, Address(rcx, rbx,
4091 Address::times_ptr, sizeof(ConstantPool)));
4092
4093 __ bind(resolved);
4094
4095 // Generate subtype check. Blows rcx, rdi
4096 // Superklass in rax. Subklass in rdx.
4097 __ gen_subtype_check(rdx, ok_is_subtype);
4098
4099 // Come here on failure
4100 __ xorl(rax, rax);
4101 __ jmpb(done);
4102 // Come here on success
4103 __ bind(ok_is_subtype);
4104 __ movl(rax, 1);
4105
4106 // Collect counts on whether this test sees NULLs a lot or not.
4107 if (ProfileInterpreter) {
4108 __ jmp(done);
4109 __ bind(is_null);
4110 __ profile_null_seen(rcx);
4111 } else {
4112 __ bind(is_null); // same as 'done'
4113 }
4114 __ bind(done);
4115 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4116 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4117 }
4118
4119
4120 //----------------------------------------------------------------------------------------------------
4121 // Breakpoints
4122 void TemplateTable::_breakpoint() {
4123 // Note: We get here even if we are single stepping..
4124 // jbug insists on setting breakpoints at every bytecode
4125 // even if we are in single step mode.
4126
4127 transition(vtos, vtos);
4128
4129 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4130
4131 // get the unpatched byte code
4132 __ get_method(rarg);
4133 __ call_VM(noreg,
4134 CAST_FROM_FN_PTR(address,
4135 InterpreterRuntime::get_original_bytecode_at),
4136 rarg, rbcp);
4137 __ mov(rbx, rax); // why?
4138
4139 // post the breakpoint event
4140 __ get_method(rarg);
4141 __ call_VM(noreg,
4142 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4143 rarg, rbcp);
4144
4145 // complete the execution of original bytecode
4146 __ dispatch_only_normal(vtos);
4147 }
4148
4149 //-----------------------------------------------------------------------------
4150 // Exceptions
4151
4152 void TemplateTable::athrow() {
4153 transition(atos, vtos);
4154 __ null_check(rax);
4155 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4156 }
4157
4158 //-----------------------------------------------------------------------------
4159 // Synchronization
4160 //
4161 // Note: monitorenter & exit are symmetric routines; which is reflected
4162 // in the assembly code structure as well
4163 //
4164 // Stack layout:
4165 //
4166 // [expressions ] <--- rsp = expression stack top
4167 // ..
4168 // [expressions ]
4169 // [monitor entry] <--- monitor block top = expression stack bot
4170 // ..
4171 // [monitor entry]
4172 // [frame data ] <--- monitor block bot
4173 // ...
4174 // [saved rbp ] <--- rbp
4175 void TemplateTable::monitorenter() {
4176 transition(atos, vtos);
4177
4178 // check for NULL object
4179 __ null_check(rax);
4180
4181 const Address monitor_block_top(
4182 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4183 const Address monitor_block_bot(
4184 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4185 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4186
4187 Label allocated;
4188
4189 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4190 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4191 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4192
4193 // initialize entry pointer
4194 __ xorl(rmon, rmon); // points to free slot or NULL
4195
4196 // find a free slot in the monitor block (result in rmon)
4197 {
4198 Label entry, loop, exit;
4199 __ movptr(rtop, monitor_block_top); // points to current entry,
4200 // starting with top-most entry
4201 __ lea(rbot, monitor_block_bot); // points to word before bottom
4202 // of monitor block
4203 __ jmpb(entry);
4204
4205 __ bind(loop);
4206 // check if current entry is used
4207 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4208 // if not used then remember entry in rmon
4209 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4210 // check if current entry is for same object
4211 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4212 // if same object then stop searching
4213 __ jccb(Assembler::equal, exit);
4214 // otherwise advance to next entry
4215 __ addptr(rtop, entry_size);
4216 __ bind(entry);
4217 // check if bottom reached
4218 __ cmpptr(rtop, rbot);
4219 // if not at bottom then check this entry
4220 __ jcc(Assembler::notEqual, loop);
4221 __ bind(exit);
4222 }
4223
4224 __ testptr(rmon, rmon); // check if a slot has been found
4225 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4226
4227 // allocate one if there's no free slot
4228 {
4229 Label entry, loop;
4230 // 1. compute new pointers // rsp: old expression stack top
4231 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4232 __ subptr(rsp, entry_size); // move expression stack top
4233 __ subptr(rmon, entry_size); // move expression stack bottom
4234 __ mov(rtop, rsp); // set start value for copy loop
4235 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4236 __ jmp(entry);
4237 // 2. move expression stack contents
4238 __ bind(loop);
4239 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4240 // word from old location
4241 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4242 __ addptr(rtop, wordSize); // advance to next word
4243 __ bind(entry);
4244 __ cmpptr(rtop, rmon); // check if bottom reached
4245 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4246 // copy next word
4247 }
4248
4249 // call run-time routine
4250 // rmon: points to monitor entry
4251 __ bind(allocated);
4252
4253 // Increment bcp to point to the next bytecode, so exception
4254 // handling for async. exceptions work correctly.
4255 // The object has already been poped from the stack, so the
4256 // expression stack looks correct.
4257 __ increment(rbcp);
4258
4259 // store object
4260 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4261 __ lock_object(rmon);
4262
4263 // check to make sure this monitor doesn't cause stack overflow after locking
4264 __ save_bcp(); // in case of exception
4265 __ generate_stack_overflow_check(0);
4266
4267 // The bcp has already been incremented. Just need to dispatch to
4268 // next instruction.
4269 __ dispatch_next(vtos);
4270 }
4271
4272 void TemplateTable::monitorexit() {
4273 transition(atos, vtos);
4274
4275 // check for NULL object
4276 __ null_check(rax);
4277
4278 const Address monitor_block_top(
4279 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4280 const Address monitor_block_bot(
4281 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4282 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4283
4284 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4285 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4286
4287 Label found;
4288
4289 // find matching slot
4290 {
4291 Label entry, loop;
4292 __ movptr(rtop, monitor_block_top); // points to current entry,
4293 // starting with top-most entry
4294 __ lea(rbot, monitor_block_bot); // points to word before bottom
4295 // of monitor block
4296 __ jmpb(entry);
4297
4298 __ bind(loop);
4299 // check if current entry is for same object
4300 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4301 // if same object then stop searching
4302 __ jcc(Assembler::equal, found);
4303 // otherwise advance to next entry
4304 __ addptr(rtop, entry_size);
4305 __ bind(entry);
4306 // check if bottom reached
4307 __ cmpptr(rtop, rbot);
4308 // if not at bottom then check this entry
4309 __ jcc(Assembler::notEqual, loop);
4310 }
4311
4312 // error handling. Unlocking was not block-structured
4313 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4314 InterpreterRuntime::throw_illegal_monitor_state_exception));
4315 __ should_not_reach_here();
4316
4317 // call run-time routine
4318 __ bind(found);
4319 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4320 __ unlock_object(rtop);
4321 __ pop_ptr(rax); // discard object
4322 }
4323
4324 // Wide instructions
4325 void TemplateTable::wide() {
4326 transition(vtos, vtos);
4327 __ load_unsigned_byte(rbx, at_bcp(1));
4328 ExternalAddress wtable((address)Interpreter::_wentry_point);
4329 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4330 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4331 }
4332
4333 // Multi arrays
4334 void TemplateTable::multianewarray() {
4335 transition(vtos, atos);
4336
4337 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4338 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4339 // last dim is on top of stack; we want address of first one:
4340 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1wordsize
4341 // the latter wordSize to point to the beginning of the array.
4342 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4343 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4344 __ load_unsigned_byte(rbx, at_bcp(3));
4345 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4346 }
4347 #endif / !CC_INTERP */
4348