1 /*
2 * Copyright (c) 1997, 2018, 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 "compiler/disassembler.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "interpreter/interp_masm.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "memory/universe.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/objArrayKlass.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/methodHandles.hpp"
37 #include "runtime/frame.inline.hpp"
38 #include "runtime/safepointMechanism.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/stubRoutines.hpp"
41 #include "runtime/synchronizer.hpp"
42 #include "utilities/macros.hpp"
43
44 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
45
46 // Global Register Names
47 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
48 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
49
50 // Platform-dependent initialization
51 void TemplateTable::pd_initialize() {
52 // No x86 specific initialization
53 }
54
55 // Address Computation: local variables
56 static inline Address iaddress(int n) {
57 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
58 }
59
60 static inline Address laddress(int n) {
61 return iaddress(n + 1);
62 }
63
64 #ifndef _LP64
65 static inline Address haddress(int n) {
66 return iaddress(n + 0);
67 }
68 #endif
69
70 static inline Address faddress(int n) {
71 return iaddress(n);
72 }
73
74 static inline Address daddress(int n) {
75 return laddress(n);
76 }
77
78 static inline Address aaddress(int n) {
79 return iaddress(n);
80 }
81
82 static inline Address iaddress(Register r) {
83 return Address(rlocals, r, Address::times_ptr);
84 }
85
86 static inline Address laddress(Register r) {
87 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
88 }
89
90 #ifndef _LP64
91 static inline Address haddress(Register r) {
92 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
93 }
94 #endif
95
96 static inline Address faddress(Register r) {
97 return iaddress(r);
98 }
99
100 static inline Address daddress(Register r) {
101 return laddress(r);
102 }
103
104 static inline Address aaddress(Register r) {
105 return iaddress(r);
106 }
107
108
109 // expression stack
110 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
111 // data beyond the rsp which is potentially unsafe in an MT environment;
112 // an interrupt may overwrite that data.)
113 static inline Address at_rsp () {
114 return Address(rsp, 0);
115 }
116
117 // At top of Java expression stack which may be different than esp(). It
118 // isn't for category 1 objects.
119 static inline Address at_tos () {
120 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
121 }
122
123 static inline Address at_tos_p1() {
124 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
125 }
126
127 static inline Address at_tos_p2() {
128 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
129 }
130
131 // Condition conversion
132 static Assembler::Condition j_not(TemplateTable::Condition cc) {
133 switch (cc) {
134 case TemplateTable::equal : return Assembler::notEqual;
135 case TemplateTable::not_equal : return Assembler::equal;
136 case TemplateTable::less : return Assembler::greaterEqual;
137 case TemplateTable::less_equal : return Assembler::greater;
138 case TemplateTable::greater : return Assembler::lessEqual;
139 case TemplateTable::greater_equal: return Assembler::less;
140 }
141 ShouldNotReachHere();
142 return Assembler::zero;
143 }
144
145
146
147 // Miscelaneous helper routines
148 // Store an oop (or NULL) at the address described by obj.
149 // If val == noreg this means store a NULL
150
151
152 static void do_oop_store(InterpreterMacroAssembler* _masm,
153 Address dst,
154 Register val,
155 DecoratorSet decorators = 0) {
156 assert(val == noreg || val == rax, "parameter is just for looks");
157 __ store_heap_oop(dst, val, rdx, rbx, decorators);
158 }
159
160 static void do_oop_load(InterpreterMacroAssembler* _masm,
161 Address src,
162 Register dst,
163 DecoratorSet decorators = 0) {
164 __ load_heap_oop(dst, src, rdx, rbx, decorators);
165 }
166
167 Address TemplateTable::at_bcp(int offset) {
168 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
169 return Address(rbcp, offset);
170 }
171
172
173 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
174 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
175 int byte_no) {
176 if (!RewriteBytecodes) return;
177 Label L_patch_done;
178
179 switch (bc) {
180 case Bytecodes::_fast_qputfield:
181 case Bytecodes::_fast_aputfield:
182 case Bytecodes::_fast_bputfield:
183 case Bytecodes::_fast_zputfield:
184 case Bytecodes::_fast_cputfield:
185 case Bytecodes::_fast_dputfield:
186 case Bytecodes::_fast_fputfield:
187 case Bytecodes::_fast_iputfield:
188 case Bytecodes::_fast_lputfield:
189 case Bytecodes::_fast_sputfield:
190 {
191 // We skip bytecode quickening for putfield instructions when
192 // the put_code written to the constant pool cache is zero.
193 // This is required so that every execution of this instruction
194 // calls out to InterpreterRuntime::resolve_get_put to do
195 // additional, required work.
196 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
197 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
198 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
199 __ movl(bc_reg, bc);
200 __ cmpl(temp_reg, (int) 0);
201 __ jcc(Assembler::zero, L_patch_done); // don't patch
202 }
203 break;
204 default:
205 assert(byte_no == -1, "sanity");
206 // the pair bytecodes have already done the load.
207 if (load_bc_into_bc_reg) {
208 __ movl(bc_reg, bc);
209 }
210 }
211
212 if (JvmtiExport::can_post_breakpoint()) {
213 Label L_fast_patch;
214 // if a breakpoint is present we can't rewrite the stream directly
215 __ movzbl(temp_reg, at_bcp(0));
216 __ cmpl(temp_reg, Bytecodes::_breakpoint);
217 __ jcc(Assembler::notEqual, L_fast_patch);
218 __ get_method(temp_reg);
219 // Let breakpoint table handling rewrite to quicker bytecode
220 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
221 #ifndef ASSERT
222 __ jmpb(L_patch_done);
223 #else
224 __ jmp(L_patch_done);
225 #endif
226 __ bind(L_fast_patch);
227 }
228
229 #ifdef ASSERT
230 Label L_okay;
231 __ load_unsigned_byte(temp_reg, at_bcp(0));
232 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
233 __ jcc(Assembler::equal, L_okay);
234 __ cmpl(temp_reg, bc_reg);
235 __ jcc(Assembler::equal, L_okay);
236 __ stop("patching the wrong bytecode");
237 __ bind(L_okay);
238 #endif
239
240 // patch bytecode
241 __ movb(at_bcp(0), bc_reg);
242 __ bind(L_patch_done);
243 }
244 // Individual instructions
245
246
247 void TemplateTable::nop() {
248 transition(vtos, vtos);
249 // nothing to do
250 }
251
252 void TemplateTable::shouldnotreachhere() {
253 transition(vtos, vtos);
254 __ stop("shouldnotreachhere bytecode");
255 }
256
257 void TemplateTable::aconst_null() {
258 transition(vtos, atos);
259 __ xorl(rax, rax);
260 }
261
262 void TemplateTable::iconst(int value) {
263 transition(vtos, itos);
264 if (value == 0) {
265 __ xorl(rax, rax);
266 } else {
267 __ movl(rax, value);
268 }
269 }
270
271 void TemplateTable::lconst(int value) {
272 transition(vtos, ltos);
273 if (value == 0) {
274 __ xorl(rax, rax);
275 } else {
276 __ movl(rax, value);
277 }
278 #ifndef _LP64
279 assert(value >= 0, "check this code");
280 __ xorptr(rdx, rdx);
281 #endif
282 }
283
284
285
286 void TemplateTable::fconst(int value) {
287 transition(vtos, ftos);
288 if (UseSSE >= 1) {
289 static float one = 1.0f, two = 2.0f;
290 switch (value) {
291 case 0:
292 __ xorps(xmm0, xmm0);
293 break;
294 case 1:
295 __ movflt(xmm0, ExternalAddress((address) &one));
296 break;
297 case 2:
298 __ movflt(xmm0, ExternalAddress((address) &two));
299 break;
300 default:
301 ShouldNotReachHere();
302 break;
303 }
304 } else {
305 #ifdef _LP64
306 ShouldNotReachHere();
307 #else
308 if (value == 0) { __ fldz();
309 } else if (value == 1) { __ fld1();
310 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
311 } else { ShouldNotReachHere();
312 }
313 #endif // _LP64
314 }
315 }
316
317 void TemplateTable::dconst(int value) {
318 transition(vtos, dtos);
319 if (UseSSE >= 2) {
320 static double one = 1.0;
321 switch (value) {
322 case 0:
323 __ xorpd(xmm0, xmm0);
324 break;
325 case 1:
326 __ movdbl(xmm0, ExternalAddress((address) &one));
327 break;
328 default:
329 ShouldNotReachHere();
330 break;
331 }
332 } else {
333 #ifdef _LP64
334 ShouldNotReachHere();
335 #else
336 if (value == 0) { __ fldz();
337 } else if (value == 1) { __ fld1();
338 } else { ShouldNotReachHere();
339 }
340 #endif
341 }
342 }
343
344 void TemplateTable::bipush() {
345 transition(vtos, itos);
346 __ load_signed_byte(rax, at_bcp(1));
347 }
348
349 void TemplateTable::sipush() {
350 transition(vtos, itos);
351 __ load_unsigned_short(rax, at_bcp(1));
352 __ bswapl(rax);
353 __ sarl(rax, 16);
354 }
355
356 void TemplateTable::ldc(bool wide) {
357 transition(vtos, vtos);
358 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
359 Label call_ldc, notFloat, notClass, notInt, Done;
360
361 if (wide) {
362 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
363 } else {
364 __ load_unsigned_byte(rbx, at_bcp(1));
365 }
366
367 __ get_cpool_and_tags(rcx, rax);
368 const int base_offset = ConstantPool::header_size() * wordSize;
369 const int tags_offset = Array<u1>::base_offset_in_bytes();
370
371 // get type
372 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
373
374 // unresolved class - get the resolved class
375 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
376 __ jccb(Assembler::equal, call_ldc);
377
378 // unresolved class in error state - call into runtime to throw the error
379 // from the first resolution attempt
380 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
381 __ jccb(Assembler::equal, call_ldc);
382
383 // resolved class - need to call vm to get java mirror of the class
384 __ cmpl(rdx, JVM_CONSTANT_Class);
385 __ jcc(Assembler::notEqual, notClass);
386
387 __ bind(call_ldc);
388
389 __ movl(rarg, wide);
390 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
391
392 __ push(atos);
393 __ jmp(Done);
394
395 __ bind(notClass);
396 __ cmpl(rdx, JVM_CONSTANT_Float);
397 __ jccb(Assembler::notEqual, notFloat);
398
399 // ftos
400 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
401 __ push(ftos);
402 __ jmp(Done);
403
404 __ bind(notFloat);
405 __ cmpl(rdx, JVM_CONSTANT_Integer);
406 __ jccb(Assembler::notEqual, notInt);
407
408 // itos
409 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
410 __ push(itos);
411 __ jmp(Done);
412
413 // assume the tag is for condy; if not, the VM runtime will tell us
414 __ bind(notInt);
415 condy_helper(Done);
416
417 __ bind(Done);
418 }
419
420 // Fast path for caching oop constants.
421 void TemplateTable::fast_aldc(bool wide) {
422 transition(vtos, atos);
423
424 Register result = rax;
425 Register tmp = rdx;
426 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
427 int index_size = wide ? sizeof(u2) : sizeof(u1);
428
429 Label resolved;
430
431 // We are resolved if the resolved reference cache entry contains a
432 // non-null object (String, MethodType, etc.)
433 assert_different_registers(result, tmp);
434 __ get_cache_index_at_bcp(tmp, 1, index_size);
435 __ load_resolved_reference_at_index(result, tmp);
436 __ testptr(result, result);
437 __ jcc(Assembler::notZero, resolved);
438
439 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
440
441 // first time invocation - must resolve first
442 __ movl(rarg, (int)bytecode());
443 __ call_VM(result, entry, rarg);
444 __ bind(resolved);
445
446 { // Check for the null sentinel.
447 // If we just called the VM, it already did the mapping for us,
448 // but it's harmless to retry.
449 Label notNull;
450 ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
451 __ movptr(tmp, null_sentinel);
452 __ cmpptr(tmp, result);
453 __ jccb(Assembler::notEqual, notNull);
454 __ xorptr(result, result); // NULL object reference
455 __ bind(notNull);
456 }
457
458 if (VerifyOops) {
459 __ verify_oop(result);
460 }
461 }
462
463 void TemplateTable::ldc2_w() {
464 transition(vtos, vtos);
465 Label notDouble, notLong, Done;
466 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
467
468 __ get_cpool_and_tags(rcx, rax);
469 const int base_offset = ConstantPool::header_size() * wordSize;
470 const int tags_offset = Array<u1>::base_offset_in_bytes();
471
472 // get type
473 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
474 __ cmpl(rdx, JVM_CONSTANT_Double);
475 __ jccb(Assembler::notEqual, notDouble);
476
477 // dtos
478 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
479 __ push(dtos);
480
481 __ jmp(Done);
482 __ bind(notDouble);
483 __ cmpl(rdx, JVM_CONSTANT_Long);
484 __ jccb(Assembler::notEqual, notLong);
485
486 // ltos
487 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
488 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
489 __ push(ltos);
490 __ jmp(Done);
491
492 __ bind(notLong);
493 condy_helper(Done);
494
495 __ bind(Done);
496 }
497
498 void TemplateTable::condy_helper(Label& Done) {
499 const Register obj = rax;
500 const Register off = rbx;
501 const Register flags = rcx;
502 const Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
503 __ movl(rarg, (int)bytecode());
504 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
505 #ifndef _LP64
506 // borrow rdi from locals
507 __ get_thread(rdi);
508 __ get_vm_result_2(flags, rdi);
509 __ restore_locals();
510 #else
511 __ get_vm_result_2(flags, r15_thread);
512 #endif
513 // VMr = obj = base address to find primitive value to push
514 // VMr2 = flags = (tos, off) using format of CPCE::_flags
515 __ movl(off, flags);
516 __ andl(off, ConstantPoolCacheEntry::field_index_mask);
517 const Address field(obj, off, Address::times_1, 0*wordSize);
518
519 // What sort of thing are we loading?
520 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
521 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
522
523 switch (bytecode()) {
524 case Bytecodes::_ldc:
525 case Bytecodes::_ldc_w:
526 {
527 // tos in (itos, ftos, stos, btos, ctos, ztos)
528 Label notInt, notFloat, notShort, notByte, notChar, notBool;
529 __ cmpl(flags, itos);
530 __ jcc(Assembler::notEqual, notInt);
531 // itos
532 __ movl(rax, field);
533 __ push(itos);
534 __ jmp(Done);
535
536 __ bind(notInt);
537 __ cmpl(flags, ftos);
538 __ jcc(Assembler::notEqual, notFloat);
539 // ftos
540 __ load_float(field);
541 __ push(ftos);
542 __ jmp(Done);
543
544 __ bind(notFloat);
545 __ cmpl(flags, stos);
546 __ jcc(Assembler::notEqual, notShort);
547 // stos
548 __ load_signed_short(rax, field);
549 __ push(stos);
550 __ jmp(Done);
551
552 __ bind(notShort);
553 __ cmpl(flags, btos);
554 __ jcc(Assembler::notEqual, notByte);
555 // btos
556 __ load_signed_byte(rax, field);
557 __ push(btos);
558 __ jmp(Done);
559
560 __ bind(notByte);
561 __ cmpl(flags, ctos);
562 __ jcc(Assembler::notEqual, notChar);
563 // ctos
564 __ load_unsigned_short(rax, field);
565 __ push(ctos);
566 __ jmp(Done);
567
568 __ bind(notChar);
569 __ cmpl(flags, ztos);
570 __ jcc(Assembler::notEqual, notBool);
571 // ztos
572 __ load_signed_byte(rax, field);
573 __ push(ztos);
574 __ jmp(Done);
575
576 __ bind(notBool);
577 break;
578 }
579
580 case Bytecodes::_ldc2_w:
581 {
582 Label notLong, notDouble;
583 __ cmpl(flags, ltos);
584 __ jcc(Assembler::notEqual, notLong);
585 // ltos
586 __ movptr(rax, field);
587 NOT_LP64(__ movptr(rdx, field.plus_disp(4)));
588 __ push(ltos);
589 __ jmp(Done);
590
591 __ bind(notLong);
592 __ cmpl(flags, dtos);
593 __ jcc(Assembler::notEqual, notDouble);
594 // dtos
595 __ load_double(field);
596 __ push(dtos);
597 __ jmp(Done);
598
599 __ bind(notDouble);
600 break;
601 }
602
603 default:
604 ShouldNotReachHere();
605 }
606
607 __ stop("bad ldc/condy");
608 }
609
610 void TemplateTable::locals_index(Register reg, int offset) {
611 __ load_unsigned_byte(reg, at_bcp(offset));
612 __ negptr(reg);
613 }
614
615 void TemplateTable::iload() {
616 iload_internal();
617 }
618
619 void TemplateTable::nofast_iload() {
620 iload_internal(may_not_rewrite);
621 }
622
623 void TemplateTable::iload_internal(RewriteControl rc) {
624 transition(vtos, itos);
625 if (RewriteFrequentPairs && rc == may_rewrite) {
626 Label rewrite, done;
627 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
628 LP64_ONLY(assert(rbx != bc, "register damaged"));
629
630 // get next byte
631 __ load_unsigned_byte(rbx,
632 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
633 // if _iload, wait to rewrite to iload2. We only want to rewrite the
634 // last two iloads in a pair. Comparing against fast_iload means that
635 // the next bytecode is neither an iload or a caload, and therefore
636 // an iload pair.
637 __ cmpl(rbx, Bytecodes::_iload);
638 __ jcc(Assembler::equal, done);
639
640 __ cmpl(rbx, Bytecodes::_fast_iload);
641 __ movl(bc, Bytecodes::_fast_iload2);
642
643 __ jccb(Assembler::equal, rewrite);
644
645 // if _caload, rewrite to fast_icaload
646 __ cmpl(rbx, Bytecodes::_caload);
647 __ movl(bc, Bytecodes::_fast_icaload);
648 __ jccb(Assembler::equal, rewrite);
649
650 // rewrite so iload doesn't check again.
651 __ movl(bc, Bytecodes::_fast_iload);
652
653 // rewrite
654 // bc: fast bytecode
655 __ bind(rewrite);
656 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
657 __ bind(done);
658 }
659
660 // Get the local value into tos
661 locals_index(rbx);
662 __ movl(rax, iaddress(rbx));
663 }
664
665 void TemplateTable::fast_iload2() {
666 transition(vtos, itos);
667 locals_index(rbx);
668 __ movl(rax, iaddress(rbx));
669 __ push(itos);
670 locals_index(rbx, 3);
671 __ movl(rax, iaddress(rbx));
672 }
673
674 void TemplateTable::fast_iload() {
675 transition(vtos, itos);
676 locals_index(rbx);
677 __ movl(rax, iaddress(rbx));
678 }
679
680 void TemplateTable::lload() {
681 transition(vtos, ltos);
682 locals_index(rbx);
683 __ movptr(rax, laddress(rbx));
684 NOT_LP64(__ movl(rdx, haddress(rbx)));
685 }
686
687 void TemplateTable::fload() {
688 transition(vtos, ftos);
689 locals_index(rbx);
690 __ load_float(faddress(rbx));
691 }
692
693 void TemplateTable::dload() {
694 transition(vtos, dtos);
695 locals_index(rbx);
696 __ load_double(daddress(rbx));
697 }
698
699 void TemplateTable::aload() {
700 transition(vtos, atos);
701 locals_index(rbx);
702 __ movptr(rax, aaddress(rbx));
703 }
704
705 void TemplateTable::locals_index_wide(Register reg) {
706 __ load_unsigned_short(reg, at_bcp(2));
707 __ bswapl(reg);
708 __ shrl(reg, 16);
709 __ negptr(reg);
710 }
711
712 void TemplateTable::wide_iload() {
713 transition(vtos, itos);
714 locals_index_wide(rbx);
715 __ movl(rax, iaddress(rbx));
716 }
717
718 void TemplateTable::wide_lload() {
719 transition(vtos, ltos);
720 locals_index_wide(rbx);
721 __ movptr(rax, laddress(rbx));
722 NOT_LP64(__ movl(rdx, haddress(rbx)));
723 }
724
725 void TemplateTable::wide_fload() {
726 transition(vtos, ftos);
727 locals_index_wide(rbx);
728 __ load_float(faddress(rbx));
729 }
730
731 void TemplateTable::wide_dload() {
732 transition(vtos, dtos);
733 locals_index_wide(rbx);
734 __ load_double(daddress(rbx));
735 }
736
737 void TemplateTable::wide_aload() {
738 transition(vtos, atos);
739 locals_index_wide(rbx);
740 __ movptr(rax, aaddress(rbx));
741 }
742
743 void TemplateTable::index_check(Register array, Register index) {
744 // Pop ptr into array
745 __ pop_ptr(array);
746 index_check_without_pop(array, index);
747 }
748
749 void TemplateTable::index_check_without_pop(Register array, Register index) {
750 // destroys rbx
751 // check array
752 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
753 // sign extend index for use by indexed load
754 __ movl2ptr(index, index);
755 // check index
756 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
757 if (index != rbx) {
758 // ??? convention: move aberrant index into rbx for exception message
759 assert(rbx != array, "different registers");
760 __ movl(rbx, index);
761 }
762 Label skip;
763 __ jccb(Assembler::below, skip);
764 // Pass array to create more detailed exceptions.
765 __ mov(NOT_LP64(rax) LP64_ONLY(c_rarg1), array);
766 __ jump(ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
767 __ bind(skip);
768 }
769
770 void TemplateTable::iaload() {
771 transition(itos, itos);
772 // rax: index
773 // rdx: array
774 index_check(rdx, rax); // kills rbx
775 __ movl(rax, Address(rdx, rax,
776 Address::times_4,
777 arrayOopDesc::base_offset_in_bytes(T_INT)));
778 }
779
780 void TemplateTable::laload() {
781 transition(itos, ltos);
782 // rax: index
783 // rdx: array
784 index_check(rdx, rax); // kills rbx
785 NOT_LP64(__ mov(rbx, rax));
786 // rbx,: index
787 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
788 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
789 }
790
791
792
793 void TemplateTable::faload() {
794 transition(itos, ftos);
795 // rax: index
796 // rdx: array
797 index_check(rdx, rax); // kills rbx
798 __ load_float(Address(rdx, rax,
799 Address::times_4,
800 arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
801 }
802
803 void TemplateTable::daload() {
804 transition(itos, dtos);
805 // rax: index
806 // rdx: array
807 index_check(rdx, rax); // kills rbx
808 __ load_double(Address(rdx, rax,
809 Address::times_8,
810 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
811 }
812
813 void TemplateTable::aaload() {
814 transition(itos, atos);
815
816 Register array = rcx;
817 Register index = rax;
818
819 index_check(array, index); // kills rbx
820 if (ValueArrayFlatten) {
821 Label is_flat_array, done;
822 __ test_flat_array_oop(array, rbx, is_flat_array);
823 do_oop_load(_masm,
824 Address(array, index,
825 UseCompressedOops ? Address::times_4 : Address::times_ptr,
826 arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
827 rax,
828 IN_HEAP_ARRAY);
829 __ jmp(done);
830 __ bind(is_flat_array);
831 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_load), array, index);
832 __ bind(done);
833 } else {
834 do_oop_load(_masm,
835 Address(array, index,
836 UseCompressedOops ? Address::times_4 : Address::times_ptr,
837 arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
838 rax,
839 IN_HEAP_ARRAY);
840 }
841 }
842
843 void TemplateTable::baload() {
844 transition(itos, itos);
845 // rax: index
846 // rdx: array
847 index_check(rdx, rax); // kills rbx
848 __ load_signed_byte(rax, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
849 }
850
851 void TemplateTable::caload() {
852 transition(itos, itos);
853 // rax: index
854 // rdx: array
855 index_check(rdx, rax); // kills rbx
856 __ load_unsigned_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
857 }
858
859 // iload followed by caload frequent pair
860 void TemplateTable::fast_icaload() {
861 transition(vtos, itos);
862 // load index out of locals
863 locals_index(rbx);
864 __ movl(rax, iaddress(rbx));
865
866 // rax: index
867 // rdx: array
868 index_check(rdx, rax); // kills rbx
869 __ load_unsigned_short(rax,
870 Address(rdx, rax,
871 Address::times_2,
872 arrayOopDesc::base_offset_in_bytes(T_CHAR)));
873 }
874
875
876 void TemplateTable::saload() {
877 transition(itos, itos);
878 // rax: index
879 // rdx: array
880 index_check(rdx, rax); // kills rbx
881 __ load_signed_short(rax, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
882 }
883
884 void TemplateTable::iload(int n) {
885 transition(vtos, itos);
886 __ movl(rax, iaddress(n));
887 }
888
889 void TemplateTable::lload(int n) {
890 transition(vtos, ltos);
891 __ movptr(rax, laddress(n));
892 NOT_LP64(__ movptr(rdx, haddress(n)));
893 }
894
895 void TemplateTable::fload(int n) {
896 transition(vtos, ftos);
897 __ load_float(faddress(n));
898 }
899
900 void TemplateTable::dload(int n) {
901 transition(vtos, dtos);
902 __ load_double(daddress(n));
903 }
904
905 void TemplateTable::aload(int n) {
906 transition(vtos, atos);
907 __ movptr(rax, aaddress(n));
908 }
909
910 void TemplateTable::aload_0() {
911 aload_0_internal();
912 }
913
914 void TemplateTable::nofast_aload_0() {
915 aload_0_internal(may_not_rewrite);
916 }
917
918 void TemplateTable::aload_0_internal(RewriteControl rc) {
919 transition(vtos, atos);
920 // According to bytecode histograms, the pairs:
921 //
922 // _aload_0, _fast_igetfield
923 // _aload_0, _fast_agetfield
924 // _aload_0, _fast_fgetfield
925 //
926 // occur frequently. If RewriteFrequentPairs is set, the (slow)
927 // _aload_0 bytecode checks if the next bytecode is either
928 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
929 // rewrites the current bytecode into a pair bytecode; otherwise it
930 // rewrites the current bytecode into _fast_aload_0 that doesn't do
931 // the pair check anymore.
932 //
933 // Note: If the next bytecode is _getfield, the rewrite must be
934 // delayed, otherwise we may miss an opportunity for a pair.
935 //
936 // Also rewrite frequent pairs
937 // aload_0, aload_1
938 // aload_0, iload_1
939 // These bytecodes with a small amount of code are most profitable
940 // to rewrite
941 if (RewriteFrequentPairs && rc == may_rewrite) {
942 Label rewrite, done;
943
944 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
945 LP64_ONLY(assert(rbx != bc, "register damaged"));
946
947 // get next byte
948 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
949
950 // if _getfield then wait with rewrite
951 __ cmpl(rbx, Bytecodes::_getfield);
952 __ jcc(Assembler::equal, done);
953
954 // if _igetfield then rewrite to _fast_iaccess_0
955 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
956 __ cmpl(rbx, Bytecodes::_fast_igetfield);
957 __ movl(bc, Bytecodes::_fast_iaccess_0);
958 __ jccb(Assembler::equal, rewrite);
959
960 // if _agetfield then rewrite to _fast_aaccess_0
961 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
962 assert(ValueTypesBufferMaxMemory == 0, "Such rewritting doesn't support flattened values yet");
963 __ cmpl(rbx, Bytecodes::_fast_agetfield);
964 __ movl(bc, Bytecodes::_fast_aaccess_0);
965 __ jccb(Assembler::equal, rewrite);
966
967 // if _fgetfield then rewrite to _fast_faccess_0
968 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
969 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
970 __ movl(bc, Bytecodes::_fast_faccess_0);
971 __ jccb(Assembler::equal, rewrite);
972
973 // else rewrite to _fast_aload0
974 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
975 __ movl(bc, Bytecodes::_fast_aload_0);
976
977 // rewrite
978 // bc: fast bytecode
979 __ bind(rewrite);
980 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
981
982 __ bind(done);
983 }
984
985 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
986 aload(0);
987 }
988
989 void TemplateTable::istore() {
990 transition(itos, vtos);
991 locals_index(rbx);
992 __ movl(iaddress(rbx), rax);
993 }
994
995
996 void TemplateTable::lstore() {
997 transition(ltos, vtos);
998 locals_index(rbx);
999 __ movptr(laddress(rbx), rax);
1000 NOT_LP64(__ movptr(haddress(rbx), rdx));
1001 }
1002
1003 void TemplateTable::fstore() {
1004 transition(ftos, vtos);
1005 locals_index(rbx);
1006 __ store_float(faddress(rbx));
1007 }
1008
1009 void TemplateTable::dstore() {
1010 transition(dtos, vtos);
1011 locals_index(rbx);
1012 __ store_double(daddress(rbx));
1013 }
1014
1015 void TemplateTable::astore() {
1016 transition(vtos, vtos);
1017 __ pop_ptr(rax);
1018 locals_index(rbx);
1019 __ movptr(aaddress(rbx), rax);
1020 }
1021
1022 void TemplateTable::wide_istore() {
1023 transition(vtos, vtos);
1024 __ pop_i();
1025 locals_index_wide(rbx);
1026 __ movl(iaddress(rbx), rax);
1027 }
1028
1029 void TemplateTable::wide_lstore() {
1030 transition(vtos, vtos);
1031 NOT_LP64(__ pop_l(rax, rdx));
1032 LP64_ONLY(__ pop_l());
1033 locals_index_wide(rbx);
1034 __ movptr(laddress(rbx), rax);
1035 NOT_LP64(__ movl(haddress(rbx), rdx));
1036 }
1037
1038 void TemplateTable::wide_fstore() {
1039 #ifdef _LP64
1040 transition(vtos, vtos);
1041 __ pop_f(xmm0);
1042 locals_index_wide(rbx);
1043 __ movflt(faddress(rbx), xmm0);
1044 #else
1045 wide_istore();
1046 #endif
1047 }
1048
1049 void TemplateTable::wide_dstore() {
1050 #ifdef _LP64
1051 transition(vtos, vtos);
1052 __ pop_d(xmm0);
1053 locals_index_wide(rbx);
1054 __ movdbl(daddress(rbx), xmm0);
1055 #else
1056 wide_lstore();
1057 #endif
1058 }
1059
1060 void TemplateTable::wide_astore() {
1061 transition(vtos, vtos);
1062 __ pop_ptr(rax);
1063 locals_index_wide(rbx);
1064 __ movptr(aaddress(rbx), rax);
1065 }
1066
1067 void TemplateTable::iastore() {
1068 transition(itos, vtos);
1069 __ pop_i(rbx);
1070 // rax: value
1071 // rbx: index
1072 // rdx: array
1073 index_check(rdx, rbx); // prefer index in rbx
1074 __ movl(Address(rdx, rbx,
1075 Address::times_4,
1076 arrayOopDesc::base_offset_in_bytes(T_INT)),
1077 rax);
1078 }
1079
1080 void TemplateTable::lastore() {
1081 transition(ltos, vtos);
1082 __ pop_i(rbx);
1083 // rax,: low(value)
1084 // rcx: array
1085 // rdx: high(value)
1086 index_check(rcx, rbx); // prefer index in rbx,
1087 // rbx,: index
1088 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
1089 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
1090 }
1091
1092
1093 void TemplateTable::fastore() {
1094 transition(ftos, vtos);
1095 __ pop_i(rbx);
1096 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1097 // rbx: index
1098 // rdx: array
1099 index_check(rdx, rbx); // prefer index in rbx
1100 __ store_float(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
1101 }
1102
1103 void TemplateTable::dastore() {
1104 transition(dtos, vtos);
1105 __ pop_i(rbx);
1106 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1107 // rbx: index
1108 // rdx: array
1109 index_check(rdx, rbx); // prefer index in rbx
1110 __ store_double(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
1111 }
1112
1113 void TemplateTable::aastore() {
1114 Label is_null, is_flat_array, ok_is_subtype, done;
1115 transition(vtos, vtos);
1116 // stack: ..., array, index, value
1117 __ movptr(rax, at_tos()); // value
1118 __ movl(rcx, at_tos_p1()); // index
1119 __ movptr(rdx, at_tos_p2()); // array
1120
1121 Address element_address(rdx, rcx,
1122 UseCompressedOops? Address::times_4 : Address::times_ptr,
1123 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1124
1125 index_check_without_pop(rdx, rcx); // kills rbx
1126
1127 __ testptr(rax, rax);
1128 __ jcc(Assembler::zero, is_null);
1129
1130 // Move array class to rdi
1131 __ load_klass(rdi, rdx);
1132 if (ValueArrayFlatten) {
1133 __ test_flat_array_klass(rdi, rbx, is_flat_array);
1134 }
1135
1136 // Move subklass into rbx
1137 __ load_klass(rbx, rax);
1138 // Move array element superklass into rax
1139 __ movptr(rax, Address(rdi,
1140 ObjArrayKlass::element_klass_offset()));
1141
1142 // Generate subtype check. Blows rcx, rdi
1143 // Superklass in rax. Subklass in rbx.
1144 // is "rbx <: rax" ? (value subclass <: array element superclass)
1145 __ gen_subtype_check(rbx, ok_is_subtype);
1146
1147 // Come here on failure
1148 // object is at TOS
1149 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1150
1151 // Come here on success
1152 __ bind(ok_is_subtype);
1153
1154 // Get the value we will store
1155 __ movptr(rax, at_tos());
1156 if (ValueTypesBufferMaxMemory > 0) {
1157 Label is_on_heap;
1158 __ test_value_is_not_buffered(rax, rbx, is_on_heap);
1159 __ push(rdx); // save precomputed element address, and convert buffer oop to heap oop
1160 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_heap_copy), rax);
1161 __ pop(rdx);
1162 __ bind(is_on_heap);
1163 }
1164 __ movl(rcx, at_tos_p1()); // index
1165 // Now store using the appropriate barrier
1166 do_oop_store(_masm, element_address, rax, IN_HEAP_ARRAY);
1167 __ jmp(done);
1168
1169 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1170 __ bind(is_null);
1171 __ profile_null_seen(rbx);
1172 if (EnableValhalla) {
1173 Label is_null_into_value_array_npe, store_null;
1174
1175 __ load_klass(rdi, rdx);
1176 // No way to store null in flat array
1177 __ test_flat_array_klass(rdi, rbx, is_null_into_value_array_npe);
1178
1179 // Use case for storing values in objArray where element_klass is specifically
1180 // a value type because they could not be flattened "for reasons",
1181 // these need to have the same semantics as flat arrays, i.e. NPE
1182 __ movptr(rdi, Address(rdi, ObjArrayKlass::element_klass_offset()));
1183 __ test_klass_is_value(rdi, rdi, is_null_into_value_array_npe);
1184 __ jmp(store_null);
1185
1186 __ bind(is_null_into_value_array_npe);
1187 __ jump(ExternalAddress(Interpreter::_throw_NullPointerException_entry));
1188
1189 __ bind(store_null);
1190 }
1191 // Store a NULL
1192 do_oop_store(_masm, element_address, noreg, IN_HEAP_ARRAY);
1193 __ jmp(done);
1194
1195 if (EnableValhalla) {
1196 Label is_type_ok;
1197 __ bind(is_flat_array); // Store non-null value to flat
1198
1199 // Simplistic type check...
1200
1201 // Profile the not-null value's klass.
1202 __ load_klass(rbx, rax);
1203 __ profile_typecheck(rcx, rbx, rax); // blows rcx, and rax
1204 // Move element klass into rax
1205 __ movptr(rax, Address(rdi, ArrayKlass::element_klass_offset()));
1206 // flat value array needs exact type match
1207 // is "rax == rbx" (value subclass == array element superclass)
1208 __ cmpptr(rax, rbx);
1209 __ jccb(Assembler::equal, is_type_ok);
1210
1211 __ profile_typecheck_failed(rcx);
1212 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1213
1214 __ bind(is_type_ok);
1215 __ movptr(rax, at_tos()); // value
1216 __ movl(rcx, at_tos_p1()); // index
1217 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::value_array_store), rax, rdx, rcx);
1218 }
1219 // Pop stack arguments
1220 __ bind(done);
1221 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1222 }
1223
1224 void TemplateTable::bastore() {
1225 transition(itos, vtos);
1226 __ pop_i(rbx);
1227 // rax: value
1228 // rbx: index
1229 // rdx: array
1230 index_check(rdx, rbx); // prefer index in rbx
1231 // Need to check whether array is boolean or byte
1232 // since both types share the bastore bytecode.
1233 __ load_klass(rcx, rdx);
1234 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1235 int diffbit = Klass::layout_helper_boolean_diffbit();
1236 __ testl(rcx, diffbit);
1237 Label L_skip;
1238 __ jccb(Assembler::zero, L_skip);
1239 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1240 __ bind(L_skip);
1241 __ movb(Address(rdx, rbx,
1242 Address::times_1,
1243 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1244 rax);
1245 }
1246
1247 void TemplateTable::castore() {
1248 transition(itos, vtos);
1249 __ pop_i(rbx);
1250 // rax: value
1251 // rbx: index
1252 // rdx: array
1253 index_check(rdx, rbx); // prefer index in rbx
1254 __ movw(Address(rdx, rbx,
1255 Address::times_2,
1256 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1257 rax);
1258 }
1259
1260
1261 void TemplateTable::sastore() {
1262 castore();
1263 }
1264
1265 void TemplateTable::istore(int n) {
1266 transition(itos, vtos);
1267 __ movl(iaddress(n), rax);
1268 }
1269
1270 void TemplateTable::lstore(int n) {
1271 transition(ltos, vtos);
1272 __ movptr(laddress(n), rax);
1273 NOT_LP64(__ movptr(haddress(n), rdx));
1274 }
1275
1276 void TemplateTable::fstore(int n) {
1277 transition(ftos, vtos);
1278 __ store_float(faddress(n));
1279 }
1280
1281 void TemplateTable::dstore(int n) {
1282 transition(dtos, vtos);
1283 __ store_double(daddress(n));
1284 }
1285
1286
1287 void TemplateTable::astore(int n) {
1288 transition(vtos, vtos);
1289 __ pop_ptr(rax);
1290 __ movptr(aaddress(n), rax);
1291 }
1292
1293 void TemplateTable::pop() {
1294 transition(vtos, vtos);
1295 __ addptr(rsp, Interpreter::stackElementSize);
1296 }
1297
1298 void TemplateTable::pop2() {
1299 transition(vtos, vtos);
1300 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1301 }
1302
1303
1304 void TemplateTable::dup() {
1305 transition(vtos, vtos);
1306 __ load_ptr(0, rax);
1307 __ push_ptr(rax);
1308 // stack: ..., a, a
1309 }
1310
1311 void TemplateTable::dup_x1() {
1312 transition(vtos, vtos);
1313 // stack: ..., a, b
1314 __ load_ptr( 0, rax); // load b
1315 __ load_ptr( 1, rcx); // load a
1316 __ store_ptr(1, rax); // store b
1317 __ store_ptr(0, rcx); // store a
1318 __ push_ptr(rax); // push b
1319 // stack: ..., b, a, b
1320 }
1321
1322 void TemplateTable::dup_x2() {
1323 transition(vtos, vtos);
1324 // stack: ..., a, b, c
1325 __ load_ptr( 0, rax); // load c
1326 __ load_ptr( 2, rcx); // load a
1327 __ store_ptr(2, rax); // store c in a
1328 __ push_ptr(rax); // push c
1329 // stack: ..., c, b, c, c
1330 __ load_ptr( 2, rax); // load b
1331 __ store_ptr(2, rcx); // store a in b
1332 // stack: ..., c, a, c, c
1333 __ store_ptr(1, rax); // store b in c
1334 // stack: ..., c, a, b, c
1335 }
1336
1337 void TemplateTable::dup2() {
1338 transition(vtos, vtos);
1339 // stack: ..., a, b
1340 __ load_ptr(1, rax); // load a
1341 __ push_ptr(rax); // push a
1342 __ load_ptr(1, rax); // load b
1343 __ push_ptr(rax); // push b
1344 // stack: ..., a, b, a, b
1345 }
1346
1347
1348 void TemplateTable::dup2_x1() {
1349 transition(vtos, vtos);
1350 // stack: ..., a, b, c
1351 __ load_ptr( 0, rcx); // load c
1352 __ load_ptr( 1, rax); // load b
1353 __ push_ptr(rax); // push b
1354 __ push_ptr(rcx); // push c
1355 // stack: ..., a, b, c, b, c
1356 __ store_ptr(3, rcx); // store c in b
1357 // stack: ..., a, c, c, b, c
1358 __ load_ptr( 4, rcx); // load a
1359 __ store_ptr(2, rcx); // store a in 2nd c
1360 // stack: ..., a, c, a, b, c
1361 __ store_ptr(4, rax); // store b in a
1362 // stack: ..., b, c, a, b, c
1363 }
1364
1365 void TemplateTable::dup2_x2() {
1366 transition(vtos, vtos);
1367 // stack: ..., a, b, c, d
1368 __ load_ptr( 0, rcx); // load d
1369 __ load_ptr( 1, rax); // load c
1370 __ push_ptr(rax); // push c
1371 __ push_ptr(rcx); // push d
1372 // stack: ..., a, b, c, d, c, d
1373 __ load_ptr( 4, rax); // load b
1374 __ store_ptr(2, rax); // store b in d
1375 __ store_ptr(4, rcx); // store d in b
1376 // stack: ..., a, d, c, b, c, d
1377 __ load_ptr( 5, rcx); // load a
1378 __ load_ptr( 3, rax); // load c
1379 __ store_ptr(3, rcx); // store a in c
1380 __ store_ptr(5, rax); // store c in a
1381 // stack: ..., c, d, a, b, c, d
1382 }
1383
1384 void TemplateTable::swap() {
1385 transition(vtos, vtos);
1386 // stack: ..., a, b
1387 __ load_ptr( 1, rcx); // load a
1388 __ load_ptr( 0, rax); // load b
1389 __ store_ptr(0, rcx); // store a in b
1390 __ store_ptr(1, rax); // store b in a
1391 // stack: ..., b, a
1392 }
1393
1394 void TemplateTable::iop2(Operation op) {
1395 transition(itos, itos);
1396 switch (op) {
1397 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1398 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1399 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1400 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1401 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1402 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1403 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1404 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1405 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1406 default : ShouldNotReachHere();
1407 }
1408 }
1409
1410 void TemplateTable::lop2(Operation op) {
1411 transition(ltos, ltos);
1412 #ifdef _LP64
1413 switch (op) {
1414 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1415 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1416 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1417 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1418 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1419 default : ShouldNotReachHere();
1420 }
1421 #else
1422 __ pop_l(rbx, rcx);
1423 switch (op) {
1424 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1425 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1426 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1427 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1428 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1429 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1430 default : ShouldNotReachHere();
1431 }
1432 #endif
1433 }
1434
1435 void TemplateTable::idiv() {
1436 transition(itos, itos);
1437 __ movl(rcx, rax);
1438 __ pop_i(rax);
1439 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1440 // they are not equal, one could do a normal division (no correction
1441 // needed), which may speed up this implementation for the common case.
1442 // (see also JVM spec., p.243 & p.271)
1443 __ corrected_idivl(rcx);
1444 }
1445
1446 void TemplateTable::irem() {
1447 transition(itos, itos);
1448 __ movl(rcx, rax);
1449 __ pop_i(rax);
1450 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1451 // they are not equal, one could do a normal division (no correction
1452 // needed), which may speed up this implementation for the common case.
1453 // (see also JVM spec., p.243 & p.271)
1454 __ corrected_idivl(rcx);
1455 __ movl(rax, rdx);
1456 }
1457
1458 void TemplateTable::lmul() {
1459 transition(ltos, ltos);
1460 #ifdef _LP64
1461 __ pop_l(rdx);
1462 __ imulq(rax, rdx);
1463 #else
1464 __ pop_l(rbx, rcx);
1465 __ push(rcx); __ push(rbx);
1466 __ push(rdx); __ push(rax);
1467 __ lmul(2 * wordSize, 0);
1468 __ addptr(rsp, 4 * wordSize); // take off temporaries
1469 #endif
1470 }
1471
1472 void TemplateTable::ldiv() {
1473 transition(ltos, ltos);
1474 #ifdef _LP64
1475 __ mov(rcx, rax);
1476 __ pop_l(rax);
1477 // generate explicit div0 check
1478 __ testq(rcx, rcx);
1479 __ jump_cc(Assembler::zero,
1480 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1481 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1482 // they are not equal, one could do a normal division (no correction
1483 // needed), which may speed up this implementation for the common case.
1484 // (see also JVM spec., p.243 & p.271)
1485 __ corrected_idivq(rcx); // kills rbx
1486 #else
1487 __ pop_l(rbx, rcx);
1488 __ push(rcx); __ push(rbx);
1489 __ push(rdx); __ push(rax);
1490 // check if y = 0
1491 __ orl(rax, rdx);
1492 __ jump_cc(Assembler::zero,
1493 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1494 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1495 __ addptr(rsp, 4 * wordSize); // take off temporaries
1496 #endif
1497 }
1498
1499 void TemplateTable::lrem() {
1500 transition(ltos, ltos);
1501 #ifdef _LP64
1502 __ mov(rcx, rax);
1503 __ pop_l(rax);
1504 __ testq(rcx, rcx);
1505 __ jump_cc(Assembler::zero,
1506 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1507 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1508 // they are not equal, one could do a normal division (no correction
1509 // needed), which may speed up this implementation for the common case.
1510 // (see also JVM spec., p.243 & p.271)
1511 __ corrected_idivq(rcx); // kills rbx
1512 __ mov(rax, rdx);
1513 #else
1514 __ pop_l(rbx, rcx);
1515 __ push(rcx); __ push(rbx);
1516 __ push(rdx); __ push(rax);
1517 // check if y = 0
1518 __ orl(rax, rdx);
1519 __ jump_cc(Assembler::zero,
1520 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1521 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1522 __ addptr(rsp, 4 * wordSize);
1523 #endif
1524 }
1525
1526 void TemplateTable::lshl() {
1527 transition(itos, ltos);
1528 __ movl(rcx, rax); // get shift count
1529 #ifdef _LP64
1530 __ pop_l(rax); // get shift value
1531 __ shlq(rax);
1532 #else
1533 __ pop_l(rax, rdx); // get shift value
1534 __ lshl(rdx, rax);
1535 #endif
1536 }
1537
1538 void TemplateTable::lshr() {
1539 #ifdef _LP64
1540 transition(itos, ltos);
1541 __ movl(rcx, rax); // get shift count
1542 __ pop_l(rax); // get shift value
1543 __ sarq(rax);
1544 #else
1545 transition(itos, ltos);
1546 __ mov(rcx, rax); // get shift count
1547 __ pop_l(rax, rdx); // get shift value
1548 __ lshr(rdx, rax, true);
1549 #endif
1550 }
1551
1552 void TemplateTable::lushr() {
1553 transition(itos, ltos);
1554 #ifdef _LP64
1555 __ movl(rcx, rax); // get shift count
1556 __ pop_l(rax); // get shift value
1557 __ shrq(rax);
1558 #else
1559 __ mov(rcx, rax); // get shift count
1560 __ pop_l(rax, rdx); // get shift value
1561 __ lshr(rdx, rax);
1562 #endif
1563 }
1564
1565 void TemplateTable::fop2(Operation op) {
1566 transition(ftos, ftos);
1567
1568 if (UseSSE >= 1) {
1569 switch (op) {
1570 case add:
1571 __ addss(xmm0, at_rsp());
1572 __ addptr(rsp, Interpreter::stackElementSize);
1573 break;
1574 case sub:
1575 __ movflt(xmm1, xmm0);
1576 __ pop_f(xmm0);
1577 __ subss(xmm0, xmm1);
1578 break;
1579 case mul:
1580 __ mulss(xmm0, at_rsp());
1581 __ addptr(rsp, Interpreter::stackElementSize);
1582 break;
1583 case div:
1584 __ movflt(xmm1, xmm0);
1585 __ pop_f(xmm0);
1586 __ divss(xmm0, xmm1);
1587 break;
1588 case rem:
1589 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1590 // modulo operation. The frem method calls the function
1591 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1592 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1593 // (signalling or quiet) is returned.
1594 //
1595 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1596 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1597 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1598 // The fprem instruction used on x86_32 is functionally equivalent to
1599 // SharedRuntime::frem in that it returns a NaN.
1600 #ifdef _LP64
1601 __ movflt(xmm1, xmm0);
1602 __ pop_f(xmm0);
1603 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1604 #else
1605 __ push_f(xmm0);
1606 __ pop_f();
1607 __ fld_s(at_rsp());
1608 __ fremr(rax);
1609 __ f2ieee();
1610 __ pop(rax); // pop second operand off the stack
1611 __ push_f();
1612 __ pop_f(xmm0);
1613 #endif
1614 break;
1615 default:
1616 ShouldNotReachHere();
1617 break;
1618 }
1619 } else {
1620 #ifdef _LP64
1621 ShouldNotReachHere();
1622 #else
1623 switch (op) {
1624 case add: __ fadd_s (at_rsp()); break;
1625 case sub: __ fsubr_s(at_rsp()); break;
1626 case mul: __ fmul_s (at_rsp()); break;
1627 case div: __ fdivr_s(at_rsp()); break;
1628 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1629 default : ShouldNotReachHere();
1630 }
1631 __ f2ieee();
1632 __ pop(rax); // pop second operand off the stack
1633 #endif // _LP64
1634 }
1635 }
1636
1637 void TemplateTable::dop2(Operation op) {
1638 transition(dtos, dtos);
1639 if (UseSSE >= 2) {
1640 switch (op) {
1641 case add:
1642 __ addsd(xmm0, at_rsp());
1643 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1644 break;
1645 case sub:
1646 __ movdbl(xmm1, xmm0);
1647 __ pop_d(xmm0);
1648 __ subsd(xmm0, xmm1);
1649 break;
1650 case mul:
1651 __ mulsd(xmm0, at_rsp());
1652 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1653 break;
1654 case div:
1655 __ movdbl(xmm1, xmm0);
1656 __ pop_d(xmm0);
1657 __ divsd(xmm0, xmm1);
1658 break;
1659 case rem:
1660 // Similar to fop2(), the modulo operation is performed using the
1661 // SharedRuntime::drem method (on x86_64 platforms) or using the
1662 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1663 #ifdef _LP64
1664 __ movdbl(xmm1, xmm0);
1665 __ pop_d(xmm0);
1666 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1667 #else
1668 __ push_d(xmm0);
1669 __ pop_d();
1670 __ fld_d(at_rsp());
1671 __ fremr(rax);
1672 __ d2ieee();
1673 __ pop(rax);
1674 __ pop(rdx);
1675 __ push_d();
1676 __ pop_d(xmm0);
1677 #endif
1678 break;
1679 default:
1680 ShouldNotReachHere();
1681 break;
1682 }
1683 } else {
1684 #ifdef _LP64
1685 ShouldNotReachHere();
1686 #else
1687 switch (op) {
1688 case add: __ fadd_d (at_rsp()); break;
1689 case sub: __ fsubr_d(at_rsp()); break;
1690 case mul: {
1691 Label L_strict;
1692 Label L_join;
1693 const Address access_flags (rcx, Method::access_flags_offset());
1694 __ get_method(rcx);
1695 __ movl(rcx, access_flags);
1696 __ testl(rcx, JVM_ACC_STRICT);
1697 __ jccb(Assembler::notZero, L_strict);
1698 __ fmul_d (at_rsp());
1699 __ jmpb(L_join);
1700 __ bind(L_strict);
1701 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1702 __ fmulp();
1703 __ fmul_d (at_rsp());
1704 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1705 __ fmulp();
1706 __ bind(L_join);
1707 break;
1708 }
1709 case div: {
1710 Label L_strict;
1711 Label L_join;
1712 const Address access_flags (rcx, Method::access_flags_offset());
1713 __ get_method(rcx);
1714 __ movl(rcx, access_flags);
1715 __ testl(rcx, JVM_ACC_STRICT);
1716 __ jccb(Assembler::notZero, L_strict);
1717 __ fdivr_d(at_rsp());
1718 __ jmp(L_join);
1719 __ bind(L_strict);
1720 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1721 __ fmul_d (at_rsp());
1722 __ fdivrp();
1723 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1724 __ fmulp();
1725 __ bind(L_join);
1726 break;
1727 }
1728 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1729 default : ShouldNotReachHere();
1730 }
1731 __ d2ieee();
1732 // Pop double precision number from rsp.
1733 __ pop(rax);
1734 __ pop(rdx);
1735 #endif
1736 }
1737 }
1738
1739 void TemplateTable::ineg() {
1740 transition(itos, itos);
1741 __ negl(rax);
1742 }
1743
1744 void TemplateTable::lneg() {
1745 transition(ltos, ltos);
1746 LP64_ONLY(__ negq(rax));
1747 NOT_LP64(__ lneg(rdx, rax));
1748 }
1749
1750 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1751 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1752 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1753 // of 128-bits operands for SSE instructions.
1754 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1755 // Store the value to a 128-bits operand.
1756 operand[0] = lo;
1757 operand[1] = hi;
1758 return operand;
1759 }
1760
1761 // Buffer for 128-bits masks used by SSE instructions.
1762 static jlong float_signflip_pool[2*2];
1763 static jlong double_signflip_pool[2*2];
1764
1765 void TemplateTable::fneg() {
1766 transition(ftos, ftos);
1767 if (UseSSE >= 1) {
1768 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1769 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1770 } else {
1771 LP64_ONLY(ShouldNotReachHere());
1772 NOT_LP64(__ fchs());
1773 }
1774 }
1775
1776 void TemplateTable::dneg() {
1777 transition(dtos, dtos);
1778 if (UseSSE >= 2) {
1779 static jlong *double_signflip =
1780 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1781 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1782 } else {
1783 #ifdef _LP64
1784 ShouldNotReachHere();
1785 #else
1786 __ fchs();
1787 #endif
1788 }
1789 }
1790
1791 void TemplateTable::iinc() {
1792 transition(vtos, vtos);
1793 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1794 locals_index(rbx);
1795 __ addl(iaddress(rbx), rdx);
1796 }
1797
1798 void TemplateTable::wide_iinc() {
1799 transition(vtos, vtos);
1800 __ movl(rdx, at_bcp(4)); // get constant
1801 locals_index_wide(rbx);
1802 __ bswapl(rdx); // swap bytes & sign-extend constant
1803 __ sarl(rdx, 16);
1804 __ addl(iaddress(rbx), rdx);
1805 // Note: should probably use only one movl to get both
1806 // the index and the constant -> fix this
1807 }
1808
1809 void TemplateTable::convert() {
1810 #ifdef _LP64
1811 // Checking
1812 #ifdef ASSERT
1813 {
1814 TosState tos_in = ilgl;
1815 TosState tos_out = ilgl;
1816 switch (bytecode()) {
1817 case Bytecodes::_i2l: // fall through
1818 case Bytecodes::_i2f: // fall through
1819 case Bytecodes::_i2d: // fall through
1820 case Bytecodes::_i2b: // fall through
1821 case Bytecodes::_i2c: // fall through
1822 case Bytecodes::_i2s: tos_in = itos; break;
1823 case Bytecodes::_l2i: // fall through
1824 case Bytecodes::_l2f: // fall through
1825 case Bytecodes::_l2d: tos_in = ltos; break;
1826 case Bytecodes::_f2i: // fall through
1827 case Bytecodes::_f2l: // fall through
1828 case Bytecodes::_f2d: tos_in = ftos; break;
1829 case Bytecodes::_d2i: // fall through
1830 case Bytecodes::_d2l: // fall through
1831 case Bytecodes::_d2f: tos_in = dtos; break;
1832 default : ShouldNotReachHere();
1833 }
1834 switch (bytecode()) {
1835 case Bytecodes::_l2i: // fall through
1836 case Bytecodes::_f2i: // fall through
1837 case Bytecodes::_d2i: // fall through
1838 case Bytecodes::_i2b: // fall through
1839 case Bytecodes::_i2c: // fall through
1840 case Bytecodes::_i2s: tos_out = itos; break;
1841 case Bytecodes::_i2l: // fall through
1842 case Bytecodes::_f2l: // fall through
1843 case Bytecodes::_d2l: tos_out = ltos; break;
1844 case Bytecodes::_i2f: // fall through
1845 case Bytecodes::_l2f: // fall through
1846 case Bytecodes::_d2f: tos_out = ftos; break;
1847 case Bytecodes::_i2d: // fall through
1848 case Bytecodes::_l2d: // fall through
1849 case Bytecodes::_f2d: tos_out = dtos; break;
1850 default : ShouldNotReachHere();
1851 }
1852 transition(tos_in, tos_out);
1853 }
1854 #endif // ASSERT
1855
1856 static const int64_t is_nan = 0x8000000000000000L;
1857
1858 // Conversion
1859 switch (bytecode()) {
1860 case Bytecodes::_i2l:
1861 __ movslq(rax, rax);
1862 break;
1863 case Bytecodes::_i2f:
1864 __ cvtsi2ssl(xmm0, rax);
1865 break;
1866 case Bytecodes::_i2d:
1867 __ cvtsi2sdl(xmm0, rax);
1868 break;
1869 case Bytecodes::_i2b:
1870 __ movsbl(rax, rax);
1871 break;
1872 case Bytecodes::_i2c:
1873 __ movzwl(rax, rax);
1874 break;
1875 case Bytecodes::_i2s:
1876 __ movswl(rax, rax);
1877 break;
1878 case Bytecodes::_l2i:
1879 __ movl(rax, rax);
1880 break;
1881 case Bytecodes::_l2f:
1882 __ cvtsi2ssq(xmm0, rax);
1883 break;
1884 case Bytecodes::_l2d:
1885 __ cvtsi2sdq(xmm0, rax);
1886 break;
1887 case Bytecodes::_f2i:
1888 {
1889 Label L;
1890 __ cvttss2sil(rax, xmm0);
1891 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1892 __ jcc(Assembler::notEqual, L);
1893 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1894 __ bind(L);
1895 }
1896 break;
1897 case Bytecodes::_f2l:
1898 {
1899 Label L;
1900 __ cvttss2siq(rax, xmm0);
1901 // NaN or overflow/underflow?
1902 __ cmp64(rax, ExternalAddress((address) &is_nan));
1903 __ jcc(Assembler::notEqual, L);
1904 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1905 __ bind(L);
1906 }
1907 break;
1908 case Bytecodes::_f2d:
1909 __ cvtss2sd(xmm0, xmm0);
1910 break;
1911 case Bytecodes::_d2i:
1912 {
1913 Label L;
1914 __ cvttsd2sil(rax, xmm0);
1915 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1916 __ jcc(Assembler::notEqual, L);
1917 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1918 __ bind(L);
1919 }
1920 break;
1921 case Bytecodes::_d2l:
1922 {
1923 Label L;
1924 __ cvttsd2siq(rax, xmm0);
1925 // NaN or overflow/underflow?
1926 __ cmp64(rax, ExternalAddress((address) &is_nan));
1927 __ jcc(Assembler::notEqual, L);
1928 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1929 __ bind(L);
1930 }
1931 break;
1932 case Bytecodes::_d2f:
1933 __ cvtsd2ss(xmm0, xmm0);
1934 break;
1935 default:
1936 ShouldNotReachHere();
1937 }
1938 #else
1939 // Checking
1940 #ifdef ASSERT
1941 { TosState tos_in = ilgl;
1942 TosState tos_out = ilgl;
1943 switch (bytecode()) {
1944 case Bytecodes::_i2l: // fall through
1945 case Bytecodes::_i2f: // fall through
1946 case Bytecodes::_i2d: // fall through
1947 case Bytecodes::_i2b: // fall through
1948 case Bytecodes::_i2c: // fall through
1949 case Bytecodes::_i2s: tos_in = itos; break;
1950 case Bytecodes::_l2i: // fall through
1951 case Bytecodes::_l2f: // fall through
1952 case Bytecodes::_l2d: tos_in = ltos; break;
1953 case Bytecodes::_f2i: // fall through
1954 case Bytecodes::_f2l: // fall through
1955 case Bytecodes::_f2d: tos_in = ftos; break;
1956 case Bytecodes::_d2i: // fall through
1957 case Bytecodes::_d2l: // fall through
1958 case Bytecodes::_d2f: tos_in = dtos; break;
1959 default : ShouldNotReachHere();
1960 }
1961 switch (bytecode()) {
1962 case Bytecodes::_l2i: // fall through
1963 case Bytecodes::_f2i: // fall through
1964 case Bytecodes::_d2i: // fall through
1965 case Bytecodes::_i2b: // fall through
1966 case Bytecodes::_i2c: // fall through
1967 case Bytecodes::_i2s: tos_out = itos; break;
1968 case Bytecodes::_i2l: // fall through
1969 case Bytecodes::_f2l: // fall through
1970 case Bytecodes::_d2l: tos_out = ltos; break;
1971 case Bytecodes::_i2f: // fall through
1972 case Bytecodes::_l2f: // fall through
1973 case Bytecodes::_d2f: tos_out = ftos; break;
1974 case Bytecodes::_i2d: // fall through
1975 case Bytecodes::_l2d: // fall through
1976 case Bytecodes::_f2d: tos_out = dtos; break;
1977 default : ShouldNotReachHere();
1978 }
1979 transition(tos_in, tos_out);
1980 }
1981 #endif // ASSERT
1982
1983 // Conversion
1984 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1985 switch (bytecode()) {
1986 case Bytecodes::_i2l:
1987 __ extend_sign(rdx, rax);
1988 break;
1989 case Bytecodes::_i2f:
1990 if (UseSSE >= 1) {
1991 __ cvtsi2ssl(xmm0, rax);
1992 } else {
1993 __ push(rax); // store int on tos
1994 __ fild_s(at_rsp()); // load int to ST0
1995 __ f2ieee(); // truncate to float size
1996 __ pop(rcx); // adjust rsp
1997 }
1998 break;
1999 case Bytecodes::_i2d:
2000 if (UseSSE >= 2) {
2001 __ cvtsi2sdl(xmm0, rax);
2002 } else {
2003 __ push(rax); // add one slot for d2ieee()
2004 __ push(rax); // store int on tos
2005 __ fild_s(at_rsp()); // load int to ST0
2006 __ d2ieee(); // truncate to double size
2007 __ pop(rcx); // adjust rsp
2008 __ pop(rcx);
2009 }
2010 break;
2011 case Bytecodes::_i2b:
2012 __ shll(rax, 24); // truncate upper 24 bits
2013 __ sarl(rax, 24); // and sign-extend byte
2014 LP64_ONLY(__ movsbl(rax, rax));
2015 break;
2016 case Bytecodes::_i2c:
2017 __ andl(rax, 0xFFFF); // truncate upper 16 bits
2018 LP64_ONLY(__ movzwl(rax, rax));
2019 break;
2020 case Bytecodes::_i2s:
2021 __ shll(rax, 16); // truncate upper 16 bits
2022 __ sarl(rax, 16); // and sign-extend short
2023 LP64_ONLY(__ movswl(rax, rax));
2024 break;
2025 case Bytecodes::_l2i:
2026 /* nothing to do */
2027 break;
2028 case Bytecodes::_l2f:
2029 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
2030 // 64-bit long values to floats. On 32-bit platforms it is not possible
2031 // to use that instruction with 64-bit operands, therefore the FPU is
2032 // used to perform the conversion.
2033 __ push(rdx); // store long on tos
2034 __ push(rax);
2035 __ fild_d(at_rsp()); // load long to ST0
2036 __ f2ieee(); // truncate to float size
2037 __ pop(rcx); // adjust rsp
2038 __ pop(rcx);
2039 if (UseSSE >= 1) {
2040 __ push_f();
2041 __ pop_f(xmm0);
2042 }
2043 break;
2044 case Bytecodes::_l2d:
2045 // On 32-bit platforms the FPU is used for conversion because on
2046 // 32-bit platforms it is not not possible to use the cvtsi2sdq
2047 // instruction with 64-bit operands.
2048 __ push(rdx); // store long on tos
2049 __ push(rax);
2050 __ fild_d(at_rsp()); // load long to ST0
2051 __ d2ieee(); // truncate to double size
2052 __ pop(rcx); // adjust rsp
2053 __ pop(rcx);
2054 if (UseSSE >= 2) {
2055 __ push_d();
2056 __ pop_d(xmm0);
2057 }
2058 break;
2059 case Bytecodes::_f2i:
2060 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
2061 // as it returns 0 for any NaN.
2062 if (UseSSE >= 1) {
2063 __ push_f(xmm0);
2064 } else {
2065 __ push(rcx); // reserve space for argument
2066 __ fstp_s(at_rsp()); // pass float argument on stack
2067 }
2068 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
2069 break;
2070 case Bytecodes::_f2l:
2071 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
2072 // as it returns 0 for any NaN.
2073 if (UseSSE >= 1) {
2074 __ push_f(xmm0);
2075 } else {
2076 __ push(rcx); // reserve space for argument
2077 __ fstp_s(at_rsp()); // pass float argument on stack
2078 }
2079 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2080 break;
2081 case Bytecodes::_f2d:
2082 if (UseSSE < 1) {
2083 /* nothing to do */
2084 } else if (UseSSE == 1) {
2085 __ push_f(xmm0);
2086 __ pop_f();
2087 } else { // UseSSE >= 2
2088 __ cvtss2sd(xmm0, xmm0);
2089 }
2090 break;
2091 case Bytecodes::_d2i:
2092 if (UseSSE >= 2) {
2093 __ push_d(xmm0);
2094 } else {
2095 __ push(rcx); // reserve space for argument
2096 __ push(rcx);
2097 __ fstp_d(at_rsp()); // pass double argument on stack
2098 }
2099 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2100 break;
2101 case Bytecodes::_d2l:
2102 if (UseSSE >= 2) {
2103 __ push_d(xmm0);
2104 } else {
2105 __ push(rcx); // reserve space for argument
2106 __ push(rcx);
2107 __ fstp_d(at_rsp()); // pass double argument on stack
2108 }
2109 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2110 break;
2111 case Bytecodes::_d2f:
2112 if (UseSSE <= 1) {
2113 __ push(rcx); // reserve space for f2ieee()
2114 __ f2ieee(); // truncate to float size
2115 __ pop(rcx); // adjust rsp
2116 if (UseSSE == 1) {
2117 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2118 // the conversion is performed using the FPU in this case.
2119 __ push_f();
2120 __ pop_f(xmm0);
2121 }
2122 } else { // UseSSE >= 2
2123 __ cvtsd2ss(xmm0, xmm0);
2124 }
2125 break;
2126 default :
2127 ShouldNotReachHere();
2128 }
2129 #endif
2130 }
2131
2132 void TemplateTable::lcmp() {
2133 transition(ltos, itos);
2134 #ifdef _LP64
2135 Label done;
2136 __ pop_l(rdx);
2137 __ cmpq(rdx, rax);
2138 __ movl(rax, -1);
2139 __ jccb(Assembler::less, done);
2140 __ setb(Assembler::notEqual, rax);
2141 __ movzbl(rax, rax);
2142 __ bind(done);
2143 #else
2144
2145 // y = rdx:rax
2146 __ pop_l(rbx, rcx); // get x = rcx:rbx
2147 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2148 __ mov(rax, rcx);
2149 #endif
2150 }
2151
2152 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2153 if ((is_float && UseSSE >= 1) ||
2154 (!is_float && UseSSE >= 2)) {
2155 Label done;
2156 if (is_float) {
2157 // XXX get rid of pop here, use ... reg, mem32
2158 __ pop_f(xmm1);
2159 __ ucomiss(xmm1, xmm0);
2160 } else {
2161 // XXX get rid of pop here, use ... reg, mem64
2162 __ pop_d(xmm1);
2163 __ ucomisd(xmm1, xmm0);
2164 }
2165 if (unordered_result < 0) {
2166 __ movl(rax, -1);
2167 __ jccb(Assembler::parity, done);
2168 __ jccb(Assembler::below, done);
2169 __ setb(Assembler::notEqual, rdx);
2170 __ movzbl(rax, rdx);
2171 } else {
2172 __ movl(rax, 1);
2173 __ jccb(Assembler::parity, done);
2174 __ jccb(Assembler::above, done);
2175 __ movl(rax, 0);
2176 __ jccb(Assembler::equal, done);
2177 __ decrementl(rax);
2178 }
2179 __ bind(done);
2180 } else {
2181 #ifdef _LP64
2182 ShouldNotReachHere();
2183 #else
2184 if (is_float) {
2185 __ fld_s(at_rsp());
2186 } else {
2187 __ fld_d(at_rsp());
2188 __ pop(rdx);
2189 }
2190 __ pop(rcx);
2191 __ fcmp2int(rax, unordered_result < 0);
2192 #endif // _LP64
2193 }
2194 }
2195
2196 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2197 if (ValueTypesThreadLocalRecycling) {
2198 Label no_vt_recycling, no_fixing_required;
2199 const Register thread1 = NOT_LP64(rbx) LP64_ONLY(r15_thread);
2200 NOT_LP64(__ get_thread(thread1));
2201 __ movptr(rbx, Address(thread1, in_bytes(JavaThread::vt_alloc_ptr_offset())));
2202 __ testptr(rbx, rbx);
2203 __ jcc(Assembler::zero, no_vt_recycling);
2204 __ movptr(rcx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize));
2205 __ testptr(rcx, rcx);
2206 __ jcc(Assembler::notZero, no_fixing_required);
2207 // vt_alloc_ptr in JavaThread is non-null but frame vt_alloc_ptr is null
2208 // which means frame vt_alloc_ptr needs to be initialized
2209 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::fix_frame_vt_alloc_ptr));
2210 __ movptr(rcx, Address(rbp, frame::interpreter_frame_vt_alloc_ptr_offset * wordSize));
2211 __ bind(no_fixing_required);
2212 __ testptr(rcx, rbx);
2213 __ jcc(Assembler::equal, no_vt_recycling);
2214 __ andptr(rcx, VTBufferChunk::chunk_mask());
2215 __ movl(rcx, Address(rcx, VTBufferChunk::index_offset()));
2216 __ andptr(rbx, VTBufferChunk::chunk_mask());
2217 __ movl(rbx, Address(rbx, VTBufferChunk::index_offset()));
2218 __ subl(rbx, rcx);
2219 __ get_method(rcx);
2220 __ movl(rcx, Address(rcx, Method::max_vt_buffer_offset()));
2221 __ cmpl(rbx, rcx);
2222 __ jcc(Assembler::lessEqual, no_vt_recycling);
2223 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::recycle_buffered_values));
2224 __ bind(no_vt_recycling);
2225 }
2226
2227 __ get_method(rcx); // rcx holds method
2228 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2229 // holds bumped taken count
2230
2231 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2232 InvocationCounter::counter_offset();
2233 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2234 InvocationCounter::counter_offset();
2235
2236 // Load up edx with the branch displacement
2237 if (is_wide) {
2238 __ movl(rdx, at_bcp(1));
2239 } else {
2240 __ load_signed_short(rdx, at_bcp(1));
2241 }
2242 __ bswapl(rdx);
2243
2244 if (!is_wide) {
2245 __ sarl(rdx, 16);
2246 }
2247 LP64_ONLY(__ movl2ptr(rdx, rdx));
2248
2249 // Handle all the JSR stuff here, then exit.
2250 // It's much shorter and cleaner than intermingling with the non-JSR
2251 // normal-branch stuff occurring below.
2252 if (is_jsr) {
2253 // Pre-load the next target bytecode into rbx
2254 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2255
2256 // compute return address as bci in rax
2257 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2258 in_bytes(ConstMethod::codes_offset())));
2259 __ subptr(rax, Address(rcx, Method::const_offset()));
2260 // Adjust the bcp in r13 by the displacement in rdx
2261 __ addptr(rbcp, rdx);
2262 // jsr returns atos that is not an oop
2263 __ push_i(rax);
2264 __ dispatch_only(vtos, true);
2265 return;
2266 }
2267
2268 // Normal (non-jsr) branch handling
2269
2270 // Adjust the bcp in r13 by the displacement in rdx
2271 __ addptr(rbcp, rdx);
2272
2273 assert(UseLoopCounter || !UseOnStackReplacement,
2274 "on-stack-replacement requires loop counters");
2275 Label backedge_counter_overflow;
2276 Label profile_method;
2277 Label dispatch;
2278 if (UseLoopCounter) {
2279 // increment backedge counter for backward branches
2280 // rax: MDO
2281 // rbx: MDO bumped taken-count
2282 // rcx: method
2283 // rdx: target offset
2284 // r13: target bcp
2285 // r14: locals pointer
2286 __ testl(rdx, rdx); // check if forward or backward branch
2287 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2288
2289 // check if MethodCounters exists
2290 Label has_counters;
2291 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2292 __ testptr(rax, rax);
2293 __ jcc(Assembler::notZero, has_counters);
2294 __ push(rdx);
2295 __ push(rcx);
2296 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2297 rcx);
2298 __ pop(rcx);
2299 __ pop(rdx);
2300 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2301 __ testptr(rax, rax);
2302 __ jcc(Assembler::zero, dispatch);
2303 __ bind(has_counters);
2304
2305 if (TieredCompilation) {
2306 Label no_mdo;
2307 int increment = InvocationCounter::count_increment;
2308 if (ProfileInterpreter) {
2309 // Are we profiling?
2310 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2311 __ testptr(rbx, rbx);
2312 __ jccb(Assembler::zero, no_mdo);
2313 // Increment the MDO backedge counter
2314 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2315 in_bytes(InvocationCounter::counter_offset()));
2316 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2317 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
2318 rax, false, Assembler::zero, &backedge_counter_overflow);
2319 __ jmp(dispatch);
2320 }
2321 __ bind(no_mdo);
2322 // Increment backedge counter in MethodCounters*
2323 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2324 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2325 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2326 rax, false, Assembler::zero, &backedge_counter_overflow);
2327 } else { // not TieredCompilation
2328 // increment counter
2329 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2330 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2331 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2332 __ movl(Address(rcx, be_offset), rax); // store counter
2333
2334 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2335
2336 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2337 __ addl(rax, Address(rcx, be_offset)); // add both counters
2338
2339 if (ProfileInterpreter) {
2340 // Test to see if we should create a method data oop
2341 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2342 __ jcc(Assembler::less, dispatch);
2343
2344 // if no method data exists, go to profile method
2345 __ test_method_data_pointer(rax, profile_method);
2346
2347 if (UseOnStackReplacement) {
2348 // check for overflow against rbx which is the MDO taken count
2349 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2350 __ jcc(Assembler::below, dispatch);
2351
2352 // When ProfileInterpreter is on, the backedge_count comes
2353 // from the MethodData*, which value does not get reset on
2354 // the call to frequency_counter_overflow(). To avoid
2355 // excessive calls to the overflow routine while the method is
2356 // being compiled, add a second test to make sure the overflow
2357 // function is called only once every overflow_frequency.
2358 const int overflow_frequency = 1024;
2359 __ andl(rbx, overflow_frequency - 1);
2360 __ jcc(Assembler::zero, backedge_counter_overflow);
2361
2362 }
2363 } else {
2364 if (UseOnStackReplacement) {
2365 // check for overflow against rax, which is the sum of the
2366 // counters
2367 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2368 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2369
2370 }
2371 }
2372 }
2373 __ bind(dispatch);
2374 }
2375
2376 // Pre-load the next target bytecode into rbx
2377 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2378
2379 // continue with the bytecode @ target
2380 // rax: return bci for jsr's, unused otherwise
2381 // rbx: target bytecode
2382 // r13: target bcp
2383 __ dispatch_only(vtos, true);
2384
2385 if (UseLoopCounter) {
2386 if (ProfileInterpreter) {
2387 // Out-of-line code to allocate method data oop.
2388 __ bind(profile_method);
2389 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2390 __ set_method_data_pointer_for_bcp();
2391 __ jmp(dispatch);
2392 }
2393
2394 if (UseOnStackReplacement) {
2395 // invocation counter overflow
2396 __ bind(backedge_counter_overflow);
2397 __ negptr(rdx);
2398 __ addptr(rdx, rbcp); // branch bcp
2399 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2400 __ call_VM(noreg,
2401 CAST_FROM_FN_PTR(address,
2402 InterpreterRuntime::frequency_counter_overflow),
2403 rdx);
2404
2405 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2406 // rdx: scratch
2407 // r14: locals pointer
2408 // r13: bcp
2409 __ testptr(rax, rax); // test result
2410 __ jcc(Assembler::zero, dispatch); // no osr if null
2411 // nmethod may have been invalidated (VM may block upon call_VM return)
2412 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2413 __ jcc(Assembler::notEqual, dispatch);
2414
2415 // We have the address of an on stack replacement routine in rax.
2416 // In preparation of invoking it, first we must migrate the locals
2417 // and monitors from off the interpreter frame on the stack.
2418 // Ensure to save the osr nmethod over the migration call,
2419 // it will be preserved in rbx.
2420 __ mov(rbx, rax);
2421
2422 NOT_LP64(__ get_thread(rcx));
2423
2424 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2425
2426 // rax is OSR buffer, move it to expected parameter location
2427 LP64_ONLY(__ mov(j_rarg0, rax));
2428 NOT_LP64(__ mov(rcx, rax));
2429 // We use j_rarg definitions here so that registers don't conflict as parameter
2430 // registers change across platforms as we are in the midst of a calling
2431 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2432
2433 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2434 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2435
2436 // pop the interpreter frame
2437 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2438 __ leave(); // remove frame anchor
2439 __ pop(retaddr); // get return address
2440 __ mov(rsp, sender_sp); // set sp to sender sp
2441 // Ensure compiled code always sees stack at proper alignment
2442 __ andptr(rsp, -(StackAlignmentInBytes));
2443
2444 // unlike x86 we need no specialized return from compiled code
2445 // to the interpreter or the call stub.
2446
2447 // push the return address
2448 __ push(retaddr);
2449
2450 // and begin the OSR nmethod
2451 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2452 }
2453 }
2454 }
2455
2456 void TemplateTable::if_0cmp(Condition cc) {
2457 transition(itos, vtos);
2458 // assume branch is more often taken than not (loops use backward branches)
2459 Label not_taken;
2460 __ testl(rax, rax);
2461 __ jcc(j_not(cc), not_taken);
2462 branch(false, false);
2463 __ bind(not_taken);
2464 __ profile_not_taken_branch(rax);
2465 }
2466
2467 void TemplateTable::if_icmp(Condition cc) {
2468 transition(itos, vtos);
2469 // assume branch is more often taken than not (loops use backward branches)
2470 Label not_taken;
2471 __ pop_i(rdx);
2472 __ cmpl(rdx, rax);
2473 __ jcc(j_not(cc), not_taken);
2474 branch(false, false);
2475 __ bind(not_taken);
2476 __ profile_not_taken_branch(rax);
2477 }
2478
2479 void TemplateTable::if_nullcmp(Condition cc) {
2480 transition(atos, vtos);
2481 // assume branch is more often taken than not (loops use backward branches)
2482 Label not_taken;
2483 __ testptr(rax, rax);
2484 __ jcc(j_not(cc), not_taken);
2485 branch(false, false);
2486 __ bind(not_taken);
2487 __ profile_not_taken_branch(rax);
2488 }
2489
2490 void TemplateTable::if_acmp(Condition cc) {
2491 transition(atos, vtos);
2492 // assume branch is more often taken than not (loops use backward branches)
2493 Label not_taken;
2494 __ pop_ptr(rdx);
2495 __ cmpoop(rdx, rax);
2496 __ jcc(j_not(cc), not_taken);
2497 branch(false, false);
2498 __ bind(not_taken);
2499 __ profile_not_taken_branch(rax);
2500 }
2501
2502 void TemplateTable::ret() {
2503 transition(vtos, vtos);
2504 locals_index(rbx);
2505 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2506 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2507 __ profile_ret(rbx, rcx);
2508 __ get_method(rax);
2509 __ movptr(rbcp, Address(rax, Method::const_offset()));
2510 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2511 ConstMethod::codes_offset()));
2512 __ dispatch_next(vtos, 0, true);
2513 }
2514
2515 void TemplateTable::wide_ret() {
2516 transition(vtos, vtos);
2517 locals_index_wide(rbx);
2518 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2519 __ profile_ret(rbx, rcx);
2520 __ get_method(rax);
2521 __ movptr(rbcp, Address(rax, Method::const_offset()));
2522 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2523 __ dispatch_next(vtos, 0, true);
2524 }
2525
2526 void TemplateTable::tableswitch() {
2527 Label default_case, continue_execution;
2528 transition(itos, vtos);
2529
2530 // align r13/rsi
2531 __ lea(rbx, at_bcp(BytesPerInt));
2532 __ andptr(rbx, -BytesPerInt);
2533 // load lo & hi
2534 __ movl(rcx, Address(rbx, BytesPerInt));
2535 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2536 __ bswapl(rcx);
2537 __ bswapl(rdx);
2538 // check against lo & hi
2539 __ cmpl(rax, rcx);
2540 __ jcc(Assembler::less, default_case);
2541 __ cmpl(rax, rdx);
2542 __ jcc(Assembler::greater, default_case);
2543 // lookup dispatch offset
2544 __ subl(rax, rcx);
2545 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2546 __ profile_switch_case(rax, rbx, rcx);
2547 // continue execution
2548 __ bind(continue_execution);
2549 __ bswapl(rdx);
2550 LP64_ONLY(__ movl2ptr(rdx, rdx));
2551 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2552 __ addptr(rbcp, rdx);
2553 __ dispatch_only(vtos, true);
2554 // handle default
2555 __ bind(default_case);
2556 __ profile_switch_default(rax);
2557 __ movl(rdx, Address(rbx, 0));
2558 __ jmp(continue_execution);
2559 }
2560
2561 void TemplateTable::lookupswitch() {
2562 transition(itos, itos);
2563 __ stop("lookupswitch bytecode should have been rewritten");
2564 }
2565
2566 void TemplateTable::fast_linearswitch() {
2567 transition(itos, vtos);
2568 Label loop_entry, loop, found, continue_execution;
2569 // bswap rax so we can avoid bswapping the table entries
2570 __ bswapl(rax);
2571 // align r13
2572 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2573 // this instruction (change offsets
2574 // below)
2575 __ andptr(rbx, -BytesPerInt);
2576 // set counter
2577 __ movl(rcx, Address(rbx, BytesPerInt));
2578 __ bswapl(rcx);
2579 __ jmpb(loop_entry);
2580 // table search
2581 __ bind(loop);
2582 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2583 __ jcc(Assembler::equal, found);
2584 __ bind(loop_entry);
2585 __ decrementl(rcx);
2586 __ jcc(Assembler::greaterEqual, loop);
2587 // default case
2588 __ profile_switch_default(rax);
2589 __ movl(rdx, Address(rbx, 0));
2590 __ jmp(continue_execution);
2591 // entry found -> get offset
2592 __ bind(found);
2593 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2594 __ profile_switch_case(rcx, rax, rbx);
2595 // continue execution
2596 __ bind(continue_execution);
2597 __ bswapl(rdx);
2598 __ movl2ptr(rdx, rdx);
2599 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2600 __ addptr(rbcp, rdx);
2601 __ dispatch_only(vtos, true);
2602 }
2603
2604 void TemplateTable::fast_binaryswitch() {
2605 transition(itos, vtos);
2606 // Implementation using the following core algorithm:
2607 //
2608 // int binary_search(int key, LookupswitchPair* array, int n) {
2609 // // Binary search according to "Methodik des Programmierens" by
2610 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2611 // int i = 0;
2612 // int j = n;
2613 // while (i+1 < j) {
2614 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2615 // // with Q: for all i: 0 <= i < n: key < a[i]
2616 // // where a stands for the array and assuming that the (inexisting)
2617 // // element a[n] is infinitely big.
2618 // int h = (i + j) >> 1;
2619 // // i < h < j
2620 // if (key < array[h].fast_match()) {
2621 // j = h;
2622 // } else {
2623 // i = h;
2624 // }
2625 // }
2626 // // R: a[i] <= key < a[i+1] or Q
2627 // // (i.e., if key is within array, i is the correct index)
2628 // return i;
2629 // }
2630
2631 // Register allocation
2632 const Register key = rax; // already set (tosca)
2633 const Register array = rbx;
2634 const Register i = rcx;
2635 const Register j = rdx;
2636 const Register h = rdi;
2637 const Register temp = rsi;
2638
2639 // Find array start
2640 NOT_LP64(__ save_bcp());
2641
2642 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2643 // get rid of this
2644 // instruction (change
2645 // offsets below)
2646 __ andptr(array, -BytesPerInt);
2647
2648 // Initialize i & j
2649 __ xorl(i, i); // i = 0;
2650 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2651
2652 // Convert j into native byteordering
2653 __ bswapl(j);
2654
2655 // And start
2656 Label entry;
2657 __ jmp(entry);
2658
2659 // binary search loop
2660 {
2661 Label loop;
2662 __ bind(loop);
2663 // int h = (i + j) >> 1;
2664 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2665 __ sarl(h, 1); // h = (i + j) >> 1;
2666 // if (key < array[h].fast_match()) {
2667 // j = h;
2668 // } else {
2669 // i = h;
2670 // }
2671 // Convert array[h].match to native byte-ordering before compare
2672 __ movl(temp, Address(array, h, Address::times_8));
2673 __ bswapl(temp);
2674 __ cmpl(key, temp);
2675 // j = h if (key < array[h].fast_match())
2676 __ cmov32(Assembler::less, j, h);
2677 // i = h if (key >= array[h].fast_match())
2678 __ cmov32(Assembler::greaterEqual, i, h);
2679 // while (i+1 < j)
2680 __ bind(entry);
2681 __ leal(h, Address(i, 1)); // i+1
2682 __ cmpl(h, j); // i+1 < j
2683 __ jcc(Assembler::less, loop);
2684 }
2685
2686 // end of binary search, result index is i (must check again!)
2687 Label default_case;
2688 // Convert array[i].match to native byte-ordering before compare
2689 __ movl(temp, Address(array, i, Address::times_8));
2690 __ bswapl(temp);
2691 __ cmpl(key, temp);
2692 __ jcc(Assembler::notEqual, default_case);
2693
2694 // entry found -> j = offset
2695 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2696 __ profile_switch_case(i, key, array);
2697 __ bswapl(j);
2698 LP64_ONLY(__ movslq(j, j));
2699
2700 NOT_LP64(__ restore_bcp());
2701 NOT_LP64(__ restore_locals()); // restore rdi
2702
2703 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2704 __ addptr(rbcp, j);
2705 __ dispatch_only(vtos, true);
2706
2707 // default case -> j = default offset
2708 __ bind(default_case);
2709 __ profile_switch_default(i);
2710 __ movl(j, Address(array, -2 * BytesPerInt));
2711 __ bswapl(j);
2712 LP64_ONLY(__ movslq(j, j));
2713
2714 NOT_LP64(__ restore_bcp());
2715 NOT_LP64(__ restore_locals());
2716
2717 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2718 __ addptr(rbcp, j);
2719 __ dispatch_only(vtos, true);
2720 }
2721
2722 void TemplateTable::_return(TosState state) {
2723 transition(state, state);
2724
2725 assert(_desc->calls_vm(),
2726 "inconsistent calls_vm information"); // call in remove_activation
2727
2728 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2729 assert(state == vtos, "only valid state");
2730 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2731 __ movptr(robj, aaddress(0));
2732 __ load_klass(rdi, robj);
2733 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2734 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2735 Label skip_register_finalizer;
2736 __ jcc(Assembler::zero, skip_register_finalizer);
2737
2738 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2739
2740 __ bind(skip_register_finalizer);
2741 }
2742
2743 if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2744 Label no_safepoint;
2745 NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2746 #ifdef _LP64
2747 __ testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2748 #else
2749 const Register thread = rdi;
2750 __ get_thread(thread);
2751 __ testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2752 #endif
2753 __ jcc(Assembler::zero, no_safepoint);
2754 __ push(state);
2755 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2756 InterpreterRuntime::at_safepoint));
2757 __ pop(state);
2758 __ bind(no_safepoint);
2759 }
2760
2761 if (state == atos) {
2762 if (ValueTypesBufferMaxMemory > 0) {
2763 Label notBuffered;
2764
2765 __ test_value_is_not_buffered(rax, rbx, notBuffered);
2766 const Register thread1 = NOT_LP64(rcx) LP64_ONLY(r15_thread);
2767 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::return_value), rax);
2768 NOT_LP64(__ get_thread(thread1));
2769 __ get_vm_result(rax, thread1);
2770 __ bind(notBuffered);
2771 }
2772 }
2773
2774 // Narrow result if state is itos but result type is smaller.
2775 // Need to narrow in the return bytecode rather than in generate_return_entry
2776 // since compiled code callers expect the result to already be narrowed.
2777 if (state == itos) {
2778 __ narrow(rax);
2779 }
2780
2781 __ remove_activation(state, rbcp, true, true, true, /*state == qtos*/ false && ValueTypeReturnedAsFields);
2782
2783 __ jmp(rbcp);
2784 }
2785
2786 // ----------------------------------------------------------------------------
2787 // Volatile variables demand their effects be made known to all CPU's
2788 // in order. Store buffers on most chips allow reads & writes to
2789 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2790 // without some kind of memory barrier (i.e., it's not sufficient that
2791 // the interpreter does not reorder volatile references, the hardware
2792 // also must not reorder them).
2793 //
2794 // According to the new Java Memory Model (JMM):
2795 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2796 // writes act as aquire & release, so:
2797 // (2) A read cannot let unrelated NON-volatile memory refs that
2798 // happen after the read float up to before the read. It's OK for
2799 // non-volatile memory refs that happen before the volatile read to
2800 // float down below it.
2801 // (3) Similar a volatile write cannot let unrelated NON-volatile
2802 // memory refs that happen BEFORE the write float down to after the
2803 // write. It's OK for non-volatile memory refs that happen after the
2804 // volatile write to float up before it.
2805 //
2806 // We only put in barriers around volatile refs (they are expensive),
2807 // not _between_ memory refs (that would require us to track the
2808 // flavor of the previous memory refs). Requirements (2) and (3)
2809 // require some barriers before volatile stores and after volatile
2810 // loads. These nearly cover requirement (1) but miss the
2811 // volatile-store-volatile-load case. This final case is placed after
2812 // volatile-stores although it could just as well go before
2813 // volatile-loads.
2814
2815 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2816 // Helper function to insert a is-volatile test and memory barrier
2817 if(!os::is_MP()) return; // Not needed on single CPU
2818 __ membar(order_constraint);
2819 }
2820
2821 void TemplateTable::resolve_cache_and_index(int byte_no,
2822 Register Rcache,
2823 Register index,
2824 size_t index_size) {
2825 const Register temp = rbx;
2826 assert_different_registers(Rcache, index, temp);
2827
2828 Label resolved;
2829
2830 Bytecodes::Code code = bytecode();
2831 switch (code) {
2832 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2833 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2834 default: break;
2835 }
2836
2837 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2838 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2839 __ cmpl(temp, code); // have we resolved this bytecode?
2840 __ jcc(Assembler::equal, resolved);
2841
2842 // resolve first time through
2843 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2844 __ movl(temp, code);
2845 __ call_VM(noreg, entry, temp);
2846 // Update registers with resolved info
2847 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2848 __ bind(resolved);
2849 }
2850
2851 // The cache and index registers must be set before call
2852 void TemplateTable::load_field_cp_cache_entry(Register obj,
2853 Register cache,
2854 Register index,
2855 Register off,
2856 Register flags,
2857 bool is_static = false) {
2858 assert_different_registers(cache, index, flags, off);
2859
2860 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2861 // Field offset
2862 __ movptr(off, Address(cache, index, Address::times_ptr,
2863 in_bytes(cp_base_offset +
2864 ConstantPoolCacheEntry::f2_offset())));
2865 // Flags
2866 __ movl(flags, Address(cache, index, Address::times_ptr,
2867 in_bytes(cp_base_offset +
2868 ConstantPoolCacheEntry::flags_offset())));
2869
2870 // klass overwrite register
2871 if (is_static) {
2872 __ movptr(obj, Address(cache, index, Address::times_ptr,
2873 in_bytes(cp_base_offset +
2874 ConstantPoolCacheEntry::f1_offset())));
2875 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2876 __ movptr(obj, Address(obj, mirror_offset));
2877 __ resolve_oop_handle(obj);
2878 }
2879 }
2880
2881 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2882 Register method,
2883 Register itable_index,
2884 Register flags,
2885 bool is_invokevirtual,
2886 bool is_invokevfinal, /*unused*/
2887 bool is_invokedynamic) {
2888 // setup registers
2889 const Register cache = rcx;
2890 const Register index = rdx;
2891 assert_different_registers(method, flags);
2892 assert_different_registers(method, cache, index);
2893 assert_different_registers(itable_index, flags);
2894 assert_different_registers(itable_index, cache, index);
2895 // determine constant pool cache field offsets
2896 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2897 const int method_offset = in_bytes(
2898 ConstantPoolCache::base_offset() +
2899 ((byte_no == f2_byte)
2900 ? ConstantPoolCacheEntry::f2_offset()
2901 : ConstantPoolCacheEntry::f1_offset()));
2902 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2903 ConstantPoolCacheEntry::flags_offset());
2904 // access constant pool cache fields
2905 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2906 ConstantPoolCacheEntry::f2_offset());
2907
2908 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2909 resolve_cache_and_index(byte_no, cache, index, index_size);
2910 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2911
2912 if (itable_index != noreg) {
2913 // pick up itable or appendix index from f2 also:
2914 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2915 }
2916 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2917 }
2918
2919 // The registers cache and index expected to be set before call.
2920 // Correct values of the cache and index registers are preserved.
2921 void TemplateTable::jvmti_post_field_access(Register cache,
2922 Register index,
2923 bool is_static,
2924 bool has_tos) {
2925 if (JvmtiExport::can_post_field_access()) {
2926 // Check to see if a field access watch has been set before we take
2927 // the time to call into the VM.
2928 Label L1;
2929 assert_different_registers(cache, index, rax);
2930 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2931 __ testl(rax,rax);
2932 __ jcc(Assembler::zero, L1);
2933
2934 // cache entry pointer
2935 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2936 __ shll(index, LogBytesPerWord);
2937 __ addptr(cache, index);
2938 if (is_static) {
2939 __ xorptr(rax, rax); // NULL object reference
2940 } else {
2941 __ pop(atos); // Get the object
2942 __ verify_oop(rax);
2943 __ push(atos); // Restore stack state
2944 }
2945 // rax,: object pointer or NULL
2946 // cache: cache entry pointer
2947 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2948 rax, cache);
2949 __ get_cache_and_index_at_bcp(cache, index, 1);
2950 __ bind(L1);
2951 }
2952 }
2953
2954 void TemplateTable::pop_and_check_object(Register r) {
2955 __ pop_ptr(r);
2956 __ null_check(r); // for field access must check obj.
2957 __ verify_oop(r);
2958 }
2959
2960 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2961 transition(vtos, vtos);
2962
2963 const Register cache = rcx;
2964 const Register index = rdx;
2965 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2966 const Register off = rbx;
2967 const Register flags = rax;
2968 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2969 const Register flags2 = rdx;
2970
2971 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2972 jvmti_post_field_access(cache, index, is_static, false);
2973 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2974
2975 const Address field(obj, off, Address::times_1, 0*wordSize);
2976 NOT_LP64(const Address hi(obj, off, Address::times_1, 1*wordSize));
2977
2978 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj, notValueType, notDouble;
2979
2980 if (!is_static) {
2981 __ movptr(rcx, Address(cache, index, Address::times_ptr,
2982 in_bytes(ConstantPoolCache::base_offset() +
2983 ConstantPoolCacheEntry::f1_offset())));
2984 }
2985
2986 __ movl(flags2, flags);
2987
2988 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2989 // Make sure we don't need to mask edx after the above shift
2990 assert(btos == 0, "change code, btos != 0");
2991
2992 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2993
2994 __ jcc(Assembler::notZero, notByte);
2995 // btos
2996 if (!is_static) pop_and_check_object(obj);
2997 __ load_signed_byte(rax, field);
2998 __ push(btos);
2999 // Rewrite bytecode to be faster
3000 if (!is_static && rc == may_rewrite) {
3001 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
3002 }
3003 __ jmp(Done);
3004
3005 __ bind(notByte);
3006
3007 __ cmpl(flags, ztos);
3008 __ jcc(Assembler::notEqual, notBool);
3009 if (!is_static) pop_and_check_object(obj);
3010 // ztos (same code as btos)
3011 __ load_signed_byte(rax, field);
3012 __ push(ztos);
3013 // Rewrite bytecode to be faster
3014 if (!is_static && rc == may_rewrite) {
3015 // use btos rewriting, no truncating to t/f bit is needed for getfield.
3016 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
3017 }
3018 __ jmp(Done);
3019
3020 __ bind(notBool);
3021 __ cmpl(flags, atos);
3022 __ jcc(Assembler::notEqual, notObj);
3023 // atos
3024 if (!EnableValhalla) {
3025 if (!is_static) pop_and_check_object(obj);
3026 do_oop_load(_masm, field, rax);
3027 __ push(atos);
3028 if (!is_static && rc == may_rewrite) {
3029 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
3030 }
3031 __ jmp(Done);
3032 } else {
3033 if (is_static) {
3034 __ load_heap_oop(rax, field);
3035 Label isFlattenable, uninitialized;
3036 // Issue below if the static field has not been initialized yet
3037 __ test_field_is_flattenable(flags2, rscratch1, isFlattenable);
3038 // Not flattenable case
3039 __ push(atos);
3040 __ jmp(Done);
3041 // Flattenable case, must not return null even if uninitialized
3042 __ bind(isFlattenable);
3043 __ testptr(rax, rax);
3044 __ jcc(Assembler::zero, uninitialized);
3045 __ push(atos);
3046 __ jmp(Done);
3047 __ bind(uninitialized);
3048 __ andl(flags2, ConstantPoolCacheEntry::field_index_mask);
3049 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_static_value_field),
3050 obj, flags2);
3051 __ verify_oop(rax);
3052 __ push(atos);
3053 __ jmp(Done);
3054 } else {
3055 Label isFlattened, nonnull, isFlattenable, rewriteFlattenable;
3056 __ test_field_is_flattenable(flags2, rscratch1, isFlattenable);
3057 // Non-flattenable field case, also covers the object case
3058 pop_and_check_object(obj);
3059 __ load_heap_oop(rax, field);
3060 __ push(atos);
3061 if (rc == may_rewrite) {
3062 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
3063 }
3064 __ jmp(Done);
3065 __ bind(isFlattenable);
3066 __ test_field_is_flattened(flags2, rscratch1, isFlattened);
3067 // Non-flattened field case
3068 pop_and_check_object(obj);
3069 __ load_heap_oop(rax, field);
3070 __ testptr(rax, rax);
3071 __ jcc(Assembler::notZero, nonnull);
3072 __ andl(flags2, ConstantPoolCacheEntry::field_index_mask);
3073 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_instance_value_field),
3074 obj, flags2);
3075 __ bind(nonnull);
3076 __ verify_oop(rax);
3077 __ push(atos);
3078 __ jmp(rewriteFlattenable);
3079 __ bind(isFlattened);
3080 __ andl(flags2, ConstantPoolCacheEntry::field_index_mask);
3081 pop_and_check_object(rbx);
3082 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flattened_field),
3083 rbx, flags2, rcx);
3084 __ verify_oop(rax);
3085 __ push(atos);
3086 __ bind(rewriteFlattenable);
3087 if (rc == may_rewrite) {
3088 patch_bytecode(Bytecodes::_fast_qgetfield, bc, rbx);
3089 }
3090 __ jmp(Done);
3091 }
3092 }
3093
3094 __ bind(notObj);
3095
3096 if (!is_static) pop_and_check_object(obj);
3097
3098 __ cmpl(flags, itos);
3099 __ jcc(Assembler::notEqual, notInt);
3100 // itos
3101 __ movl(rax, field);
3102 __ push(itos);
3103 // Rewrite bytecode to be faster
3104 if (!is_static && rc == may_rewrite) {
3105 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
3106 }
3107 __ jmp(Done);
3108
3109 __ bind(notInt);
3110 __ cmpl(flags, ctos);
3111 __ jcc(Assembler::notEqual, notChar);
3112 // ctos
3113 __ load_unsigned_short(rax, field);
3114 __ push(ctos);
3115 // Rewrite bytecode to be faster
3116 if (!is_static && rc == may_rewrite) {
3117 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
3118 }
3119 __ jmp(Done);
3120
3121 __ bind(notChar);
3122 __ cmpl(flags, stos);
3123 __ jcc(Assembler::notEqual, notShort);
3124 // stos
3125 __ load_signed_short(rax, field);
3126 __ push(stos);
3127 // Rewrite bytecode to be faster
3128 if (!is_static && rc == may_rewrite) {
3129 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
3130 }
3131 __ jmp(Done);
3132
3133 __ bind(notShort);
3134 __ cmpl(flags, ltos);
3135 __ jcc(Assembler::notEqual, notLong);
3136 // ltos
3137
3138 #ifndef _LP64
3139 // Generate code as if volatile. There just aren't enough registers to
3140 // save that information and this code is faster than the test.
3141 __ fild_d(field); // Must load atomically
3142 __ subptr(rsp,2*wordSize); // Make space for store
3143 __ fistp_d(Address(rsp,0));
3144 __ pop(rax);
3145 __ pop(rdx);
3146 #else
3147 __ movq(rax, field);
3148 #endif
3149
3150 __ push(ltos);
3151 // Rewrite bytecode to be faster
3152 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
3153 __ jmp(Done);
3154
3155 __ bind(notLong);
3156 __ cmpl(flags, ftos);
3157 __ jcc(Assembler::notEqual, notFloat);
3158 // ftos
3159
3160 __ load_float(field);
3161 __ push(ftos);
3162 // Rewrite bytecode to be faster
3163 if (!is_static && rc == may_rewrite) {
3164 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
3165 }
3166 __ jmp(Done);
3167
3168 __ bind(notFloat);
3169 #ifdef ASSERT
3170 __ cmpl(flags, dtos);
3171 __ jcc(Assembler::notEqual, notDouble);
3172 #endif
3173 // dtos
3174 __ load_double(field);
3175 __ push(dtos);
3176 // Rewrite bytecode to be faster
3177 if (!is_static && rc == may_rewrite) {
3178 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
3179 }
3180 #ifdef ASSERT
3181 __ jmp(Done);
3182
3183
3184 __ bind(notDouble);
3185 __ stop("Bad state");
3186 #endif
3187
3188 __ bind(Done);
3189 // [jk] not needed currently
3190 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3191 // Assembler::LoadStore));
3192 }
3193
3194 void TemplateTable::getfield(int byte_no) {
3195 getfield_or_static(byte_no, false);
3196 }
3197
3198 void TemplateTable::nofast_getfield(int byte_no) {
3199 getfield_or_static(byte_no, false, may_not_rewrite);
3200 }
3201
3202 void TemplateTable::getstatic(int byte_no) {
3203 getfield_or_static(byte_no, true);
3204 }
3205
3206 void TemplateTable::withfield() {
3207 transition(vtos, atos);
3208
3209 Register cache = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
3210 Register index = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
3211
3212 resolve_cache_and_index(f2_byte, cache, index, sizeof(u2));
3213
3214 call_VM(rbx, CAST_FROM_FN_PTR(address, InterpreterRuntime::withfield), cache);
3215 // new value type is returned in rbx
3216 // stack adjustement is returned in rax
3217 __ verify_oop(rbx);
3218 __ addptr(rsp, rax);
3219 __ movptr(rax, rbx);
3220 }
3221
3222 // The registers cache and index expected to be set before call.
3223 // The function may destroy various registers, just not the cache and index registers.
3224 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3225
3226 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
3227 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
3228 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3229 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
3230
3231 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3232
3233 if (JvmtiExport::can_post_field_modification()) {
3234 // Check to see if a field modification watch has been set before
3235 // we take the time to call into the VM.
3236 Label L1;
3237 assert_different_registers(cache, index, rax);
3238 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3239 __ testl(rax, rax);
3240 __ jcc(Assembler::zero, L1);
3241
3242 __ get_cache_and_index_at_bcp(robj, RDX, 1);
3243
3244
3245 if (is_static) {
3246 // Life is simple. Null out the object pointer.
3247 __ xorl(RBX, RBX);
3248
3249 } else {
3250 // Life is harder. The stack holds the value on top, followed by
3251 // the object. We don't know the size of the value, though; it
3252 // could be one or two words depending on its type. As a result,
3253 // we must find the type to determine where the object is.
3254 #ifndef _LP64
3255 Label two_word, valsize_known;
3256 #endif
3257 __ movl(RCX, Address(robj, RDX,
3258 Address::times_ptr,
3259 in_bytes(cp_base_offset +
3260 ConstantPoolCacheEntry::flags_offset())));
3261 NOT_LP64(__ mov(rbx, rsp));
3262 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3263
3264 // Make sure we don't need to mask rcx after the above shift
3265 ConstantPoolCacheEntry::verify_tos_state_shift();
3266 #ifdef _LP64
3267 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
3268 __ cmpl(c_rarg3, ltos);
3269 __ cmovptr(Assembler::equal,
3270 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3271 __ cmpl(c_rarg3, dtos);
3272 __ cmovptr(Assembler::equal,
3273 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3274 #else
3275 __ cmpl(rcx, ltos);
3276 __ jccb(Assembler::equal, two_word);
3277 __ cmpl(rcx, dtos);
3278 __ jccb(Assembler::equal, two_word);
3279 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3280 __ jmpb(valsize_known);
3281
3282 __ bind(two_word);
3283 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3284
3285 __ bind(valsize_known);
3286 // setup object pointer
3287 __ movptr(rbx, Address(rbx, 0));
3288 #endif
3289 }
3290 // cache entry pointer
3291 __ addptr(robj, in_bytes(cp_base_offset));
3292 __ shll(RDX, LogBytesPerWord);
3293 __ addptr(robj, RDX);
3294 // object (tos)
3295 __ mov(RCX, rsp);
3296 // c_rarg1: object pointer set up above (NULL if static)
3297 // c_rarg2: cache entry pointer
3298 // c_rarg3: jvalue object on the stack
3299 __ call_VM(noreg,
3300 CAST_FROM_FN_PTR(address,
3301 InterpreterRuntime::post_field_modification),
3302 RBX, robj, RCX);
3303 __ get_cache_and_index_at_bcp(cache, index, 1);
3304 __ bind(L1);
3305 }
3306 }
3307
3308 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3309 transition(vtos, vtos);
3310
3311 const Register cache = rcx;
3312 const Register index = rdx;
3313 const Register obj = rcx;
3314 const Register off = rbx;
3315 const Register flags = rax;
3316 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3317 const Register flags2 = rdx;
3318
3319 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3320 jvmti_post_field_mod(cache, index, is_static);
3321 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3322
3323 // [jk] not needed currently
3324 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3325 // Assembler::StoreStore));
3326
3327 Label notVolatile, Done;
3328
3329 __ movl(flags2, flags);
3330
3331 // field addresses
3332 const Address field(obj, off, Address::times_1, 0*wordSize);
3333 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3334
3335 Label notByte, notBool, notInt, notShort, notChar,
3336 notLong, notFloat, notObj, notValueType, notDouble;
3337
3338 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3339
3340 assert(btos == 0, "change code, btos != 0");
3341 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3342 __ jcc(Assembler::notZero, notByte);
3343
3344 // btos
3345 {
3346 __ pop(btos);
3347 if (!is_static) pop_and_check_object(obj);
3348 __ movb(field, rax);
3349 if (!is_static && rc == may_rewrite) {
3350 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3351 }
3352 __ jmp(Done);
3353 }
3354
3355 __ bind(notByte);
3356 __ cmpl(flags, ztos);
3357 __ jcc(Assembler::notEqual, notBool);
3358
3359 // ztos
3360 {
3361 __ pop(ztos);
3362 if (!is_static) pop_and_check_object(obj);
3363 __ andl(rax, 0x1);
3364 __ movb(field, rax);
3365 if (!is_static && rc == may_rewrite) {
3366 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3367 }
3368 __ jmp(Done);
3369 }
3370
3371 __ bind(notBool);
3372 __ cmpl(flags, atos);
3373 __ jcc(Assembler::notEqual, notObj);
3374
3375 // atos
3376 {
3377 if (!EnableValhalla) {
3378 __ pop(atos);
3379 if (!is_static) pop_and_check_object(obj);
3380 // Store into the field
3381 do_oop_store(_masm, field, rax);
3382 if (!is_static && rc == may_rewrite) {
3383 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3384 }
3385 __ jmp(Done);
3386 } else {
3387 __ pop(atos);
3388 if (is_static) {
3389 Label notFlattenable, notBuffered;
3390 __ test_field_is_not_flattenable(flags2, rscratch1, notFlattenable);
3391 __ null_check(rax);
3392 __ bind(notFlattenable);
3393 if (ValueTypesBufferMaxMemory > 0) {
3394 __ test_value_is_not_buffered(rax, rscratch1, notBuffered);
3395 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_heap_copy),
3396 rax, off, obj);
3397 __ jmp(Done);
3398 __ bind(notBuffered);
3399 }
3400 do_oop_store(_masm, field, rax);
3401 __ jmp(Done);
3402 } else {
3403 Label isFlattenable, isFlattened, notBuffered, notBuffered2, rewriteNotFlattenable, rewriteFlattenable;
3404 __ test_field_is_flattenable(flags2, rscratch1, isFlattenable);
3405 // Not flattenable case, covers not flattenable values and objects
3406 pop_and_check_object(obj);
3407 // Store into the field
3408 if (ValueTypesBufferMaxMemory > 0) {
3409 __ test_value_is_not_buffered(rax, rscratch1, notBuffered);
3410 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_heap_copy),
3411 rax, off, obj);
3412 __ jmp(rewriteNotFlattenable);
3413 __ bind(notBuffered);
3414 }
3415 do_oop_store(_masm, field, rax);
3416 __ bind(rewriteNotFlattenable);
3417 if (rc == may_rewrite) {
3418 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3419 }
3420 __ jmp(Done);
3421 // Implementation of the flattenable semantic
3422 __ bind(isFlattenable);
3423 __ null_check(rax);
3424 __ test_field_is_flattened(flags2, rscratch1, isFlattened);
3425 // Not flattened case
3426 if (ValueTypesBufferMaxMemory > 0) {
3427 __ test_value_is_not_buffered(rax, rscratch1, notBuffered2);
3428 pop_and_check_object(obj);
3429 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_heap_copy),
3430 rax, off, obj);
3431 __ jmp(rewriteFlattenable);
3432 __ bind(notBuffered2);
3433 }
3434 pop_and_check_object(obj);
3435 // Store into the field
3436 do_oop_store(_masm, field, rax);
3437 __ jmp(rewriteFlattenable);
3438 __ bind(isFlattened);
3439 pop_and_check_object(obj);
3440 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_flattened_value),
3441 rax, off, obj);
3442 __ bind(rewriteFlattenable);
3443 if (rc == may_rewrite) {
3444 patch_bytecode(Bytecodes::_fast_qputfield, bc, rbx, true, byte_no);
3445 }
3446 __ jmp(Done);
3447 }
3448 }
3449 }
3450
3451 __ bind(notObj);
3452 __ cmpl(flags, itos);
3453 __ jcc(Assembler::notEqual, notInt);
3454
3455 // itos
3456 {
3457 __ pop(itos);
3458 if (!is_static) pop_and_check_object(obj);
3459 __ movl(field, rax);
3460 if (!is_static && rc == may_rewrite) {
3461 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3462 }
3463 __ jmp(Done);
3464 }
3465
3466 __ bind(notInt);
3467 __ cmpl(flags, ctos);
3468 __ jcc(Assembler::notEqual, notChar);
3469
3470 // ctos
3471 {
3472 __ pop(ctos);
3473 if (!is_static) pop_and_check_object(obj);
3474 __ movw(field, rax);
3475 if (!is_static && rc == may_rewrite) {
3476 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3477 }
3478 __ jmp(Done);
3479 }
3480
3481 __ bind(notChar);
3482 __ cmpl(flags, stos);
3483 __ jcc(Assembler::notEqual, notShort);
3484
3485 // stos
3486 {
3487 __ pop(stos);
3488 if (!is_static) pop_and_check_object(obj);
3489 __ movw(field, rax);
3490 if (!is_static && rc == may_rewrite) {
3491 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3492 }
3493 __ jmp(Done);
3494 }
3495
3496 __ bind(notShort);
3497 __ cmpl(flags, ltos);
3498 __ jcc(Assembler::notEqual, notLong);
3499
3500 // ltos
3501 #ifdef _LP64
3502 {
3503 __ pop(ltos);
3504 if (!is_static) pop_and_check_object(obj);
3505 __ movq(field, rax);
3506 if (!is_static && rc == may_rewrite) {
3507 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3508 }
3509 __ jmp(Done);
3510 }
3511 #else
3512 {
3513 Label notVolatileLong;
3514 __ testl(rdx, rdx);
3515 __ jcc(Assembler::zero, notVolatileLong);
3516
3517 __ pop(ltos); // overwrites rdx, do this after testing volatile.
3518 if (!is_static) pop_and_check_object(obj);
3519
3520 // Replace with real volatile test
3521 __ push(rdx);
3522 __ push(rax); // Must update atomically with FIST
3523 __ fild_d(Address(rsp,0)); // So load into FPU register
3524 __ fistp_d(field); // and put into memory atomically
3525 __ addptr(rsp, 2*wordSize);
3526 // volatile_barrier();
3527 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3528 Assembler::StoreStore));
3529 // Don't rewrite volatile version
3530 __ jmp(notVolatile);
3531
3532 __ bind(notVolatileLong);
3533
3534 __ pop(ltos); // overwrites rdx
3535 if (!is_static) pop_and_check_object(obj);
3536 __ movptr(hi, rdx);
3537 __ movptr(field, rax);
3538 // Don't rewrite to _fast_lputfield for potential volatile case.
3539 __ jmp(notVolatile);
3540 }
3541 #endif // _LP64
3542
3543 __ bind(notLong);
3544 __ cmpl(flags, ftos);
3545 __ jcc(Assembler::notEqual, notFloat);
3546
3547 // ftos
3548 {
3549 __ pop(ftos);
3550 if (!is_static) pop_and_check_object(obj);
3551 __ store_float(field);
3552 if (!is_static && rc == may_rewrite) {
3553 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3554 }
3555 __ jmp(Done);
3556 }
3557
3558 __ bind(notFloat);
3559 #ifdef ASSERT
3560 __ cmpl(flags, dtos);
3561 __ jcc(Assembler::notEqual, notDouble);
3562 #endif
3563
3564 // dtos
3565 {
3566 __ pop(dtos);
3567 if (!is_static) pop_and_check_object(obj);
3568 __ store_double(field);
3569 if (!is_static && rc == may_rewrite) {
3570 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3571 }
3572 }
3573
3574 #ifdef ASSERT
3575 __ jmp(Done);
3576
3577 __ bind(notDouble);
3578 __ stop("Bad state");
3579 #endif
3580
3581 __ bind(Done);
3582
3583 __ shrl(flags2, ConstantPoolCacheEntry::is_volatile_shift);
3584 __ andl(flags2, 0x1);
3585
3586 // Check for volatile store
3587 __ testl(flags2, flags2);
3588 __ jcc(Assembler::zero, notVolatile);
3589 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3590 Assembler::StoreStore));
3591 __ bind(notVolatile);
3592 }
3593
3594 void TemplateTable::putfield(int byte_no) {
3595 putfield_or_static(byte_no, false);
3596 }
3597
3598 void TemplateTable::nofast_putfield(int byte_no) {
3599 putfield_or_static(byte_no, false, may_not_rewrite);
3600 }
3601
3602 void TemplateTable::putstatic(int byte_no) {
3603 putfield_or_static(byte_no, true);
3604 }
3605
3606 void TemplateTable::jvmti_post_fast_field_mod() {
3607
3608 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3609
3610 if (JvmtiExport::can_post_field_modification()) {
3611 // Check to see if a field modification watch has been set before
3612 // we take the time to call into the VM.
3613 Label L2;
3614 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3615 __ testl(scratch, scratch);
3616 __ jcc(Assembler::zero, L2);
3617 __ pop_ptr(rbx); // copy the object pointer from tos
3618 __ verify_oop(rbx);
3619 __ push_ptr(rbx); // put the object pointer back on tos
3620 // Save tos values before call_VM() clobbers them. Since we have
3621 // to do it for every data type, we use the saved values as the
3622 // jvalue object.
3623 switch (bytecode()) { // load values into the jvalue object
3624 case Bytecodes::_fast_qputfield: //fall through
3625 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3626 case Bytecodes::_fast_bputfield: // fall through
3627 case Bytecodes::_fast_zputfield: // fall through
3628 case Bytecodes::_fast_sputfield: // fall through
3629 case Bytecodes::_fast_cputfield: // fall through
3630 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3631 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3632 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3633 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3634
3635 default:
3636 ShouldNotReachHere();
3637 }
3638 __ mov(scratch, rsp); // points to jvalue on the stack
3639 // access constant pool cache entry
3640 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3641 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3642 __ verify_oop(rbx);
3643 // rbx: object pointer copied above
3644 // c_rarg2: cache entry pointer
3645 // c_rarg3: jvalue object on the stack
3646 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3647 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3648
3649 switch (bytecode()) { // restore tos values
3650 case Bytecodes::_fast_qputfield: // fall through
3651 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3652 case Bytecodes::_fast_bputfield: // fall through
3653 case Bytecodes::_fast_zputfield: // fall through
3654 case Bytecodes::_fast_sputfield: // fall through
3655 case Bytecodes::_fast_cputfield: // fall through
3656 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3657 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3658 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3659 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3660 default: break;
3661 }
3662 __ bind(L2);
3663 }
3664 }
3665
3666 void TemplateTable::fast_storefield(TosState state) {
3667 transition(state, vtos);
3668
3669 ByteSize base = ConstantPoolCache::base_offset();
3670
3671 jvmti_post_fast_field_mod();
3672
3673 // access constant pool cache
3674 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3675
3676 // test for volatile with rdx but rdx is tos register for lputfield.
3677 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3678 in_bytes(base +
3679 ConstantPoolCacheEntry::flags_offset())));
3680
3681 // replace index with field offset from cache entry
3682 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3683 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3684
3685 // [jk] not needed currently
3686 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3687 // Assembler::StoreStore));
3688
3689 if (bytecode() == Bytecodes::_fast_qputfield) {
3690 __ movl(rscratch2, rdx);
3691 }
3692
3693 Label notVolatile;
3694 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3695 __ andl(rdx, 0x1);
3696
3697 // Get object from stack
3698 pop_and_check_object(rcx);
3699
3700 // field address
3701 const Address field(rcx, rbx, Address::times_1);
3702
3703 // access field
3704 switch (bytecode()) {
3705 case Bytecodes::_fast_qputfield:
3706 {
3707 Label isFlattened, notBuffered, done;
3708 __ null_check(rax);
3709 __ test_field_is_flattened(rscratch2, rscratch1, isFlattened);
3710 // No Flattened case
3711 if (ValueTypesBufferMaxMemory > 0) {
3712 __ test_value_is_not_buffered(rax, rscratch1, notBuffered);
3713 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_heap_copy),
3714 rax, rbx, rcx);
3715 __ jmp(done);
3716 __ bind(notBuffered);
3717 }
3718 do_oop_store(_masm, field, rax);
3719 __ jmp(done);
3720 __ bind(isFlattened);
3721 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_flattened_value),
3722 rax, rbx, rcx);
3723 __ bind(done);
3724 }
3725 break;
3726 case Bytecodes::_fast_aputfield:
3727 {
3728 Label notBuffered, done;
3729 if (ValueTypesBufferMaxMemory > 0) {
3730 __ test_value_is_not_buffered(rax, rscratch1, notBuffered);
3731 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::write_heap_copy),
3732 rax, rbx, rcx);
3733 __ jmp(done);
3734 __ bind(notBuffered);
3735 }
3736 do_oop_store(_masm, field, rax);
3737 __ bind(done);
3738 }
3739 break;
3740 case Bytecodes::_fast_lputfield:
3741 #ifdef _LP64
3742 __ movq(field, rax);
3743 #else
3744 __ stop("should not be rewritten");
3745 #endif
3746 break;
3747 case Bytecodes::_fast_iputfield:
3748 __ movl(field, rax);
3749 break;
3750 case Bytecodes::_fast_zputfield:
3751 __ andl(rax, 0x1); // boolean is true if LSB is 1
3752 // fall through to bputfield
3753 case Bytecodes::_fast_bputfield:
3754 __ movb(field, rax);
3755 break;
3756 case Bytecodes::_fast_sputfield:
3757 // fall through
3758 case Bytecodes::_fast_cputfield:
3759 __ movw(field, rax);
3760 break;
3761 case Bytecodes::_fast_fputfield:
3762 __ store_float(field);
3763 break;
3764 case Bytecodes::_fast_dputfield:
3765 __ store_double(field);
3766 break;
3767 default:
3768 ShouldNotReachHere();
3769 }
3770
3771 // Check for volatile store
3772 __ testl(rdx, rdx);
3773 __ jcc(Assembler::zero, notVolatile);
3774 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3775 Assembler::StoreStore));
3776 __ bind(notVolatile);
3777 }
3778
3779 void TemplateTable::fast_accessfield(TosState state) {
3780 transition(atos, state);
3781
3782 // Do the JVMTI work here to avoid disturbing the register state below
3783 if (JvmtiExport::can_post_field_access()) {
3784 // Check to see if a field access watch has been set before we
3785 // take the time to call into the VM.
3786 Label L1;
3787 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3788 __ testl(rcx, rcx);
3789 __ jcc(Assembler::zero, L1);
3790 // access constant pool cache entry
3791 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3792 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3793 __ verify_oop(rax);
3794 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3795 LP64_ONLY(__ mov(c_rarg1, rax));
3796 // c_rarg1: object pointer copied above
3797 // c_rarg2: cache entry pointer
3798 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3799 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3800 __ pop_ptr(rax); // restore object pointer
3801 __ bind(L1);
3802 }
3803
3804 // access constant pool cache
3805 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3806 // replace index with field offset from cache entry
3807 // [jk] not needed currently
3808 // if (os::is_MP()) {
3809 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3810 // in_bytes(ConstantPoolCache::base_offset() +
3811 // ConstantPoolCacheEntry::flags_offset())));
3812 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3813 // __ andl(rdx, 0x1);
3814 // }
3815 __ movptr(rdx, Address(rcx, rbx, Address::times_ptr,
3816 in_bytes(ConstantPoolCache::base_offset() +
3817 ConstantPoolCacheEntry::f2_offset())));
3818
3819 // rax: object
3820 __ verify_oop(rax);
3821 __ null_check(rax);
3822 Address field(rax, rdx, Address::times_1);
3823
3824 // access field
3825 switch (bytecode()) {
3826 case Bytecodes::_fast_qgetfield:
3827 {
3828 Label isFlattened, nonnull, Done;
3829 __ movptr(rscratch1, Address(rcx, rbx, Address::times_ptr,
3830 in_bytes(ConstantPoolCache::base_offset() +
3831 ConstantPoolCacheEntry::flags_offset())));
3832 __ test_field_is_flattened(rscratch1, rscratch2, isFlattened);
3833 // Non-flattened field case
3834 __ movptr(rscratch1, rax);
3835 __ load_heap_oop(rax, field);
3836 __ testptr(rax, rax);
3837 __ jcc(Assembler::notZero, nonnull);
3838 __ movptr(rax, rscratch1);
3839 __ movl(rcx, Address(rcx, rbx, Address::times_ptr,
3840 in_bytes(ConstantPoolCache::base_offset() +
3841 ConstantPoolCacheEntry::flags_offset())));
3842 __ andl(rcx, ConstantPoolCacheEntry::field_index_mask);
3843 __ call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::uninitialized_instance_value_field),
3844 rax, rcx);
3845 __ bind(nonnull);
3846 __ verify_oop(rax);
3847 __ jmp(Done);
3848 __ bind(isFlattened);
3849 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3850 in_bytes(ConstantPoolCache::base_offset() +
3851 ConstantPoolCacheEntry::flags_offset())));
3852 __ andl(rdx, ConstantPoolCacheEntry::field_index_mask);
3853 __ movptr(rcx, Address(rcx, rbx, Address::times_ptr,
3854 in_bytes(ConstantPoolCache::base_offset() +
3855 ConstantPoolCacheEntry::f1_offset())));
3856 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::read_flattened_field),
3857 rax, rdx, rcx);
3858 __ verify_oop(rax);
3859 __ bind(Done);
3860 }
3861 break;
3862 case Bytecodes::_fast_agetfield:
3863 do_oop_load(_masm, field, rax);
3864 __ verify_oop(rax);
3865 break;
3866 case Bytecodes::_fast_lgetfield:
3867 #ifdef _LP64
3868 __ movq(rax, field);
3869 #else
3870 __ stop("should not be rewritten");
3871 #endif
3872 break;
3873 case Bytecodes::_fast_igetfield:
3874 __ movl(rax, field);
3875 break;
3876 case Bytecodes::_fast_bgetfield:
3877 __ movsbl(rax, field);
3878 break;
3879 case Bytecodes::_fast_sgetfield:
3880 __ load_signed_short(rax, field);
3881 break;
3882 case Bytecodes::_fast_cgetfield:
3883 __ load_unsigned_short(rax, field);
3884 break;
3885 case Bytecodes::_fast_fgetfield:
3886 __ load_float(field);
3887 break;
3888 case Bytecodes::_fast_dgetfield:
3889 __ load_double(field);
3890 break;
3891 default:
3892 ShouldNotReachHere();
3893 }
3894 // [jk] not needed currently
3895 // if (os::is_MP()) {
3896 // Label notVolatile;
3897 // __ testl(rdx, rdx);
3898 // __ jcc(Assembler::zero, notVolatile);
3899 // __ membar(Assembler::LoadLoad);
3900 // __ bind(notVolatile);
3901 //};
3902 }
3903
3904 void TemplateTable::fast_xaccess(TosState state) {
3905 transition(vtos, state);
3906
3907 // get receiver
3908 __ movptr(rax, aaddress(0));
3909 // access constant pool cache
3910 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3911 __ movptr(rbx,
3912 Address(rcx, rdx, Address::times_ptr,
3913 in_bytes(ConstantPoolCache::base_offset() +
3914 ConstantPoolCacheEntry::f2_offset())));
3915 // make sure exception is reported in correct bcp range (getfield is
3916 // next instruction)
3917 __ increment(rbcp);
3918 __ null_check(rax);
3919 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3920 switch (state) {
3921 case itos:
3922 __ movl(rax, field);
3923 break;
3924 case atos:
3925 do_oop_load(_masm, field, rax);
3926 __ verify_oop(rax);
3927 break;
3928 case ftos:
3929 __ load_float(field);
3930 break;
3931 default:
3932 ShouldNotReachHere();
3933 }
3934
3935 // [jk] not needed currently
3936 // if (os::is_MP()) {
3937 // Label notVolatile;
3938 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3939 // in_bytes(ConstantPoolCache::base_offset() +
3940 // ConstantPoolCacheEntry::flags_offset())));
3941 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3942 // __ testl(rdx, 0x1);
3943 // __ jcc(Assembler::zero, notVolatile);
3944 // __ membar(Assembler::LoadLoad);
3945 // __ bind(notVolatile);
3946 // }
3947
3948 __ decrement(rbcp);
3949 }
3950
3951 //-----------------------------------------------------------------------------
3952 // Calls
3953
3954 void TemplateTable::count_calls(Register method, Register temp) {
3955 // implemented elsewhere
3956 ShouldNotReachHere();
3957 }
3958
3959 void TemplateTable::prepare_invoke(int byte_no,
3960 Register method, // linked method (or i-klass)
3961 Register index, // itable index, MethodType, etc.
3962 Register recv, // if caller wants to see it
3963 Register flags // if caller wants to test it
3964 ) {
3965 // determine flags
3966 const Bytecodes::Code code = bytecode();
3967 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3968 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3969 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3970 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3971 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3972 const bool load_receiver = (recv != noreg);
3973 const bool save_flags = (flags != noreg);
3974 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3975 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3976 assert(flags == noreg || flags == rdx, "");
3977 assert(recv == noreg || recv == rcx, "");
3978
3979 // setup registers & access constant pool cache
3980 if (recv == noreg) recv = rcx;
3981 if (flags == noreg) flags = rdx;
3982 assert_different_registers(method, index, recv, flags);
3983
3984 // save 'interpreter return address'
3985 __ save_bcp();
3986
3987 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3988
3989 // maybe push appendix to arguments (just before return address)
3990 if (is_invokedynamic || is_invokehandle) {
3991 Label L_no_push;
3992 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3993 __ jcc(Assembler::zero, L_no_push);
3994 // Push the appendix as a trailing parameter.
3995 // This must be done before we get the receiver,
3996 // since the parameter_size includes it.
3997 __ push(rbx);
3998 __ mov(rbx, index);
3999 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
4000 __ load_resolved_reference_at_index(index, rbx);
4001 __ pop(rbx);
4002 __ push(index); // push appendix (MethodType, CallSite, etc.)
4003 __ bind(L_no_push);
4004 }
4005
4006 // load receiver if needed (after appendix is pushed so parameter size is correct)
4007 // Note: no return address pushed yet
4008 if (load_receiver) {
4009 __ movl(recv, flags);
4010 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
4011 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
4012 const int receiver_is_at_end = -1; // back off one slot to get receiver
4013 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
4014 __ movptr(recv, recv_addr);
4015 __ verify_oop(recv);
4016 }
4017
4018 if (save_flags) {
4019 __ movl(rbcp, flags);
4020 }
4021
4022 // compute return type
4023 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
4024 // Make sure we don't need to mask flags after the above shift
4025 ConstantPoolCacheEntry::verify_tos_state_shift();
4026 // load return address
4027 {
4028 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
4029 ExternalAddress table(table_addr);
4030 LP64_ONLY(__ lea(rscratch1, table));
4031 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
4032 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
4033 }
4034
4035 // push return address
4036 __ push(flags);
4037
4038 // Restore flags value from the constant pool cache, and restore rsi
4039 // for later null checks. r13 is the bytecode pointer
4040 if (save_flags) {
4041 __ movl(flags, rbcp);
4042 __ restore_bcp();
4043 }
4044 }
4045
4046 void TemplateTable::invokevirtual_helper(Register index,
4047 Register recv,
4048 Register flags) {
4049 // Uses temporary registers rax, rdx
4050 assert_different_registers(index, recv, rax, rdx);
4051 assert(index == rbx, "");
4052 assert(recv == rcx, "");
4053
4054 // Test for an invoke of a final method
4055 Label notFinal;
4056 __ movl(rax, flags);
4057 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
4058 __ jcc(Assembler::zero, notFinal);
4059
4060 const Register method = index; // method must be rbx
4061 assert(method == rbx,
4062 "Method* must be rbx for interpreter calling convention");
4063
4064 // do the call - the index is actually the method to call
4065 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
4066
4067 // It's final, need a null check here!
4068 __ null_check(recv);
4069
4070 // profile this call
4071 __ profile_final_call(rax);
4072 __ profile_arguments_type(rax, method, rbcp, true);
4073
4074 __ jump_from_interpreted(method, rax);
4075
4076 __ bind(notFinal);
4077
4078 // get receiver klass
4079 __ null_check(recv, oopDesc::klass_offset_in_bytes());
4080 __ load_klass(rax, recv);
4081
4082 // profile this call
4083 __ profile_virtual_call(rax, rlocals, rdx);
4084 // get target Method* & entry point
4085 __ lookup_virtual_method(rax, index, method);
4086 __ profile_called_method(method, rdx, rbcp);
4087
4088 __ profile_arguments_type(rdx, method, rbcp, true);
4089 __ jump_from_interpreted(method, rdx);
4090 }
4091
4092 void TemplateTable::invokevirtual(int byte_no) {
4093 transition(vtos, vtos);
4094 assert(byte_no == f2_byte, "use this argument");
4095 prepare_invoke(byte_no,
4096 rbx, // method or vtable index
4097 noreg, // unused itable index
4098 rcx, rdx); // recv, flags
4099
4100 // rbx: index
4101 // rcx: receiver
4102 // rdx: flags
4103
4104 invokevirtual_helper(rbx, rcx, rdx);
4105 }
4106
4107 void TemplateTable::invokespecial(int byte_no) {
4108 transition(vtos, vtos);
4109 assert(byte_no == f1_byte, "use this argument");
4110 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
4111 rcx); // get receiver also for null check
4112 __ verify_oop(rcx);
4113 __ null_check(rcx);
4114 // do the call
4115 __ profile_call(rax);
4116 __ profile_arguments_type(rax, rbx, rbcp, false);
4117 __ jump_from_interpreted(rbx, rax);
4118 }
4119
4120 void TemplateTable::invokestatic(int byte_no) {
4121 transition(vtos, vtos);
4122 assert(byte_no == f1_byte, "use this argument");
4123 prepare_invoke(byte_no, rbx); // get f1 Method*
4124 // do the call
4125 __ profile_call(rax);
4126 __ profile_arguments_type(rax, rbx, rbcp, false);
4127 __ jump_from_interpreted(rbx, rax);
4128 }
4129
4130
4131 void TemplateTable::fast_invokevfinal(int byte_no) {
4132 transition(vtos, vtos);
4133 assert(byte_no == f2_byte, "use this argument");
4134 __ stop("fast_invokevfinal not used on x86");
4135 }
4136
4137
4138 void TemplateTable::invokeinterface(int byte_no) {
4139 transition(vtos, vtos);
4140 assert(byte_no == f1_byte, "use this argument");
4141 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 Method*
4142 rcx, rdx); // recv, flags
4143
4144 // rax: reference klass (from f1)
4145 // rbx: method (from f2)
4146 // rcx: receiver
4147 // rdx: flags
4148
4149 // Special case of invokeinterface called for virtual method of
4150 // java.lang.Object. See cpCacheOop.cpp for details.
4151 // This code isn't produced by javac, but could be produced by
4152 // another compliant java compiler.
4153 Label notMethod;
4154 __ movl(rlocals, rdx);
4155 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
4156
4157 __ jcc(Assembler::zero, notMethod);
4158
4159 invokevirtual_helper(rbx, rcx, rdx);
4160 __ bind(notMethod);
4161
4162 // Get receiver klass into rdx - also a null check
4163 __ restore_locals(); // restore r14
4164 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
4165 __ load_klass(rdx, rcx);
4166
4167 Label no_such_interface, no_such_method;
4168
4169 // Preserve method for throw_AbstractMethodErrorVerbose.
4170 __ mov(rcx, rbx);
4171 // Receiver subtype check against REFC.
4172 // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
4173 __ lookup_interface_method(// inputs: rec. class, interface, itable index
4174 rdx, rax, noreg,
4175 // outputs: scan temp. reg, scan temp. reg
4176 rbcp, rlocals,
4177 no_such_interface,
4178 /*return_method=*/false);
4179
4180 // profile this call
4181 __ restore_bcp(); // rbcp was destroyed by receiver type check
4182 __ profile_virtual_call(rdx, rbcp, rlocals);
4183
4184 // Get declaring interface class from method, and itable index
4185 __ movptr(rax, Address(rbx, Method::const_offset()));
4186 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
4187 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
4188 __ movl(rbx, Address(rbx, Method::itable_index_offset()));
4189 __ subl(rbx, Method::itable_index_max);
4190 __ negl(rbx);
4191
4192 // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
4193 __ mov(rlocals, rdx);
4194 __ lookup_interface_method(// inputs: rec. class, interface, itable index
4195 rlocals, rax, rbx,
4196 // outputs: method, scan temp. reg
4197 rbx, rbcp,
4198 no_such_interface);
4199
4200 // rbx: Method* to call
4201 // rcx: receiver
4202 // Check for abstract method error
4203 // Note: This should be done more efficiently via a throw_abstract_method_error
4204 // interpreter entry point and a conditional jump to it in case of a null
4205 // method.
4206 __ testptr(rbx, rbx);
4207 __ jcc(Assembler::zero, no_such_method);
4208
4209 __ profile_called_method(rbx, rbcp, rdx);
4210 __ profile_arguments_type(rdx, rbx, rbcp, true);
4211
4212 // do the call
4213 // rcx: receiver
4214 // rbx,: Method*
4215 __ jump_from_interpreted(rbx, rdx);
4216 __ should_not_reach_here();
4217
4218 // exception handling code follows...
4219 // note: must restore interpreter registers to canonical
4220 // state for exception handling to work correctly!
4221
4222 __ bind(no_such_method);
4223 // throw exception
4224 __ pop(rbx); // pop return address (pushed by prepare_invoke)
4225 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
4226 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
4227 // Pass arguments for generating a verbose error message.
4228 #ifdef _LP64
4229 Register recvKlass = c_rarg1;
4230 Register method = c_rarg2;
4231 if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
4232 if (method != rcx) { __ movq(method, rcx); }
4233 #else
4234 Register recvKlass = rdx;
4235 Register method = rcx;
4236 #endif
4237 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
4238 recvKlass, method);
4239 // The call_VM checks for exception, so we should never return here.
4240 __ should_not_reach_here();
4241
4242 __ bind(no_such_interface);
4243 // throw exception
4244 __ pop(rbx); // pop return address (pushed by prepare_invoke)
4245 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
4246 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
4247 // Pass arguments for generating a verbose error message.
4248 LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
4249 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
4250 recvKlass, rax);
4251 // the call_VM checks for exception, so we should never return here.
4252 __ should_not_reach_here();
4253 }
4254
4255 void TemplateTable::invokehandle(int byte_no) {
4256 transition(vtos, vtos);
4257 assert(byte_no == f1_byte, "use this argument");
4258 const Register rbx_method = rbx;
4259 const Register rax_mtype = rax;
4260 const Register rcx_recv = rcx;
4261 const Register rdx_flags = rdx;
4262
4263 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
4264 __ verify_method_ptr(rbx_method);
4265 __ verify_oop(rcx_recv);
4266 __ null_check(rcx_recv);
4267
4268 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
4269 // rbx: MH.invokeExact_MT method (from f2)
4270
4271 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
4272
4273 // FIXME: profile the LambdaForm also
4274 __ profile_final_call(rax);
4275 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
4276
4277 __ jump_from_interpreted(rbx_method, rdx);
4278 }
4279
4280 void TemplateTable::invokedynamic(int byte_no) {
4281 transition(vtos, vtos);
4282 assert(byte_no == f1_byte, "use this argument");
4283
4284 const Register rbx_method = rbx;
4285 const Register rax_callsite = rax;
4286
4287 prepare_invoke(byte_no, rbx_method, rax_callsite);
4288
4289 // rax: CallSite object (from cpool->resolved_references[f1])
4290 // rbx: MH.linkToCallSite method (from f2)
4291
4292 // Note: rax_callsite is already pushed by prepare_invoke
4293
4294 // %%% should make a type profile for any invokedynamic that takes a ref argument
4295 // profile this call
4296 __ profile_call(rbcp);
4297 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
4298
4299 __ verify_oop(rax_callsite);
4300
4301 __ jump_from_interpreted(rbx_method, rdx);
4302 }
4303
4304 //-----------------------------------------------------------------------------
4305 // Allocation
4306
4307 void TemplateTable::_new() {
4308 transition(vtos, atos);
4309 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
4310 Label slow_case;
4311 Label slow_case_no_pop;
4312 Label done;
4313 Label initialize_header;
4314 Label initialize_object; // including clearing the fields
4315
4316 __ get_cpool_and_tags(rcx, rax);
4317
4318 // Make sure the class we're about to instantiate has been resolved.
4319 // This is done before loading InstanceKlass to be consistent with the order
4320 // how Constant Pool is updated (see ConstantPool::klass_at_put)
4321 const int tags_offset = Array<u1>::base_offset_in_bytes();
4322 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
4323 __ jcc(Assembler::notEqual, slow_case_no_pop);
4324
4325 // get InstanceKlass
4326 __ load_resolved_klass_at_index(rcx, rdx, rcx);
4327 __ push(rcx); // save the contexts of klass for initializing the header
4328
4329 // make sure klass is initialized & doesn't have finalizer
4330 // make sure klass is fully initialized
4331 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4332 __ jcc(Assembler::notEqual, slow_case);
4333
4334 // get instance_size in InstanceKlass (scaled to a count of bytes)
4335 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
4336 // test to see if it has a finalizer or is malformed in some way
4337 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
4338 __ jcc(Assembler::notZero, slow_case);
4339
4340 // Allocate the instance:
4341 // If TLAB is enabled:
4342 // Try to allocate in the TLAB.
4343 // If fails, go to the slow path.
4344 // Else If inline contiguous allocations are enabled:
4345 // Try to allocate in eden.
4346 // If fails due to heap end, go to slow path.
4347 //
4348 // If TLAB is enabled OR inline contiguous is enabled:
4349 // Initialize the allocation.
4350 // Exit.
4351 //
4352 // Go to slow path.
4353
4354 const bool allow_shared_alloc =
4355 Universe::heap()->supports_inline_contig_alloc();
4356
4357 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4358 #ifndef _LP64
4359 if (UseTLAB || allow_shared_alloc) {
4360 __ get_thread(thread);
4361 }
4362 #endif // _LP64
4363
4364 if (UseTLAB) {
4365 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
4366 __ lea(rbx, Address(rax, rdx, Address::times_1));
4367 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
4368 __ jcc(Assembler::above, slow_case);
4369 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
4370 if (ZeroTLAB) {
4371 // the fields have been already cleared
4372 __ jmp(initialize_header);
4373 } else {
4374 // initialize both the header and fields
4375 __ jmp(initialize_object);
4376 }
4377 } else {
4378 // Allocation in the shared Eden, if allowed.
4379 //
4380 // rdx: instance size in bytes
4381 if (allow_shared_alloc) {
4382 ExternalAddress heap_top((address)Universe::heap()->top_addr());
4383 ExternalAddress heap_end((address)Universe::heap()->end_addr());
4384
4385 Label retry;
4386 __ bind(retry);
4387 __ movptr(rax, heap_top);
4388 __ lea(rbx, Address(rax, rdx, Address::times_1));
4389 __ cmpptr(rbx, heap_end);
4390 __ jcc(Assembler::above, slow_case);
4391
4392 // Compare rax, with the top addr, and if still equal, store the new
4393 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
4394 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
4395 //
4396 // rax,: object begin
4397 // rbx,: object end
4398 // rdx: instance size in bytes
4399 __ locked_cmpxchgptr(rbx, heap_top);
4400
4401 // if someone beat us on the allocation, try again, otherwise continue
4402 __ jcc(Assembler::notEqual, retry);
4403
4404 __ incr_allocated_bytes(thread, rdx, 0);
4405 }
4406 }
4407
4408 // If UseTLAB or allow_shared_alloc are true, the object is created above and
4409 // there is an initialize need. Otherwise, skip and go to the slow path.
4410 if (UseTLAB || allow_shared_alloc) {
4411 // The object is initialized before the header. If the object size is
4412 // zero, go directly to the header initialization.
4413 __ bind(initialize_object);
4414 __ decrement(rdx, sizeof(oopDesc));
4415 __ jcc(Assembler::zero, initialize_header);
4416
4417 // Initialize topmost object field, divide rdx by 8, check if odd and
4418 // test if zero.
4419 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
4420 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4421
4422 // rdx must have been multiple of 8
4423 #ifdef ASSERT
4424 // make sure rdx was multiple of 8
4425 Label L;
4426 // Ignore partial flag stall after shrl() since it is debug VM
4427 __ jccb(Assembler::carryClear, L);
4428 __ stop("object size is not multiple of 2 - adjust this code");
4429 __ bind(L);
4430 // rdx must be > 0, no extra check needed here
4431 #endif
4432
4433 // initialize remaining object fields: rdx was a multiple of 8
4434 { Label loop;
4435 __ bind(loop);
4436 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4437 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4438 __ decrement(rdx);
4439 __ jcc(Assembler::notZero, loop);
4440 }
4441
4442 // initialize object header only.
4443 __ bind(initialize_header);
4444 if (UseBiasedLocking) {
4445 __ pop(rcx); // get saved klass back in the register.
4446 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4447 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4448 } else {
4449 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4450 (intptr_t)markOopDesc::prototype()); // header
4451 __ pop(rcx); // get saved klass back in the register.
4452 }
4453 #ifdef _LP64
4454 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4455 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
4456 #endif
4457 __ store_klass(rax, rcx); // klass
4458
4459 {
4460 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4461 // Trigger dtrace event for fastpath
4462 __ push(atos);
4463 __ call_VM_leaf(
4464 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4465 __ pop(atos);
4466 }
4467
4468 __ jmp(done);
4469 }
4470
4471 // slow case
4472 __ bind(slow_case);
4473 __ pop(rcx); // restore stack pointer to what it was when we came in.
4474 __ bind(slow_case_no_pop);
4475
4476 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4477 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4478
4479 __ get_constant_pool(rarg1);
4480 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4481 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4482 __ verify_oop(rax);
4483
4484 // continue
4485 __ bind(done);
4486 }
4487
4488 void TemplateTable::defaultvalue() {
4489 transition(vtos, atos);
4490
4491 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4492 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4493
4494 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4495 __ get_constant_pool(rarg1);
4496
4497 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::defaultvalue),
4498 rarg1, rarg2);
4499 __ verify_oop(rax);
4500 }
4501
4502 void TemplateTable::newarray() {
4503 transition(itos, atos);
4504 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4505 __ load_unsigned_byte(rarg1, at_bcp(1));
4506 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4507 rarg1, rax);
4508 }
4509
4510 void TemplateTable::anewarray() {
4511 transition(itos, atos);
4512
4513 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4514 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4515
4516 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4517 __ get_constant_pool(rarg1);
4518 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4519 rarg1, rarg2, rax);
4520 }
4521
4522 void TemplateTable::arraylength() {
4523 transition(atos, itos);
4524 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4525 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4526 }
4527
4528 void TemplateTable::checkcast() {
4529 transition(atos, atos);
4530 Label done, is_null, ok_is_subtype, quicked, resolved;
4531 __ testptr(rax, rax); // object is in rax
4532 __ jcc(Assembler::zero, is_null);
4533
4534 // Get cpool & tags index
4535 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4536 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4537 // See if bytecode has already been quicked
4538 __ cmpb(Address(rdx, rbx,
4539 Address::times_1,
4540 Array<u1>::base_offset_in_bytes()),
4541 JVM_CONSTANT_Class);
4542 __ jcc(Assembler::equal, quicked);
4543 __ push(atos); // save receiver for result, and for GC
4544 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4545
4546 // vm_result_2 has metadata result
4547 #ifndef _LP64
4548 // borrow rdi from locals
4549 __ get_thread(rdi);
4550 __ get_vm_result_2(rax, rdi);
4551 __ restore_locals();
4552 #else
4553 __ get_vm_result_2(rax, r15_thread);
4554 #endif
4555
4556 __ pop_ptr(rdx); // restore receiver
4557 __ jmpb(resolved);
4558
4559 // Get superklass in rax and subklass in rbx
4560 __ bind(quicked);
4561 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4562 __ load_resolved_klass_at_index(rcx, rbx, rax);
4563
4564 __ bind(resolved);
4565 __ load_klass(rbx, rdx);
4566
4567 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4568 // Superklass in rax. Subklass in rbx.
4569 __ gen_subtype_check(rbx, ok_is_subtype);
4570
4571 // Come here on failure
4572 __ push_ptr(rdx);
4573 // object is at TOS
4574 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4575
4576 // Come here on success
4577 __ bind(ok_is_subtype);
4578 __ mov(rax, rdx); // Restore object in rdx
4579
4580 // Collect counts on whether this check-cast sees NULLs a lot or not.
4581 if (ProfileInterpreter) {
4582 __ jmp(done);
4583 __ bind(is_null);
4584 __ profile_null_seen(rcx);
4585 } else {
4586 __ bind(is_null); // same as 'done'
4587 }
4588 __ bind(done);
4589 }
4590
4591 void TemplateTable::instanceof() {
4592 transition(atos, itos);
4593 Label done, is_null, ok_is_subtype, quicked, resolved;
4594 __ testptr(rax, rax);
4595 __ jcc(Assembler::zero, is_null);
4596
4597 // Get cpool & tags index
4598 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4599 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4600 // See if bytecode has already been quicked
4601 __ cmpb(Address(rdx, rbx,
4602 Address::times_1,
4603 Array<u1>::base_offset_in_bytes()),
4604 JVM_CONSTANT_Class);
4605 __ jcc(Assembler::equal, quicked);
4606
4607 __ push(atos); // save receiver for result, and for GC
4608 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4609 // vm_result_2 has metadata result
4610
4611 #ifndef _LP64
4612 // borrow rdi from locals
4613 __ get_thread(rdi);
4614 __ get_vm_result_2(rax, rdi);
4615 __ restore_locals();
4616 #else
4617 __ get_vm_result_2(rax, r15_thread);
4618 #endif
4619
4620 __ pop_ptr(rdx); // restore receiver
4621 __ verify_oop(rdx);
4622 __ load_klass(rdx, rdx);
4623 __ jmpb(resolved);
4624
4625 // Get superklass in rax and subklass in rdx
4626 __ bind(quicked);
4627 __ load_klass(rdx, rax);
4628 __ load_resolved_klass_at_index(rcx, rbx, rax);
4629
4630 __ bind(resolved);
4631
4632 // Generate subtype check. Blows rcx, rdi
4633 // Superklass in rax. Subklass in rdx.
4634 __ gen_subtype_check(rdx, ok_is_subtype);
4635
4636 // Come here on failure
4637 __ xorl(rax, rax);
4638 __ jmpb(done);
4639 // Come here on success
4640 __ bind(ok_is_subtype);
4641 __ movl(rax, 1);
4642
4643 // Collect counts on whether this test sees NULLs a lot or not.
4644 if (ProfileInterpreter) {
4645 __ jmp(done);
4646 __ bind(is_null);
4647 __ profile_null_seen(rcx);
4648 } else {
4649 __ bind(is_null); // same as 'done'
4650 }
4651 __ bind(done);
4652 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4653 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4654 }
4655
4656 //----------------------------------------------------------------------------------------------------
4657 // Breakpoints
4658 void TemplateTable::_breakpoint() {
4659 // Note: We get here even if we are single stepping..
4660 // jbug insists on setting breakpoints at every bytecode
4661 // even if we are in single step mode.
4662
4663 transition(vtos, vtos);
4664
4665 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4666
4667 // get the unpatched byte code
4668 __ get_method(rarg);
4669 __ call_VM(noreg,
4670 CAST_FROM_FN_PTR(address,
4671 InterpreterRuntime::get_original_bytecode_at),
4672 rarg, rbcp);
4673 __ mov(rbx, rax); // why?
4674
4675 // post the breakpoint event
4676 __ get_method(rarg);
4677 __ call_VM(noreg,
4678 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4679 rarg, rbcp);
4680
4681 // complete the execution of original bytecode
4682 __ dispatch_only_normal(vtos);
4683 }
4684
4685 //-----------------------------------------------------------------------------
4686 // Exceptions
4687
4688 void TemplateTable::athrow() {
4689 transition(atos, vtos);
4690 __ null_check(rax);
4691 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4692 }
4693
4694 //-----------------------------------------------------------------------------
4695 // Synchronization
4696 //
4697 // Note: monitorenter & exit are symmetric routines; which is reflected
4698 // in the assembly code structure as well
4699 //
4700 // Stack layout:
4701 //
4702 // [expressions ] <--- rsp = expression stack top
4703 // ..
4704 // [expressions ]
4705 // [monitor entry] <--- monitor block top = expression stack bot
4706 // ..
4707 // [monitor entry]
4708 // [frame data ] <--- monitor block bot
4709 // ...
4710 // [saved rbp ] <--- rbp
4711 void TemplateTable::monitorenter() {
4712 transition(atos, vtos);
4713
4714 // check for NULL object
4715 __ null_check(rax);
4716
4717 const Address monitor_block_top(
4718 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4719 const Address monitor_block_bot(
4720 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4721 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4722
4723 Label allocated;
4724
4725 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4726 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4727 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4728
4729 // initialize entry pointer
4730 __ xorl(rmon, rmon); // points to free slot or NULL
4731
4732 // find a free slot in the monitor block (result in rmon)
4733 {
4734 Label entry, loop, exit;
4735 __ movptr(rtop, monitor_block_top); // points to current entry,
4736 // starting with top-most entry
4737 __ lea(rbot, monitor_block_bot); // points to word before bottom
4738 // of monitor block
4739 __ jmpb(entry);
4740
4741 __ bind(loop);
4742 // check if current entry is used
4743 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4744 // if not used then remember entry in rmon
4745 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4746 // check if current entry is for same object
4747 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4748 // if same object then stop searching
4749 __ jccb(Assembler::equal, exit);
4750 // otherwise advance to next entry
4751 __ addptr(rtop, entry_size);
4752 __ bind(entry);
4753 // check if bottom reached
4754 __ cmpptr(rtop, rbot);
4755 // if not at bottom then check this entry
4756 __ jcc(Assembler::notEqual, loop);
4757 __ bind(exit);
4758 }
4759
4760 __ testptr(rmon, rmon); // check if a slot has been found
4761 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4762
4763 // allocate one if there's no free slot
4764 {
4765 Label entry, loop;
4766 // 1. compute new pointers // rsp: old expression stack top
4767 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4768 __ subptr(rsp, entry_size); // move expression stack top
4769 __ subptr(rmon, entry_size); // move expression stack bottom
4770 __ mov(rtop, rsp); // set start value for copy loop
4771 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4772 __ jmp(entry);
4773 // 2. move expression stack contents
4774 __ bind(loop);
4775 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4776 // word from old location
4777 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4778 __ addptr(rtop, wordSize); // advance to next word
4779 __ bind(entry);
4780 __ cmpptr(rtop, rmon); // check if bottom reached
4781 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4782 // copy next word
4783 }
4784
4785 // call run-time routine
4786 // rmon: points to monitor entry
4787 __ bind(allocated);
4788
4789 // Increment bcp to point to the next bytecode, so exception
4790 // handling for async. exceptions work correctly.
4791 // The object has already been poped from the stack, so the
4792 // expression stack looks correct.
4793 __ increment(rbcp);
4794
4795 // store object
4796 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4797 __ lock_object(rmon);
4798
4799 // check to make sure this monitor doesn't cause stack overflow after locking
4800 __ save_bcp(); // in case of exception
4801 __ generate_stack_overflow_check(0);
4802
4803 // The bcp has already been incremented. Just need to dispatch to
4804 // next instruction.
4805 __ dispatch_next(vtos);
4806 }
4807
4808 void TemplateTable::monitorexit() {
4809 transition(atos, vtos);
4810
4811 // check for NULL object
4812 __ null_check(rax);
4813
4814 const Address monitor_block_top(
4815 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4816 const Address monitor_block_bot(
4817 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4818 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4819
4820 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4821 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4822
4823 Label found;
4824
4825 // find matching slot
4826 {
4827 Label entry, loop;
4828 __ movptr(rtop, monitor_block_top); // points to current entry,
4829 // starting with top-most entry
4830 __ lea(rbot, monitor_block_bot); // points to word before bottom
4831 // of monitor block
4832 __ jmpb(entry);
4833
4834 __ bind(loop);
4835 // check if current entry is for same object
4836 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4837 // if same object then stop searching
4838 __ jcc(Assembler::equal, found);
4839 // otherwise advance to next entry
4840 __ addptr(rtop, entry_size);
4841 __ bind(entry);
4842 // check if bottom reached
4843 __ cmpptr(rtop, rbot);
4844 // if not at bottom then check this entry
4845 __ jcc(Assembler::notEqual, loop);
4846 }
4847
4848 // error handling. Unlocking was not block-structured
4849 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4850 InterpreterRuntime::throw_illegal_monitor_state_exception));
4851 __ should_not_reach_here();
4852
4853 // call run-time routine
4854 __ bind(found);
4855 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4856 __ unlock_object(rtop);
4857 __ pop_ptr(rax); // discard object
4858 }
4859
4860 // Wide instructions
4861 void TemplateTable::wide() {
4862 transition(vtos, vtos);
4863 __ load_unsigned_byte(rbx, at_bcp(1));
4864 ExternalAddress wtable((address)Interpreter::_wentry_point);
4865 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4866 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4867 }
4868
4869 // Multi arrays
4870 void TemplateTable::multianewarray() {
4871 transition(vtos, atos);
4872
4873 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4874 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4875 // last dim is on top of stack; we want address of first one:
4876 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4877 // the latter wordSize to point to the beginning of the array.
4878 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4879 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4880 __ load_unsigned_byte(rbx, at_bcp(3));
4881 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4882 }