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