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