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