1 /*
2 * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "compiler/disassembler.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterRuntime.hpp"
30 #include "interpreter/interp_masm.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "memory/universe.hpp"
33 #include "oops/methodData.hpp"
34 #include "oops/objArrayKlass.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/methodHandles.hpp"
37 #include "runtime/frame.inline.hpp"
38 #include "runtime/safepointMechanism.hpp"
39 #include "runtime/sharedRuntime.hpp"
40 #include "runtime/stubRoutines.hpp"
41 #include "runtime/synchronizer.hpp"
42 #include "utilities/macros.hpp"
43
44 #define __ Disassembler::hook<InterpreterMacroAssembler>(__FILE__, __LINE__, _masm)->
45
46 // Global Register Names
47 static const Register rbcp = LP64_ONLY(r13) NOT_LP64(rsi);
48 static const Register rlocals = LP64_ONLY(r14) NOT_LP64(rdi);
49
50 // Platform-dependent initialization
51 void TemplateTable::pd_initialize() {
52 // No x86 specific initialization
53 }
54
55 // Address Computation: local variables
56 static inline Address iaddress(int n) {
57 return Address(rlocals, Interpreter::local_offset_in_bytes(n));
58 }
59
60 static inline Address laddress(int n) {
61 return iaddress(n + 1);
62 }
63
64 #ifndef _LP64
65 static inline Address haddress(int n) {
66 return iaddress(n + 0);
67 }
68 #endif
69
70 static inline Address faddress(int n) {
71 return iaddress(n);
72 }
73
74 static inline Address daddress(int n) {
75 return laddress(n);
76 }
77
78 static inline Address aaddress(int n) {
79 return iaddress(n);
80 }
81
82 static inline Address iaddress(Register r) {
83 return Address(rlocals, r, Address::times_ptr);
84 }
85
86 static inline Address laddress(Register r) {
87 return Address(rlocals, r, Address::times_ptr, Interpreter::local_offset_in_bytes(1));
88 }
89
90 #ifndef _LP64
91 static inline Address haddress(Register r) {
92 return Address(rlocals, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
93 }
94 #endif
95
96 static inline Address faddress(Register r) {
97 return iaddress(r);
98 }
99
100 static inline Address daddress(Register r) {
101 return laddress(r);
102 }
103
104 static inline Address aaddress(Register r) {
105 return iaddress(r);
106 }
107
108
109 // expression stack
110 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
111 // data beyond the rsp which is potentially unsafe in an MT environment;
112 // an interrupt may overwrite that data.)
113 static inline Address at_rsp () {
114 return Address(rsp, 0);
115 }
116
117 // At top of Java expression stack which may be different than esp(). It
118 // isn't for category 1 objects.
119 static inline Address at_tos () {
120 return Address(rsp, Interpreter::expr_offset_in_bytes(0));
121 }
122
123 static inline Address at_tos_p1() {
124 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
125 }
126
127 static inline Address at_tos_p2() {
128 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
129 }
130
131 // Condition conversion
132 static Assembler::Condition j_not(TemplateTable::Condition cc) {
133 switch (cc) {
134 case TemplateTable::equal : return Assembler::notEqual;
135 case TemplateTable::not_equal : return Assembler::equal;
136 case TemplateTable::less : return Assembler::greaterEqual;
137 case TemplateTable::less_equal : return Assembler::greater;
138 case TemplateTable::greater : return Assembler::lessEqual;
139 case TemplateTable::greater_equal: return Assembler::less;
140 }
141 ShouldNotReachHere();
142 return Assembler::zero;
143 }
144
145
146
147 // Miscelaneous helper routines
148 // Store an oop (or NULL) at the address described by obj.
149 // If val == noreg this means store a NULL
150
151
152 static void do_oop_store(InterpreterMacroAssembler* _masm,
153 Address dst,
154 Register val,
155 DecoratorSet decorators = 0) {
156 assert(val == noreg || val == rax, "parameter is just for looks");
157 __ store_heap_oop(dst, val, rdx, rbx, decorators);
158 }
159
160 static void do_oop_load(InterpreterMacroAssembler* _masm,
161 Address src,
162 Register dst,
163 DecoratorSet decorators = 0) {
164 __ load_heap_oop(dst, src, rdx, rbx, decorators);
165 }
166
167 Address TemplateTable::at_bcp(int offset) {
168 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
169 return Address(rbcp, offset);
170 }
171
172
173 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
174 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
175 int byte_no) {
176 if (!RewriteBytecodes) return;
177 Label L_patch_done;
178
179 switch (bc) {
180 case Bytecodes::_fast_aputfield:
181 case Bytecodes::_fast_bputfield:
182 case Bytecodes::_fast_zputfield:
183 case Bytecodes::_fast_cputfield:
184 case Bytecodes::_fast_dputfield:
185 case Bytecodes::_fast_fputfield:
186 case Bytecodes::_fast_iputfield:
187 case Bytecodes::_fast_lputfield:
188 case Bytecodes::_fast_sputfield:
189 {
190 // We skip bytecode quickening for putfield instructions when
191 // the put_code written to the constant pool cache is zero.
192 // This is required so that every execution of this instruction
193 // calls out to InterpreterRuntime::resolve_get_put to do
194 // additional, required work.
195 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
196 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
197 __ get_cache_and_index_and_bytecode_at_bcp(temp_reg, bc_reg, temp_reg, byte_no, 1);
198 __ movl(bc_reg, bc);
199 __ cmpl(temp_reg, (int) 0);
200 __ jcc(Assembler::zero, L_patch_done); // don't patch
201 }
202 break;
203 default:
204 assert(byte_no == -1, "sanity");
205 // the pair bytecodes have already done the load.
206 if (load_bc_into_bc_reg) {
207 __ movl(bc_reg, bc);
208 }
209 }
210
211 if (JvmtiExport::can_post_breakpoint()) {
212 Label L_fast_patch;
213 // if a breakpoint is present we can't rewrite the stream directly
214 __ movzbl(temp_reg, at_bcp(0));
215 __ cmpl(temp_reg, Bytecodes::_breakpoint);
216 __ jcc(Assembler::notEqual, L_fast_patch);
217 __ get_method(temp_reg);
218 // Let breakpoint table handling rewrite to quicker bytecode
219 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rbcp, bc_reg);
220 #ifndef ASSERT
221 __ jmpb(L_patch_done);
222 #else
223 __ jmp(L_patch_done);
224 #endif
225 __ bind(L_fast_patch);
226 }
227
228 #ifdef ASSERT
229 Label L_okay;
230 __ load_unsigned_byte(temp_reg, at_bcp(0));
231 __ cmpl(temp_reg, (int) Bytecodes::java_code(bc));
232 __ jcc(Assembler::equal, L_okay);
233 __ cmpl(temp_reg, bc_reg);
234 __ jcc(Assembler::equal, L_okay);
235 __ stop("patching the wrong bytecode");
236 __ bind(L_okay);
237 #endif
238
239 // patch bytecode
240 __ movb(at_bcp(0), bc_reg);
241 __ bind(L_patch_done);
242 }
243 // Individual instructions
244
245
246 void TemplateTable::nop() {
247 transition(vtos, vtos);
248 // nothing to do
249 }
250
251 void TemplateTable::shouldnotreachhere() {
252 transition(vtos, vtos);
253 __ stop("shouldnotreachhere bytecode");
254 }
255
256 void TemplateTable::aconst_null() {
257 transition(vtos, atos);
258 __ xorl(rax, rax);
259 }
260
261 void TemplateTable::iconst(int value) {
262 transition(vtos, itos);
263 if (value == 0) {
264 __ xorl(rax, rax);
265 } else {
266 __ movl(rax, value);
267 }
268 }
269
270 void TemplateTable::lconst(int value) {
271 transition(vtos, ltos);
272 if (value == 0) {
273 __ xorl(rax, rax);
274 } else {
275 __ movl(rax, value);
276 }
277 #ifndef _LP64
278 assert(value >= 0, "check this code");
279 __ xorptr(rdx, rdx);
280 #endif
281 }
282
283
284
285 void TemplateTable::fconst(int value) {
286 transition(vtos, ftos);
287 if (UseSSE >= 1) {
288 static float one = 1.0f, two = 2.0f;
289 switch (value) {
290 case 0:
291 __ xorps(xmm0, xmm0);
292 break;
293 case 1:
294 __ movflt(xmm0, ExternalAddress((address) &one));
295 break;
296 case 2:
297 __ movflt(xmm0, ExternalAddress((address) &two));
298 break;
299 default:
300 ShouldNotReachHere();
301 break;
302 }
303 } else {
304 #ifdef _LP64
305 ShouldNotReachHere();
306 #else
307 if (value == 0) { __ fldz();
308 } else if (value == 1) { __ fld1();
309 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
310 } else { ShouldNotReachHere();
311 }
312 #endif // _LP64
313 }
314 }
315
316 void TemplateTable::dconst(int value) {
317 transition(vtos, dtos);
318 if (UseSSE >= 2) {
319 static double one = 1.0;
320 switch (value) {
321 case 0:
322 __ xorpd(xmm0, xmm0);
323 break;
324 case 1:
325 __ movdbl(xmm0, ExternalAddress((address) &one));
326 break;
327 default:
328 ShouldNotReachHere();
329 break;
330 }
331 } else {
332 #ifdef _LP64
333 ShouldNotReachHere();
334 #else
335 if (value == 0) { __ fldz();
336 } else if (value == 1) { __ fld1();
337 } else { ShouldNotReachHere();
338 }
339 #endif
340 }
341 }
342
343 void TemplateTable::bipush() {
344 transition(vtos, itos);
345 __ load_signed_byte(rax, at_bcp(1));
346 }
347
348 void TemplateTable::sipush() {
349 transition(vtos, itos);
350 __ load_unsigned_short(rax, at_bcp(1));
351 __ bswapl(rax);
352 __ sarl(rax, 16);
353 }
354
355 void TemplateTable::ldc(bool wide) {
356 transition(vtos, vtos);
357 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
358 Label call_ldc, notFloat, notClass, notInt, Done;
359
360 if (wide) {
361 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
362 } else {
363 __ load_unsigned_byte(rbx, at_bcp(1));
364 }
365
366 __ get_cpool_and_tags(rcx, rax);
367 const int base_offset = ConstantPool::header_size() * wordSize;
368 const int tags_offset = Array<u1>::base_offset_in_bytes();
369
370 // get type
371 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
372
373 // unresolved class - get the resolved class
374 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
375 __ jccb(Assembler::equal, call_ldc);
376
377 // unresolved class in error state - call into runtime to throw the error
378 // from the first resolution attempt
379 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
380 __ jccb(Assembler::equal, call_ldc);
381
382 // resolved class - need to call vm to get java mirror of the class
383 __ cmpl(rdx, JVM_CONSTANT_Class);
384 __ jcc(Assembler::notEqual, notClass);
385
386 __ bind(call_ldc);
387
388 __ movl(rarg, wide);
389 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rarg);
390
391 __ push(atos);
392 __ jmp(Done);
393
394 __ bind(notClass);
395 __ cmpl(rdx, JVM_CONSTANT_Float);
396 __ jccb(Assembler::notEqual, notFloat);
397
398 // ftos
399 __ load_float(Address(rcx, rbx, Address::times_ptr, base_offset));
400 __ push(ftos);
401 __ jmp(Done);
402
403 __ bind(notFloat);
404 __ cmpl(rdx, JVM_CONSTANT_Integer);
405 __ jccb(Assembler::notEqual, notInt);
406
407 // itos
408 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
409 __ push(itos);
410 __ jmp(Done);
411
412 // assume the tag is for condy; if not, the VM runtime will tell us
413 __ bind(notInt);
414 condy_helper(Done);
415
416 __ bind(Done);
417 }
418
419 // Fast path for caching oop constants.
420 void TemplateTable::fast_aldc(bool wide) {
421 transition(vtos, atos);
422
423 Register result = rax;
424 Register tmp = rdx;
425 Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
426 int index_size = wide ? sizeof(u2) : sizeof(u1);
427
428 Label resolved;
429
430 // We are resolved if the resolved reference cache entry contains a
431 // non-null object (String, MethodType, etc.)
432 assert_different_registers(result, tmp);
433 __ get_cache_index_at_bcp(tmp, 1, index_size);
434 __ load_resolved_reference_at_index(result, tmp);
435 __ testptr(result, result);
436 __ jcc(Assembler::notZero, resolved);
437
438 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
439
440 // first time invocation - must resolve first
441 __ movl(rarg, (int)bytecode());
442 __ call_VM(result, entry, rarg);
443 __ bind(resolved);
444
445 { // Check for the null sentinel.
446 // If we just called the VM, it already did the mapping for us,
447 // but it's harmless to retry.
448 Label notNull;
449 ExternalAddress null_sentinel((address)Universe::the_null_sentinel_addr());
450 __ movptr(tmp, null_sentinel);
451 __ cmpoop(tmp, result);
452 __ jccb(Assembler::notEqual, notNull);
453 __ xorptr(result, result); // NULL object reference
454 __ bind(notNull);
455 }
456
457 if (VerifyOops) {
458 __ verify_oop(result);
459 }
460 }
461
462 void TemplateTable::ldc2_w() {
463 transition(vtos, vtos);
464 Label notDouble, notLong, Done;
465 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
466
467 __ get_cpool_and_tags(rcx, rax);
468 const int base_offset = ConstantPool::header_size() * wordSize;
469 const int tags_offset = Array<u1>::base_offset_in_bytes();
470
471 // get type
472 __ movzbl(rdx, Address(rax, rbx, Address::times_1, tags_offset));
473 __ cmpl(rdx, JVM_CONSTANT_Double);
474 __ jccb(Assembler::notEqual, notDouble);
475
476 // dtos
477 __ load_double(Address(rcx, rbx, Address::times_ptr, base_offset));
478 __ push(dtos);
479
480 __ jmp(Done);
481 __ bind(notDouble);
482 __ cmpl(rdx, JVM_CONSTANT_Long);
483 __ jccb(Assembler::notEqual, notLong);
484
485 // ltos
486 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
487 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
488 __ push(ltos);
489 __ jmp(Done);
490
491 __ bind(notLong);
492 condy_helper(Done);
493
494 __ bind(Done);
495 }
496
497 void TemplateTable::condy_helper(Label& Done) {
498 const Register obj = rax;
499 const Register off = rbx;
500 const Register flags = rcx;
501 const Register rarg = NOT_LP64(rcx) LP64_ONLY(c_rarg1);
502 __ movl(rarg, (int)bytecode());
503 call_VM(obj, CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc), rarg);
504 #ifndef _LP64
505 // borrow rdi from locals
506 __ get_thread(rdi);
507 __ get_vm_result_2(flags, rdi);
508 __ restore_locals();
509 #else
510 __ get_vm_result_2(flags, r15_thread);
511 #endif
512 // VMr = obj = base address to find primitive value to push
513 // VMr2 = flags = (tos, off) using format of CPCE::_flags
514 __ movl(off, flags);
515 __ andl(off, ConstantPoolCacheEntry::field_index_mask);
516 const Address field(obj, off, Address::times_1, 0*wordSize);
517
518 // What sort of thing are we loading?
519 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
520 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
521
522 switch (bytecode()) {
523 case Bytecodes::_ldc:
524 case Bytecodes::_ldc_w:
525 {
526 // tos in (itos, ftos, stos, btos, ctos, ztos)
527 Label notInt, notFloat, notShort, notByte, notChar, notBool;
528 __ cmpl(flags, itos);
529 __ jcc(Assembler::notEqual, notInt);
530 // itos
531 __ movl(rax, field);
532 __ push(itos);
533 __ jmp(Done);
534
535 __ bind(notInt);
536 __ cmpl(flags, ftos);
537 __ jcc(Assembler::notEqual, notFloat);
538 // ftos
539 __ load_float(field);
540 __ push(ftos);
541 __ jmp(Done);
542
543 __ bind(notFloat);
544 __ cmpl(flags, stos);
545 __ jcc(Assembler::notEqual, notShort);
546 // stos
547 __ load_signed_short(rax, field);
548 __ push(stos);
549 __ jmp(Done);
550
551 __ bind(notShort);
552 __ cmpl(flags, btos);
553 __ jcc(Assembler::notEqual, notByte);
554 // btos
555 __ load_signed_byte(rax, field);
556 __ push(btos);
557 __ jmp(Done);
558
559 __ bind(notByte);
560 __ cmpl(flags, ctos);
561 __ jcc(Assembler::notEqual, notChar);
562 // ctos
563 __ load_unsigned_short(rax, field);
564 __ push(ctos);
565 __ jmp(Done);
566
567 __ bind(notChar);
568 __ cmpl(flags, ztos);
569 __ jcc(Assembler::notEqual, notBool);
570 // ztos
571 __ load_signed_byte(rax, field);
572 __ push(ztos);
573 __ jmp(Done);
574
575 __ bind(notBool);
576 break;
577 }
578
579 case Bytecodes::_ldc2_w:
580 {
581 Label notLong, notDouble;
582 __ cmpl(flags, ltos);
583 __ jcc(Assembler::notEqual, notLong);
584 // ltos
585 __ movptr(rax, field);
586 NOT_LP64(__ movptr(rdx, field.plus_disp(4)));
587 __ push(ltos);
588 __ jmp(Done);
589
590 __ bind(notLong);
591 __ cmpl(flags, dtos);
592 __ jcc(Assembler::notEqual, notDouble);
593 // dtos
594 __ load_double(field);
595 __ push(dtos);
596 __ jmp(Done);
597
598 __ bind(notDouble);
599 break;
600 }
601
602 default:
603 ShouldNotReachHere();
604 }
605
606 __ stop("bad ldc/condy");
607 }
608
609 void TemplateTable::locals_index(Register reg, int offset) {
610 __ load_unsigned_byte(reg, at_bcp(offset));
611 __ negptr(reg);
612 }
613
614 void TemplateTable::iload() {
615 iload_internal();
616 }
617
618 void TemplateTable::nofast_iload() {
619 iload_internal(may_not_rewrite);
620 }
621
622 void TemplateTable::iload_internal(RewriteControl rc) {
623 transition(vtos, itos);
624 if (RewriteFrequentPairs && rc == may_rewrite) {
625 Label rewrite, done;
626 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
627 LP64_ONLY(assert(rbx != bc, "register damaged"));
628
629 // get next byte
630 __ load_unsigned_byte(rbx,
631 at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
632 // if _iload, wait to rewrite to iload2. We only want to rewrite the
633 // last two iloads in a pair. Comparing against fast_iload means that
634 // the next bytecode is neither an iload or a caload, and therefore
635 // an iload pair.
636 __ cmpl(rbx, Bytecodes::_iload);
637 __ jcc(Assembler::equal, done);
638
639 __ cmpl(rbx, Bytecodes::_fast_iload);
640 __ movl(bc, Bytecodes::_fast_iload2);
641
642 __ jccb(Assembler::equal, rewrite);
643
644 // if _caload, rewrite to fast_icaload
645 __ cmpl(rbx, Bytecodes::_caload);
646 __ movl(bc, Bytecodes::_fast_icaload);
647 __ jccb(Assembler::equal, rewrite);
648
649 // rewrite so iload doesn't check again.
650 __ movl(bc, Bytecodes::_fast_iload);
651
652 // rewrite
653 // bc: fast bytecode
654 __ bind(rewrite);
655 patch_bytecode(Bytecodes::_iload, bc, rbx, false);
656 __ bind(done);
657 }
658
659 // Get the local value into tos
660 locals_index(rbx);
661 __ movl(rax, iaddress(rbx));
662 }
663
664 void TemplateTable::fast_iload2() {
665 transition(vtos, itos);
666 locals_index(rbx);
667 __ movl(rax, iaddress(rbx));
668 __ push(itos);
669 locals_index(rbx, 3);
670 __ movl(rax, iaddress(rbx));
671 }
672
673 void TemplateTable::fast_iload() {
674 transition(vtos, itos);
675 locals_index(rbx);
676 __ movl(rax, iaddress(rbx));
677 }
678
679 void TemplateTable::lload() {
680 transition(vtos, ltos);
681 locals_index(rbx);
682 __ movptr(rax, laddress(rbx));
683 NOT_LP64(__ movl(rdx, haddress(rbx)));
684 }
685
686 void TemplateTable::fload() {
687 transition(vtos, ftos);
688 locals_index(rbx);
689 __ load_float(faddress(rbx));
690 }
691
692 void TemplateTable::dload() {
693 transition(vtos, dtos);
694 locals_index(rbx);
695 __ load_double(daddress(rbx));
696 }
697
698 void TemplateTable::aload() {
699 transition(vtos, atos);
700 locals_index(rbx);
701 __ movptr(rax, aaddress(rbx));
702 }
703
704 void TemplateTable::locals_index_wide(Register reg) {
705 __ load_unsigned_short(reg, at_bcp(2));
706 __ bswapl(reg);
707 __ shrl(reg, 16);
708 __ negptr(reg);
709 }
710
711 void TemplateTable::wide_iload() {
712 transition(vtos, itos);
713 locals_index_wide(rbx);
714 __ movl(rax, iaddress(rbx));
715 }
716
717 void TemplateTable::wide_lload() {
718 transition(vtos, ltos);
719 locals_index_wide(rbx);
720 __ movptr(rax, laddress(rbx));
721 NOT_LP64(__ movl(rdx, haddress(rbx)));
722 }
723
724 void TemplateTable::wide_fload() {
725 transition(vtos, ftos);
726 locals_index_wide(rbx);
727 __ load_float(faddress(rbx));
728 }
729
730 void TemplateTable::wide_dload() {
731 transition(vtos, dtos);
732 locals_index_wide(rbx);
733 __ load_double(daddress(rbx));
734 }
735
736 void TemplateTable::wide_aload() {
737 transition(vtos, atos);
738 locals_index_wide(rbx);
739 __ movptr(rax, aaddress(rbx));
740 }
741
742 void TemplateTable::index_check(Register array, Register index) {
743 // Pop ptr into array
744 __ pop_ptr(array);
745 index_check_without_pop(array, index);
746 }
747
748 void TemplateTable::index_check_without_pop(Register array, Register index) {
749 // destroys rbx
750 // check array
751 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
752 // sign extend index for use by indexed load
753 __ movl2ptr(index, index);
754 // check index
755 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
756 if (index != rbx) {
757 // ??? convention: move aberrant index into rbx for exception message
758 assert(rbx != array, "different registers");
759 __ movl(rbx, index);
760 }
761 Label skip;
762 __ jccb(Assembler::below, skip);
763 // Pass array to create more detailed exceptions.
764 __ mov(NOT_LP64(rax) LP64_ONLY(c_rarg1), array);
765 __ jump(ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
766 __ bind(skip);
767 }
768
769 void TemplateTable::iaload() {
770 transition(itos, itos);
771 // rax: index
772 // rdx: array
773 index_check(rdx, rax); // kills rbx
774 __ access_load_at(T_INT, IN_HEAP | IS_ARRAY, rax,
775 Address(rdx, rax, Address::times_4,
776 arrayOopDesc::base_offset_in_bytes(T_INT)),
777 noreg, noreg);
778 }
779
780 void TemplateTable::laload() {
781 transition(itos, ltos);
782 // rax: index
783 // rdx: array
784 index_check(rdx, rax); // kills rbx
785 NOT_LP64(__ mov(rbx, rax));
786 // rbx,: index
787 __ access_load_at(T_LONG, IN_HEAP | IS_ARRAY, noreg /* ltos */,
788 Address(rdx, rbx, Address::times_8,
789 arrayOopDesc::base_offset_in_bytes(T_LONG)),
790 noreg, noreg);
791 }
792
793
794
795 void TemplateTable::faload() {
796 transition(itos, ftos);
797 // rax: index
798 // rdx: array
799 index_check(rdx, rax); // kills rbx
800 __ access_load_at(T_FLOAT, IN_HEAP | IS_ARRAY, noreg /* ftos */,
801 Address(rdx, rax,
802 Address::times_4,
803 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
804 noreg, noreg);
805 }
806
807 void TemplateTable::daload() {
808 transition(itos, dtos);
809 // rax: index
810 // rdx: array
811 index_check(rdx, rax); // kills rbx
812 __ access_load_at(T_DOUBLE, IN_HEAP | IS_ARRAY, noreg /* dtos */,
813 Address(rdx, rax,
814 Address::times_8,
815 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
816 noreg, noreg);
817 }
818
819 void TemplateTable::aaload() {
820 transition(itos, atos);
821 // rax: index
822 // rdx: array
823 index_check(rdx, rax); // kills rbx
824 do_oop_load(_masm,
825 Address(rdx, rax,
826 UseCompressedOops ? Address::times_4 : Address::times_ptr,
827 arrayOopDesc::base_offset_in_bytes(T_OBJECT)),
828 rax,
829 IS_ARRAY);
830 }
831
832 void TemplateTable::baload() {
833 transition(itos, itos);
834 // rax: index
835 // rdx: array
836 index_check(rdx, rax); // kills rbx
837 __ access_load_at(T_BYTE, IN_HEAP | IS_ARRAY, rax,
838 Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)),
839 noreg, noreg);
840 }
841
842 void TemplateTable::caload() {
843 transition(itos, itos);
844 // rax: index
845 // rdx: array
846 index_check(rdx, rax); // kills rbx
847 __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
848 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
849 noreg, noreg);
850 }
851
852 // iload followed by caload frequent pair
853 void TemplateTable::fast_icaload() {
854 transition(vtos, itos);
855 // load index out of locals
856 locals_index(rbx);
857 __ movl(rax, iaddress(rbx));
858
859 // rax: index
860 // rdx: array
861 index_check(rdx, rax); // kills rbx
862 __ access_load_at(T_CHAR, IN_HEAP | IS_ARRAY, rax,
863 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)),
864 noreg, noreg);
865 }
866
867
868 void TemplateTable::saload() {
869 transition(itos, itos);
870 // rax: index
871 // rdx: array
872 index_check(rdx, rax); // kills rbx
873 __ access_load_at(T_SHORT, IN_HEAP | IS_ARRAY, rax,
874 Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)),
875 noreg, noreg);
876 }
877
878 void TemplateTable::iload(int n) {
879 transition(vtos, itos);
880 __ movl(rax, iaddress(n));
881 }
882
883 void TemplateTable::lload(int n) {
884 transition(vtos, ltos);
885 __ movptr(rax, laddress(n));
886 NOT_LP64(__ movptr(rdx, haddress(n)));
887 }
888
889 void TemplateTable::fload(int n) {
890 transition(vtos, ftos);
891 __ load_float(faddress(n));
892 }
893
894 void TemplateTable::dload(int n) {
895 transition(vtos, dtos);
896 __ load_double(daddress(n));
897 }
898
899 void TemplateTable::aload(int n) {
900 transition(vtos, atos);
901 __ movptr(rax, aaddress(n));
902 }
903
904 void TemplateTable::aload_0() {
905 aload_0_internal();
906 }
907
908 void TemplateTable::nofast_aload_0() {
909 aload_0_internal(may_not_rewrite);
910 }
911
912 void TemplateTable::aload_0_internal(RewriteControl rc) {
913 transition(vtos, atos);
914 // According to bytecode histograms, the pairs:
915 //
916 // _aload_0, _fast_igetfield
917 // _aload_0, _fast_agetfield
918 // _aload_0, _fast_fgetfield
919 //
920 // occur frequently. If RewriteFrequentPairs is set, the (slow)
921 // _aload_0 bytecode checks if the next bytecode is either
922 // _fast_igetfield, _fast_agetfield or _fast_fgetfield and then
923 // rewrites the current bytecode into a pair bytecode; otherwise it
924 // rewrites the current bytecode into _fast_aload_0 that doesn't do
925 // the pair check anymore.
926 //
927 // Note: If the next bytecode is _getfield, the rewrite must be
928 // delayed, otherwise we may miss an opportunity for a pair.
929 //
930 // Also rewrite frequent pairs
931 // aload_0, aload_1
932 // aload_0, iload_1
933 // These bytecodes with a small amount of code are most profitable
934 // to rewrite
935 if (RewriteFrequentPairs && rc == may_rewrite) {
936 Label rewrite, done;
937
938 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
939 LP64_ONLY(assert(rbx != bc, "register damaged"));
940
941 // get next byte
942 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
943
944 // if _getfield then wait with rewrite
945 __ cmpl(rbx, Bytecodes::_getfield);
946 __ jcc(Assembler::equal, done);
947
948 // if _igetfield then rewrite to _fast_iaccess_0
949 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
950 __ cmpl(rbx, Bytecodes::_fast_igetfield);
951 __ movl(bc, Bytecodes::_fast_iaccess_0);
952 __ jccb(Assembler::equal, rewrite);
953
954 // if _agetfield then rewrite to _fast_aaccess_0
955 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
956 __ cmpl(rbx, Bytecodes::_fast_agetfield);
957 __ movl(bc, Bytecodes::_fast_aaccess_0);
958 __ jccb(Assembler::equal, rewrite);
959
960 // if _fgetfield then rewrite to _fast_faccess_0
961 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
962 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
963 __ movl(bc, Bytecodes::_fast_faccess_0);
964 __ jccb(Assembler::equal, rewrite);
965
966 // else rewrite to _fast_aload0
967 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
968 __ movl(bc, Bytecodes::_fast_aload_0);
969
970 // rewrite
971 // bc: fast bytecode
972 __ bind(rewrite);
973 patch_bytecode(Bytecodes::_aload_0, bc, rbx, false);
974
975 __ bind(done);
976 }
977
978 // Do actual aload_0 (must do this after patch_bytecode which might call VM and GC might change oop).
979 aload(0);
980 }
981
982 void TemplateTable::istore() {
983 transition(itos, vtos);
984 locals_index(rbx);
985 __ movl(iaddress(rbx), rax);
986 }
987
988
989 void TemplateTable::lstore() {
990 transition(ltos, vtos);
991 locals_index(rbx);
992 __ movptr(laddress(rbx), rax);
993 NOT_LP64(__ movptr(haddress(rbx), rdx));
994 }
995
996 void TemplateTable::fstore() {
997 transition(ftos, vtos);
998 locals_index(rbx);
999 __ store_float(faddress(rbx));
1000 }
1001
1002 void TemplateTable::dstore() {
1003 transition(dtos, vtos);
1004 locals_index(rbx);
1005 __ store_double(daddress(rbx));
1006 }
1007
1008 void TemplateTable::astore() {
1009 transition(vtos, vtos);
1010 __ pop_ptr(rax);
1011 locals_index(rbx);
1012 __ movptr(aaddress(rbx), rax);
1013 }
1014
1015 void TemplateTable::wide_istore() {
1016 transition(vtos, vtos);
1017 __ pop_i();
1018 locals_index_wide(rbx);
1019 __ movl(iaddress(rbx), rax);
1020 }
1021
1022 void TemplateTable::wide_lstore() {
1023 transition(vtos, vtos);
1024 NOT_LP64(__ pop_l(rax, rdx));
1025 LP64_ONLY(__ pop_l());
1026 locals_index_wide(rbx);
1027 __ movptr(laddress(rbx), rax);
1028 NOT_LP64(__ movl(haddress(rbx), rdx));
1029 }
1030
1031 void TemplateTable::wide_fstore() {
1032 #ifdef _LP64
1033 transition(vtos, vtos);
1034 __ pop_f(xmm0);
1035 locals_index_wide(rbx);
1036 __ movflt(faddress(rbx), xmm0);
1037 #else
1038 wide_istore();
1039 #endif
1040 }
1041
1042 void TemplateTable::wide_dstore() {
1043 #ifdef _LP64
1044 transition(vtos, vtos);
1045 __ pop_d(xmm0);
1046 locals_index_wide(rbx);
1047 __ movdbl(daddress(rbx), xmm0);
1048 #else
1049 wide_lstore();
1050 #endif
1051 }
1052
1053 void TemplateTable::wide_astore() {
1054 transition(vtos, vtos);
1055 __ pop_ptr(rax);
1056 locals_index_wide(rbx);
1057 __ movptr(aaddress(rbx), rax);
1058 }
1059
1060 void TemplateTable::iastore() {
1061 transition(itos, vtos);
1062 __ pop_i(rbx);
1063 // rax: value
1064 // rbx: index
1065 // rdx: array
1066 index_check(rdx, rbx); // prefer index in rbx
1067 __ access_store_at(T_INT, IN_HEAP | IS_ARRAY,
1068 Address(rdx, rbx, Address::times_4,
1069 arrayOopDesc::base_offset_in_bytes(T_INT)),
1070 rax, noreg, noreg);
1071 }
1072
1073 void TemplateTable::lastore() {
1074 transition(ltos, vtos);
1075 __ pop_i(rbx);
1076 // rax,: low(value)
1077 // rcx: array
1078 // rdx: high(value)
1079 index_check(rcx, rbx); // prefer index in rbx,
1080 // rbx,: index
1081 __ access_store_at(T_LONG, IN_HEAP | IS_ARRAY,
1082 Address(rcx, rbx, Address::times_8,
1083 arrayOopDesc::base_offset_in_bytes(T_LONG)),
1084 noreg /* ltos */, noreg, noreg);
1085 }
1086
1087
1088 void TemplateTable::fastore() {
1089 transition(ftos, vtos);
1090 __ pop_i(rbx);
1091 // value is in UseSSE >= 1 ? xmm0 : ST(0)
1092 // rbx: index
1093 // rdx: array
1094 index_check(rdx, rbx); // prefer index in rbx
1095 __ access_store_at(T_FLOAT, IN_HEAP | IS_ARRAY,
1096 Address(rdx, rbx, Address::times_4,
1097 arrayOopDesc::base_offset_in_bytes(T_FLOAT)),
1098 noreg /* ftos */, noreg, noreg);
1099 }
1100
1101 void TemplateTable::dastore() {
1102 transition(dtos, vtos);
1103 __ pop_i(rbx);
1104 // value is in UseSSE >= 2 ? xmm0 : ST(0)
1105 // rbx: index
1106 // rdx: array
1107 index_check(rdx, rbx); // prefer index in rbx
1108 __ access_store_at(T_DOUBLE, IN_HEAP | IS_ARRAY,
1109 Address(rdx, rbx, Address::times_8,
1110 arrayOopDesc::base_offset_in_bytes(T_DOUBLE)),
1111 noreg /* dtos */, noreg, noreg);
1112 }
1113
1114 void TemplateTable::aastore() {
1115 Label is_null, ok_is_subtype, done;
1116 transition(vtos, vtos);
1117 // stack: ..., array, index, value
1118 __ movptr(rax, at_tos()); // value
1119 __ movl(rcx, at_tos_p1()); // index
1120 __ movptr(rdx, at_tos_p2()); // array
1121
1122 Address element_address(rdx, rcx,
1123 UseCompressedOops? Address::times_4 : Address::times_ptr,
1124 arrayOopDesc::base_offset_in_bytes(T_OBJECT));
1125
1126 index_check_without_pop(rdx, rcx); // kills rbx
1127 __ testptr(rax, rax);
1128 __ jcc(Assembler::zero, is_null);
1129
1130 // Move subklass into rbx
1131 __ load_klass(rbx, rax);
1132 // Move superklass into rax
1133 __ load_klass(rax, rdx);
1134 __ movptr(rax, Address(rax,
1135 ObjArrayKlass::element_klass_offset()));
1136
1137 // Generate subtype check. Blows rcx, rdi
1138 // Superklass in rax. Subklass in rbx.
1139 __ gen_subtype_check(rbx, ok_is_subtype);
1140
1141 // Come here on failure
1142 // object is at TOS
1143 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
1144
1145 // Come here on success
1146 __ bind(ok_is_subtype);
1147
1148 // Get the value we will store
1149 __ movptr(rax, at_tos());
1150 __ movl(rcx, at_tos_p1()); // index
1151 // Now store using the appropriate barrier
1152 do_oop_store(_masm, element_address, rax, IS_ARRAY);
1153 __ jmp(done);
1154
1155 // Have a NULL in rax, rdx=array, ecx=index. Store NULL at ary[idx]
1156 __ bind(is_null);
1157 __ profile_null_seen(rbx);
1158
1159 // Store a NULL
1160 do_oop_store(_masm, element_address, noreg, IS_ARRAY);
1161
1162 // Pop stack arguments
1163 __ bind(done);
1164 __ addptr(rsp, 3 * Interpreter::stackElementSize);
1165 }
1166
1167 void TemplateTable::bastore() {
1168 transition(itos, vtos);
1169 __ pop_i(rbx);
1170 // rax: value
1171 // rbx: index
1172 // rdx: array
1173 index_check(rdx, rbx); // prefer index in rbx
1174 // Need to check whether array is boolean or byte
1175 // since both types share the bastore bytecode.
1176 __ load_klass(rcx, rdx);
1177 __ movl(rcx, Address(rcx, Klass::layout_helper_offset()));
1178 int diffbit = Klass::layout_helper_boolean_diffbit();
1179 __ testl(rcx, diffbit);
1180 Label L_skip;
1181 __ jccb(Assembler::zero, L_skip);
1182 __ andl(rax, 1); // if it is a T_BOOLEAN array, mask the stored value to 0/1
1183 __ bind(L_skip);
1184 __ access_store_at(T_BYTE, IN_HEAP | IS_ARRAY,
1185 Address(rdx, rbx,Address::times_1,
1186 arrayOopDesc::base_offset_in_bytes(T_BYTE)),
1187 rax, noreg, noreg);
1188 }
1189
1190 void TemplateTable::castore() {
1191 transition(itos, vtos);
1192 __ pop_i(rbx);
1193 // rax: value
1194 // rbx: index
1195 // rdx: array
1196 index_check(rdx, rbx); // prefer index in rbx
1197 __ access_store_at(T_CHAR, IN_HEAP | IS_ARRAY,
1198 Address(rdx, rbx, Address::times_2,
1199 arrayOopDesc::base_offset_in_bytes(T_CHAR)),
1200 rax, noreg, noreg);
1201 }
1202
1203
1204 void TemplateTable::sastore() {
1205 castore();
1206 }
1207
1208 void TemplateTable::istore(int n) {
1209 transition(itos, vtos);
1210 __ movl(iaddress(n), rax);
1211 }
1212
1213 void TemplateTable::lstore(int n) {
1214 transition(ltos, vtos);
1215 __ movptr(laddress(n), rax);
1216 NOT_LP64(__ movptr(haddress(n), rdx));
1217 }
1218
1219 void TemplateTable::fstore(int n) {
1220 transition(ftos, vtos);
1221 __ store_float(faddress(n));
1222 }
1223
1224 void TemplateTable::dstore(int n) {
1225 transition(dtos, vtos);
1226 __ store_double(daddress(n));
1227 }
1228
1229
1230 void TemplateTable::astore(int n) {
1231 transition(vtos, vtos);
1232 __ pop_ptr(rax);
1233 __ movptr(aaddress(n), rax);
1234 }
1235
1236 void TemplateTable::pop() {
1237 transition(vtos, vtos);
1238 __ addptr(rsp, Interpreter::stackElementSize);
1239 }
1240
1241 void TemplateTable::pop2() {
1242 transition(vtos, vtos);
1243 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1244 }
1245
1246
1247 void TemplateTable::dup() {
1248 transition(vtos, vtos);
1249 __ load_ptr(0, rax);
1250 __ push_ptr(rax);
1251 // stack: ..., a, a
1252 }
1253
1254 void TemplateTable::dup_x1() {
1255 transition(vtos, vtos);
1256 // stack: ..., a, b
1257 __ load_ptr( 0, rax); // load b
1258 __ load_ptr( 1, rcx); // load a
1259 __ store_ptr(1, rax); // store b
1260 __ store_ptr(0, rcx); // store a
1261 __ push_ptr(rax); // push b
1262 // stack: ..., b, a, b
1263 }
1264
1265 void TemplateTable::dup_x2() {
1266 transition(vtos, vtos);
1267 // stack: ..., a, b, c
1268 __ load_ptr( 0, rax); // load c
1269 __ load_ptr( 2, rcx); // load a
1270 __ store_ptr(2, rax); // store c in a
1271 __ push_ptr(rax); // push c
1272 // stack: ..., c, b, c, c
1273 __ load_ptr( 2, rax); // load b
1274 __ store_ptr(2, rcx); // store a in b
1275 // stack: ..., c, a, c, c
1276 __ store_ptr(1, rax); // store b in c
1277 // stack: ..., c, a, b, c
1278 }
1279
1280 void TemplateTable::dup2() {
1281 transition(vtos, vtos);
1282 // stack: ..., a, b
1283 __ load_ptr(1, rax); // load a
1284 __ push_ptr(rax); // push a
1285 __ load_ptr(1, rax); // load b
1286 __ push_ptr(rax); // push b
1287 // stack: ..., a, b, a, b
1288 }
1289
1290
1291 void TemplateTable::dup2_x1() {
1292 transition(vtos, vtos);
1293 // stack: ..., a, b, c
1294 __ load_ptr( 0, rcx); // load c
1295 __ load_ptr( 1, rax); // load b
1296 __ push_ptr(rax); // push b
1297 __ push_ptr(rcx); // push c
1298 // stack: ..., a, b, c, b, c
1299 __ store_ptr(3, rcx); // store c in b
1300 // stack: ..., a, c, c, b, c
1301 __ load_ptr( 4, rcx); // load a
1302 __ store_ptr(2, rcx); // store a in 2nd c
1303 // stack: ..., a, c, a, b, c
1304 __ store_ptr(4, rax); // store b in a
1305 // stack: ..., b, c, a, b, c
1306 }
1307
1308 void TemplateTable::dup2_x2() {
1309 transition(vtos, vtos);
1310 // stack: ..., a, b, c, d
1311 __ load_ptr( 0, rcx); // load d
1312 __ load_ptr( 1, rax); // load c
1313 __ push_ptr(rax); // push c
1314 __ push_ptr(rcx); // push d
1315 // stack: ..., a, b, c, d, c, d
1316 __ load_ptr( 4, rax); // load b
1317 __ store_ptr(2, rax); // store b in d
1318 __ store_ptr(4, rcx); // store d in b
1319 // stack: ..., a, d, c, b, c, d
1320 __ load_ptr( 5, rcx); // load a
1321 __ load_ptr( 3, rax); // load c
1322 __ store_ptr(3, rcx); // store a in c
1323 __ store_ptr(5, rax); // store c in a
1324 // stack: ..., c, d, a, b, c, d
1325 }
1326
1327 void TemplateTable::swap() {
1328 transition(vtos, vtos);
1329 // stack: ..., a, b
1330 __ load_ptr( 1, rcx); // load a
1331 __ load_ptr( 0, rax); // load b
1332 __ store_ptr(0, rcx); // store a in b
1333 __ store_ptr(1, rax); // store b in a
1334 // stack: ..., b, a
1335 }
1336
1337 void TemplateTable::iop2(Operation op) {
1338 transition(itos, itos);
1339 switch (op) {
1340 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1341 case sub : __ movl(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1342 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1343 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1344 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1345 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1346 case shl : __ movl(rcx, rax); __ pop_i(rax); __ shll (rax); break;
1347 case shr : __ movl(rcx, rax); __ pop_i(rax); __ sarl (rax); break;
1348 case ushr : __ movl(rcx, rax); __ pop_i(rax); __ shrl (rax); break;
1349 default : ShouldNotReachHere();
1350 }
1351 }
1352
1353 void TemplateTable::lop2(Operation op) {
1354 transition(ltos, ltos);
1355 #ifdef _LP64
1356 switch (op) {
1357 case add : __ pop_l(rdx); __ addptr(rax, rdx); break;
1358 case sub : __ mov(rdx, rax); __ pop_l(rax); __ subptr(rax, rdx); break;
1359 case _and : __ pop_l(rdx); __ andptr(rax, rdx); break;
1360 case _or : __ pop_l(rdx); __ orptr (rax, rdx); break;
1361 case _xor : __ pop_l(rdx); __ xorptr(rax, rdx); break;
1362 default : ShouldNotReachHere();
1363 }
1364 #else
1365 __ pop_l(rbx, rcx);
1366 switch (op) {
1367 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1368 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1369 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1370 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1371 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1372 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1373 default : ShouldNotReachHere();
1374 }
1375 #endif
1376 }
1377
1378 void TemplateTable::idiv() {
1379 transition(itos, itos);
1380 __ movl(rcx, rax);
1381 __ pop_i(rax);
1382 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1383 // they are not equal, one could do a normal division (no correction
1384 // needed), which may speed up this implementation for the common case.
1385 // (see also JVM spec., p.243 & p.271)
1386 __ corrected_idivl(rcx);
1387 }
1388
1389 void TemplateTable::irem() {
1390 transition(itos, itos);
1391 __ movl(rcx, rax);
1392 __ pop_i(rax);
1393 // Note: could xor rax and ecx and compare with (-1 ^ min_int). If
1394 // they are not equal, one could do a normal division (no correction
1395 // needed), which may speed up this implementation for the common case.
1396 // (see also JVM spec., p.243 & p.271)
1397 __ corrected_idivl(rcx);
1398 __ movl(rax, rdx);
1399 }
1400
1401 void TemplateTable::lmul() {
1402 transition(ltos, ltos);
1403 #ifdef _LP64
1404 __ pop_l(rdx);
1405 __ imulq(rax, rdx);
1406 #else
1407 __ pop_l(rbx, rcx);
1408 __ push(rcx); __ push(rbx);
1409 __ push(rdx); __ push(rax);
1410 __ lmul(2 * wordSize, 0);
1411 __ addptr(rsp, 4 * wordSize); // take off temporaries
1412 #endif
1413 }
1414
1415 void TemplateTable::ldiv() {
1416 transition(ltos, ltos);
1417 #ifdef _LP64
1418 __ mov(rcx, rax);
1419 __ pop_l(rax);
1420 // generate explicit div0 check
1421 __ testq(rcx, rcx);
1422 __ jump_cc(Assembler::zero,
1423 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1424 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1425 // they are not equal, one could do a normal division (no correction
1426 // needed), which may speed up this implementation for the common case.
1427 // (see also JVM spec., p.243 & p.271)
1428 __ corrected_idivq(rcx); // kills rbx
1429 #else
1430 __ pop_l(rbx, rcx);
1431 __ push(rcx); __ push(rbx);
1432 __ push(rdx); __ push(rax);
1433 // check if y = 0
1434 __ orl(rax, rdx);
1435 __ jump_cc(Assembler::zero,
1436 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1437 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1438 __ addptr(rsp, 4 * wordSize); // take off temporaries
1439 #endif
1440 }
1441
1442 void TemplateTable::lrem() {
1443 transition(ltos, ltos);
1444 #ifdef _LP64
1445 __ mov(rcx, rax);
1446 __ pop_l(rax);
1447 __ testq(rcx, rcx);
1448 __ jump_cc(Assembler::zero,
1449 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1450 // Note: could xor rax and rcx and compare with (-1 ^ min_int). If
1451 // they are not equal, one could do a normal division (no correction
1452 // needed), which may speed up this implementation for the common case.
1453 // (see also JVM spec., p.243 & p.271)
1454 __ corrected_idivq(rcx); // kills rbx
1455 __ mov(rax, rdx);
1456 #else
1457 __ pop_l(rbx, rcx);
1458 __ push(rcx); __ push(rbx);
1459 __ push(rdx); __ push(rax);
1460 // check if y = 0
1461 __ orl(rax, rdx);
1462 __ jump_cc(Assembler::zero,
1463 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1464 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1465 __ addptr(rsp, 4 * wordSize);
1466 #endif
1467 }
1468
1469 void TemplateTable::lshl() {
1470 transition(itos, ltos);
1471 __ movl(rcx, rax); // get shift count
1472 #ifdef _LP64
1473 __ pop_l(rax); // get shift value
1474 __ shlq(rax);
1475 #else
1476 __ pop_l(rax, rdx); // get shift value
1477 __ lshl(rdx, rax);
1478 #endif
1479 }
1480
1481 void TemplateTable::lshr() {
1482 #ifdef _LP64
1483 transition(itos, ltos);
1484 __ movl(rcx, rax); // get shift count
1485 __ pop_l(rax); // get shift value
1486 __ sarq(rax);
1487 #else
1488 transition(itos, ltos);
1489 __ mov(rcx, rax); // get shift count
1490 __ pop_l(rax, rdx); // get shift value
1491 __ lshr(rdx, rax, true);
1492 #endif
1493 }
1494
1495 void TemplateTable::lushr() {
1496 transition(itos, ltos);
1497 #ifdef _LP64
1498 __ movl(rcx, rax); // get shift count
1499 __ pop_l(rax); // get shift value
1500 __ shrq(rax);
1501 #else
1502 __ mov(rcx, rax); // get shift count
1503 __ pop_l(rax, rdx); // get shift value
1504 __ lshr(rdx, rax);
1505 #endif
1506 }
1507
1508 void TemplateTable::fop2(Operation op) {
1509 transition(ftos, ftos);
1510
1511 if (UseSSE >= 1) {
1512 switch (op) {
1513 case add:
1514 __ addss(xmm0, at_rsp());
1515 __ addptr(rsp, Interpreter::stackElementSize);
1516 break;
1517 case sub:
1518 __ movflt(xmm1, xmm0);
1519 __ pop_f(xmm0);
1520 __ subss(xmm0, xmm1);
1521 break;
1522 case mul:
1523 __ mulss(xmm0, at_rsp());
1524 __ addptr(rsp, Interpreter::stackElementSize);
1525 break;
1526 case div:
1527 __ movflt(xmm1, xmm0);
1528 __ pop_f(xmm0);
1529 __ divss(xmm0, xmm1);
1530 break;
1531 case rem:
1532 // On x86_64 platforms the SharedRuntime::frem method is called to perform the
1533 // modulo operation. The frem method calls the function
1534 // double fmod(double x, double y) in math.h. The documentation of fmod states:
1535 // "If x or y is a NaN, a NaN is returned." without specifying what type of NaN
1536 // (signalling or quiet) is returned.
1537 //
1538 // On x86_32 platforms the FPU is used to perform the modulo operation. The
1539 // reason is that on 32-bit Windows the sign of modulo operations diverges from
1540 // what is considered the standard (e.g., -0.0f % -3.14f is 0.0f (and not -0.0f).
1541 // The fprem instruction used on x86_32 is functionally equivalent to
1542 // SharedRuntime::frem in that it returns a NaN.
1543 #ifdef _LP64
1544 __ movflt(xmm1, xmm0);
1545 __ pop_f(xmm0);
1546 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::frem), 2);
1547 #else
1548 __ push_f(xmm0);
1549 __ pop_f();
1550 __ fld_s(at_rsp());
1551 __ fremr(rax);
1552 __ f2ieee();
1553 __ pop(rax); // pop second operand off the stack
1554 __ push_f();
1555 __ pop_f(xmm0);
1556 #endif
1557 break;
1558 default:
1559 ShouldNotReachHere();
1560 break;
1561 }
1562 } else {
1563 #ifdef _LP64
1564 ShouldNotReachHere();
1565 #else
1566 switch (op) {
1567 case add: __ fadd_s (at_rsp()); break;
1568 case sub: __ fsubr_s(at_rsp()); break;
1569 case mul: __ fmul_s (at_rsp()); break;
1570 case div: __ fdivr_s(at_rsp()); break;
1571 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1572 default : ShouldNotReachHere();
1573 }
1574 __ f2ieee();
1575 __ pop(rax); // pop second operand off the stack
1576 #endif // _LP64
1577 }
1578 }
1579
1580 void TemplateTable::dop2(Operation op) {
1581 transition(dtos, dtos);
1582 if (UseSSE >= 2) {
1583 switch (op) {
1584 case add:
1585 __ addsd(xmm0, at_rsp());
1586 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1587 break;
1588 case sub:
1589 __ movdbl(xmm1, xmm0);
1590 __ pop_d(xmm0);
1591 __ subsd(xmm0, xmm1);
1592 break;
1593 case mul:
1594 __ mulsd(xmm0, at_rsp());
1595 __ addptr(rsp, 2 * Interpreter::stackElementSize);
1596 break;
1597 case div:
1598 __ movdbl(xmm1, xmm0);
1599 __ pop_d(xmm0);
1600 __ divsd(xmm0, xmm1);
1601 break;
1602 case rem:
1603 // Similar to fop2(), the modulo operation is performed using the
1604 // SharedRuntime::drem method (on x86_64 platforms) or using the
1605 // FPU (on x86_32 platforms) for the same reasons as mentioned in fop2().
1606 #ifdef _LP64
1607 __ movdbl(xmm1, xmm0);
1608 __ pop_d(xmm0);
1609 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::drem), 2);
1610 #else
1611 __ push_d(xmm0);
1612 __ pop_d();
1613 __ fld_d(at_rsp());
1614 __ fremr(rax);
1615 __ d2ieee();
1616 __ pop(rax);
1617 __ pop(rdx);
1618 __ push_d();
1619 __ pop_d(xmm0);
1620 #endif
1621 break;
1622 default:
1623 ShouldNotReachHere();
1624 break;
1625 }
1626 } else {
1627 #ifdef _LP64
1628 ShouldNotReachHere();
1629 #else
1630 switch (op) {
1631 case add: __ fadd_d (at_rsp()); break;
1632 case sub: __ fsubr_d(at_rsp()); break;
1633 case mul: {
1634 Label L_strict;
1635 Label L_join;
1636 const Address access_flags (rcx, Method::access_flags_offset());
1637 __ get_method(rcx);
1638 __ movl(rcx, access_flags);
1639 __ testl(rcx, JVM_ACC_STRICT);
1640 __ jccb(Assembler::notZero, L_strict);
1641 __ fmul_d (at_rsp());
1642 __ jmpb(L_join);
1643 __ bind(L_strict);
1644 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1645 __ fmulp();
1646 __ fmul_d (at_rsp());
1647 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1648 __ fmulp();
1649 __ bind(L_join);
1650 break;
1651 }
1652 case div: {
1653 Label L_strict;
1654 Label L_join;
1655 const Address access_flags (rcx, Method::access_flags_offset());
1656 __ get_method(rcx);
1657 __ movl(rcx, access_flags);
1658 __ testl(rcx, JVM_ACC_STRICT);
1659 __ jccb(Assembler::notZero, L_strict);
1660 __ fdivr_d(at_rsp());
1661 __ jmp(L_join);
1662 __ bind(L_strict);
1663 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1664 __ fmul_d (at_rsp());
1665 __ fdivrp();
1666 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1667 __ fmulp();
1668 __ bind(L_join);
1669 break;
1670 }
1671 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1672 default : ShouldNotReachHere();
1673 }
1674 __ d2ieee();
1675 // Pop double precision number from rsp.
1676 __ pop(rax);
1677 __ pop(rdx);
1678 #endif
1679 }
1680 }
1681
1682 void TemplateTable::ineg() {
1683 transition(itos, itos);
1684 __ negl(rax);
1685 }
1686
1687 void TemplateTable::lneg() {
1688 transition(ltos, ltos);
1689 LP64_ONLY(__ negq(rax));
1690 NOT_LP64(__ lneg(rdx, rax));
1691 }
1692
1693 // Note: 'double' and 'long long' have 32-bits alignment on x86.
1694 static jlong* double_quadword(jlong *adr, jlong lo, jlong hi) {
1695 // Use the expression (adr)&(~0xF) to provide 128-bits aligned address
1696 // of 128-bits operands for SSE instructions.
1697 jlong *operand = (jlong*)(((intptr_t)adr)&((intptr_t)(~0xF)));
1698 // Store the value to a 128-bits operand.
1699 operand[0] = lo;
1700 operand[1] = hi;
1701 return operand;
1702 }
1703
1704 // Buffer for 128-bits masks used by SSE instructions.
1705 static jlong float_signflip_pool[2*2];
1706 static jlong double_signflip_pool[2*2];
1707
1708 void TemplateTable::fneg() {
1709 transition(ftos, ftos);
1710 if (UseSSE >= 1) {
1711 static jlong *float_signflip = double_quadword(&float_signflip_pool[1], CONST64(0x8000000080000000), CONST64(0x8000000080000000));
1712 __ xorps(xmm0, ExternalAddress((address) float_signflip));
1713 } else {
1714 LP64_ONLY(ShouldNotReachHere());
1715 NOT_LP64(__ fchs());
1716 }
1717 }
1718
1719 void TemplateTable::dneg() {
1720 transition(dtos, dtos);
1721 if (UseSSE >= 2) {
1722 static jlong *double_signflip =
1723 double_quadword(&double_signflip_pool[1], CONST64(0x8000000000000000), CONST64(0x8000000000000000));
1724 __ xorpd(xmm0, ExternalAddress((address) double_signflip));
1725 } else {
1726 #ifdef _LP64
1727 ShouldNotReachHere();
1728 #else
1729 __ fchs();
1730 #endif
1731 }
1732 }
1733
1734 void TemplateTable::iinc() {
1735 transition(vtos, vtos);
1736 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1737 locals_index(rbx);
1738 __ addl(iaddress(rbx), rdx);
1739 }
1740
1741 void TemplateTable::wide_iinc() {
1742 transition(vtos, vtos);
1743 __ movl(rdx, at_bcp(4)); // get constant
1744 locals_index_wide(rbx);
1745 __ bswapl(rdx); // swap bytes & sign-extend constant
1746 __ sarl(rdx, 16);
1747 __ addl(iaddress(rbx), rdx);
1748 // Note: should probably use only one movl to get both
1749 // the index and the constant -> fix this
1750 }
1751
1752 void TemplateTable::convert() {
1753 #ifdef _LP64
1754 // Checking
1755 #ifdef ASSERT
1756 {
1757 TosState tos_in = ilgl;
1758 TosState tos_out = ilgl;
1759 switch (bytecode()) {
1760 case Bytecodes::_i2l: // fall through
1761 case Bytecodes::_i2f: // fall through
1762 case Bytecodes::_i2d: // fall through
1763 case Bytecodes::_i2b: // fall through
1764 case Bytecodes::_i2c: // fall through
1765 case Bytecodes::_i2s: tos_in = itos; break;
1766 case Bytecodes::_l2i: // fall through
1767 case Bytecodes::_l2f: // fall through
1768 case Bytecodes::_l2d: tos_in = ltos; break;
1769 case Bytecodes::_f2i: // fall through
1770 case Bytecodes::_f2l: // fall through
1771 case Bytecodes::_f2d: tos_in = ftos; break;
1772 case Bytecodes::_d2i: // fall through
1773 case Bytecodes::_d2l: // fall through
1774 case Bytecodes::_d2f: tos_in = dtos; break;
1775 default : ShouldNotReachHere();
1776 }
1777 switch (bytecode()) {
1778 case Bytecodes::_l2i: // fall through
1779 case Bytecodes::_f2i: // fall through
1780 case Bytecodes::_d2i: // fall through
1781 case Bytecodes::_i2b: // fall through
1782 case Bytecodes::_i2c: // fall through
1783 case Bytecodes::_i2s: tos_out = itos; break;
1784 case Bytecodes::_i2l: // fall through
1785 case Bytecodes::_f2l: // fall through
1786 case Bytecodes::_d2l: tos_out = ltos; break;
1787 case Bytecodes::_i2f: // fall through
1788 case Bytecodes::_l2f: // fall through
1789 case Bytecodes::_d2f: tos_out = ftos; break;
1790 case Bytecodes::_i2d: // fall through
1791 case Bytecodes::_l2d: // fall through
1792 case Bytecodes::_f2d: tos_out = dtos; break;
1793 default : ShouldNotReachHere();
1794 }
1795 transition(tos_in, tos_out);
1796 }
1797 #endif // ASSERT
1798
1799 static const int64_t is_nan = 0x8000000000000000L;
1800
1801 // Conversion
1802 switch (bytecode()) {
1803 case Bytecodes::_i2l:
1804 __ movslq(rax, rax);
1805 break;
1806 case Bytecodes::_i2f:
1807 __ cvtsi2ssl(xmm0, rax);
1808 break;
1809 case Bytecodes::_i2d:
1810 __ cvtsi2sdl(xmm0, rax);
1811 break;
1812 case Bytecodes::_i2b:
1813 __ movsbl(rax, rax);
1814 break;
1815 case Bytecodes::_i2c:
1816 __ movzwl(rax, rax);
1817 break;
1818 case Bytecodes::_i2s:
1819 __ movswl(rax, rax);
1820 break;
1821 case Bytecodes::_l2i:
1822 __ movl(rax, rax);
1823 break;
1824 case Bytecodes::_l2f:
1825 __ cvtsi2ssq(xmm0, rax);
1826 break;
1827 case Bytecodes::_l2d:
1828 __ cvtsi2sdq(xmm0, rax);
1829 break;
1830 case Bytecodes::_f2i:
1831 {
1832 Label L;
1833 __ cvttss2sil(rax, xmm0);
1834 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1835 __ jcc(Assembler::notEqual, L);
1836 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1837 __ bind(L);
1838 }
1839 break;
1840 case Bytecodes::_f2l:
1841 {
1842 Label L;
1843 __ cvttss2siq(rax, xmm0);
1844 // NaN or overflow/underflow?
1845 __ cmp64(rax, ExternalAddress((address) &is_nan));
1846 __ jcc(Assembler::notEqual, L);
1847 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1848 __ bind(L);
1849 }
1850 break;
1851 case Bytecodes::_f2d:
1852 __ cvtss2sd(xmm0, xmm0);
1853 break;
1854 case Bytecodes::_d2i:
1855 {
1856 Label L;
1857 __ cvttsd2sil(rax, xmm0);
1858 __ cmpl(rax, 0x80000000); // NaN or overflow/underflow?
1859 __ jcc(Assembler::notEqual, L);
1860 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 1);
1861 __ bind(L);
1862 }
1863 break;
1864 case Bytecodes::_d2l:
1865 {
1866 Label L;
1867 __ cvttsd2siq(rax, xmm0);
1868 // NaN or overflow/underflow?
1869 __ cmp64(rax, ExternalAddress((address) &is_nan));
1870 __ jcc(Assembler::notEqual, L);
1871 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 1);
1872 __ bind(L);
1873 }
1874 break;
1875 case Bytecodes::_d2f:
1876 __ cvtsd2ss(xmm0, xmm0);
1877 break;
1878 default:
1879 ShouldNotReachHere();
1880 }
1881 #else
1882 // Checking
1883 #ifdef ASSERT
1884 { TosState tos_in = ilgl;
1885 TosState tos_out = ilgl;
1886 switch (bytecode()) {
1887 case Bytecodes::_i2l: // fall through
1888 case Bytecodes::_i2f: // fall through
1889 case Bytecodes::_i2d: // fall through
1890 case Bytecodes::_i2b: // fall through
1891 case Bytecodes::_i2c: // fall through
1892 case Bytecodes::_i2s: tos_in = itos; break;
1893 case Bytecodes::_l2i: // fall through
1894 case Bytecodes::_l2f: // fall through
1895 case Bytecodes::_l2d: tos_in = ltos; break;
1896 case Bytecodes::_f2i: // fall through
1897 case Bytecodes::_f2l: // fall through
1898 case Bytecodes::_f2d: tos_in = ftos; break;
1899 case Bytecodes::_d2i: // fall through
1900 case Bytecodes::_d2l: // fall through
1901 case Bytecodes::_d2f: tos_in = dtos; break;
1902 default : ShouldNotReachHere();
1903 }
1904 switch (bytecode()) {
1905 case Bytecodes::_l2i: // fall through
1906 case Bytecodes::_f2i: // fall through
1907 case Bytecodes::_d2i: // fall through
1908 case Bytecodes::_i2b: // fall through
1909 case Bytecodes::_i2c: // fall through
1910 case Bytecodes::_i2s: tos_out = itos; break;
1911 case Bytecodes::_i2l: // fall through
1912 case Bytecodes::_f2l: // fall through
1913 case Bytecodes::_d2l: tos_out = ltos; break;
1914 case Bytecodes::_i2f: // fall through
1915 case Bytecodes::_l2f: // fall through
1916 case Bytecodes::_d2f: tos_out = ftos; break;
1917 case Bytecodes::_i2d: // fall through
1918 case Bytecodes::_l2d: // fall through
1919 case Bytecodes::_f2d: tos_out = dtos; break;
1920 default : ShouldNotReachHere();
1921 }
1922 transition(tos_in, tos_out);
1923 }
1924 #endif // ASSERT
1925
1926 // Conversion
1927 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1928 switch (bytecode()) {
1929 case Bytecodes::_i2l:
1930 __ extend_sign(rdx, rax);
1931 break;
1932 case Bytecodes::_i2f:
1933 if (UseSSE >= 1) {
1934 __ cvtsi2ssl(xmm0, rax);
1935 } else {
1936 __ push(rax); // store int on tos
1937 __ fild_s(at_rsp()); // load int to ST0
1938 __ f2ieee(); // truncate to float size
1939 __ pop(rcx); // adjust rsp
1940 }
1941 break;
1942 case Bytecodes::_i2d:
1943 if (UseSSE >= 2) {
1944 __ cvtsi2sdl(xmm0, rax);
1945 } else {
1946 __ push(rax); // add one slot for d2ieee()
1947 __ push(rax); // store int on tos
1948 __ fild_s(at_rsp()); // load int to ST0
1949 __ d2ieee(); // truncate to double size
1950 __ pop(rcx); // adjust rsp
1951 __ pop(rcx);
1952 }
1953 break;
1954 case Bytecodes::_i2b:
1955 __ shll(rax, 24); // truncate upper 24 bits
1956 __ sarl(rax, 24); // and sign-extend byte
1957 LP64_ONLY(__ movsbl(rax, rax));
1958 break;
1959 case Bytecodes::_i2c:
1960 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1961 LP64_ONLY(__ movzwl(rax, rax));
1962 break;
1963 case Bytecodes::_i2s:
1964 __ shll(rax, 16); // truncate upper 16 bits
1965 __ sarl(rax, 16); // and sign-extend short
1966 LP64_ONLY(__ movswl(rax, rax));
1967 break;
1968 case Bytecodes::_l2i:
1969 /* nothing to do */
1970 break;
1971 case Bytecodes::_l2f:
1972 // On 64-bit platforms, the cvtsi2ssq instruction is used to convert
1973 // 64-bit long values to floats. On 32-bit platforms it is not possible
1974 // to use that instruction with 64-bit operands, therefore the FPU is
1975 // used to perform the conversion.
1976 __ push(rdx); // store long on tos
1977 __ push(rax);
1978 __ fild_d(at_rsp()); // load long to ST0
1979 __ f2ieee(); // truncate to float size
1980 __ pop(rcx); // adjust rsp
1981 __ pop(rcx);
1982 if (UseSSE >= 1) {
1983 __ push_f();
1984 __ pop_f(xmm0);
1985 }
1986 break;
1987 case Bytecodes::_l2d:
1988 // On 32-bit platforms the FPU is used for conversion because on
1989 // 32-bit platforms it is not not possible to use the cvtsi2sdq
1990 // instruction with 64-bit operands.
1991 __ push(rdx); // store long on tos
1992 __ push(rax);
1993 __ fild_d(at_rsp()); // load long to ST0
1994 __ d2ieee(); // truncate to double size
1995 __ pop(rcx); // adjust rsp
1996 __ pop(rcx);
1997 if (UseSSE >= 2) {
1998 __ push_d();
1999 __ pop_d(xmm0);
2000 }
2001 break;
2002 case Bytecodes::_f2i:
2003 // SharedRuntime::f2i does not differentiate between sNaNs and qNaNs
2004 // as it returns 0 for any NaN.
2005 if (UseSSE >= 1) {
2006 __ push_f(xmm0);
2007 } else {
2008 __ push(rcx); // reserve space for argument
2009 __ fstp_s(at_rsp()); // pass float argument on stack
2010 }
2011 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
2012 break;
2013 case Bytecodes::_f2l:
2014 // SharedRuntime::f2l does not differentiate between sNaNs and qNaNs
2015 // as it returns 0 for any NaN.
2016 if (UseSSE >= 1) {
2017 __ push_f(xmm0);
2018 } else {
2019 __ push(rcx); // reserve space for argument
2020 __ fstp_s(at_rsp()); // pass float argument on stack
2021 }
2022 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
2023 break;
2024 case Bytecodes::_f2d:
2025 if (UseSSE < 1) {
2026 /* nothing to do */
2027 } else if (UseSSE == 1) {
2028 __ push_f(xmm0);
2029 __ pop_f();
2030 } else { // UseSSE >= 2
2031 __ cvtss2sd(xmm0, xmm0);
2032 }
2033 break;
2034 case Bytecodes::_d2i:
2035 if (UseSSE >= 2) {
2036 __ push_d(xmm0);
2037 } else {
2038 __ push(rcx); // reserve space for argument
2039 __ push(rcx);
2040 __ fstp_d(at_rsp()); // pass double argument on stack
2041 }
2042 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
2043 break;
2044 case Bytecodes::_d2l:
2045 if (UseSSE >= 2) {
2046 __ push_d(xmm0);
2047 } else {
2048 __ push(rcx); // reserve space for argument
2049 __ push(rcx);
2050 __ fstp_d(at_rsp()); // pass double argument on stack
2051 }
2052 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
2053 break;
2054 case Bytecodes::_d2f:
2055 if (UseSSE <= 1) {
2056 __ push(rcx); // reserve space for f2ieee()
2057 __ f2ieee(); // truncate to float size
2058 __ pop(rcx); // adjust rsp
2059 if (UseSSE == 1) {
2060 // The cvtsd2ss instruction is not available if UseSSE==1, therefore
2061 // the conversion is performed using the FPU in this case.
2062 __ push_f();
2063 __ pop_f(xmm0);
2064 }
2065 } else { // UseSSE >= 2
2066 __ cvtsd2ss(xmm0, xmm0);
2067 }
2068 break;
2069 default :
2070 ShouldNotReachHere();
2071 }
2072 #endif
2073 }
2074
2075 void TemplateTable::lcmp() {
2076 transition(ltos, itos);
2077 #ifdef _LP64
2078 Label done;
2079 __ pop_l(rdx);
2080 __ cmpq(rdx, rax);
2081 __ movl(rax, -1);
2082 __ jccb(Assembler::less, done);
2083 __ setb(Assembler::notEqual, rax);
2084 __ movzbl(rax, rax);
2085 __ bind(done);
2086 #else
2087
2088 // y = rdx:rax
2089 __ pop_l(rbx, rcx); // get x = rcx:rbx
2090 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
2091 __ mov(rax, rcx);
2092 #endif
2093 }
2094
2095 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
2096 if ((is_float && UseSSE >= 1) ||
2097 (!is_float && UseSSE >= 2)) {
2098 Label done;
2099 if (is_float) {
2100 // XXX get rid of pop here, use ... reg, mem32
2101 __ pop_f(xmm1);
2102 __ ucomiss(xmm1, xmm0);
2103 } else {
2104 // XXX get rid of pop here, use ... reg, mem64
2105 __ pop_d(xmm1);
2106 __ ucomisd(xmm1, xmm0);
2107 }
2108 if (unordered_result < 0) {
2109 __ movl(rax, -1);
2110 __ jccb(Assembler::parity, done);
2111 __ jccb(Assembler::below, done);
2112 __ setb(Assembler::notEqual, rdx);
2113 __ movzbl(rax, rdx);
2114 } else {
2115 __ movl(rax, 1);
2116 __ jccb(Assembler::parity, done);
2117 __ jccb(Assembler::above, done);
2118 __ movl(rax, 0);
2119 __ jccb(Assembler::equal, done);
2120 __ decrementl(rax);
2121 }
2122 __ bind(done);
2123 } else {
2124 #ifdef _LP64
2125 ShouldNotReachHere();
2126 #else
2127 if (is_float) {
2128 __ fld_s(at_rsp());
2129 } else {
2130 __ fld_d(at_rsp());
2131 __ pop(rdx);
2132 }
2133 __ pop(rcx);
2134 __ fcmp2int(rax, unordered_result < 0);
2135 #endif // _LP64
2136 }
2137 }
2138
2139 void TemplateTable::branch(bool is_jsr, bool is_wide) {
2140 __ get_method(rcx); // rcx holds method
2141 __ profile_taken_branch(rax, rbx); // rax holds updated MDP, rbx
2142 // holds bumped taken count
2143
2144 const ByteSize be_offset = MethodCounters::backedge_counter_offset() +
2145 InvocationCounter::counter_offset();
2146 const ByteSize inv_offset = MethodCounters::invocation_counter_offset() +
2147 InvocationCounter::counter_offset();
2148
2149 // Load up edx with the branch displacement
2150 if (is_wide) {
2151 __ movl(rdx, at_bcp(1));
2152 } else {
2153 __ load_signed_short(rdx, at_bcp(1));
2154 }
2155 __ bswapl(rdx);
2156
2157 if (!is_wide) {
2158 __ sarl(rdx, 16);
2159 }
2160 LP64_ONLY(__ movl2ptr(rdx, rdx));
2161
2162 // Handle all the JSR stuff here, then exit.
2163 // It's much shorter and cleaner than intermingling with the non-JSR
2164 // normal-branch stuff occurring below.
2165 if (is_jsr) {
2166 // Pre-load the next target bytecode into rbx
2167 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1, 0));
2168
2169 // compute return address as bci in rax
2170 __ lea(rax, at_bcp((is_wide ? 5 : 3) -
2171 in_bytes(ConstMethod::codes_offset())));
2172 __ subptr(rax, Address(rcx, Method::const_offset()));
2173 // Adjust the bcp in r13 by the displacement in rdx
2174 __ addptr(rbcp, rdx);
2175 // jsr returns atos that is not an oop
2176 __ push_i(rax);
2177 __ dispatch_only(vtos, true);
2178 return;
2179 }
2180
2181 // Normal (non-jsr) branch handling
2182
2183 // Adjust the bcp in r13 by the displacement in rdx
2184 __ addptr(rbcp, rdx);
2185
2186 assert(UseLoopCounter || !UseOnStackReplacement,
2187 "on-stack-replacement requires loop counters");
2188 Label backedge_counter_overflow;
2189 Label profile_method;
2190 Label dispatch;
2191 if (UseLoopCounter) {
2192 // increment backedge counter for backward branches
2193 // rax: MDO
2194 // rbx: MDO bumped taken-count
2195 // rcx: method
2196 // rdx: target offset
2197 // r13: target bcp
2198 // r14: locals pointer
2199 __ testl(rdx, rdx); // check if forward or backward branch
2200 __ jcc(Assembler::positive, dispatch); // count only if backward branch
2201
2202 // check if MethodCounters exists
2203 Label has_counters;
2204 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2205 __ testptr(rax, rax);
2206 __ jcc(Assembler::notZero, has_counters);
2207 __ push(rdx);
2208 __ push(rcx);
2209 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::build_method_counters),
2210 rcx);
2211 __ pop(rcx);
2212 __ pop(rdx);
2213 __ movptr(rax, Address(rcx, Method::method_counters_offset()));
2214 __ testptr(rax, rax);
2215 __ jcc(Assembler::zero, dispatch);
2216 __ bind(has_counters);
2217
2218 if (TieredCompilation) {
2219 Label no_mdo;
2220 int increment = InvocationCounter::count_increment;
2221 if (ProfileInterpreter) {
2222 // Are we profiling?
2223 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
2224 __ testptr(rbx, rbx);
2225 __ jccb(Assembler::zero, no_mdo);
2226 // Increment the MDO backedge counter
2227 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
2228 in_bytes(InvocationCounter::counter_offset()));
2229 const Address mask(rbx, in_bytes(MethodData::backedge_mask_offset()));
2230 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask, rax, false, Assembler::zero,
2231 UseOnStackReplacement ? &backedge_counter_overflow : NULL);
2232 __ jmp(dispatch);
2233 }
2234 __ bind(no_mdo);
2235 // Increment backedge counter in MethodCounters*
2236 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2237 const Address mask(rcx, in_bytes(MethodCounters::backedge_mask_offset()));
2238 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
2239 rax, false, Assembler::zero,
2240 UseOnStackReplacement ? &backedge_counter_overflow : NULL);
2241 } else { // not TieredCompilation
2242 // increment counter
2243 __ movptr(rcx, Address(rcx, Method::method_counters_offset()));
2244 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
2245 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
2246 __ movl(Address(rcx, be_offset), rax); // store counter
2247
2248 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
2249
2250 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
2251 __ addl(rax, Address(rcx, be_offset)); // add both counters
2252
2253 if (ProfileInterpreter) {
2254 // Test to see if we should create a method data oop
2255 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_profile_limit_offset())));
2256 __ jcc(Assembler::less, dispatch);
2257
2258 // if no method data exists, go to profile method
2259 __ test_method_data_pointer(rax, profile_method);
2260
2261 if (UseOnStackReplacement) {
2262 // check for overflow against rbx which is the MDO taken count
2263 __ cmp32(rbx, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2264 __ jcc(Assembler::below, dispatch);
2265
2266 // When ProfileInterpreter is on, the backedge_count comes
2267 // from the MethodData*, which value does not get reset on
2268 // the call to frequency_counter_overflow(). To avoid
2269 // excessive calls to the overflow routine while the method is
2270 // being compiled, add a second test to make sure the overflow
2271 // function is called only once every overflow_frequency.
2272 const int overflow_frequency = 1024;
2273 __ andl(rbx, overflow_frequency - 1);
2274 __ jcc(Assembler::zero, backedge_counter_overflow);
2275
2276 }
2277 } else {
2278 if (UseOnStackReplacement) {
2279 // check for overflow against rax, which is the sum of the
2280 // counters
2281 __ cmp32(rax, Address(rcx, in_bytes(MethodCounters::interpreter_backward_branch_limit_offset())));
2282 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
2283
2284 }
2285 }
2286 }
2287 __ bind(dispatch);
2288 }
2289
2290 // Pre-load the next target bytecode into rbx
2291 __ load_unsigned_byte(rbx, Address(rbcp, 0));
2292
2293 // continue with the bytecode @ target
2294 // rax: return bci for jsr's, unused otherwise
2295 // rbx: target bytecode
2296 // r13: target bcp
2297 __ dispatch_only(vtos, true);
2298
2299 if (UseLoopCounter) {
2300 if (ProfileInterpreter) {
2301 // Out-of-line code to allocate method data oop.
2302 __ bind(profile_method);
2303 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
2304 __ set_method_data_pointer_for_bcp();
2305 __ jmp(dispatch);
2306 }
2307
2308 if (UseOnStackReplacement) {
2309 // invocation counter overflow
2310 __ bind(backedge_counter_overflow);
2311 __ negptr(rdx);
2312 __ addptr(rdx, rbcp); // branch bcp
2313 // IcoResult frequency_counter_overflow([JavaThread*], address branch_bcp)
2314 __ call_VM(noreg,
2315 CAST_FROM_FN_PTR(address,
2316 InterpreterRuntime::frequency_counter_overflow),
2317 rdx);
2318
2319 // rax: osr nmethod (osr ok) or NULL (osr not possible)
2320 // rdx: scratch
2321 // r14: locals pointer
2322 // r13: bcp
2323 __ testptr(rax, rax); // test result
2324 __ jcc(Assembler::zero, dispatch); // no osr if null
2325 // nmethod may have been invalidated (VM may block upon call_VM return)
2326 __ cmpb(Address(rax, nmethod::state_offset()), nmethod::in_use);
2327 __ jcc(Assembler::notEqual, dispatch);
2328
2329 // We have the address of an on stack replacement routine in rax.
2330 // In preparation of invoking it, first we must migrate the locals
2331 // and monitors from off the interpreter frame on the stack.
2332 // Ensure to save the osr nmethod over the migration call,
2333 // it will be preserved in rbx.
2334 __ mov(rbx, rax);
2335
2336 NOT_LP64(__ get_thread(rcx));
2337
2338 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
2339
2340 // rax is OSR buffer, move it to expected parameter location
2341 LP64_ONLY(__ mov(j_rarg0, rax));
2342 NOT_LP64(__ mov(rcx, rax));
2343 // We use j_rarg definitions here so that registers don't conflict as parameter
2344 // registers change across platforms as we are in the midst of a calling
2345 // sequence to the OSR nmethod and we don't want collision. These are NOT parameters.
2346
2347 const Register retaddr = LP64_ONLY(j_rarg2) NOT_LP64(rdi);
2348 const Register sender_sp = LP64_ONLY(j_rarg1) NOT_LP64(rdx);
2349
2350 // pop the interpreter frame
2351 __ movptr(sender_sp, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
2352 __ leave(); // remove frame anchor
2353 __ pop(retaddr); // get return address
2354 __ mov(rsp, sender_sp); // set sp to sender sp
2355 // Ensure compiled code always sees stack at proper alignment
2356 __ andptr(rsp, -(StackAlignmentInBytes));
2357
2358 // unlike x86 we need no specialized return from compiled code
2359 // to the interpreter or the call stub.
2360
2361 // push the return address
2362 __ push(retaddr);
2363
2364 // and begin the OSR nmethod
2365 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
2366 }
2367 }
2368 }
2369
2370 void TemplateTable::if_0cmp(Condition cc) {
2371 transition(itos, vtos);
2372 // assume branch is more often taken than not (loops use backward branches)
2373 Label not_taken;
2374 __ testl(rax, rax);
2375 __ jcc(j_not(cc), not_taken);
2376 branch(false, false);
2377 __ bind(not_taken);
2378 __ profile_not_taken_branch(rax);
2379 }
2380
2381 void TemplateTable::if_icmp(Condition cc) {
2382 transition(itos, vtos);
2383 // assume branch is more often taken than not (loops use backward branches)
2384 Label not_taken;
2385 __ pop_i(rdx);
2386 __ cmpl(rdx, rax);
2387 __ jcc(j_not(cc), not_taken);
2388 branch(false, false);
2389 __ bind(not_taken);
2390 __ profile_not_taken_branch(rax);
2391 }
2392
2393 void TemplateTable::if_nullcmp(Condition cc) {
2394 transition(atos, vtos);
2395 // assume branch is more often taken than not (loops use backward branches)
2396 Label not_taken;
2397 __ testptr(rax, rax);
2398 __ jcc(j_not(cc), not_taken);
2399 branch(false, false);
2400 __ bind(not_taken);
2401 __ profile_not_taken_branch(rax);
2402 }
2403
2404 void TemplateTable::if_acmp(Condition cc) {
2405 transition(atos, vtos);
2406 // assume branch is more often taken than not (loops use backward branches)
2407 Label not_taken;
2408 __ pop_ptr(rdx);
2409 __ cmpoop(rdx, rax);
2410 __ jcc(j_not(cc), not_taken);
2411 branch(false, false);
2412 __ bind(not_taken);
2413 __ profile_not_taken_branch(rax);
2414 }
2415
2416 void TemplateTable::ret() {
2417 transition(vtos, vtos);
2418 locals_index(rbx);
2419 LP64_ONLY(__ movslq(rbx, iaddress(rbx))); // get return bci, compute return bcp
2420 NOT_LP64(__ movptr(rbx, iaddress(rbx)));
2421 __ profile_ret(rbx, rcx);
2422 __ get_method(rax);
2423 __ movptr(rbcp, Address(rax, Method::const_offset()));
2424 __ lea(rbcp, Address(rbcp, rbx, Address::times_1,
2425 ConstMethod::codes_offset()));
2426 __ dispatch_next(vtos, 0, true);
2427 }
2428
2429 void TemplateTable::wide_ret() {
2430 transition(vtos, vtos);
2431 locals_index_wide(rbx);
2432 __ movptr(rbx, aaddress(rbx)); // get return bci, compute return bcp
2433 __ profile_ret(rbx, rcx);
2434 __ get_method(rax);
2435 __ movptr(rbcp, Address(rax, Method::const_offset()));
2436 __ lea(rbcp, Address(rbcp, rbx, Address::times_1, ConstMethod::codes_offset()));
2437 __ dispatch_next(vtos, 0, true);
2438 }
2439
2440 void TemplateTable::tableswitch() {
2441 Label default_case, continue_execution;
2442 transition(itos, vtos);
2443
2444 // align r13/rsi
2445 __ lea(rbx, at_bcp(BytesPerInt));
2446 __ andptr(rbx, -BytesPerInt);
2447 // load lo & hi
2448 __ movl(rcx, Address(rbx, BytesPerInt));
2449 __ movl(rdx, Address(rbx, 2 * BytesPerInt));
2450 __ bswapl(rcx);
2451 __ bswapl(rdx);
2452 // check against lo & hi
2453 __ cmpl(rax, rcx);
2454 __ jcc(Assembler::less, default_case);
2455 __ cmpl(rax, rdx);
2456 __ jcc(Assembler::greater, default_case);
2457 // lookup dispatch offset
2458 __ subl(rax, rcx);
2459 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
2460 __ profile_switch_case(rax, rbx, rcx);
2461 // continue execution
2462 __ bind(continue_execution);
2463 __ bswapl(rdx);
2464 LP64_ONLY(__ movl2ptr(rdx, rdx));
2465 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2466 __ addptr(rbcp, rdx);
2467 __ dispatch_only(vtos, true);
2468 // handle default
2469 __ bind(default_case);
2470 __ profile_switch_default(rax);
2471 __ movl(rdx, Address(rbx, 0));
2472 __ jmp(continue_execution);
2473 }
2474
2475 void TemplateTable::lookupswitch() {
2476 transition(itos, itos);
2477 __ stop("lookupswitch bytecode should have been rewritten");
2478 }
2479
2480 void TemplateTable::fast_linearswitch() {
2481 transition(itos, vtos);
2482 Label loop_entry, loop, found, continue_execution;
2483 // bswap rax so we can avoid bswapping the table entries
2484 __ bswapl(rax);
2485 // align r13
2486 __ lea(rbx, at_bcp(BytesPerInt)); // btw: should be able to get rid of
2487 // this instruction (change offsets
2488 // below)
2489 __ andptr(rbx, -BytesPerInt);
2490 // set counter
2491 __ movl(rcx, Address(rbx, BytesPerInt));
2492 __ bswapl(rcx);
2493 __ jmpb(loop_entry);
2494 // table search
2495 __ bind(loop);
2496 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * BytesPerInt));
2497 __ jcc(Assembler::equal, found);
2498 __ bind(loop_entry);
2499 __ decrementl(rcx);
2500 __ jcc(Assembler::greaterEqual, loop);
2501 // default case
2502 __ profile_switch_default(rax);
2503 __ movl(rdx, Address(rbx, 0));
2504 __ jmp(continue_execution);
2505 // entry found -> get offset
2506 __ bind(found);
2507 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * BytesPerInt));
2508 __ profile_switch_case(rcx, rax, rbx);
2509 // continue execution
2510 __ bind(continue_execution);
2511 __ bswapl(rdx);
2512 __ movl2ptr(rdx, rdx);
2513 __ load_unsigned_byte(rbx, Address(rbcp, rdx, Address::times_1));
2514 __ addptr(rbcp, rdx);
2515 __ dispatch_only(vtos, true);
2516 }
2517
2518 void TemplateTable::fast_binaryswitch() {
2519 transition(itos, vtos);
2520 // Implementation using the following core algorithm:
2521 //
2522 // int binary_search(int key, LookupswitchPair* array, int n) {
2523 // // Binary search according to "Methodik des Programmierens" by
2524 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
2525 // int i = 0;
2526 // int j = n;
2527 // while (i+1 < j) {
2528 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
2529 // // with Q: for all i: 0 <= i < n: key < a[i]
2530 // // where a stands for the array and assuming that the (inexisting)
2531 // // element a[n] is infinitely big.
2532 // int h = (i + j) >> 1;
2533 // // i < h < j
2534 // if (key < array[h].fast_match()) {
2535 // j = h;
2536 // } else {
2537 // i = h;
2538 // }
2539 // }
2540 // // R: a[i] <= key < a[i+1] or Q
2541 // // (i.e., if key is within array, i is the correct index)
2542 // return i;
2543 // }
2544
2545 // Register allocation
2546 const Register key = rax; // already set (tosca)
2547 const Register array = rbx;
2548 const Register i = rcx;
2549 const Register j = rdx;
2550 const Register h = rdi;
2551 const Register temp = rsi;
2552
2553 // Find array start
2554 NOT_LP64(__ save_bcp());
2555
2556 __ lea(array, at_bcp(3 * BytesPerInt)); // btw: should be able to
2557 // get rid of this
2558 // instruction (change
2559 // offsets below)
2560 __ andptr(array, -BytesPerInt);
2561
2562 // Initialize i & j
2563 __ xorl(i, i); // i = 0;
2564 __ movl(j, Address(array, -BytesPerInt)); // j = length(array);
2565
2566 // Convert j into native byteordering
2567 __ bswapl(j);
2568
2569 // And start
2570 Label entry;
2571 __ jmp(entry);
2572
2573 // binary search loop
2574 {
2575 Label loop;
2576 __ bind(loop);
2577 // int h = (i + j) >> 1;
2578 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
2579 __ sarl(h, 1); // h = (i + j) >> 1;
2580 // if (key < array[h].fast_match()) {
2581 // j = h;
2582 // } else {
2583 // i = h;
2584 // }
2585 // Convert array[h].match to native byte-ordering before compare
2586 __ movl(temp, Address(array, h, Address::times_8));
2587 __ bswapl(temp);
2588 __ cmpl(key, temp);
2589 // j = h if (key < array[h].fast_match())
2590 __ cmov32(Assembler::less, j, h);
2591 // i = h if (key >= array[h].fast_match())
2592 __ cmov32(Assembler::greaterEqual, i, h);
2593 // while (i+1 < j)
2594 __ bind(entry);
2595 __ leal(h, Address(i, 1)); // i+1
2596 __ cmpl(h, j); // i+1 < j
2597 __ jcc(Assembler::less, loop);
2598 }
2599
2600 // end of binary search, result index is i (must check again!)
2601 Label default_case;
2602 // Convert array[i].match to native byte-ordering before compare
2603 __ movl(temp, Address(array, i, Address::times_8));
2604 __ bswapl(temp);
2605 __ cmpl(key, temp);
2606 __ jcc(Assembler::notEqual, default_case);
2607
2608 // entry found -> j = offset
2609 __ movl(j , Address(array, i, Address::times_8, BytesPerInt));
2610 __ profile_switch_case(i, key, array);
2611 __ bswapl(j);
2612 LP64_ONLY(__ movslq(j, j));
2613
2614 NOT_LP64(__ restore_bcp());
2615 NOT_LP64(__ restore_locals()); // restore rdi
2616
2617 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2618 __ addptr(rbcp, j);
2619 __ dispatch_only(vtos, true);
2620
2621 // default case -> j = default offset
2622 __ bind(default_case);
2623 __ profile_switch_default(i);
2624 __ movl(j, Address(array, -2 * BytesPerInt));
2625 __ bswapl(j);
2626 LP64_ONLY(__ movslq(j, j));
2627
2628 NOT_LP64(__ restore_bcp());
2629 NOT_LP64(__ restore_locals());
2630
2631 __ load_unsigned_byte(rbx, Address(rbcp, j, Address::times_1));
2632 __ addptr(rbcp, j);
2633 __ dispatch_only(vtos, true);
2634 }
2635
2636 void TemplateTable::_return(TosState state) {
2637 transition(state, state);
2638
2639 assert(_desc->calls_vm(),
2640 "inconsistent calls_vm information"); // call in remove_activation
2641
2642 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2643 assert(state == vtos, "only valid state");
2644 Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rax);
2645 __ movptr(robj, aaddress(0));
2646 __ load_klass(rdi, robj);
2647 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2648 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2649 Label skip_register_finalizer;
2650 __ jcc(Assembler::zero, skip_register_finalizer);
2651
2652 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), robj);
2653
2654 __ bind(skip_register_finalizer);
2655 }
2656
2657 if (SafepointMechanism::uses_thread_local_poll() && _desc->bytecode() != Bytecodes::_return_register_finalizer) {
2658 Label no_safepoint;
2659 NOT_PRODUCT(__ block_comment("Thread-local Safepoint poll"));
2660 #ifdef _LP64
2661 __ testb(Address(r15_thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2662 #else
2663 const Register thread = rdi;
2664 __ get_thread(thread);
2665 __ testb(Address(thread, Thread::polling_page_offset()), SafepointMechanism::poll_bit());
2666 #endif
2667 __ jcc(Assembler::zero, no_safepoint);
2668 __ push(state);
2669 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
2670 InterpreterRuntime::at_safepoint));
2671 __ pop(state);
2672 __ bind(no_safepoint);
2673 }
2674
2675 // Narrow result if state is itos but result type is smaller.
2676 // Need to narrow in the return bytecode rather than in generate_return_entry
2677 // since compiled code callers expect the result to already be narrowed.
2678 if (state == itos) {
2679 __ narrow(rax);
2680 }
2681 __ remove_activation(state, rbcp);
2682
2683 __ jmp(rbcp);
2684 }
2685
2686 // ----------------------------------------------------------------------------
2687 // Volatile variables demand their effects be made known to all CPU's
2688 // in order. Store buffers on most chips allow reads & writes to
2689 // reorder; the JMM's ReadAfterWrite.java test fails in -Xint mode
2690 // without some kind of memory barrier (i.e., it's not sufficient that
2691 // the interpreter does not reorder volatile references, the hardware
2692 // also must not reorder them).
2693 //
2694 // According to the new Java Memory Model (JMM):
2695 // (1) All volatiles are serialized wrt to each other. ALSO reads &
2696 // writes act as aquire & release, so:
2697 // (2) A read cannot let unrelated NON-volatile memory refs that
2698 // happen after the read float up to before the read. It's OK for
2699 // non-volatile memory refs that happen before the volatile read to
2700 // float down below it.
2701 // (3) Similar a volatile write cannot let unrelated NON-volatile
2702 // memory refs that happen BEFORE the write float down to after the
2703 // write. It's OK for non-volatile memory refs that happen after the
2704 // volatile write to float up before it.
2705 //
2706 // We only put in barriers around volatile refs (they are expensive),
2707 // not _between_ memory refs (that would require us to track the
2708 // flavor of the previous memory refs). Requirements (2) and (3)
2709 // require some barriers before volatile stores and after volatile
2710 // loads. These nearly cover requirement (1) but miss the
2711 // volatile-store-volatile-load case. This final case is placed after
2712 // volatile-stores although it could just as well go before
2713 // volatile-loads.
2714
2715 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2716 // Helper function to insert a is-volatile test and memory barrier
2717 __ membar(order_constraint);
2718 }
2719
2720 void TemplateTable::resolve_cache_and_index(int byte_no,
2721 Register Rcache,
2722 Register index,
2723 size_t index_size) {
2724 const Register temp = rbx;
2725 assert_different_registers(Rcache, index, temp);
2726
2727 Label resolved;
2728
2729 Bytecodes::Code code = bytecode();
2730 switch (code) {
2731 case Bytecodes::_nofast_getfield: code = Bytecodes::_getfield; break;
2732 case Bytecodes::_nofast_putfield: code = Bytecodes::_putfield; break;
2733 default: break;
2734 }
2735
2736 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2737 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2738 __ cmpl(temp, code); // have we resolved this bytecode?
2739 __ jcc(Assembler::equal, resolved);
2740
2741 // resolve first time through
2742 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_from_cache);
2743 __ movl(temp, code);
2744 __ call_VM(noreg, entry, temp);
2745 // Update registers with resolved info
2746 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2747 __ bind(resolved);
2748 }
2749
2750 // The cache and index registers must be set before call
2751 void TemplateTable::load_field_cp_cache_entry(Register obj,
2752 Register cache,
2753 Register index,
2754 Register off,
2755 Register flags,
2756 bool is_static = false) {
2757 assert_different_registers(cache, index, flags, off);
2758
2759 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2760 // Field offset
2761 __ movptr(off, Address(cache, index, Address::times_ptr,
2762 in_bytes(cp_base_offset +
2763 ConstantPoolCacheEntry::f2_offset())));
2764 // Flags
2765 __ movl(flags, Address(cache, index, Address::times_ptr,
2766 in_bytes(cp_base_offset +
2767 ConstantPoolCacheEntry::flags_offset())));
2768
2769 // klass overwrite register
2770 if (is_static) {
2771 __ movptr(obj, Address(cache, index, Address::times_ptr,
2772 in_bytes(cp_base_offset +
2773 ConstantPoolCacheEntry::f1_offset())));
2774 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2775 __ movptr(obj, Address(obj, mirror_offset));
2776 __ resolve_oop_handle(obj);
2777 }
2778 }
2779
2780 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2781 Register method,
2782 Register itable_index,
2783 Register flags,
2784 bool is_invokevirtual,
2785 bool is_invokevfinal, /*unused*/
2786 bool is_invokedynamic) {
2787 // setup registers
2788 const Register cache = rcx;
2789 const Register index = rdx;
2790 assert_different_registers(method, flags);
2791 assert_different_registers(method, cache, index);
2792 assert_different_registers(itable_index, flags);
2793 assert_different_registers(itable_index, cache, index);
2794 // determine constant pool cache field offsets
2795 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2796 const int method_offset = in_bytes(
2797 ConstantPoolCache::base_offset() +
2798 ((byte_no == f2_byte)
2799 ? ConstantPoolCacheEntry::f2_offset()
2800 : ConstantPoolCacheEntry::f1_offset()));
2801 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2802 ConstantPoolCacheEntry::flags_offset());
2803 // access constant pool cache fields
2804 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2805 ConstantPoolCacheEntry::f2_offset());
2806
2807 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2808 resolve_cache_and_index(byte_no, cache, index, index_size);
2809 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2810
2811 if (itable_index != noreg) {
2812 // pick up itable or appendix index from f2 also:
2813 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2814 }
2815 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2816 }
2817
2818 // The registers cache and index expected to be set before call.
2819 // Correct values of the cache and index registers are preserved.
2820 void TemplateTable::jvmti_post_field_access(Register cache,
2821 Register index,
2822 bool is_static,
2823 bool has_tos) {
2824 if (JvmtiExport::can_post_field_access()) {
2825 // Check to see if a field access watch has been set before we take
2826 // the time to call into the VM.
2827 Label L1;
2828 assert_different_registers(cache, index, rax);
2829 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2830 __ testl(rax,rax);
2831 __ jcc(Assembler::zero, L1);
2832
2833 // cache entry pointer
2834 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2835 __ shll(index, LogBytesPerWord);
2836 __ addptr(cache, index);
2837 if (is_static) {
2838 __ xorptr(rax, rax); // NULL object reference
2839 } else {
2840 __ pop(atos); // Get the object
2841 __ verify_oop(rax);
2842 __ push(atos); // Restore stack state
2843 }
2844 // rax,: object pointer or NULL
2845 // cache: cache entry pointer
2846 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2847 rax, cache);
2848 __ get_cache_and_index_at_bcp(cache, index, 1);
2849 __ bind(L1);
2850 }
2851 }
2852
2853 void TemplateTable::pop_and_check_object(Register r) {
2854 __ pop_ptr(r);
2855 __ null_check(r); // for field access must check obj.
2856 __ verify_oop(r);
2857 }
2858
2859 void TemplateTable::getfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
2860 transition(vtos, vtos);
2861
2862 const Register cache = rcx;
2863 const Register index = rdx;
2864 const Register obj = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
2865 const Register off = rbx;
2866 const Register flags = rax;
2867 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // uses same reg as obj, so don't mix them
2868
2869 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2870 jvmti_post_field_access(cache, index, is_static, false);
2871 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2872
2873 if (!is_static) pop_and_check_object(obj);
2874
2875 const Address field(obj, off, Address::times_1, 0*wordSize);
2876
2877 Label Done, notByte, notBool, notInt, notShort, notChar, notLong, notFloat, notObj;
2878
2879 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2880 // Make sure we don't need to mask edx after the above shift
2881 assert(btos == 0, "change code, btos != 0");
2882
2883 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2884
2885 __ jcc(Assembler::notZero, notByte);
2886 // btos
2887 __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
2888 __ push(btos);
2889 // Rewrite bytecode to be faster
2890 if (!is_static && rc == may_rewrite) {
2891 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2892 }
2893 __ jmp(Done);
2894
2895 __ bind(notByte);
2896 __ cmpl(flags, ztos);
2897 __ jcc(Assembler::notEqual, notBool);
2898
2899 // ztos (same code as btos)
2900 __ access_load_at(T_BOOLEAN, IN_HEAP, rax, field, noreg, noreg);
2901 __ push(ztos);
2902 // Rewrite bytecode to be faster
2903 if (!is_static && rc == may_rewrite) {
2904 // use btos rewriting, no truncating to t/f bit is needed for getfield.
2905 patch_bytecode(Bytecodes::_fast_bgetfield, bc, rbx);
2906 }
2907 __ jmp(Done);
2908
2909 __ bind(notBool);
2910 __ cmpl(flags, atos);
2911 __ jcc(Assembler::notEqual, notObj);
2912 // atos
2913 do_oop_load(_masm, field, rax);
2914 __ push(atos);
2915 if (!is_static && rc == may_rewrite) {
2916 patch_bytecode(Bytecodes::_fast_agetfield, bc, rbx);
2917 }
2918 __ jmp(Done);
2919
2920 __ bind(notObj);
2921 __ cmpl(flags, itos);
2922 __ jcc(Assembler::notEqual, notInt);
2923 // itos
2924 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
2925 __ push(itos);
2926 // Rewrite bytecode to be faster
2927 if (!is_static && rc == may_rewrite) {
2928 patch_bytecode(Bytecodes::_fast_igetfield, bc, rbx);
2929 }
2930 __ jmp(Done);
2931
2932 __ bind(notInt);
2933 __ cmpl(flags, ctos);
2934 __ jcc(Assembler::notEqual, notChar);
2935 // ctos
2936 __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
2937 __ push(ctos);
2938 // Rewrite bytecode to be faster
2939 if (!is_static && rc == may_rewrite) {
2940 patch_bytecode(Bytecodes::_fast_cgetfield, bc, rbx);
2941 }
2942 __ jmp(Done);
2943
2944 __ bind(notChar);
2945 __ cmpl(flags, stos);
2946 __ jcc(Assembler::notEqual, notShort);
2947 // stos
2948 __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
2949 __ push(stos);
2950 // Rewrite bytecode to be faster
2951 if (!is_static && rc == may_rewrite) {
2952 patch_bytecode(Bytecodes::_fast_sgetfield, bc, rbx);
2953 }
2954 __ jmp(Done);
2955
2956 __ bind(notShort);
2957 __ cmpl(flags, ltos);
2958 __ jcc(Assembler::notEqual, notLong);
2959 // ltos
2960 // Generate code as if volatile (x86_32). There just aren't enough registers to
2961 // save that information and this code is faster than the test.
2962 __ access_load_at(T_LONG, IN_HEAP | MO_RELAXED, noreg /* ltos */, field, noreg, noreg);
2963 __ push(ltos);
2964 // Rewrite bytecode to be faster
2965 LP64_ONLY(if (!is_static && rc == may_rewrite) patch_bytecode(Bytecodes::_fast_lgetfield, bc, rbx));
2966 __ jmp(Done);
2967
2968 __ bind(notLong);
2969 __ cmpl(flags, ftos);
2970 __ jcc(Assembler::notEqual, notFloat);
2971 // ftos
2972
2973 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
2974 __ push(ftos);
2975 // Rewrite bytecode to be faster
2976 if (!is_static && rc == may_rewrite) {
2977 patch_bytecode(Bytecodes::_fast_fgetfield, bc, rbx);
2978 }
2979 __ jmp(Done);
2980
2981 __ bind(notFloat);
2982 #ifdef ASSERT
2983 Label notDouble;
2984 __ cmpl(flags, dtos);
2985 __ jcc(Assembler::notEqual, notDouble);
2986 #endif
2987 // dtos
2988 __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
2989 __ push(dtos);
2990 // Rewrite bytecode to be faster
2991 if (!is_static && rc == may_rewrite) {
2992 patch_bytecode(Bytecodes::_fast_dgetfield, bc, rbx);
2993 }
2994 #ifdef ASSERT
2995 __ jmp(Done);
2996
2997 __ bind(notDouble);
2998 __ stop("Bad state");
2999 #endif
3000
3001 __ bind(Done);
3002 // [jk] not needed currently
3003 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadLoad |
3004 // Assembler::LoadStore));
3005 }
3006
3007 void TemplateTable::getfield(int byte_no) {
3008 getfield_or_static(byte_no, false);
3009 }
3010
3011 void TemplateTable::nofast_getfield(int byte_no) {
3012 getfield_or_static(byte_no, false, may_not_rewrite);
3013 }
3014
3015 void TemplateTable::getstatic(int byte_no) {
3016 getfield_or_static(byte_no, true);
3017 }
3018
3019
3020 // The registers cache and index expected to be set before call.
3021 // The function may destroy various registers, just not the cache and index registers.
3022 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
3023
3024 const Register robj = LP64_ONLY(c_rarg2) NOT_LP64(rax);
3025 const Register RBX = LP64_ONLY(c_rarg1) NOT_LP64(rbx);
3026 const Register RCX = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3027 const Register RDX = LP64_ONLY(rscratch1) NOT_LP64(rdx);
3028
3029 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
3030
3031 if (JvmtiExport::can_post_field_modification()) {
3032 // Check to see if a field modification watch has been set before
3033 // we take the time to call into the VM.
3034 Label L1;
3035 assert_different_registers(cache, index, rax);
3036 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3037 __ testl(rax, rax);
3038 __ jcc(Assembler::zero, L1);
3039
3040 __ get_cache_and_index_at_bcp(robj, RDX, 1);
3041
3042
3043 if (is_static) {
3044 // Life is simple. Null out the object pointer.
3045 __ xorl(RBX, RBX);
3046
3047 } else {
3048 // Life is harder. The stack holds the value on top, followed by
3049 // the object. We don't know the size of the value, though; it
3050 // could be one or two words depending on its type. As a result,
3051 // we must find the type to determine where the object is.
3052 #ifndef _LP64
3053 Label two_word, valsize_known;
3054 #endif
3055 __ movl(RCX, Address(robj, RDX,
3056 Address::times_ptr,
3057 in_bytes(cp_base_offset +
3058 ConstantPoolCacheEntry::flags_offset())));
3059 NOT_LP64(__ mov(rbx, rsp));
3060 __ shrl(RCX, ConstantPoolCacheEntry::tos_state_shift);
3061
3062 // Make sure we don't need to mask rcx after the above shift
3063 ConstantPoolCacheEntry::verify_tos_state_shift();
3064 #ifdef _LP64
3065 __ movptr(c_rarg1, at_tos_p1()); // initially assume a one word jvalue
3066 __ cmpl(c_rarg3, ltos);
3067 __ cmovptr(Assembler::equal,
3068 c_rarg1, at_tos_p2()); // ltos (two word jvalue)
3069 __ cmpl(c_rarg3, dtos);
3070 __ cmovptr(Assembler::equal,
3071 c_rarg1, at_tos_p2()); // dtos (two word jvalue)
3072 #else
3073 __ cmpl(rcx, ltos);
3074 __ jccb(Assembler::equal, two_word);
3075 __ cmpl(rcx, dtos);
3076 __ jccb(Assembler::equal, two_word);
3077 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
3078 __ jmpb(valsize_known);
3079
3080 __ bind(two_word);
3081 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
3082
3083 __ bind(valsize_known);
3084 // setup object pointer
3085 __ movptr(rbx, Address(rbx, 0));
3086 #endif
3087 }
3088 // cache entry pointer
3089 __ addptr(robj, in_bytes(cp_base_offset));
3090 __ shll(RDX, LogBytesPerWord);
3091 __ addptr(robj, RDX);
3092 // object (tos)
3093 __ mov(RCX, rsp);
3094 // c_rarg1: object pointer set up above (NULL if static)
3095 // c_rarg2: cache entry pointer
3096 // c_rarg3: jvalue object on the stack
3097 __ call_VM(noreg,
3098 CAST_FROM_FN_PTR(address,
3099 InterpreterRuntime::post_field_modification),
3100 RBX, robj, RCX);
3101 __ get_cache_and_index_at_bcp(cache, index, 1);
3102 __ bind(L1);
3103 }
3104 }
3105
3106 void TemplateTable::putfield_or_static(int byte_no, bool is_static, RewriteControl rc) {
3107 transition(vtos, vtos);
3108
3109 const Register cache = rcx;
3110 const Register index = rdx;
3111 const Register obj = rcx;
3112 const Register off = rbx;
3113 const Register flags = rax;
3114
3115 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
3116 jvmti_post_field_mod(cache, index, is_static);
3117 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
3118
3119 // [jk] not needed currently
3120 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3121 // Assembler::StoreStore));
3122
3123 Label notVolatile, Done;
3124 __ movl(rdx, flags);
3125 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3126 __ andl(rdx, 0x1);
3127
3128 // Check for volatile store
3129 __ testl(rdx, rdx);
3130 __ jcc(Assembler::zero, notVolatile);
3131
3132 putfield_or_static_helper(byte_no, is_static, rc, obj, off, flags);
3133 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3134 Assembler::StoreStore));
3135 __ jmp(Done);
3136 __ bind(notVolatile);
3137
3138 putfield_or_static_helper(byte_no, is_static, rc, obj, off, flags);
3139
3140 __ bind(Done);
3141 }
3142
3143 void TemplateTable::putfield_or_static_helper(int byte_no, bool is_static, RewriteControl rc,
3144 Register obj, Register off, Register flags) {
3145
3146 // field addresses
3147 const Address field(obj, off, Address::times_1, 0*wordSize);
3148 NOT_LP64( const Address hi(obj, off, Address::times_1, 1*wordSize);)
3149
3150 Label notByte, notBool, notInt, notShort, notChar,
3151 notLong, notFloat, notObj;
3152 Label Done;
3153
3154 const Register bc = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3155
3156 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3157
3158 assert(btos == 0, "change code, btos != 0");
3159 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
3160 __ jcc(Assembler::notZero, notByte);
3161
3162 // btos
3163 {
3164 __ pop(btos);
3165 if (!is_static) pop_and_check_object(obj);
3166 __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg);
3167 if (!is_static && rc == may_rewrite) {
3168 patch_bytecode(Bytecodes::_fast_bputfield, bc, rbx, true, byte_no);
3169 }
3170 __ jmp(Done);
3171 }
3172
3173 __ bind(notByte);
3174 __ cmpl(flags, ztos);
3175 __ jcc(Assembler::notEqual, notBool);
3176
3177 // ztos
3178 {
3179 __ pop(ztos);
3180 if (!is_static) pop_and_check_object(obj);
3181 __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg);
3182 if (!is_static && rc == may_rewrite) {
3183 patch_bytecode(Bytecodes::_fast_zputfield, bc, rbx, true, byte_no);
3184 }
3185 __ jmp(Done);
3186 }
3187
3188 __ bind(notBool);
3189 __ cmpl(flags, atos);
3190 __ jcc(Assembler::notEqual, notObj);
3191
3192 // atos
3193 {
3194 __ pop(atos);
3195 if (!is_static) pop_and_check_object(obj);
3196 // Store into the field
3197 do_oop_store(_masm, field, rax);
3198 if (!is_static && rc == may_rewrite) {
3199 patch_bytecode(Bytecodes::_fast_aputfield, bc, rbx, true, byte_no);
3200 }
3201 __ jmp(Done);
3202 }
3203
3204 __ bind(notObj);
3205 __ cmpl(flags, itos);
3206 __ jcc(Assembler::notEqual, notInt);
3207
3208 // itos
3209 {
3210 __ pop(itos);
3211 if (!is_static) pop_and_check_object(obj);
3212 __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg);
3213 if (!is_static && rc == may_rewrite) {
3214 patch_bytecode(Bytecodes::_fast_iputfield, bc, rbx, true, byte_no);
3215 }
3216 __ jmp(Done);
3217 }
3218
3219 __ bind(notInt);
3220 __ cmpl(flags, ctos);
3221 __ jcc(Assembler::notEqual, notChar);
3222
3223 // ctos
3224 {
3225 __ pop(ctos);
3226 if (!is_static) pop_and_check_object(obj);
3227 __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg);
3228 if (!is_static && rc == may_rewrite) {
3229 patch_bytecode(Bytecodes::_fast_cputfield, bc, rbx, true, byte_no);
3230 }
3231 __ jmp(Done);
3232 }
3233
3234 __ bind(notChar);
3235 __ cmpl(flags, stos);
3236 __ jcc(Assembler::notEqual, notShort);
3237
3238 // stos
3239 {
3240 __ pop(stos);
3241 if (!is_static) pop_and_check_object(obj);
3242 __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg);
3243 if (!is_static && rc == may_rewrite) {
3244 patch_bytecode(Bytecodes::_fast_sputfield, bc, rbx, true, byte_no);
3245 }
3246 __ jmp(Done);
3247 }
3248
3249 __ bind(notShort);
3250 __ cmpl(flags, ltos);
3251 __ jcc(Assembler::notEqual, notLong);
3252
3253 // ltos
3254 {
3255 __ pop(ltos);
3256 if (!is_static) pop_and_check_object(obj);
3257 __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos*/, noreg, noreg);
3258 #ifdef _LP64
3259 if (!is_static && rc == may_rewrite) {
3260 patch_bytecode(Bytecodes::_fast_lputfield, bc, rbx, true, byte_no);
3261 }
3262 #endif // _LP64
3263 __ jmp(Done);
3264 }
3265
3266 __ bind(notLong);
3267 __ cmpl(flags, ftos);
3268 __ jcc(Assembler::notEqual, notFloat);
3269
3270 // ftos
3271 {
3272 __ pop(ftos);
3273 if (!is_static) pop_and_check_object(obj);
3274 __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos */, noreg, noreg);
3275 if (!is_static && rc == may_rewrite) {
3276 patch_bytecode(Bytecodes::_fast_fputfield, bc, rbx, true, byte_no);
3277 }
3278 __ jmp(Done);
3279 }
3280
3281 __ bind(notFloat);
3282 #ifdef ASSERT
3283 Label notDouble;
3284 __ cmpl(flags, dtos);
3285 __ jcc(Assembler::notEqual, notDouble);
3286 #endif
3287
3288 // dtos
3289 {
3290 __ pop(dtos);
3291 if (!is_static) pop_and_check_object(obj);
3292 __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos */, noreg, noreg);
3293 if (!is_static && rc == may_rewrite) {
3294 patch_bytecode(Bytecodes::_fast_dputfield, bc, rbx, true, byte_no);
3295 }
3296 }
3297
3298 #ifdef ASSERT
3299 __ jmp(Done);
3300
3301 __ bind(notDouble);
3302 __ stop("Bad state");
3303 #endif
3304
3305 __ bind(Done);
3306 }
3307
3308 void TemplateTable::putfield(int byte_no) {
3309 putfield_or_static(byte_no, false);
3310 }
3311
3312 void TemplateTable::nofast_putfield(int byte_no) {
3313 putfield_or_static(byte_no, false, may_not_rewrite);
3314 }
3315
3316 void TemplateTable::putstatic(int byte_no) {
3317 putfield_or_static(byte_no, true);
3318 }
3319
3320 void TemplateTable::jvmti_post_fast_field_mod() {
3321
3322 const Register scratch = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
3323
3324 if (JvmtiExport::can_post_field_modification()) {
3325 // Check to see if a field modification watch has been set before
3326 // we take the time to call into the VM.
3327 Label L2;
3328 __ mov32(scratch, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
3329 __ testl(scratch, scratch);
3330 __ jcc(Assembler::zero, L2);
3331 __ pop_ptr(rbx); // copy the object pointer from tos
3332 __ verify_oop(rbx);
3333 __ push_ptr(rbx); // put the object pointer back on tos
3334 // Save tos values before call_VM() clobbers them. Since we have
3335 // to do it for every data type, we use the saved values as the
3336 // jvalue object.
3337 switch (bytecode()) { // load values into the jvalue object
3338 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
3339 case Bytecodes::_fast_bputfield: // fall through
3340 case Bytecodes::_fast_zputfield: // fall through
3341 case Bytecodes::_fast_sputfield: // fall through
3342 case Bytecodes::_fast_cputfield: // fall through
3343 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
3344 case Bytecodes::_fast_dputfield: __ push(dtos); break;
3345 case Bytecodes::_fast_fputfield: __ push(ftos); break;
3346 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
3347
3348 default:
3349 ShouldNotReachHere();
3350 }
3351 __ mov(scratch, rsp); // points to jvalue on the stack
3352 // access constant pool cache entry
3353 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rax, 1));
3354 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rax, rdx, 1));
3355 __ verify_oop(rbx);
3356 // rbx: object pointer copied above
3357 // c_rarg2: cache entry pointer
3358 // c_rarg3: jvalue object on the stack
3359 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, c_rarg2, c_rarg3));
3360 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx));
3361
3362 switch (bytecode()) { // restore tos values
3363 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
3364 case Bytecodes::_fast_bputfield: // fall through
3365 case Bytecodes::_fast_zputfield: // fall through
3366 case Bytecodes::_fast_sputfield: // fall through
3367 case Bytecodes::_fast_cputfield: // fall through
3368 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
3369 case Bytecodes::_fast_dputfield: __ pop(dtos); break;
3370 case Bytecodes::_fast_fputfield: __ pop(ftos); break;
3371 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
3372 default: break;
3373 }
3374 __ bind(L2);
3375 }
3376 }
3377
3378 void TemplateTable::fast_storefield(TosState state) {
3379 transition(state, vtos);
3380
3381 ByteSize base = ConstantPoolCache::base_offset();
3382
3383 jvmti_post_fast_field_mod();
3384
3385 // access constant pool cache
3386 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3387
3388 // test for volatile with rdx but rdx is tos register for lputfield.
3389 __ movl(rdx, Address(rcx, rbx, Address::times_ptr,
3390 in_bytes(base +
3391 ConstantPoolCacheEntry::flags_offset())));
3392
3393 // replace index with field offset from cache entry
3394 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3395 in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
3396
3397 // [jk] not needed currently
3398 // volatile_barrier(Assembler::Membar_mask_bits(Assembler::LoadStore |
3399 // Assembler::StoreStore));
3400
3401 Label notVolatile, Done;
3402 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3403 __ andl(rdx, 0x1);
3404
3405 // Get object from stack
3406 pop_and_check_object(rcx);
3407
3408 // field address
3409 const Address field(rcx, rbx, Address::times_1);
3410
3411 // Check for volatile store
3412 __ testl(rdx, rdx);
3413 __ jcc(Assembler::zero, notVolatile);
3414
3415 fast_storefield_helper(field, rax);
3416 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
3417 Assembler::StoreStore));
3418 __ jmp(Done);
3419 __ bind(notVolatile);
3420
3421 fast_storefield_helper(field, rax);
3422
3423 __ bind(Done);
3424 }
3425
3426 void TemplateTable::fast_storefield_helper(Address field, Register rax) {
3427
3428 // access field
3429 switch (bytecode()) {
3430 case Bytecodes::_fast_aputfield:
3431 do_oop_store(_masm, field, rax);
3432 break;
3433 case Bytecodes::_fast_lputfield:
3434 #ifdef _LP64
3435 __ access_store_at(T_LONG, IN_HEAP, field, noreg /* ltos */, noreg, noreg);
3436 #else
3437 __ stop("should not be rewritten");
3438 #endif
3439 break;
3440 case Bytecodes::_fast_iputfield:
3441 __ access_store_at(T_INT, IN_HEAP, field, rax, noreg, noreg);
3442 break;
3443 case Bytecodes::_fast_zputfield:
3444 __ access_store_at(T_BOOLEAN, IN_HEAP, field, rax, noreg, noreg);
3445 break;
3446 case Bytecodes::_fast_bputfield:
3447 __ access_store_at(T_BYTE, IN_HEAP, field, rax, noreg, noreg);
3448 break;
3449 case Bytecodes::_fast_sputfield:
3450 __ access_store_at(T_SHORT, IN_HEAP, field, rax, noreg, noreg);
3451 break;
3452 case Bytecodes::_fast_cputfield:
3453 __ access_store_at(T_CHAR, IN_HEAP, field, rax, noreg, noreg);
3454 break;
3455 case Bytecodes::_fast_fputfield:
3456 __ access_store_at(T_FLOAT, IN_HEAP, field, noreg /* ftos*/, noreg, noreg);
3457 break;
3458 case Bytecodes::_fast_dputfield:
3459 __ access_store_at(T_DOUBLE, IN_HEAP, field, noreg /* dtos*/, noreg, noreg);
3460 break;
3461 default:
3462 ShouldNotReachHere();
3463 }
3464 }
3465
3466 void TemplateTable::fast_accessfield(TosState state) {
3467 transition(atos, state);
3468
3469 // Do the JVMTI work here to avoid disturbing the register state below
3470 if (JvmtiExport::can_post_field_access()) {
3471 // Check to see if a field access watch has been set before we
3472 // take the time to call into the VM.
3473 Label L1;
3474 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
3475 __ testl(rcx, rcx);
3476 __ jcc(Assembler::zero, L1);
3477 // access constant pool cache entry
3478 LP64_ONLY(__ get_cache_entry_pointer_at_bcp(c_rarg2, rcx, 1));
3479 NOT_LP64(__ get_cache_entry_pointer_at_bcp(rcx, rdx, 1));
3480 __ verify_oop(rax);
3481 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
3482 LP64_ONLY(__ mov(c_rarg1, rax));
3483 // c_rarg1: object pointer copied above
3484 // c_rarg2: cache entry pointer
3485 LP64_ONLY(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), c_rarg1, c_rarg2));
3486 NOT_LP64(__ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx));
3487 __ pop_ptr(rax); // restore object pointer
3488 __ bind(L1);
3489 }
3490
3491 // access constant pool cache
3492 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
3493 // replace index with field offset from cache entry
3494 // [jk] not needed currently
3495 // __ movl(rdx, Address(rcx, rbx, Address::times_8,
3496 // in_bytes(ConstantPoolCache::base_offset() +
3497 // ConstantPoolCacheEntry::flags_offset())));
3498 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3499 // __ andl(rdx, 0x1);
3500 //
3501 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr,
3502 in_bytes(ConstantPoolCache::base_offset() +
3503 ConstantPoolCacheEntry::f2_offset())));
3504
3505 // rax: object
3506 __ verify_oop(rax);
3507 __ null_check(rax);
3508 Address field(rax, rbx, Address::times_1);
3509
3510 // access field
3511 switch (bytecode()) {
3512 case Bytecodes::_fast_agetfield:
3513 do_oop_load(_masm, field, rax);
3514 __ verify_oop(rax);
3515 break;
3516 case Bytecodes::_fast_lgetfield:
3517 #ifdef _LP64
3518 __ access_load_at(T_LONG, IN_HEAP, noreg /* ltos */, field, noreg, noreg);
3519 #else
3520 __ stop("should not be rewritten");
3521 #endif
3522 break;
3523 case Bytecodes::_fast_igetfield:
3524 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3525 break;
3526 case Bytecodes::_fast_bgetfield:
3527 __ access_load_at(T_BYTE, IN_HEAP, rax, field, noreg, noreg);
3528 break;
3529 case Bytecodes::_fast_sgetfield:
3530 __ access_load_at(T_SHORT, IN_HEAP, rax, field, noreg, noreg);
3531 break;
3532 case Bytecodes::_fast_cgetfield:
3533 __ access_load_at(T_CHAR, IN_HEAP, rax, field, noreg, noreg);
3534 break;
3535 case Bytecodes::_fast_fgetfield:
3536 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3537 break;
3538 case Bytecodes::_fast_dgetfield:
3539 __ access_load_at(T_DOUBLE, IN_HEAP, noreg /* dtos */, field, noreg, noreg);
3540 break;
3541 default:
3542 ShouldNotReachHere();
3543 }
3544 // [jk] not needed currently
3545 // Label notVolatile;
3546 // __ testl(rdx, rdx);
3547 // __ jcc(Assembler::zero, notVolatile);
3548 // __ membar(Assembler::LoadLoad);
3549 // __ bind(notVolatile);
3550 }
3551
3552 void TemplateTable::fast_xaccess(TosState state) {
3553 transition(vtos, state);
3554
3555 // get receiver
3556 __ movptr(rax, aaddress(0));
3557 // access constant pool cache
3558 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
3559 __ movptr(rbx,
3560 Address(rcx, rdx, Address::times_ptr,
3561 in_bytes(ConstantPoolCache::base_offset() +
3562 ConstantPoolCacheEntry::f2_offset())));
3563 // make sure exception is reported in correct bcp range (getfield is
3564 // next instruction)
3565 __ increment(rbcp);
3566 __ null_check(rax);
3567 const Address field = Address(rax, rbx, Address::times_1, 0*wordSize);
3568 switch (state) {
3569 case itos:
3570 __ access_load_at(T_INT, IN_HEAP, rax, field, noreg, noreg);
3571 break;
3572 case atos:
3573 do_oop_load(_masm, field, rax);
3574 __ verify_oop(rax);
3575 break;
3576 case ftos:
3577 __ access_load_at(T_FLOAT, IN_HEAP, noreg /* ftos */, field, noreg, noreg);
3578 break;
3579 default:
3580 ShouldNotReachHere();
3581 }
3582
3583 // [jk] not needed currently
3584 // Label notVolatile;
3585 // __ movl(rdx, Address(rcx, rdx, Address::times_8,
3586 // in_bytes(ConstantPoolCache::base_offset() +
3587 // ConstantPoolCacheEntry::flags_offset())));
3588 // __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
3589 // __ testl(rdx, 0x1);
3590 // __ jcc(Assembler::zero, notVolatile);
3591 // __ membar(Assembler::LoadLoad);
3592 // __ bind(notVolatile);
3593
3594 __ decrement(rbcp);
3595 }
3596
3597 //-----------------------------------------------------------------------------
3598 // Calls
3599
3600 void TemplateTable::count_calls(Register method, Register temp) {
3601 // implemented elsewhere
3602 ShouldNotReachHere();
3603 }
3604
3605 void TemplateTable::prepare_invoke(int byte_no,
3606 Register method, // linked method (or i-klass)
3607 Register index, // itable index, MethodType, etc.
3608 Register recv, // if caller wants to see it
3609 Register flags // if caller wants to test it
3610 ) {
3611 // determine flags
3612 const Bytecodes::Code code = bytecode();
3613 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
3614 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
3615 const bool is_invokehandle = code == Bytecodes::_invokehandle;
3616 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
3617 const bool is_invokespecial = code == Bytecodes::_invokespecial;
3618 const bool load_receiver = (recv != noreg);
3619 const bool save_flags = (flags != noreg);
3620 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
3621 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
3622 assert(flags == noreg || flags == rdx, "");
3623 assert(recv == noreg || recv == rcx, "");
3624
3625 // setup registers & access constant pool cache
3626 if (recv == noreg) recv = rcx;
3627 if (flags == noreg) flags = rdx;
3628 assert_different_registers(method, index, recv, flags);
3629
3630 // save 'interpreter return address'
3631 __ save_bcp();
3632
3633 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
3634
3635 // maybe push appendix to arguments (just before return address)
3636 if (is_invokedynamic || is_invokehandle) {
3637 Label L_no_push;
3638 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
3639 __ jcc(Assembler::zero, L_no_push);
3640 // Push the appendix as a trailing parameter.
3641 // This must be done before we get the receiver,
3642 // since the parameter_size includes it.
3643 __ push(rbx);
3644 __ mov(rbx, index);
3645 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
3646 __ load_resolved_reference_at_index(index, rbx);
3647 __ pop(rbx);
3648 __ push(index); // push appendix (MethodType, CallSite, etc.)
3649 __ bind(L_no_push);
3650 }
3651
3652 // load receiver if needed (after appendix is pushed so parameter size is correct)
3653 // Note: no return address pushed yet
3654 if (load_receiver) {
3655 __ movl(recv, flags);
3656 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
3657 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
3658 const int receiver_is_at_end = -1; // back off one slot to get receiver
3659 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
3660 __ movptr(recv, recv_addr);
3661 __ verify_oop(recv);
3662 }
3663
3664 if (save_flags) {
3665 __ movl(rbcp, flags);
3666 }
3667
3668 // compute return type
3669 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
3670 // Make sure we don't need to mask flags after the above shift
3671 ConstantPoolCacheEntry::verify_tos_state_shift();
3672 // load return address
3673 {
3674 const address table_addr = (address) Interpreter::invoke_return_entry_table_for(code);
3675 ExternalAddress table(table_addr);
3676 LP64_ONLY(__ lea(rscratch1, table));
3677 LP64_ONLY(__ movptr(flags, Address(rscratch1, flags, Address::times_ptr)));
3678 NOT_LP64(__ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr))));
3679 }
3680
3681 // push return address
3682 __ push(flags);
3683
3684 // Restore flags value from the constant pool cache, and restore rsi
3685 // for later null checks. r13 is the bytecode pointer
3686 if (save_flags) {
3687 __ movl(flags, rbcp);
3688 __ restore_bcp();
3689 }
3690 }
3691
3692 void TemplateTable::invokevirtual_helper(Register index,
3693 Register recv,
3694 Register flags) {
3695 // Uses temporary registers rax, rdx
3696 assert_different_registers(index, recv, rax, rdx);
3697 assert(index == rbx, "");
3698 assert(recv == rcx, "");
3699
3700 // Test for an invoke of a final method
3701 Label notFinal;
3702 __ movl(rax, flags);
3703 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3704 __ jcc(Assembler::zero, notFinal);
3705
3706 const Register method = index; // method must be rbx
3707 assert(method == rbx,
3708 "Method* must be rbx for interpreter calling convention");
3709
3710 // do the call - the index is actually the method to call
3711 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
3712
3713 // It's final, need a null check here!
3714 __ null_check(recv);
3715
3716 // profile this call
3717 __ profile_final_call(rax);
3718 __ profile_arguments_type(rax, method, rbcp, true);
3719
3720 __ jump_from_interpreted(method, rax);
3721
3722 __ bind(notFinal);
3723
3724 // get receiver klass
3725 __ null_check(recv, oopDesc::klass_offset_in_bytes());
3726 __ load_klass(rax, recv);
3727
3728 // profile this call
3729 __ profile_virtual_call(rax, rlocals, rdx);
3730 // get target Method* & entry point
3731 __ lookup_virtual_method(rax, index, method);
3732 __ profile_called_method(method, rdx, rbcp);
3733
3734 __ profile_arguments_type(rdx, method, rbcp, true);
3735 __ jump_from_interpreted(method, rdx);
3736 }
3737
3738 void TemplateTable::invokevirtual(int byte_no) {
3739 transition(vtos, vtos);
3740 assert(byte_no == f2_byte, "use this argument");
3741 prepare_invoke(byte_no,
3742 rbx, // method or vtable index
3743 noreg, // unused itable index
3744 rcx, rdx); // recv, flags
3745
3746 // rbx: index
3747 // rcx: receiver
3748 // rdx: flags
3749
3750 invokevirtual_helper(rbx, rcx, rdx);
3751 }
3752
3753 void TemplateTable::invokespecial(int byte_no) {
3754 transition(vtos, vtos);
3755 assert(byte_no == f1_byte, "use this argument");
3756 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
3757 rcx); // get receiver also for null check
3758 __ verify_oop(rcx);
3759 __ null_check(rcx);
3760 // do the call
3761 __ profile_call(rax);
3762 __ profile_arguments_type(rax, rbx, rbcp, false);
3763 __ jump_from_interpreted(rbx, rax);
3764 }
3765
3766 void TemplateTable::invokestatic(int byte_no) {
3767 transition(vtos, vtos);
3768 assert(byte_no == f1_byte, "use this argument");
3769 prepare_invoke(byte_no, rbx); // get f1 Method*
3770 // do the call
3771 __ profile_call(rax);
3772 __ profile_arguments_type(rax, rbx, rbcp, false);
3773 __ jump_from_interpreted(rbx, rax);
3774 }
3775
3776
3777 void TemplateTable::fast_invokevfinal(int byte_no) {
3778 transition(vtos, vtos);
3779 assert(byte_no == f2_byte, "use this argument");
3780 __ stop("fast_invokevfinal not used on x86");
3781 }
3782
3783
3784 void TemplateTable::invokeinterface(int byte_no) {
3785 transition(vtos, vtos);
3786 assert(byte_no == f1_byte, "use this argument");
3787 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 Method*
3788 rcx, rdx); // recv, flags
3789
3790 // rax: reference klass (from f1) if interface method
3791 // rbx: method (from f2)
3792 // rcx: receiver
3793 // rdx: flags
3794
3795 // First check for Object case, then private interface method,
3796 // then regular interface method.
3797
3798 // Special case of invokeinterface called for virtual method of
3799 // java.lang.Object. See cpCache.cpp for details.
3800 Label notObjectMethod;
3801 __ movl(rlocals, rdx);
3802 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3803 __ jcc(Assembler::zero, notObjectMethod);
3804 invokevirtual_helper(rbx, rcx, rdx);
3805 // no return from above
3806 __ bind(notObjectMethod);
3807
3808 Label no_such_interface; // for receiver subtype check
3809 Register recvKlass; // used for exception processing
3810
3811 // Check for private method invocation - indicated by vfinal
3812 Label notVFinal;
3813 __ movl(rlocals, rdx);
3814 __ andl(rlocals, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
3815 __ jcc(Assembler::zero, notVFinal);
3816
3817 // Get receiver klass into rlocals - also a null check
3818 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3819 __ load_klass(rlocals, rcx);
3820
3821 Label subtype;
3822 __ check_klass_subtype(rlocals, rax, rbcp, subtype);
3823 // If we get here the typecheck failed
3824 recvKlass = rdx;
3825 __ mov(recvKlass, rlocals); // shuffle receiver class for exception use
3826 __ jmp(no_such_interface);
3827
3828 __ bind(subtype);
3829
3830 // do the call - rbx is actually the method to call
3831
3832 __ profile_final_call(rdx);
3833 __ profile_arguments_type(rdx, rbx, rbcp, true);
3834
3835 __ jump_from_interpreted(rbx, rdx);
3836 // no return from above
3837 __ bind(notVFinal);
3838
3839 // Get receiver klass into rdx - also a null check
3840 __ restore_locals(); // restore r14
3841 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3842 __ load_klass(rdx, rcx);
3843
3844 Label no_such_method;
3845
3846 // Preserve method for throw_AbstractMethodErrorVerbose.
3847 __ mov(rcx, rbx);
3848 // Receiver subtype check against REFC.
3849 // Superklass in rax. Subklass in rdx. Blows rcx, rdi.
3850 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3851 rdx, rax, noreg,
3852 // outputs: scan temp. reg, scan temp. reg
3853 rbcp, rlocals,
3854 no_such_interface,
3855 /*return_method=*/false);
3856
3857 // profile this call
3858 __ restore_bcp(); // rbcp was destroyed by receiver type check
3859 __ profile_virtual_call(rdx, rbcp, rlocals);
3860
3861 // Get declaring interface class from method, and itable index
3862 __ movptr(rax, Address(rbx, Method::const_offset()));
3863 __ movptr(rax, Address(rax, ConstMethod::constants_offset()));
3864 __ movptr(rax, Address(rax, ConstantPool::pool_holder_offset_in_bytes()));
3865 __ movl(rbx, Address(rbx, Method::itable_index_offset()));
3866 __ subl(rbx, Method::itable_index_max);
3867 __ negl(rbx);
3868
3869 // Preserve recvKlass for throw_AbstractMethodErrorVerbose.
3870 __ mov(rlocals, rdx);
3871 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3872 rlocals, rax, rbx,
3873 // outputs: method, scan temp. reg
3874 rbx, rbcp,
3875 no_such_interface);
3876
3877 // rbx: Method* to call
3878 // rcx: receiver
3879 // Check for abstract method error
3880 // Note: This should be done more efficiently via a throw_abstract_method_error
3881 // interpreter entry point and a conditional jump to it in case of a null
3882 // method.
3883 __ testptr(rbx, rbx);
3884 __ jcc(Assembler::zero, no_such_method);
3885
3886 __ profile_called_method(rbx, rbcp, rdx);
3887 __ profile_arguments_type(rdx, rbx, rbcp, true);
3888
3889 // do the call
3890 // rcx: receiver
3891 // rbx,: Method*
3892 __ jump_from_interpreted(rbx, rdx);
3893 __ should_not_reach_here();
3894
3895 // exception handling code follows...
3896 // note: must restore interpreter registers to canonical
3897 // state for exception handling to work correctly!
3898
3899 __ bind(no_such_method);
3900 // throw exception
3901 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3902 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3903 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3904 // Pass arguments for generating a verbose error message.
3905 #ifdef _LP64
3906 recvKlass = c_rarg1;
3907 Register method = c_rarg2;
3908 if (recvKlass != rdx) { __ movq(recvKlass, rdx); }
3909 if (method != rcx) { __ movq(method, rcx); }
3910 #else
3911 recvKlass = rdx;
3912 Register method = rcx;
3913 #endif
3914 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodErrorVerbose),
3915 recvKlass, method);
3916 // The call_VM checks for exception, so we should never return here.
3917 __ should_not_reach_here();
3918
3919 __ bind(no_such_interface);
3920 // throw exception
3921 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3922 __ restore_bcp(); // rbcp must be correct for exception handler (was destroyed)
3923 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3924 // Pass arguments for generating a verbose error message.
3925 LP64_ONLY( if (recvKlass != rdx) { __ movq(recvKlass, rdx); } )
3926 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_IncompatibleClassChangeErrorVerbose),
3927 recvKlass, rax);
3928 // the call_VM checks for exception, so we should never return here.
3929 __ should_not_reach_here();
3930 }
3931
3932 void TemplateTable::invokehandle(int byte_no) {
3933 transition(vtos, vtos);
3934 assert(byte_no == f1_byte, "use this argument");
3935 const Register rbx_method = rbx;
3936 const Register rax_mtype = rax;
3937 const Register rcx_recv = rcx;
3938 const Register rdx_flags = rdx;
3939
3940 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3941 __ verify_method_ptr(rbx_method);
3942 __ verify_oop(rcx_recv);
3943 __ null_check(rcx_recv);
3944
3945 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3946 // rbx: MH.invokeExact_MT method (from f2)
3947
3948 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3949
3950 // FIXME: profile the LambdaForm also
3951 __ profile_final_call(rax);
3952 __ profile_arguments_type(rdx, rbx_method, rbcp, true);
3953
3954 __ jump_from_interpreted(rbx_method, rdx);
3955 }
3956
3957 void TemplateTable::invokedynamic(int byte_no) {
3958 transition(vtos, vtos);
3959 assert(byte_no == f1_byte, "use this argument");
3960
3961 const Register rbx_method = rbx;
3962 const Register rax_callsite = rax;
3963
3964 prepare_invoke(byte_no, rbx_method, rax_callsite);
3965
3966 // rax: CallSite object (from cpool->resolved_references[f1])
3967 // rbx: MH.linkToCallSite method (from f2)
3968
3969 // Note: rax_callsite is already pushed by prepare_invoke
3970
3971 // %%% should make a type profile for any invokedynamic that takes a ref argument
3972 // profile this call
3973 __ profile_call(rbcp);
3974 __ profile_arguments_type(rdx, rbx_method, rbcp, false);
3975
3976 __ verify_oop(rax_callsite);
3977
3978 __ jump_from_interpreted(rbx_method, rdx);
3979 }
3980
3981 //-----------------------------------------------------------------------------
3982 // Allocation
3983
3984 void TemplateTable::_new() {
3985 transition(vtos, atos);
3986 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3987 Label slow_case;
3988 Label slow_case_no_pop;
3989 Label done;
3990 Label initialize_header;
3991 Label initialize_object; // including clearing the fields
3992
3993 __ get_cpool_and_tags(rcx, rax);
3994
3995 // Make sure the class we're about to instantiate has been resolved.
3996 // This is done before loading InstanceKlass to be consistent with the order
3997 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3998 const int tags_offset = Array<u1>::base_offset_in_bytes();
3999 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
4000 __ jcc(Assembler::notEqual, slow_case_no_pop);
4001
4002 // get InstanceKlass
4003 __ load_resolved_klass_at_index(rcx, rdx, rcx);
4004 __ push(rcx); // save the contexts of klass for initializing the header
4005
4006 // make sure klass is initialized & doesn't have finalizer
4007 // make sure klass is fully initialized
4008 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
4009 __ jcc(Assembler::notEqual, slow_case);
4010
4011 // get instance_size in InstanceKlass (scaled to a count of bytes)
4012 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
4013 // test to see if it has a finalizer or is malformed in some way
4014 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
4015 __ jcc(Assembler::notZero, slow_case);
4016
4017 // Allocate the instance:
4018 // If TLAB is enabled:
4019 // Try to allocate in the TLAB.
4020 // If fails, go to the slow path.
4021 // Else If inline contiguous allocations are enabled:
4022 // Try to allocate in eden.
4023 // If fails due to heap end, go to slow path.
4024 //
4025 // If TLAB is enabled OR inline contiguous is enabled:
4026 // Initialize the allocation.
4027 // Exit.
4028 //
4029 // Go to slow path.
4030
4031 const bool allow_shared_alloc =
4032 Universe::heap()->supports_inline_contig_alloc();
4033
4034 const Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx);
4035 #ifndef _LP64
4036 if (UseTLAB || allow_shared_alloc) {
4037 __ get_thread(thread);
4038 }
4039 #endif // _LP64
4040
4041 if (UseTLAB) {
4042 __ tlab_allocate(thread, rax, rdx, 0, rcx, rbx, slow_case);
4043 if (ZeroTLAB) {
4044 // the fields have been already cleared
4045 __ jmp(initialize_header);
4046 } else {
4047 // initialize both the header and fields
4048 __ jmp(initialize_object);
4049 }
4050 } else {
4051 // Allocation in the shared Eden, if allowed.
4052 //
4053 // rdx: instance size in bytes
4054 __ eden_allocate(thread, rax, rdx, 0, rbx, slow_case);
4055 }
4056
4057 // If UseTLAB or allow_shared_alloc are true, the object is created above and
4058 // there is an initialize need. Otherwise, skip and go to the slow path.
4059 if (UseTLAB || allow_shared_alloc) {
4060 // The object is initialized before the header. If the object size is
4061 // zero, go directly to the header initialization.
4062 __ bind(initialize_object);
4063 __ decrement(rdx, sizeof(oopDesc));
4064 __ jcc(Assembler::zero, initialize_header);
4065
4066 // Initialize topmost object field, divide rdx by 8, check if odd and
4067 // test if zero.
4068 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
4069 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
4070
4071 // rdx must have been multiple of 8
4072 #ifdef ASSERT
4073 // make sure rdx was multiple of 8
4074 Label L;
4075 // Ignore partial flag stall after shrl() since it is debug VM
4076 __ jcc(Assembler::carryClear, L);
4077 __ stop("object size is not multiple of 2 - adjust this code");
4078 __ bind(L);
4079 // rdx must be > 0, no extra check needed here
4080 #endif
4081
4082 // initialize remaining object fields: rdx was a multiple of 8
4083 { Label loop;
4084 __ bind(loop);
4085 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
4086 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
4087 __ decrement(rdx);
4088 __ jcc(Assembler::notZero, loop);
4089 }
4090
4091 // initialize object header only.
4092 __ bind(initialize_header);
4093 if (UseBiasedLocking) {
4094 __ pop(rcx); // get saved klass back in the register.
4095 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
4096 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
4097 } else {
4098 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
4099 (intptr_t)markOopDesc::prototype()); // header
4100 __ pop(rcx); // get saved klass back in the register.
4101 }
4102 #ifdef _LP64
4103 __ xorl(rsi, rsi); // use zero reg to clear memory (shorter code)
4104 __ store_klass_gap(rax, rsi); // zero klass gap for compressed oops
4105 #endif
4106 __ store_klass(rax, rcx); // klass
4107
4108 {
4109 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
4110 // Trigger dtrace event for fastpath
4111 __ push(atos);
4112 __ call_VM_leaf(
4113 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
4114 __ pop(atos);
4115 }
4116
4117 __ jmp(done);
4118 }
4119
4120 // slow case
4121 __ bind(slow_case);
4122 __ pop(rcx); // restore stack pointer to what it was when we came in.
4123 __ bind(slow_case_no_pop);
4124
4125 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4126 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4127
4128 __ get_constant_pool(rarg1);
4129 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4130 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rarg1, rarg2);
4131 __ verify_oop(rax);
4132
4133 // continue
4134 __ bind(done);
4135 }
4136
4137 void TemplateTable::newarray() {
4138 transition(itos, atos);
4139 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4140 __ load_unsigned_byte(rarg1, at_bcp(1));
4141 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray),
4142 rarg1, rax);
4143 }
4144
4145 void TemplateTable::anewarray() {
4146 transition(itos, atos);
4147
4148 Register rarg1 = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4149 Register rarg2 = LP64_ONLY(c_rarg2) NOT_LP64(rdx);
4150
4151 __ get_unsigned_2_byte_index_at_bcp(rarg2, 1);
4152 __ get_constant_pool(rarg1);
4153 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray),
4154 rarg1, rarg2, rax);
4155 }
4156
4157 void TemplateTable::arraylength() {
4158 transition(atos, itos);
4159 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
4160 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
4161 }
4162
4163 void TemplateTable::checkcast() {
4164 transition(atos, atos);
4165 Label done, is_null, ok_is_subtype, quicked, resolved;
4166 __ testptr(rax, rax); // object is in rax
4167 __ jcc(Assembler::zero, is_null);
4168
4169 // Get cpool & tags index
4170 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4171 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4172 // See if bytecode has already been quicked
4173 __ cmpb(Address(rdx, rbx,
4174 Address::times_1,
4175 Array<u1>::base_offset_in_bytes()),
4176 JVM_CONSTANT_Class);
4177 __ jcc(Assembler::equal, quicked);
4178 __ push(atos); // save receiver for result, and for GC
4179 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4180
4181 // vm_result_2 has metadata result
4182 #ifndef _LP64
4183 // borrow rdi from locals
4184 __ get_thread(rdi);
4185 __ get_vm_result_2(rax, rdi);
4186 __ restore_locals();
4187 #else
4188 __ get_vm_result_2(rax, r15_thread);
4189 #endif
4190
4191 __ pop_ptr(rdx); // restore receiver
4192 __ jmpb(resolved);
4193
4194 // Get superklass in rax and subklass in rbx
4195 __ bind(quicked);
4196 __ mov(rdx, rax); // Save object in rdx; rax needed for subtype check
4197 __ load_resolved_klass_at_index(rcx, rbx, rax);
4198
4199 __ bind(resolved);
4200 __ load_klass(rbx, rdx);
4201
4202 // Generate subtype check. Blows rcx, rdi. Object in rdx.
4203 // Superklass in rax. Subklass in rbx.
4204 __ gen_subtype_check(rbx, ok_is_subtype);
4205
4206 // Come here on failure
4207 __ push_ptr(rdx);
4208 // object is at TOS
4209 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
4210
4211 // Come here on success
4212 __ bind(ok_is_subtype);
4213 __ mov(rax, rdx); // Restore object in rdx
4214
4215 // Collect counts on whether this check-cast sees NULLs a lot or not.
4216 if (ProfileInterpreter) {
4217 __ jmp(done);
4218 __ bind(is_null);
4219 __ profile_null_seen(rcx);
4220 } else {
4221 __ bind(is_null); // same as 'done'
4222 }
4223 __ bind(done);
4224 }
4225
4226 void TemplateTable::instanceof() {
4227 transition(atos, itos);
4228 Label done, is_null, ok_is_subtype, quicked, resolved;
4229 __ testptr(rax, rax);
4230 __ jcc(Assembler::zero, is_null);
4231
4232 // Get cpool & tags index
4233 __ get_cpool_and_tags(rcx, rdx); // rcx=cpool, rdx=tags array
4234 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // rbx=index
4235 // See if bytecode has already been quicked
4236 __ cmpb(Address(rdx, rbx,
4237 Address::times_1,
4238 Array<u1>::base_offset_in_bytes()),
4239 JVM_CONSTANT_Class);
4240 __ jcc(Assembler::equal, quicked);
4241
4242 __ push(atos); // save receiver for result, and for GC
4243 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc));
4244 // vm_result_2 has metadata result
4245
4246 #ifndef _LP64
4247 // borrow rdi from locals
4248 __ get_thread(rdi);
4249 __ get_vm_result_2(rax, rdi);
4250 __ restore_locals();
4251 #else
4252 __ get_vm_result_2(rax, r15_thread);
4253 #endif
4254
4255 __ pop_ptr(rdx); // restore receiver
4256 __ verify_oop(rdx);
4257 __ load_klass(rdx, rdx);
4258 __ jmpb(resolved);
4259
4260 // Get superklass in rax and subklass in rdx
4261 __ bind(quicked);
4262 __ load_klass(rdx, rax);
4263 __ load_resolved_klass_at_index(rcx, rbx, rax);
4264
4265 __ bind(resolved);
4266
4267 // Generate subtype check. Blows rcx, rdi
4268 // Superklass in rax. Subklass in rdx.
4269 __ gen_subtype_check(rdx, ok_is_subtype);
4270
4271 // Come here on failure
4272 __ xorl(rax, rax);
4273 __ jmpb(done);
4274 // Come here on success
4275 __ bind(ok_is_subtype);
4276 __ movl(rax, 1);
4277
4278 // Collect counts on whether this test sees NULLs a lot or not.
4279 if (ProfileInterpreter) {
4280 __ jmp(done);
4281 __ bind(is_null);
4282 __ profile_null_seen(rcx);
4283 } else {
4284 __ bind(is_null); // same as 'done'
4285 }
4286 __ bind(done);
4287 // rax = 0: obj == NULL or obj is not an instanceof the specified klass
4288 // rax = 1: obj != NULL and obj is an instanceof the specified klass
4289 }
4290
4291
4292 //----------------------------------------------------------------------------------------------------
4293 // Breakpoints
4294 void TemplateTable::_breakpoint() {
4295 // Note: We get here even if we are single stepping..
4296 // jbug insists on setting breakpoints at every bytecode
4297 // even if we are in single step mode.
4298
4299 transition(vtos, vtos);
4300
4301 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rcx);
4302
4303 // get the unpatched byte code
4304 __ get_method(rarg);
4305 __ call_VM(noreg,
4306 CAST_FROM_FN_PTR(address,
4307 InterpreterRuntime::get_original_bytecode_at),
4308 rarg, rbcp);
4309 __ mov(rbx, rax); // why?
4310
4311 // post the breakpoint event
4312 __ get_method(rarg);
4313 __ call_VM(noreg,
4314 CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint),
4315 rarg, rbcp);
4316
4317 // complete the execution of original bytecode
4318 __ dispatch_only_normal(vtos);
4319 }
4320
4321 //-----------------------------------------------------------------------------
4322 // Exceptions
4323
4324 void TemplateTable::athrow() {
4325 transition(atos, vtos);
4326 __ null_check(rax);
4327 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
4328 }
4329
4330 //-----------------------------------------------------------------------------
4331 // Synchronization
4332 //
4333 // Note: monitorenter & exit are symmetric routines; which is reflected
4334 // in the assembly code structure as well
4335 //
4336 // Stack layout:
4337 //
4338 // [expressions ] <--- rsp = expression stack top
4339 // ..
4340 // [expressions ]
4341 // [monitor entry] <--- monitor block top = expression stack bot
4342 // ..
4343 // [monitor entry]
4344 // [frame data ] <--- monitor block bot
4345 // ...
4346 // [saved rbp ] <--- rbp
4347 void TemplateTable::monitorenter() {
4348 transition(atos, vtos);
4349
4350 // check for NULL object
4351 __ null_check(rax);
4352
4353 __ resolve(IS_NOT_NULL, rax);
4354
4355 const Address monitor_block_top(
4356 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4357 const Address monitor_block_bot(
4358 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4359 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4360
4361 Label allocated;
4362
4363 Register rtop = LP64_ONLY(c_rarg3) NOT_LP64(rcx);
4364 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4365 Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4366
4367 // initialize entry pointer
4368 __ xorl(rmon, rmon); // points to free slot or NULL
4369
4370 // find a free slot in the monitor block (result in rmon)
4371 {
4372 Label entry, loop, exit;
4373 __ movptr(rtop, monitor_block_top); // points to current entry,
4374 // starting with top-most entry
4375 __ lea(rbot, monitor_block_bot); // points to word before bottom
4376 // of monitor block
4377 __ jmpb(entry);
4378
4379 __ bind(loop);
4380 // check if current entry is used
4381 __ cmpptr(Address(rtop, BasicObjectLock::obj_offset_in_bytes()), (int32_t) NULL_WORD);
4382 // if not used then remember entry in rmon
4383 __ cmovptr(Assembler::equal, rmon, rtop); // cmov => cmovptr
4384 // check if current entry is for same object
4385 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4386 // if same object then stop searching
4387 __ jccb(Assembler::equal, exit);
4388 // otherwise advance to next entry
4389 __ addptr(rtop, entry_size);
4390 __ bind(entry);
4391 // check if bottom reached
4392 __ cmpptr(rtop, rbot);
4393 // if not at bottom then check this entry
4394 __ jcc(Assembler::notEqual, loop);
4395 __ bind(exit);
4396 }
4397
4398 __ testptr(rmon, rmon); // check if a slot has been found
4399 __ jcc(Assembler::notZero, allocated); // if found, continue with that one
4400
4401 // allocate one if there's no free slot
4402 {
4403 Label entry, loop;
4404 // 1. compute new pointers // rsp: old expression stack top
4405 __ movptr(rmon, monitor_block_bot); // rmon: old expression stack bottom
4406 __ subptr(rsp, entry_size); // move expression stack top
4407 __ subptr(rmon, entry_size); // move expression stack bottom
4408 __ mov(rtop, rsp); // set start value for copy loop
4409 __ movptr(monitor_block_bot, rmon); // set new monitor block bottom
4410 __ jmp(entry);
4411 // 2. move expression stack contents
4412 __ bind(loop);
4413 __ movptr(rbot, Address(rtop, entry_size)); // load expression stack
4414 // word from old location
4415 __ movptr(Address(rtop, 0), rbot); // and store it at new location
4416 __ addptr(rtop, wordSize); // advance to next word
4417 __ bind(entry);
4418 __ cmpptr(rtop, rmon); // check if bottom reached
4419 __ jcc(Assembler::notEqual, loop); // if not at bottom then
4420 // copy next word
4421 }
4422
4423 // call run-time routine
4424 // rmon: points to monitor entry
4425 __ bind(allocated);
4426
4427 // Increment bcp to point to the next bytecode, so exception
4428 // handling for async. exceptions work correctly.
4429 // The object has already been poped from the stack, so the
4430 // expression stack looks correct.
4431 __ increment(rbcp);
4432
4433 // store object
4434 __ movptr(Address(rmon, BasicObjectLock::obj_offset_in_bytes()), rax);
4435 __ lock_object(rmon);
4436
4437 // check to make sure this monitor doesn't cause stack overflow after locking
4438 __ save_bcp(); // in case of exception
4439 __ generate_stack_overflow_check(0);
4440
4441 // The bcp has already been incremented. Just need to dispatch to
4442 // next instruction.
4443 __ dispatch_next(vtos);
4444 }
4445
4446 void TemplateTable::monitorexit() {
4447 transition(atos, vtos);
4448
4449 // check for NULL object
4450 __ null_check(rax);
4451
4452 __ resolve(IS_NOT_NULL, rax);
4453
4454 const Address monitor_block_top(
4455 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
4456 const Address monitor_block_bot(
4457 rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
4458 const int entry_size = frame::interpreter_frame_monitor_size() * wordSize;
4459
4460 Register rtop = LP64_ONLY(c_rarg1) NOT_LP64(rdx);
4461 Register rbot = LP64_ONLY(c_rarg2) NOT_LP64(rbx);
4462
4463 Label found;
4464
4465 // find matching slot
4466 {
4467 Label entry, loop;
4468 __ movptr(rtop, monitor_block_top); // points to current entry,
4469 // starting with top-most entry
4470 __ lea(rbot, monitor_block_bot); // points to word before bottom
4471 // of monitor block
4472 __ jmpb(entry);
4473
4474 __ bind(loop);
4475 // check if current entry is for same object
4476 __ cmpptr(rax, Address(rtop, BasicObjectLock::obj_offset_in_bytes()));
4477 // if same object then stop searching
4478 __ jcc(Assembler::equal, found);
4479 // otherwise advance to next entry
4480 __ addptr(rtop, entry_size);
4481 __ bind(entry);
4482 // check if bottom reached
4483 __ cmpptr(rtop, rbot);
4484 // if not at bottom then check this entry
4485 __ jcc(Assembler::notEqual, loop);
4486 }
4487
4488 // error handling. Unlocking was not block-structured
4489 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
4490 InterpreterRuntime::throw_illegal_monitor_state_exception));
4491 __ should_not_reach_here();
4492
4493 // call run-time routine
4494 __ bind(found);
4495 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
4496 __ unlock_object(rtop);
4497 __ pop_ptr(rax); // discard object
4498 }
4499
4500 // Wide instructions
4501 void TemplateTable::wide() {
4502 transition(vtos, vtos);
4503 __ load_unsigned_byte(rbx, at_bcp(1));
4504 ExternalAddress wtable((address)Interpreter::_wentry_point);
4505 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
4506 // Note: the rbcp increment step is part of the individual wide bytecode implementations
4507 }
4508
4509 // Multi arrays
4510 void TemplateTable::multianewarray() {
4511 transition(vtos, atos);
4512
4513 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rax);
4514 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
4515 // last dim is on top of stack; we want address of first one:
4516 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
4517 // the latter wordSize to point to the beginning of the array.
4518 __ lea(rarg, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
4519 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rarg);
4520 __ load_unsigned_byte(rbx, at_bcp(3));
4521 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
4522 }