Thu, 14 Apr 2011 13:45:41 -0700
Merge
1 /*
2 * Copyright (c) 1997, 2011, 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 */
25 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/templateTable.hpp"
30 #include "memory/universe.inline.hpp"
31 #include "oops/methodDataOop.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "oops/oop.inline.hpp"
34 #include "prims/methodHandles.hpp"
35 #include "runtime/sharedRuntime.hpp"
36 #include "runtime/stubRoutines.hpp"
37 #include "runtime/synchronizer.hpp"
39 #ifndef CC_INTERP
40 #define __ _masm->
42 //----------------------------------------------------------------------------------------------------
43 // Platform-dependent initialization
45 void TemplateTable::pd_initialize() {
46 // No i486 specific initialization
47 }
49 //----------------------------------------------------------------------------------------------------
50 // Address computation
52 // local variables
53 static inline Address iaddress(int n) {
54 return Address(rdi, Interpreter::local_offset_in_bytes(n));
55 }
57 static inline Address laddress(int n) { return iaddress(n + 1); }
58 static inline Address haddress(int n) { return iaddress(n + 0); }
59 static inline Address faddress(int n) { return iaddress(n); }
60 static inline Address daddress(int n) { return laddress(n); }
61 static inline Address aaddress(int n) { return iaddress(n); }
63 static inline Address iaddress(Register r) {
64 return Address(rdi, r, Interpreter::stackElementScale());
65 }
66 static inline Address laddress(Register r) {
67 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(1));
68 }
69 static inline Address haddress(Register r) {
70 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
71 }
73 static inline Address faddress(Register r) { return iaddress(r); }
74 static inline Address daddress(Register r) { return laddress(r); }
75 static inline Address aaddress(Register r) { return iaddress(r); }
77 // expression stack
78 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
79 // data beyond the rsp which is potentially unsafe in an MT environment;
80 // an interrupt may overwrite that data.)
81 static inline Address at_rsp () {
82 return Address(rsp, 0);
83 }
85 // At top of Java expression stack which may be different than rsp(). It
86 // isn't for category 1 objects.
87 static inline Address at_tos () {
88 Address tos = Address(rsp, Interpreter::expr_offset_in_bytes(0));
89 return tos;
90 }
92 static inline Address at_tos_p1() {
93 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
94 }
96 static inline Address at_tos_p2() {
97 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
98 }
100 // Condition conversion
101 static Assembler::Condition j_not(TemplateTable::Condition cc) {
102 switch (cc) {
103 case TemplateTable::equal : return Assembler::notEqual;
104 case TemplateTable::not_equal : return Assembler::equal;
105 case TemplateTable::less : return Assembler::greaterEqual;
106 case TemplateTable::less_equal : return Assembler::greater;
107 case TemplateTable::greater : return Assembler::lessEqual;
108 case TemplateTable::greater_equal: return Assembler::less;
109 }
110 ShouldNotReachHere();
111 return Assembler::zero;
112 }
115 //----------------------------------------------------------------------------------------------------
116 // Miscelaneous helper routines
118 // Store an oop (or NULL) at the address described by obj.
119 // If val == noreg this means store a NULL
121 static void do_oop_store(InterpreterMacroAssembler* _masm,
122 Address obj,
123 Register val,
124 BarrierSet::Name barrier,
125 bool precise) {
126 assert(val == noreg || val == rax, "parameter is just for looks");
127 switch (barrier) {
128 #ifndef SERIALGC
129 case BarrierSet::G1SATBCT:
130 case BarrierSet::G1SATBCTLogging:
131 {
132 // flatten object address if needed
133 // We do it regardless of precise because we need the registers
134 if (obj.index() == noreg && obj.disp() == 0) {
135 if (obj.base() != rdx) {
136 __ movl(rdx, obj.base());
137 }
138 } else {
139 __ leal(rdx, obj);
140 }
141 __ get_thread(rcx);
142 __ save_bcp();
143 __ g1_write_barrier_pre(rdx /* obj */,
144 rbx /* pre_val */,
145 rcx /* thread */,
146 rsi /* tmp */,
147 val != noreg /* tosca_live */,
148 false /* expand_call */);
150 // Do the actual store
151 // noreg means NULL
152 if (val == noreg) {
153 __ movptr(Address(rdx, 0), NULL_WORD);
154 // No post barrier for NULL
155 } else {
156 __ movl(Address(rdx, 0), val);
157 __ g1_write_barrier_post(rdx /* store_adr */,
158 val /* new_val */,
159 rcx /* thread */,
160 rbx /* tmp */,
161 rsi /* tmp2 */);
162 }
163 __ restore_bcp();
165 }
166 break;
167 #endif // SERIALGC
168 case BarrierSet::CardTableModRef:
169 case BarrierSet::CardTableExtension:
170 {
171 if (val == noreg) {
172 __ movptr(obj, NULL_WORD);
173 } else {
174 __ movl(obj, val);
175 // flatten object address if needed
176 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
177 __ store_check(obj.base());
178 } else {
179 __ leal(rdx, obj);
180 __ store_check(rdx);
181 }
182 }
183 }
184 break;
185 case BarrierSet::ModRef:
186 case BarrierSet::Other:
187 if (val == noreg) {
188 __ movptr(obj, NULL_WORD);
189 } else {
190 __ movl(obj, val);
191 }
192 break;
193 default :
194 ShouldNotReachHere();
196 }
197 }
199 Address TemplateTable::at_bcp(int offset) {
200 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
201 return Address(rsi, offset);
202 }
205 void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
206 Register scratch,
207 bool load_bc_into_scratch/*=true*/) {
209 if (!RewriteBytecodes) return;
210 // the pair bytecodes have already done the load.
211 if (load_bc_into_scratch) {
212 __ movl(bc, bytecode);
213 }
214 Label patch_done;
215 if (JvmtiExport::can_post_breakpoint()) {
216 Label fast_patch;
217 // if a breakpoint is present we can't rewrite the stream directly
218 __ movzbl(scratch, at_bcp(0));
219 __ cmpl(scratch, Bytecodes::_breakpoint);
220 __ jcc(Assembler::notEqual, fast_patch);
221 __ get_method(scratch);
222 // Let breakpoint table handling rewrite to quicker bytecode
223 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, rsi, bc);
224 #ifndef ASSERT
225 __ jmpb(patch_done);
226 #else
227 __ jmp(patch_done);
228 #endif
229 __ bind(fast_patch);
230 }
231 #ifdef ASSERT
232 Label okay;
233 __ load_unsigned_byte(scratch, at_bcp(0));
234 __ cmpl(scratch, (int)Bytecodes::java_code(bytecode));
235 __ jccb(Assembler::equal, okay);
236 __ cmpl(scratch, bc);
237 __ jcc(Assembler::equal, okay);
238 __ stop("patching the wrong bytecode");
239 __ bind(okay);
240 #endif
241 // patch bytecode
242 __ movb(at_bcp(0), bc);
243 __ bind(patch_done);
244 }
246 //----------------------------------------------------------------------------------------------------
247 // Individual instructions
249 void TemplateTable::nop() {
250 transition(vtos, vtos);
251 // nothing to do
252 }
254 void TemplateTable::shouldnotreachhere() {
255 transition(vtos, vtos);
256 __ stop("shouldnotreachhere bytecode");
257 }
261 void TemplateTable::aconst_null() {
262 transition(vtos, atos);
263 __ xorptr(rax, rax);
264 }
267 void TemplateTable::iconst(int value) {
268 transition(vtos, itos);
269 if (value == 0) {
270 __ xorptr(rax, rax);
271 } else {
272 __ movptr(rax, value);
273 }
274 }
277 void TemplateTable::lconst(int value) {
278 transition(vtos, ltos);
279 if (value == 0) {
280 __ xorptr(rax, rax);
281 } else {
282 __ movptr(rax, value);
283 }
284 assert(value >= 0, "check this code");
285 __ xorptr(rdx, rdx);
286 }
289 void TemplateTable::fconst(int value) {
290 transition(vtos, ftos);
291 if (value == 0) { __ fldz();
292 } else if (value == 1) { __ fld1();
293 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
294 } else { ShouldNotReachHere();
295 }
296 }
299 void TemplateTable::dconst(int value) {
300 transition(vtos, dtos);
301 if (value == 0) { __ fldz();
302 } else if (value == 1) { __ fld1();
303 } else { ShouldNotReachHere();
304 }
305 }
308 void TemplateTable::bipush() {
309 transition(vtos, itos);
310 __ load_signed_byte(rax, at_bcp(1));
311 }
314 void TemplateTable::sipush() {
315 transition(vtos, itos);
316 __ load_unsigned_short(rax, at_bcp(1));
317 __ bswapl(rax);
318 __ sarl(rax, 16);
319 }
321 void TemplateTable::ldc(bool wide) {
322 transition(vtos, vtos);
323 Label call_ldc, notFloat, notClass, Done;
325 if (wide) {
326 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
327 } else {
328 __ load_unsigned_byte(rbx, at_bcp(1));
329 }
330 __ get_cpool_and_tags(rcx, rax);
331 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
332 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
334 // get type
335 __ xorptr(rdx, rdx);
336 __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset));
338 // unresolved string - get the resolved string
339 __ cmpl(rdx, JVM_CONSTANT_UnresolvedString);
340 __ jccb(Assembler::equal, call_ldc);
342 // unresolved class - get the resolved class
343 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
344 __ jccb(Assembler::equal, call_ldc);
346 // unresolved class in error (resolution failed) - call into runtime
347 // so that the same error from first resolution attempt is thrown.
348 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
349 __ jccb(Assembler::equal, call_ldc);
351 // resolved class - need to call vm to get java mirror of the class
352 __ cmpl(rdx, JVM_CONSTANT_Class);
353 __ jcc(Assembler::notEqual, notClass);
355 __ bind(call_ldc);
356 __ movl(rcx, wide);
357 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx);
358 __ push(atos);
359 __ jmp(Done);
361 __ bind(notClass);
362 __ cmpl(rdx, JVM_CONSTANT_Float);
363 __ jccb(Assembler::notEqual, notFloat);
364 // ftos
365 __ fld_s( Address(rcx, rbx, Address::times_ptr, base_offset));
366 __ push(ftos);
367 __ jmp(Done);
369 __ bind(notFloat);
370 #ifdef ASSERT
371 { Label L;
372 __ cmpl(rdx, JVM_CONSTANT_Integer);
373 __ jcc(Assembler::equal, L);
374 __ cmpl(rdx, JVM_CONSTANT_String);
375 __ jcc(Assembler::equal, L);
376 __ stop("unexpected tag type in ldc");
377 __ bind(L);
378 }
379 #endif
380 Label isOop;
381 // atos and itos
382 // String is only oop type we will see here
383 __ cmpl(rdx, JVM_CONSTANT_String);
384 __ jccb(Assembler::equal, isOop);
385 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
386 __ push(itos);
387 __ jmp(Done);
388 __ bind(isOop);
389 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
390 __ push(atos);
392 if (VerifyOops) {
393 __ verify_oop(rax);
394 }
395 __ bind(Done);
396 }
398 // Fast path for caching oop constants.
399 // %%% We should use this to handle Class and String constants also.
400 // %%% It will simplify the ldc/primitive path considerably.
401 void TemplateTable::fast_aldc(bool wide) {
402 transition(vtos, atos);
404 if (!EnableInvokeDynamic) {
405 // We should not encounter this bytecode if !EnableInvokeDynamic.
406 // The verifier will stop it. However, if we get past the verifier,
407 // this will stop the thread in a reasonable way, without crashing the JVM.
408 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
409 InterpreterRuntime::throw_IncompatibleClassChangeError));
410 // the call_VM checks for exception, so we should never return here.
411 __ should_not_reach_here();
412 return;
413 }
415 const Register cache = rcx;
416 const Register index = rdx;
418 resolve_cache_and_index(f1_oop, rax, cache, index, wide ? sizeof(u2) : sizeof(u1));
419 if (VerifyOops) {
420 __ verify_oop(rax);
421 }
423 Label L_done, L_throw_exception;
424 const Register con_klass_temp = rcx; // same as Rcache
425 __ movptr(con_klass_temp, Address(rax, oopDesc::klass_offset_in_bytes()));
426 __ cmpptr(con_klass_temp, ExternalAddress((address)Universe::systemObjArrayKlassObj_addr()));
427 __ jcc(Assembler::notEqual, L_done);
428 __ cmpl(Address(rax, arrayOopDesc::length_offset_in_bytes()), 0);
429 __ jcc(Assembler::notEqual, L_throw_exception);
430 __ xorptr(rax, rax);
431 __ jmp(L_done);
433 // Load the exception from the system-array which wraps it:
434 __ bind(L_throw_exception);
435 __ movptr(rax, Address(rax, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
436 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
438 __ bind(L_done);
439 }
441 void TemplateTable::ldc2_w() {
442 transition(vtos, vtos);
443 Label Long, Done;
444 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
446 __ get_cpool_and_tags(rcx, rax);
447 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
448 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
450 // get type
451 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double);
452 __ jccb(Assembler::notEqual, Long);
453 // dtos
454 __ fld_d( Address(rcx, rbx, Address::times_ptr, base_offset));
455 __ push(dtos);
456 __ jmpb(Done);
458 __ bind(Long);
459 // ltos
460 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
461 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
463 __ push(ltos);
465 __ bind(Done);
466 }
469 void TemplateTable::locals_index(Register reg, int offset) {
470 __ load_unsigned_byte(reg, at_bcp(offset));
471 __ negptr(reg);
472 }
475 void TemplateTable::iload() {
476 transition(vtos, itos);
477 if (RewriteFrequentPairs) {
478 Label rewrite, done;
480 // get next byte
481 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
482 // if _iload, wait to rewrite to iload2. We only want to rewrite the
483 // last two iloads in a pair. Comparing against fast_iload means that
484 // the next bytecode is neither an iload or a caload, and therefore
485 // an iload pair.
486 __ cmpl(rbx, Bytecodes::_iload);
487 __ jcc(Assembler::equal, done);
489 __ cmpl(rbx, Bytecodes::_fast_iload);
490 __ movl(rcx, Bytecodes::_fast_iload2);
491 __ jccb(Assembler::equal, rewrite);
493 // if _caload, rewrite to fast_icaload
494 __ cmpl(rbx, Bytecodes::_caload);
495 __ movl(rcx, Bytecodes::_fast_icaload);
496 __ jccb(Assembler::equal, rewrite);
498 // rewrite so iload doesn't check again.
499 __ movl(rcx, Bytecodes::_fast_iload);
501 // rewrite
502 // rcx: fast bytecode
503 __ bind(rewrite);
504 patch_bytecode(Bytecodes::_iload, rcx, rbx, false);
505 __ bind(done);
506 }
508 // Get the local value into tos
509 locals_index(rbx);
510 __ movl(rax, iaddress(rbx));
511 }
514 void TemplateTable::fast_iload2() {
515 transition(vtos, itos);
516 locals_index(rbx);
517 __ movl(rax, iaddress(rbx));
518 __ push(itos);
519 locals_index(rbx, 3);
520 __ movl(rax, iaddress(rbx));
521 }
523 void TemplateTable::fast_iload() {
524 transition(vtos, itos);
525 locals_index(rbx);
526 __ movl(rax, iaddress(rbx));
527 }
530 void TemplateTable::lload() {
531 transition(vtos, ltos);
532 locals_index(rbx);
533 __ movptr(rax, laddress(rbx));
534 NOT_LP64(__ movl(rdx, haddress(rbx)));
535 }
538 void TemplateTable::fload() {
539 transition(vtos, ftos);
540 locals_index(rbx);
541 __ fld_s(faddress(rbx));
542 }
545 void TemplateTable::dload() {
546 transition(vtos, dtos);
547 locals_index(rbx);
548 __ fld_d(daddress(rbx));
549 }
552 void TemplateTable::aload() {
553 transition(vtos, atos);
554 locals_index(rbx);
555 __ movptr(rax, aaddress(rbx));
556 }
559 void TemplateTable::locals_index_wide(Register reg) {
560 __ movl(reg, at_bcp(2));
561 __ bswapl(reg);
562 __ shrl(reg, 16);
563 __ negptr(reg);
564 }
567 void TemplateTable::wide_iload() {
568 transition(vtos, itos);
569 locals_index_wide(rbx);
570 __ movl(rax, iaddress(rbx));
571 }
574 void TemplateTable::wide_lload() {
575 transition(vtos, ltos);
576 locals_index_wide(rbx);
577 __ movptr(rax, laddress(rbx));
578 NOT_LP64(__ movl(rdx, haddress(rbx)));
579 }
582 void TemplateTable::wide_fload() {
583 transition(vtos, ftos);
584 locals_index_wide(rbx);
585 __ fld_s(faddress(rbx));
586 }
589 void TemplateTable::wide_dload() {
590 transition(vtos, dtos);
591 locals_index_wide(rbx);
592 __ fld_d(daddress(rbx));
593 }
596 void TemplateTable::wide_aload() {
597 transition(vtos, atos);
598 locals_index_wide(rbx);
599 __ movptr(rax, aaddress(rbx));
600 }
602 void TemplateTable::index_check(Register array, Register index) {
603 // Pop ptr into array
604 __ pop_ptr(array);
605 index_check_without_pop(array, index);
606 }
608 void TemplateTable::index_check_without_pop(Register array, Register index) {
609 // destroys rbx,
610 // check array
611 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
612 LP64_ONLY(__ movslq(index, index));
613 // check index
614 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
615 if (index != rbx) {
616 // ??? convention: move aberrant index into rbx, for exception message
617 assert(rbx != array, "different registers");
618 __ mov(rbx, index);
619 }
620 __ jump_cc(Assembler::aboveEqual,
621 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
622 }
625 void TemplateTable::iaload() {
626 transition(itos, itos);
627 // rdx: array
628 index_check(rdx, rax); // kills rbx,
629 // rax,: index
630 __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)));
631 }
634 void TemplateTable::laload() {
635 transition(itos, ltos);
636 // rax,: index
637 // rdx: array
638 index_check(rdx, rax);
639 __ mov(rbx, rax);
640 // rbx,: index
641 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
642 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
643 }
646 void TemplateTable::faload() {
647 transition(itos, ftos);
648 // rdx: array
649 index_check(rdx, rax); // kills rbx,
650 // rax,: index
651 __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
652 }
655 void TemplateTable::daload() {
656 transition(itos, dtos);
657 // rdx: array
658 index_check(rdx, rax); // kills rbx,
659 // rax,: index
660 __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
661 }
664 void TemplateTable::aaload() {
665 transition(itos, atos);
666 // rdx: array
667 index_check(rdx, rax); // kills rbx,
668 // rax,: index
669 __ movptr(rax, Address(rdx, rax, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
670 }
673 void TemplateTable::baload() {
674 transition(itos, itos);
675 // rdx: array
676 index_check(rdx, rax); // kills rbx,
677 // rax,: index
678 // can do better code for P5 - fix this at some point
679 __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
680 __ mov(rax, rbx);
681 }
684 void TemplateTable::caload() {
685 transition(itos, itos);
686 // rdx: array
687 index_check(rdx, rax); // kills rbx,
688 // rax,: index
689 // can do better code for P5 - may want to improve this at some point
690 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
691 __ mov(rax, rbx);
692 }
694 // iload followed by caload frequent pair
695 void TemplateTable::fast_icaload() {
696 transition(vtos, itos);
697 // load index out of locals
698 locals_index(rbx);
699 __ movl(rax, iaddress(rbx));
701 // rdx: array
702 index_check(rdx, rax);
703 // rax,: index
704 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
705 __ mov(rax, rbx);
706 }
708 void TemplateTable::saload() {
709 transition(itos, itos);
710 // rdx: array
711 index_check(rdx, rax); // kills rbx,
712 // rax,: index
713 // can do better code for P5 - may want to improve this at some point
714 __ load_signed_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
715 __ mov(rax, rbx);
716 }
719 void TemplateTable::iload(int n) {
720 transition(vtos, itos);
721 __ movl(rax, iaddress(n));
722 }
725 void TemplateTable::lload(int n) {
726 transition(vtos, ltos);
727 __ movptr(rax, laddress(n));
728 NOT_LP64(__ movptr(rdx, haddress(n)));
729 }
732 void TemplateTable::fload(int n) {
733 transition(vtos, ftos);
734 __ fld_s(faddress(n));
735 }
738 void TemplateTable::dload(int n) {
739 transition(vtos, dtos);
740 __ fld_d(daddress(n));
741 }
744 void TemplateTable::aload(int n) {
745 transition(vtos, atos);
746 __ movptr(rax, aaddress(n));
747 }
750 void TemplateTable::aload_0() {
751 transition(vtos, atos);
752 // According to bytecode histograms, the pairs:
753 //
754 // _aload_0, _fast_igetfield
755 // _aload_0, _fast_agetfield
756 // _aload_0, _fast_fgetfield
757 //
758 // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
759 // bytecode checks if the next bytecode is either _fast_igetfield,
760 // _fast_agetfield or _fast_fgetfield and then rewrites the
761 // current bytecode into a pair bytecode; otherwise it rewrites the current
762 // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
763 //
764 // Note: If the next bytecode is _getfield, the rewrite must be delayed,
765 // otherwise we may miss an opportunity for a pair.
766 //
767 // Also rewrite frequent pairs
768 // aload_0, aload_1
769 // aload_0, iload_1
770 // These bytecodes with a small amount of code are most profitable to rewrite
771 if (RewriteFrequentPairs) {
772 Label rewrite, done;
773 // get next byte
774 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
776 // do actual aload_0
777 aload(0);
779 // if _getfield then wait with rewrite
780 __ cmpl(rbx, Bytecodes::_getfield);
781 __ jcc(Assembler::equal, done);
783 // if _igetfield then reqrite to _fast_iaccess_0
784 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
785 __ cmpl(rbx, Bytecodes::_fast_igetfield);
786 __ movl(rcx, Bytecodes::_fast_iaccess_0);
787 __ jccb(Assembler::equal, rewrite);
789 // if _agetfield then reqrite to _fast_aaccess_0
790 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
791 __ cmpl(rbx, Bytecodes::_fast_agetfield);
792 __ movl(rcx, Bytecodes::_fast_aaccess_0);
793 __ jccb(Assembler::equal, rewrite);
795 // if _fgetfield then reqrite to _fast_faccess_0
796 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
797 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
798 __ movl(rcx, Bytecodes::_fast_faccess_0);
799 __ jccb(Assembler::equal, rewrite);
801 // else rewrite to _fast_aload0
802 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
803 __ movl(rcx, Bytecodes::_fast_aload_0);
805 // rewrite
806 // rcx: fast bytecode
807 __ bind(rewrite);
808 patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false);
810 __ bind(done);
811 } else {
812 aload(0);
813 }
814 }
816 void TemplateTable::istore() {
817 transition(itos, vtos);
818 locals_index(rbx);
819 __ movl(iaddress(rbx), rax);
820 }
823 void TemplateTable::lstore() {
824 transition(ltos, vtos);
825 locals_index(rbx);
826 __ movptr(laddress(rbx), rax);
827 NOT_LP64(__ movptr(haddress(rbx), rdx));
828 }
831 void TemplateTable::fstore() {
832 transition(ftos, vtos);
833 locals_index(rbx);
834 __ fstp_s(faddress(rbx));
835 }
838 void TemplateTable::dstore() {
839 transition(dtos, vtos);
840 locals_index(rbx);
841 __ fstp_d(daddress(rbx));
842 }
845 void TemplateTable::astore() {
846 transition(vtos, vtos);
847 __ pop_ptr(rax);
848 locals_index(rbx);
849 __ movptr(aaddress(rbx), rax);
850 }
853 void TemplateTable::wide_istore() {
854 transition(vtos, vtos);
855 __ pop_i(rax);
856 locals_index_wide(rbx);
857 __ movl(iaddress(rbx), rax);
858 }
861 void TemplateTable::wide_lstore() {
862 transition(vtos, vtos);
863 __ pop_l(rax, rdx);
864 locals_index_wide(rbx);
865 __ movptr(laddress(rbx), rax);
866 NOT_LP64(__ movl(haddress(rbx), rdx));
867 }
870 void TemplateTable::wide_fstore() {
871 wide_istore();
872 }
875 void TemplateTable::wide_dstore() {
876 wide_lstore();
877 }
880 void TemplateTable::wide_astore() {
881 transition(vtos, vtos);
882 __ pop_ptr(rax);
883 locals_index_wide(rbx);
884 __ movptr(aaddress(rbx), rax);
885 }
888 void TemplateTable::iastore() {
889 transition(itos, vtos);
890 __ pop_i(rbx);
891 // rax,: value
892 // rdx: array
893 index_check(rdx, rbx); // prefer index in rbx,
894 // rbx,: index
895 __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax);
896 }
899 void TemplateTable::lastore() {
900 transition(ltos, vtos);
901 __ pop_i(rbx);
902 // rax,: low(value)
903 // rcx: array
904 // rdx: high(value)
905 index_check(rcx, rbx); // prefer index in rbx,
906 // rbx,: index
907 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
908 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
909 }
912 void TemplateTable::fastore() {
913 transition(ftos, vtos);
914 __ pop_i(rbx);
915 // rdx: array
916 // st0: value
917 index_check(rdx, rbx); // prefer index in rbx,
918 // rbx,: index
919 __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
920 }
923 void TemplateTable::dastore() {
924 transition(dtos, vtos);
925 __ pop_i(rbx);
926 // rdx: array
927 // st0: value
928 index_check(rdx, rbx); // prefer index in rbx,
929 // rbx,: index
930 __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
931 }
934 void TemplateTable::aastore() {
935 Label is_null, ok_is_subtype, done;
936 transition(vtos, vtos);
937 // stack: ..., array, index, value
938 __ movptr(rax, at_tos()); // Value
939 __ movl(rcx, at_tos_p1()); // Index
940 __ movptr(rdx, at_tos_p2()); // Array
942 Address element_address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
943 index_check_without_pop(rdx, rcx); // kills rbx,
944 // do array store check - check for NULL value first
945 __ testptr(rax, rax);
946 __ jcc(Assembler::zero, is_null);
948 // Move subklass into EBX
949 __ movptr(rbx, Address(rax, oopDesc::klass_offset_in_bytes()));
950 // Move superklass into EAX
951 __ movptr(rax, Address(rdx, oopDesc::klass_offset_in_bytes()));
952 __ movptr(rax, Address(rax, sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes()));
953 // Compress array+index*wordSize+12 into a single register. Frees ECX.
954 __ lea(rdx, element_address);
956 // Generate subtype check. Blows ECX. Resets EDI to locals.
957 // Superklass in EAX. Subklass in EBX.
958 __ gen_subtype_check( rbx, ok_is_subtype );
960 // Come here on failure
961 // object is at TOS
962 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
964 // Come here on success
965 __ bind(ok_is_subtype);
967 // Get the value to store
968 __ movptr(rax, at_rsp());
969 // and store it with appropriate barrier
970 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
972 __ jmp(done);
974 // Have a NULL in EAX, EDX=array, ECX=index. Store NULL at ary[idx]
975 __ bind(is_null);
976 __ profile_null_seen(rbx);
978 // Store NULL, (noreg means NULL to do_oop_store)
979 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
981 // Pop stack arguments
982 __ bind(done);
983 __ addptr(rsp, 3 * Interpreter::stackElementSize);
984 }
987 void TemplateTable::bastore() {
988 transition(itos, vtos);
989 __ pop_i(rbx);
990 // rax,: value
991 // rdx: array
992 index_check(rdx, rbx); // prefer index in rbx,
993 // rbx,: index
994 __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax);
995 }
998 void TemplateTable::castore() {
999 transition(itos, vtos);
1000 __ pop_i(rbx);
1001 // rax,: value
1002 // rdx: array
1003 index_check(rdx, rbx); // prefer index in rbx,
1004 // rbx,: index
1005 __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax);
1006 }
1009 void TemplateTable::sastore() {
1010 castore();
1011 }
1014 void TemplateTable::istore(int n) {
1015 transition(itos, vtos);
1016 __ movl(iaddress(n), rax);
1017 }
1020 void TemplateTable::lstore(int n) {
1021 transition(ltos, vtos);
1022 __ movptr(laddress(n), rax);
1023 NOT_LP64(__ movptr(haddress(n), rdx));
1024 }
1027 void TemplateTable::fstore(int n) {
1028 transition(ftos, vtos);
1029 __ fstp_s(faddress(n));
1030 }
1033 void TemplateTable::dstore(int n) {
1034 transition(dtos, vtos);
1035 __ fstp_d(daddress(n));
1036 }
1039 void TemplateTable::astore(int n) {
1040 transition(vtos, vtos);
1041 __ pop_ptr(rax);
1042 __ movptr(aaddress(n), rax);
1043 }
1046 void TemplateTable::pop() {
1047 transition(vtos, vtos);
1048 __ addptr(rsp, Interpreter::stackElementSize);
1049 }
1052 void TemplateTable::pop2() {
1053 transition(vtos, vtos);
1054 __ addptr(rsp, 2*Interpreter::stackElementSize);
1055 }
1058 void TemplateTable::dup() {
1059 transition(vtos, vtos);
1060 // stack: ..., a
1061 __ load_ptr(0, rax);
1062 __ push_ptr(rax);
1063 // stack: ..., a, a
1064 }
1067 void TemplateTable::dup_x1() {
1068 transition(vtos, vtos);
1069 // stack: ..., a, b
1070 __ load_ptr( 0, rax); // load b
1071 __ load_ptr( 1, rcx); // load a
1072 __ store_ptr(1, rax); // store b
1073 __ store_ptr(0, rcx); // store a
1074 __ push_ptr(rax); // push b
1075 // stack: ..., b, a, b
1076 }
1079 void TemplateTable::dup_x2() {
1080 transition(vtos, vtos);
1081 // stack: ..., a, b, c
1082 __ load_ptr( 0, rax); // load c
1083 __ load_ptr( 2, rcx); // load a
1084 __ store_ptr(2, rax); // store c in a
1085 __ push_ptr(rax); // push c
1086 // stack: ..., c, b, c, c
1087 __ load_ptr( 2, rax); // load b
1088 __ store_ptr(2, rcx); // store a in b
1089 // stack: ..., c, a, c, c
1090 __ store_ptr(1, rax); // store b in c
1091 // stack: ..., c, a, b, c
1092 }
1095 void TemplateTable::dup2() {
1096 transition(vtos, vtos);
1097 // stack: ..., a, b
1098 __ load_ptr(1, rax); // load a
1099 __ push_ptr(rax); // push a
1100 __ load_ptr(1, rax); // load b
1101 __ push_ptr(rax); // push b
1102 // stack: ..., a, b, a, b
1103 }
1106 void TemplateTable::dup2_x1() {
1107 transition(vtos, vtos);
1108 // stack: ..., a, b, c
1109 __ load_ptr( 0, rcx); // load c
1110 __ load_ptr( 1, rax); // load b
1111 __ push_ptr(rax); // push b
1112 __ push_ptr(rcx); // push c
1113 // stack: ..., a, b, c, b, c
1114 __ store_ptr(3, rcx); // store c in b
1115 // stack: ..., a, c, c, b, c
1116 __ load_ptr( 4, rcx); // load a
1117 __ store_ptr(2, rcx); // store a in 2nd c
1118 // stack: ..., a, c, a, b, c
1119 __ store_ptr(4, rax); // store b in a
1120 // stack: ..., b, c, a, b, c
1121 // stack: ..., b, c, a, b, c
1122 }
1125 void TemplateTable::dup2_x2() {
1126 transition(vtos, vtos);
1127 // stack: ..., a, b, c, d
1128 __ load_ptr( 0, rcx); // load d
1129 __ load_ptr( 1, rax); // load c
1130 __ push_ptr(rax); // push c
1131 __ push_ptr(rcx); // push d
1132 // stack: ..., a, b, c, d, c, d
1133 __ load_ptr( 4, rax); // load b
1134 __ store_ptr(2, rax); // store b in d
1135 __ store_ptr(4, rcx); // store d in b
1136 // stack: ..., a, d, c, b, c, d
1137 __ load_ptr( 5, rcx); // load a
1138 __ load_ptr( 3, rax); // load c
1139 __ store_ptr(3, rcx); // store a in c
1140 __ store_ptr(5, rax); // store c in a
1141 // stack: ..., c, d, a, b, c, d
1142 // stack: ..., c, d, a, b, c, d
1143 }
1146 void TemplateTable::swap() {
1147 transition(vtos, vtos);
1148 // stack: ..., a, b
1149 __ load_ptr( 1, rcx); // load a
1150 __ load_ptr( 0, rax); // load b
1151 __ store_ptr(0, rcx); // store a in b
1152 __ store_ptr(1, rax); // store b in a
1153 // stack: ..., b, a
1154 }
1157 void TemplateTable::iop2(Operation op) {
1158 transition(itos, itos);
1159 switch (op) {
1160 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1161 case sub : __ mov(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1162 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1163 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1164 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1165 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1166 case shl : __ mov(rcx, rax); __ pop_i(rax); __ shll (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1167 case shr : __ mov(rcx, rax); __ pop_i(rax); __ sarl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1168 case ushr : __ mov(rcx, rax); __ pop_i(rax); __ shrl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1169 default : ShouldNotReachHere();
1170 }
1171 }
1174 void TemplateTable::lop2(Operation op) {
1175 transition(ltos, ltos);
1176 __ pop_l(rbx, rcx);
1177 switch (op) {
1178 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1179 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1180 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1181 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1182 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1183 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1184 default : ShouldNotReachHere();
1185 }
1186 }
1189 void TemplateTable::idiv() {
1190 transition(itos, itos);
1191 __ mov(rcx, rax);
1192 __ pop_i(rax);
1193 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1194 // they are not equal, one could do a normal division (no correction
1195 // needed), which may speed up this implementation for the common case.
1196 // (see also JVM spec., p.243 & p.271)
1197 __ corrected_idivl(rcx);
1198 }
1201 void TemplateTable::irem() {
1202 transition(itos, itos);
1203 __ mov(rcx, rax);
1204 __ pop_i(rax);
1205 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1206 // they are not equal, one could do a normal division (no correction
1207 // needed), which may speed up this implementation for the common case.
1208 // (see also JVM spec., p.243 & p.271)
1209 __ corrected_idivl(rcx);
1210 __ mov(rax, rdx);
1211 }
1214 void TemplateTable::lmul() {
1215 transition(ltos, ltos);
1216 __ pop_l(rbx, rcx);
1217 __ push(rcx); __ push(rbx);
1218 __ push(rdx); __ push(rax);
1219 __ lmul(2 * wordSize, 0);
1220 __ addptr(rsp, 4 * wordSize); // take off temporaries
1221 }
1224 void TemplateTable::ldiv() {
1225 transition(ltos, ltos);
1226 __ pop_l(rbx, rcx);
1227 __ push(rcx); __ push(rbx);
1228 __ push(rdx); __ push(rax);
1229 // check if y = 0
1230 __ orl(rax, rdx);
1231 __ jump_cc(Assembler::zero,
1232 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1233 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1234 __ addptr(rsp, 4 * wordSize); // take off temporaries
1235 }
1238 void TemplateTable::lrem() {
1239 transition(ltos, ltos);
1240 __ pop_l(rbx, rcx);
1241 __ push(rcx); __ push(rbx);
1242 __ push(rdx); __ push(rax);
1243 // check if y = 0
1244 __ orl(rax, rdx);
1245 __ jump_cc(Assembler::zero,
1246 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1247 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1248 __ addptr(rsp, 4 * wordSize);
1249 }
1252 void TemplateTable::lshl() {
1253 transition(itos, ltos);
1254 __ movl(rcx, rax); // get shift count
1255 __ pop_l(rax, rdx); // get shift value
1256 __ lshl(rdx, rax);
1257 }
1260 void TemplateTable::lshr() {
1261 transition(itos, ltos);
1262 __ mov(rcx, rax); // get shift count
1263 __ pop_l(rax, rdx); // get shift value
1264 __ lshr(rdx, rax, true);
1265 }
1268 void TemplateTable::lushr() {
1269 transition(itos, ltos);
1270 __ mov(rcx, rax); // get shift count
1271 __ pop_l(rax, rdx); // get shift value
1272 __ lshr(rdx, rax);
1273 }
1276 void TemplateTable::fop2(Operation op) {
1277 transition(ftos, ftos);
1278 switch (op) {
1279 case add: __ fadd_s (at_rsp()); break;
1280 case sub: __ fsubr_s(at_rsp()); break;
1281 case mul: __ fmul_s (at_rsp()); break;
1282 case div: __ fdivr_s(at_rsp()); break;
1283 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1284 default : ShouldNotReachHere();
1285 }
1286 __ f2ieee();
1287 __ pop(rax); // pop float thing off
1288 }
1291 void TemplateTable::dop2(Operation op) {
1292 transition(dtos, dtos);
1294 switch (op) {
1295 case add: __ fadd_d (at_rsp()); break;
1296 case sub: __ fsubr_d(at_rsp()); break;
1297 case mul: {
1298 Label L_strict;
1299 Label L_join;
1300 const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1301 __ get_method(rcx);
1302 __ movl(rcx, access_flags);
1303 __ testl(rcx, JVM_ACC_STRICT);
1304 __ jccb(Assembler::notZero, L_strict);
1305 __ fmul_d (at_rsp());
1306 __ jmpb(L_join);
1307 __ bind(L_strict);
1308 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1309 __ fmulp();
1310 __ fmul_d (at_rsp());
1311 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1312 __ fmulp();
1313 __ bind(L_join);
1314 break;
1315 }
1316 case div: {
1317 Label L_strict;
1318 Label L_join;
1319 const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1320 __ get_method(rcx);
1321 __ movl(rcx, access_flags);
1322 __ testl(rcx, JVM_ACC_STRICT);
1323 __ jccb(Assembler::notZero, L_strict);
1324 __ fdivr_d(at_rsp());
1325 __ jmp(L_join);
1326 __ bind(L_strict);
1327 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1328 __ fmul_d (at_rsp());
1329 __ fdivrp();
1330 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1331 __ fmulp();
1332 __ bind(L_join);
1333 break;
1334 }
1335 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1336 default : ShouldNotReachHere();
1337 }
1338 __ d2ieee();
1339 // Pop double precision number from rsp.
1340 __ pop(rax);
1341 __ pop(rdx);
1342 }
1345 void TemplateTable::ineg() {
1346 transition(itos, itos);
1347 __ negl(rax);
1348 }
1351 void TemplateTable::lneg() {
1352 transition(ltos, ltos);
1353 __ lneg(rdx, rax);
1354 }
1357 void TemplateTable::fneg() {
1358 transition(ftos, ftos);
1359 __ fchs();
1360 }
1363 void TemplateTable::dneg() {
1364 transition(dtos, dtos);
1365 __ fchs();
1366 }
1369 void TemplateTable::iinc() {
1370 transition(vtos, vtos);
1371 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1372 locals_index(rbx);
1373 __ addl(iaddress(rbx), rdx);
1374 }
1377 void TemplateTable::wide_iinc() {
1378 transition(vtos, vtos);
1379 __ movl(rdx, at_bcp(4)); // get constant
1380 locals_index_wide(rbx);
1381 __ bswapl(rdx); // swap bytes & sign-extend constant
1382 __ sarl(rdx, 16);
1383 __ addl(iaddress(rbx), rdx);
1384 // Note: should probably use only one movl to get both
1385 // the index and the constant -> fix this
1386 }
1389 void TemplateTable::convert() {
1390 // Checking
1391 #ifdef ASSERT
1392 { TosState tos_in = ilgl;
1393 TosState tos_out = ilgl;
1394 switch (bytecode()) {
1395 case Bytecodes::_i2l: // fall through
1396 case Bytecodes::_i2f: // fall through
1397 case Bytecodes::_i2d: // fall through
1398 case Bytecodes::_i2b: // fall through
1399 case Bytecodes::_i2c: // fall through
1400 case Bytecodes::_i2s: tos_in = itos; break;
1401 case Bytecodes::_l2i: // fall through
1402 case Bytecodes::_l2f: // fall through
1403 case Bytecodes::_l2d: tos_in = ltos; break;
1404 case Bytecodes::_f2i: // fall through
1405 case Bytecodes::_f2l: // fall through
1406 case Bytecodes::_f2d: tos_in = ftos; break;
1407 case Bytecodes::_d2i: // fall through
1408 case Bytecodes::_d2l: // fall through
1409 case Bytecodes::_d2f: tos_in = dtos; break;
1410 default : ShouldNotReachHere();
1411 }
1412 switch (bytecode()) {
1413 case Bytecodes::_l2i: // fall through
1414 case Bytecodes::_f2i: // fall through
1415 case Bytecodes::_d2i: // fall through
1416 case Bytecodes::_i2b: // fall through
1417 case Bytecodes::_i2c: // fall through
1418 case Bytecodes::_i2s: tos_out = itos; break;
1419 case Bytecodes::_i2l: // fall through
1420 case Bytecodes::_f2l: // fall through
1421 case Bytecodes::_d2l: tos_out = ltos; break;
1422 case Bytecodes::_i2f: // fall through
1423 case Bytecodes::_l2f: // fall through
1424 case Bytecodes::_d2f: tos_out = ftos; break;
1425 case Bytecodes::_i2d: // fall through
1426 case Bytecodes::_l2d: // fall through
1427 case Bytecodes::_f2d: tos_out = dtos; break;
1428 default : ShouldNotReachHere();
1429 }
1430 transition(tos_in, tos_out);
1431 }
1432 #endif // ASSERT
1434 // Conversion
1435 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1436 switch (bytecode()) {
1437 case Bytecodes::_i2l:
1438 __ extend_sign(rdx, rax);
1439 break;
1440 case Bytecodes::_i2f:
1441 __ push(rax); // store int on tos
1442 __ fild_s(at_rsp()); // load int to ST0
1443 __ f2ieee(); // truncate to float size
1444 __ pop(rcx); // adjust rsp
1445 break;
1446 case Bytecodes::_i2d:
1447 __ push(rax); // add one slot for d2ieee()
1448 __ push(rax); // store int on tos
1449 __ fild_s(at_rsp()); // load int to ST0
1450 __ d2ieee(); // truncate to double size
1451 __ pop(rcx); // adjust rsp
1452 __ pop(rcx);
1453 break;
1454 case Bytecodes::_i2b:
1455 __ shll(rax, 24); // truncate upper 24 bits
1456 __ sarl(rax, 24); // and sign-extend byte
1457 LP64_ONLY(__ movsbl(rax, rax));
1458 break;
1459 case Bytecodes::_i2c:
1460 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1461 LP64_ONLY(__ movzwl(rax, rax));
1462 break;
1463 case Bytecodes::_i2s:
1464 __ shll(rax, 16); // truncate upper 16 bits
1465 __ sarl(rax, 16); // and sign-extend short
1466 LP64_ONLY(__ movswl(rax, rax));
1467 break;
1468 case Bytecodes::_l2i:
1469 /* nothing to do */
1470 break;
1471 case Bytecodes::_l2f:
1472 __ push(rdx); // store long on tos
1473 __ push(rax);
1474 __ fild_d(at_rsp()); // load long to ST0
1475 __ f2ieee(); // truncate to float size
1476 __ pop(rcx); // adjust rsp
1477 __ pop(rcx);
1478 break;
1479 case Bytecodes::_l2d:
1480 __ push(rdx); // store long on tos
1481 __ push(rax);
1482 __ fild_d(at_rsp()); // load long to ST0
1483 __ d2ieee(); // truncate to double size
1484 __ pop(rcx); // adjust rsp
1485 __ pop(rcx);
1486 break;
1487 case Bytecodes::_f2i:
1488 __ push(rcx); // reserve space for argument
1489 __ fstp_s(at_rsp()); // pass float argument on stack
1490 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1491 break;
1492 case Bytecodes::_f2l:
1493 __ push(rcx); // reserve space for argument
1494 __ fstp_s(at_rsp()); // pass float argument on stack
1495 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1496 break;
1497 case Bytecodes::_f2d:
1498 /* nothing to do */
1499 break;
1500 case Bytecodes::_d2i:
1501 __ push(rcx); // reserve space for argument
1502 __ push(rcx);
1503 __ fstp_d(at_rsp()); // pass double argument on stack
1504 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1505 break;
1506 case Bytecodes::_d2l:
1507 __ push(rcx); // reserve space for argument
1508 __ push(rcx);
1509 __ fstp_d(at_rsp()); // pass double argument on stack
1510 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1511 break;
1512 case Bytecodes::_d2f:
1513 __ push(rcx); // reserve space for f2ieee()
1514 __ f2ieee(); // truncate to float size
1515 __ pop(rcx); // adjust rsp
1516 break;
1517 default :
1518 ShouldNotReachHere();
1519 }
1520 }
1523 void TemplateTable::lcmp() {
1524 transition(ltos, itos);
1525 // y = rdx:rax
1526 __ pop_l(rbx, rcx); // get x = rcx:rbx
1527 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1528 __ mov(rax, rcx);
1529 }
1532 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1533 if (is_float) {
1534 __ fld_s(at_rsp());
1535 } else {
1536 __ fld_d(at_rsp());
1537 __ pop(rdx);
1538 }
1539 __ pop(rcx);
1540 __ fcmp2int(rax, unordered_result < 0);
1541 }
1544 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1545 __ get_method(rcx); // ECX holds method
1546 __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count
1548 const ByteSize be_offset = methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset();
1549 const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset();
1550 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1552 // Load up EDX with the branch displacement
1553 __ movl(rdx, at_bcp(1));
1554 __ bswapl(rdx);
1555 if (!is_wide) __ sarl(rdx, 16);
1556 LP64_ONLY(__ movslq(rdx, rdx));
1559 // Handle all the JSR stuff here, then exit.
1560 // It's much shorter and cleaner than intermingling with the
1561 // non-JSR normal-branch stuff occurring below.
1562 if (is_jsr) {
1563 // Pre-load the next target bytecode into EBX
1564 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0));
1566 // compute return address as bci in rax,
1567 __ lea(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(constMethodOopDesc::codes_offset())));
1568 __ subptr(rax, Address(rcx, methodOopDesc::const_offset()));
1569 // Adjust the bcp in RSI by the displacement in EDX
1570 __ addptr(rsi, rdx);
1571 // Push return address
1572 __ push_i(rax);
1573 // jsr returns vtos
1574 __ dispatch_only_noverify(vtos);
1575 return;
1576 }
1578 // Normal (non-jsr) branch handling
1580 // Adjust the bcp in RSI by the displacement in EDX
1581 __ addptr(rsi, rdx);
1583 assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1584 Label backedge_counter_overflow;
1585 Label profile_method;
1586 Label dispatch;
1587 if (UseLoopCounter) {
1588 // increment backedge counter for backward branches
1589 // rax,: MDO
1590 // rbx,: MDO bumped taken-count
1591 // rcx: method
1592 // rdx: target offset
1593 // rsi: target bcp
1594 // rdi: locals pointer
1595 __ testl(rdx, rdx); // check if forward or backward branch
1596 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1598 if (TieredCompilation) {
1599 Label no_mdo;
1600 int increment = InvocationCounter::count_increment;
1601 int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1602 if (ProfileInterpreter) {
1603 // Are we profiling?
1604 __ movptr(rbx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1605 __ testptr(rbx, rbx);
1606 __ jccb(Assembler::zero, no_mdo);
1607 // Increment the MDO backedge counter
1608 const Address mdo_backedge_counter(rbx, in_bytes(methodDataOopDesc::backedge_counter_offset()) +
1609 in_bytes(InvocationCounter::counter_offset()));
1610 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1611 rax, false, Assembler::zero, &backedge_counter_overflow);
1612 __ jmp(dispatch);
1613 }
1614 __ bind(no_mdo);
1615 // Increment backedge counter in methodOop
1616 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
1617 rax, false, Assembler::zero, &backedge_counter_overflow);
1618 } else {
1619 // increment counter
1620 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1621 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1622 __ movl(Address(rcx, be_offset), rax); // store counter
1624 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1625 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1626 __ addl(rax, Address(rcx, be_offset)); // add both counters
1628 if (ProfileInterpreter) {
1629 // Test to see if we should create a method data oop
1630 __ cmp32(rax,
1631 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1632 __ jcc(Assembler::less, dispatch);
1634 // if no method data exists, go to profile method
1635 __ test_method_data_pointer(rax, profile_method);
1637 if (UseOnStackReplacement) {
1638 // check for overflow against rbx, which is the MDO taken count
1639 __ cmp32(rbx,
1640 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1641 __ jcc(Assembler::below, dispatch);
1643 // When ProfileInterpreter is on, the backedge_count comes from the
1644 // methodDataOop, which value does not get reset on the call to
1645 // frequency_counter_overflow(). To avoid excessive calls to the overflow
1646 // routine while the method is being compiled, add a second test to make
1647 // sure the overflow function is called only once every overflow_frequency.
1648 const int overflow_frequency = 1024;
1649 __ andptr(rbx, overflow_frequency-1);
1650 __ jcc(Assembler::zero, backedge_counter_overflow);
1651 }
1652 } else {
1653 if (UseOnStackReplacement) {
1654 // check for overflow against rax, which is the sum of the counters
1655 __ cmp32(rax,
1656 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1657 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1659 }
1660 }
1661 }
1662 __ bind(dispatch);
1663 }
1665 // Pre-load the next target bytecode into EBX
1666 __ load_unsigned_byte(rbx, Address(rsi, 0));
1668 // continue with the bytecode @ target
1669 // rax,: return bci for jsr's, unused otherwise
1670 // rbx,: target bytecode
1671 // rsi: target bcp
1672 __ dispatch_only(vtos);
1674 if (UseLoopCounter) {
1675 if (ProfileInterpreter) {
1676 // Out-of-line code to allocate method data oop.
1677 __ bind(profile_method);
1678 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1679 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1680 __ set_method_data_pointer_for_bcp();
1681 __ jmp(dispatch);
1682 }
1684 if (UseOnStackReplacement) {
1686 // invocation counter overflow
1687 __ bind(backedge_counter_overflow);
1688 __ negptr(rdx);
1689 __ addptr(rdx, rsi); // branch bcp
1690 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx);
1691 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1693 // rax,: osr nmethod (osr ok) or NULL (osr not possible)
1694 // rbx,: target bytecode
1695 // rdx: scratch
1696 // rdi: locals pointer
1697 // rsi: bcp
1698 __ testptr(rax, rax); // test result
1699 __ jcc(Assembler::zero, dispatch); // no osr if null
1700 // nmethod may have been invalidated (VM may block upon call_VM return)
1701 __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1702 __ cmpl(rcx, InvalidOSREntryBci);
1703 __ jcc(Assembler::equal, dispatch);
1705 // We have the address of an on stack replacement routine in rax,
1706 // We need to prepare to execute the OSR method. First we must
1707 // migrate the locals and monitors off of the stack.
1709 __ mov(rbx, rax); // save the nmethod
1711 const Register thread = rcx;
1712 __ get_thread(thread);
1713 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1714 // rax, is OSR buffer, move it to expected parameter location
1715 __ mov(rcx, rax);
1717 // pop the interpreter frame
1718 __ movptr(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1719 __ leave(); // remove frame anchor
1720 __ pop(rdi); // get return address
1721 __ mov(rsp, rdx); // set sp to sender sp
1723 // Align stack pointer for compiled code (note that caller is
1724 // responsible for undoing this fixup by remembering the old SP
1725 // in an rbp,-relative location)
1726 __ andptr(rsp, -(StackAlignmentInBytes));
1728 // push the (possibly adjusted) return address
1729 __ push(rdi);
1731 // and begin the OSR nmethod
1732 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
1733 }
1734 }
1735 }
1738 void TemplateTable::if_0cmp(Condition cc) {
1739 transition(itos, vtos);
1740 // assume branch is more often taken than not (loops use backward branches)
1741 Label not_taken;
1742 __ testl(rax, rax);
1743 __ jcc(j_not(cc), not_taken);
1744 branch(false, false);
1745 __ bind(not_taken);
1746 __ profile_not_taken_branch(rax);
1747 }
1750 void TemplateTable::if_icmp(Condition cc) {
1751 transition(itos, vtos);
1752 // assume branch is more often taken than not (loops use backward branches)
1753 Label not_taken;
1754 __ pop_i(rdx);
1755 __ cmpl(rdx, rax);
1756 __ jcc(j_not(cc), not_taken);
1757 branch(false, false);
1758 __ bind(not_taken);
1759 __ profile_not_taken_branch(rax);
1760 }
1763 void TemplateTable::if_nullcmp(Condition cc) {
1764 transition(atos, vtos);
1765 // assume branch is more often taken than not (loops use backward branches)
1766 Label not_taken;
1767 __ testptr(rax, rax);
1768 __ jcc(j_not(cc), not_taken);
1769 branch(false, false);
1770 __ bind(not_taken);
1771 __ profile_not_taken_branch(rax);
1772 }
1775 void TemplateTable::if_acmp(Condition cc) {
1776 transition(atos, vtos);
1777 // assume branch is more often taken than not (loops use backward branches)
1778 Label not_taken;
1779 __ pop_ptr(rdx);
1780 __ cmpptr(rdx, rax);
1781 __ jcc(j_not(cc), not_taken);
1782 branch(false, false);
1783 __ bind(not_taken);
1784 __ profile_not_taken_branch(rax);
1785 }
1788 void TemplateTable::ret() {
1789 transition(vtos, vtos);
1790 locals_index(rbx);
1791 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1792 __ profile_ret(rbx, rcx);
1793 __ get_method(rax);
1794 __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1795 __ lea(rsi, Address(rsi, rbx, Address::times_1,
1796 constMethodOopDesc::codes_offset()));
1797 __ dispatch_next(vtos);
1798 }
1801 void TemplateTable::wide_ret() {
1802 transition(vtos, vtos);
1803 locals_index_wide(rbx);
1804 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1805 __ profile_ret(rbx, rcx);
1806 __ get_method(rax);
1807 __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1808 __ lea(rsi, Address(rsi, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1809 __ dispatch_next(vtos);
1810 }
1813 void TemplateTable::tableswitch() {
1814 Label default_case, continue_execution;
1815 transition(itos, vtos);
1816 // align rsi
1817 __ lea(rbx, at_bcp(wordSize));
1818 __ andptr(rbx, -wordSize);
1819 // load lo & hi
1820 __ movl(rcx, Address(rbx, 1 * wordSize));
1821 __ movl(rdx, Address(rbx, 2 * wordSize));
1822 __ bswapl(rcx);
1823 __ bswapl(rdx);
1824 // check against lo & hi
1825 __ cmpl(rax, rcx);
1826 __ jccb(Assembler::less, default_case);
1827 __ cmpl(rax, rdx);
1828 __ jccb(Assembler::greater, default_case);
1829 // lookup dispatch offset
1830 __ subl(rax, rcx);
1831 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1832 __ profile_switch_case(rax, rbx, rcx);
1833 // continue execution
1834 __ bind(continue_execution);
1835 __ bswapl(rdx);
1836 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1837 __ addptr(rsi, rdx);
1838 __ dispatch_only(vtos);
1839 // handle default
1840 __ bind(default_case);
1841 __ profile_switch_default(rax);
1842 __ movl(rdx, Address(rbx, 0));
1843 __ jmp(continue_execution);
1844 }
1847 void TemplateTable::lookupswitch() {
1848 transition(itos, itos);
1849 __ stop("lookupswitch bytecode should have been rewritten");
1850 }
1853 void TemplateTable::fast_linearswitch() {
1854 transition(itos, vtos);
1855 Label loop_entry, loop, found, continue_execution;
1856 // bswapl rax, so we can avoid bswapping the table entries
1857 __ bswapl(rax);
1858 // align rsi
1859 __ lea(rbx, at_bcp(wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1860 __ andptr(rbx, -wordSize);
1861 // set counter
1862 __ movl(rcx, Address(rbx, wordSize));
1863 __ bswapl(rcx);
1864 __ jmpb(loop_entry);
1865 // table search
1866 __ bind(loop);
1867 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize));
1868 __ jccb(Assembler::equal, found);
1869 __ bind(loop_entry);
1870 __ decrementl(rcx);
1871 __ jcc(Assembler::greaterEqual, loop);
1872 // default case
1873 __ profile_switch_default(rax);
1874 __ movl(rdx, Address(rbx, 0));
1875 __ jmpb(continue_execution);
1876 // entry found -> get offset
1877 __ bind(found);
1878 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize));
1879 __ profile_switch_case(rcx, rax, rbx);
1880 // continue execution
1881 __ bind(continue_execution);
1882 __ bswapl(rdx);
1883 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1884 __ addptr(rsi, rdx);
1885 __ dispatch_only(vtos);
1886 }
1889 void TemplateTable::fast_binaryswitch() {
1890 transition(itos, vtos);
1891 // Implementation using the following core algorithm:
1892 //
1893 // int binary_search(int key, LookupswitchPair* array, int n) {
1894 // // Binary search according to "Methodik des Programmierens" by
1895 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1896 // int i = 0;
1897 // int j = n;
1898 // while (i+1 < j) {
1899 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1900 // // with Q: for all i: 0 <= i < n: key < a[i]
1901 // // where a stands for the array and assuming that the (inexisting)
1902 // // element a[n] is infinitely big.
1903 // int h = (i + j) >> 1;
1904 // // i < h < j
1905 // if (key < array[h].fast_match()) {
1906 // j = h;
1907 // } else {
1908 // i = h;
1909 // }
1910 // }
1911 // // R: a[i] <= key < a[i+1] or Q
1912 // // (i.e., if key is within array, i is the correct index)
1913 // return i;
1914 // }
1916 // register allocation
1917 const Register key = rax; // already set (tosca)
1918 const Register array = rbx;
1919 const Register i = rcx;
1920 const Register j = rdx;
1921 const Register h = rdi; // needs to be restored
1922 const Register temp = rsi;
1923 // setup array
1924 __ save_bcp();
1926 __ lea(array, at_bcp(3*wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1927 __ andptr(array, -wordSize);
1928 // initialize i & j
1929 __ xorl(i, i); // i = 0;
1930 __ movl(j, Address(array, -wordSize)); // j = length(array);
1931 // Convert j into native byteordering
1932 __ bswapl(j);
1933 // and start
1934 Label entry;
1935 __ jmp(entry);
1937 // binary search loop
1938 { Label loop;
1939 __ bind(loop);
1940 // int h = (i + j) >> 1;
1941 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1942 __ sarl(h, 1); // h = (i + j) >> 1;
1943 // if (key < array[h].fast_match()) {
1944 // j = h;
1945 // } else {
1946 // i = h;
1947 // }
1948 // Convert array[h].match to native byte-ordering before compare
1949 __ movl(temp, Address(array, h, Address::times_8, 0*wordSize));
1950 __ bswapl(temp);
1951 __ cmpl(key, temp);
1952 // j = h if (key < array[h].fast_match())
1953 __ cmov32(Assembler::less , j, h);
1954 // i = h if (key >= array[h].fast_match())
1955 __ cmov32(Assembler::greaterEqual, i, h);
1956 // while (i+1 < j)
1957 __ bind(entry);
1958 __ leal(h, Address(i, 1)); // i+1
1959 __ cmpl(h, j); // i+1 < j
1960 __ jcc(Assembler::less, loop);
1961 }
1963 // end of binary search, result index is i (must check again!)
1964 Label default_case;
1965 // Convert array[i].match to native byte-ordering before compare
1966 __ movl(temp, Address(array, i, Address::times_8, 0*wordSize));
1967 __ bswapl(temp);
1968 __ cmpl(key, temp);
1969 __ jcc(Assembler::notEqual, default_case);
1971 // entry found -> j = offset
1972 __ movl(j , Address(array, i, Address::times_8, 1*wordSize));
1973 __ profile_switch_case(i, key, array);
1974 __ bswapl(j);
1975 LP64_ONLY(__ movslq(j, j));
1976 __ restore_bcp();
1977 __ restore_locals(); // restore rdi
1978 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1980 __ addptr(rsi, j);
1981 __ dispatch_only(vtos);
1983 // default case -> j = default offset
1984 __ bind(default_case);
1985 __ profile_switch_default(i);
1986 __ movl(j, Address(array, -2*wordSize));
1987 __ bswapl(j);
1988 LP64_ONLY(__ movslq(j, j));
1989 __ restore_bcp();
1990 __ restore_locals(); // restore rdi
1991 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1992 __ addptr(rsi, j);
1993 __ dispatch_only(vtos);
1994 }
1997 void TemplateTable::_return(TosState state) {
1998 transition(state, state);
1999 assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation
2001 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2002 assert(state == vtos, "only valid state");
2003 __ movptr(rax, aaddress(0));
2004 __ movptr(rdi, Address(rax, oopDesc::klass_offset_in_bytes()));
2005 __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
2006 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2007 Label skip_register_finalizer;
2008 __ jcc(Assembler::zero, skip_register_finalizer);
2010 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax);
2012 __ bind(skip_register_finalizer);
2013 }
2015 __ remove_activation(state, rsi);
2016 __ jmp(rsi);
2017 }
2020 // ----------------------------------------------------------------------------
2021 // Volatile variables demand their effects be made known to all CPU's in
2022 // order. Store buffers on most chips allow reads & writes to reorder; the
2023 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2024 // memory barrier (i.e., it's not sufficient that the interpreter does not
2025 // reorder volatile references, the hardware also must not reorder them).
2026 //
2027 // According to the new Java Memory Model (JMM):
2028 // (1) All volatiles are serialized wrt to each other.
2029 // ALSO reads & writes act as aquire & release, so:
2030 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
2031 // the read float up to before the read. It's OK for non-volatile memory refs
2032 // that happen before the volatile read to float down below it.
2033 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2034 // that happen BEFORE the write float down to after the write. It's OK for
2035 // non-volatile memory refs that happen after the volatile write to float up
2036 // before it.
2037 //
2038 // We only put in barriers around volatile refs (they are expensive), not
2039 // _between_ memory refs (that would require us to track the flavor of the
2040 // previous memory refs). Requirements (2) and (3) require some barriers
2041 // before volatile stores and after volatile loads. These nearly cover
2042 // requirement (1) but miss the volatile-store-volatile-load case. This final
2043 // case is placed after volatile-stores although it could just as well go
2044 // before volatile-loads.
2045 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2046 // Helper function to insert a is-volatile test and memory barrier
2047 if( !os::is_MP() ) return; // Not needed on single CPU
2048 __ membar(order_constraint);
2049 }
2051 void TemplateTable::resolve_cache_and_index(int byte_no,
2052 Register result,
2053 Register Rcache,
2054 Register index,
2055 size_t index_size) {
2056 Register temp = rbx;
2058 assert_different_registers(result, Rcache, index, temp);
2060 Label resolved;
2061 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2062 if (byte_no == f1_oop) {
2063 // We are resolved if the f1 field contains a non-null object (CallSite, etc.)
2064 // This kind of CP cache entry does not need to match the flags byte, because
2065 // there is a 1-1 relation between bytecode type and CP entry type.
2066 assert(result != noreg, ""); //else do cmpptr(Address(...), (int32_t) NULL_WORD)
2067 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2068 __ testptr(result, result);
2069 __ jcc(Assembler::notEqual, resolved);
2070 } else {
2071 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2072 assert(result == noreg, ""); //else change code for setting result
2073 const int shift_count = (1 + byte_no)*BitsPerByte;
2074 __ movl(temp, Address(Rcache, index, Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
2075 __ shrl(temp, shift_count);
2076 // have we resolved this bytecode?
2077 __ andl(temp, 0xFF);
2078 __ cmpl(temp, (int)bytecode());
2079 __ jcc(Assembler::equal, resolved);
2080 }
2082 // resolve first time through
2083 address entry;
2084 switch (bytecode()) {
2085 case Bytecodes::_getstatic : // fall through
2086 case Bytecodes::_putstatic : // fall through
2087 case Bytecodes::_getfield : // fall through
2088 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2089 case Bytecodes::_invokevirtual : // fall through
2090 case Bytecodes::_invokespecial : // fall through
2091 case Bytecodes::_invokestatic : // fall through
2092 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2093 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2094 case Bytecodes::_fast_aldc : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); break;
2095 case Bytecodes::_fast_aldc_w : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); break;
2096 default : ShouldNotReachHere(); break;
2097 }
2098 __ movl(temp, (int)bytecode());
2099 __ call_VM(noreg, entry, temp);
2100 // Update registers with resolved info
2101 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2102 if (result != noreg)
2103 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2104 __ bind(resolved);
2105 }
2108 // The cache and index registers must be set before call
2109 void TemplateTable::load_field_cp_cache_entry(Register obj,
2110 Register cache,
2111 Register index,
2112 Register off,
2113 Register flags,
2114 bool is_static = false) {
2115 assert_different_registers(cache, index, flags, off);
2117 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2118 // Field offset
2119 __ movptr(off, Address(cache, index, Address::times_ptr,
2120 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset())));
2121 // Flags
2122 __ movl(flags, Address(cache, index, Address::times_ptr,
2123 in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset())));
2125 // klass overwrite register
2126 if (is_static) {
2127 __ movptr(obj, Address(cache, index, Address::times_ptr,
2128 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset())));
2129 }
2130 }
2132 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2133 Register method,
2134 Register itable_index,
2135 Register flags,
2136 bool is_invokevirtual,
2137 bool is_invokevfinal /*unused*/,
2138 bool is_invokedynamic) {
2139 // setup registers
2140 const Register cache = rcx;
2141 const Register index = rdx;
2142 assert_different_registers(method, flags);
2143 assert_different_registers(method, cache, index);
2144 assert_different_registers(itable_index, flags);
2145 assert_different_registers(itable_index, cache, index);
2146 // determine constant pool cache field offsets
2147 const int method_offset = in_bytes(
2148 constantPoolCacheOopDesc::base_offset() +
2149 (is_invokevirtual
2150 ? ConstantPoolCacheEntry::f2_offset()
2151 : ConstantPoolCacheEntry::f1_offset()
2152 )
2153 );
2154 const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2155 ConstantPoolCacheEntry::flags_offset());
2156 // access constant pool cache fields
2157 const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2158 ConstantPoolCacheEntry::f2_offset());
2160 if (byte_no == f1_oop) {
2161 // Resolved f1_oop goes directly into 'method' register.
2162 assert(is_invokedynamic, "");
2163 resolve_cache_and_index(byte_no, method, cache, index, sizeof(u4));
2164 } else {
2165 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2166 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2167 }
2168 if (itable_index != noreg) {
2169 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2170 }
2171 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2172 }
2175 // The registers cache and index expected to be set before call.
2176 // Correct values of the cache and index registers are preserved.
2177 void TemplateTable::jvmti_post_field_access(Register cache,
2178 Register index,
2179 bool is_static,
2180 bool has_tos) {
2181 if (JvmtiExport::can_post_field_access()) {
2182 // Check to see if a field access watch has been set before we take
2183 // the time to call into the VM.
2184 Label L1;
2185 assert_different_registers(cache, index, rax);
2186 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2187 __ testl(rax,rax);
2188 __ jcc(Assembler::zero, L1);
2190 // cache entry pointer
2191 __ addptr(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
2192 __ shll(index, LogBytesPerWord);
2193 __ addptr(cache, index);
2194 if (is_static) {
2195 __ xorptr(rax, rax); // NULL object reference
2196 } else {
2197 __ pop(atos); // Get the object
2198 __ verify_oop(rax);
2199 __ push(atos); // Restore stack state
2200 }
2201 // rax,: object pointer or NULL
2202 // cache: cache entry pointer
2203 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2204 rax, cache);
2205 __ get_cache_and_index_at_bcp(cache, index, 1);
2206 __ bind(L1);
2207 }
2208 }
2210 void TemplateTable::pop_and_check_object(Register r) {
2211 __ pop_ptr(r);
2212 __ null_check(r); // for field access must check obj.
2213 __ verify_oop(r);
2214 }
2216 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2217 transition(vtos, vtos);
2219 const Register cache = rcx;
2220 const Register index = rdx;
2221 const Register obj = rcx;
2222 const Register off = rbx;
2223 const Register flags = rax;
2225 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2226 jvmti_post_field_access(cache, index, is_static, false);
2227 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2229 if (!is_static) pop_and_check_object(obj);
2231 const Address lo(obj, off, Address::times_1, 0*wordSize);
2232 const Address hi(obj, off, Address::times_1, 1*wordSize);
2234 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2236 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2237 assert(btos == 0, "change code, btos != 0");
2238 // btos
2239 __ andptr(flags, 0x0f);
2240 __ jcc(Assembler::notZero, notByte);
2242 __ load_signed_byte(rax, lo );
2243 __ push(btos);
2244 // Rewrite bytecode to be faster
2245 if (!is_static) {
2246 patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx);
2247 }
2248 __ jmp(Done);
2250 __ bind(notByte);
2251 // itos
2252 __ cmpl(flags, itos );
2253 __ jcc(Assembler::notEqual, notInt);
2255 __ movl(rax, lo );
2256 __ push(itos);
2257 // Rewrite bytecode to be faster
2258 if (!is_static) {
2259 patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx);
2260 }
2261 __ jmp(Done);
2263 __ bind(notInt);
2264 // atos
2265 __ cmpl(flags, atos );
2266 __ jcc(Assembler::notEqual, notObj);
2268 __ movl(rax, lo );
2269 __ push(atos);
2270 if (!is_static) {
2271 patch_bytecode(Bytecodes::_fast_agetfield, rcx, rbx);
2272 }
2273 __ jmp(Done);
2275 __ bind(notObj);
2276 // ctos
2277 __ cmpl(flags, ctos );
2278 __ jcc(Assembler::notEqual, notChar);
2280 __ load_unsigned_short(rax, lo );
2281 __ push(ctos);
2282 if (!is_static) {
2283 patch_bytecode(Bytecodes::_fast_cgetfield, rcx, rbx);
2284 }
2285 __ jmp(Done);
2287 __ bind(notChar);
2288 // stos
2289 __ cmpl(flags, stos );
2290 __ jcc(Assembler::notEqual, notShort);
2292 __ load_signed_short(rax, lo );
2293 __ push(stos);
2294 if (!is_static) {
2295 patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx);
2296 }
2297 __ jmp(Done);
2299 __ bind(notShort);
2300 // ltos
2301 __ cmpl(flags, ltos );
2302 __ jcc(Assembler::notEqual, notLong);
2304 // Generate code as if volatile. There just aren't enough registers to
2305 // save that information and this code is faster than the test.
2306 __ fild_d(lo); // Must load atomically
2307 __ subptr(rsp,2*wordSize); // Make space for store
2308 __ fistp_d(Address(rsp,0));
2309 __ pop(rax);
2310 __ pop(rdx);
2312 __ push(ltos);
2313 // Don't rewrite to _fast_lgetfield for potential volatile case.
2314 __ jmp(Done);
2316 __ bind(notLong);
2317 // ftos
2318 __ cmpl(flags, ftos );
2319 __ jcc(Assembler::notEqual, notFloat);
2321 __ fld_s(lo);
2322 __ push(ftos);
2323 if (!is_static) {
2324 patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx);
2325 }
2326 __ jmp(Done);
2328 __ bind(notFloat);
2329 // dtos
2330 __ cmpl(flags, dtos );
2331 __ jcc(Assembler::notEqual, notDouble);
2333 __ fld_d(lo);
2334 __ push(dtos);
2335 if (!is_static) {
2336 patch_bytecode(Bytecodes::_fast_dgetfield, rcx, rbx);
2337 }
2338 __ jmpb(Done);
2340 __ bind(notDouble);
2342 __ stop("Bad state");
2344 __ bind(Done);
2345 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2346 // volatile_barrier( );
2347 }
2350 void TemplateTable::getfield(int byte_no) {
2351 getfield_or_static(byte_no, false);
2352 }
2355 void TemplateTable::getstatic(int byte_no) {
2356 getfield_or_static(byte_no, true);
2357 }
2359 // The registers cache and index expected to be set before call.
2360 // The function may destroy various registers, just not the cache and index registers.
2361 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2363 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2365 if (JvmtiExport::can_post_field_modification()) {
2366 // Check to see if a field modification watch has been set before we take
2367 // the time to call into the VM.
2368 Label L1;
2369 assert_different_registers(cache, index, rax);
2370 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2371 __ testl(rax, rax);
2372 __ jcc(Assembler::zero, L1);
2374 // The cache and index registers have been already set.
2375 // This allows to eliminate this call but the cache and index
2376 // registers have to be correspondingly used after this line.
2377 __ get_cache_and_index_at_bcp(rax, rdx, 1);
2379 if (is_static) {
2380 // Life is simple. Null out the object pointer.
2381 __ xorptr(rbx, rbx);
2382 } else {
2383 // Life is harder. The stack holds the value on top, followed by the object.
2384 // We don't know the size of the value, though; it could be one or two words
2385 // depending on its type. As a result, we must find the type to determine where
2386 // the object is.
2387 Label two_word, valsize_known;
2388 __ movl(rcx, Address(rax, rdx, Address::times_ptr, in_bytes(cp_base_offset +
2389 ConstantPoolCacheEntry::flags_offset())));
2390 __ mov(rbx, rsp);
2391 __ shrl(rcx, ConstantPoolCacheEntry::tosBits);
2392 // Make sure we don't need to mask rcx for tosBits after the above shift
2393 ConstantPoolCacheEntry::verify_tosBits();
2394 __ cmpl(rcx, ltos);
2395 __ jccb(Assembler::equal, two_word);
2396 __ cmpl(rcx, dtos);
2397 __ jccb(Assembler::equal, two_word);
2398 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2399 __ jmpb(valsize_known);
2401 __ bind(two_word);
2402 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2404 __ bind(valsize_known);
2405 // setup object pointer
2406 __ movptr(rbx, Address(rbx, 0));
2407 }
2408 // cache entry pointer
2409 __ addptr(rax, in_bytes(cp_base_offset));
2410 __ shll(rdx, LogBytesPerWord);
2411 __ addptr(rax, rdx);
2412 // object (tos)
2413 __ mov(rcx, rsp);
2414 // rbx,: object pointer set up above (NULL if static)
2415 // rax,: cache entry pointer
2416 // rcx: jvalue object on the stack
2417 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2418 rbx, rax, rcx);
2419 __ get_cache_and_index_at_bcp(cache, index, 1);
2420 __ bind(L1);
2421 }
2422 }
2425 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2426 transition(vtos, vtos);
2428 const Register cache = rcx;
2429 const Register index = rdx;
2430 const Register obj = rcx;
2431 const Register off = rbx;
2432 const Register flags = rax;
2434 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2435 jvmti_post_field_mod(cache, index, is_static);
2436 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2438 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2439 // volatile_barrier( );
2441 Label notVolatile, Done;
2442 __ movl(rdx, flags);
2443 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2444 __ andl(rdx, 0x1);
2446 // field addresses
2447 const Address lo(obj, off, Address::times_1, 0*wordSize);
2448 const Address hi(obj, off, Address::times_1, 1*wordSize);
2450 Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2452 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2453 assert(btos == 0, "change code, btos != 0");
2454 // btos
2455 __ andl(flags, 0x0f);
2456 __ jcc(Assembler::notZero, notByte);
2458 __ pop(btos);
2459 if (!is_static) pop_and_check_object(obj);
2460 __ movb(lo, rax );
2461 if (!is_static) {
2462 patch_bytecode(Bytecodes::_fast_bputfield, rcx, rbx);
2463 }
2464 __ jmp(Done);
2466 __ bind(notByte);
2467 // itos
2468 __ cmpl(flags, itos );
2469 __ jcc(Assembler::notEqual, notInt);
2471 __ pop(itos);
2472 if (!is_static) pop_and_check_object(obj);
2474 __ movl(lo, rax );
2475 if (!is_static) {
2476 patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx);
2477 }
2478 __ jmp(Done);
2480 __ bind(notInt);
2481 // atos
2482 __ cmpl(flags, atos );
2483 __ jcc(Assembler::notEqual, notObj);
2485 __ pop(atos);
2486 if (!is_static) pop_and_check_object(obj);
2488 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2490 if (!is_static) {
2491 patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx);
2492 }
2494 __ jmp(Done);
2496 __ bind(notObj);
2497 // ctos
2498 __ cmpl(flags, ctos );
2499 __ jcc(Assembler::notEqual, notChar);
2501 __ pop(ctos);
2502 if (!is_static) pop_and_check_object(obj);
2503 __ movw(lo, rax );
2504 if (!is_static) {
2505 patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx);
2506 }
2507 __ jmp(Done);
2509 __ bind(notChar);
2510 // stos
2511 __ cmpl(flags, stos );
2512 __ jcc(Assembler::notEqual, notShort);
2514 __ pop(stos);
2515 if (!is_static) pop_and_check_object(obj);
2516 __ movw(lo, rax );
2517 if (!is_static) {
2518 patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx);
2519 }
2520 __ jmp(Done);
2522 __ bind(notShort);
2523 // ltos
2524 __ cmpl(flags, ltos );
2525 __ jcc(Assembler::notEqual, notLong);
2527 Label notVolatileLong;
2528 __ testl(rdx, rdx);
2529 __ jcc(Assembler::zero, notVolatileLong);
2531 __ pop(ltos); // overwrites rdx, do this after testing volatile.
2532 if (!is_static) pop_and_check_object(obj);
2534 // Replace with real volatile test
2535 __ push(rdx);
2536 __ push(rax); // Must update atomically with FIST
2537 __ fild_d(Address(rsp,0)); // So load into FPU register
2538 __ fistp_d(lo); // and put into memory atomically
2539 __ addptr(rsp, 2*wordSize);
2540 // volatile_barrier();
2541 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2542 Assembler::StoreStore));
2543 // Don't rewrite volatile version
2544 __ jmp(notVolatile);
2546 __ bind(notVolatileLong);
2548 __ pop(ltos); // overwrites rdx
2549 if (!is_static) pop_and_check_object(obj);
2550 NOT_LP64(__ movptr(hi, rdx));
2551 __ movptr(lo, rax);
2552 if (!is_static) {
2553 patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx);
2554 }
2555 __ jmp(notVolatile);
2557 __ bind(notLong);
2558 // ftos
2559 __ cmpl(flags, ftos );
2560 __ jcc(Assembler::notEqual, notFloat);
2562 __ pop(ftos);
2563 if (!is_static) pop_and_check_object(obj);
2564 __ fstp_s(lo);
2565 if (!is_static) {
2566 patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx);
2567 }
2568 __ jmp(Done);
2570 __ bind(notFloat);
2571 // dtos
2572 __ cmpl(flags, dtos );
2573 __ jcc(Assembler::notEqual, notDouble);
2575 __ pop(dtos);
2576 if (!is_static) pop_and_check_object(obj);
2577 __ fstp_d(lo);
2578 if (!is_static) {
2579 patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx);
2580 }
2581 __ jmp(Done);
2583 __ bind(notDouble);
2585 __ stop("Bad state");
2587 __ bind(Done);
2589 // Check for volatile store
2590 __ testl(rdx, rdx);
2591 __ jcc(Assembler::zero, notVolatile);
2592 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2593 Assembler::StoreStore));
2594 __ bind(notVolatile);
2595 }
2598 void TemplateTable::putfield(int byte_no) {
2599 putfield_or_static(byte_no, false);
2600 }
2603 void TemplateTable::putstatic(int byte_no) {
2604 putfield_or_static(byte_no, true);
2605 }
2607 void TemplateTable::jvmti_post_fast_field_mod() {
2608 if (JvmtiExport::can_post_field_modification()) {
2609 // Check to see if a field modification watch has been set before we take
2610 // the time to call into the VM.
2611 Label L2;
2612 __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2613 __ testl(rcx,rcx);
2614 __ jcc(Assembler::zero, L2);
2615 __ pop_ptr(rbx); // copy the object pointer from tos
2616 __ verify_oop(rbx);
2617 __ push_ptr(rbx); // put the object pointer back on tos
2618 __ subptr(rsp, sizeof(jvalue)); // add space for a jvalue object
2619 __ mov(rcx, rsp);
2620 __ push_ptr(rbx); // save object pointer so we can steal rbx,
2621 __ xorptr(rbx, rbx);
2622 const Address lo_value(rcx, rbx, Address::times_1, 0*wordSize);
2623 const Address hi_value(rcx, rbx, Address::times_1, 1*wordSize);
2624 switch (bytecode()) { // load values into the jvalue object
2625 case Bytecodes::_fast_bputfield: __ movb(lo_value, rax); break;
2626 case Bytecodes::_fast_sputfield: __ movw(lo_value, rax); break;
2627 case Bytecodes::_fast_cputfield: __ movw(lo_value, rax); break;
2628 case Bytecodes::_fast_iputfield: __ movl(lo_value, rax); break;
2629 case Bytecodes::_fast_lputfield:
2630 NOT_LP64(__ movptr(hi_value, rdx));
2631 __ movptr(lo_value, rax);
2632 break;
2634 // need to call fld_s() after fstp_s() to restore the value for below
2635 case Bytecodes::_fast_fputfield: __ fstp_s(lo_value); __ fld_s(lo_value); break;
2637 // need to call fld_d() after fstp_d() to restore the value for below
2638 case Bytecodes::_fast_dputfield: __ fstp_d(lo_value); __ fld_d(lo_value); break;
2640 // since rcx is not an object we don't call store_check() here
2641 case Bytecodes::_fast_aputfield: __ movptr(lo_value, rax); break;
2643 default: ShouldNotReachHere();
2644 }
2645 __ pop_ptr(rbx); // restore copy of object pointer
2647 // Save rax, and sometimes rdx because call_VM() will clobber them,
2648 // then use them for JVM/DI purposes
2649 __ push(rax);
2650 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2651 // access constant pool cache entry
2652 __ get_cache_entry_pointer_at_bcp(rax, rdx, 1);
2653 __ verify_oop(rbx);
2654 // rbx,: object pointer copied above
2655 // rax,: cache entry pointer
2656 // rcx: jvalue object on the stack
2657 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx);
2658 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); // restore high value
2659 __ pop(rax); // restore lower value
2660 __ addptr(rsp, sizeof(jvalue)); // release jvalue object space
2661 __ bind(L2);
2662 }
2663 }
2665 void TemplateTable::fast_storefield(TosState state) {
2666 transition(state, vtos);
2668 ByteSize base = constantPoolCacheOopDesc::base_offset();
2670 jvmti_post_fast_field_mod();
2672 // access constant pool cache
2673 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2675 // test for volatile with rdx but rdx is tos register for lputfield.
2676 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2677 __ movl(rdx, Address(rcx, rbx, Address::times_ptr, in_bytes(base +
2678 ConstantPoolCacheEntry::flags_offset())));
2680 // replace index with field offset from cache entry
2681 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2683 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2684 // volatile_barrier( );
2686 Label notVolatile, Done;
2687 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2688 __ andl(rdx, 0x1);
2689 // Check for volatile store
2690 __ testl(rdx, rdx);
2691 __ jcc(Assembler::zero, notVolatile);
2693 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2695 // Get object from stack
2696 pop_and_check_object(rcx);
2698 // field addresses
2699 const Address lo(rcx, rbx, Address::times_1, 0*wordSize);
2700 const Address hi(rcx, rbx, Address::times_1, 1*wordSize);
2702 // access field
2703 switch (bytecode()) {
2704 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2705 case Bytecodes::_fast_sputfield: // fall through
2706 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2707 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2708 case Bytecodes::_fast_lputfield:
2709 NOT_LP64(__ movptr(hi, rdx));
2710 __ movptr(lo, rax);
2711 break;
2712 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2713 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2714 case Bytecodes::_fast_aputfield: {
2715 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2716 break;
2717 }
2718 default:
2719 ShouldNotReachHere();
2720 }
2722 Label done;
2723 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2724 Assembler::StoreStore));
2725 // Barriers are so large that short branch doesn't reach!
2726 __ jmp(done);
2728 // Same code as above, but don't need rdx to test for volatile.
2729 __ bind(notVolatile);
2731 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2733 // Get object from stack
2734 pop_and_check_object(rcx);
2736 // access field
2737 switch (bytecode()) {
2738 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2739 case Bytecodes::_fast_sputfield: // fall through
2740 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2741 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2742 case Bytecodes::_fast_lputfield:
2743 NOT_LP64(__ movptr(hi, rdx));
2744 __ movptr(lo, rax);
2745 break;
2746 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2747 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2748 case Bytecodes::_fast_aputfield: {
2749 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2750 break;
2751 }
2752 default:
2753 ShouldNotReachHere();
2754 }
2755 __ bind(done);
2756 }
2759 void TemplateTable::fast_accessfield(TosState state) {
2760 transition(atos, state);
2762 // do the JVMTI work here to avoid disturbing the register state below
2763 if (JvmtiExport::can_post_field_access()) {
2764 // Check to see if a field access watch has been set before we take
2765 // the time to call into the VM.
2766 Label L1;
2767 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2768 __ testl(rcx,rcx);
2769 __ jcc(Assembler::zero, L1);
2770 // access constant pool cache entry
2771 __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1);
2772 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
2773 __ verify_oop(rax);
2774 // rax,: object pointer copied above
2775 // rcx: cache entry pointer
2776 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx);
2777 __ pop_ptr(rax); // restore object pointer
2778 __ bind(L1);
2779 }
2781 // access constant pool cache
2782 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2783 // replace index with field offset from cache entry
2784 __ movptr(rbx, Address(rcx,
2785 rbx,
2786 Address::times_ptr,
2787 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2790 // rax,: object
2791 __ verify_oop(rax);
2792 __ null_check(rax);
2793 // field addresses
2794 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2795 const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize);
2797 // access field
2798 switch (bytecode()) {
2799 case Bytecodes::_fast_bgetfield: __ movsbl(rax, lo ); break;
2800 case Bytecodes::_fast_sgetfield: __ load_signed_short(rax, lo ); break;
2801 case Bytecodes::_fast_cgetfield: __ load_unsigned_short(rax, lo ); break;
2802 case Bytecodes::_fast_igetfield: __ movl(rax, lo); break;
2803 case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten"); break;
2804 case Bytecodes::_fast_fgetfield: __ fld_s(lo); break;
2805 case Bytecodes::_fast_dgetfield: __ fld_d(lo); break;
2806 case Bytecodes::_fast_agetfield: __ movptr(rax, lo); __ verify_oop(rax); break;
2807 default:
2808 ShouldNotReachHere();
2809 }
2811 // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
2812 // volatile_barrier( );
2813 }
2815 void TemplateTable::fast_xaccess(TosState state) {
2816 transition(vtos, state);
2817 // get receiver
2818 __ movptr(rax, aaddress(0));
2819 // access constant pool cache
2820 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2821 __ movptr(rbx, Address(rcx,
2822 rdx,
2823 Address::times_ptr,
2824 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2825 // make sure exception is reported in correct bcp range (getfield is next instruction)
2826 __ increment(rsi);
2827 __ null_check(rax);
2828 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2829 if (state == itos) {
2830 __ movl(rax, lo);
2831 } else if (state == atos) {
2832 __ movptr(rax, lo);
2833 __ verify_oop(rax);
2834 } else if (state == ftos) {
2835 __ fld_s(lo);
2836 } else {
2837 ShouldNotReachHere();
2838 }
2839 __ decrement(rsi);
2840 }
2844 //----------------------------------------------------------------------------------------------------
2845 // Calls
2847 void TemplateTable::count_calls(Register method, Register temp) {
2848 // implemented elsewhere
2849 ShouldNotReachHere();
2850 }
2853 void TemplateTable::prepare_invoke(Register method, Register index, int byte_no) {
2854 // determine flags
2855 Bytecodes::Code code = bytecode();
2856 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2857 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
2858 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2859 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2860 const bool load_receiver = (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic);
2861 const bool receiver_null_check = is_invokespecial;
2862 const bool save_flags = is_invokeinterface || is_invokevirtual;
2863 // setup registers & access constant pool cache
2864 const Register recv = rcx;
2865 const Register flags = rdx;
2866 assert_different_registers(method, index, recv, flags);
2868 // save 'interpreter return address'
2869 __ save_bcp();
2871 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2873 // load receiver if needed (note: no return address pushed yet)
2874 if (load_receiver) {
2875 assert(!is_invokedynamic, "");
2876 __ movl(recv, flags);
2877 __ andl(recv, 0xFF);
2878 // recv count is 0 based?
2879 Address recv_addr(rsp, recv, Interpreter::stackElementScale(), -Interpreter::expr_offset_in_bytes(1));
2880 __ movptr(recv, recv_addr);
2881 __ verify_oop(recv);
2882 }
2884 // do null check if needed
2885 if (receiver_null_check) {
2886 __ null_check(recv);
2887 }
2889 if (save_flags) {
2890 __ mov(rsi, flags);
2891 }
2893 // compute return type
2894 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2895 // Make sure we don't need to mask flags for tosBits after the above shift
2896 ConstantPoolCacheEntry::verify_tosBits();
2897 // load return address
2898 {
2899 address table_addr;
2900 if (is_invokeinterface || is_invokedynamic)
2901 table_addr = (address)Interpreter::return_5_addrs_by_index_table();
2902 else
2903 table_addr = (address)Interpreter::return_3_addrs_by_index_table();
2904 ExternalAddress table(table_addr);
2905 __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));
2906 }
2908 // push return address
2909 __ push(flags);
2911 // Restore flag value from the constant pool cache, and restore rsi
2912 // for later null checks. rsi is the bytecode pointer
2913 if (save_flags) {
2914 __ mov(flags, rsi);
2915 __ restore_bcp();
2916 }
2917 }
2920 void TemplateTable::invokevirtual_helper(Register index, Register recv,
2921 Register flags) {
2923 // Uses temporary registers rax, rdx
2924 assert_different_registers(index, recv, rax, rdx);
2926 // Test for an invoke of a final method
2927 Label notFinal;
2928 __ movl(rax, flags);
2929 __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
2930 __ jcc(Assembler::zero, notFinal);
2932 Register method = index; // method must be rbx,
2933 assert(method == rbx, "methodOop must be rbx, for interpreter calling convention");
2935 // do the call - the index is actually the method to call
2936 __ verify_oop(method);
2938 // It's final, need a null check here!
2939 __ null_check(recv);
2941 // profile this call
2942 __ profile_final_call(rax);
2944 __ jump_from_interpreted(method, rax);
2946 __ bind(notFinal);
2948 // get receiver klass
2949 __ null_check(recv, oopDesc::klass_offset_in_bytes());
2950 // Keep recv in rcx for callee expects it there
2951 __ movptr(rax, Address(recv, oopDesc::klass_offset_in_bytes()));
2952 __ verify_oop(rax);
2954 // profile this call
2955 __ profile_virtual_call(rax, rdi, rdx);
2957 // get target methodOop & entry point
2958 const int base = instanceKlass::vtable_start_offset() * wordSize;
2959 assert(vtableEntry::size() * wordSize == 4, "adjust the scaling in the code below");
2960 __ movptr(method, Address(rax, index, Address::times_ptr, base + vtableEntry::method_offset_in_bytes()));
2961 __ jump_from_interpreted(method, rdx);
2962 }
2965 void TemplateTable::invokevirtual(int byte_no) {
2966 transition(vtos, vtos);
2967 assert(byte_no == f2_byte, "use this argument");
2968 prepare_invoke(rbx, noreg, byte_no);
2970 // rbx,: index
2971 // rcx: receiver
2972 // rdx: flags
2974 invokevirtual_helper(rbx, rcx, rdx);
2975 }
2978 void TemplateTable::invokespecial(int byte_no) {
2979 transition(vtos, vtos);
2980 assert(byte_no == f1_byte, "use this argument");
2981 prepare_invoke(rbx, noreg, byte_no);
2982 // do the call
2983 __ verify_oop(rbx);
2984 __ profile_call(rax);
2985 __ jump_from_interpreted(rbx, rax);
2986 }
2989 void TemplateTable::invokestatic(int byte_no) {
2990 transition(vtos, vtos);
2991 assert(byte_no == f1_byte, "use this argument");
2992 prepare_invoke(rbx, noreg, byte_no);
2993 // do the call
2994 __ verify_oop(rbx);
2995 __ profile_call(rax);
2996 __ jump_from_interpreted(rbx, rax);
2997 }
3000 void TemplateTable::fast_invokevfinal(int byte_no) {
3001 transition(vtos, vtos);
3002 assert(byte_no == f2_byte, "use this argument");
3003 __ stop("fast_invokevfinal not used on x86");
3004 }
3007 void TemplateTable::invokeinterface(int byte_no) {
3008 transition(vtos, vtos);
3009 assert(byte_no == f1_byte, "use this argument");
3010 prepare_invoke(rax, rbx, byte_no);
3012 // rax,: Interface
3013 // rbx,: index
3014 // rcx: receiver
3015 // rdx: flags
3017 // Special case of invokeinterface called for virtual method of
3018 // java.lang.Object. See cpCacheOop.cpp for details.
3019 // This code isn't produced by javac, but could be produced by
3020 // another compliant java compiler.
3021 Label notMethod;
3022 __ movl(rdi, rdx);
3023 __ andl(rdi, (1 << ConstantPoolCacheEntry::methodInterface));
3024 __ jcc(Assembler::zero, notMethod);
3026 invokevirtual_helper(rbx, rcx, rdx);
3027 __ bind(notMethod);
3029 // Get receiver klass into rdx - also a null check
3030 __ restore_locals(); // restore rdi
3031 __ movptr(rdx, Address(rcx, oopDesc::klass_offset_in_bytes()));
3032 __ verify_oop(rdx);
3034 // profile this call
3035 __ profile_virtual_call(rdx, rsi, rdi);
3037 Label no_such_interface, no_such_method;
3039 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3040 rdx, rax, rbx,
3041 // outputs: method, scan temp. reg
3042 rbx, rsi,
3043 no_such_interface);
3045 // rbx,: methodOop to call
3046 // rcx: receiver
3047 // Check for abstract method error
3048 // Note: This should be done more efficiently via a throw_abstract_method_error
3049 // interpreter entry point and a conditional jump to it in case of a null
3050 // method.
3051 __ testptr(rbx, rbx);
3052 __ jcc(Assembler::zero, no_such_method);
3054 // do the call
3055 // rcx: receiver
3056 // rbx,: methodOop
3057 __ jump_from_interpreted(rbx, rdx);
3058 __ should_not_reach_here();
3060 // exception handling code follows...
3061 // note: must restore interpreter registers to canonical
3062 // state for exception handling to work correctly!
3064 __ bind(no_such_method);
3065 // throw exception
3066 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3067 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3068 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3069 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3070 // the call_VM checks for exception, so we should never return here.
3071 __ should_not_reach_here();
3073 __ bind(no_such_interface);
3074 // throw exception
3075 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3076 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3077 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3078 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3079 InterpreterRuntime::throw_IncompatibleClassChangeError));
3080 // the call_VM checks for exception, so we should never return here.
3081 __ should_not_reach_here();
3082 }
3084 void TemplateTable::invokedynamic(int byte_no) {
3085 transition(vtos, vtos);
3087 if (!EnableInvokeDynamic) {
3088 // We should not encounter this bytecode if !EnableInvokeDynamic.
3089 // The verifier will stop it. However, if we get past the verifier,
3090 // this will stop the thread in a reasonable way, without crashing the JVM.
3091 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3092 InterpreterRuntime::throw_IncompatibleClassChangeError));
3093 // the call_VM checks for exception, so we should never return here.
3094 __ should_not_reach_here();
3095 return;
3096 }
3098 assert(byte_no == f1_oop, "use this argument");
3099 prepare_invoke(rax, rbx, byte_no);
3101 // rax: CallSite object (f1)
3102 // rbx: unused (f2)
3103 // rcx: receiver address
3104 // rdx: flags (unused)
3106 Register rax_callsite = rax;
3107 Register rcx_method_handle = rcx;
3109 if (ProfileInterpreter) {
3110 // %%% should make a type profile for any invokedynamic that takes a ref argument
3111 // profile this call
3112 __ profile_call(rsi);
3113 }
3115 __ movptr(rcx_method_handle, Address(rax_callsite, __ delayed_value(java_lang_invoke_CallSite::target_offset_in_bytes, rcx)));
3116 __ null_check(rcx_method_handle);
3117 __ prepare_to_jump_from_interpreted();
3118 __ jump_to_method_handle_entry(rcx_method_handle, rdx);
3119 }
3121 //----------------------------------------------------------------------------------------------------
3122 // Allocation
3124 void TemplateTable::_new() {
3125 transition(vtos, atos);
3126 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3127 Label slow_case;
3128 Label slow_case_no_pop;
3129 Label done;
3130 Label initialize_header;
3131 Label initialize_object; // including clearing the fields
3132 Label allocate_shared;
3134 __ get_cpool_and_tags(rcx, rax);
3136 // Make sure the class we're about to instantiate has been resolved.
3137 // This is done before loading instanceKlass to be consistent with the order
3138 // how Constant Pool is updated (see constantPoolOopDesc::klass_at_put)
3139 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
3140 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3141 __ jcc(Assembler::notEqual, slow_case_no_pop);
3143 // get instanceKlass
3144 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3145 __ push(rcx); // save the contexts of klass for initializing the header
3147 // make sure klass is initialized & doesn't have finalizer
3148 // make sure klass is fully initialized
3149 __ cmpl(Address(rcx, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)), instanceKlass::fully_initialized);
3150 __ jcc(Assembler::notEqual, slow_case);
3152 // get instance_size in instanceKlass (scaled to a count of bytes)
3153 __ movl(rdx, Address(rcx, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
3154 // test to see if it has a finalizer or is malformed in some way
3155 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3156 __ jcc(Assembler::notZero, slow_case);
3158 //
3159 // Allocate the instance
3160 // 1) Try to allocate in the TLAB
3161 // 2) if fail and the object is large allocate in the shared Eden
3162 // 3) if the above fails (or is not applicable), go to a slow case
3163 // (creates a new TLAB, etc.)
3165 const bool allow_shared_alloc =
3166 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3168 const Register thread = rcx;
3169 if (UseTLAB || allow_shared_alloc) {
3170 __ get_thread(thread);
3171 }
3173 if (UseTLAB) {
3174 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3175 __ lea(rbx, Address(rax, rdx, Address::times_1));
3176 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3177 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3178 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3179 if (ZeroTLAB) {
3180 // the fields have been already cleared
3181 __ jmp(initialize_header);
3182 } else {
3183 // initialize both the header and fields
3184 __ jmp(initialize_object);
3185 }
3186 }
3188 // Allocation in the shared Eden, if allowed.
3189 //
3190 // rdx: instance size in bytes
3191 if (allow_shared_alloc) {
3192 __ bind(allocate_shared);
3194 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3196 Label retry;
3197 __ bind(retry);
3198 __ movptr(rax, heap_top);
3199 __ lea(rbx, Address(rax, rdx, Address::times_1));
3200 __ cmpptr(rbx, ExternalAddress((address)Universe::heap()->end_addr()));
3201 __ jcc(Assembler::above, slow_case);
3203 // Compare rax, with the top addr, and if still equal, store the new
3204 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3205 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3206 //
3207 // rax,: object begin
3208 // rbx,: object end
3209 // rdx: instance size in bytes
3210 __ locked_cmpxchgptr(rbx, heap_top);
3212 // if someone beat us on the allocation, try again, otherwise continue
3213 __ jcc(Assembler::notEqual, retry);
3215 __ incr_allocated_bytes(thread, rdx, 0);
3216 }
3218 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3219 // The object is initialized before the header. If the object size is
3220 // zero, go directly to the header initialization.
3221 __ bind(initialize_object);
3222 __ decrement(rdx, sizeof(oopDesc));
3223 __ jcc(Assembler::zero, initialize_header);
3225 // Initialize topmost object field, divide rdx by 8, check if odd and
3226 // test if zero.
3227 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3228 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3230 // rdx must have been multiple of 8
3231 #ifdef ASSERT
3232 // make sure rdx was multiple of 8
3233 Label L;
3234 // Ignore partial flag stall after shrl() since it is debug VM
3235 __ jccb(Assembler::carryClear, L);
3236 __ stop("object size is not multiple of 2 - adjust this code");
3237 __ bind(L);
3238 // rdx must be > 0, no extra check needed here
3239 #endif
3241 // initialize remaining object fields: rdx was a multiple of 8
3242 { Label loop;
3243 __ bind(loop);
3244 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3245 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3246 __ decrement(rdx);
3247 __ jcc(Assembler::notZero, loop);
3248 }
3250 // initialize object header only.
3251 __ bind(initialize_header);
3252 if (UseBiasedLocking) {
3253 __ pop(rcx); // get saved klass back in the register.
3254 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
3255 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3256 } else {
3257 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3258 (int32_t)markOopDesc::prototype()); // header
3259 __ pop(rcx); // get saved klass back in the register.
3260 }
3261 __ movptr(Address(rax, oopDesc::klass_offset_in_bytes()), rcx); // klass
3263 {
3264 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3265 // Trigger dtrace event for fastpath
3266 __ push(atos);
3267 __ call_VM_leaf(
3268 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3269 __ pop(atos);
3270 }
3272 __ jmp(done);
3273 }
3275 // slow case
3276 __ bind(slow_case);
3277 __ pop(rcx); // restore stack pointer to what it was when we came in.
3278 __ bind(slow_case_no_pop);
3279 __ get_constant_pool(rax);
3280 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3281 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx);
3283 // continue
3284 __ bind(done);
3285 }
3288 void TemplateTable::newarray() {
3289 transition(itos, atos);
3290 __ push_i(rax); // make sure everything is on the stack
3291 __ load_unsigned_byte(rdx, at_bcp(1));
3292 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax);
3293 __ pop_i(rdx); // discard size
3294 }
3297 void TemplateTable::anewarray() {
3298 transition(itos, atos);
3299 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3300 __ get_constant_pool(rcx);
3301 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax);
3302 }
3305 void TemplateTable::arraylength() {
3306 transition(atos, itos);
3307 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3308 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3309 }
3312 void TemplateTable::checkcast() {
3313 transition(atos, atos);
3314 Label done, is_null, ok_is_subtype, quicked, resolved;
3315 __ testptr(rax, rax); // Object is in EAX
3316 __ jcc(Assembler::zero, is_null);
3318 // Get cpool & tags index
3319 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3320 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3321 // See if bytecode has already been quicked
3322 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3323 __ jcc(Assembler::equal, quicked);
3325 __ push(atos);
3326 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3327 __ pop_ptr(rdx);
3328 __ jmpb(resolved);
3330 // Get superklass in EAX and subklass in EBX
3331 __ bind(quicked);
3332 __ mov(rdx, rax); // Save object in EDX; EAX needed for subtype check
3333 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3335 __ bind(resolved);
3336 __ movptr(rbx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3338 // Generate subtype check. Blows ECX. Resets EDI. Object in EDX.
3339 // Superklass in EAX. Subklass in EBX.
3340 __ gen_subtype_check( rbx, ok_is_subtype );
3342 // Come here on failure
3343 __ push(rdx);
3344 // object is at TOS
3345 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3347 // Come here on success
3348 __ bind(ok_is_subtype);
3349 __ mov(rax,rdx); // Restore object in EDX
3351 // Collect counts on whether this check-cast sees NULLs a lot or not.
3352 if (ProfileInterpreter) {
3353 __ jmp(done);
3354 __ bind(is_null);
3355 __ profile_null_seen(rcx);
3356 } else {
3357 __ bind(is_null); // same as 'done'
3358 }
3359 __ bind(done);
3360 }
3363 void TemplateTable::instanceof() {
3364 transition(atos, itos);
3365 Label done, is_null, ok_is_subtype, quicked, resolved;
3366 __ testptr(rax, rax);
3367 __ jcc(Assembler::zero, is_null);
3369 // Get cpool & tags index
3370 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3371 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3372 // See if bytecode has already been quicked
3373 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3374 __ jcc(Assembler::equal, quicked);
3376 __ push(atos);
3377 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3378 __ pop_ptr(rdx);
3379 __ movptr(rdx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3380 __ jmp(resolved);
3382 // Get superklass in EAX and subklass in EDX
3383 __ bind(quicked);
3384 __ movptr(rdx, Address(rax, oopDesc::klass_offset_in_bytes()));
3385 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3387 __ bind(resolved);
3389 // Generate subtype check. Blows ECX. Resets EDI.
3390 // Superklass in EAX. Subklass in EDX.
3391 __ gen_subtype_check( rdx, ok_is_subtype );
3393 // Come here on failure
3394 __ xorl(rax,rax);
3395 __ jmpb(done);
3396 // Come here on success
3397 __ bind(ok_is_subtype);
3398 __ movl(rax, 1);
3400 // Collect counts on whether this test sees NULLs a lot or not.
3401 if (ProfileInterpreter) {
3402 __ jmp(done);
3403 __ bind(is_null);
3404 __ profile_null_seen(rcx);
3405 } else {
3406 __ bind(is_null); // same as 'done'
3407 }
3408 __ bind(done);
3409 // rax, = 0: obj == NULL or obj is not an instanceof the specified klass
3410 // rax, = 1: obj != NULL and obj is an instanceof the specified klass
3411 }
3414 //----------------------------------------------------------------------------------------------------
3415 // Breakpoints
3416 void TemplateTable::_breakpoint() {
3418 // Note: We get here even if we are single stepping..
3419 // jbug inists on setting breakpoints at every bytecode
3420 // even if we are in single step mode.
3422 transition(vtos, vtos);
3424 // get the unpatched byte code
3425 __ get_method(rcx);
3426 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi);
3427 __ mov(rbx, rax);
3429 // post the breakpoint event
3430 __ get_method(rcx);
3431 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi);
3433 // complete the execution of original bytecode
3434 __ dispatch_only_normal(vtos);
3435 }
3438 //----------------------------------------------------------------------------------------------------
3439 // Exceptions
3441 void TemplateTable::athrow() {
3442 transition(atos, vtos);
3443 __ null_check(rax);
3444 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3445 }
3448 //----------------------------------------------------------------------------------------------------
3449 // Synchronization
3450 //
3451 // Note: monitorenter & exit are symmetric routines; which is reflected
3452 // in the assembly code structure as well
3453 //
3454 // Stack layout:
3455 //
3456 // [expressions ] <--- rsp = expression stack top
3457 // ..
3458 // [expressions ]
3459 // [monitor entry] <--- monitor block top = expression stack bot
3460 // ..
3461 // [monitor entry]
3462 // [frame data ] <--- monitor block bot
3463 // ...
3464 // [saved rbp, ] <--- rbp,
3467 void TemplateTable::monitorenter() {
3468 transition(atos, vtos);
3470 // check for NULL object
3471 __ null_check(rax);
3473 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3474 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3475 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3476 Label allocated;
3478 // initialize entry pointer
3479 __ xorl(rdx, rdx); // points to free slot or NULL
3481 // find a free slot in the monitor block (result in rdx)
3482 { Label entry, loop, exit;
3483 __ movptr(rcx, monitor_block_top); // points to current entry, starting with top-most entry
3485 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3486 __ jmpb(entry);
3488 __ bind(loop);
3489 __ cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); // check if current entry is used
3490 __ cmovptr(Assembler::equal, rdx, rcx); // if not used then remember entry in rdx
3491 __ cmpptr(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3492 __ jccb(Assembler::equal, exit); // if same object then stop searching
3493 __ addptr(rcx, entry_size); // otherwise advance to next entry
3494 __ bind(entry);
3495 __ cmpptr(rcx, rbx); // check if bottom reached
3496 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3497 __ bind(exit);
3498 }
3500 __ testptr(rdx, rdx); // check if a slot has been found
3501 __ jccb(Assembler::notZero, allocated); // if found, continue with that one
3503 // allocate one if there's no free slot
3504 { Label entry, loop;
3505 // 1. compute new pointers // rsp: old expression stack top
3506 __ movptr(rdx, monitor_block_bot); // rdx: old expression stack bottom
3507 __ subptr(rsp, entry_size); // move expression stack top
3508 __ subptr(rdx, entry_size); // move expression stack bottom
3509 __ mov(rcx, rsp); // set start value for copy loop
3510 __ movptr(monitor_block_bot, rdx); // set new monitor block top
3511 __ jmp(entry);
3512 // 2. move expression stack contents
3513 __ bind(loop);
3514 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location
3515 __ movptr(Address(rcx, 0), rbx); // and store it at new location
3516 __ addptr(rcx, wordSize); // advance to next word
3517 __ bind(entry);
3518 __ cmpptr(rcx, rdx); // check if bottom reached
3519 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word
3520 }
3522 // call run-time routine
3523 // rdx: points to monitor entry
3524 __ bind(allocated);
3526 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3527 // The object has already been poped from the stack, so the expression stack looks correct.
3528 __ increment(rsi);
3530 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
3531 __ lock_object(rdx);
3533 // check to make sure this monitor doesn't cause stack overflow after locking
3534 __ save_bcp(); // in case of exception
3535 __ generate_stack_overflow_check(0);
3537 // The bcp has already been incremented. Just need to dispatch to next instruction.
3538 __ dispatch_next(vtos);
3539 }
3542 void TemplateTable::monitorexit() {
3543 transition(atos, vtos);
3545 // check for NULL object
3546 __ null_check(rax);
3548 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3549 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3550 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3551 Label found;
3553 // find matching slot
3554 { Label entry, loop;
3555 __ movptr(rdx, monitor_block_top); // points to current entry, starting with top-most entry
3556 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3557 __ jmpb(entry);
3559 __ bind(loop);
3560 __ cmpptr(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3561 __ jcc(Assembler::equal, found); // if same object then stop searching
3562 __ addptr(rdx, entry_size); // otherwise advance to next entry
3563 __ bind(entry);
3564 __ cmpptr(rdx, rbx); // check if bottom reached
3565 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3566 }
3568 // error handling. Unlocking was not block-structured
3569 Label end;
3570 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3571 __ should_not_reach_here();
3573 // call run-time routine
3574 // rcx: points to monitor entry
3575 __ bind(found);
3576 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3577 __ unlock_object(rdx);
3578 __ pop_ptr(rax); // discard object
3579 __ bind(end);
3580 }
3583 //----------------------------------------------------------------------------------------------------
3584 // Wide instructions
3586 void TemplateTable::wide() {
3587 transition(vtos, vtos);
3588 __ load_unsigned_byte(rbx, at_bcp(1));
3589 ExternalAddress wtable((address)Interpreter::_wentry_point);
3590 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
3591 // Note: the rsi increment step is part of the individual wide bytecode implementations
3592 }
3595 //----------------------------------------------------------------------------------------------------
3596 // Multi arrays
3598 void TemplateTable::multianewarray() {
3599 transition(vtos, atos);
3600 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3601 // last dim is on top of stack; we want address of first one:
3602 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
3603 // the latter wordSize to point to the beginning of the array.
3604 __ lea( rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
3605 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax); // pass in rax,
3606 __ load_unsigned_byte(rbx, at_bcp(3));
3607 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
3608 }
3610 #endif /* !CC_INTERP */