Fri, 07 Jan 2011 10:42:32 -0500
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
Summary: Track allocated bytes in Thread's, update on TLAB retirement and direct allocation in Eden and tenured, add JNI methods for ThreadMXBean.
Reviewed-by: coleenp, kvn, dholmes, ysr
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 "interpreter/interpreter.hpp"
27 #include "interpreter/interpreterRuntime.hpp"
28 #include "interpreter/templateTable.hpp"
29 #include "memory/universe.inline.hpp"
30 #include "oops/methodDataOop.hpp"
31 #include "oops/objArrayKlass.hpp"
32 #include "oops/oop.inline.hpp"
33 #include "prims/methodHandles.hpp"
34 #include "runtime/sharedRuntime.hpp"
35 #include "runtime/stubRoutines.hpp"
36 #include "runtime/synchronizer.hpp"
38 #ifndef CC_INTERP
39 #define __ _masm->
41 //----------------------------------------------------------------------------------------------------
42 // Platform-dependent initialization
44 void TemplateTable::pd_initialize() {
45 // No i486 specific initialization
46 }
48 //----------------------------------------------------------------------------------------------------
49 // Address computation
51 // local variables
52 static inline Address iaddress(int n) {
53 return Address(rdi, Interpreter::local_offset_in_bytes(n));
54 }
56 static inline Address laddress(int n) { return iaddress(n + 1); }
57 static inline Address haddress(int n) { return iaddress(n + 0); }
58 static inline Address faddress(int n) { return iaddress(n); }
59 static inline Address daddress(int n) { return laddress(n); }
60 static inline Address aaddress(int n) { return iaddress(n); }
62 static inline Address iaddress(Register r) {
63 return Address(rdi, r, Interpreter::stackElementScale());
64 }
65 static inline Address laddress(Register r) {
66 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(1));
67 }
68 static inline Address haddress(Register r) {
69 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
70 }
72 static inline Address faddress(Register r) { return iaddress(r); }
73 static inline Address daddress(Register r) { return laddress(r); }
74 static inline Address aaddress(Register r) { return iaddress(r); }
76 // expression stack
77 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
78 // data beyond the rsp which is potentially unsafe in an MT environment;
79 // an interrupt may overwrite that data.)
80 static inline Address at_rsp () {
81 return Address(rsp, 0);
82 }
84 // At top of Java expression stack which may be different than rsp(). It
85 // isn't for category 1 objects.
86 static inline Address at_tos () {
87 Address tos = Address(rsp, Interpreter::expr_offset_in_bytes(0));
88 return tos;
89 }
91 static inline Address at_tos_p1() {
92 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
93 }
95 static inline Address at_tos_p2() {
96 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
97 }
99 // Condition conversion
100 static Assembler::Condition j_not(TemplateTable::Condition cc) {
101 switch (cc) {
102 case TemplateTable::equal : return Assembler::notEqual;
103 case TemplateTable::not_equal : return Assembler::equal;
104 case TemplateTable::less : return Assembler::greaterEqual;
105 case TemplateTable::less_equal : return Assembler::greater;
106 case TemplateTable::greater : return Assembler::lessEqual;
107 case TemplateTable::greater_equal: return Assembler::less;
108 }
109 ShouldNotReachHere();
110 return Assembler::zero;
111 }
114 //----------------------------------------------------------------------------------------------------
115 // Miscelaneous helper routines
117 // Store an oop (or NULL) at the address described by obj.
118 // If val == noreg this means store a NULL
120 static void do_oop_store(InterpreterMacroAssembler* _masm,
121 Address obj,
122 Register val,
123 BarrierSet::Name barrier,
124 bool precise) {
125 assert(val == noreg || val == rax, "parameter is just for looks");
126 switch (barrier) {
127 #ifndef SERIALGC
128 case BarrierSet::G1SATBCT:
129 case BarrierSet::G1SATBCTLogging:
130 {
131 // flatten object address if needed
132 // We do it regardless of precise because we need the registers
133 if (obj.index() == noreg && obj.disp() == 0) {
134 if (obj.base() != rdx) {
135 __ movl(rdx, obj.base());
136 }
137 } else {
138 __ leal(rdx, obj);
139 }
140 __ get_thread(rcx);
141 __ save_bcp();
142 __ g1_write_barrier_pre(rdx, rcx, rsi, rbx, val != noreg);
144 // Do the actual store
145 // noreg means NULL
146 if (val == noreg) {
147 __ movptr(Address(rdx, 0), NULL_WORD);
148 // No post barrier for NULL
149 } else {
150 __ movl(Address(rdx, 0), val);
151 __ g1_write_barrier_post(rdx, rax, rcx, rbx, rsi);
152 }
153 __ restore_bcp();
155 }
156 break;
157 #endif // SERIALGC
158 case BarrierSet::CardTableModRef:
159 case BarrierSet::CardTableExtension:
160 {
161 if (val == noreg) {
162 __ movptr(obj, NULL_WORD);
163 } else {
164 __ movl(obj, val);
165 // flatten object address if needed
166 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
167 __ store_check(obj.base());
168 } else {
169 __ leal(rdx, obj);
170 __ store_check(rdx);
171 }
172 }
173 }
174 break;
175 case BarrierSet::ModRef:
176 case BarrierSet::Other:
177 if (val == noreg) {
178 __ movptr(obj, NULL_WORD);
179 } else {
180 __ movl(obj, val);
181 }
182 break;
183 default :
184 ShouldNotReachHere();
186 }
187 }
189 Address TemplateTable::at_bcp(int offset) {
190 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
191 return Address(rsi, offset);
192 }
195 void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
196 Register scratch,
197 bool load_bc_into_scratch/*=true*/) {
199 if (!RewriteBytecodes) return;
200 // the pair bytecodes have already done the load.
201 if (load_bc_into_scratch) {
202 __ movl(bc, bytecode);
203 }
204 Label patch_done;
205 if (JvmtiExport::can_post_breakpoint()) {
206 Label fast_patch;
207 // if a breakpoint is present we can't rewrite the stream directly
208 __ movzbl(scratch, at_bcp(0));
209 __ cmpl(scratch, Bytecodes::_breakpoint);
210 __ jcc(Assembler::notEqual, fast_patch);
211 __ get_method(scratch);
212 // Let breakpoint table handling rewrite to quicker bytecode
213 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, rsi, bc);
214 #ifndef ASSERT
215 __ jmpb(patch_done);
216 #else
217 __ jmp(patch_done);
218 #endif
219 __ bind(fast_patch);
220 }
221 #ifdef ASSERT
222 Label okay;
223 __ load_unsigned_byte(scratch, at_bcp(0));
224 __ cmpl(scratch, (int)Bytecodes::java_code(bytecode));
225 __ jccb(Assembler::equal, okay);
226 __ cmpl(scratch, bc);
227 __ jcc(Assembler::equal, okay);
228 __ stop("patching the wrong bytecode");
229 __ bind(okay);
230 #endif
231 // patch bytecode
232 __ movb(at_bcp(0), bc);
233 __ bind(patch_done);
234 }
236 //----------------------------------------------------------------------------------------------------
237 // Individual instructions
239 void TemplateTable::nop() {
240 transition(vtos, vtos);
241 // nothing to do
242 }
244 void TemplateTable::shouldnotreachhere() {
245 transition(vtos, vtos);
246 __ stop("shouldnotreachhere bytecode");
247 }
251 void TemplateTable::aconst_null() {
252 transition(vtos, atos);
253 __ xorptr(rax, rax);
254 }
257 void TemplateTable::iconst(int value) {
258 transition(vtos, itos);
259 if (value == 0) {
260 __ xorptr(rax, rax);
261 } else {
262 __ movptr(rax, value);
263 }
264 }
267 void TemplateTable::lconst(int value) {
268 transition(vtos, ltos);
269 if (value == 0) {
270 __ xorptr(rax, rax);
271 } else {
272 __ movptr(rax, value);
273 }
274 assert(value >= 0, "check this code");
275 __ xorptr(rdx, rdx);
276 }
279 void TemplateTable::fconst(int value) {
280 transition(vtos, ftos);
281 if (value == 0) { __ fldz();
282 } else if (value == 1) { __ fld1();
283 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
284 } else { ShouldNotReachHere();
285 }
286 }
289 void TemplateTable::dconst(int value) {
290 transition(vtos, dtos);
291 if (value == 0) { __ fldz();
292 } else if (value == 1) { __ fld1();
293 } else { ShouldNotReachHere();
294 }
295 }
298 void TemplateTable::bipush() {
299 transition(vtos, itos);
300 __ load_signed_byte(rax, at_bcp(1));
301 }
304 void TemplateTable::sipush() {
305 transition(vtos, itos);
306 __ load_unsigned_short(rax, at_bcp(1));
307 __ bswapl(rax);
308 __ sarl(rax, 16);
309 }
311 void TemplateTable::ldc(bool wide) {
312 transition(vtos, vtos);
313 Label call_ldc, notFloat, notClass, Done;
315 if (wide) {
316 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
317 } else {
318 __ load_unsigned_byte(rbx, at_bcp(1));
319 }
320 __ get_cpool_and_tags(rcx, rax);
321 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
322 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
324 // get type
325 __ xorptr(rdx, rdx);
326 __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset));
328 // unresolved string - get the resolved string
329 __ cmpl(rdx, JVM_CONSTANT_UnresolvedString);
330 __ jccb(Assembler::equal, call_ldc);
332 // unresolved class - get the resolved class
333 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
334 __ jccb(Assembler::equal, call_ldc);
336 // unresolved class in error (resolution failed) - call into runtime
337 // so that the same error from first resolution attempt is thrown.
338 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
339 __ jccb(Assembler::equal, call_ldc);
341 // resolved class - need to call vm to get java mirror of the class
342 __ cmpl(rdx, JVM_CONSTANT_Class);
343 __ jcc(Assembler::notEqual, notClass);
345 __ bind(call_ldc);
346 __ movl(rcx, wide);
347 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx);
348 __ push(atos);
349 __ jmp(Done);
351 __ bind(notClass);
352 __ cmpl(rdx, JVM_CONSTANT_Float);
353 __ jccb(Assembler::notEqual, notFloat);
354 // ftos
355 __ fld_s( Address(rcx, rbx, Address::times_ptr, base_offset));
356 __ push(ftos);
357 __ jmp(Done);
359 __ bind(notFloat);
360 #ifdef ASSERT
361 { Label L;
362 __ cmpl(rdx, JVM_CONSTANT_Integer);
363 __ jcc(Assembler::equal, L);
364 __ cmpl(rdx, JVM_CONSTANT_String);
365 __ jcc(Assembler::equal, L);
366 __ stop("unexpected tag type in ldc");
367 __ bind(L);
368 }
369 #endif
370 Label isOop;
371 // atos and itos
372 // String is only oop type we will see here
373 __ cmpl(rdx, JVM_CONSTANT_String);
374 __ jccb(Assembler::equal, isOop);
375 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
376 __ push(itos);
377 __ jmp(Done);
378 __ bind(isOop);
379 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
380 __ push(atos);
382 if (VerifyOops) {
383 __ verify_oop(rax);
384 }
385 __ bind(Done);
386 }
388 // Fast path for caching oop constants.
389 // %%% We should use this to handle Class and String constants also.
390 // %%% It will simplify the ldc/primitive path considerably.
391 void TemplateTable::fast_aldc(bool wide) {
392 transition(vtos, atos);
394 if (!EnableMethodHandles) {
395 // We should not encounter this bytecode if !EnableMethodHandles.
396 // The verifier will stop it. However, if we get past the verifier,
397 // this will stop the thread in a reasonable way, without crashing the JVM.
398 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
399 InterpreterRuntime::throw_IncompatibleClassChangeError));
400 // the call_VM checks for exception, so we should never return here.
401 __ should_not_reach_here();
402 return;
403 }
405 const Register cache = rcx;
406 const Register index = rdx;
408 resolve_cache_and_index(f1_oop, rax, cache, index, wide ? sizeof(u2) : sizeof(u1));
409 if (VerifyOops) {
410 __ verify_oop(rax);
411 }
413 Label L_done, L_throw_exception;
414 const Register con_klass_temp = rcx; // same as Rcache
415 __ movptr(con_klass_temp, Address(rax, oopDesc::klass_offset_in_bytes()));
416 __ cmpptr(con_klass_temp, ExternalAddress((address)Universe::systemObjArrayKlassObj_addr()));
417 __ jcc(Assembler::notEqual, L_done);
418 __ cmpl(Address(rax, arrayOopDesc::length_offset_in_bytes()), 0);
419 __ jcc(Assembler::notEqual, L_throw_exception);
420 __ xorptr(rax, rax);
421 __ jmp(L_done);
423 // Load the exception from the system-array which wraps it:
424 __ bind(L_throw_exception);
425 __ movptr(rax, Address(rax, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
426 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
428 __ bind(L_done);
429 }
431 void TemplateTable::ldc2_w() {
432 transition(vtos, vtos);
433 Label Long, Done;
434 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
436 __ get_cpool_and_tags(rcx, rax);
437 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
438 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
440 // get type
441 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double);
442 __ jccb(Assembler::notEqual, Long);
443 // dtos
444 __ fld_d( Address(rcx, rbx, Address::times_ptr, base_offset));
445 __ push(dtos);
446 __ jmpb(Done);
448 __ bind(Long);
449 // ltos
450 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
451 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
453 __ push(ltos);
455 __ bind(Done);
456 }
459 void TemplateTable::locals_index(Register reg, int offset) {
460 __ load_unsigned_byte(reg, at_bcp(offset));
461 __ negptr(reg);
462 }
465 void TemplateTable::iload() {
466 transition(vtos, itos);
467 if (RewriteFrequentPairs) {
468 Label rewrite, done;
470 // get next byte
471 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
472 // if _iload, wait to rewrite to iload2. We only want to rewrite the
473 // last two iloads in a pair. Comparing against fast_iload means that
474 // the next bytecode is neither an iload or a caload, and therefore
475 // an iload pair.
476 __ cmpl(rbx, Bytecodes::_iload);
477 __ jcc(Assembler::equal, done);
479 __ cmpl(rbx, Bytecodes::_fast_iload);
480 __ movl(rcx, Bytecodes::_fast_iload2);
481 __ jccb(Assembler::equal, rewrite);
483 // if _caload, rewrite to fast_icaload
484 __ cmpl(rbx, Bytecodes::_caload);
485 __ movl(rcx, Bytecodes::_fast_icaload);
486 __ jccb(Assembler::equal, rewrite);
488 // rewrite so iload doesn't check again.
489 __ movl(rcx, Bytecodes::_fast_iload);
491 // rewrite
492 // rcx: fast bytecode
493 __ bind(rewrite);
494 patch_bytecode(Bytecodes::_iload, rcx, rbx, false);
495 __ bind(done);
496 }
498 // Get the local value into tos
499 locals_index(rbx);
500 __ movl(rax, iaddress(rbx));
501 }
504 void TemplateTable::fast_iload2() {
505 transition(vtos, itos);
506 locals_index(rbx);
507 __ movl(rax, iaddress(rbx));
508 __ push(itos);
509 locals_index(rbx, 3);
510 __ movl(rax, iaddress(rbx));
511 }
513 void TemplateTable::fast_iload() {
514 transition(vtos, itos);
515 locals_index(rbx);
516 __ movl(rax, iaddress(rbx));
517 }
520 void TemplateTable::lload() {
521 transition(vtos, ltos);
522 locals_index(rbx);
523 __ movptr(rax, laddress(rbx));
524 NOT_LP64(__ movl(rdx, haddress(rbx)));
525 }
528 void TemplateTable::fload() {
529 transition(vtos, ftos);
530 locals_index(rbx);
531 __ fld_s(faddress(rbx));
532 }
535 void TemplateTable::dload() {
536 transition(vtos, dtos);
537 locals_index(rbx);
538 __ fld_d(daddress(rbx));
539 }
542 void TemplateTable::aload() {
543 transition(vtos, atos);
544 locals_index(rbx);
545 __ movptr(rax, aaddress(rbx));
546 }
549 void TemplateTable::locals_index_wide(Register reg) {
550 __ movl(reg, at_bcp(2));
551 __ bswapl(reg);
552 __ shrl(reg, 16);
553 __ negptr(reg);
554 }
557 void TemplateTable::wide_iload() {
558 transition(vtos, itos);
559 locals_index_wide(rbx);
560 __ movl(rax, iaddress(rbx));
561 }
564 void TemplateTable::wide_lload() {
565 transition(vtos, ltos);
566 locals_index_wide(rbx);
567 __ movptr(rax, laddress(rbx));
568 NOT_LP64(__ movl(rdx, haddress(rbx)));
569 }
572 void TemplateTable::wide_fload() {
573 transition(vtos, ftos);
574 locals_index_wide(rbx);
575 __ fld_s(faddress(rbx));
576 }
579 void TemplateTable::wide_dload() {
580 transition(vtos, dtos);
581 locals_index_wide(rbx);
582 __ fld_d(daddress(rbx));
583 }
586 void TemplateTable::wide_aload() {
587 transition(vtos, atos);
588 locals_index_wide(rbx);
589 __ movptr(rax, aaddress(rbx));
590 }
592 void TemplateTable::index_check(Register array, Register index) {
593 // Pop ptr into array
594 __ pop_ptr(array);
595 index_check_without_pop(array, index);
596 }
598 void TemplateTable::index_check_without_pop(Register array, Register index) {
599 // destroys rbx,
600 // check array
601 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
602 LP64_ONLY(__ movslq(index, index));
603 // check index
604 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
605 if (index != rbx) {
606 // ??? convention: move aberrant index into rbx, for exception message
607 assert(rbx != array, "different registers");
608 __ mov(rbx, index);
609 }
610 __ jump_cc(Assembler::aboveEqual,
611 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
612 }
615 void TemplateTable::iaload() {
616 transition(itos, itos);
617 // rdx: array
618 index_check(rdx, rax); // kills rbx,
619 // rax,: index
620 __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)));
621 }
624 void TemplateTable::laload() {
625 transition(itos, ltos);
626 // rax,: index
627 // rdx: array
628 index_check(rdx, rax);
629 __ mov(rbx, rax);
630 // rbx,: index
631 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
632 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
633 }
636 void TemplateTable::faload() {
637 transition(itos, ftos);
638 // rdx: array
639 index_check(rdx, rax); // kills rbx,
640 // rax,: index
641 __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
642 }
645 void TemplateTable::daload() {
646 transition(itos, dtos);
647 // rdx: array
648 index_check(rdx, rax); // kills rbx,
649 // rax,: index
650 __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
651 }
654 void TemplateTable::aaload() {
655 transition(itos, atos);
656 // rdx: array
657 index_check(rdx, rax); // kills rbx,
658 // rax,: index
659 __ movptr(rax, Address(rdx, rax, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
660 }
663 void TemplateTable::baload() {
664 transition(itos, itos);
665 // rdx: array
666 index_check(rdx, rax); // kills rbx,
667 // rax,: index
668 // can do better code for P5 - fix this at some point
669 __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
670 __ mov(rax, rbx);
671 }
674 void TemplateTable::caload() {
675 transition(itos, itos);
676 // rdx: array
677 index_check(rdx, rax); // kills rbx,
678 // rax,: index
679 // can do better code for P5 - may want to improve this at some point
680 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
681 __ mov(rax, rbx);
682 }
684 // iload followed by caload frequent pair
685 void TemplateTable::fast_icaload() {
686 transition(vtos, itos);
687 // load index out of locals
688 locals_index(rbx);
689 __ movl(rax, iaddress(rbx));
691 // rdx: array
692 index_check(rdx, rax);
693 // rax,: index
694 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
695 __ mov(rax, rbx);
696 }
698 void TemplateTable::saload() {
699 transition(itos, itos);
700 // rdx: array
701 index_check(rdx, rax); // kills rbx,
702 // rax,: index
703 // can do better code for P5 - may want to improve this at some point
704 __ load_signed_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
705 __ mov(rax, rbx);
706 }
709 void TemplateTable::iload(int n) {
710 transition(vtos, itos);
711 __ movl(rax, iaddress(n));
712 }
715 void TemplateTable::lload(int n) {
716 transition(vtos, ltos);
717 __ movptr(rax, laddress(n));
718 NOT_LP64(__ movptr(rdx, haddress(n)));
719 }
722 void TemplateTable::fload(int n) {
723 transition(vtos, ftos);
724 __ fld_s(faddress(n));
725 }
728 void TemplateTable::dload(int n) {
729 transition(vtos, dtos);
730 __ fld_d(daddress(n));
731 }
734 void TemplateTable::aload(int n) {
735 transition(vtos, atos);
736 __ movptr(rax, aaddress(n));
737 }
740 void TemplateTable::aload_0() {
741 transition(vtos, atos);
742 // According to bytecode histograms, the pairs:
743 //
744 // _aload_0, _fast_igetfield
745 // _aload_0, _fast_agetfield
746 // _aload_0, _fast_fgetfield
747 //
748 // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
749 // bytecode checks if the next bytecode is either _fast_igetfield,
750 // _fast_agetfield or _fast_fgetfield and then rewrites the
751 // current bytecode into a pair bytecode; otherwise it rewrites the current
752 // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
753 //
754 // Note: If the next bytecode is _getfield, the rewrite must be delayed,
755 // otherwise we may miss an opportunity for a pair.
756 //
757 // Also rewrite frequent pairs
758 // aload_0, aload_1
759 // aload_0, iload_1
760 // These bytecodes with a small amount of code are most profitable to rewrite
761 if (RewriteFrequentPairs) {
762 Label rewrite, done;
763 // get next byte
764 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
766 // do actual aload_0
767 aload(0);
769 // if _getfield then wait with rewrite
770 __ cmpl(rbx, Bytecodes::_getfield);
771 __ jcc(Assembler::equal, done);
773 // if _igetfield then reqrite to _fast_iaccess_0
774 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
775 __ cmpl(rbx, Bytecodes::_fast_igetfield);
776 __ movl(rcx, Bytecodes::_fast_iaccess_0);
777 __ jccb(Assembler::equal, rewrite);
779 // if _agetfield then reqrite to _fast_aaccess_0
780 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
781 __ cmpl(rbx, Bytecodes::_fast_agetfield);
782 __ movl(rcx, Bytecodes::_fast_aaccess_0);
783 __ jccb(Assembler::equal, rewrite);
785 // if _fgetfield then reqrite to _fast_faccess_0
786 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
787 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
788 __ movl(rcx, Bytecodes::_fast_faccess_0);
789 __ jccb(Assembler::equal, rewrite);
791 // else rewrite to _fast_aload0
792 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
793 __ movl(rcx, Bytecodes::_fast_aload_0);
795 // rewrite
796 // rcx: fast bytecode
797 __ bind(rewrite);
798 patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false);
800 __ bind(done);
801 } else {
802 aload(0);
803 }
804 }
806 void TemplateTable::istore() {
807 transition(itos, vtos);
808 locals_index(rbx);
809 __ movl(iaddress(rbx), rax);
810 }
813 void TemplateTable::lstore() {
814 transition(ltos, vtos);
815 locals_index(rbx);
816 __ movptr(laddress(rbx), rax);
817 NOT_LP64(__ movptr(haddress(rbx), rdx));
818 }
821 void TemplateTable::fstore() {
822 transition(ftos, vtos);
823 locals_index(rbx);
824 __ fstp_s(faddress(rbx));
825 }
828 void TemplateTable::dstore() {
829 transition(dtos, vtos);
830 locals_index(rbx);
831 __ fstp_d(daddress(rbx));
832 }
835 void TemplateTable::astore() {
836 transition(vtos, vtos);
837 __ pop_ptr(rax);
838 locals_index(rbx);
839 __ movptr(aaddress(rbx), rax);
840 }
843 void TemplateTable::wide_istore() {
844 transition(vtos, vtos);
845 __ pop_i(rax);
846 locals_index_wide(rbx);
847 __ movl(iaddress(rbx), rax);
848 }
851 void TemplateTable::wide_lstore() {
852 transition(vtos, vtos);
853 __ pop_l(rax, rdx);
854 locals_index_wide(rbx);
855 __ movptr(laddress(rbx), rax);
856 NOT_LP64(__ movl(haddress(rbx), rdx));
857 }
860 void TemplateTable::wide_fstore() {
861 wide_istore();
862 }
865 void TemplateTable::wide_dstore() {
866 wide_lstore();
867 }
870 void TemplateTable::wide_astore() {
871 transition(vtos, vtos);
872 __ pop_ptr(rax);
873 locals_index_wide(rbx);
874 __ movptr(aaddress(rbx), rax);
875 }
878 void TemplateTable::iastore() {
879 transition(itos, vtos);
880 __ pop_i(rbx);
881 // rax,: value
882 // rdx: array
883 index_check(rdx, rbx); // prefer index in rbx,
884 // rbx,: index
885 __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax);
886 }
889 void TemplateTable::lastore() {
890 transition(ltos, vtos);
891 __ pop_i(rbx);
892 // rax,: low(value)
893 // rcx: array
894 // rdx: high(value)
895 index_check(rcx, rbx); // prefer index in rbx,
896 // rbx,: index
897 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
898 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
899 }
902 void TemplateTable::fastore() {
903 transition(ftos, vtos);
904 __ pop_i(rbx);
905 // rdx: array
906 // st0: value
907 index_check(rdx, rbx); // prefer index in rbx,
908 // rbx,: index
909 __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
910 }
913 void TemplateTable::dastore() {
914 transition(dtos, vtos);
915 __ pop_i(rbx);
916 // rdx: array
917 // st0: value
918 index_check(rdx, rbx); // prefer index in rbx,
919 // rbx,: index
920 __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
921 }
924 void TemplateTable::aastore() {
925 Label is_null, ok_is_subtype, done;
926 transition(vtos, vtos);
927 // stack: ..., array, index, value
928 __ movptr(rax, at_tos()); // Value
929 __ movl(rcx, at_tos_p1()); // Index
930 __ movptr(rdx, at_tos_p2()); // Array
932 Address element_address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
933 index_check_without_pop(rdx, rcx); // kills rbx,
934 // do array store check - check for NULL value first
935 __ testptr(rax, rax);
936 __ jcc(Assembler::zero, is_null);
938 // Move subklass into EBX
939 __ movptr(rbx, Address(rax, oopDesc::klass_offset_in_bytes()));
940 // Move superklass into EAX
941 __ movptr(rax, Address(rdx, oopDesc::klass_offset_in_bytes()));
942 __ movptr(rax, Address(rax, sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes()));
943 // Compress array+index*wordSize+12 into a single register. Frees ECX.
944 __ lea(rdx, element_address);
946 // Generate subtype check. Blows ECX. Resets EDI to locals.
947 // Superklass in EAX. Subklass in EBX.
948 __ gen_subtype_check( rbx, ok_is_subtype );
950 // Come here on failure
951 // object is at TOS
952 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
954 // Come here on success
955 __ bind(ok_is_subtype);
957 // Get the value to store
958 __ movptr(rax, at_rsp());
959 // and store it with appropriate barrier
960 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
962 __ jmp(done);
964 // Have a NULL in EAX, EDX=array, ECX=index. Store NULL at ary[idx]
965 __ bind(is_null);
966 __ profile_null_seen(rbx);
968 // Store NULL, (noreg means NULL to do_oop_store)
969 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
971 // Pop stack arguments
972 __ bind(done);
973 __ addptr(rsp, 3 * Interpreter::stackElementSize);
974 }
977 void TemplateTable::bastore() {
978 transition(itos, vtos);
979 __ pop_i(rbx);
980 // rax,: value
981 // rdx: array
982 index_check(rdx, rbx); // prefer index in rbx,
983 // rbx,: index
984 __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax);
985 }
988 void TemplateTable::castore() {
989 transition(itos, vtos);
990 __ pop_i(rbx);
991 // rax,: value
992 // rdx: array
993 index_check(rdx, rbx); // prefer index in rbx,
994 // rbx,: index
995 __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax);
996 }
999 void TemplateTable::sastore() {
1000 castore();
1001 }
1004 void TemplateTable::istore(int n) {
1005 transition(itos, vtos);
1006 __ movl(iaddress(n), rax);
1007 }
1010 void TemplateTable::lstore(int n) {
1011 transition(ltos, vtos);
1012 __ movptr(laddress(n), rax);
1013 NOT_LP64(__ movptr(haddress(n), rdx));
1014 }
1017 void TemplateTable::fstore(int n) {
1018 transition(ftos, vtos);
1019 __ fstp_s(faddress(n));
1020 }
1023 void TemplateTable::dstore(int n) {
1024 transition(dtos, vtos);
1025 __ fstp_d(daddress(n));
1026 }
1029 void TemplateTable::astore(int n) {
1030 transition(vtos, vtos);
1031 __ pop_ptr(rax);
1032 __ movptr(aaddress(n), rax);
1033 }
1036 void TemplateTable::pop() {
1037 transition(vtos, vtos);
1038 __ addptr(rsp, Interpreter::stackElementSize);
1039 }
1042 void TemplateTable::pop2() {
1043 transition(vtos, vtos);
1044 __ addptr(rsp, 2*Interpreter::stackElementSize);
1045 }
1048 void TemplateTable::dup() {
1049 transition(vtos, vtos);
1050 // stack: ..., a
1051 __ load_ptr(0, rax);
1052 __ push_ptr(rax);
1053 // stack: ..., a, a
1054 }
1057 void TemplateTable::dup_x1() {
1058 transition(vtos, vtos);
1059 // stack: ..., a, b
1060 __ load_ptr( 0, rax); // load b
1061 __ load_ptr( 1, rcx); // load a
1062 __ store_ptr(1, rax); // store b
1063 __ store_ptr(0, rcx); // store a
1064 __ push_ptr(rax); // push b
1065 // stack: ..., b, a, b
1066 }
1069 void TemplateTable::dup_x2() {
1070 transition(vtos, vtos);
1071 // stack: ..., a, b, c
1072 __ load_ptr( 0, rax); // load c
1073 __ load_ptr( 2, rcx); // load a
1074 __ store_ptr(2, rax); // store c in a
1075 __ push_ptr(rax); // push c
1076 // stack: ..., c, b, c, c
1077 __ load_ptr( 2, rax); // load b
1078 __ store_ptr(2, rcx); // store a in b
1079 // stack: ..., c, a, c, c
1080 __ store_ptr(1, rax); // store b in c
1081 // stack: ..., c, a, b, c
1082 }
1085 void TemplateTable::dup2() {
1086 transition(vtos, vtos);
1087 // stack: ..., a, b
1088 __ load_ptr(1, rax); // load a
1089 __ push_ptr(rax); // push a
1090 __ load_ptr(1, rax); // load b
1091 __ push_ptr(rax); // push b
1092 // stack: ..., a, b, a, b
1093 }
1096 void TemplateTable::dup2_x1() {
1097 transition(vtos, vtos);
1098 // stack: ..., a, b, c
1099 __ load_ptr( 0, rcx); // load c
1100 __ load_ptr( 1, rax); // load b
1101 __ push_ptr(rax); // push b
1102 __ push_ptr(rcx); // push c
1103 // stack: ..., a, b, c, b, c
1104 __ store_ptr(3, rcx); // store c in b
1105 // stack: ..., a, c, c, b, c
1106 __ load_ptr( 4, rcx); // load a
1107 __ store_ptr(2, rcx); // store a in 2nd c
1108 // stack: ..., a, c, a, b, c
1109 __ store_ptr(4, rax); // store b in a
1110 // stack: ..., b, c, a, b, c
1111 // stack: ..., b, c, a, b, c
1112 }
1115 void TemplateTable::dup2_x2() {
1116 transition(vtos, vtos);
1117 // stack: ..., a, b, c, d
1118 __ load_ptr( 0, rcx); // load d
1119 __ load_ptr( 1, rax); // load c
1120 __ push_ptr(rax); // push c
1121 __ push_ptr(rcx); // push d
1122 // stack: ..., a, b, c, d, c, d
1123 __ load_ptr( 4, rax); // load b
1124 __ store_ptr(2, rax); // store b in d
1125 __ store_ptr(4, rcx); // store d in b
1126 // stack: ..., a, d, c, b, c, d
1127 __ load_ptr( 5, rcx); // load a
1128 __ load_ptr( 3, rax); // load c
1129 __ store_ptr(3, rcx); // store a in c
1130 __ store_ptr(5, rax); // store c in a
1131 // stack: ..., c, d, a, b, c, d
1132 // stack: ..., c, d, a, b, c, d
1133 }
1136 void TemplateTable::swap() {
1137 transition(vtos, vtos);
1138 // stack: ..., a, b
1139 __ load_ptr( 1, rcx); // load a
1140 __ load_ptr( 0, rax); // load b
1141 __ store_ptr(0, rcx); // store a in b
1142 __ store_ptr(1, rax); // store b in a
1143 // stack: ..., b, a
1144 }
1147 void TemplateTable::iop2(Operation op) {
1148 transition(itos, itos);
1149 switch (op) {
1150 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1151 case sub : __ mov(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1152 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1153 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1154 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1155 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1156 case shl : __ mov(rcx, rax); __ pop_i(rax); __ shll (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1157 case shr : __ mov(rcx, rax); __ pop_i(rax); __ sarl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1158 case ushr : __ mov(rcx, rax); __ pop_i(rax); __ shrl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1159 default : ShouldNotReachHere();
1160 }
1161 }
1164 void TemplateTable::lop2(Operation op) {
1165 transition(ltos, ltos);
1166 __ pop_l(rbx, rcx);
1167 switch (op) {
1168 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1169 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1170 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1171 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1172 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1173 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1174 default : ShouldNotReachHere();
1175 }
1176 }
1179 void TemplateTable::idiv() {
1180 transition(itos, itos);
1181 __ mov(rcx, rax);
1182 __ pop_i(rax);
1183 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1184 // they are not equal, one could do a normal division (no correction
1185 // needed), which may speed up this implementation for the common case.
1186 // (see also JVM spec., p.243 & p.271)
1187 __ corrected_idivl(rcx);
1188 }
1191 void TemplateTable::irem() {
1192 transition(itos, itos);
1193 __ mov(rcx, rax);
1194 __ pop_i(rax);
1195 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1196 // they are not equal, one could do a normal division (no correction
1197 // needed), which may speed up this implementation for the common case.
1198 // (see also JVM spec., p.243 & p.271)
1199 __ corrected_idivl(rcx);
1200 __ mov(rax, rdx);
1201 }
1204 void TemplateTable::lmul() {
1205 transition(ltos, ltos);
1206 __ pop_l(rbx, rcx);
1207 __ push(rcx); __ push(rbx);
1208 __ push(rdx); __ push(rax);
1209 __ lmul(2 * wordSize, 0);
1210 __ addptr(rsp, 4 * wordSize); // take off temporaries
1211 }
1214 void TemplateTable::ldiv() {
1215 transition(ltos, ltos);
1216 __ pop_l(rbx, rcx);
1217 __ push(rcx); __ push(rbx);
1218 __ push(rdx); __ push(rax);
1219 // check if y = 0
1220 __ orl(rax, rdx);
1221 __ jump_cc(Assembler::zero,
1222 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1223 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1224 __ addptr(rsp, 4 * wordSize); // take off temporaries
1225 }
1228 void TemplateTable::lrem() {
1229 transition(ltos, ltos);
1230 __ pop_l(rbx, rcx);
1231 __ push(rcx); __ push(rbx);
1232 __ push(rdx); __ push(rax);
1233 // check if y = 0
1234 __ orl(rax, rdx);
1235 __ jump_cc(Assembler::zero,
1236 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1237 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1238 __ addptr(rsp, 4 * wordSize);
1239 }
1242 void TemplateTable::lshl() {
1243 transition(itos, ltos);
1244 __ movl(rcx, rax); // get shift count
1245 __ pop_l(rax, rdx); // get shift value
1246 __ lshl(rdx, rax);
1247 }
1250 void TemplateTable::lshr() {
1251 transition(itos, ltos);
1252 __ mov(rcx, rax); // get shift count
1253 __ pop_l(rax, rdx); // get shift value
1254 __ lshr(rdx, rax, true);
1255 }
1258 void TemplateTable::lushr() {
1259 transition(itos, ltos);
1260 __ mov(rcx, rax); // get shift count
1261 __ pop_l(rax, rdx); // get shift value
1262 __ lshr(rdx, rax);
1263 }
1266 void TemplateTable::fop2(Operation op) {
1267 transition(ftos, ftos);
1268 switch (op) {
1269 case add: __ fadd_s (at_rsp()); break;
1270 case sub: __ fsubr_s(at_rsp()); break;
1271 case mul: __ fmul_s (at_rsp()); break;
1272 case div: __ fdivr_s(at_rsp()); break;
1273 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1274 default : ShouldNotReachHere();
1275 }
1276 __ f2ieee();
1277 __ pop(rax); // pop float thing off
1278 }
1281 void TemplateTable::dop2(Operation op) {
1282 transition(dtos, dtos);
1284 switch (op) {
1285 case add: __ fadd_d (at_rsp()); break;
1286 case sub: __ fsubr_d(at_rsp()); break;
1287 case mul: {
1288 Label L_strict;
1289 Label L_join;
1290 const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1291 __ get_method(rcx);
1292 __ movl(rcx, access_flags);
1293 __ testl(rcx, JVM_ACC_STRICT);
1294 __ jccb(Assembler::notZero, L_strict);
1295 __ fmul_d (at_rsp());
1296 __ jmpb(L_join);
1297 __ bind(L_strict);
1298 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1299 __ fmulp();
1300 __ fmul_d (at_rsp());
1301 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1302 __ fmulp();
1303 __ bind(L_join);
1304 break;
1305 }
1306 case div: {
1307 Label L_strict;
1308 Label L_join;
1309 const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1310 __ get_method(rcx);
1311 __ movl(rcx, access_flags);
1312 __ testl(rcx, JVM_ACC_STRICT);
1313 __ jccb(Assembler::notZero, L_strict);
1314 __ fdivr_d(at_rsp());
1315 __ jmp(L_join);
1316 __ bind(L_strict);
1317 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1318 __ fmul_d (at_rsp());
1319 __ fdivrp();
1320 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1321 __ fmulp();
1322 __ bind(L_join);
1323 break;
1324 }
1325 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1326 default : ShouldNotReachHere();
1327 }
1328 __ d2ieee();
1329 // Pop double precision number from rsp.
1330 __ pop(rax);
1331 __ pop(rdx);
1332 }
1335 void TemplateTable::ineg() {
1336 transition(itos, itos);
1337 __ negl(rax);
1338 }
1341 void TemplateTable::lneg() {
1342 transition(ltos, ltos);
1343 __ lneg(rdx, rax);
1344 }
1347 void TemplateTable::fneg() {
1348 transition(ftos, ftos);
1349 __ fchs();
1350 }
1353 void TemplateTable::dneg() {
1354 transition(dtos, dtos);
1355 __ fchs();
1356 }
1359 void TemplateTable::iinc() {
1360 transition(vtos, vtos);
1361 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1362 locals_index(rbx);
1363 __ addl(iaddress(rbx), rdx);
1364 }
1367 void TemplateTable::wide_iinc() {
1368 transition(vtos, vtos);
1369 __ movl(rdx, at_bcp(4)); // get constant
1370 locals_index_wide(rbx);
1371 __ bswapl(rdx); // swap bytes & sign-extend constant
1372 __ sarl(rdx, 16);
1373 __ addl(iaddress(rbx), rdx);
1374 // Note: should probably use only one movl to get both
1375 // the index and the constant -> fix this
1376 }
1379 void TemplateTable::convert() {
1380 // Checking
1381 #ifdef ASSERT
1382 { TosState tos_in = ilgl;
1383 TosState tos_out = ilgl;
1384 switch (bytecode()) {
1385 case Bytecodes::_i2l: // fall through
1386 case Bytecodes::_i2f: // fall through
1387 case Bytecodes::_i2d: // fall through
1388 case Bytecodes::_i2b: // fall through
1389 case Bytecodes::_i2c: // fall through
1390 case Bytecodes::_i2s: tos_in = itos; break;
1391 case Bytecodes::_l2i: // fall through
1392 case Bytecodes::_l2f: // fall through
1393 case Bytecodes::_l2d: tos_in = ltos; break;
1394 case Bytecodes::_f2i: // fall through
1395 case Bytecodes::_f2l: // fall through
1396 case Bytecodes::_f2d: tos_in = ftos; break;
1397 case Bytecodes::_d2i: // fall through
1398 case Bytecodes::_d2l: // fall through
1399 case Bytecodes::_d2f: tos_in = dtos; break;
1400 default : ShouldNotReachHere();
1401 }
1402 switch (bytecode()) {
1403 case Bytecodes::_l2i: // fall through
1404 case Bytecodes::_f2i: // fall through
1405 case Bytecodes::_d2i: // fall through
1406 case Bytecodes::_i2b: // fall through
1407 case Bytecodes::_i2c: // fall through
1408 case Bytecodes::_i2s: tos_out = itos; break;
1409 case Bytecodes::_i2l: // fall through
1410 case Bytecodes::_f2l: // fall through
1411 case Bytecodes::_d2l: tos_out = ltos; break;
1412 case Bytecodes::_i2f: // fall through
1413 case Bytecodes::_l2f: // fall through
1414 case Bytecodes::_d2f: tos_out = ftos; break;
1415 case Bytecodes::_i2d: // fall through
1416 case Bytecodes::_l2d: // fall through
1417 case Bytecodes::_f2d: tos_out = dtos; break;
1418 default : ShouldNotReachHere();
1419 }
1420 transition(tos_in, tos_out);
1421 }
1422 #endif // ASSERT
1424 // Conversion
1425 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1426 switch (bytecode()) {
1427 case Bytecodes::_i2l:
1428 __ extend_sign(rdx, rax);
1429 break;
1430 case Bytecodes::_i2f:
1431 __ push(rax); // store int on tos
1432 __ fild_s(at_rsp()); // load int to ST0
1433 __ f2ieee(); // truncate to float size
1434 __ pop(rcx); // adjust rsp
1435 break;
1436 case Bytecodes::_i2d:
1437 __ push(rax); // add one slot for d2ieee()
1438 __ push(rax); // store int on tos
1439 __ fild_s(at_rsp()); // load int to ST0
1440 __ d2ieee(); // truncate to double size
1441 __ pop(rcx); // adjust rsp
1442 __ pop(rcx);
1443 break;
1444 case Bytecodes::_i2b:
1445 __ shll(rax, 24); // truncate upper 24 bits
1446 __ sarl(rax, 24); // and sign-extend byte
1447 LP64_ONLY(__ movsbl(rax, rax));
1448 break;
1449 case Bytecodes::_i2c:
1450 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1451 LP64_ONLY(__ movzwl(rax, rax));
1452 break;
1453 case Bytecodes::_i2s:
1454 __ shll(rax, 16); // truncate upper 16 bits
1455 __ sarl(rax, 16); // and sign-extend short
1456 LP64_ONLY(__ movswl(rax, rax));
1457 break;
1458 case Bytecodes::_l2i:
1459 /* nothing to do */
1460 break;
1461 case Bytecodes::_l2f:
1462 __ push(rdx); // store long on tos
1463 __ push(rax);
1464 __ fild_d(at_rsp()); // load long to ST0
1465 __ f2ieee(); // truncate to float size
1466 __ pop(rcx); // adjust rsp
1467 __ pop(rcx);
1468 break;
1469 case Bytecodes::_l2d:
1470 __ push(rdx); // store long on tos
1471 __ push(rax);
1472 __ fild_d(at_rsp()); // load long to ST0
1473 __ d2ieee(); // truncate to double size
1474 __ pop(rcx); // adjust rsp
1475 __ pop(rcx);
1476 break;
1477 case Bytecodes::_f2i:
1478 __ push(rcx); // reserve space for argument
1479 __ fstp_s(at_rsp()); // pass float argument on stack
1480 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1481 break;
1482 case Bytecodes::_f2l:
1483 __ push(rcx); // reserve space for argument
1484 __ fstp_s(at_rsp()); // pass float argument on stack
1485 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1486 break;
1487 case Bytecodes::_f2d:
1488 /* nothing to do */
1489 break;
1490 case Bytecodes::_d2i:
1491 __ push(rcx); // reserve space for argument
1492 __ push(rcx);
1493 __ fstp_d(at_rsp()); // pass double argument on stack
1494 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1495 break;
1496 case Bytecodes::_d2l:
1497 __ push(rcx); // reserve space for argument
1498 __ push(rcx);
1499 __ fstp_d(at_rsp()); // pass double argument on stack
1500 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1501 break;
1502 case Bytecodes::_d2f:
1503 __ push(rcx); // reserve space for f2ieee()
1504 __ f2ieee(); // truncate to float size
1505 __ pop(rcx); // adjust rsp
1506 break;
1507 default :
1508 ShouldNotReachHere();
1509 }
1510 }
1513 void TemplateTable::lcmp() {
1514 transition(ltos, itos);
1515 // y = rdx:rax
1516 __ pop_l(rbx, rcx); // get x = rcx:rbx
1517 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1518 __ mov(rax, rcx);
1519 }
1522 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1523 if (is_float) {
1524 __ fld_s(at_rsp());
1525 } else {
1526 __ fld_d(at_rsp());
1527 __ pop(rdx);
1528 }
1529 __ pop(rcx);
1530 __ fcmp2int(rax, unordered_result < 0);
1531 }
1534 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1535 __ get_method(rcx); // ECX holds method
1536 __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count
1538 const ByteSize be_offset = methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset();
1539 const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset();
1540 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1542 // Load up EDX with the branch displacement
1543 __ movl(rdx, at_bcp(1));
1544 __ bswapl(rdx);
1545 if (!is_wide) __ sarl(rdx, 16);
1546 LP64_ONLY(__ movslq(rdx, rdx));
1549 // Handle all the JSR stuff here, then exit.
1550 // It's much shorter and cleaner than intermingling with the
1551 // non-JSR normal-branch stuff occurring below.
1552 if (is_jsr) {
1553 // Pre-load the next target bytecode into EBX
1554 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0));
1556 // compute return address as bci in rax,
1557 __ lea(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(constMethodOopDesc::codes_offset())));
1558 __ subptr(rax, Address(rcx, methodOopDesc::const_offset()));
1559 // Adjust the bcp in RSI by the displacement in EDX
1560 __ addptr(rsi, rdx);
1561 // Push return address
1562 __ push_i(rax);
1563 // jsr returns vtos
1564 __ dispatch_only_noverify(vtos);
1565 return;
1566 }
1568 // Normal (non-jsr) branch handling
1570 // Adjust the bcp in RSI by the displacement in EDX
1571 __ addptr(rsi, rdx);
1573 assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1574 Label backedge_counter_overflow;
1575 Label profile_method;
1576 Label dispatch;
1577 if (UseLoopCounter) {
1578 // increment backedge counter for backward branches
1579 // rax,: MDO
1580 // rbx,: MDO bumped taken-count
1581 // rcx: method
1582 // rdx: target offset
1583 // rsi: target bcp
1584 // rdi: locals pointer
1585 __ testl(rdx, rdx); // check if forward or backward branch
1586 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1588 if (TieredCompilation) {
1589 Label no_mdo;
1590 int increment = InvocationCounter::count_increment;
1591 int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1592 if (ProfileInterpreter) {
1593 // Are we profiling?
1594 __ movptr(rbx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1595 __ testptr(rbx, rbx);
1596 __ jccb(Assembler::zero, no_mdo);
1597 // Increment the MDO backedge counter
1598 const Address mdo_backedge_counter(rbx, in_bytes(methodDataOopDesc::backedge_counter_offset()) +
1599 in_bytes(InvocationCounter::counter_offset()));
1600 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1601 rax, false, Assembler::zero, &backedge_counter_overflow);
1602 __ jmp(dispatch);
1603 }
1604 __ bind(no_mdo);
1605 // Increment backedge counter in methodOop
1606 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
1607 rax, false, Assembler::zero, &backedge_counter_overflow);
1608 } else {
1609 // increment counter
1610 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1611 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1612 __ movl(Address(rcx, be_offset), rax); // store counter
1614 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1615 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1616 __ addl(rax, Address(rcx, be_offset)); // add both counters
1618 if (ProfileInterpreter) {
1619 // Test to see if we should create a method data oop
1620 __ cmp32(rax,
1621 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1622 __ jcc(Assembler::less, dispatch);
1624 // if no method data exists, go to profile method
1625 __ test_method_data_pointer(rax, profile_method);
1627 if (UseOnStackReplacement) {
1628 // check for overflow against rbx, which is the MDO taken count
1629 __ cmp32(rbx,
1630 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1631 __ jcc(Assembler::below, dispatch);
1633 // When ProfileInterpreter is on, the backedge_count comes from the
1634 // methodDataOop, which value does not get reset on the call to
1635 // frequency_counter_overflow(). To avoid excessive calls to the overflow
1636 // routine while the method is being compiled, add a second test to make
1637 // sure the overflow function is called only once every overflow_frequency.
1638 const int overflow_frequency = 1024;
1639 __ andptr(rbx, overflow_frequency-1);
1640 __ jcc(Assembler::zero, backedge_counter_overflow);
1641 }
1642 } else {
1643 if (UseOnStackReplacement) {
1644 // check for overflow against rax, which is the sum of the counters
1645 __ cmp32(rax,
1646 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1647 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1649 }
1650 }
1651 }
1652 __ bind(dispatch);
1653 }
1655 // Pre-load the next target bytecode into EBX
1656 __ load_unsigned_byte(rbx, Address(rsi, 0));
1658 // continue with the bytecode @ target
1659 // rax,: return bci for jsr's, unused otherwise
1660 // rbx,: target bytecode
1661 // rsi: target bcp
1662 __ dispatch_only(vtos);
1664 if (UseLoopCounter) {
1665 if (ProfileInterpreter) {
1666 // Out-of-line code to allocate method data oop.
1667 __ bind(profile_method);
1668 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi);
1669 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1670 __ movptr(rcx, Address(rbp, method_offset));
1671 __ movptr(rcx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1672 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1673 __ test_method_data_pointer(rcx, dispatch);
1674 // offset non-null mdp by MDO::data_offset() + IR::profile_method()
1675 __ addptr(rcx, in_bytes(methodDataOopDesc::data_offset()));
1676 __ addptr(rcx, rax);
1677 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1678 __ jmp(dispatch);
1679 }
1681 if (UseOnStackReplacement) {
1683 // invocation counter overflow
1684 __ bind(backedge_counter_overflow);
1685 __ negptr(rdx);
1686 __ addptr(rdx, rsi); // branch bcp
1687 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx);
1688 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1690 // rax,: osr nmethod (osr ok) or NULL (osr not possible)
1691 // rbx,: target bytecode
1692 // rdx: scratch
1693 // rdi: locals pointer
1694 // rsi: bcp
1695 __ testptr(rax, rax); // test result
1696 __ jcc(Assembler::zero, dispatch); // no osr if null
1697 // nmethod may have been invalidated (VM may block upon call_VM return)
1698 __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1699 __ cmpl(rcx, InvalidOSREntryBci);
1700 __ jcc(Assembler::equal, dispatch);
1702 // We have the address of an on stack replacement routine in rax,
1703 // We need to prepare to execute the OSR method. First we must
1704 // migrate the locals and monitors off of the stack.
1706 __ mov(rbx, rax); // save the nmethod
1708 const Register thread = rcx;
1709 __ get_thread(thread);
1710 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1711 // rax, is OSR buffer, move it to expected parameter location
1712 __ mov(rcx, rax);
1714 // pop the interpreter frame
1715 __ movptr(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1716 __ leave(); // remove frame anchor
1717 __ pop(rdi); // get return address
1718 __ mov(rsp, rdx); // set sp to sender sp
1721 Label skip;
1722 Label chkint;
1724 // The interpreter frame we have removed may be returning to
1725 // either the callstub or the interpreter. Since we will
1726 // now be returning from a compiled (OSR) nmethod we must
1727 // adjust the return to the return were it can handler compiled
1728 // results and clean the fpu stack. This is very similar to
1729 // what a i2c adapter must do.
1731 // Are we returning to the call stub?
1733 __ cmp32(rdi, ExternalAddress(StubRoutines::_call_stub_return_address));
1734 __ jcc(Assembler::notEqual, chkint);
1736 // yes adjust to the specialized call stub return.
1737 assert(StubRoutines::x86::get_call_stub_compiled_return() != NULL, "must be set");
1738 __ lea(rdi, ExternalAddress(StubRoutines::x86::get_call_stub_compiled_return()));
1739 __ jmp(skip);
1741 __ bind(chkint);
1743 // Are we returning to the interpreter? Look for sentinel
1745 __ cmpl(Address(rdi, -2*wordSize), Interpreter::return_sentinel);
1746 __ jcc(Assembler::notEqual, skip);
1748 // Adjust to compiled return back to interpreter
1750 __ movptr(rdi, Address(rdi, -wordSize));
1751 __ bind(skip);
1753 // Align stack pointer for compiled code (note that caller is
1754 // responsible for undoing this fixup by remembering the old SP
1755 // in an rbp,-relative location)
1756 __ andptr(rsp, -(StackAlignmentInBytes));
1758 // push the (possibly adjusted) return address
1759 __ push(rdi);
1761 // and begin the OSR nmethod
1762 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
1763 }
1764 }
1765 }
1768 void TemplateTable::if_0cmp(Condition cc) {
1769 transition(itos, vtos);
1770 // assume branch is more often taken than not (loops use backward branches)
1771 Label not_taken;
1772 __ testl(rax, rax);
1773 __ jcc(j_not(cc), not_taken);
1774 branch(false, false);
1775 __ bind(not_taken);
1776 __ profile_not_taken_branch(rax);
1777 }
1780 void TemplateTable::if_icmp(Condition cc) {
1781 transition(itos, vtos);
1782 // assume branch is more often taken than not (loops use backward branches)
1783 Label not_taken;
1784 __ pop_i(rdx);
1785 __ cmpl(rdx, rax);
1786 __ jcc(j_not(cc), not_taken);
1787 branch(false, false);
1788 __ bind(not_taken);
1789 __ profile_not_taken_branch(rax);
1790 }
1793 void TemplateTable::if_nullcmp(Condition cc) {
1794 transition(atos, vtos);
1795 // assume branch is more often taken than not (loops use backward branches)
1796 Label not_taken;
1797 __ testptr(rax, rax);
1798 __ jcc(j_not(cc), not_taken);
1799 branch(false, false);
1800 __ bind(not_taken);
1801 __ profile_not_taken_branch(rax);
1802 }
1805 void TemplateTable::if_acmp(Condition cc) {
1806 transition(atos, vtos);
1807 // assume branch is more often taken than not (loops use backward branches)
1808 Label not_taken;
1809 __ pop_ptr(rdx);
1810 __ cmpptr(rdx, rax);
1811 __ jcc(j_not(cc), not_taken);
1812 branch(false, false);
1813 __ bind(not_taken);
1814 __ profile_not_taken_branch(rax);
1815 }
1818 void TemplateTable::ret() {
1819 transition(vtos, vtos);
1820 locals_index(rbx);
1821 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1822 __ profile_ret(rbx, rcx);
1823 __ get_method(rax);
1824 __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1825 __ lea(rsi, Address(rsi, rbx, Address::times_1,
1826 constMethodOopDesc::codes_offset()));
1827 __ dispatch_next(vtos);
1828 }
1831 void TemplateTable::wide_ret() {
1832 transition(vtos, vtos);
1833 locals_index_wide(rbx);
1834 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1835 __ profile_ret(rbx, rcx);
1836 __ get_method(rax);
1837 __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1838 __ lea(rsi, Address(rsi, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1839 __ dispatch_next(vtos);
1840 }
1843 void TemplateTable::tableswitch() {
1844 Label default_case, continue_execution;
1845 transition(itos, vtos);
1846 // align rsi
1847 __ lea(rbx, at_bcp(wordSize));
1848 __ andptr(rbx, -wordSize);
1849 // load lo & hi
1850 __ movl(rcx, Address(rbx, 1 * wordSize));
1851 __ movl(rdx, Address(rbx, 2 * wordSize));
1852 __ bswapl(rcx);
1853 __ bswapl(rdx);
1854 // check against lo & hi
1855 __ cmpl(rax, rcx);
1856 __ jccb(Assembler::less, default_case);
1857 __ cmpl(rax, rdx);
1858 __ jccb(Assembler::greater, default_case);
1859 // lookup dispatch offset
1860 __ subl(rax, rcx);
1861 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1862 __ profile_switch_case(rax, rbx, rcx);
1863 // continue execution
1864 __ bind(continue_execution);
1865 __ bswapl(rdx);
1866 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1867 __ addptr(rsi, rdx);
1868 __ dispatch_only(vtos);
1869 // handle default
1870 __ bind(default_case);
1871 __ profile_switch_default(rax);
1872 __ movl(rdx, Address(rbx, 0));
1873 __ jmp(continue_execution);
1874 }
1877 void TemplateTable::lookupswitch() {
1878 transition(itos, itos);
1879 __ stop("lookupswitch bytecode should have been rewritten");
1880 }
1883 void TemplateTable::fast_linearswitch() {
1884 transition(itos, vtos);
1885 Label loop_entry, loop, found, continue_execution;
1886 // bswapl rax, so we can avoid bswapping the table entries
1887 __ bswapl(rax);
1888 // align rsi
1889 __ lea(rbx, at_bcp(wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1890 __ andptr(rbx, -wordSize);
1891 // set counter
1892 __ movl(rcx, Address(rbx, wordSize));
1893 __ bswapl(rcx);
1894 __ jmpb(loop_entry);
1895 // table search
1896 __ bind(loop);
1897 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize));
1898 __ jccb(Assembler::equal, found);
1899 __ bind(loop_entry);
1900 __ decrementl(rcx);
1901 __ jcc(Assembler::greaterEqual, loop);
1902 // default case
1903 __ profile_switch_default(rax);
1904 __ movl(rdx, Address(rbx, 0));
1905 __ jmpb(continue_execution);
1906 // entry found -> get offset
1907 __ bind(found);
1908 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize));
1909 __ profile_switch_case(rcx, rax, rbx);
1910 // continue execution
1911 __ bind(continue_execution);
1912 __ bswapl(rdx);
1913 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1914 __ addptr(rsi, rdx);
1915 __ dispatch_only(vtos);
1916 }
1919 void TemplateTable::fast_binaryswitch() {
1920 transition(itos, vtos);
1921 // Implementation using the following core algorithm:
1922 //
1923 // int binary_search(int key, LookupswitchPair* array, int n) {
1924 // // Binary search according to "Methodik des Programmierens" by
1925 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1926 // int i = 0;
1927 // int j = n;
1928 // while (i+1 < j) {
1929 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1930 // // with Q: for all i: 0 <= i < n: key < a[i]
1931 // // where a stands for the array and assuming that the (inexisting)
1932 // // element a[n] is infinitely big.
1933 // int h = (i + j) >> 1;
1934 // // i < h < j
1935 // if (key < array[h].fast_match()) {
1936 // j = h;
1937 // } else {
1938 // i = h;
1939 // }
1940 // }
1941 // // R: a[i] <= key < a[i+1] or Q
1942 // // (i.e., if key is within array, i is the correct index)
1943 // return i;
1944 // }
1946 // register allocation
1947 const Register key = rax; // already set (tosca)
1948 const Register array = rbx;
1949 const Register i = rcx;
1950 const Register j = rdx;
1951 const Register h = rdi; // needs to be restored
1952 const Register temp = rsi;
1953 // setup array
1954 __ save_bcp();
1956 __ lea(array, at_bcp(3*wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1957 __ andptr(array, -wordSize);
1958 // initialize i & j
1959 __ xorl(i, i); // i = 0;
1960 __ movl(j, Address(array, -wordSize)); // j = length(array);
1961 // Convert j into native byteordering
1962 __ bswapl(j);
1963 // and start
1964 Label entry;
1965 __ jmp(entry);
1967 // binary search loop
1968 { Label loop;
1969 __ bind(loop);
1970 // int h = (i + j) >> 1;
1971 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1972 __ sarl(h, 1); // h = (i + j) >> 1;
1973 // if (key < array[h].fast_match()) {
1974 // j = h;
1975 // } else {
1976 // i = h;
1977 // }
1978 // Convert array[h].match to native byte-ordering before compare
1979 __ movl(temp, Address(array, h, Address::times_8, 0*wordSize));
1980 __ bswapl(temp);
1981 __ cmpl(key, temp);
1982 if (VM_Version::supports_cmov()) {
1983 __ cmovl(Assembler::less , j, h); // j = h if (key < array[h].fast_match())
1984 __ cmovl(Assembler::greaterEqual, i, h); // i = h if (key >= array[h].fast_match())
1985 } else {
1986 Label set_i, end_of_if;
1987 __ jccb(Assembler::greaterEqual, set_i); // {
1988 __ mov(j, h); // j = h;
1989 __ jmp(end_of_if); // }
1990 __ bind(set_i); // else {
1991 __ mov(i, h); // i = h;
1992 __ bind(end_of_if); // }
1993 }
1994 // while (i+1 < j)
1995 __ bind(entry);
1996 __ leal(h, Address(i, 1)); // i+1
1997 __ cmpl(h, j); // i+1 < j
1998 __ jcc(Assembler::less, loop);
1999 }
2001 // end of binary search, result index is i (must check again!)
2002 Label default_case;
2003 // Convert array[i].match to native byte-ordering before compare
2004 __ movl(temp, Address(array, i, Address::times_8, 0*wordSize));
2005 __ bswapl(temp);
2006 __ cmpl(key, temp);
2007 __ jcc(Assembler::notEqual, default_case);
2009 // entry found -> j = offset
2010 __ movl(j , Address(array, i, Address::times_8, 1*wordSize));
2011 __ profile_switch_case(i, key, array);
2012 __ bswapl(j);
2013 LP64_ONLY(__ movslq(j, j));
2014 __ restore_bcp();
2015 __ restore_locals(); // restore rdi
2016 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
2018 __ addptr(rsi, j);
2019 __ dispatch_only(vtos);
2021 // default case -> j = default offset
2022 __ bind(default_case);
2023 __ profile_switch_default(i);
2024 __ movl(j, Address(array, -2*wordSize));
2025 __ bswapl(j);
2026 LP64_ONLY(__ movslq(j, j));
2027 __ restore_bcp();
2028 __ restore_locals(); // restore rdi
2029 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
2030 __ addptr(rsi, j);
2031 __ dispatch_only(vtos);
2032 }
2035 void TemplateTable::_return(TosState state) {
2036 transition(state, state);
2037 assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation
2039 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2040 assert(state == vtos, "only valid state");
2041 __ movptr(rax, aaddress(0));
2042 __ movptr(rdi, Address(rax, oopDesc::klass_offset_in_bytes()));
2043 __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
2044 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2045 Label skip_register_finalizer;
2046 __ jcc(Assembler::zero, skip_register_finalizer);
2048 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax);
2050 __ bind(skip_register_finalizer);
2051 }
2053 __ remove_activation(state, rsi);
2054 __ jmp(rsi);
2055 }
2058 // ----------------------------------------------------------------------------
2059 // Volatile variables demand their effects be made known to all CPU's in
2060 // order. Store buffers on most chips allow reads & writes to reorder; the
2061 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2062 // memory barrier (i.e., it's not sufficient that the interpreter does not
2063 // reorder volatile references, the hardware also must not reorder them).
2064 //
2065 // According to the new Java Memory Model (JMM):
2066 // (1) All volatiles are serialized wrt to each other.
2067 // ALSO reads & writes act as aquire & release, so:
2068 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
2069 // the read float up to before the read. It's OK for non-volatile memory refs
2070 // that happen before the volatile read to float down below it.
2071 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2072 // that happen BEFORE the write float down to after the write. It's OK for
2073 // non-volatile memory refs that happen after the volatile write to float up
2074 // before it.
2075 //
2076 // We only put in barriers around volatile refs (they are expensive), not
2077 // _between_ memory refs (that would require us to track the flavor of the
2078 // previous memory refs). Requirements (2) and (3) require some barriers
2079 // before volatile stores and after volatile loads. These nearly cover
2080 // requirement (1) but miss the volatile-store-volatile-load case. This final
2081 // case is placed after volatile-stores although it could just as well go
2082 // before volatile-loads.
2083 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2084 // Helper function to insert a is-volatile test and memory barrier
2085 if( !os::is_MP() ) return; // Not needed on single CPU
2086 __ membar(order_constraint);
2087 }
2089 void TemplateTable::resolve_cache_and_index(int byte_no,
2090 Register result,
2091 Register Rcache,
2092 Register index,
2093 size_t index_size) {
2094 Register temp = rbx;
2096 assert_different_registers(result, Rcache, index, temp);
2098 Label resolved;
2099 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2100 if (byte_no == f1_oop) {
2101 // We are resolved if the f1 field contains a non-null object (CallSite, etc.)
2102 // This kind of CP cache entry does not need to match the flags byte, because
2103 // there is a 1-1 relation between bytecode type and CP entry type.
2104 assert(result != noreg, ""); //else do cmpptr(Address(...), (int32_t) NULL_WORD)
2105 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2106 __ testptr(result, result);
2107 __ jcc(Assembler::notEqual, resolved);
2108 } else {
2109 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2110 assert(result == noreg, ""); //else change code for setting result
2111 const int shift_count = (1 + byte_no)*BitsPerByte;
2112 __ movl(temp, Address(Rcache, index, Address::times_4, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
2113 __ shrl(temp, shift_count);
2114 // have we resolved this bytecode?
2115 __ andl(temp, 0xFF);
2116 __ cmpl(temp, (int)bytecode());
2117 __ jcc(Assembler::equal, resolved);
2118 }
2120 // resolve first time through
2121 address entry;
2122 switch (bytecode()) {
2123 case Bytecodes::_getstatic : // fall through
2124 case Bytecodes::_putstatic : // fall through
2125 case Bytecodes::_getfield : // fall through
2126 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2127 case Bytecodes::_invokevirtual : // fall through
2128 case Bytecodes::_invokespecial : // fall through
2129 case Bytecodes::_invokestatic : // fall through
2130 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2131 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2132 case Bytecodes::_fast_aldc : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); break;
2133 case Bytecodes::_fast_aldc_w : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc); break;
2134 default : ShouldNotReachHere(); break;
2135 }
2136 __ movl(temp, (int)bytecode());
2137 __ call_VM(noreg, entry, temp);
2138 // Update registers with resolved info
2139 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2140 if (result != noreg)
2141 __ movptr(result, Address(Rcache, index, Address::times_ptr, constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f1_offset()));
2142 __ bind(resolved);
2143 }
2146 // The cache and index registers must be set before call
2147 void TemplateTable::load_field_cp_cache_entry(Register obj,
2148 Register cache,
2149 Register index,
2150 Register off,
2151 Register flags,
2152 bool is_static = false) {
2153 assert_different_registers(cache, index, flags, off);
2155 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2156 // Field offset
2157 __ movptr(off, Address(cache, index, Address::times_ptr,
2158 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset())));
2159 // Flags
2160 __ movl(flags, Address(cache, index, Address::times_ptr,
2161 in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset())));
2163 // klass overwrite register
2164 if (is_static) {
2165 __ movptr(obj, Address(cache, index, Address::times_ptr,
2166 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset())));
2167 }
2168 }
2170 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2171 Register method,
2172 Register itable_index,
2173 Register flags,
2174 bool is_invokevirtual,
2175 bool is_invokevfinal /*unused*/,
2176 bool is_invokedynamic) {
2177 // setup registers
2178 const Register cache = rcx;
2179 const Register index = rdx;
2180 assert_different_registers(method, flags);
2181 assert_different_registers(method, cache, index);
2182 assert_different_registers(itable_index, flags);
2183 assert_different_registers(itable_index, cache, index);
2184 // determine constant pool cache field offsets
2185 const int method_offset = in_bytes(
2186 constantPoolCacheOopDesc::base_offset() +
2187 (is_invokevirtual
2188 ? ConstantPoolCacheEntry::f2_offset()
2189 : ConstantPoolCacheEntry::f1_offset()
2190 )
2191 );
2192 const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2193 ConstantPoolCacheEntry::flags_offset());
2194 // access constant pool cache fields
2195 const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2196 ConstantPoolCacheEntry::f2_offset());
2198 if (byte_no == f1_oop) {
2199 // Resolved f1_oop goes directly into 'method' register.
2200 assert(is_invokedynamic, "");
2201 resolve_cache_and_index(byte_no, method, cache, index, sizeof(u4));
2202 } else {
2203 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2204 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2205 }
2206 if (itable_index != noreg) {
2207 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2208 }
2209 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2210 }
2213 // The registers cache and index expected to be set before call.
2214 // Correct values of the cache and index registers are preserved.
2215 void TemplateTable::jvmti_post_field_access(Register cache,
2216 Register index,
2217 bool is_static,
2218 bool has_tos) {
2219 if (JvmtiExport::can_post_field_access()) {
2220 // Check to see if a field access watch has been set before we take
2221 // the time to call into the VM.
2222 Label L1;
2223 assert_different_registers(cache, index, rax);
2224 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2225 __ testl(rax,rax);
2226 __ jcc(Assembler::zero, L1);
2228 // cache entry pointer
2229 __ addptr(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
2230 __ shll(index, LogBytesPerWord);
2231 __ addptr(cache, index);
2232 if (is_static) {
2233 __ xorptr(rax, rax); // NULL object reference
2234 } else {
2235 __ pop(atos); // Get the object
2236 __ verify_oop(rax);
2237 __ push(atos); // Restore stack state
2238 }
2239 // rax,: object pointer or NULL
2240 // cache: cache entry pointer
2241 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2242 rax, cache);
2243 __ get_cache_and_index_at_bcp(cache, index, 1);
2244 __ bind(L1);
2245 }
2246 }
2248 void TemplateTable::pop_and_check_object(Register r) {
2249 __ pop_ptr(r);
2250 __ null_check(r); // for field access must check obj.
2251 __ verify_oop(r);
2252 }
2254 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2255 transition(vtos, vtos);
2257 const Register cache = rcx;
2258 const Register index = rdx;
2259 const Register obj = rcx;
2260 const Register off = rbx;
2261 const Register flags = rax;
2263 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2264 jvmti_post_field_access(cache, index, is_static, false);
2265 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2267 if (!is_static) pop_and_check_object(obj);
2269 const Address lo(obj, off, Address::times_1, 0*wordSize);
2270 const Address hi(obj, off, Address::times_1, 1*wordSize);
2272 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2274 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2275 assert(btos == 0, "change code, btos != 0");
2276 // btos
2277 __ andptr(flags, 0x0f);
2278 __ jcc(Assembler::notZero, notByte);
2280 __ load_signed_byte(rax, lo );
2281 __ push(btos);
2282 // Rewrite bytecode to be faster
2283 if (!is_static) {
2284 patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx);
2285 }
2286 __ jmp(Done);
2288 __ bind(notByte);
2289 // itos
2290 __ cmpl(flags, itos );
2291 __ jcc(Assembler::notEqual, notInt);
2293 __ movl(rax, lo );
2294 __ push(itos);
2295 // Rewrite bytecode to be faster
2296 if (!is_static) {
2297 patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx);
2298 }
2299 __ jmp(Done);
2301 __ bind(notInt);
2302 // atos
2303 __ cmpl(flags, atos );
2304 __ jcc(Assembler::notEqual, notObj);
2306 __ movl(rax, lo );
2307 __ push(atos);
2308 if (!is_static) {
2309 patch_bytecode(Bytecodes::_fast_agetfield, rcx, rbx);
2310 }
2311 __ jmp(Done);
2313 __ bind(notObj);
2314 // ctos
2315 __ cmpl(flags, ctos );
2316 __ jcc(Assembler::notEqual, notChar);
2318 __ load_unsigned_short(rax, lo );
2319 __ push(ctos);
2320 if (!is_static) {
2321 patch_bytecode(Bytecodes::_fast_cgetfield, rcx, rbx);
2322 }
2323 __ jmp(Done);
2325 __ bind(notChar);
2326 // stos
2327 __ cmpl(flags, stos );
2328 __ jcc(Assembler::notEqual, notShort);
2330 __ load_signed_short(rax, lo );
2331 __ push(stos);
2332 if (!is_static) {
2333 patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx);
2334 }
2335 __ jmp(Done);
2337 __ bind(notShort);
2338 // ltos
2339 __ cmpl(flags, ltos );
2340 __ jcc(Assembler::notEqual, notLong);
2342 // Generate code as if volatile. There just aren't enough registers to
2343 // save that information and this code is faster than the test.
2344 __ fild_d(lo); // Must load atomically
2345 __ subptr(rsp,2*wordSize); // Make space for store
2346 __ fistp_d(Address(rsp,0));
2347 __ pop(rax);
2348 __ pop(rdx);
2350 __ push(ltos);
2351 // Don't rewrite to _fast_lgetfield for potential volatile case.
2352 __ jmp(Done);
2354 __ bind(notLong);
2355 // ftos
2356 __ cmpl(flags, ftos );
2357 __ jcc(Assembler::notEqual, notFloat);
2359 __ fld_s(lo);
2360 __ push(ftos);
2361 if (!is_static) {
2362 patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx);
2363 }
2364 __ jmp(Done);
2366 __ bind(notFloat);
2367 // dtos
2368 __ cmpl(flags, dtos );
2369 __ jcc(Assembler::notEqual, notDouble);
2371 __ fld_d(lo);
2372 __ push(dtos);
2373 if (!is_static) {
2374 patch_bytecode(Bytecodes::_fast_dgetfield, rcx, rbx);
2375 }
2376 __ jmpb(Done);
2378 __ bind(notDouble);
2380 __ stop("Bad state");
2382 __ bind(Done);
2383 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2384 // volatile_barrier( );
2385 }
2388 void TemplateTable::getfield(int byte_no) {
2389 getfield_or_static(byte_no, false);
2390 }
2393 void TemplateTable::getstatic(int byte_no) {
2394 getfield_or_static(byte_no, true);
2395 }
2397 // The registers cache and index expected to be set before call.
2398 // The function may destroy various registers, just not the cache and index registers.
2399 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2401 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2403 if (JvmtiExport::can_post_field_modification()) {
2404 // Check to see if a field modification watch has been set before we take
2405 // the time to call into the VM.
2406 Label L1;
2407 assert_different_registers(cache, index, rax);
2408 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2409 __ testl(rax, rax);
2410 __ jcc(Assembler::zero, L1);
2412 // The cache and index registers have been already set.
2413 // This allows to eliminate this call but the cache and index
2414 // registers have to be correspondingly used after this line.
2415 __ get_cache_and_index_at_bcp(rax, rdx, 1);
2417 if (is_static) {
2418 // Life is simple. Null out the object pointer.
2419 __ xorptr(rbx, rbx);
2420 } else {
2421 // Life is harder. The stack holds the value on top, followed by the object.
2422 // We don't know the size of the value, though; it could be one or two words
2423 // depending on its type. As a result, we must find the type to determine where
2424 // the object is.
2425 Label two_word, valsize_known;
2426 __ movl(rcx, Address(rax, rdx, Address::times_ptr, in_bytes(cp_base_offset +
2427 ConstantPoolCacheEntry::flags_offset())));
2428 __ mov(rbx, rsp);
2429 __ shrl(rcx, ConstantPoolCacheEntry::tosBits);
2430 // Make sure we don't need to mask rcx for tosBits after the above shift
2431 ConstantPoolCacheEntry::verify_tosBits();
2432 __ cmpl(rcx, ltos);
2433 __ jccb(Assembler::equal, two_word);
2434 __ cmpl(rcx, dtos);
2435 __ jccb(Assembler::equal, two_word);
2436 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2437 __ jmpb(valsize_known);
2439 __ bind(two_word);
2440 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2442 __ bind(valsize_known);
2443 // setup object pointer
2444 __ movptr(rbx, Address(rbx, 0));
2445 }
2446 // cache entry pointer
2447 __ addptr(rax, in_bytes(cp_base_offset));
2448 __ shll(rdx, LogBytesPerWord);
2449 __ addptr(rax, rdx);
2450 // object (tos)
2451 __ mov(rcx, rsp);
2452 // rbx,: object pointer set up above (NULL if static)
2453 // rax,: cache entry pointer
2454 // rcx: jvalue object on the stack
2455 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2456 rbx, rax, rcx);
2457 __ get_cache_and_index_at_bcp(cache, index, 1);
2458 __ bind(L1);
2459 }
2460 }
2463 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2464 transition(vtos, vtos);
2466 const Register cache = rcx;
2467 const Register index = rdx;
2468 const Register obj = rcx;
2469 const Register off = rbx;
2470 const Register flags = rax;
2472 resolve_cache_and_index(byte_no, noreg, cache, index, sizeof(u2));
2473 jvmti_post_field_mod(cache, index, is_static);
2474 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2476 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2477 // volatile_barrier( );
2479 Label notVolatile, Done;
2480 __ movl(rdx, flags);
2481 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2482 __ andl(rdx, 0x1);
2484 // field addresses
2485 const Address lo(obj, off, Address::times_1, 0*wordSize);
2486 const Address hi(obj, off, Address::times_1, 1*wordSize);
2488 Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2490 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2491 assert(btos == 0, "change code, btos != 0");
2492 // btos
2493 __ andl(flags, 0x0f);
2494 __ jcc(Assembler::notZero, notByte);
2496 __ pop(btos);
2497 if (!is_static) pop_and_check_object(obj);
2498 __ movb(lo, rax );
2499 if (!is_static) {
2500 patch_bytecode(Bytecodes::_fast_bputfield, rcx, rbx);
2501 }
2502 __ jmp(Done);
2504 __ bind(notByte);
2505 // itos
2506 __ cmpl(flags, itos );
2507 __ jcc(Assembler::notEqual, notInt);
2509 __ pop(itos);
2510 if (!is_static) pop_and_check_object(obj);
2512 __ movl(lo, rax );
2513 if (!is_static) {
2514 patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx);
2515 }
2516 __ jmp(Done);
2518 __ bind(notInt);
2519 // atos
2520 __ cmpl(flags, atos );
2521 __ jcc(Assembler::notEqual, notObj);
2523 __ pop(atos);
2524 if (!is_static) pop_and_check_object(obj);
2526 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2528 if (!is_static) {
2529 patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx);
2530 }
2532 __ jmp(Done);
2534 __ bind(notObj);
2535 // ctos
2536 __ cmpl(flags, ctos );
2537 __ jcc(Assembler::notEqual, notChar);
2539 __ pop(ctos);
2540 if (!is_static) pop_and_check_object(obj);
2541 __ movw(lo, rax );
2542 if (!is_static) {
2543 patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx);
2544 }
2545 __ jmp(Done);
2547 __ bind(notChar);
2548 // stos
2549 __ cmpl(flags, stos );
2550 __ jcc(Assembler::notEqual, notShort);
2552 __ pop(stos);
2553 if (!is_static) pop_and_check_object(obj);
2554 __ movw(lo, rax );
2555 if (!is_static) {
2556 patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx);
2557 }
2558 __ jmp(Done);
2560 __ bind(notShort);
2561 // ltos
2562 __ cmpl(flags, ltos );
2563 __ jcc(Assembler::notEqual, notLong);
2565 Label notVolatileLong;
2566 __ testl(rdx, rdx);
2567 __ jcc(Assembler::zero, notVolatileLong);
2569 __ pop(ltos); // overwrites rdx, do this after testing volatile.
2570 if (!is_static) pop_and_check_object(obj);
2572 // Replace with real volatile test
2573 __ push(rdx);
2574 __ push(rax); // Must update atomically with FIST
2575 __ fild_d(Address(rsp,0)); // So load into FPU register
2576 __ fistp_d(lo); // and put into memory atomically
2577 __ addptr(rsp, 2*wordSize);
2578 // volatile_barrier();
2579 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2580 Assembler::StoreStore));
2581 // Don't rewrite volatile version
2582 __ jmp(notVolatile);
2584 __ bind(notVolatileLong);
2586 __ pop(ltos); // overwrites rdx
2587 if (!is_static) pop_and_check_object(obj);
2588 NOT_LP64(__ movptr(hi, rdx));
2589 __ movptr(lo, rax);
2590 if (!is_static) {
2591 patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx);
2592 }
2593 __ jmp(notVolatile);
2595 __ bind(notLong);
2596 // ftos
2597 __ cmpl(flags, ftos );
2598 __ jcc(Assembler::notEqual, notFloat);
2600 __ pop(ftos);
2601 if (!is_static) pop_and_check_object(obj);
2602 __ fstp_s(lo);
2603 if (!is_static) {
2604 patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx);
2605 }
2606 __ jmp(Done);
2608 __ bind(notFloat);
2609 // dtos
2610 __ cmpl(flags, dtos );
2611 __ jcc(Assembler::notEqual, notDouble);
2613 __ pop(dtos);
2614 if (!is_static) pop_and_check_object(obj);
2615 __ fstp_d(lo);
2616 if (!is_static) {
2617 patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx);
2618 }
2619 __ jmp(Done);
2621 __ bind(notDouble);
2623 __ stop("Bad state");
2625 __ bind(Done);
2627 // Check for volatile store
2628 __ testl(rdx, rdx);
2629 __ jcc(Assembler::zero, notVolatile);
2630 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2631 Assembler::StoreStore));
2632 __ bind(notVolatile);
2633 }
2636 void TemplateTable::putfield(int byte_no) {
2637 putfield_or_static(byte_no, false);
2638 }
2641 void TemplateTable::putstatic(int byte_no) {
2642 putfield_or_static(byte_no, true);
2643 }
2645 void TemplateTable::jvmti_post_fast_field_mod() {
2646 if (JvmtiExport::can_post_field_modification()) {
2647 // Check to see if a field modification watch has been set before we take
2648 // the time to call into the VM.
2649 Label L2;
2650 __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2651 __ testl(rcx,rcx);
2652 __ jcc(Assembler::zero, L2);
2653 __ pop_ptr(rbx); // copy the object pointer from tos
2654 __ verify_oop(rbx);
2655 __ push_ptr(rbx); // put the object pointer back on tos
2656 __ subptr(rsp, sizeof(jvalue)); // add space for a jvalue object
2657 __ mov(rcx, rsp);
2658 __ push_ptr(rbx); // save object pointer so we can steal rbx,
2659 __ xorptr(rbx, rbx);
2660 const Address lo_value(rcx, rbx, Address::times_1, 0*wordSize);
2661 const Address hi_value(rcx, rbx, Address::times_1, 1*wordSize);
2662 switch (bytecode()) { // load values into the jvalue object
2663 case Bytecodes::_fast_bputfield: __ movb(lo_value, rax); break;
2664 case Bytecodes::_fast_sputfield: __ movw(lo_value, rax); break;
2665 case Bytecodes::_fast_cputfield: __ movw(lo_value, rax); break;
2666 case Bytecodes::_fast_iputfield: __ movl(lo_value, rax); break;
2667 case Bytecodes::_fast_lputfield:
2668 NOT_LP64(__ movptr(hi_value, rdx));
2669 __ movptr(lo_value, rax);
2670 break;
2672 // need to call fld_s() after fstp_s() to restore the value for below
2673 case Bytecodes::_fast_fputfield: __ fstp_s(lo_value); __ fld_s(lo_value); break;
2675 // need to call fld_d() after fstp_d() to restore the value for below
2676 case Bytecodes::_fast_dputfield: __ fstp_d(lo_value); __ fld_d(lo_value); break;
2678 // since rcx is not an object we don't call store_check() here
2679 case Bytecodes::_fast_aputfield: __ movptr(lo_value, rax); break;
2681 default: ShouldNotReachHere();
2682 }
2683 __ pop_ptr(rbx); // restore copy of object pointer
2685 // Save rax, and sometimes rdx because call_VM() will clobber them,
2686 // then use them for JVM/DI purposes
2687 __ push(rax);
2688 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2689 // access constant pool cache entry
2690 __ get_cache_entry_pointer_at_bcp(rax, rdx, 1);
2691 __ verify_oop(rbx);
2692 // rbx,: object pointer copied above
2693 // rax,: cache entry pointer
2694 // rcx: jvalue object on the stack
2695 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx);
2696 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); // restore high value
2697 __ pop(rax); // restore lower value
2698 __ addptr(rsp, sizeof(jvalue)); // release jvalue object space
2699 __ bind(L2);
2700 }
2701 }
2703 void TemplateTable::fast_storefield(TosState state) {
2704 transition(state, vtos);
2706 ByteSize base = constantPoolCacheOopDesc::base_offset();
2708 jvmti_post_fast_field_mod();
2710 // access constant pool cache
2711 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2713 // test for volatile with rdx but rdx is tos register for lputfield.
2714 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2715 __ movl(rdx, Address(rcx, rbx, Address::times_ptr, in_bytes(base +
2716 ConstantPoolCacheEntry::flags_offset())));
2718 // replace index with field offset from cache entry
2719 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2721 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2722 // volatile_barrier( );
2724 Label notVolatile, Done;
2725 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2726 __ andl(rdx, 0x1);
2727 // Check for volatile store
2728 __ testl(rdx, rdx);
2729 __ jcc(Assembler::zero, notVolatile);
2731 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2733 // Get object from stack
2734 pop_and_check_object(rcx);
2736 // field addresses
2737 const Address lo(rcx, rbx, Address::times_1, 0*wordSize);
2738 const Address hi(rcx, rbx, Address::times_1, 1*wordSize);
2740 // access field
2741 switch (bytecode()) {
2742 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2743 case Bytecodes::_fast_sputfield: // fall through
2744 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2745 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2746 case Bytecodes::_fast_lputfield:
2747 NOT_LP64(__ movptr(hi, rdx));
2748 __ movptr(lo, rax);
2749 break;
2750 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2751 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2752 case Bytecodes::_fast_aputfield: {
2753 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2754 break;
2755 }
2756 default:
2757 ShouldNotReachHere();
2758 }
2760 Label done;
2761 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2762 Assembler::StoreStore));
2763 // Barriers are so large that short branch doesn't reach!
2764 __ jmp(done);
2766 // Same code as above, but don't need rdx to test for volatile.
2767 __ bind(notVolatile);
2769 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2771 // Get object from stack
2772 pop_and_check_object(rcx);
2774 // access field
2775 switch (bytecode()) {
2776 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2777 case Bytecodes::_fast_sputfield: // fall through
2778 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2779 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2780 case Bytecodes::_fast_lputfield:
2781 NOT_LP64(__ movptr(hi, rdx));
2782 __ movptr(lo, rax);
2783 break;
2784 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2785 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2786 case Bytecodes::_fast_aputfield: {
2787 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2788 break;
2789 }
2790 default:
2791 ShouldNotReachHere();
2792 }
2793 __ bind(done);
2794 }
2797 void TemplateTable::fast_accessfield(TosState state) {
2798 transition(atos, state);
2800 // do the JVMTI work here to avoid disturbing the register state below
2801 if (JvmtiExport::can_post_field_access()) {
2802 // Check to see if a field access watch has been set before we take
2803 // the time to call into the VM.
2804 Label L1;
2805 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2806 __ testl(rcx,rcx);
2807 __ jcc(Assembler::zero, L1);
2808 // access constant pool cache entry
2809 __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1);
2810 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
2811 __ verify_oop(rax);
2812 // rax,: object pointer copied above
2813 // rcx: cache entry pointer
2814 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx);
2815 __ pop_ptr(rax); // restore object pointer
2816 __ bind(L1);
2817 }
2819 // access constant pool cache
2820 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2821 // replace index with field offset from cache entry
2822 __ movptr(rbx, Address(rcx,
2823 rbx,
2824 Address::times_ptr,
2825 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2828 // rax,: object
2829 __ verify_oop(rax);
2830 __ null_check(rax);
2831 // field addresses
2832 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2833 const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize);
2835 // access field
2836 switch (bytecode()) {
2837 case Bytecodes::_fast_bgetfield: __ movsbl(rax, lo ); break;
2838 case Bytecodes::_fast_sgetfield: __ load_signed_short(rax, lo ); break;
2839 case Bytecodes::_fast_cgetfield: __ load_unsigned_short(rax, lo ); break;
2840 case Bytecodes::_fast_igetfield: __ movl(rax, lo); break;
2841 case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten"); break;
2842 case Bytecodes::_fast_fgetfield: __ fld_s(lo); break;
2843 case Bytecodes::_fast_dgetfield: __ fld_d(lo); break;
2844 case Bytecodes::_fast_agetfield: __ movptr(rax, lo); __ verify_oop(rax); break;
2845 default:
2846 ShouldNotReachHere();
2847 }
2849 // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
2850 // volatile_barrier( );
2851 }
2853 void TemplateTable::fast_xaccess(TosState state) {
2854 transition(vtos, state);
2855 // get receiver
2856 __ movptr(rax, aaddress(0));
2857 // access constant pool cache
2858 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2859 __ movptr(rbx, Address(rcx,
2860 rdx,
2861 Address::times_ptr,
2862 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2863 // make sure exception is reported in correct bcp range (getfield is next instruction)
2864 __ increment(rsi);
2865 __ null_check(rax);
2866 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2867 if (state == itos) {
2868 __ movl(rax, lo);
2869 } else if (state == atos) {
2870 __ movptr(rax, lo);
2871 __ verify_oop(rax);
2872 } else if (state == ftos) {
2873 __ fld_s(lo);
2874 } else {
2875 ShouldNotReachHere();
2876 }
2877 __ decrement(rsi);
2878 }
2882 //----------------------------------------------------------------------------------------------------
2883 // Calls
2885 void TemplateTable::count_calls(Register method, Register temp) {
2886 // implemented elsewhere
2887 ShouldNotReachHere();
2888 }
2891 void TemplateTable::prepare_invoke(Register method, Register index, int byte_no) {
2892 // determine flags
2893 Bytecodes::Code code = bytecode();
2894 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2895 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
2896 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2897 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2898 const bool load_receiver = (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic);
2899 const bool receiver_null_check = is_invokespecial;
2900 const bool save_flags = is_invokeinterface || is_invokevirtual;
2901 // setup registers & access constant pool cache
2902 const Register recv = rcx;
2903 const Register flags = rdx;
2904 assert_different_registers(method, index, recv, flags);
2906 // save 'interpreter return address'
2907 __ save_bcp();
2909 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2911 // load receiver if needed (note: no return address pushed yet)
2912 if (load_receiver) {
2913 assert(!is_invokedynamic, "");
2914 __ movl(recv, flags);
2915 __ andl(recv, 0xFF);
2916 // recv count is 0 based?
2917 Address recv_addr(rsp, recv, Interpreter::stackElementScale(), -Interpreter::expr_offset_in_bytes(1));
2918 __ movptr(recv, recv_addr);
2919 __ verify_oop(recv);
2920 }
2922 // do null check if needed
2923 if (receiver_null_check) {
2924 __ null_check(recv);
2925 }
2927 if (save_flags) {
2928 __ mov(rsi, flags);
2929 }
2931 // compute return type
2932 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2933 // Make sure we don't need to mask flags for tosBits after the above shift
2934 ConstantPoolCacheEntry::verify_tosBits();
2935 // load return address
2936 {
2937 address table_addr;
2938 if (is_invokeinterface || is_invokedynamic)
2939 table_addr = (address)Interpreter::return_5_addrs_by_index_table();
2940 else
2941 table_addr = (address)Interpreter::return_3_addrs_by_index_table();
2942 ExternalAddress table(table_addr);
2943 __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));
2944 }
2946 // push return address
2947 __ push(flags);
2949 // Restore flag value from the constant pool cache, and restore rsi
2950 // for later null checks. rsi is the bytecode pointer
2951 if (save_flags) {
2952 __ mov(flags, rsi);
2953 __ restore_bcp();
2954 }
2955 }
2958 void TemplateTable::invokevirtual_helper(Register index, Register recv,
2959 Register flags) {
2961 // Uses temporary registers rax, rdx
2962 assert_different_registers(index, recv, rax, rdx);
2964 // Test for an invoke of a final method
2965 Label notFinal;
2966 __ movl(rax, flags);
2967 __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
2968 __ jcc(Assembler::zero, notFinal);
2970 Register method = index; // method must be rbx,
2971 assert(method == rbx, "methodOop must be rbx, for interpreter calling convention");
2973 // do the call - the index is actually the method to call
2974 __ verify_oop(method);
2976 // It's final, need a null check here!
2977 __ null_check(recv);
2979 // profile this call
2980 __ profile_final_call(rax);
2982 __ jump_from_interpreted(method, rax);
2984 __ bind(notFinal);
2986 // get receiver klass
2987 __ null_check(recv, oopDesc::klass_offset_in_bytes());
2988 // Keep recv in rcx for callee expects it there
2989 __ movptr(rax, Address(recv, oopDesc::klass_offset_in_bytes()));
2990 __ verify_oop(rax);
2992 // profile this call
2993 __ profile_virtual_call(rax, rdi, rdx);
2995 // get target methodOop & entry point
2996 const int base = instanceKlass::vtable_start_offset() * wordSize;
2997 assert(vtableEntry::size() * wordSize == 4, "adjust the scaling in the code below");
2998 __ movptr(method, Address(rax, index, Address::times_ptr, base + vtableEntry::method_offset_in_bytes()));
2999 __ jump_from_interpreted(method, rdx);
3000 }
3003 void TemplateTable::invokevirtual(int byte_no) {
3004 transition(vtos, vtos);
3005 assert(byte_no == f2_byte, "use this argument");
3006 prepare_invoke(rbx, noreg, byte_no);
3008 // rbx,: index
3009 // rcx: receiver
3010 // rdx: flags
3012 invokevirtual_helper(rbx, rcx, rdx);
3013 }
3016 void TemplateTable::invokespecial(int byte_no) {
3017 transition(vtos, vtos);
3018 assert(byte_no == f1_byte, "use this argument");
3019 prepare_invoke(rbx, noreg, byte_no);
3020 // do the call
3021 __ verify_oop(rbx);
3022 __ profile_call(rax);
3023 __ jump_from_interpreted(rbx, rax);
3024 }
3027 void TemplateTable::invokestatic(int byte_no) {
3028 transition(vtos, vtos);
3029 assert(byte_no == f1_byte, "use this argument");
3030 prepare_invoke(rbx, noreg, byte_no);
3031 // do the call
3032 __ verify_oop(rbx);
3033 __ profile_call(rax);
3034 __ jump_from_interpreted(rbx, rax);
3035 }
3038 void TemplateTable::fast_invokevfinal(int byte_no) {
3039 transition(vtos, vtos);
3040 assert(byte_no == f2_byte, "use this argument");
3041 __ stop("fast_invokevfinal not used on x86");
3042 }
3045 void TemplateTable::invokeinterface(int byte_no) {
3046 transition(vtos, vtos);
3047 assert(byte_no == f1_byte, "use this argument");
3048 prepare_invoke(rax, rbx, byte_no);
3050 // rax,: Interface
3051 // rbx,: index
3052 // rcx: receiver
3053 // rdx: flags
3055 // Special case of invokeinterface called for virtual method of
3056 // java.lang.Object. See cpCacheOop.cpp for details.
3057 // This code isn't produced by javac, but could be produced by
3058 // another compliant java compiler.
3059 Label notMethod;
3060 __ movl(rdi, rdx);
3061 __ andl(rdi, (1 << ConstantPoolCacheEntry::methodInterface));
3062 __ jcc(Assembler::zero, notMethod);
3064 invokevirtual_helper(rbx, rcx, rdx);
3065 __ bind(notMethod);
3067 // Get receiver klass into rdx - also a null check
3068 __ restore_locals(); // restore rdi
3069 __ movptr(rdx, Address(rcx, oopDesc::klass_offset_in_bytes()));
3070 __ verify_oop(rdx);
3072 // profile this call
3073 __ profile_virtual_call(rdx, rsi, rdi);
3075 Label no_such_interface, no_such_method;
3077 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3078 rdx, rax, rbx,
3079 // outputs: method, scan temp. reg
3080 rbx, rsi,
3081 no_such_interface);
3083 // rbx,: methodOop to call
3084 // rcx: receiver
3085 // Check for abstract method error
3086 // Note: This should be done more efficiently via a throw_abstract_method_error
3087 // interpreter entry point and a conditional jump to it in case of a null
3088 // method.
3089 __ testptr(rbx, rbx);
3090 __ jcc(Assembler::zero, no_such_method);
3092 // do the call
3093 // rcx: receiver
3094 // rbx,: methodOop
3095 __ jump_from_interpreted(rbx, rdx);
3096 __ should_not_reach_here();
3098 // exception handling code follows...
3099 // note: must restore interpreter registers to canonical
3100 // state for exception handling to work correctly!
3102 __ bind(no_such_method);
3103 // throw exception
3104 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3105 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3106 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3107 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3108 // the call_VM checks for exception, so we should never return here.
3109 __ should_not_reach_here();
3111 __ bind(no_such_interface);
3112 // throw exception
3113 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3114 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3115 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3116 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3117 InterpreterRuntime::throw_IncompatibleClassChangeError));
3118 // the call_VM checks for exception, so we should never return here.
3119 __ should_not_reach_here();
3120 }
3122 void TemplateTable::invokedynamic(int byte_no) {
3123 transition(vtos, vtos);
3125 if (!EnableInvokeDynamic) {
3126 // We should not encounter this bytecode if !EnableInvokeDynamic.
3127 // The verifier will stop it. However, if we get past the verifier,
3128 // this will stop the thread in a reasonable way, without crashing the JVM.
3129 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3130 InterpreterRuntime::throw_IncompatibleClassChangeError));
3131 // the call_VM checks for exception, so we should never return here.
3132 __ should_not_reach_here();
3133 return;
3134 }
3136 assert(byte_no == f1_oop, "use this argument");
3137 prepare_invoke(rax, rbx, byte_no);
3139 // rax: CallSite object (f1)
3140 // rbx: unused (f2)
3141 // rcx: receiver address
3142 // rdx: flags (unused)
3144 Register rax_callsite = rax;
3145 Register rcx_method_handle = rcx;
3147 if (ProfileInterpreter) {
3148 // %%% should make a type profile for any invokedynamic that takes a ref argument
3149 // profile this call
3150 __ profile_call(rsi);
3151 }
3153 __ movptr(rcx_method_handle, Address(rax_callsite, __ delayed_value(java_dyn_CallSite::target_offset_in_bytes, rcx)));
3154 __ null_check(rcx_method_handle);
3155 __ prepare_to_jump_from_interpreted();
3156 __ jump_to_method_handle_entry(rcx_method_handle, rdx);
3157 }
3159 //----------------------------------------------------------------------------------------------------
3160 // Allocation
3162 void TemplateTable::_new() {
3163 transition(vtos, atos);
3164 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3165 Label slow_case;
3166 Label slow_case_no_pop;
3167 Label done;
3168 Label initialize_header;
3169 Label initialize_object; // including clearing the fields
3170 Label allocate_shared;
3172 __ get_cpool_and_tags(rcx, rax);
3174 // Make sure the class we're about to instantiate has been resolved.
3175 // This is done before loading instanceKlass to be consistent with the order
3176 // how Constant Pool is updated (see constantPoolOopDesc::klass_at_put)
3177 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
3178 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3179 __ jcc(Assembler::notEqual, slow_case_no_pop);
3181 // get instanceKlass
3182 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3183 __ push(rcx); // save the contexts of klass for initializing the header
3185 // make sure klass is initialized & doesn't have finalizer
3186 // make sure klass is fully initialized
3187 __ cmpl(Address(rcx, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)), instanceKlass::fully_initialized);
3188 __ jcc(Assembler::notEqual, slow_case);
3190 // get instance_size in instanceKlass (scaled to a count of bytes)
3191 __ movl(rdx, Address(rcx, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
3192 // test to see if it has a finalizer or is malformed in some way
3193 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3194 __ jcc(Assembler::notZero, slow_case);
3196 //
3197 // Allocate the instance
3198 // 1) Try to allocate in the TLAB
3199 // 2) if fail and the object is large allocate in the shared Eden
3200 // 3) if the above fails (or is not applicable), go to a slow case
3201 // (creates a new TLAB, etc.)
3203 const bool allow_shared_alloc =
3204 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3206 const Register thread = rcx;
3207 if (UseTLAB || allow_shared_alloc) {
3208 __ get_thread(thread);
3209 }
3211 if (UseTLAB) {
3212 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3213 __ lea(rbx, Address(rax, rdx, Address::times_1));
3214 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3215 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3216 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3217 if (ZeroTLAB) {
3218 // the fields have been already cleared
3219 __ jmp(initialize_header);
3220 } else {
3221 // initialize both the header and fields
3222 __ jmp(initialize_object);
3223 }
3224 }
3226 // Allocation in the shared Eden, if allowed.
3227 //
3228 // rdx: instance size in bytes
3229 if (allow_shared_alloc) {
3230 __ bind(allocate_shared);
3232 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3234 Label retry;
3235 __ bind(retry);
3236 __ movptr(rax, heap_top);
3237 __ lea(rbx, Address(rax, rdx, Address::times_1));
3238 __ cmpptr(rbx, ExternalAddress((address)Universe::heap()->end_addr()));
3239 __ jcc(Assembler::above, slow_case);
3241 // Compare rax, with the top addr, and if still equal, store the new
3242 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3243 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3244 //
3245 // rax,: object begin
3246 // rbx,: object end
3247 // rdx: instance size in bytes
3248 __ locked_cmpxchgptr(rbx, heap_top);
3250 // if someone beat us on the allocation, try again, otherwise continue
3251 __ jcc(Assembler::notEqual, retry);
3253 __ incr_allocated_bytes(thread, rdx, 0);
3254 }
3256 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3257 // The object is initialized before the header. If the object size is
3258 // zero, go directly to the header initialization.
3259 __ bind(initialize_object);
3260 __ decrement(rdx, sizeof(oopDesc));
3261 __ jcc(Assembler::zero, initialize_header);
3263 // Initialize topmost object field, divide rdx by 8, check if odd and
3264 // test if zero.
3265 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3266 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3268 // rdx must have been multiple of 8
3269 #ifdef ASSERT
3270 // make sure rdx was multiple of 8
3271 Label L;
3272 // Ignore partial flag stall after shrl() since it is debug VM
3273 __ jccb(Assembler::carryClear, L);
3274 __ stop("object size is not multiple of 2 - adjust this code");
3275 __ bind(L);
3276 // rdx must be > 0, no extra check needed here
3277 #endif
3279 // initialize remaining object fields: rdx was a multiple of 8
3280 { Label loop;
3281 __ bind(loop);
3282 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3283 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3284 __ decrement(rdx);
3285 __ jcc(Assembler::notZero, loop);
3286 }
3288 // initialize object header only.
3289 __ bind(initialize_header);
3290 if (UseBiasedLocking) {
3291 __ pop(rcx); // get saved klass back in the register.
3292 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
3293 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3294 } else {
3295 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3296 (int32_t)markOopDesc::prototype()); // header
3297 __ pop(rcx); // get saved klass back in the register.
3298 }
3299 __ movptr(Address(rax, oopDesc::klass_offset_in_bytes()), rcx); // klass
3301 {
3302 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3303 // Trigger dtrace event for fastpath
3304 __ push(atos);
3305 __ call_VM_leaf(
3306 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3307 __ pop(atos);
3308 }
3310 __ jmp(done);
3311 }
3313 // slow case
3314 __ bind(slow_case);
3315 __ pop(rcx); // restore stack pointer to what it was when we came in.
3316 __ bind(slow_case_no_pop);
3317 __ get_constant_pool(rax);
3318 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3319 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx);
3321 // continue
3322 __ bind(done);
3323 }
3326 void TemplateTable::newarray() {
3327 transition(itos, atos);
3328 __ push_i(rax); // make sure everything is on the stack
3329 __ load_unsigned_byte(rdx, at_bcp(1));
3330 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax);
3331 __ pop_i(rdx); // discard size
3332 }
3335 void TemplateTable::anewarray() {
3336 transition(itos, atos);
3337 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3338 __ get_constant_pool(rcx);
3339 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax);
3340 }
3343 void TemplateTable::arraylength() {
3344 transition(atos, itos);
3345 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3346 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3347 }
3350 void TemplateTable::checkcast() {
3351 transition(atos, atos);
3352 Label done, is_null, ok_is_subtype, quicked, resolved;
3353 __ testptr(rax, rax); // Object is in EAX
3354 __ jcc(Assembler::zero, is_null);
3356 // Get cpool & tags index
3357 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3358 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3359 // See if bytecode has already been quicked
3360 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3361 __ jcc(Assembler::equal, quicked);
3363 __ push(atos);
3364 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3365 __ pop_ptr(rdx);
3366 __ jmpb(resolved);
3368 // Get superklass in EAX and subklass in EBX
3369 __ bind(quicked);
3370 __ mov(rdx, rax); // Save object in EDX; EAX needed for subtype check
3371 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3373 __ bind(resolved);
3374 __ movptr(rbx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3376 // Generate subtype check. Blows ECX. Resets EDI. Object in EDX.
3377 // Superklass in EAX. Subklass in EBX.
3378 __ gen_subtype_check( rbx, ok_is_subtype );
3380 // Come here on failure
3381 __ push(rdx);
3382 // object is at TOS
3383 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3385 // Come here on success
3386 __ bind(ok_is_subtype);
3387 __ mov(rax,rdx); // Restore object in EDX
3389 // Collect counts on whether this check-cast sees NULLs a lot or not.
3390 if (ProfileInterpreter) {
3391 __ jmp(done);
3392 __ bind(is_null);
3393 __ profile_null_seen(rcx);
3394 } else {
3395 __ bind(is_null); // same as 'done'
3396 }
3397 __ bind(done);
3398 }
3401 void TemplateTable::instanceof() {
3402 transition(atos, itos);
3403 Label done, is_null, ok_is_subtype, quicked, resolved;
3404 __ testptr(rax, rax);
3405 __ jcc(Assembler::zero, is_null);
3407 // Get cpool & tags index
3408 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3409 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3410 // See if bytecode has already been quicked
3411 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3412 __ jcc(Assembler::equal, quicked);
3414 __ push(atos);
3415 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3416 __ pop_ptr(rdx);
3417 __ movptr(rdx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3418 __ jmp(resolved);
3420 // Get superklass in EAX and subklass in EDX
3421 __ bind(quicked);
3422 __ movptr(rdx, Address(rax, oopDesc::klass_offset_in_bytes()));
3423 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3425 __ bind(resolved);
3427 // Generate subtype check. Blows ECX. Resets EDI.
3428 // Superklass in EAX. Subklass in EDX.
3429 __ gen_subtype_check( rdx, ok_is_subtype );
3431 // Come here on failure
3432 __ xorl(rax,rax);
3433 __ jmpb(done);
3434 // Come here on success
3435 __ bind(ok_is_subtype);
3436 __ movl(rax, 1);
3438 // Collect counts on whether this test sees NULLs a lot or not.
3439 if (ProfileInterpreter) {
3440 __ jmp(done);
3441 __ bind(is_null);
3442 __ profile_null_seen(rcx);
3443 } else {
3444 __ bind(is_null); // same as 'done'
3445 }
3446 __ bind(done);
3447 // rax, = 0: obj == NULL or obj is not an instanceof the specified klass
3448 // rax, = 1: obj != NULL and obj is an instanceof the specified klass
3449 }
3452 //----------------------------------------------------------------------------------------------------
3453 // Breakpoints
3454 void TemplateTable::_breakpoint() {
3456 // Note: We get here even if we are single stepping..
3457 // jbug inists on setting breakpoints at every bytecode
3458 // even if we are in single step mode.
3460 transition(vtos, vtos);
3462 // get the unpatched byte code
3463 __ get_method(rcx);
3464 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi);
3465 __ mov(rbx, rax);
3467 // post the breakpoint event
3468 __ get_method(rcx);
3469 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi);
3471 // complete the execution of original bytecode
3472 __ dispatch_only_normal(vtos);
3473 }
3476 //----------------------------------------------------------------------------------------------------
3477 // Exceptions
3479 void TemplateTable::athrow() {
3480 transition(atos, vtos);
3481 __ null_check(rax);
3482 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3483 }
3486 //----------------------------------------------------------------------------------------------------
3487 // Synchronization
3488 //
3489 // Note: monitorenter & exit are symmetric routines; which is reflected
3490 // in the assembly code structure as well
3491 //
3492 // Stack layout:
3493 //
3494 // [expressions ] <--- rsp = expression stack top
3495 // ..
3496 // [expressions ]
3497 // [monitor entry] <--- monitor block top = expression stack bot
3498 // ..
3499 // [monitor entry]
3500 // [frame data ] <--- monitor block bot
3501 // ...
3502 // [saved rbp, ] <--- rbp,
3505 void TemplateTable::monitorenter() {
3506 transition(atos, vtos);
3508 // check for NULL object
3509 __ null_check(rax);
3511 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3512 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3513 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3514 Label allocated;
3516 // initialize entry pointer
3517 __ xorl(rdx, rdx); // points to free slot or NULL
3519 // find a free slot in the monitor block (result in rdx)
3520 { Label entry, loop, exit;
3521 __ movptr(rcx, monitor_block_top); // points to current entry, starting with top-most entry
3522 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3523 __ jmpb(entry);
3525 __ bind(loop);
3526 __ cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); // check if current entry is used
3528 // TODO - need new func here - kbt
3529 if (VM_Version::supports_cmov()) {
3530 __ cmov(Assembler::equal, rdx, rcx); // if not used then remember entry in rdx
3531 } else {
3532 Label L;
3533 __ jccb(Assembler::notEqual, L);
3534 __ mov(rdx, rcx); // if not used then remember entry in rdx
3535 __ bind(L);
3536 }
3537 __ cmpptr(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3538 __ jccb(Assembler::equal, exit); // if same object then stop searching
3539 __ addptr(rcx, entry_size); // otherwise advance to next entry
3540 __ bind(entry);
3541 __ cmpptr(rcx, rbx); // check if bottom reached
3542 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3543 __ bind(exit);
3544 }
3546 __ testptr(rdx, rdx); // check if a slot has been found
3547 __ jccb(Assembler::notZero, allocated); // if found, continue with that one
3549 // allocate one if there's no free slot
3550 { Label entry, loop;
3551 // 1. compute new pointers // rsp: old expression stack top
3552 __ movptr(rdx, monitor_block_bot); // rdx: old expression stack bottom
3553 __ subptr(rsp, entry_size); // move expression stack top
3554 __ subptr(rdx, entry_size); // move expression stack bottom
3555 __ mov(rcx, rsp); // set start value for copy loop
3556 __ movptr(monitor_block_bot, rdx); // set new monitor block top
3557 __ jmp(entry);
3558 // 2. move expression stack contents
3559 __ bind(loop);
3560 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location
3561 __ movptr(Address(rcx, 0), rbx); // and store it at new location
3562 __ addptr(rcx, wordSize); // advance to next word
3563 __ bind(entry);
3564 __ cmpptr(rcx, rdx); // check if bottom reached
3565 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word
3566 }
3568 // call run-time routine
3569 // rdx: points to monitor entry
3570 __ bind(allocated);
3572 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3573 // The object has already been poped from the stack, so the expression stack looks correct.
3574 __ increment(rsi);
3576 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
3577 __ lock_object(rdx);
3579 // check to make sure this monitor doesn't cause stack overflow after locking
3580 __ save_bcp(); // in case of exception
3581 __ generate_stack_overflow_check(0);
3583 // The bcp has already been incremented. Just need to dispatch to next instruction.
3584 __ dispatch_next(vtos);
3585 }
3588 void TemplateTable::monitorexit() {
3589 transition(atos, vtos);
3591 // check for NULL object
3592 __ null_check(rax);
3594 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3595 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3596 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3597 Label found;
3599 // find matching slot
3600 { Label entry, loop;
3601 __ movptr(rdx, monitor_block_top); // points to current entry, starting with top-most entry
3602 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3603 __ jmpb(entry);
3605 __ bind(loop);
3606 __ cmpptr(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3607 __ jcc(Assembler::equal, found); // if same object then stop searching
3608 __ addptr(rdx, entry_size); // otherwise advance to next entry
3609 __ bind(entry);
3610 __ cmpptr(rdx, rbx); // check if bottom reached
3611 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3612 }
3614 // error handling. Unlocking was not block-structured
3615 Label end;
3616 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3617 __ should_not_reach_here();
3619 // call run-time routine
3620 // rcx: points to monitor entry
3621 __ bind(found);
3622 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3623 __ unlock_object(rdx);
3624 __ pop_ptr(rax); // discard object
3625 __ bind(end);
3626 }
3629 //----------------------------------------------------------------------------------------------------
3630 // Wide instructions
3632 void TemplateTable::wide() {
3633 transition(vtos, vtos);
3634 __ load_unsigned_byte(rbx, at_bcp(1));
3635 ExternalAddress wtable((address)Interpreter::_wentry_point);
3636 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
3637 // Note: the rsi increment step is part of the individual wide bytecode implementations
3638 }
3641 //----------------------------------------------------------------------------------------------------
3642 // Multi arrays
3644 void TemplateTable::multianewarray() {
3645 transition(vtos, atos);
3646 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3647 // last dim is on top of stack; we want address of first one:
3648 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
3649 // the latter wordSize to point to the beginning of the array.
3650 __ lea( rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
3651 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax); // pass in rax,
3652 __ load_unsigned_byte(rbx, at_bcp(3));
3653 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
3654 }
3656 #endif /* !CC_INTERP */