Wed, 23 Jan 2013 13:02:39 -0500
8005915: Unify SERIALGC and INCLUDE_ALTERNATE_GCS
Summary: Rename INCLUDE_ALTERNATE_GCS to INCLUDE_ALL_GCS and replace SERIALGC with INCLUDE_ALL_GCS.
Reviewed-by: coleenp, stefank
1 /*
2 * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "interpreter/interpreterRuntime.hpp"
29 #include "interpreter/templateTable.hpp"
30 #include "memory/universe.inline.hpp"
31 #include "oops/methodData.hpp"
32 #include "oops/objArrayKlass.hpp"
33 #include "oops/oop.inline.hpp"
34 #include "prims/methodHandles.hpp"
35 #include "runtime/sharedRuntime.hpp"
36 #include "runtime/stubRoutines.hpp"
37 #include "runtime/synchronizer.hpp"
38 #include "utilities/macros.hpp"
40 #ifndef CC_INTERP
41 #define __ _masm->
43 //----------------------------------------------------------------------------------------------------
44 // Platform-dependent initialization
46 void TemplateTable::pd_initialize() {
47 // No i486 specific initialization
48 }
50 //----------------------------------------------------------------------------------------------------
51 // Address computation
53 // local variables
54 static inline Address iaddress(int n) {
55 return Address(rdi, Interpreter::local_offset_in_bytes(n));
56 }
58 static inline Address laddress(int n) { return iaddress(n + 1); }
59 static inline Address haddress(int n) { return iaddress(n + 0); }
60 static inline Address faddress(int n) { return iaddress(n); }
61 static inline Address daddress(int n) { return laddress(n); }
62 static inline Address aaddress(int n) { return iaddress(n); }
64 static inline Address iaddress(Register r) {
65 return Address(rdi, r, Interpreter::stackElementScale());
66 }
67 static inline Address laddress(Register r) {
68 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(1));
69 }
70 static inline Address haddress(Register r) {
71 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
72 }
74 static inline Address faddress(Register r) { return iaddress(r); }
75 static inline Address daddress(Register r) { return laddress(r); }
76 static inline Address aaddress(Register r) { return iaddress(r); }
78 // expression stack
79 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
80 // data beyond the rsp which is potentially unsafe in an MT environment;
81 // an interrupt may overwrite that data.)
82 static inline Address at_rsp () {
83 return Address(rsp, 0);
84 }
86 // At top of Java expression stack which may be different than rsp(). It
87 // isn't for category 1 objects.
88 static inline Address at_tos () {
89 Address tos = Address(rsp, Interpreter::expr_offset_in_bytes(0));
90 return tos;
91 }
93 static inline Address at_tos_p1() {
94 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
95 }
97 static inline Address at_tos_p2() {
98 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
99 }
101 // Condition conversion
102 static Assembler::Condition j_not(TemplateTable::Condition cc) {
103 switch (cc) {
104 case TemplateTable::equal : return Assembler::notEqual;
105 case TemplateTable::not_equal : return Assembler::equal;
106 case TemplateTable::less : return Assembler::greaterEqual;
107 case TemplateTable::less_equal : return Assembler::greater;
108 case TemplateTable::greater : return Assembler::lessEqual;
109 case TemplateTable::greater_equal: return Assembler::less;
110 }
111 ShouldNotReachHere();
112 return Assembler::zero;
113 }
116 //----------------------------------------------------------------------------------------------------
117 // Miscelaneous helper routines
119 // Store an oop (or NULL) at the address described by obj.
120 // If val == noreg this means store a NULL
122 static void do_oop_store(InterpreterMacroAssembler* _masm,
123 Address obj,
124 Register val,
125 BarrierSet::Name barrier,
126 bool precise) {
127 assert(val == noreg || val == rax, "parameter is just for looks");
128 switch (barrier) {
129 #if INCLUDE_ALL_GCS
130 case BarrierSet::G1SATBCT:
131 case BarrierSet::G1SATBCTLogging:
132 {
133 // flatten object address if needed
134 // We do it regardless of precise because we need the registers
135 if (obj.index() == noreg && obj.disp() == 0) {
136 if (obj.base() != rdx) {
137 __ movl(rdx, obj.base());
138 }
139 } else {
140 __ leal(rdx, obj);
141 }
142 __ get_thread(rcx);
143 __ save_bcp();
144 __ g1_write_barrier_pre(rdx /* obj */,
145 rbx /* pre_val */,
146 rcx /* thread */,
147 rsi /* tmp */,
148 val != noreg /* tosca_live */,
149 false /* expand_call */);
151 // Do the actual store
152 // noreg means NULL
153 if (val == noreg) {
154 __ movptr(Address(rdx, 0), NULL_WORD);
155 // No post barrier for NULL
156 } else {
157 __ movl(Address(rdx, 0), val);
158 __ g1_write_barrier_post(rdx /* store_adr */,
159 val /* new_val */,
160 rcx /* thread */,
161 rbx /* tmp */,
162 rsi /* tmp2 */);
163 }
164 __ restore_bcp();
166 }
167 break;
168 #endif // INCLUDE_ALL_GCS
169 case BarrierSet::CardTableModRef:
170 case BarrierSet::CardTableExtension:
171 {
172 if (val == noreg) {
173 __ movptr(obj, NULL_WORD);
174 } else {
175 __ movl(obj, val);
176 // flatten object address if needed
177 if (!precise || (obj.index() == noreg && obj.disp() == 0)) {
178 __ store_check(obj.base());
179 } else {
180 __ leal(rdx, obj);
181 __ store_check(rdx);
182 }
183 }
184 }
185 break;
186 case BarrierSet::ModRef:
187 case BarrierSet::Other:
188 if (val == noreg) {
189 __ movptr(obj, NULL_WORD);
190 } else {
191 __ movl(obj, val);
192 }
193 break;
194 default :
195 ShouldNotReachHere();
197 }
198 }
200 Address TemplateTable::at_bcp(int offset) {
201 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
202 return Address(rsi, offset);
203 }
206 void TemplateTable::patch_bytecode(Bytecodes::Code bc, Register bc_reg,
207 Register temp_reg, bool load_bc_into_bc_reg/*=true*/,
208 int byte_no) {
209 if (!RewriteBytecodes) return;
210 Label L_patch_done;
212 switch (bc) {
213 case Bytecodes::_fast_aputfield:
214 case Bytecodes::_fast_bputfield:
215 case Bytecodes::_fast_cputfield:
216 case Bytecodes::_fast_dputfield:
217 case Bytecodes::_fast_fputfield:
218 case Bytecodes::_fast_iputfield:
219 case Bytecodes::_fast_lputfield:
220 case Bytecodes::_fast_sputfield:
221 {
222 // We skip bytecode quickening for putfield instructions when
223 // the put_code written to the constant pool cache is zero.
224 // This is required so that every execution of this instruction
225 // calls out to InterpreterRuntime::resolve_get_put to do
226 // additional, required work.
227 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
228 assert(load_bc_into_bc_reg, "we use bc_reg as temp");
229 __ get_cache_and_index_and_bytecode_at_bcp(bc_reg, temp_reg, temp_reg, byte_no, 1);
230 __ movl(bc_reg, bc);
231 __ cmpl(temp_reg, (int) 0);
232 __ jcc(Assembler::zero, L_patch_done); // don't patch
233 }
234 break;
235 default:
236 assert(byte_no == -1, "sanity");
237 // the pair bytecodes have already done the load.
238 if (load_bc_into_bc_reg) {
239 __ movl(bc_reg, bc);
240 }
241 }
243 if (JvmtiExport::can_post_breakpoint()) {
244 Label L_fast_patch;
245 // if a breakpoint is present we can't rewrite the stream directly
246 __ movzbl(temp_reg, at_bcp(0));
247 __ cmpl(temp_reg, Bytecodes::_breakpoint);
248 __ jcc(Assembler::notEqual, L_fast_patch);
249 __ get_method(temp_reg);
250 // Let breakpoint table handling rewrite to quicker bytecode
251 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), temp_reg, rsi, bc_reg);
252 #ifndef ASSERT
253 __ jmpb(L_patch_done);
254 #else
255 __ jmp(L_patch_done);
256 #endif
257 __ bind(L_fast_patch);
258 }
260 #ifdef ASSERT
261 Label L_okay;
262 __ load_unsigned_byte(temp_reg, at_bcp(0));
263 __ cmpl(temp_reg, (int)Bytecodes::java_code(bc));
264 __ jccb(Assembler::equal, L_okay);
265 __ cmpl(temp_reg, bc_reg);
266 __ jcc(Assembler::equal, L_okay);
267 __ stop("patching the wrong bytecode");
268 __ bind(L_okay);
269 #endif
271 // patch bytecode
272 __ movb(at_bcp(0), bc_reg);
273 __ bind(L_patch_done);
274 }
276 //----------------------------------------------------------------------------------------------------
277 // Individual instructions
279 void TemplateTable::nop() {
280 transition(vtos, vtos);
281 // nothing to do
282 }
284 void TemplateTable::shouldnotreachhere() {
285 transition(vtos, vtos);
286 __ stop("shouldnotreachhere bytecode");
287 }
291 void TemplateTable::aconst_null() {
292 transition(vtos, atos);
293 __ xorptr(rax, rax);
294 }
297 void TemplateTable::iconst(int value) {
298 transition(vtos, itos);
299 if (value == 0) {
300 __ xorptr(rax, rax);
301 } else {
302 __ movptr(rax, value);
303 }
304 }
307 void TemplateTable::lconst(int value) {
308 transition(vtos, ltos);
309 if (value == 0) {
310 __ xorptr(rax, rax);
311 } else {
312 __ movptr(rax, value);
313 }
314 assert(value >= 0, "check this code");
315 __ xorptr(rdx, rdx);
316 }
319 void TemplateTable::fconst(int value) {
320 transition(vtos, ftos);
321 if (value == 0) { __ fldz();
322 } else if (value == 1) { __ fld1();
323 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
324 } else { ShouldNotReachHere();
325 }
326 }
329 void TemplateTable::dconst(int value) {
330 transition(vtos, dtos);
331 if (value == 0) { __ fldz();
332 } else if (value == 1) { __ fld1();
333 } else { ShouldNotReachHere();
334 }
335 }
338 void TemplateTable::bipush() {
339 transition(vtos, itos);
340 __ load_signed_byte(rax, at_bcp(1));
341 }
344 void TemplateTable::sipush() {
345 transition(vtos, itos);
346 __ load_unsigned_short(rax, at_bcp(1));
347 __ bswapl(rax);
348 __ sarl(rax, 16);
349 }
351 void TemplateTable::ldc(bool wide) {
352 transition(vtos, vtos);
353 Label call_ldc, notFloat, notClass, Done;
355 if (wide) {
356 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
357 } else {
358 __ load_unsigned_byte(rbx, at_bcp(1));
359 }
360 __ get_cpool_and_tags(rcx, rax);
361 const int base_offset = ConstantPool::header_size() * wordSize;
362 const int tags_offset = Array<u1>::base_offset_in_bytes();
364 // get type
365 __ xorptr(rdx, rdx);
366 __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset));
368 // unresolved class - get the resolved class
369 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
370 __ jccb(Assembler::equal, call_ldc);
372 // unresolved class in error (resolution failed) - call into runtime
373 // so that the same error from first resolution attempt is thrown.
374 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
375 __ jccb(Assembler::equal, call_ldc);
377 // resolved class - need to call vm to get java mirror of the class
378 __ cmpl(rdx, JVM_CONSTANT_Class);
379 __ jcc(Assembler::notEqual, notClass);
381 __ bind(call_ldc);
382 __ movl(rcx, wide);
383 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx);
384 __ push(atos);
385 __ jmp(Done);
387 __ bind(notClass);
388 __ cmpl(rdx, JVM_CONSTANT_Float);
389 __ jccb(Assembler::notEqual, notFloat);
390 // ftos
391 __ fld_s( Address(rcx, rbx, Address::times_ptr, base_offset));
392 __ push(ftos);
393 __ jmp(Done);
395 __ bind(notFloat);
396 #ifdef ASSERT
397 { Label L;
398 __ cmpl(rdx, JVM_CONSTANT_Integer);
399 __ jcc(Assembler::equal, L);
400 // String and Object are rewritten to fast_aldc
401 __ stop("unexpected tag type in ldc");
402 __ bind(L);
403 }
404 #endif
405 // itos JVM_CONSTANT_Integer only
406 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
407 __ push(itos);
408 __ bind(Done);
409 }
411 // Fast path for caching oop constants.
412 void TemplateTable::fast_aldc(bool wide) {
413 transition(vtos, atos);
415 Register result = rax;
416 Register tmp = rdx;
417 int index_size = wide ? sizeof(u2) : sizeof(u1);
419 Label resolved;
421 // We are resolved if the resolved reference cache entry contains a
422 // non-null object (String, MethodType, etc.)
423 assert_different_registers(result, tmp);
424 __ get_cache_index_at_bcp(tmp, 1, index_size);
425 __ load_resolved_reference_at_index(result, tmp);
426 __ testl(result, result);
427 __ jcc(Assembler::notZero, resolved);
429 address entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_ldc);
431 // first time invocation - must resolve first
432 __ movl(tmp, (int)bytecode());
433 __ call_VM(result, entry, tmp);
435 __ bind(resolved);
437 if (VerifyOops) {
438 __ verify_oop(result);
439 }
440 }
442 void TemplateTable::ldc2_w() {
443 transition(vtos, vtos);
444 Label Long, Done;
445 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
447 __ get_cpool_and_tags(rcx, rax);
448 const int base_offset = ConstantPool::header_size() * wordSize;
449 const int tags_offset = Array<u1>::base_offset_in_bytes();
451 // get type
452 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double);
453 __ jccb(Assembler::notEqual, Long);
454 // dtos
455 __ fld_d( Address(rcx, rbx, Address::times_ptr, base_offset));
456 __ push(dtos);
457 __ jmpb(Done);
459 __ bind(Long);
460 // ltos
461 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
462 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
464 __ push(ltos);
466 __ bind(Done);
467 }
470 void TemplateTable::locals_index(Register reg, int offset) {
471 __ load_unsigned_byte(reg, at_bcp(offset));
472 __ negptr(reg);
473 }
476 void TemplateTable::iload() {
477 transition(vtos, itos);
478 if (RewriteFrequentPairs) {
479 Label rewrite, done;
481 // get next byte
482 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
483 // if _iload, wait to rewrite to iload2. We only want to rewrite the
484 // last two iloads in a pair. Comparing against fast_iload means that
485 // the next bytecode is neither an iload or a caload, and therefore
486 // an iload pair.
487 __ cmpl(rbx, Bytecodes::_iload);
488 __ jcc(Assembler::equal, done);
490 __ cmpl(rbx, Bytecodes::_fast_iload);
491 __ movl(rcx, Bytecodes::_fast_iload2);
492 __ jccb(Assembler::equal, rewrite);
494 // if _caload, rewrite to fast_icaload
495 __ cmpl(rbx, Bytecodes::_caload);
496 __ movl(rcx, Bytecodes::_fast_icaload);
497 __ jccb(Assembler::equal, rewrite);
499 // rewrite so iload doesn't check again.
500 __ movl(rcx, Bytecodes::_fast_iload);
502 // rewrite
503 // rcx: fast bytecode
504 __ bind(rewrite);
505 patch_bytecode(Bytecodes::_iload, rcx, rbx, false);
506 __ bind(done);
507 }
509 // Get the local value into tos
510 locals_index(rbx);
511 __ movl(rax, iaddress(rbx));
512 }
515 void TemplateTable::fast_iload2() {
516 transition(vtos, itos);
517 locals_index(rbx);
518 __ movl(rax, iaddress(rbx));
519 __ push(itos);
520 locals_index(rbx, 3);
521 __ movl(rax, iaddress(rbx));
522 }
524 void TemplateTable::fast_iload() {
525 transition(vtos, itos);
526 locals_index(rbx);
527 __ movl(rax, iaddress(rbx));
528 }
531 void TemplateTable::lload() {
532 transition(vtos, ltos);
533 locals_index(rbx);
534 __ movptr(rax, laddress(rbx));
535 NOT_LP64(__ movl(rdx, haddress(rbx)));
536 }
539 void TemplateTable::fload() {
540 transition(vtos, ftos);
541 locals_index(rbx);
542 __ fld_s(faddress(rbx));
543 }
546 void TemplateTable::dload() {
547 transition(vtos, dtos);
548 locals_index(rbx);
549 __ fld_d(daddress(rbx));
550 }
553 void TemplateTable::aload() {
554 transition(vtos, atos);
555 locals_index(rbx);
556 __ movptr(rax, aaddress(rbx));
557 }
560 void TemplateTable::locals_index_wide(Register reg) {
561 __ movl(reg, at_bcp(2));
562 __ bswapl(reg);
563 __ shrl(reg, 16);
564 __ negptr(reg);
565 }
568 void TemplateTable::wide_iload() {
569 transition(vtos, itos);
570 locals_index_wide(rbx);
571 __ movl(rax, iaddress(rbx));
572 }
575 void TemplateTable::wide_lload() {
576 transition(vtos, ltos);
577 locals_index_wide(rbx);
578 __ movptr(rax, laddress(rbx));
579 NOT_LP64(__ movl(rdx, haddress(rbx)));
580 }
583 void TemplateTable::wide_fload() {
584 transition(vtos, ftos);
585 locals_index_wide(rbx);
586 __ fld_s(faddress(rbx));
587 }
590 void TemplateTable::wide_dload() {
591 transition(vtos, dtos);
592 locals_index_wide(rbx);
593 __ fld_d(daddress(rbx));
594 }
597 void TemplateTable::wide_aload() {
598 transition(vtos, atos);
599 locals_index_wide(rbx);
600 __ movptr(rax, aaddress(rbx));
601 }
603 void TemplateTable::index_check(Register array, Register index) {
604 // Pop ptr into array
605 __ pop_ptr(array);
606 index_check_without_pop(array, index);
607 }
609 void TemplateTable::index_check_without_pop(Register array, Register index) {
610 // destroys rbx,
611 // check array
612 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
613 LP64_ONLY(__ movslq(index, index));
614 // check index
615 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
616 if (index != rbx) {
617 // ??? convention: move aberrant index into rbx, for exception message
618 assert(rbx != array, "different registers");
619 __ mov(rbx, index);
620 }
621 __ jump_cc(Assembler::aboveEqual,
622 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
623 }
626 void TemplateTable::iaload() {
627 transition(itos, itos);
628 // rdx: array
629 index_check(rdx, rax); // kills rbx,
630 // rax,: index
631 __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)));
632 }
635 void TemplateTable::laload() {
636 transition(itos, ltos);
637 // rax,: index
638 // rdx: array
639 index_check(rdx, rax);
640 __ mov(rbx, rax);
641 // rbx,: index
642 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
643 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
644 }
647 void TemplateTable::faload() {
648 transition(itos, ftos);
649 // rdx: array
650 index_check(rdx, rax); // kills rbx,
651 // rax,: index
652 __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
653 }
656 void TemplateTable::daload() {
657 transition(itos, dtos);
658 // rdx: array
659 index_check(rdx, rax); // kills rbx,
660 // rax,: index
661 __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
662 }
665 void TemplateTable::aaload() {
666 transition(itos, atos);
667 // rdx: array
668 index_check(rdx, rax); // kills rbx,
669 // rax,: index
670 __ movptr(rax, Address(rdx, rax, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
671 }
674 void TemplateTable::baload() {
675 transition(itos, itos);
676 // rdx: array
677 index_check(rdx, rax); // kills rbx,
678 // rax,: index
679 // can do better code for P5 - fix this at some point
680 __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
681 __ mov(rax, rbx);
682 }
685 void TemplateTable::caload() {
686 transition(itos, itos);
687 // rdx: array
688 index_check(rdx, rax); // kills rbx,
689 // rax,: index
690 // can do better code for P5 - may want to improve this at some point
691 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
692 __ mov(rax, rbx);
693 }
695 // iload followed by caload frequent pair
696 void TemplateTable::fast_icaload() {
697 transition(vtos, itos);
698 // load index out of locals
699 locals_index(rbx);
700 __ movl(rax, iaddress(rbx));
702 // rdx: array
703 index_check(rdx, rax);
704 // rax,: index
705 __ load_unsigned_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
706 __ mov(rax, rbx);
707 }
709 void TemplateTable::saload() {
710 transition(itos, itos);
711 // rdx: array
712 index_check(rdx, rax); // kills rbx,
713 // rax,: index
714 // can do better code for P5 - may want to improve this at some point
715 __ load_signed_short(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
716 __ mov(rax, rbx);
717 }
720 void TemplateTable::iload(int n) {
721 transition(vtos, itos);
722 __ movl(rax, iaddress(n));
723 }
726 void TemplateTable::lload(int n) {
727 transition(vtos, ltos);
728 __ movptr(rax, laddress(n));
729 NOT_LP64(__ movptr(rdx, haddress(n)));
730 }
733 void TemplateTable::fload(int n) {
734 transition(vtos, ftos);
735 __ fld_s(faddress(n));
736 }
739 void TemplateTable::dload(int n) {
740 transition(vtos, dtos);
741 __ fld_d(daddress(n));
742 }
745 void TemplateTable::aload(int n) {
746 transition(vtos, atos);
747 __ movptr(rax, aaddress(n));
748 }
751 void TemplateTable::aload_0() {
752 transition(vtos, atos);
753 // According to bytecode histograms, the pairs:
754 //
755 // _aload_0, _fast_igetfield
756 // _aload_0, _fast_agetfield
757 // _aload_0, _fast_fgetfield
758 //
759 // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
760 // bytecode checks if the next bytecode is either _fast_igetfield,
761 // _fast_agetfield or _fast_fgetfield and then rewrites the
762 // current bytecode into a pair bytecode; otherwise it rewrites the current
763 // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
764 //
765 // Note: If the next bytecode is _getfield, the rewrite must be delayed,
766 // otherwise we may miss an opportunity for a pair.
767 //
768 // Also rewrite frequent pairs
769 // aload_0, aload_1
770 // aload_0, iload_1
771 // These bytecodes with a small amount of code are most profitable to rewrite
772 if (RewriteFrequentPairs) {
773 Label rewrite, done;
774 // get next byte
775 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
777 // do actual aload_0
778 aload(0);
780 // if _getfield then wait with rewrite
781 __ cmpl(rbx, Bytecodes::_getfield);
782 __ jcc(Assembler::equal, done);
784 // if _igetfield then reqrite to _fast_iaccess_0
785 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
786 __ cmpl(rbx, Bytecodes::_fast_igetfield);
787 __ movl(rcx, Bytecodes::_fast_iaccess_0);
788 __ jccb(Assembler::equal, rewrite);
790 // if _agetfield then reqrite to _fast_aaccess_0
791 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
792 __ cmpl(rbx, Bytecodes::_fast_agetfield);
793 __ movl(rcx, Bytecodes::_fast_aaccess_0);
794 __ jccb(Assembler::equal, rewrite);
796 // if _fgetfield then reqrite to _fast_faccess_0
797 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
798 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
799 __ movl(rcx, Bytecodes::_fast_faccess_0);
800 __ jccb(Assembler::equal, rewrite);
802 // else rewrite to _fast_aload0
803 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
804 __ movl(rcx, Bytecodes::_fast_aload_0);
806 // rewrite
807 // rcx: fast bytecode
808 __ bind(rewrite);
809 patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false);
811 __ bind(done);
812 } else {
813 aload(0);
814 }
815 }
817 void TemplateTable::istore() {
818 transition(itos, vtos);
819 locals_index(rbx);
820 __ movl(iaddress(rbx), rax);
821 }
824 void TemplateTable::lstore() {
825 transition(ltos, vtos);
826 locals_index(rbx);
827 __ movptr(laddress(rbx), rax);
828 NOT_LP64(__ movptr(haddress(rbx), rdx));
829 }
832 void TemplateTable::fstore() {
833 transition(ftos, vtos);
834 locals_index(rbx);
835 __ fstp_s(faddress(rbx));
836 }
839 void TemplateTable::dstore() {
840 transition(dtos, vtos);
841 locals_index(rbx);
842 __ fstp_d(daddress(rbx));
843 }
846 void TemplateTable::astore() {
847 transition(vtos, vtos);
848 __ pop_ptr(rax);
849 locals_index(rbx);
850 __ movptr(aaddress(rbx), rax);
851 }
854 void TemplateTable::wide_istore() {
855 transition(vtos, vtos);
856 __ pop_i(rax);
857 locals_index_wide(rbx);
858 __ movl(iaddress(rbx), rax);
859 }
862 void TemplateTable::wide_lstore() {
863 transition(vtos, vtos);
864 __ pop_l(rax, rdx);
865 locals_index_wide(rbx);
866 __ movptr(laddress(rbx), rax);
867 NOT_LP64(__ movl(haddress(rbx), rdx));
868 }
871 void TemplateTable::wide_fstore() {
872 wide_istore();
873 }
876 void TemplateTable::wide_dstore() {
877 wide_lstore();
878 }
881 void TemplateTable::wide_astore() {
882 transition(vtos, vtos);
883 __ pop_ptr(rax);
884 locals_index_wide(rbx);
885 __ movptr(aaddress(rbx), rax);
886 }
889 void TemplateTable::iastore() {
890 transition(itos, vtos);
891 __ pop_i(rbx);
892 // rax,: value
893 // rdx: array
894 index_check(rdx, rbx); // prefer index in rbx,
895 // rbx,: index
896 __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax);
897 }
900 void TemplateTable::lastore() {
901 transition(ltos, vtos);
902 __ pop_i(rbx);
903 // rax,: low(value)
904 // rcx: array
905 // rdx: high(value)
906 index_check(rcx, rbx); // prefer index in rbx,
907 // rbx,: index
908 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
909 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
910 }
913 void TemplateTable::fastore() {
914 transition(ftos, 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_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
921 }
924 void TemplateTable::dastore() {
925 transition(dtos, vtos);
926 __ pop_i(rbx);
927 // rdx: array
928 // st0: value
929 index_check(rdx, rbx); // prefer index in rbx,
930 // rbx,: index
931 __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
932 }
935 void TemplateTable::aastore() {
936 Label is_null, ok_is_subtype, done;
937 transition(vtos, vtos);
938 // stack: ..., array, index, value
939 __ movptr(rax, at_tos()); // Value
940 __ movl(rcx, at_tos_p1()); // Index
941 __ movptr(rdx, at_tos_p2()); // Array
943 Address element_address(rdx, rcx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_OBJECT));
944 index_check_without_pop(rdx, rcx); // kills rbx,
945 // do array store check - check for NULL value first
946 __ testptr(rax, rax);
947 __ jcc(Assembler::zero, is_null);
949 // Move subklass into EBX
950 __ load_klass(rbx, rax);
951 // Move superklass into EAX
952 __ load_klass(rax, rdx);
953 __ movptr(rax, Address(rax, ObjArrayKlass::element_klass_offset()));
954 // Compress array+index*wordSize+12 into a single register. Frees ECX.
955 __ lea(rdx, element_address);
957 // Generate subtype check. Blows ECX. Resets EDI to locals.
958 // Superklass in EAX. Subklass in EBX.
959 __ gen_subtype_check( rbx, ok_is_subtype );
961 // Come here on failure
962 // object is at TOS
963 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
965 // Come here on success
966 __ bind(ok_is_subtype);
968 // Get the value to store
969 __ movptr(rax, at_rsp());
970 // and store it with appropriate barrier
971 do_oop_store(_masm, Address(rdx, 0), rax, _bs->kind(), true);
973 __ jmp(done);
975 // Have a NULL in EAX, EDX=array, ECX=index. Store NULL at ary[idx]
976 __ bind(is_null);
977 __ profile_null_seen(rbx);
979 // Store NULL, (noreg means NULL to do_oop_store)
980 do_oop_store(_masm, element_address, noreg, _bs->kind(), true);
982 // Pop stack arguments
983 __ bind(done);
984 __ addptr(rsp, 3 * Interpreter::stackElementSize);
985 }
988 void TemplateTable::bastore() {
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 __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax);
996 }
999 void TemplateTable::castore() {
1000 transition(itos, vtos);
1001 __ pop_i(rbx);
1002 // rax,: value
1003 // rdx: array
1004 index_check(rdx, rbx); // prefer index in rbx,
1005 // rbx,: index
1006 __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax);
1007 }
1010 void TemplateTable::sastore() {
1011 castore();
1012 }
1015 void TemplateTable::istore(int n) {
1016 transition(itos, vtos);
1017 __ movl(iaddress(n), rax);
1018 }
1021 void TemplateTable::lstore(int n) {
1022 transition(ltos, vtos);
1023 __ movptr(laddress(n), rax);
1024 NOT_LP64(__ movptr(haddress(n), rdx));
1025 }
1028 void TemplateTable::fstore(int n) {
1029 transition(ftos, vtos);
1030 __ fstp_s(faddress(n));
1031 }
1034 void TemplateTable::dstore(int n) {
1035 transition(dtos, vtos);
1036 __ fstp_d(daddress(n));
1037 }
1040 void TemplateTable::astore(int n) {
1041 transition(vtos, vtos);
1042 __ pop_ptr(rax);
1043 __ movptr(aaddress(n), rax);
1044 }
1047 void TemplateTable::pop() {
1048 transition(vtos, vtos);
1049 __ addptr(rsp, Interpreter::stackElementSize);
1050 }
1053 void TemplateTable::pop2() {
1054 transition(vtos, vtos);
1055 __ addptr(rsp, 2*Interpreter::stackElementSize);
1056 }
1059 void TemplateTable::dup() {
1060 transition(vtos, vtos);
1061 // stack: ..., a
1062 __ load_ptr(0, rax);
1063 __ push_ptr(rax);
1064 // stack: ..., a, a
1065 }
1068 void TemplateTable::dup_x1() {
1069 transition(vtos, vtos);
1070 // stack: ..., a, b
1071 __ load_ptr( 0, rax); // load b
1072 __ load_ptr( 1, rcx); // load a
1073 __ store_ptr(1, rax); // store b
1074 __ store_ptr(0, rcx); // store a
1075 __ push_ptr(rax); // push b
1076 // stack: ..., b, a, b
1077 }
1080 void TemplateTable::dup_x2() {
1081 transition(vtos, vtos);
1082 // stack: ..., a, b, c
1083 __ load_ptr( 0, rax); // load c
1084 __ load_ptr( 2, rcx); // load a
1085 __ store_ptr(2, rax); // store c in a
1086 __ push_ptr(rax); // push c
1087 // stack: ..., c, b, c, c
1088 __ load_ptr( 2, rax); // load b
1089 __ store_ptr(2, rcx); // store a in b
1090 // stack: ..., c, a, c, c
1091 __ store_ptr(1, rax); // store b in c
1092 // stack: ..., c, a, b, c
1093 }
1096 void TemplateTable::dup2() {
1097 transition(vtos, vtos);
1098 // stack: ..., a, b
1099 __ load_ptr(1, rax); // load a
1100 __ push_ptr(rax); // push a
1101 __ load_ptr(1, rax); // load b
1102 __ push_ptr(rax); // push b
1103 // stack: ..., a, b, a, b
1104 }
1107 void TemplateTable::dup2_x1() {
1108 transition(vtos, vtos);
1109 // stack: ..., a, b, c
1110 __ load_ptr( 0, rcx); // load c
1111 __ load_ptr( 1, rax); // load b
1112 __ push_ptr(rax); // push b
1113 __ push_ptr(rcx); // push c
1114 // stack: ..., a, b, c, b, c
1115 __ store_ptr(3, rcx); // store c in b
1116 // stack: ..., a, c, c, b, c
1117 __ load_ptr( 4, rcx); // load a
1118 __ store_ptr(2, rcx); // store a in 2nd c
1119 // stack: ..., a, c, a, b, c
1120 __ store_ptr(4, rax); // store b in a
1121 // stack: ..., b, c, a, b, c
1122 // stack: ..., b, c, a, b, c
1123 }
1126 void TemplateTable::dup2_x2() {
1127 transition(vtos, vtos);
1128 // stack: ..., a, b, c, d
1129 __ load_ptr( 0, rcx); // load d
1130 __ load_ptr( 1, rax); // load c
1131 __ push_ptr(rax); // push c
1132 __ push_ptr(rcx); // push d
1133 // stack: ..., a, b, c, d, c, d
1134 __ load_ptr( 4, rax); // load b
1135 __ store_ptr(2, rax); // store b in d
1136 __ store_ptr(4, rcx); // store d in b
1137 // stack: ..., a, d, c, b, c, d
1138 __ load_ptr( 5, rcx); // load a
1139 __ load_ptr( 3, rax); // load c
1140 __ store_ptr(3, rcx); // store a in c
1141 __ store_ptr(5, rax); // store c in a
1142 // stack: ..., c, d, a, b, c, d
1143 // stack: ..., c, d, a, b, c, d
1144 }
1147 void TemplateTable::swap() {
1148 transition(vtos, vtos);
1149 // stack: ..., a, b
1150 __ load_ptr( 1, rcx); // load a
1151 __ load_ptr( 0, rax); // load b
1152 __ store_ptr(0, rcx); // store a in b
1153 __ store_ptr(1, rax); // store b in a
1154 // stack: ..., b, a
1155 }
1158 void TemplateTable::iop2(Operation op) {
1159 transition(itos, itos);
1160 switch (op) {
1161 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1162 case sub : __ mov(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1163 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1164 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1165 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1166 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1167 case shl : __ mov(rcx, rax); __ pop_i(rax); __ shll (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1168 case shr : __ mov(rcx, rax); __ pop_i(rax); __ sarl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1169 case ushr : __ mov(rcx, rax); __ pop_i(rax); __ shrl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1170 default : ShouldNotReachHere();
1171 }
1172 }
1175 void TemplateTable::lop2(Operation op) {
1176 transition(ltos, ltos);
1177 __ pop_l(rbx, rcx);
1178 switch (op) {
1179 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1180 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1181 __ mov (rax, rbx); __ mov (rdx, rcx); break;
1182 case _and : __ andl(rax, rbx); __ andl(rdx, rcx); break;
1183 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1184 case _xor : __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1185 default : ShouldNotReachHere();
1186 }
1187 }
1190 void TemplateTable::idiv() {
1191 transition(itos, itos);
1192 __ mov(rcx, rax);
1193 __ pop_i(rax);
1194 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1195 // they are not equal, one could do a normal division (no correction
1196 // needed), which may speed up this implementation for the common case.
1197 // (see also JVM spec., p.243 & p.271)
1198 __ corrected_idivl(rcx);
1199 }
1202 void TemplateTable::irem() {
1203 transition(itos, itos);
1204 __ mov(rcx, rax);
1205 __ pop_i(rax);
1206 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1207 // they are not equal, one could do a normal division (no correction
1208 // needed), which may speed up this implementation for the common case.
1209 // (see also JVM spec., p.243 & p.271)
1210 __ corrected_idivl(rcx);
1211 __ mov(rax, rdx);
1212 }
1215 void TemplateTable::lmul() {
1216 transition(ltos, ltos);
1217 __ pop_l(rbx, rcx);
1218 __ push(rcx); __ push(rbx);
1219 __ push(rdx); __ push(rax);
1220 __ lmul(2 * wordSize, 0);
1221 __ addptr(rsp, 4 * wordSize); // take off temporaries
1222 }
1225 void TemplateTable::ldiv() {
1226 transition(ltos, ltos);
1227 __ pop_l(rbx, rcx);
1228 __ push(rcx); __ push(rbx);
1229 __ push(rdx); __ push(rax);
1230 // check if y = 0
1231 __ orl(rax, rdx);
1232 __ jump_cc(Assembler::zero,
1233 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1234 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1235 __ addptr(rsp, 4 * wordSize); // take off temporaries
1236 }
1239 void TemplateTable::lrem() {
1240 transition(ltos, ltos);
1241 __ pop_l(rbx, rcx);
1242 __ push(rcx); __ push(rbx);
1243 __ push(rdx); __ push(rax);
1244 // check if y = 0
1245 __ orl(rax, rdx);
1246 __ jump_cc(Assembler::zero,
1247 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1248 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1249 __ addptr(rsp, 4 * wordSize);
1250 }
1253 void TemplateTable::lshl() {
1254 transition(itos, ltos);
1255 __ movl(rcx, rax); // get shift count
1256 __ pop_l(rax, rdx); // get shift value
1257 __ lshl(rdx, rax);
1258 }
1261 void TemplateTable::lshr() {
1262 transition(itos, ltos);
1263 __ mov(rcx, rax); // get shift count
1264 __ pop_l(rax, rdx); // get shift value
1265 __ lshr(rdx, rax, true);
1266 }
1269 void TemplateTable::lushr() {
1270 transition(itos, ltos);
1271 __ mov(rcx, rax); // get shift count
1272 __ pop_l(rax, rdx); // get shift value
1273 __ lshr(rdx, rax);
1274 }
1277 void TemplateTable::fop2(Operation op) {
1278 transition(ftos, ftos);
1279 switch (op) {
1280 case add: __ fadd_s (at_rsp()); break;
1281 case sub: __ fsubr_s(at_rsp()); break;
1282 case mul: __ fmul_s (at_rsp()); break;
1283 case div: __ fdivr_s(at_rsp()); break;
1284 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1285 default : ShouldNotReachHere();
1286 }
1287 __ f2ieee();
1288 __ pop(rax); // pop float thing off
1289 }
1292 void TemplateTable::dop2(Operation op) {
1293 transition(dtos, dtos);
1295 switch (op) {
1296 case add: __ fadd_d (at_rsp()); break;
1297 case sub: __ fsubr_d(at_rsp()); break;
1298 case mul: {
1299 Label L_strict;
1300 Label L_join;
1301 const Address access_flags (rcx, Method::access_flags_offset());
1302 __ get_method(rcx);
1303 __ movl(rcx, access_flags);
1304 __ testl(rcx, JVM_ACC_STRICT);
1305 __ jccb(Assembler::notZero, L_strict);
1306 __ fmul_d (at_rsp());
1307 __ jmpb(L_join);
1308 __ bind(L_strict);
1309 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1310 __ fmulp();
1311 __ fmul_d (at_rsp());
1312 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1313 __ fmulp();
1314 __ bind(L_join);
1315 break;
1316 }
1317 case div: {
1318 Label L_strict;
1319 Label L_join;
1320 const Address access_flags (rcx, Method::access_flags_offset());
1321 __ get_method(rcx);
1322 __ movl(rcx, access_flags);
1323 __ testl(rcx, JVM_ACC_STRICT);
1324 __ jccb(Assembler::notZero, L_strict);
1325 __ fdivr_d(at_rsp());
1326 __ jmp(L_join);
1327 __ bind(L_strict);
1328 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1329 __ fmul_d (at_rsp());
1330 __ fdivrp();
1331 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1332 __ fmulp();
1333 __ bind(L_join);
1334 break;
1335 }
1336 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1337 default : ShouldNotReachHere();
1338 }
1339 __ d2ieee();
1340 // Pop double precision number from rsp.
1341 __ pop(rax);
1342 __ pop(rdx);
1343 }
1346 void TemplateTable::ineg() {
1347 transition(itos, itos);
1348 __ negl(rax);
1349 }
1352 void TemplateTable::lneg() {
1353 transition(ltos, ltos);
1354 __ lneg(rdx, rax);
1355 }
1358 void TemplateTable::fneg() {
1359 transition(ftos, ftos);
1360 __ fchs();
1361 }
1364 void TemplateTable::dneg() {
1365 transition(dtos, dtos);
1366 __ fchs();
1367 }
1370 void TemplateTable::iinc() {
1371 transition(vtos, vtos);
1372 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1373 locals_index(rbx);
1374 __ addl(iaddress(rbx), rdx);
1375 }
1378 void TemplateTable::wide_iinc() {
1379 transition(vtos, vtos);
1380 __ movl(rdx, at_bcp(4)); // get constant
1381 locals_index_wide(rbx);
1382 __ bswapl(rdx); // swap bytes & sign-extend constant
1383 __ sarl(rdx, 16);
1384 __ addl(iaddress(rbx), rdx);
1385 // Note: should probably use only one movl to get both
1386 // the index and the constant -> fix this
1387 }
1390 void TemplateTable::convert() {
1391 // Checking
1392 #ifdef ASSERT
1393 { TosState tos_in = ilgl;
1394 TosState tos_out = ilgl;
1395 switch (bytecode()) {
1396 case Bytecodes::_i2l: // fall through
1397 case Bytecodes::_i2f: // fall through
1398 case Bytecodes::_i2d: // fall through
1399 case Bytecodes::_i2b: // fall through
1400 case Bytecodes::_i2c: // fall through
1401 case Bytecodes::_i2s: tos_in = itos; break;
1402 case Bytecodes::_l2i: // fall through
1403 case Bytecodes::_l2f: // fall through
1404 case Bytecodes::_l2d: tos_in = ltos; break;
1405 case Bytecodes::_f2i: // fall through
1406 case Bytecodes::_f2l: // fall through
1407 case Bytecodes::_f2d: tos_in = ftos; break;
1408 case Bytecodes::_d2i: // fall through
1409 case Bytecodes::_d2l: // fall through
1410 case Bytecodes::_d2f: tos_in = dtos; break;
1411 default : ShouldNotReachHere();
1412 }
1413 switch (bytecode()) {
1414 case Bytecodes::_l2i: // fall through
1415 case Bytecodes::_f2i: // fall through
1416 case Bytecodes::_d2i: // fall through
1417 case Bytecodes::_i2b: // fall through
1418 case Bytecodes::_i2c: // fall through
1419 case Bytecodes::_i2s: tos_out = itos; break;
1420 case Bytecodes::_i2l: // fall through
1421 case Bytecodes::_f2l: // fall through
1422 case Bytecodes::_d2l: tos_out = ltos; break;
1423 case Bytecodes::_i2f: // fall through
1424 case Bytecodes::_l2f: // fall through
1425 case Bytecodes::_d2f: tos_out = ftos; break;
1426 case Bytecodes::_i2d: // fall through
1427 case Bytecodes::_l2d: // fall through
1428 case Bytecodes::_f2d: tos_out = dtos; break;
1429 default : ShouldNotReachHere();
1430 }
1431 transition(tos_in, tos_out);
1432 }
1433 #endif // ASSERT
1435 // Conversion
1436 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1437 switch (bytecode()) {
1438 case Bytecodes::_i2l:
1439 __ extend_sign(rdx, rax);
1440 break;
1441 case Bytecodes::_i2f:
1442 __ push(rax); // store int on tos
1443 __ fild_s(at_rsp()); // load int to ST0
1444 __ f2ieee(); // truncate to float size
1445 __ pop(rcx); // adjust rsp
1446 break;
1447 case Bytecodes::_i2d:
1448 __ push(rax); // add one slot for d2ieee()
1449 __ push(rax); // store int on tos
1450 __ fild_s(at_rsp()); // load int to ST0
1451 __ d2ieee(); // truncate to double size
1452 __ pop(rcx); // adjust rsp
1453 __ pop(rcx);
1454 break;
1455 case Bytecodes::_i2b:
1456 __ shll(rax, 24); // truncate upper 24 bits
1457 __ sarl(rax, 24); // and sign-extend byte
1458 LP64_ONLY(__ movsbl(rax, rax));
1459 break;
1460 case Bytecodes::_i2c:
1461 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1462 LP64_ONLY(__ movzwl(rax, rax));
1463 break;
1464 case Bytecodes::_i2s:
1465 __ shll(rax, 16); // truncate upper 16 bits
1466 __ sarl(rax, 16); // and sign-extend short
1467 LP64_ONLY(__ movswl(rax, rax));
1468 break;
1469 case Bytecodes::_l2i:
1470 /* nothing to do */
1471 break;
1472 case Bytecodes::_l2f:
1473 __ push(rdx); // store long on tos
1474 __ push(rax);
1475 __ fild_d(at_rsp()); // load long to ST0
1476 __ f2ieee(); // truncate to float size
1477 __ pop(rcx); // adjust rsp
1478 __ pop(rcx);
1479 break;
1480 case Bytecodes::_l2d:
1481 __ push(rdx); // store long on tos
1482 __ push(rax);
1483 __ fild_d(at_rsp()); // load long to ST0
1484 __ d2ieee(); // truncate to double size
1485 __ pop(rcx); // adjust rsp
1486 __ pop(rcx);
1487 break;
1488 case Bytecodes::_f2i:
1489 __ push(rcx); // reserve space for argument
1490 __ fstp_s(at_rsp()); // pass float argument on stack
1491 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1492 break;
1493 case Bytecodes::_f2l:
1494 __ push(rcx); // reserve space for argument
1495 __ fstp_s(at_rsp()); // pass float argument on stack
1496 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1497 break;
1498 case Bytecodes::_f2d:
1499 /* nothing to do */
1500 break;
1501 case Bytecodes::_d2i:
1502 __ push(rcx); // reserve space for argument
1503 __ push(rcx);
1504 __ fstp_d(at_rsp()); // pass double argument on stack
1505 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1506 break;
1507 case Bytecodes::_d2l:
1508 __ push(rcx); // reserve space for argument
1509 __ push(rcx);
1510 __ fstp_d(at_rsp()); // pass double argument on stack
1511 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1512 break;
1513 case Bytecodes::_d2f:
1514 __ push(rcx); // reserve space for f2ieee()
1515 __ f2ieee(); // truncate to float size
1516 __ pop(rcx); // adjust rsp
1517 break;
1518 default :
1519 ShouldNotReachHere();
1520 }
1521 }
1524 void TemplateTable::lcmp() {
1525 transition(ltos, itos);
1526 // y = rdx:rax
1527 __ pop_l(rbx, rcx); // get x = rcx:rbx
1528 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1529 __ mov(rax, rcx);
1530 }
1533 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1534 if (is_float) {
1535 __ fld_s(at_rsp());
1536 } else {
1537 __ fld_d(at_rsp());
1538 __ pop(rdx);
1539 }
1540 __ pop(rcx);
1541 __ fcmp2int(rax, unordered_result < 0);
1542 }
1545 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1546 __ get_method(rcx); // ECX holds method
1547 __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count
1549 const ByteSize be_offset = Method::backedge_counter_offset() + InvocationCounter::counter_offset();
1550 const ByteSize inv_offset = Method::invocation_counter_offset() + InvocationCounter::counter_offset();
1551 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1553 // Load up EDX with the branch displacement
1554 __ movl(rdx, at_bcp(1));
1555 __ bswapl(rdx);
1556 if (!is_wide) __ sarl(rdx, 16);
1557 LP64_ONLY(__ movslq(rdx, rdx));
1560 // Handle all the JSR stuff here, then exit.
1561 // It's much shorter and cleaner than intermingling with the
1562 // non-JSR normal-branch stuff occurring below.
1563 if (is_jsr) {
1564 // Pre-load the next target bytecode into EBX
1565 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0));
1567 // compute return address as bci in rax,
1568 __ lea(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(ConstMethod::codes_offset())));
1569 __ subptr(rax, Address(rcx, Method::const_offset()));
1570 // Adjust the bcp in RSI by the displacement in EDX
1571 __ addptr(rsi, rdx);
1572 // Push return address
1573 __ push_i(rax);
1574 // jsr returns vtos
1575 __ dispatch_only_noverify(vtos);
1576 return;
1577 }
1579 // Normal (non-jsr) branch handling
1581 // Adjust the bcp in RSI by the displacement in EDX
1582 __ addptr(rsi, rdx);
1584 assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1585 Label backedge_counter_overflow;
1586 Label profile_method;
1587 Label dispatch;
1588 if (UseLoopCounter) {
1589 // increment backedge counter for backward branches
1590 // rax,: MDO
1591 // rbx,: MDO bumped taken-count
1592 // rcx: method
1593 // rdx: target offset
1594 // rsi: target bcp
1595 // rdi: locals pointer
1596 __ testl(rdx, rdx); // check if forward or backward branch
1597 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1599 if (TieredCompilation) {
1600 Label no_mdo;
1601 int increment = InvocationCounter::count_increment;
1602 int mask = ((1 << Tier0BackedgeNotifyFreqLog) - 1) << InvocationCounter::count_shift;
1603 if (ProfileInterpreter) {
1604 // Are we profiling?
1605 __ movptr(rbx, Address(rcx, in_bytes(Method::method_data_offset())));
1606 __ testptr(rbx, rbx);
1607 __ jccb(Assembler::zero, no_mdo);
1608 // Increment the MDO backedge counter
1609 const Address mdo_backedge_counter(rbx, in_bytes(MethodData::backedge_counter_offset()) +
1610 in_bytes(InvocationCounter::counter_offset()));
1611 __ increment_mask_and_jump(mdo_backedge_counter, increment, mask,
1612 rax, false, Assembler::zero, &backedge_counter_overflow);
1613 __ jmp(dispatch);
1614 }
1615 __ bind(no_mdo);
1616 // Increment backedge counter in Method*
1617 __ increment_mask_and_jump(Address(rcx, be_offset), increment, mask,
1618 rax, false, Assembler::zero, &backedge_counter_overflow);
1619 } else {
1620 // increment counter
1621 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1622 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1623 __ movl(Address(rcx, be_offset), rax); // store counter
1625 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1626 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1627 __ addl(rax, Address(rcx, be_offset)); // add both counters
1629 if (ProfileInterpreter) {
1630 // Test to see if we should create a method data oop
1631 __ cmp32(rax,
1632 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1633 __ jcc(Assembler::less, dispatch);
1635 // if no method data exists, go to profile method
1636 __ test_method_data_pointer(rax, profile_method);
1638 if (UseOnStackReplacement) {
1639 // check for overflow against rbx, which is the MDO taken count
1640 __ cmp32(rbx,
1641 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1642 __ jcc(Assembler::below, dispatch);
1644 // When ProfileInterpreter is on, the backedge_count comes from the
1645 // MethodData*, which value does not get reset on the call to
1646 // frequency_counter_overflow(). To avoid excessive calls to the overflow
1647 // routine while the method is being compiled, add a second test to make
1648 // sure the overflow function is called only once every overflow_frequency.
1649 const int overflow_frequency = 1024;
1650 __ andptr(rbx, overflow_frequency-1);
1651 __ jcc(Assembler::zero, backedge_counter_overflow);
1652 }
1653 } else {
1654 if (UseOnStackReplacement) {
1655 // check for overflow against rax, which is the sum of the counters
1656 __ cmp32(rax,
1657 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1658 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1660 }
1661 }
1662 }
1663 __ bind(dispatch);
1664 }
1666 // Pre-load the next target bytecode into EBX
1667 __ load_unsigned_byte(rbx, Address(rsi, 0));
1669 // continue with the bytecode @ target
1670 // rax,: return bci for jsr's, unused otherwise
1671 // rbx,: target bytecode
1672 // rsi: target bcp
1673 __ dispatch_only(vtos);
1675 if (UseLoopCounter) {
1676 if (ProfileInterpreter) {
1677 // Out-of-line code to allocate method data oop.
1678 __ bind(profile_method);
1679 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method));
1680 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1681 __ set_method_data_pointer_for_bcp();
1682 __ jmp(dispatch);
1683 }
1685 if (UseOnStackReplacement) {
1687 // invocation counter overflow
1688 __ bind(backedge_counter_overflow);
1689 __ negptr(rdx);
1690 __ addptr(rdx, rsi); // branch bcp
1691 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx);
1692 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1694 // rax,: osr nmethod (osr ok) or NULL (osr not possible)
1695 // rbx,: target bytecode
1696 // rdx: scratch
1697 // rdi: locals pointer
1698 // rsi: bcp
1699 __ testptr(rax, rax); // test result
1700 __ jcc(Assembler::zero, dispatch); // no osr if null
1701 // nmethod may have been invalidated (VM may block upon call_VM return)
1702 __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1703 __ cmpl(rcx, InvalidOSREntryBci);
1704 __ jcc(Assembler::equal, dispatch);
1706 // We have the address of an on stack replacement routine in rax,
1707 // We need to prepare to execute the OSR method. First we must
1708 // migrate the locals and monitors off of the stack.
1710 __ mov(rbx, rax); // save the nmethod
1712 const Register thread = rcx;
1713 __ get_thread(thread);
1714 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1715 // rax, is OSR buffer, move it to expected parameter location
1716 __ mov(rcx, rax);
1718 // pop the interpreter frame
1719 __ movptr(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1720 __ leave(); // remove frame anchor
1721 __ pop(rdi); // get return address
1722 __ mov(rsp, rdx); // set sp to sender sp
1724 // Align stack pointer for compiled code (note that caller is
1725 // responsible for undoing this fixup by remembering the old SP
1726 // in an rbp,-relative location)
1727 __ andptr(rsp, -(StackAlignmentInBytes));
1729 // push the (possibly adjusted) return address
1730 __ push(rdi);
1732 // and begin the OSR nmethod
1733 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
1734 }
1735 }
1736 }
1739 void TemplateTable::if_0cmp(Condition cc) {
1740 transition(itos, vtos);
1741 // assume branch is more often taken than not (loops use backward branches)
1742 Label not_taken;
1743 __ testl(rax, rax);
1744 __ jcc(j_not(cc), not_taken);
1745 branch(false, false);
1746 __ bind(not_taken);
1747 __ profile_not_taken_branch(rax);
1748 }
1751 void TemplateTable::if_icmp(Condition cc) {
1752 transition(itos, vtos);
1753 // assume branch is more often taken than not (loops use backward branches)
1754 Label not_taken;
1755 __ pop_i(rdx);
1756 __ cmpl(rdx, rax);
1757 __ jcc(j_not(cc), not_taken);
1758 branch(false, false);
1759 __ bind(not_taken);
1760 __ profile_not_taken_branch(rax);
1761 }
1764 void TemplateTable::if_nullcmp(Condition cc) {
1765 transition(atos, vtos);
1766 // assume branch is more often taken than not (loops use backward branches)
1767 Label not_taken;
1768 __ testptr(rax, rax);
1769 __ jcc(j_not(cc), not_taken);
1770 branch(false, false);
1771 __ bind(not_taken);
1772 __ profile_not_taken_branch(rax);
1773 }
1776 void TemplateTable::if_acmp(Condition cc) {
1777 transition(atos, vtos);
1778 // assume branch is more often taken than not (loops use backward branches)
1779 Label not_taken;
1780 __ pop_ptr(rdx);
1781 __ cmpptr(rdx, rax);
1782 __ jcc(j_not(cc), not_taken);
1783 branch(false, false);
1784 __ bind(not_taken);
1785 __ profile_not_taken_branch(rax);
1786 }
1789 void TemplateTable::ret() {
1790 transition(vtos, vtos);
1791 locals_index(rbx);
1792 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1793 __ profile_ret(rbx, rcx);
1794 __ get_method(rax);
1795 __ movptr(rsi, Address(rax, Method::const_offset()));
1796 __ lea(rsi, Address(rsi, rbx, Address::times_1,
1797 ConstMethod::codes_offset()));
1798 __ dispatch_next(vtos);
1799 }
1802 void TemplateTable::wide_ret() {
1803 transition(vtos, vtos);
1804 locals_index_wide(rbx);
1805 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1806 __ profile_ret(rbx, rcx);
1807 __ get_method(rax);
1808 __ movptr(rsi, Address(rax, Method::const_offset()));
1809 __ lea(rsi, Address(rsi, rbx, Address::times_1, ConstMethod::codes_offset()));
1810 __ dispatch_next(vtos);
1811 }
1814 void TemplateTable::tableswitch() {
1815 Label default_case, continue_execution;
1816 transition(itos, vtos);
1817 // align rsi
1818 __ lea(rbx, at_bcp(wordSize));
1819 __ andptr(rbx, -wordSize);
1820 // load lo & hi
1821 __ movl(rcx, Address(rbx, 1 * wordSize));
1822 __ movl(rdx, Address(rbx, 2 * wordSize));
1823 __ bswapl(rcx);
1824 __ bswapl(rdx);
1825 // check against lo & hi
1826 __ cmpl(rax, rcx);
1827 __ jccb(Assembler::less, default_case);
1828 __ cmpl(rax, rdx);
1829 __ jccb(Assembler::greater, default_case);
1830 // lookup dispatch offset
1831 __ subl(rax, rcx);
1832 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1833 __ profile_switch_case(rax, rbx, rcx);
1834 // continue execution
1835 __ bind(continue_execution);
1836 __ bswapl(rdx);
1837 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1838 __ addptr(rsi, rdx);
1839 __ dispatch_only(vtos);
1840 // handle default
1841 __ bind(default_case);
1842 __ profile_switch_default(rax);
1843 __ movl(rdx, Address(rbx, 0));
1844 __ jmp(continue_execution);
1845 }
1848 void TemplateTable::lookupswitch() {
1849 transition(itos, itos);
1850 __ stop("lookupswitch bytecode should have been rewritten");
1851 }
1854 void TemplateTable::fast_linearswitch() {
1855 transition(itos, vtos);
1856 Label loop_entry, loop, found, continue_execution;
1857 // bswapl rax, so we can avoid bswapping the table entries
1858 __ bswapl(rax);
1859 // align rsi
1860 __ lea(rbx, at_bcp(wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1861 __ andptr(rbx, -wordSize);
1862 // set counter
1863 __ movl(rcx, Address(rbx, wordSize));
1864 __ bswapl(rcx);
1865 __ jmpb(loop_entry);
1866 // table search
1867 __ bind(loop);
1868 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize));
1869 __ jccb(Assembler::equal, found);
1870 __ bind(loop_entry);
1871 __ decrementl(rcx);
1872 __ jcc(Assembler::greaterEqual, loop);
1873 // default case
1874 __ profile_switch_default(rax);
1875 __ movl(rdx, Address(rbx, 0));
1876 __ jmpb(continue_execution);
1877 // entry found -> get offset
1878 __ bind(found);
1879 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize));
1880 __ profile_switch_case(rcx, rax, rbx);
1881 // continue execution
1882 __ bind(continue_execution);
1883 __ bswapl(rdx);
1884 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1885 __ addptr(rsi, rdx);
1886 __ dispatch_only(vtos);
1887 }
1890 void TemplateTable::fast_binaryswitch() {
1891 transition(itos, vtos);
1892 // Implementation using the following core algorithm:
1893 //
1894 // int binary_search(int key, LookupswitchPair* array, int n) {
1895 // // Binary search according to "Methodik des Programmierens" by
1896 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1897 // int i = 0;
1898 // int j = n;
1899 // while (i+1 < j) {
1900 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1901 // // with Q: for all i: 0 <= i < n: key < a[i]
1902 // // where a stands for the array and assuming that the (inexisting)
1903 // // element a[n] is infinitely big.
1904 // int h = (i + j) >> 1;
1905 // // i < h < j
1906 // if (key < array[h].fast_match()) {
1907 // j = h;
1908 // } else {
1909 // i = h;
1910 // }
1911 // }
1912 // // R: a[i] <= key < a[i+1] or Q
1913 // // (i.e., if key is within array, i is the correct index)
1914 // return i;
1915 // }
1917 // register allocation
1918 const Register key = rax; // already set (tosca)
1919 const Register array = rbx;
1920 const Register i = rcx;
1921 const Register j = rdx;
1922 const Register h = rdi; // needs to be restored
1923 const Register temp = rsi;
1924 // setup array
1925 __ save_bcp();
1927 __ lea(array, at_bcp(3*wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1928 __ andptr(array, -wordSize);
1929 // initialize i & j
1930 __ xorl(i, i); // i = 0;
1931 __ movl(j, Address(array, -wordSize)); // j = length(array);
1932 // Convert j into native byteordering
1933 __ bswapl(j);
1934 // and start
1935 Label entry;
1936 __ jmp(entry);
1938 // binary search loop
1939 { Label loop;
1940 __ bind(loop);
1941 // int h = (i + j) >> 1;
1942 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1943 __ sarl(h, 1); // h = (i + j) >> 1;
1944 // if (key < array[h].fast_match()) {
1945 // j = h;
1946 // } else {
1947 // i = h;
1948 // }
1949 // Convert array[h].match to native byte-ordering before compare
1950 __ movl(temp, Address(array, h, Address::times_8, 0*wordSize));
1951 __ bswapl(temp);
1952 __ cmpl(key, temp);
1953 // j = h if (key < array[h].fast_match())
1954 __ cmov32(Assembler::less , j, h);
1955 // i = h if (key >= array[h].fast_match())
1956 __ cmov32(Assembler::greaterEqual, i, h);
1957 // while (i+1 < j)
1958 __ bind(entry);
1959 __ leal(h, Address(i, 1)); // i+1
1960 __ cmpl(h, j); // i+1 < j
1961 __ jcc(Assembler::less, loop);
1962 }
1964 // end of binary search, result index is i (must check again!)
1965 Label default_case;
1966 // Convert array[i].match to native byte-ordering before compare
1967 __ movl(temp, Address(array, i, Address::times_8, 0*wordSize));
1968 __ bswapl(temp);
1969 __ cmpl(key, temp);
1970 __ jcc(Assembler::notEqual, default_case);
1972 // entry found -> j = offset
1973 __ movl(j , Address(array, i, Address::times_8, 1*wordSize));
1974 __ profile_switch_case(i, key, array);
1975 __ bswapl(j);
1976 LP64_ONLY(__ movslq(j, j));
1977 __ restore_bcp();
1978 __ restore_locals(); // restore rdi
1979 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1981 __ addptr(rsi, j);
1982 __ dispatch_only(vtos);
1984 // default case -> j = default offset
1985 __ bind(default_case);
1986 __ profile_switch_default(i);
1987 __ movl(j, Address(array, -2*wordSize));
1988 __ bswapl(j);
1989 LP64_ONLY(__ movslq(j, j));
1990 __ restore_bcp();
1991 __ restore_locals(); // restore rdi
1992 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1993 __ addptr(rsi, j);
1994 __ dispatch_only(vtos);
1995 }
1998 void TemplateTable::_return(TosState state) {
1999 transition(state, state);
2000 assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation
2002 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
2003 assert(state == vtos, "only valid state");
2004 __ movptr(rax, aaddress(0));
2005 __ load_klass(rdi, rax);
2006 __ movl(rdi, Address(rdi, Klass::access_flags_offset()));
2007 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
2008 Label skip_register_finalizer;
2009 __ jcc(Assembler::zero, skip_register_finalizer);
2011 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax);
2013 __ bind(skip_register_finalizer);
2014 }
2016 __ remove_activation(state, rsi);
2017 __ jmp(rsi);
2018 }
2021 // ----------------------------------------------------------------------------
2022 // Volatile variables demand their effects be made known to all CPU's in
2023 // order. Store buffers on most chips allow reads & writes to reorder; the
2024 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2025 // memory barrier (i.e., it's not sufficient that the interpreter does not
2026 // reorder volatile references, the hardware also must not reorder them).
2027 //
2028 // According to the new Java Memory Model (JMM):
2029 // (1) All volatiles are serialized wrt to each other.
2030 // ALSO reads & writes act as aquire & release, so:
2031 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
2032 // the read float up to before the read. It's OK for non-volatile memory refs
2033 // that happen before the volatile read to float down below it.
2034 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2035 // that happen BEFORE the write float down to after the write. It's OK for
2036 // non-volatile memory refs that happen after the volatile write to float up
2037 // before it.
2038 //
2039 // We only put in barriers around volatile refs (they are expensive), not
2040 // _between_ memory refs (that would require us to track the flavor of the
2041 // previous memory refs). Requirements (2) and (3) require some barriers
2042 // before volatile stores and after volatile loads. These nearly cover
2043 // requirement (1) but miss the volatile-store-volatile-load case. This final
2044 // case is placed after volatile-stores although it could just as well go
2045 // before volatile-loads.
2046 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2047 // Helper function to insert a is-volatile test and memory barrier
2048 if( !os::is_MP() ) return; // Not needed on single CPU
2049 __ membar(order_constraint);
2050 }
2052 void TemplateTable::resolve_cache_and_index(int byte_no,
2053 Register Rcache,
2054 Register index,
2055 size_t index_size) {
2056 const Register temp = rbx;
2057 assert_different_registers(Rcache, index, temp);
2059 Label resolved;
2060 assert(byte_no == f1_byte || byte_no == f2_byte, "byte_no out of range");
2061 __ get_cache_and_index_and_bytecode_at_bcp(Rcache, index, temp, byte_no, 1, index_size);
2062 __ cmpl(temp, (int) bytecode()); // have we resolved this bytecode?
2063 __ jcc(Assembler::equal, resolved);
2065 // resolve first time through
2066 address entry;
2067 switch (bytecode()) {
2068 case Bytecodes::_getstatic : // fall through
2069 case Bytecodes::_putstatic : // fall through
2070 case Bytecodes::_getfield : // fall through
2071 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2072 case Bytecodes::_invokevirtual : // fall through
2073 case Bytecodes::_invokespecial : // fall through
2074 case Bytecodes::_invokestatic : // fall through
2075 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2076 case Bytecodes::_invokehandle : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokehandle); break;
2077 case Bytecodes::_invokedynamic : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invokedynamic); break;
2078 default:
2079 fatal(err_msg("unexpected bytecode: %s", Bytecodes::name(bytecode())));
2080 break;
2081 }
2082 __ movl(temp, (int)bytecode());
2083 __ call_VM(noreg, entry, temp);
2084 // Update registers with resolved info
2085 __ get_cache_and_index_at_bcp(Rcache, index, 1, index_size);
2086 __ bind(resolved);
2087 }
2090 // The cache and index registers must be set before call
2091 void TemplateTable::load_field_cp_cache_entry(Register obj,
2092 Register cache,
2093 Register index,
2094 Register off,
2095 Register flags,
2096 bool is_static = false) {
2097 assert_different_registers(cache, index, flags, off);
2099 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2100 // Field offset
2101 __ movptr(off, Address(cache, index, Address::times_ptr,
2102 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset())));
2103 // Flags
2104 __ movl(flags, Address(cache, index, Address::times_ptr,
2105 in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset())));
2107 // klass overwrite register
2108 if (is_static) {
2109 __ movptr(obj, Address(cache, index, Address::times_ptr,
2110 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset())));
2111 const int mirror_offset = in_bytes(Klass::java_mirror_offset());
2112 __ movptr(obj, Address(obj, mirror_offset));
2113 }
2114 }
2116 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2117 Register method,
2118 Register itable_index,
2119 Register flags,
2120 bool is_invokevirtual,
2121 bool is_invokevfinal, /*unused*/
2122 bool is_invokedynamic) {
2123 // setup registers
2124 const Register cache = rcx;
2125 const Register index = rdx;
2126 assert_different_registers(method, flags);
2127 assert_different_registers(method, cache, index);
2128 assert_different_registers(itable_index, flags);
2129 assert_different_registers(itable_index, cache, index);
2130 // determine constant pool cache field offsets
2131 assert(is_invokevirtual == (byte_no == f2_byte), "is_invokevirtual flag redundant");
2132 const int method_offset = in_bytes(
2133 ConstantPoolCache::base_offset() +
2134 ((byte_no == f2_byte)
2135 ? ConstantPoolCacheEntry::f2_offset()
2136 : ConstantPoolCacheEntry::f1_offset()));
2137 const int flags_offset = in_bytes(ConstantPoolCache::base_offset() +
2138 ConstantPoolCacheEntry::flags_offset());
2139 // access constant pool cache fields
2140 const int index_offset = in_bytes(ConstantPoolCache::base_offset() +
2141 ConstantPoolCacheEntry::f2_offset());
2143 size_t index_size = (is_invokedynamic ? sizeof(u4) : sizeof(u2));
2144 resolve_cache_and_index(byte_no, cache, index, index_size);
2145 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2147 if (itable_index != noreg) {
2148 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2149 }
2150 __ movl(flags, Address(cache, index, Address::times_ptr, flags_offset));
2151 }
2154 // The registers cache and index expected to be set before call.
2155 // Correct values of the cache and index registers are preserved.
2156 void TemplateTable::jvmti_post_field_access(Register cache,
2157 Register index,
2158 bool is_static,
2159 bool has_tos) {
2160 if (JvmtiExport::can_post_field_access()) {
2161 // Check to see if a field access watch has been set before we take
2162 // the time to call into the VM.
2163 Label L1;
2164 assert_different_registers(cache, index, rax);
2165 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2166 __ testl(rax,rax);
2167 __ jcc(Assembler::zero, L1);
2169 // cache entry pointer
2170 __ addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
2171 __ shll(index, LogBytesPerWord);
2172 __ addptr(cache, index);
2173 if (is_static) {
2174 __ xorptr(rax, rax); // NULL object reference
2175 } else {
2176 __ pop(atos); // Get the object
2177 __ verify_oop(rax);
2178 __ push(atos); // Restore stack state
2179 }
2180 // rax,: object pointer or NULL
2181 // cache: cache entry pointer
2182 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2183 rax, cache);
2184 __ get_cache_and_index_at_bcp(cache, index, 1);
2185 __ bind(L1);
2186 }
2187 }
2189 void TemplateTable::pop_and_check_object(Register r) {
2190 __ pop_ptr(r);
2191 __ null_check(r); // for field access must check obj.
2192 __ verify_oop(r);
2193 }
2195 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2196 transition(vtos, vtos);
2198 const Register cache = rcx;
2199 const Register index = rdx;
2200 const Register obj = rcx;
2201 const Register off = rbx;
2202 const Register flags = rax;
2204 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2205 jvmti_post_field_access(cache, index, is_static, false);
2206 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2208 if (!is_static) pop_and_check_object(obj);
2210 const Address lo(obj, off, Address::times_1, 0*wordSize);
2211 const Address hi(obj, off, Address::times_1, 1*wordSize);
2213 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2215 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2216 assert(btos == 0, "change code, btos != 0");
2217 // btos
2218 __ andptr(flags, ConstantPoolCacheEntry::tos_state_mask);
2219 __ jcc(Assembler::notZero, notByte);
2221 __ load_signed_byte(rax, lo );
2222 __ push(btos);
2223 // Rewrite bytecode to be faster
2224 if (!is_static) {
2225 patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx);
2226 }
2227 __ jmp(Done);
2229 __ bind(notByte);
2230 // itos
2231 __ cmpl(flags, itos );
2232 __ jcc(Assembler::notEqual, notInt);
2234 __ movl(rax, lo );
2235 __ push(itos);
2236 // Rewrite bytecode to be faster
2237 if (!is_static) {
2238 patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx);
2239 }
2240 __ jmp(Done);
2242 __ bind(notInt);
2243 // atos
2244 __ cmpl(flags, atos );
2245 __ jcc(Assembler::notEqual, notObj);
2247 __ movl(rax, lo );
2248 __ push(atos);
2249 if (!is_static) {
2250 patch_bytecode(Bytecodes::_fast_agetfield, rcx, rbx);
2251 }
2252 __ jmp(Done);
2254 __ bind(notObj);
2255 // ctos
2256 __ cmpl(flags, ctos );
2257 __ jcc(Assembler::notEqual, notChar);
2259 __ load_unsigned_short(rax, lo );
2260 __ push(ctos);
2261 if (!is_static) {
2262 patch_bytecode(Bytecodes::_fast_cgetfield, rcx, rbx);
2263 }
2264 __ jmp(Done);
2266 __ bind(notChar);
2267 // stos
2268 __ cmpl(flags, stos );
2269 __ jcc(Assembler::notEqual, notShort);
2271 __ load_signed_short(rax, lo );
2272 __ push(stos);
2273 if (!is_static) {
2274 patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx);
2275 }
2276 __ jmp(Done);
2278 __ bind(notShort);
2279 // ltos
2280 __ cmpl(flags, ltos );
2281 __ jcc(Assembler::notEqual, notLong);
2283 // Generate code as if volatile. There just aren't enough registers to
2284 // save that information and this code is faster than the test.
2285 __ fild_d(lo); // Must load atomically
2286 __ subptr(rsp,2*wordSize); // Make space for store
2287 __ fistp_d(Address(rsp,0));
2288 __ pop(rax);
2289 __ pop(rdx);
2291 __ push(ltos);
2292 // Don't rewrite to _fast_lgetfield for potential volatile case.
2293 __ jmp(Done);
2295 __ bind(notLong);
2296 // ftos
2297 __ cmpl(flags, ftos );
2298 __ jcc(Assembler::notEqual, notFloat);
2300 __ fld_s(lo);
2301 __ push(ftos);
2302 if (!is_static) {
2303 patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx);
2304 }
2305 __ jmp(Done);
2307 __ bind(notFloat);
2308 // dtos
2309 __ cmpl(flags, dtos );
2310 __ jcc(Assembler::notEqual, notDouble);
2312 __ fld_d(lo);
2313 __ push(dtos);
2314 if (!is_static) {
2315 patch_bytecode(Bytecodes::_fast_dgetfield, rcx, rbx);
2316 }
2317 __ jmpb(Done);
2319 __ bind(notDouble);
2321 __ stop("Bad state");
2323 __ bind(Done);
2324 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2325 // volatile_barrier( );
2326 }
2329 void TemplateTable::getfield(int byte_no) {
2330 getfield_or_static(byte_no, false);
2331 }
2334 void TemplateTable::getstatic(int byte_no) {
2335 getfield_or_static(byte_no, true);
2336 }
2338 // The registers cache and index expected to be set before call.
2339 // The function may destroy various registers, just not the cache and index registers.
2340 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2342 ByteSize cp_base_offset = ConstantPoolCache::base_offset();
2344 if (JvmtiExport::can_post_field_modification()) {
2345 // Check to see if a field modification watch has been set before we take
2346 // the time to call into the VM.
2347 Label L1;
2348 assert_different_registers(cache, index, rax);
2349 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2350 __ testl(rax, rax);
2351 __ jcc(Assembler::zero, L1);
2353 // The cache and index registers have been already set.
2354 // This allows to eliminate this call but the cache and index
2355 // registers have to be correspondingly used after this line.
2356 __ get_cache_and_index_at_bcp(rax, rdx, 1);
2358 if (is_static) {
2359 // Life is simple. Null out the object pointer.
2360 __ xorptr(rbx, rbx);
2361 } else {
2362 // Life is harder. The stack holds the value on top, followed by the object.
2363 // We don't know the size of the value, though; it could be one or two words
2364 // depending on its type. As a result, we must find the type to determine where
2365 // the object is.
2366 Label two_word, valsize_known;
2367 __ movl(rcx, Address(rax, rdx, Address::times_ptr, in_bytes(cp_base_offset +
2368 ConstantPoolCacheEntry::flags_offset())));
2369 __ mov(rbx, rsp);
2370 __ shrl(rcx, ConstantPoolCacheEntry::tos_state_shift);
2371 // Make sure we don't need to mask rcx after the above shift
2372 ConstantPoolCacheEntry::verify_tos_state_shift();
2373 __ cmpl(rcx, ltos);
2374 __ jccb(Assembler::equal, two_word);
2375 __ cmpl(rcx, dtos);
2376 __ jccb(Assembler::equal, two_word);
2377 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2378 __ jmpb(valsize_known);
2380 __ bind(two_word);
2381 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2383 __ bind(valsize_known);
2384 // setup object pointer
2385 __ movptr(rbx, Address(rbx, 0));
2386 }
2387 // cache entry pointer
2388 __ addptr(rax, in_bytes(cp_base_offset));
2389 __ shll(rdx, LogBytesPerWord);
2390 __ addptr(rax, rdx);
2391 // object (tos)
2392 __ mov(rcx, rsp);
2393 // rbx,: object pointer set up above (NULL if static)
2394 // rax,: cache entry pointer
2395 // rcx: jvalue object on the stack
2396 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2397 rbx, rax, rcx);
2398 __ get_cache_and_index_at_bcp(cache, index, 1);
2399 __ bind(L1);
2400 }
2401 }
2404 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2405 transition(vtos, vtos);
2407 const Register cache = rcx;
2408 const Register index = rdx;
2409 const Register obj = rcx;
2410 const Register off = rbx;
2411 const Register flags = rax;
2413 resolve_cache_and_index(byte_no, cache, index, sizeof(u2));
2414 jvmti_post_field_mod(cache, index, is_static);
2415 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2417 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2418 // volatile_barrier( );
2420 Label notVolatile, Done;
2421 __ movl(rdx, flags);
2422 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2423 __ andl(rdx, 0x1);
2425 // field addresses
2426 const Address lo(obj, off, Address::times_1, 0*wordSize);
2427 const Address hi(obj, off, Address::times_1, 1*wordSize);
2429 Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2431 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2432 assert(btos == 0, "change code, btos != 0");
2433 __ andl(flags, ConstantPoolCacheEntry::tos_state_mask);
2434 __ jcc(Assembler::notZero, notByte);
2436 // btos
2437 {
2438 __ pop(btos);
2439 if (!is_static) pop_and_check_object(obj);
2440 __ movb(lo, rax);
2441 if (!is_static) {
2442 patch_bytecode(Bytecodes::_fast_bputfield, rcx, rbx, true, byte_no);
2443 }
2444 __ jmp(Done);
2445 }
2447 __ bind(notByte);
2448 __ cmpl(flags, itos);
2449 __ jcc(Assembler::notEqual, notInt);
2451 // itos
2452 {
2453 __ pop(itos);
2454 if (!is_static) pop_and_check_object(obj);
2455 __ movl(lo, rax);
2456 if (!is_static) {
2457 patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx, true, byte_no);
2458 }
2459 __ jmp(Done);
2460 }
2462 __ bind(notInt);
2463 __ cmpl(flags, atos);
2464 __ jcc(Assembler::notEqual, notObj);
2466 // atos
2467 {
2468 __ pop(atos);
2469 if (!is_static) pop_and_check_object(obj);
2470 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2471 if (!is_static) {
2472 patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx, true, byte_no);
2473 }
2474 __ jmp(Done);
2475 }
2477 __ bind(notObj);
2478 __ cmpl(flags, ctos);
2479 __ jcc(Assembler::notEqual, notChar);
2481 // ctos
2482 {
2483 __ pop(ctos);
2484 if (!is_static) pop_and_check_object(obj);
2485 __ movw(lo, rax);
2486 if (!is_static) {
2487 patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx, true, byte_no);
2488 }
2489 __ jmp(Done);
2490 }
2492 __ bind(notChar);
2493 __ cmpl(flags, stos);
2494 __ jcc(Assembler::notEqual, notShort);
2496 // stos
2497 {
2498 __ pop(stos);
2499 if (!is_static) pop_and_check_object(obj);
2500 __ movw(lo, rax);
2501 if (!is_static) {
2502 patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx, true, byte_no);
2503 }
2504 __ jmp(Done);
2505 }
2507 __ bind(notShort);
2508 __ cmpl(flags, ltos);
2509 __ jcc(Assembler::notEqual, notLong);
2511 // ltos
2512 {
2513 Label notVolatileLong;
2514 __ testl(rdx, rdx);
2515 __ jcc(Assembler::zero, notVolatileLong);
2517 __ pop(ltos); // overwrites rdx, do this after testing volatile.
2518 if (!is_static) pop_and_check_object(obj);
2520 // Replace with real volatile test
2521 __ push(rdx);
2522 __ push(rax); // Must update atomically with FIST
2523 __ fild_d(Address(rsp,0)); // So load into FPU register
2524 __ fistp_d(lo); // and put into memory atomically
2525 __ addptr(rsp, 2*wordSize);
2526 // volatile_barrier();
2527 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2528 Assembler::StoreStore));
2529 // Don't rewrite volatile version
2530 __ jmp(notVolatile);
2532 __ bind(notVolatileLong);
2534 __ pop(ltos); // overwrites rdx
2535 if (!is_static) pop_and_check_object(obj);
2536 NOT_LP64(__ movptr(hi, rdx));
2537 __ movptr(lo, rax);
2538 if (!is_static) {
2539 patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx, true, byte_no);
2540 }
2541 __ jmp(notVolatile);
2542 }
2544 __ bind(notLong);
2545 __ cmpl(flags, ftos);
2546 __ jcc(Assembler::notEqual, notFloat);
2548 // ftos
2549 {
2550 __ pop(ftos);
2551 if (!is_static) pop_and_check_object(obj);
2552 __ fstp_s(lo);
2553 if (!is_static) {
2554 patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx, true, byte_no);
2555 }
2556 __ jmp(Done);
2557 }
2559 __ bind(notFloat);
2560 #ifdef ASSERT
2561 __ cmpl(flags, dtos);
2562 __ jcc(Assembler::notEqual, notDouble);
2563 #endif
2565 // dtos
2566 {
2567 __ pop(dtos);
2568 if (!is_static) pop_and_check_object(obj);
2569 __ fstp_d(lo);
2570 if (!is_static) {
2571 patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx, true, byte_no);
2572 }
2573 __ jmp(Done);
2574 }
2576 #ifdef ASSERT
2577 __ bind(notDouble);
2578 __ stop("Bad state");
2579 #endif
2581 __ bind(Done);
2583 // Check for volatile store
2584 __ testl(rdx, rdx);
2585 __ jcc(Assembler::zero, notVolatile);
2586 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2587 Assembler::StoreStore));
2588 __ bind(notVolatile);
2589 }
2592 void TemplateTable::putfield(int byte_no) {
2593 putfield_or_static(byte_no, false);
2594 }
2597 void TemplateTable::putstatic(int byte_no) {
2598 putfield_or_static(byte_no, true);
2599 }
2601 void TemplateTable::jvmti_post_fast_field_mod() {
2602 if (JvmtiExport::can_post_field_modification()) {
2603 // Check to see if a field modification watch has been set before we take
2604 // the time to call into the VM.
2605 Label L2;
2606 __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2607 __ testl(rcx,rcx);
2608 __ jcc(Assembler::zero, L2);
2609 __ pop_ptr(rbx); // copy the object pointer from tos
2610 __ verify_oop(rbx);
2611 __ push_ptr(rbx); // put the object pointer back on tos
2613 // Save tos values before call_VM() clobbers them. Since we have
2614 // to do it for every data type, we use the saved values as the
2615 // jvalue object.
2616 switch (bytecode()) { // load values into the jvalue object
2617 case Bytecodes::_fast_aputfield: __ push_ptr(rax); break;
2618 case Bytecodes::_fast_bputfield: // fall through
2619 case Bytecodes::_fast_sputfield: // fall through
2620 case Bytecodes::_fast_cputfield: // fall through
2621 case Bytecodes::_fast_iputfield: __ push_i(rax); break;
2622 case Bytecodes::_fast_dputfield: __ push_d(); break;
2623 case Bytecodes::_fast_fputfield: __ push_f(); break;
2624 case Bytecodes::_fast_lputfield: __ push_l(rax); break;
2626 default:
2627 ShouldNotReachHere();
2628 }
2629 __ mov(rcx, rsp); // points to jvalue on the stack
2630 // access constant pool cache entry
2631 __ get_cache_entry_pointer_at_bcp(rax, rdx, 1);
2632 __ verify_oop(rbx);
2633 // rbx,: object pointer copied above
2634 // rax,: cache entry pointer
2635 // rcx: jvalue object on the stack
2636 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx);
2638 switch (bytecode()) { // restore tos values
2639 case Bytecodes::_fast_aputfield: __ pop_ptr(rax); break;
2640 case Bytecodes::_fast_bputfield: // fall through
2641 case Bytecodes::_fast_sputfield: // fall through
2642 case Bytecodes::_fast_cputfield: // fall through
2643 case Bytecodes::_fast_iputfield: __ pop_i(rax); break;
2644 case Bytecodes::_fast_dputfield: __ pop_d(); break;
2645 case Bytecodes::_fast_fputfield: __ pop_f(); break;
2646 case Bytecodes::_fast_lputfield: __ pop_l(rax); break;
2647 }
2648 __ bind(L2);
2649 }
2650 }
2652 void TemplateTable::fast_storefield(TosState state) {
2653 transition(state, vtos);
2655 ByteSize base = ConstantPoolCache::base_offset();
2657 jvmti_post_fast_field_mod();
2659 // access constant pool cache
2660 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2662 // test for volatile with rdx but rdx is tos register for lputfield.
2663 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2664 __ movl(rdx, Address(rcx, rbx, Address::times_ptr, in_bytes(base +
2665 ConstantPoolCacheEntry::flags_offset())));
2667 // replace index with field offset from cache entry
2668 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2670 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2671 // volatile_barrier( );
2673 Label notVolatile, Done;
2674 __ shrl(rdx, ConstantPoolCacheEntry::is_volatile_shift);
2675 __ andl(rdx, 0x1);
2676 // Check for volatile store
2677 __ testl(rdx, rdx);
2678 __ jcc(Assembler::zero, notVolatile);
2680 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2682 // Get object from stack
2683 pop_and_check_object(rcx);
2685 // field addresses
2686 const Address lo(rcx, rbx, Address::times_1, 0*wordSize);
2687 const Address hi(rcx, rbx, Address::times_1, 1*wordSize);
2689 // access field
2690 switch (bytecode()) {
2691 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2692 case Bytecodes::_fast_sputfield: // fall through
2693 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2694 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2695 case Bytecodes::_fast_lputfield:
2696 NOT_LP64(__ movptr(hi, rdx));
2697 __ movptr(lo, rax);
2698 break;
2699 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2700 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2701 case Bytecodes::_fast_aputfield: {
2702 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2703 break;
2704 }
2705 default:
2706 ShouldNotReachHere();
2707 }
2709 Label done;
2710 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2711 Assembler::StoreStore));
2712 // Barriers are so large that short branch doesn't reach!
2713 __ jmp(done);
2715 // Same code as above, but don't need rdx to test for volatile.
2716 __ bind(notVolatile);
2718 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2720 // Get object from stack
2721 pop_and_check_object(rcx);
2723 // access field
2724 switch (bytecode()) {
2725 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2726 case Bytecodes::_fast_sputfield: // fall through
2727 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2728 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2729 case Bytecodes::_fast_lputfield:
2730 NOT_LP64(__ movptr(hi, rdx));
2731 __ movptr(lo, rax);
2732 break;
2733 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2734 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2735 case Bytecodes::_fast_aputfield: {
2736 do_oop_store(_masm, lo, rax, _bs->kind(), false);
2737 break;
2738 }
2739 default:
2740 ShouldNotReachHere();
2741 }
2742 __ bind(done);
2743 }
2746 void TemplateTable::fast_accessfield(TosState state) {
2747 transition(atos, state);
2749 // do the JVMTI work here to avoid disturbing the register state below
2750 if (JvmtiExport::can_post_field_access()) {
2751 // Check to see if a field access watch has been set before we take
2752 // the time to call into the VM.
2753 Label L1;
2754 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2755 __ testl(rcx,rcx);
2756 __ jcc(Assembler::zero, L1);
2757 // access constant pool cache entry
2758 __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1);
2759 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
2760 __ verify_oop(rax);
2761 // rax,: object pointer copied above
2762 // rcx: cache entry pointer
2763 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx);
2764 __ pop_ptr(rax); // restore object pointer
2765 __ bind(L1);
2766 }
2768 // access constant pool cache
2769 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2770 // replace index with field offset from cache entry
2771 __ movptr(rbx, Address(rcx,
2772 rbx,
2773 Address::times_ptr,
2774 in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2777 // rax,: object
2778 __ verify_oop(rax);
2779 __ null_check(rax);
2780 // field addresses
2781 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2782 const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize);
2784 // access field
2785 switch (bytecode()) {
2786 case Bytecodes::_fast_bgetfield: __ movsbl(rax, lo ); break;
2787 case Bytecodes::_fast_sgetfield: __ load_signed_short(rax, lo ); break;
2788 case Bytecodes::_fast_cgetfield: __ load_unsigned_short(rax, lo ); break;
2789 case Bytecodes::_fast_igetfield: __ movl(rax, lo); break;
2790 case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten"); break;
2791 case Bytecodes::_fast_fgetfield: __ fld_s(lo); break;
2792 case Bytecodes::_fast_dgetfield: __ fld_d(lo); break;
2793 case Bytecodes::_fast_agetfield: __ movptr(rax, lo); __ verify_oop(rax); break;
2794 default:
2795 ShouldNotReachHere();
2796 }
2798 // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
2799 // volatile_barrier( );
2800 }
2802 void TemplateTable::fast_xaccess(TosState state) {
2803 transition(vtos, state);
2804 // get receiver
2805 __ movptr(rax, aaddress(0));
2806 // access constant pool cache
2807 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2808 __ movptr(rbx, Address(rcx,
2809 rdx,
2810 Address::times_ptr,
2811 in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2812 // make sure exception is reported in correct bcp range (getfield is next instruction)
2813 __ increment(rsi);
2814 __ null_check(rax);
2815 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2816 if (state == itos) {
2817 __ movl(rax, lo);
2818 } else if (state == atos) {
2819 __ movptr(rax, lo);
2820 __ verify_oop(rax);
2821 } else if (state == ftos) {
2822 __ fld_s(lo);
2823 } else {
2824 ShouldNotReachHere();
2825 }
2826 __ decrement(rsi);
2827 }
2831 //----------------------------------------------------------------------------------------------------
2832 // Calls
2834 void TemplateTable::count_calls(Register method, Register temp) {
2835 // implemented elsewhere
2836 ShouldNotReachHere();
2837 }
2840 void TemplateTable::prepare_invoke(int byte_no,
2841 Register method, // linked method (or i-klass)
2842 Register index, // itable index, MethodType, etc.
2843 Register recv, // if caller wants to see it
2844 Register flags // if caller wants to test it
2845 ) {
2846 // determine flags
2847 const Bytecodes::Code code = bytecode();
2848 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2849 const bool is_invokedynamic = code == Bytecodes::_invokedynamic;
2850 const bool is_invokehandle = code == Bytecodes::_invokehandle;
2851 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2852 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2853 const bool load_receiver = (recv != noreg);
2854 const bool save_flags = (flags != noreg);
2855 assert(load_receiver == (code != Bytecodes::_invokestatic && code != Bytecodes::_invokedynamic), "");
2856 assert(save_flags == (is_invokeinterface || is_invokevirtual), "need flags for vfinal");
2857 assert(flags == noreg || flags == rdx, "");
2858 assert(recv == noreg || recv == rcx, "");
2860 // setup registers & access constant pool cache
2861 if (recv == noreg) recv = rcx;
2862 if (flags == noreg) flags = rdx;
2863 assert_different_registers(method, index, recv, flags);
2865 // save 'interpreter return address'
2866 __ save_bcp();
2868 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual, false, is_invokedynamic);
2870 // maybe push appendix to arguments (just before return address)
2871 if (is_invokedynamic || is_invokehandle) {
2872 Label L_no_push;
2873 __ testl(flags, (1 << ConstantPoolCacheEntry::has_appendix_shift));
2874 __ jccb(Assembler::zero, L_no_push);
2875 // Push the appendix as a trailing parameter.
2876 // This must be done before we get the receiver,
2877 // since the parameter_size includes it.
2878 __ push(rbx);
2879 __ mov(rbx, index);
2880 assert(ConstantPoolCacheEntry::_indy_resolved_references_appendix_offset == 0, "appendix expected at index+0");
2881 __ load_resolved_reference_at_index(index, rbx);
2882 __ pop(rbx);
2883 __ push(index); // push appendix (MethodType, CallSite, etc.)
2884 __ bind(L_no_push);
2885 }
2887 // load receiver if needed (note: no return address pushed yet)
2888 if (load_receiver) {
2889 __ movl(recv, flags);
2890 __ andl(recv, ConstantPoolCacheEntry::parameter_size_mask);
2891 const int no_return_pc_pushed_yet = -1; // argument slot correction before we push return address
2892 const int receiver_is_at_end = -1; // back off one slot to get receiver
2893 Address recv_addr = __ argument_address(recv, no_return_pc_pushed_yet + receiver_is_at_end);
2894 __ movptr(recv, recv_addr);
2895 __ verify_oop(recv);
2896 }
2898 if (save_flags) {
2899 __ mov(rsi, flags);
2900 }
2902 // compute return type
2903 __ shrl(flags, ConstantPoolCacheEntry::tos_state_shift);
2904 // Make sure we don't need to mask flags after the above shift
2905 ConstantPoolCacheEntry::verify_tos_state_shift();
2906 // load return address
2907 {
2908 const address table_addr = (is_invokeinterface || is_invokedynamic) ?
2909 (address)Interpreter::return_5_addrs_by_index_table() :
2910 (address)Interpreter::return_3_addrs_by_index_table();
2911 ExternalAddress table(table_addr);
2912 __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));
2913 }
2915 // push return address
2916 __ push(flags);
2918 // Restore flags value from the constant pool cache, and restore rsi
2919 // for later null checks. rsi is the bytecode pointer
2920 if (save_flags) {
2921 __ mov(flags, rsi);
2922 __ restore_bcp();
2923 }
2924 }
2927 void TemplateTable::invokevirtual_helper(Register index,
2928 Register recv,
2929 Register flags) {
2930 // Uses temporary registers rax, rdx
2931 assert_different_registers(index, recv, rax, rdx);
2932 assert(index == rbx, "");
2933 assert(recv == rcx, "");
2935 // Test for an invoke of a final method
2936 Label notFinal;
2937 __ movl(rax, flags);
2938 __ andl(rax, (1 << ConstantPoolCacheEntry::is_vfinal_shift));
2939 __ jcc(Assembler::zero, notFinal);
2941 const Register method = index; // method must be rbx
2942 assert(method == rbx,
2943 "Method* must be rbx for interpreter calling convention");
2945 // do the call - the index is actually the method to call
2946 // that is, f2 is a vtable index if !is_vfinal, else f2 is a Method*
2948 // It's final, need a null check here!
2949 __ null_check(recv);
2951 // profile this call
2952 __ profile_final_call(rax);
2954 __ jump_from_interpreted(method, rax);
2956 __ bind(notFinal);
2958 // get receiver klass
2959 __ null_check(recv, oopDesc::klass_offset_in_bytes());
2960 __ load_klass(rax, recv);
2962 // profile this call
2963 __ profile_virtual_call(rax, rdi, rdx);
2965 // get target Method* & entry point
2966 __ lookup_virtual_method(rax, index, method);
2967 __ jump_from_interpreted(method, rdx);
2968 }
2971 void TemplateTable::invokevirtual(int byte_no) {
2972 transition(vtos, vtos);
2973 assert(byte_no == f2_byte, "use this argument");
2974 prepare_invoke(byte_no,
2975 rbx, // method or vtable index
2976 noreg, // unused itable index
2977 rcx, rdx); // recv, flags
2979 // rbx: index
2980 // rcx: receiver
2981 // rdx: flags
2983 invokevirtual_helper(rbx, rcx, rdx);
2984 }
2987 void TemplateTable::invokespecial(int byte_no) {
2988 transition(vtos, vtos);
2989 assert(byte_no == f1_byte, "use this argument");
2990 prepare_invoke(byte_no, rbx, noreg, // get f1 Method*
2991 rcx); // get receiver also for null check
2992 __ verify_oop(rcx);
2993 __ null_check(rcx);
2994 // do the call
2995 __ profile_call(rax);
2996 __ jump_from_interpreted(rbx, rax);
2997 }
3000 void TemplateTable::invokestatic(int byte_no) {
3001 transition(vtos, vtos);
3002 assert(byte_no == f1_byte, "use this argument");
3003 prepare_invoke(byte_no, rbx); // get f1 Method*
3004 // do the call
3005 __ profile_call(rax);
3006 __ jump_from_interpreted(rbx, rax);
3007 }
3010 void TemplateTable::fast_invokevfinal(int byte_no) {
3011 transition(vtos, vtos);
3012 assert(byte_no == f2_byte, "use this argument");
3013 __ stop("fast_invokevfinal not used on x86");
3014 }
3017 void TemplateTable::invokeinterface(int byte_no) {
3018 transition(vtos, vtos);
3019 assert(byte_no == f1_byte, "use this argument");
3020 prepare_invoke(byte_no, rax, rbx, // get f1 Klass*, f2 itable index
3021 rcx, rdx); // recv, flags
3023 // rax: interface klass (from f1)
3024 // rbx: itable index (from f2)
3025 // rcx: receiver
3026 // rdx: flags
3028 // Special case of invokeinterface called for virtual method of
3029 // java.lang.Object. See cpCacheOop.cpp for details.
3030 // This code isn't produced by javac, but could be produced by
3031 // another compliant java compiler.
3032 Label notMethod;
3033 __ movl(rdi, rdx);
3034 __ andl(rdi, (1 << ConstantPoolCacheEntry::is_forced_virtual_shift));
3035 __ jcc(Assembler::zero, notMethod);
3037 invokevirtual_helper(rbx, rcx, rdx);
3038 __ bind(notMethod);
3040 // Get receiver klass into rdx - also a null check
3041 __ restore_locals(); // restore rdi
3042 __ null_check(rcx, oopDesc::klass_offset_in_bytes());
3043 __ load_klass(rdx, rcx);
3045 // profile this call
3046 __ profile_virtual_call(rdx, rsi, rdi);
3048 Label no_such_interface, no_such_method;
3050 __ lookup_interface_method(// inputs: rec. class, interface, itable index
3051 rdx, rax, rbx,
3052 // outputs: method, scan temp. reg
3053 rbx, rsi,
3054 no_such_interface);
3056 // rbx: Method* to call
3057 // rcx: receiver
3058 // Check for abstract method error
3059 // Note: This should be done more efficiently via a throw_abstract_method_error
3060 // interpreter entry point and a conditional jump to it in case of a null
3061 // method.
3062 __ testptr(rbx, rbx);
3063 __ jcc(Assembler::zero, no_such_method);
3065 // do the call
3066 // rcx: receiver
3067 // rbx,: Method*
3068 __ jump_from_interpreted(rbx, rdx);
3069 __ should_not_reach_here();
3071 // exception handling code follows...
3072 // note: must restore interpreter registers to canonical
3073 // state for exception handling to work correctly!
3075 __ bind(no_such_method);
3076 // throw exception
3077 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3078 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3079 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3080 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3081 // the call_VM checks for exception, so we should never return here.
3082 __ should_not_reach_here();
3084 __ bind(no_such_interface);
3085 // throw exception
3086 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3087 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3088 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3089 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3090 InterpreterRuntime::throw_IncompatibleClassChangeError));
3091 // the call_VM checks for exception, so we should never return here.
3092 __ should_not_reach_here();
3093 }
3095 void TemplateTable::invokehandle(int byte_no) {
3096 transition(vtos, vtos);
3097 assert(byte_no == f1_byte, "use this argument");
3098 const Register rbx_method = rbx;
3099 const Register rax_mtype = rax;
3100 const Register rcx_recv = rcx;
3101 const Register rdx_flags = rdx;
3103 if (!EnableInvokeDynamic) {
3104 // rewriter does not generate this bytecode
3105 __ should_not_reach_here();
3106 return;
3107 }
3109 prepare_invoke(byte_no, rbx_method, rax_mtype, rcx_recv);
3110 __ verify_method_ptr(rbx_method);
3111 __ verify_oop(rcx_recv);
3112 __ null_check(rcx_recv);
3114 // rax: MethodType object (from cpool->resolved_references[f1], if necessary)
3115 // rbx: MH.invokeExact_MT method (from f2)
3117 // Note: rax_mtype is already pushed (if necessary) by prepare_invoke
3119 // FIXME: profile the LambdaForm also
3120 __ profile_final_call(rax);
3122 __ jump_from_interpreted(rbx_method, rdx);
3123 }
3126 void TemplateTable::invokedynamic(int byte_no) {
3127 transition(vtos, vtos);
3128 assert(byte_no == f1_byte, "use this argument");
3130 if (!EnableInvokeDynamic) {
3131 // We should not encounter this bytecode if !EnableInvokeDynamic.
3132 // The verifier will stop it. However, if we get past the verifier,
3133 // this will stop the thread in a reasonable way, without crashing the JVM.
3134 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3135 InterpreterRuntime::throw_IncompatibleClassChangeError));
3136 // the call_VM checks for exception, so we should never return here.
3137 __ should_not_reach_here();
3138 return;
3139 }
3141 const Register rbx_method = rbx;
3142 const Register rax_callsite = rax;
3144 prepare_invoke(byte_no, rbx_method, rax_callsite);
3146 // rax: CallSite object (from cpool->resolved_references[f1])
3147 // rbx: MH.linkToCallSite method (from f2)
3149 // Note: rax_callsite is already pushed by prepare_invoke
3151 // %%% should make a type profile for any invokedynamic that takes a ref argument
3152 // profile this call
3153 __ profile_call(rsi);
3155 __ verify_oop(rax_callsite);
3157 __ jump_from_interpreted(rbx_method, rdx);
3158 }
3160 //----------------------------------------------------------------------------------------------------
3161 // Allocation
3163 void TemplateTable::_new() {
3164 transition(vtos, atos);
3165 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3166 Label slow_case;
3167 Label slow_case_no_pop;
3168 Label done;
3169 Label initialize_header;
3170 Label initialize_object; // including clearing the fields
3171 Label allocate_shared;
3173 __ get_cpool_and_tags(rcx, rax);
3175 // Make sure the class we're about to instantiate has been resolved.
3176 // This is done before loading InstanceKlass to be consistent with the order
3177 // how Constant Pool is updated (see ConstantPool::klass_at_put)
3178 const int tags_offset = Array<u1>::base_offset_in_bytes();
3179 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3180 __ jcc(Assembler::notEqual, slow_case_no_pop);
3182 // get InstanceKlass
3183 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(ConstantPool)));
3184 __ push(rcx); // save the contexts of klass for initializing the header
3186 // make sure klass is initialized & doesn't have finalizer
3187 // make sure klass is fully initialized
3188 __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
3189 __ jcc(Assembler::notEqual, slow_case);
3191 // get instance_size in InstanceKlass (scaled to a count of bytes)
3192 __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
3193 // test to see if it has a finalizer or is malformed in some way
3194 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3195 __ jcc(Assembler::notZero, slow_case);
3197 //
3198 // Allocate the instance
3199 // 1) Try to allocate in the TLAB
3200 // 2) if fail and the object is large allocate in the shared Eden
3201 // 3) if the above fails (or is not applicable), go to a slow case
3202 // (creates a new TLAB, etc.)
3204 const bool allow_shared_alloc =
3205 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3207 const Register thread = rcx;
3208 if (UseTLAB || allow_shared_alloc) {
3209 __ get_thread(thread);
3210 }
3212 if (UseTLAB) {
3213 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3214 __ lea(rbx, Address(rax, rdx, Address::times_1));
3215 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3216 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3217 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3218 if (ZeroTLAB) {
3219 // the fields have been already cleared
3220 __ jmp(initialize_header);
3221 } else {
3222 // initialize both the header and fields
3223 __ jmp(initialize_object);
3224 }
3225 }
3227 // Allocation in the shared Eden, if allowed.
3228 //
3229 // rdx: instance size in bytes
3230 if (allow_shared_alloc) {
3231 __ bind(allocate_shared);
3233 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3235 Label retry;
3236 __ bind(retry);
3237 __ movptr(rax, heap_top);
3238 __ lea(rbx, Address(rax, rdx, Address::times_1));
3239 __ cmpptr(rbx, ExternalAddress((address)Universe::heap()->end_addr()));
3240 __ jcc(Assembler::above, slow_case);
3242 // Compare rax, with the top addr, and if still equal, store the new
3243 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3244 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3245 //
3246 // rax,: object begin
3247 // rbx,: object end
3248 // rdx: instance size in bytes
3249 __ locked_cmpxchgptr(rbx, heap_top);
3251 // if someone beat us on the allocation, try again, otherwise continue
3252 __ jcc(Assembler::notEqual, retry);
3254 __ incr_allocated_bytes(thread, rdx, 0);
3255 }
3257 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3258 // The object is initialized before the header. If the object size is
3259 // zero, go directly to the header initialization.
3260 __ bind(initialize_object);
3261 __ decrement(rdx, sizeof(oopDesc));
3262 __ jcc(Assembler::zero, initialize_header);
3264 // Initialize topmost object field, divide rdx by 8, check if odd and
3265 // test if zero.
3266 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3267 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3269 // rdx must have been multiple of 8
3270 #ifdef ASSERT
3271 // make sure rdx was multiple of 8
3272 Label L;
3273 // Ignore partial flag stall after shrl() since it is debug VM
3274 __ jccb(Assembler::carryClear, L);
3275 __ stop("object size is not multiple of 2 - adjust this code");
3276 __ bind(L);
3277 // rdx must be > 0, no extra check needed here
3278 #endif
3280 // initialize remaining object fields: rdx was a multiple of 8
3281 { Label loop;
3282 __ bind(loop);
3283 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3284 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3285 __ decrement(rdx);
3286 __ jcc(Assembler::notZero, loop);
3287 }
3289 // initialize object header only.
3290 __ bind(initialize_header);
3291 if (UseBiasedLocking) {
3292 __ pop(rcx); // get saved klass back in the register.
3293 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset()));
3294 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3295 } else {
3296 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3297 (int32_t)markOopDesc::prototype()); // header
3298 __ pop(rcx); // get saved klass back in the register.
3299 }
3300 __ store_klass(rax, rcx); // klass
3302 {
3303 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3304 // Trigger dtrace event for fastpath
3305 __ push(atos);
3306 __ call_VM_leaf(
3307 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3308 __ pop(atos);
3309 }
3311 __ jmp(done);
3312 }
3314 // slow case
3315 __ bind(slow_case);
3316 __ pop(rcx); // restore stack pointer to what it was when we came in.
3317 __ bind(slow_case_no_pop);
3318 __ get_constant_pool(rax);
3319 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3320 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx);
3322 // continue
3323 __ bind(done);
3324 }
3327 void TemplateTable::newarray() {
3328 transition(itos, atos);
3329 __ push_i(rax); // make sure everything is on the stack
3330 __ load_unsigned_byte(rdx, at_bcp(1));
3331 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax);
3332 __ pop_i(rdx); // discard size
3333 }
3336 void TemplateTable::anewarray() {
3337 transition(itos, atos);
3338 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3339 __ get_constant_pool(rcx);
3340 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax);
3341 }
3344 void TemplateTable::arraylength() {
3345 transition(atos, itos);
3346 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3347 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3348 }
3351 void TemplateTable::checkcast() {
3352 transition(atos, atos);
3353 Label done, is_null, ok_is_subtype, quicked, resolved;
3354 __ testptr(rax, rax); // Object is in EAX
3355 __ jcc(Assembler::zero, is_null);
3357 // Get cpool & tags index
3358 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3359 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3360 // See if bytecode has already been quicked
3361 __ cmpb(Address(rdx, rbx, Address::times_1, Array<u1>::base_offset_in_bytes()), JVM_CONSTANT_Class);
3362 __ jcc(Assembler::equal, quicked);
3364 __ push(atos);
3365 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3366 // vm_result_2 has metadata result
3367 // borrow rdi from locals
3368 __ get_thread(rdi);
3369 __ get_vm_result_2(rax, rdi);
3370 __ restore_locals();
3371 __ pop_ptr(rdx);
3372 __ jmpb(resolved);
3374 // Get superklass in EAX and subklass in EBX
3375 __ bind(quicked);
3376 __ mov(rdx, rax); // Save object in EDX; EAX needed for subtype check
3377 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(ConstantPool)));
3379 __ bind(resolved);
3380 __ load_klass(rbx, rdx);
3382 // Generate subtype check. Blows ECX. Resets EDI. Object in EDX.
3383 // Superklass in EAX. Subklass in EBX.
3384 __ gen_subtype_check( rbx, ok_is_subtype );
3386 // Come here on failure
3387 __ push(rdx);
3388 // object is at TOS
3389 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3391 // Come here on success
3392 __ bind(ok_is_subtype);
3393 __ mov(rax,rdx); // Restore object in EDX
3395 // Collect counts on whether this check-cast sees NULLs a lot or not.
3396 if (ProfileInterpreter) {
3397 __ jmp(done);
3398 __ bind(is_null);
3399 __ profile_null_seen(rcx);
3400 } else {
3401 __ bind(is_null); // same as 'done'
3402 }
3403 __ bind(done);
3404 }
3407 void TemplateTable::instanceof() {
3408 transition(atos, itos);
3409 Label done, is_null, ok_is_subtype, quicked, resolved;
3410 __ testptr(rax, rax);
3411 __ jcc(Assembler::zero, is_null);
3413 // Get cpool & tags index
3414 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3415 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3416 // See if bytecode has already been quicked
3417 __ cmpb(Address(rdx, rbx, Address::times_1, Array<u1>::base_offset_in_bytes()), JVM_CONSTANT_Class);
3418 __ jcc(Assembler::equal, quicked);
3420 __ push(atos);
3421 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3422 // vm_result_2 has metadata result
3423 // borrow rdi from locals
3424 __ get_thread(rdi);
3425 __ get_vm_result_2(rax, rdi);
3426 __ restore_locals();
3427 __ pop_ptr(rdx);
3428 __ load_klass(rdx, rdx);
3429 __ jmp(resolved);
3431 // Get superklass in EAX and subklass in EDX
3432 __ bind(quicked);
3433 __ load_klass(rdx, rax);
3434 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(ConstantPool)));
3436 __ bind(resolved);
3438 // Generate subtype check. Blows ECX. Resets EDI.
3439 // Superklass in EAX. Subklass in EDX.
3440 __ gen_subtype_check( rdx, ok_is_subtype );
3442 // Come here on failure
3443 __ xorl(rax,rax);
3444 __ jmpb(done);
3445 // Come here on success
3446 __ bind(ok_is_subtype);
3447 __ movl(rax, 1);
3449 // Collect counts on whether this test sees NULLs a lot or not.
3450 if (ProfileInterpreter) {
3451 __ jmp(done);
3452 __ bind(is_null);
3453 __ profile_null_seen(rcx);
3454 } else {
3455 __ bind(is_null); // same as 'done'
3456 }
3457 __ bind(done);
3458 // rax, = 0: obj == NULL or obj is not an instanceof the specified klass
3459 // rax, = 1: obj != NULL and obj is an instanceof the specified klass
3460 }
3463 //----------------------------------------------------------------------------------------------------
3464 // Breakpoints
3465 void TemplateTable::_breakpoint() {
3467 // Note: We get here even if we are single stepping..
3468 // jbug inists on setting breakpoints at every bytecode
3469 // even if we are in single step mode.
3471 transition(vtos, vtos);
3473 // get the unpatched byte code
3474 __ get_method(rcx);
3475 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi);
3476 __ mov(rbx, rax);
3478 // post the breakpoint event
3479 __ get_method(rcx);
3480 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi);
3482 // complete the execution of original bytecode
3483 __ dispatch_only_normal(vtos);
3484 }
3487 //----------------------------------------------------------------------------------------------------
3488 // Exceptions
3490 void TemplateTable::athrow() {
3491 transition(atos, vtos);
3492 __ null_check(rax);
3493 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3494 }
3497 //----------------------------------------------------------------------------------------------------
3498 // Synchronization
3499 //
3500 // Note: monitorenter & exit are symmetric routines; which is reflected
3501 // in the assembly code structure as well
3502 //
3503 // Stack layout:
3504 //
3505 // [expressions ] <--- rsp = expression stack top
3506 // ..
3507 // [expressions ]
3508 // [monitor entry] <--- monitor block top = expression stack bot
3509 // ..
3510 // [monitor entry]
3511 // [frame data ] <--- monitor block bot
3512 // ...
3513 // [saved rbp, ] <--- rbp,
3516 void TemplateTable::monitorenter() {
3517 transition(atos, vtos);
3519 // check for NULL object
3520 __ null_check(rax);
3522 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3523 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3524 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3525 Label allocated;
3527 // initialize entry pointer
3528 __ xorl(rdx, rdx); // points to free slot or NULL
3530 // find a free slot in the monitor block (result in rdx)
3531 { Label entry, loop, exit;
3532 __ movptr(rcx, monitor_block_top); // points to current entry, starting with top-most entry
3534 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3535 __ jmpb(entry);
3537 __ bind(loop);
3538 __ cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); // check if current entry is used
3539 __ cmovptr(Assembler::equal, rdx, rcx); // if not used then remember entry in rdx
3540 __ cmpptr(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3541 __ jccb(Assembler::equal, exit); // if same object then stop searching
3542 __ addptr(rcx, entry_size); // otherwise advance to next entry
3543 __ bind(entry);
3544 __ cmpptr(rcx, rbx); // check if bottom reached
3545 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3546 __ bind(exit);
3547 }
3549 __ testptr(rdx, rdx); // check if a slot has been found
3550 __ jccb(Assembler::notZero, allocated); // if found, continue with that one
3552 // allocate one if there's no free slot
3553 { Label entry, loop;
3554 // 1. compute new pointers // rsp: old expression stack top
3555 __ movptr(rdx, monitor_block_bot); // rdx: old expression stack bottom
3556 __ subptr(rsp, entry_size); // move expression stack top
3557 __ subptr(rdx, entry_size); // move expression stack bottom
3558 __ mov(rcx, rsp); // set start value for copy loop
3559 __ movptr(monitor_block_bot, rdx); // set new monitor block top
3560 __ jmp(entry);
3561 // 2. move expression stack contents
3562 __ bind(loop);
3563 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location
3564 __ movptr(Address(rcx, 0), rbx); // and store it at new location
3565 __ addptr(rcx, wordSize); // advance to next word
3566 __ bind(entry);
3567 __ cmpptr(rcx, rdx); // check if bottom reached
3568 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word
3569 }
3571 // call run-time routine
3572 // rdx: points to monitor entry
3573 __ bind(allocated);
3575 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3576 // The object has already been poped from the stack, so the expression stack looks correct.
3577 __ increment(rsi);
3579 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
3580 __ lock_object(rdx);
3582 // check to make sure this monitor doesn't cause stack overflow after locking
3583 __ save_bcp(); // in case of exception
3584 __ generate_stack_overflow_check(0);
3586 // The bcp has already been incremented. Just need to dispatch to next instruction.
3587 __ dispatch_next(vtos);
3588 }
3591 void TemplateTable::monitorexit() {
3592 transition(atos, vtos);
3594 // check for NULL object
3595 __ null_check(rax);
3597 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3598 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3599 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3600 Label found;
3602 // find matching slot
3603 { Label entry, loop;
3604 __ movptr(rdx, monitor_block_top); // points to current entry, starting with top-most entry
3605 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3606 __ jmpb(entry);
3608 __ bind(loop);
3609 __ cmpptr(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3610 __ jcc(Assembler::equal, found); // if same object then stop searching
3611 __ addptr(rdx, entry_size); // otherwise advance to next entry
3612 __ bind(entry);
3613 __ cmpptr(rdx, rbx); // check if bottom reached
3614 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3615 }
3617 // error handling. Unlocking was not block-structured
3618 Label end;
3619 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3620 __ should_not_reach_here();
3622 // call run-time routine
3623 // rcx: points to monitor entry
3624 __ bind(found);
3625 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3626 __ unlock_object(rdx);
3627 __ pop_ptr(rax); // discard object
3628 __ bind(end);
3629 }
3632 //----------------------------------------------------------------------------------------------------
3633 // Wide instructions
3635 void TemplateTable::wide() {
3636 transition(vtos, vtos);
3637 __ load_unsigned_byte(rbx, at_bcp(1));
3638 ExternalAddress wtable((address)Interpreter::_wentry_point);
3639 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
3640 // Note: the rsi increment step is part of the individual wide bytecode implementations
3641 }
3644 //----------------------------------------------------------------------------------------------------
3645 // Multi arrays
3647 void TemplateTable::multianewarray() {
3648 transition(vtos, atos);
3649 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3650 // last dim is on top of stack; we want address of first one:
3651 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
3652 // the latter wordSize to point to the beginning of the array.
3653 __ lea( rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
3654 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax); // pass in rax,
3655 __ load_unsigned_byte(rbx, at_bcp(3));
3656 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
3657 }
3659 #endif /* !CC_INTERP */