Mercurial > jdk8-mips64-public > hotspot / file revision
src/cpu/x86/vm/templateTable_x86_32.cpp@9ee9cf798b59
src/cpu/x86/vm/templateTable_x86_32.cpp
Thu, 02 Oct 2008 19:58:19 -0700
- author
- xdono
- date
- Thu, 02 Oct 2008 19:58:19 -0700
- changeset 772
- 9ee9cf798b59
- parent 739
- dc7f315e41f7
- child 797
- f8199438385b
- permissions
- -rw-r--r--
6754988: Update copyright year
Summary: Update for files that have been modified starting July 2008
Reviewed-by: ohair, tbell
1 /*
2 * Copyright 1997-2008 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
25 #include "incls/_precompiled.incl"
26 #include "incls/_templateTable_x86_32.cpp.incl"
28 #ifndef CC_INTERP
29 #define __ _masm->
31 //----------------------------------------------------------------------------------------------------
32 // Platform-dependent initialization
34 void TemplateTable::pd_initialize() {
35 // No i486 specific initialization
36 }
38 //----------------------------------------------------------------------------------------------------
39 // Address computation
41 // local variables
42 static inline Address iaddress(int n) {
43 return Address(rdi, Interpreter::local_offset_in_bytes(n));
44 }
46 static inline Address laddress(int n) { return iaddress(n + 1); }
47 static inline Address haddress(int n) { return iaddress(n + 0); }
48 static inline Address faddress(int n) { return iaddress(n); }
49 static inline Address daddress(int n) { return laddress(n); }
50 static inline Address aaddress(int n) { return iaddress(n); }
52 static inline Address iaddress(Register r) {
53 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::value_offset_in_bytes());
54 }
55 static inline Address laddress(Register r) {
56 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(1));
57 }
58 static inline Address haddress(Register r) {
59 return Address(rdi, r, Interpreter::stackElementScale(), Interpreter::local_offset_in_bytes(0));
60 }
62 static inline Address faddress(Register r) { return iaddress(r); };
63 static inline Address daddress(Register r) {
64 assert(!TaggedStackInterpreter, "This doesn't work");
65 return laddress(r);
66 };
67 static inline Address aaddress(Register r) { return iaddress(r); };
69 // expression stack
70 // (Note: Must not use symmetric equivalents at_rsp_m1/2 since they store
71 // data beyond the rsp which is potentially unsafe in an MT environment;
72 // an interrupt may overwrite that data.)
73 static inline Address at_rsp () {
74 return Address(rsp, 0);
75 }
77 // At top of Java expression stack which may be different than rsp(). It
78 // isn't for category 1 objects.
79 static inline Address at_tos () {
80 Address tos = Address(rsp, Interpreter::expr_offset_in_bytes(0));
81 return tos;
82 }
84 static inline Address at_tos_p1() {
85 return Address(rsp, Interpreter::expr_offset_in_bytes(1));
86 }
88 static inline Address at_tos_p2() {
89 return Address(rsp, Interpreter::expr_offset_in_bytes(2));
90 }
92 // Condition conversion
93 static Assembler::Condition j_not(TemplateTable::Condition cc) {
94 switch (cc) {
95 case TemplateTable::equal : return Assembler::notEqual;
96 case TemplateTable::not_equal : return Assembler::equal;
97 case TemplateTable::less : return Assembler::greaterEqual;
98 case TemplateTable::less_equal : return Assembler::greater;
99 case TemplateTable::greater : return Assembler::lessEqual;
100 case TemplateTable::greater_equal: return Assembler::less;
101 }
102 ShouldNotReachHere();
103 return Assembler::zero;
104 }
107 //----------------------------------------------------------------------------------------------------
108 // Miscelaneous helper routines
110 Address TemplateTable::at_bcp(int offset) {
111 assert(_desc->uses_bcp(), "inconsistent uses_bcp information");
112 return Address(rsi, offset);
113 }
116 void TemplateTable::patch_bytecode(Bytecodes::Code bytecode, Register bc,
117 Register scratch,
118 bool load_bc_into_scratch/*=true*/) {
120 if (!RewriteBytecodes) return;
121 // the pair bytecodes have already done the load.
122 if (load_bc_into_scratch) {
123 __ movl(bc, bytecode);
124 }
125 Label patch_done;
126 if (JvmtiExport::can_post_breakpoint()) {
127 Label fast_patch;
128 // if a breakpoint is present we can't rewrite the stream directly
129 __ movzbl(scratch, at_bcp(0));
130 __ cmpl(scratch, Bytecodes::_breakpoint);
131 __ jcc(Assembler::notEqual, fast_patch);
132 __ get_method(scratch);
133 // Let breakpoint table handling rewrite to quicker bytecode
134 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::set_original_bytecode_at), scratch, rsi, bc);
135 #ifndef ASSERT
136 __ jmpb(patch_done);
137 __ bind(fast_patch);
138 }
139 #else
140 __ jmp(patch_done);
141 __ bind(fast_patch);
142 }
143 Label okay;
144 __ load_unsigned_byte(scratch, at_bcp(0));
145 __ cmpl(scratch, (int)Bytecodes::java_code(bytecode));
146 __ jccb(Assembler::equal, okay);
147 __ cmpl(scratch, bc);
148 __ jcc(Assembler::equal, okay);
149 __ stop("patching the wrong bytecode");
150 __ bind(okay);
151 #endif
152 // patch bytecode
153 __ movb(at_bcp(0), bc);
154 __ bind(patch_done);
155 }
157 //----------------------------------------------------------------------------------------------------
158 // Individual instructions
160 void TemplateTable::nop() {
161 transition(vtos, vtos);
162 // nothing to do
163 }
165 void TemplateTable::shouldnotreachhere() {
166 transition(vtos, vtos);
167 __ stop("shouldnotreachhere bytecode");
168 }
172 void TemplateTable::aconst_null() {
173 transition(vtos, atos);
174 __ xorptr(rax, rax);
175 }
178 void TemplateTable::iconst(int value) {
179 transition(vtos, itos);
180 if (value == 0) {
181 __ xorptr(rax, rax);
182 } else {
183 __ movptr(rax, value);
184 }
185 }
188 void TemplateTable::lconst(int value) {
189 transition(vtos, ltos);
190 if (value == 0) {
191 __ xorptr(rax, rax);
192 } else {
193 __ movptr(rax, value);
194 }
195 assert(value >= 0, "check this code");
196 __ xorptr(rdx, rdx);
197 }
200 void TemplateTable::fconst(int value) {
201 transition(vtos, ftos);
202 if (value == 0) { __ fldz();
203 } else if (value == 1) { __ fld1();
204 } else if (value == 2) { __ fld1(); __ fld1(); __ faddp(); // should do a better solution here
205 } else { ShouldNotReachHere();
206 }
207 }
210 void TemplateTable::dconst(int value) {
211 transition(vtos, dtos);
212 if (value == 0) { __ fldz();
213 } else if (value == 1) { __ fld1();
214 } else { ShouldNotReachHere();
215 }
216 }
219 void TemplateTable::bipush() {
220 transition(vtos, itos);
221 __ load_signed_byte(rax, at_bcp(1));
222 }
225 void TemplateTable::sipush() {
226 transition(vtos, itos);
227 __ load_unsigned_word(rax, at_bcp(1));
228 __ bswapl(rax);
229 __ sarl(rax, 16);
230 }
232 void TemplateTable::ldc(bool wide) {
233 transition(vtos, vtos);
234 Label call_ldc, notFloat, notClass, Done;
236 if (wide) {
237 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
238 } else {
239 __ load_unsigned_byte(rbx, at_bcp(1));
240 }
241 __ get_cpool_and_tags(rcx, rax);
242 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
243 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
245 // get type
246 __ xorptr(rdx, rdx);
247 __ movb(rdx, Address(rax, rbx, Address::times_1, tags_offset));
249 // unresolved string - get the resolved string
250 __ cmpl(rdx, JVM_CONSTANT_UnresolvedString);
251 __ jccb(Assembler::equal, call_ldc);
253 // unresolved class - get the resolved class
254 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClass);
255 __ jccb(Assembler::equal, call_ldc);
257 // unresolved class in error (resolution failed) - call into runtime
258 // so that the same error from first resolution attempt is thrown.
259 __ cmpl(rdx, JVM_CONSTANT_UnresolvedClassInError);
260 __ jccb(Assembler::equal, call_ldc);
262 // resolved class - need to call vm to get java mirror of the class
263 __ cmpl(rdx, JVM_CONSTANT_Class);
264 __ jcc(Assembler::notEqual, notClass);
266 __ bind(call_ldc);
267 __ movl(rcx, wide);
268 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::ldc), rcx);
269 __ push(atos);
270 __ jmp(Done);
272 __ bind(notClass);
273 __ cmpl(rdx, JVM_CONSTANT_Float);
274 __ jccb(Assembler::notEqual, notFloat);
275 // ftos
276 __ fld_s( Address(rcx, rbx, Address::times_ptr, base_offset));
277 __ push(ftos);
278 __ jmp(Done);
280 __ bind(notFloat);
281 #ifdef ASSERT
282 { Label L;
283 __ cmpl(rdx, JVM_CONSTANT_Integer);
284 __ jcc(Assembler::equal, L);
285 __ cmpl(rdx, JVM_CONSTANT_String);
286 __ jcc(Assembler::equal, L);
287 __ stop("unexpected tag type in ldc");
288 __ bind(L);
289 }
290 #endif
291 Label isOop;
292 // atos and itos
293 // String is only oop type we will see here
294 __ cmpl(rdx, JVM_CONSTANT_String);
295 __ jccb(Assembler::equal, isOop);
296 __ movl(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
297 __ push(itos);
298 __ jmp(Done);
299 __ bind(isOop);
300 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset));
301 __ push(atos);
303 if (VerifyOops) {
304 __ verify_oop(rax);
305 }
306 __ bind(Done);
307 }
309 void TemplateTable::ldc2_w() {
310 transition(vtos, vtos);
311 Label Long, Done;
312 __ get_unsigned_2_byte_index_at_bcp(rbx, 1);
314 __ get_cpool_and_tags(rcx, rax);
315 const int base_offset = constantPoolOopDesc::header_size() * wordSize;
316 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
318 // get type
319 __ cmpb(Address(rax, rbx, Address::times_1, tags_offset), JVM_CONSTANT_Double);
320 __ jccb(Assembler::notEqual, Long);
321 // dtos
322 __ fld_d( Address(rcx, rbx, Address::times_ptr, base_offset));
323 __ push(dtos);
324 __ jmpb(Done);
326 __ bind(Long);
327 // ltos
328 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, base_offset + 0 * wordSize));
329 NOT_LP64(__ movptr(rdx, Address(rcx, rbx, Address::times_ptr, base_offset + 1 * wordSize)));
331 __ push(ltos);
333 __ bind(Done);
334 }
337 void TemplateTable::locals_index(Register reg, int offset) {
338 __ load_unsigned_byte(reg, at_bcp(offset));
339 __ negptr(reg);
340 }
343 void TemplateTable::iload() {
344 transition(vtos, itos);
345 if (RewriteFrequentPairs) {
346 Label rewrite, done;
348 // get next byte
349 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_iload)));
350 // if _iload, wait to rewrite to iload2. We only want to rewrite the
351 // last two iloads in a pair. Comparing against fast_iload means that
352 // the next bytecode is neither an iload or a caload, and therefore
353 // an iload pair.
354 __ cmpl(rbx, Bytecodes::_iload);
355 __ jcc(Assembler::equal, done);
357 __ cmpl(rbx, Bytecodes::_fast_iload);
358 __ movl(rcx, Bytecodes::_fast_iload2);
359 __ jccb(Assembler::equal, rewrite);
361 // if _caload, rewrite to fast_icaload
362 __ cmpl(rbx, Bytecodes::_caload);
363 __ movl(rcx, Bytecodes::_fast_icaload);
364 __ jccb(Assembler::equal, rewrite);
366 // rewrite so iload doesn't check again.
367 __ movl(rcx, Bytecodes::_fast_iload);
369 // rewrite
370 // rcx: fast bytecode
371 __ bind(rewrite);
372 patch_bytecode(Bytecodes::_iload, rcx, rbx, false);
373 __ bind(done);
374 }
376 // Get the local value into tos
377 locals_index(rbx);
378 __ movl(rax, iaddress(rbx));
379 debug_only(__ verify_local_tag(frame::TagValue, rbx));
380 }
383 void TemplateTable::fast_iload2() {
384 transition(vtos, itos);
385 locals_index(rbx);
386 __ movl(rax, iaddress(rbx));
387 debug_only(__ verify_local_tag(frame::TagValue, rbx));
388 __ push(itos);
389 locals_index(rbx, 3);
390 __ movl(rax, iaddress(rbx));
391 debug_only(__ verify_local_tag(frame::TagValue, rbx));
392 }
394 void TemplateTable::fast_iload() {
395 transition(vtos, itos);
396 locals_index(rbx);
397 __ movl(rax, iaddress(rbx));
398 debug_only(__ verify_local_tag(frame::TagValue, rbx));
399 }
402 void TemplateTable::lload() {
403 transition(vtos, ltos);
404 locals_index(rbx);
405 __ movptr(rax, laddress(rbx));
406 NOT_LP64(__ movl(rdx, haddress(rbx)));
407 debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
408 }
411 void TemplateTable::fload() {
412 transition(vtos, ftos);
413 locals_index(rbx);
414 __ fld_s(faddress(rbx));
415 debug_only(__ verify_local_tag(frame::TagValue, rbx));
416 }
419 void TemplateTable::dload() {
420 transition(vtos, dtos);
421 locals_index(rbx);
422 if (TaggedStackInterpreter) {
423 // Get double out of locals array, onto temp stack and load with
424 // float instruction into ST0
425 __ movl(rax, laddress(rbx));
426 __ movl(rdx, haddress(rbx));
427 __ push(rdx); // push hi first
428 __ push(rax);
429 __ fld_d(Address(rsp, 0));
430 __ addptr(rsp, 2*wordSize);
431 debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
432 } else {
433 __ fld_d(daddress(rbx));
434 }
435 }
438 void TemplateTable::aload() {
439 transition(vtos, atos);
440 locals_index(rbx);
441 __ movptr(rax, aaddress(rbx));
442 debug_only(__ verify_local_tag(frame::TagReference, rbx));
443 }
446 void TemplateTable::locals_index_wide(Register reg) {
447 __ movl(reg, at_bcp(2));
448 __ bswapl(reg);
449 __ shrl(reg, 16);
450 __ negptr(reg);
451 }
454 void TemplateTable::wide_iload() {
455 transition(vtos, itos);
456 locals_index_wide(rbx);
457 __ movl(rax, iaddress(rbx));
458 debug_only(__ verify_local_tag(frame::TagValue, rbx));
459 }
462 void TemplateTable::wide_lload() {
463 transition(vtos, ltos);
464 locals_index_wide(rbx);
465 __ movptr(rax, laddress(rbx));
466 NOT_LP64(__ movl(rdx, haddress(rbx)));
467 debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
468 }
471 void TemplateTable::wide_fload() {
472 transition(vtos, ftos);
473 locals_index_wide(rbx);
474 __ fld_s(faddress(rbx));
475 debug_only(__ verify_local_tag(frame::TagValue, rbx));
476 }
479 void TemplateTable::wide_dload() {
480 transition(vtos, dtos);
481 locals_index_wide(rbx);
482 if (TaggedStackInterpreter) {
483 // Get double out of locals array, onto temp stack and load with
484 // float instruction into ST0
485 __ movl(rax, laddress(rbx));
486 __ movl(rdx, haddress(rbx));
487 __ push(rdx); // push hi first
488 __ push(rax);
489 __ fld_d(Address(rsp, 0));
490 __ addl(rsp, 2*wordSize);
491 debug_only(__ verify_local_tag(frame::TagCategory2, rbx));
492 } else {
493 __ fld_d(daddress(rbx));
494 }
495 }
498 void TemplateTable::wide_aload() {
499 transition(vtos, atos);
500 locals_index_wide(rbx);
501 __ movptr(rax, aaddress(rbx));
502 debug_only(__ verify_local_tag(frame::TagReference, rbx));
503 }
505 void TemplateTable::index_check(Register array, Register index) {
506 // Pop ptr into array
507 __ pop_ptr(array);
508 index_check_without_pop(array, index);
509 }
511 void TemplateTable::index_check_without_pop(Register array, Register index) {
512 // destroys rbx,
513 // check array
514 __ null_check(array, arrayOopDesc::length_offset_in_bytes());
515 LP64_ONLY(__ movslq(index, index));
516 // check index
517 __ cmpl(index, Address(array, arrayOopDesc::length_offset_in_bytes()));
518 if (index != rbx) {
519 // ??? convention: move aberrant index into rbx, for exception message
520 assert(rbx != array, "different registers");
521 __ mov(rbx, index);
522 }
523 __ jump_cc(Assembler::aboveEqual,
524 ExternalAddress(Interpreter::_throw_ArrayIndexOutOfBoundsException_entry));
525 }
528 void TemplateTable::iaload() {
529 transition(itos, itos);
530 // rdx: array
531 index_check(rdx, rax); // kills rbx,
532 // rax,: index
533 __ movl(rax, Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)));
534 }
537 void TemplateTable::laload() {
538 transition(itos, ltos);
539 // rax,: index
540 // rdx: array
541 index_check(rdx, rax);
542 __ mov(rbx, rax);
543 // rbx,: index
544 __ movptr(rax, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize));
545 NOT_LP64(__ movl(rdx, Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize)));
546 }
549 void TemplateTable::faload() {
550 transition(itos, ftos);
551 // rdx: array
552 index_check(rdx, rax); // kills rbx,
553 // rax,: index
554 __ fld_s(Address(rdx, rax, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
555 }
558 void TemplateTable::daload() {
559 transition(itos, dtos);
560 // rdx: array
561 index_check(rdx, rax); // kills rbx,
562 // rax,: index
563 __ fld_d(Address(rdx, rax, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
564 }
567 void TemplateTable::aaload() {
568 transition(itos, atos);
569 // rdx: array
570 index_check(rdx, rax); // kills rbx,
571 // rax,: index
572 __ movptr(rax, Address(rdx, rax, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
573 }
576 void TemplateTable::baload() {
577 transition(itos, itos);
578 // rdx: array
579 index_check(rdx, rax); // kills rbx,
580 // rax,: index
581 // can do better code for P5 - fix this at some point
582 __ load_signed_byte(rbx, Address(rdx, rax, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)));
583 __ mov(rax, rbx);
584 }
587 void TemplateTable::caload() {
588 transition(itos, itos);
589 // rdx: array
590 index_check(rdx, rax); // kills rbx,
591 // rax,: index
592 // can do better code for P5 - may want to improve this at some point
593 __ load_unsigned_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
594 __ mov(rax, rbx);
595 }
597 // iload followed by caload frequent pair
598 void TemplateTable::fast_icaload() {
599 transition(vtos, itos);
600 // load index out of locals
601 locals_index(rbx);
602 __ movl(rax, iaddress(rbx));
603 debug_only(__ verify_local_tag(frame::TagValue, rbx));
605 // rdx: array
606 index_check(rdx, rax);
607 // rax,: index
608 __ load_unsigned_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)));
609 __ mov(rax, rbx);
610 }
612 void TemplateTable::saload() {
613 transition(itos, itos);
614 // rdx: array
615 index_check(rdx, rax); // kills rbx,
616 // rax,: index
617 // can do better code for P5 - may want to improve this at some point
618 __ load_signed_word(rbx, Address(rdx, rax, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_SHORT)));
619 __ mov(rax, rbx);
620 }
623 void TemplateTable::iload(int n) {
624 transition(vtos, itos);
625 __ movl(rax, iaddress(n));
626 debug_only(__ verify_local_tag(frame::TagValue, n));
627 }
630 void TemplateTable::lload(int n) {
631 transition(vtos, ltos);
632 __ movptr(rax, laddress(n));
633 NOT_LP64(__ movptr(rdx, haddress(n)));
634 debug_only(__ verify_local_tag(frame::TagCategory2, n));
635 }
638 void TemplateTable::fload(int n) {
639 transition(vtos, ftos);
640 __ fld_s(faddress(n));
641 debug_only(__ verify_local_tag(frame::TagValue, n));
642 }
645 void TemplateTable::dload(int n) {
646 transition(vtos, dtos);
647 if (TaggedStackInterpreter) {
648 // Get double out of locals array, onto temp stack and load with
649 // float instruction into ST0
650 __ movl(rax, laddress(n));
651 __ movl(rdx, haddress(n));
652 __ push(rdx); // push hi first
653 __ push(rax);
654 __ fld_d(Address(rsp, 0));
655 __ addptr(rsp, 2*wordSize); // reset rsp
656 debug_only(__ verify_local_tag(frame::TagCategory2, n));
657 } else {
658 __ fld_d(daddress(n));
659 }
660 }
663 void TemplateTable::aload(int n) {
664 transition(vtos, atos);
665 __ movptr(rax, aaddress(n));
666 debug_only(__ verify_local_tag(frame::TagReference, n));
667 }
670 void TemplateTable::aload_0() {
671 transition(vtos, atos);
672 // According to bytecode histograms, the pairs:
673 //
674 // _aload_0, _fast_igetfield
675 // _aload_0, _fast_agetfield
676 // _aload_0, _fast_fgetfield
677 //
678 // occur frequently. If RewriteFrequentPairs is set, the (slow) _aload_0
679 // bytecode checks if the next bytecode is either _fast_igetfield,
680 // _fast_agetfield or _fast_fgetfield and then rewrites the
681 // current bytecode into a pair bytecode; otherwise it rewrites the current
682 // bytecode into _fast_aload_0 that doesn't do the pair check anymore.
683 //
684 // Note: If the next bytecode is _getfield, the rewrite must be delayed,
685 // otherwise we may miss an opportunity for a pair.
686 //
687 // Also rewrite frequent pairs
688 // aload_0, aload_1
689 // aload_0, iload_1
690 // These bytecodes with a small amount of code are most profitable to rewrite
691 if (RewriteFrequentPairs) {
692 Label rewrite, done;
693 // get next byte
694 __ load_unsigned_byte(rbx, at_bcp(Bytecodes::length_for(Bytecodes::_aload_0)));
696 // do actual aload_0
697 aload(0);
699 // if _getfield then wait with rewrite
700 __ cmpl(rbx, Bytecodes::_getfield);
701 __ jcc(Assembler::equal, done);
703 // if _igetfield then reqrite to _fast_iaccess_0
704 assert(Bytecodes::java_code(Bytecodes::_fast_iaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
705 __ cmpl(rbx, Bytecodes::_fast_igetfield);
706 __ movl(rcx, Bytecodes::_fast_iaccess_0);
707 __ jccb(Assembler::equal, rewrite);
709 // if _agetfield then reqrite to _fast_aaccess_0
710 assert(Bytecodes::java_code(Bytecodes::_fast_aaccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
711 __ cmpl(rbx, Bytecodes::_fast_agetfield);
712 __ movl(rcx, Bytecodes::_fast_aaccess_0);
713 __ jccb(Assembler::equal, rewrite);
715 // if _fgetfield then reqrite to _fast_faccess_0
716 assert(Bytecodes::java_code(Bytecodes::_fast_faccess_0) == Bytecodes::_aload_0, "fix bytecode definition");
717 __ cmpl(rbx, Bytecodes::_fast_fgetfield);
718 __ movl(rcx, Bytecodes::_fast_faccess_0);
719 __ jccb(Assembler::equal, rewrite);
721 // else rewrite to _fast_aload0
722 assert(Bytecodes::java_code(Bytecodes::_fast_aload_0) == Bytecodes::_aload_0, "fix bytecode definition");
723 __ movl(rcx, Bytecodes::_fast_aload_0);
725 // rewrite
726 // rcx: fast bytecode
727 __ bind(rewrite);
728 patch_bytecode(Bytecodes::_aload_0, rcx, rbx, false);
730 __ bind(done);
731 } else {
732 aload(0);
733 }
734 }
736 void TemplateTable::istore() {
737 transition(itos, vtos);
738 locals_index(rbx);
739 __ movl(iaddress(rbx), rax);
740 __ tag_local(frame::TagValue, rbx);
741 }
744 void TemplateTable::lstore() {
745 transition(ltos, vtos);
746 locals_index(rbx);
747 __ movptr(laddress(rbx), rax);
748 NOT_LP64(__ movptr(haddress(rbx), rdx));
749 __ tag_local(frame::TagCategory2, rbx);
750 }
753 void TemplateTable::fstore() {
754 transition(ftos, vtos);
755 locals_index(rbx);
756 __ fstp_s(faddress(rbx));
757 __ tag_local(frame::TagValue, rbx);
758 }
761 void TemplateTable::dstore() {
762 transition(dtos, vtos);
763 locals_index(rbx);
764 if (TaggedStackInterpreter) {
765 // Store double on stack and reload into locals nonadjacently
766 __ subptr(rsp, 2 * wordSize);
767 __ fstp_d(Address(rsp, 0));
768 __ pop(rax);
769 __ pop(rdx);
770 __ movptr(laddress(rbx), rax);
771 __ movptr(haddress(rbx), rdx);
772 __ tag_local(frame::TagCategory2, rbx);
773 } else {
774 __ fstp_d(daddress(rbx));
775 }
776 }
779 void TemplateTable::astore() {
780 transition(vtos, vtos);
781 __ pop_ptr(rax, rdx); // will need to pop tag too
782 locals_index(rbx);
783 __ movptr(aaddress(rbx), rax);
784 __ tag_local(rdx, rbx); // need to store same tag in local may be returnAddr
785 }
788 void TemplateTable::wide_istore() {
789 transition(vtos, vtos);
790 __ pop_i(rax);
791 locals_index_wide(rbx);
792 __ movl(iaddress(rbx), rax);
793 __ tag_local(frame::TagValue, rbx);
794 }
797 void TemplateTable::wide_lstore() {
798 transition(vtos, vtos);
799 __ pop_l(rax, rdx);
800 locals_index_wide(rbx);
801 __ movptr(laddress(rbx), rax);
802 NOT_LP64(__ movl(haddress(rbx), rdx));
803 __ tag_local(frame::TagCategory2, rbx);
804 }
807 void TemplateTable::wide_fstore() {
808 wide_istore();
809 }
812 void TemplateTable::wide_dstore() {
813 wide_lstore();
814 }
817 void TemplateTable::wide_astore() {
818 transition(vtos, vtos);
819 __ pop_ptr(rax, rdx);
820 locals_index_wide(rbx);
821 __ movptr(aaddress(rbx), rax);
822 __ tag_local(rdx, rbx);
823 }
826 void TemplateTable::iastore() {
827 transition(itos, vtos);
828 __ pop_i(rbx);
829 // rax,: value
830 // rdx: array
831 index_check(rdx, rbx); // prefer index in rbx,
832 // rbx,: index
833 __ movl(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_INT)), rax);
834 }
837 void TemplateTable::lastore() {
838 transition(ltos, vtos);
839 __ pop_i(rbx);
840 // rax,: low(value)
841 // rcx: array
842 // rdx: high(value)
843 index_check(rcx, rbx); // prefer index in rbx,
844 // rbx,: index
845 __ movptr(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 0 * wordSize), rax);
846 NOT_LP64(__ movl(Address(rcx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_LONG) + 1 * wordSize), rdx));
847 }
850 void TemplateTable::fastore() {
851 transition(ftos, vtos);
852 __ pop_i(rbx);
853 // rdx: array
854 // st0: value
855 index_check(rdx, rbx); // prefer index in rbx,
856 // rbx,: index
857 __ fstp_s(Address(rdx, rbx, Address::times_4, arrayOopDesc::base_offset_in_bytes(T_FLOAT)));
858 }
861 void TemplateTable::dastore() {
862 transition(dtos, vtos);
863 __ pop_i(rbx);
864 // rdx: array
865 // st0: value
866 index_check(rdx, rbx); // prefer index in rbx,
867 // rbx,: index
868 __ fstp_d(Address(rdx, rbx, Address::times_8, arrayOopDesc::base_offset_in_bytes(T_DOUBLE)));
869 }
872 void TemplateTable::aastore() {
873 Label is_null, ok_is_subtype, done;
874 transition(vtos, vtos);
875 // stack: ..., array, index, value
876 __ movptr(rax, at_tos()); // Value
877 __ movl(rcx, at_tos_p1()); // Index
878 __ movptr(rdx, at_tos_p2()); // Array
879 index_check_without_pop(rdx, rcx); // kills rbx,
880 // do array store check - check for NULL value first
881 __ testptr(rax, rax);
882 __ jcc(Assembler::zero, is_null);
884 // Move subklass into EBX
885 __ movptr(rbx, Address(rax, oopDesc::klass_offset_in_bytes()));
886 // Move superklass into EAX
887 __ movptr(rax, Address(rdx, oopDesc::klass_offset_in_bytes()));
888 __ movptr(rax, Address(rax, sizeof(oopDesc) + objArrayKlass::element_klass_offset_in_bytes()));
889 // Compress array+index*wordSize+12 into a single register. Frees ECX.
890 __ lea(rdx, Address(rdx, rcx, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
892 // Generate subtype check. Blows ECX. Resets EDI to locals.
893 // Superklass in EAX. Subklass in EBX.
894 __ gen_subtype_check( rbx, ok_is_subtype );
896 // Come here on failure
897 // object is at TOS
898 __ jump(ExternalAddress(Interpreter::_throw_ArrayStoreException_entry));
900 // Come here on success
901 __ bind(ok_is_subtype);
902 __ movptr(rax, at_rsp()); // Value
903 __ movptr(Address(rdx, 0), rax);
904 __ store_check(rdx);
905 __ jmpb(done);
907 // Have a NULL in EAX, EDX=array, ECX=index. Store NULL at ary[idx]
908 __ bind(is_null);
909 __ profile_null_seen(rbx);
910 __ movptr(Address(rdx, rcx, Address::times_ptr, arrayOopDesc::base_offset_in_bytes(T_OBJECT)), rax);
912 // Pop stack arguments
913 __ bind(done);
914 __ addptr(rsp, 3 * Interpreter::stackElementSize());
915 }
918 void TemplateTable::bastore() {
919 transition(itos, vtos);
920 __ pop_i(rbx);
921 // rax,: value
922 // rdx: array
923 index_check(rdx, rbx); // prefer index in rbx,
924 // rbx,: index
925 __ movb(Address(rdx, rbx, Address::times_1, arrayOopDesc::base_offset_in_bytes(T_BYTE)), rax);
926 }
929 void TemplateTable::castore() {
930 transition(itos, vtos);
931 __ pop_i(rbx);
932 // rax,: value
933 // rdx: array
934 index_check(rdx, rbx); // prefer index in rbx,
935 // rbx,: index
936 __ movw(Address(rdx, rbx, Address::times_2, arrayOopDesc::base_offset_in_bytes(T_CHAR)), rax);
937 }
940 void TemplateTable::sastore() {
941 castore();
942 }
945 void TemplateTable::istore(int n) {
946 transition(itos, vtos);
947 __ movl(iaddress(n), rax);
948 __ tag_local(frame::TagValue, n);
949 }
952 void TemplateTable::lstore(int n) {
953 transition(ltos, vtos);
954 __ movptr(laddress(n), rax);
955 NOT_LP64(__ movptr(haddress(n), rdx));
956 __ tag_local(frame::TagCategory2, n);
957 }
960 void TemplateTable::fstore(int n) {
961 transition(ftos, vtos);
962 __ fstp_s(faddress(n));
963 __ tag_local(frame::TagValue, n);
964 }
967 void TemplateTable::dstore(int n) {
968 transition(dtos, vtos);
969 if (TaggedStackInterpreter) {
970 __ subptr(rsp, 2 * wordSize);
971 __ fstp_d(Address(rsp, 0));
972 __ pop(rax);
973 __ pop(rdx);
974 __ movl(laddress(n), rax);
975 __ movl(haddress(n), rdx);
976 __ tag_local(frame::TagCategory2, n);
977 } else {
978 __ fstp_d(daddress(n));
979 }
980 }
983 void TemplateTable::astore(int n) {
984 transition(vtos, vtos);
985 __ pop_ptr(rax, rdx);
986 __ movptr(aaddress(n), rax);
987 __ tag_local(rdx, n);
988 }
991 void TemplateTable::pop() {
992 transition(vtos, vtos);
993 __ addptr(rsp, Interpreter::stackElementSize());
994 }
997 void TemplateTable::pop2() {
998 transition(vtos, vtos);
999 __ addptr(rsp, 2*Interpreter::stackElementSize());
1000 }
1003 void TemplateTable::dup() {
1004 transition(vtos, vtos);
1005 // stack: ..., a
1006 __ load_ptr_and_tag(0, rax, rdx);
1007 __ push_ptr(rax, rdx);
1008 // stack: ..., a, a
1009 }
1012 void TemplateTable::dup_x1() {
1013 transition(vtos, vtos);
1014 // stack: ..., a, b
1015 __ load_ptr_and_tag(0, rax, rdx); // load b
1016 __ load_ptr_and_tag(1, rcx, rbx); // load a
1017 __ store_ptr_and_tag(1, rax, rdx); // store b
1018 __ store_ptr_and_tag(0, rcx, rbx); // store a
1019 __ push_ptr(rax, rdx); // push b
1020 // stack: ..., b, a, b
1021 }
1024 void TemplateTable::dup_x2() {
1025 transition(vtos, vtos);
1026 // stack: ..., a, b, c
1027 __ load_ptr_and_tag(0, rax, rdx); // load c
1028 __ load_ptr_and_tag(2, rcx, rbx); // load a
1029 __ store_ptr_and_tag(2, rax, rdx); // store c in a
1030 __ push_ptr(rax, rdx); // push c
1031 // stack: ..., c, b, c, c
1032 __ load_ptr_and_tag(2, rax, rdx); // load b
1033 __ store_ptr_and_tag(2, rcx, rbx); // store a in b
1034 // stack: ..., c, a, c, c
1035 __ store_ptr_and_tag(1, rax, rdx); // store b in c
1036 // stack: ..., c, a, b, c
1037 }
1040 void TemplateTable::dup2() {
1041 transition(vtos, vtos);
1042 // stack: ..., a, b
1043 __ load_ptr_and_tag(1, rax, rdx); // load a
1044 __ push_ptr(rax, rdx); // push a
1045 __ load_ptr_and_tag(1, rax, rdx); // load b
1046 __ push_ptr(rax, rdx); // push b
1047 // stack: ..., a, b, a, b
1048 }
1051 void TemplateTable::dup2_x1() {
1052 transition(vtos, vtos);
1053 // stack: ..., a, b, c
1054 __ load_ptr_and_tag(0, rcx, rbx); // load c
1055 __ load_ptr_and_tag(1, rax, rdx); // load b
1056 __ push_ptr(rax, rdx); // push b
1057 __ push_ptr(rcx, rbx); // push c
1058 // stack: ..., a, b, c, b, c
1059 __ store_ptr_and_tag(3, rcx, rbx); // store c in b
1060 // stack: ..., a, c, c, b, c
1061 __ load_ptr_and_tag(4, rcx, rbx); // load a
1062 __ store_ptr_and_tag(2, rcx, rbx); // store a in 2nd c
1063 // stack: ..., a, c, a, b, c
1064 __ store_ptr_and_tag(4, rax, rdx); // store b in a
1065 // stack: ..., b, c, a, b, c
1066 // stack: ..., b, c, a, b, c
1067 }
1070 void TemplateTable::dup2_x2() {
1071 transition(vtos, vtos);
1072 // stack: ..., a, b, c, d
1073 __ load_ptr_and_tag(0, rcx, rbx); // load d
1074 __ load_ptr_and_tag(1, rax, rdx); // load c
1075 __ push_ptr(rax, rdx); // push c
1076 __ push_ptr(rcx, rbx); // push d
1077 // stack: ..., a, b, c, d, c, d
1078 __ load_ptr_and_tag(4, rax, rdx); // load b
1079 __ store_ptr_and_tag(2, rax, rdx); // store b in d
1080 __ store_ptr_and_tag(4, rcx, rbx); // store d in b
1081 // stack: ..., a, d, c, b, c, d
1082 __ load_ptr_and_tag(5, rcx, rbx); // load a
1083 __ load_ptr_and_tag(3, rax, rdx); // load c
1084 __ store_ptr_and_tag(3, rcx, rbx); // store a in c
1085 __ store_ptr_and_tag(5, rax, rdx); // store c in a
1086 // stack: ..., c, d, a, b, c, d
1087 // stack: ..., c, d, a, b, c, d
1088 }
1091 void TemplateTable::swap() {
1092 transition(vtos, vtos);
1093 // stack: ..., a, b
1094 __ load_ptr_and_tag(1, rcx, rbx); // load a
1095 __ load_ptr_and_tag(0, rax, rdx); // load b
1096 __ store_ptr_and_tag(0, rcx, rbx); // store a in b
1097 __ store_ptr_and_tag(1, rax, rdx); // store b in a
1098 // stack: ..., b, a
1099 }
1102 void TemplateTable::iop2(Operation op) {
1103 transition(itos, itos);
1104 switch (op) {
1105 case add : __ pop_i(rdx); __ addl (rax, rdx); break;
1106 case sub : __ mov(rdx, rax); __ pop_i(rax); __ subl (rax, rdx); break;
1107 case mul : __ pop_i(rdx); __ imull(rax, rdx); break;
1108 case _and : __ pop_i(rdx); __ andl (rax, rdx); break;
1109 case _or : __ pop_i(rdx); __ orl (rax, rdx); break;
1110 case _xor : __ pop_i(rdx); __ xorl (rax, rdx); break;
1111 case shl : __ mov(rcx, rax); __ pop_i(rax); __ shll (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1112 case shr : __ mov(rcx, rax); __ pop_i(rax); __ sarl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1113 case ushr : __ mov(rcx, rax); __ pop_i(rax); __ shrl (rax); break; // implicit masking of lower 5 bits by Intel shift instr.
1114 default : ShouldNotReachHere();
1115 }
1116 }
1119 void TemplateTable::lop2(Operation op) {
1120 transition(ltos, ltos);
1121 __ pop_l(rbx, rcx);
1122 switch (op) {
1123 case add : __ addl(rax, rbx); __ adcl(rdx, rcx); break;
1124 case sub : __ subl(rbx, rax); __ sbbl(rcx, rdx);
1125 __ mov(rax, rbx); __ mov(rdx, rcx); break;
1126 case _and: __ andl(rax, rbx); __ andl(rdx, rcx); break;
1127 case _or : __ orl (rax, rbx); __ orl (rdx, rcx); break;
1128 case _xor: __ xorl(rax, rbx); __ xorl(rdx, rcx); break;
1129 default : ShouldNotReachHere();
1130 }
1131 }
1134 void TemplateTable::idiv() {
1135 transition(itos, itos);
1136 __ mov(rcx, rax);
1137 __ pop_i(rax);
1138 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1139 // they are not equal, one could do a normal division (no correction
1140 // needed), which may speed up this implementation for the common case.
1141 // (see also JVM spec., p.243 & p.271)
1142 __ corrected_idivl(rcx);
1143 }
1146 void TemplateTable::irem() {
1147 transition(itos, itos);
1148 __ mov(rcx, rax);
1149 __ pop_i(rax);
1150 // Note: could xor rax, and rcx and compare with (-1 ^ min_int). If
1151 // they are not equal, one could do a normal division (no correction
1152 // needed), which may speed up this implementation for the common case.
1153 // (see also JVM spec., p.243 & p.271)
1154 __ corrected_idivl(rcx);
1155 __ mov(rax, rdx);
1156 }
1159 void TemplateTable::lmul() {
1160 transition(ltos, ltos);
1161 __ pop_l(rbx, rcx);
1162 __ push(rcx); __ push(rbx);
1163 __ push(rdx); __ push(rax);
1164 __ lmul(2 * wordSize, 0);
1165 __ addptr(rsp, 4 * wordSize); // take off temporaries
1166 }
1169 void TemplateTable::ldiv() {
1170 transition(ltos, ltos);
1171 __ pop_l(rbx, rcx);
1172 __ push(rcx); __ push(rbx);
1173 __ push(rdx); __ push(rax);
1174 // check if y = 0
1175 __ orl(rax, rdx);
1176 __ jump_cc(Assembler::zero,
1177 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1178 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::ldiv));
1179 __ addptr(rsp, 4 * wordSize); // take off temporaries
1180 }
1183 void TemplateTable::lrem() {
1184 transition(ltos, ltos);
1185 __ pop_l(rbx, rcx);
1186 __ push(rcx); __ push(rbx);
1187 __ push(rdx); __ push(rax);
1188 // check if y = 0
1189 __ orl(rax, rdx);
1190 __ jump_cc(Assembler::zero,
1191 ExternalAddress(Interpreter::_throw_ArithmeticException_entry));
1192 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::lrem));
1193 __ addptr(rsp, 4 * wordSize);
1194 }
1197 void TemplateTable::lshl() {
1198 transition(itos, ltos);
1199 __ movl(rcx, rax); // get shift count
1200 __ pop_l(rax, rdx); // get shift value
1201 __ lshl(rdx, rax);
1202 }
1205 void TemplateTable::lshr() {
1206 transition(itos, ltos);
1207 __ mov(rcx, rax); // get shift count
1208 __ pop_l(rax, rdx); // get shift value
1209 __ lshr(rdx, rax, true);
1210 }
1213 void TemplateTable::lushr() {
1214 transition(itos, ltos);
1215 __ mov(rcx, rax); // get shift count
1216 __ pop_l(rax, rdx); // get shift value
1217 __ lshr(rdx, rax);
1218 }
1221 void TemplateTable::fop2(Operation op) {
1222 transition(ftos, ftos);
1223 __ pop_ftos_to_rsp(); // pop ftos into rsp
1224 switch (op) {
1225 case add: __ fadd_s (at_rsp()); break;
1226 case sub: __ fsubr_s(at_rsp()); break;
1227 case mul: __ fmul_s (at_rsp()); break;
1228 case div: __ fdivr_s(at_rsp()); break;
1229 case rem: __ fld_s (at_rsp()); __ fremr(rax); break;
1230 default : ShouldNotReachHere();
1231 }
1232 __ f2ieee();
1233 __ pop(rax); // pop float thing off
1234 }
1237 void TemplateTable::dop2(Operation op) {
1238 transition(dtos, dtos);
1239 __ pop_dtos_to_rsp(); // pop dtos into rsp
1241 switch (op) {
1242 case add: __ fadd_d (at_rsp()); break;
1243 case sub: __ fsubr_d(at_rsp()); break;
1244 case mul: {
1245 Label L_strict;
1246 Label L_join;
1247 const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1248 __ get_method(rcx);
1249 __ movl(rcx, access_flags);
1250 __ testl(rcx, JVM_ACC_STRICT);
1251 __ jccb(Assembler::notZero, L_strict);
1252 __ fmul_d (at_rsp());
1253 __ jmpb(L_join);
1254 __ bind(L_strict);
1255 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1256 __ fmulp();
1257 __ fmul_d (at_rsp());
1258 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1259 __ fmulp();
1260 __ bind(L_join);
1261 break;
1262 }
1263 case div: {
1264 Label L_strict;
1265 Label L_join;
1266 const Address access_flags (rcx, methodOopDesc::access_flags_offset());
1267 __ get_method(rcx);
1268 __ movl(rcx, access_flags);
1269 __ testl(rcx, JVM_ACC_STRICT);
1270 __ jccb(Assembler::notZero, L_strict);
1271 __ fdivr_d(at_rsp());
1272 __ jmp(L_join);
1273 __ bind(L_strict);
1274 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias1()));
1275 __ fmul_d (at_rsp());
1276 __ fdivrp();
1277 __ fld_x(ExternalAddress(StubRoutines::addr_fpu_subnormal_bias2()));
1278 __ fmulp();
1279 __ bind(L_join);
1280 break;
1281 }
1282 case rem: __ fld_d (at_rsp()); __ fremr(rax); break;
1283 default : ShouldNotReachHere();
1284 }
1285 __ d2ieee();
1286 // Pop double precision number from rsp.
1287 __ pop(rax);
1288 __ pop(rdx);
1289 }
1292 void TemplateTable::ineg() {
1293 transition(itos, itos);
1294 __ negl(rax);
1295 }
1298 void TemplateTable::lneg() {
1299 transition(ltos, ltos);
1300 __ lneg(rdx, rax);
1301 }
1304 void TemplateTable::fneg() {
1305 transition(ftos, ftos);
1306 __ fchs();
1307 }
1310 void TemplateTable::dneg() {
1311 transition(dtos, dtos);
1312 __ fchs();
1313 }
1316 void TemplateTable::iinc() {
1317 transition(vtos, vtos);
1318 __ load_signed_byte(rdx, at_bcp(2)); // get constant
1319 locals_index(rbx);
1320 __ addl(iaddress(rbx), rdx);
1321 }
1324 void TemplateTable::wide_iinc() {
1325 transition(vtos, vtos);
1326 __ movl(rdx, at_bcp(4)); // get constant
1327 locals_index_wide(rbx);
1328 __ bswapl(rdx); // swap bytes & sign-extend constant
1329 __ sarl(rdx, 16);
1330 __ addl(iaddress(rbx), rdx);
1331 // Note: should probably use only one movl to get both
1332 // the index and the constant -> fix this
1333 }
1336 void TemplateTable::convert() {
1337 // Checking
1338 #ifdef ASSERT
1339 { TosState tos_in = ilgl;
1340 TosState tos_out = ilgl;
1341 switch (bytecode()) {
1342 case Bytecodes::_i2l: // fall through
1343 case Bytecodes::_i2f: // fall through
1344 case Bytecodes::_i2d: // fall through
1345 case Bytecodes::_i2b: // fall through
1346 case Bytecodes::_i2c: // fall through
1347 case Bytecodes::_i2s: tos_in = itos; break;
1348 case Bytecodes::_l2i: // fall through
1349 case Bytecodes::_l2f: // fall through
1350 case Bytecodes::_l2d: tos_in = ltos; break;
1351 case Bytecodes::_f2i: // fall through
1352 case Bytecodes::_f2l: // fall through
1353 case Bytecodes::_f2d: tos_in = ftos; break;
1354 case Bytecodes::_d2i: // fall through
1355 case Bytecodes::_d2l: // fall through
1356 case Bytecodes::_d2f: tos_in = dtos; break;
1357 default : ShouldNotReachHere();
1358 }
1359 switch (bytecode()) {
1360 case Bytecodes::_l2i: // fall through
1361 case Bytecodes::_f2i: // fall through
1362 case Bytecodes::_d2i: // fall through
1363 case Bytecodes::_i2b: // fall through
1364 case Bytecodes::_i2c: // fall through
1365 case Bytecodes::_i2s: tos_out = itos; break;
1366 case Bytecodes::_i2l: // fall through
1367 case Bytecodes::_f2l: // fall through
1368 case Bytecodes::_d2l: tos_out = ltos; break;
1369 case Bytecodes::_i2f: // fall through
1370 case Bytecodes::_l2f: // fall through
1371 case Bytecodes::_d2f: tos_out = ftos; break;
1372 case Bytecodes::_i2d: // fall through
1373 case Bytecodes::_l2d: // fall through
1374 case Bytecodes::_f2d: tos_out = dtos; break;
1375 default : ShouldNotReachHere();
1376 }
1377 transition(tos_in, tos_out);
1378 }
1379 #endif // ASSERT
1381 // Conversion
1382 // (Note: use push(rcx)/pop(rcx) for 1/2-word stack-ptr manipulation)
1383 switch (bytecode()) {
1384 case Bytecodes::_i2l:
1385 __ extend_sign(rdx, rax);
1386 break;
1387 case Bytecodes::_i2f:
1388 __ push(rax); // store int on tos
1389 __ fild_s(at_rsp()); // load int to ST0
1390 __ f2ieee(); // truncate to float size
1391 __ pop(rcx); // adjust rsp
1392 break;
1393 case Bytecodes::_i2d:
1394 __ push(rax); // add one slot for d2ieee()
1395 __ push(rax); // store int on tos
1396 __ fild_s(at_rsp()); // load int to ST0
1397 __ d2ieee(); // truncate to double size
1398 __ pop(rcx); // adjust rsp
1399 __ pop(rcx);
1400 break;
1401 case Bytecodes::_i2b:
1402 __ shll(rax, 24); // truncate upper 24 bits
1403 __ sarl(rax, 24); // and sign-extend byte
1404 LP64_ONLY(__ movsbl(rax, rax));
1405 break;
1406 case Bytecodes::_i2c:
1407 __ andl(rax, 0xFFFF); // truncate upper 16 bits
1408 LP64_ONLY(__ movzwl(rax, rax));
1409 break;
1410 case Bytecodes::_i2s:
1411 __ shll(rax, 16); // truncate upper 16 bits
1412 __ sarl(rax, 16); // and sign-extend short
1413 LP64_ONLY(__ movswl(rax, rax));
1414 break;
1415 case Bytecodes::_l2i:
1416 /* nothing to do */
1417 break;
1418 case Bytecodes::_l2f:
1419 __ push(rdx); // store long on tos
1420 __ push(rax);
1421 __ fild_d(at_rsp()); // load long to ST0
1422 __ f2ieee(); // truncate to float size
1423 __ pop(rcx); // adjust rsp
1424 __ pop(rcx);
1425 break;
1426 case Bytecodes::_l2d:
1427 __ push(rdx); // store long on tos
1428 __ push(rax);
1429 __ fild_d(at_rsp()); // load long to ST0
1430 __ d2ieee(); // truncate to double size
1431 __ pop(rcx); // adjust rsp
1432 __ pop(rcx);
1433 break;
1434 case Bytecodes::_f2i:
1435 __ push(rcx); // reserve space for argument
1436 __ fstp_s(at_rsp()); // pass float argument on stack
1437 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2i), 1);
1438 break;
1439 case Bytecodes::_f2l:
1440 __ push(rcx); // reserve space for argument
1441 __ fstp_s(at_rsp()); // pass float argument on stack
1442 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::f2l), 1);
1443 break;
1444 case Bytecodes::_f2d:
1445 /* nothing to do */
1446 break;
1447 case Bytecodes::_d2i:
1448 __ push(rcx); // reserve space for argument
1449 __ push(rcx);
1450 __ fstp_d(at_rsp()); // pass double argument on stack
1451 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2i), 2);
1452 break;
1453 case Bytecodes::_d2l:
1454 __ push(rcx); // reserve space for argument
1455 __ push(rcx);
1456 __ fstp_d(at_rsp()); // pass double argument on stack
1457 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::d2l), 2);
1458 break;
1459 case Bytecodes::_d2f:
1460 __ push(rcx); // reserve space for f2ieee()
1461 __ f2ieee(); // truncate to float size
1462 __ pop(rcx); // adjust rsp
1463 break;
1464 default :
1465 ShouldNotReachHere();
1466 }
1467 }
1470 void TemplateTable::lcmp() {
1471 transition(ltos, itos);
1472 // y = rdx:rax
1473 __ pop_l(rbx, rcx); // get x = rcx:rbx
1474 __ lcmp2int(rcx, rbx, rdx, rax);// rcx := cmp(x, y)
1475 __ mov(rax, rcx);
1476 }
1479 void TemplateTable::float_cmp(bool is_float, int unordered_result) {
1480 if (is_float) {
1481 __ pop_ftos_to_rsp();
1482 __ fld_s(at_rsp());
1483 } else {
1484 __ pop_dtos_to_rsp();
1485 __ fld_d(at_rsp());
1486 __ pop(rdx);
1487 }
1488 __ pop(rcx);
1489 __ fcmp2int(rax, unordered_result < 0);
1490 }
1493 void TemplateTable::branch(bool is_jsr, bool is_wide) {
1494 __ get_method(rcx); // ECX holds method
1495 __ profile_taken_branch(rax,rbx); // EAX holds updated MDP, EBX holds bumped taken count
1497 const ByteSize be_offset = methodOopDesc::backedge_counter_offset() + InvocationCounter::counter_offset();
1498 const ByteSize inv_offset = methodOopDesc::invocation_counter_offset() + InvocationCounter::counter_offset();
1499 const int method_offset = frame::interpreter_frame_method_offset * wordSize;
1501 // Load up EDX with the branch displacement
1502 __ movl(rdx, at_bcp(1));
1503 __ bswapl(rdx);
1504 if (!is_wide) __ sarl(rdx, 16);
1505 LP64_ONLY(__ movslq(rdx, rdx));
1508 // Handle all the JSR stuff here, then exit.
1509 // It's much shorter and cleaner than intermingling with the
1510 // non-JSR normal-branch stuff occuring below.
1511 if (is_jsr) {
1512 // Pre-load the next target bytecode into EBX
1513 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1, 0));
1515 // compute return address as bci in rax,
1516 __ lea(rax, at_bcp((is_wide ? 5 : 3) - in_bytes(constMethodOopDesc::codes_offset())));
1517 __ subptr(rax, Address(rcx, methodOopDesc::const_offset()));
1518 // Adjust the bcp in ESI by the displacement in EDX
1519 __ addptr(rsi, rdx);
1520 // Push return address
1521 __ push_i(rax);
1522 // jsr returns vtos
1523 __ dispatch_only_noverify(vtos);
1524 return;
1525 }
1527 // Normal (non-jsr) branch handling
1529 // Adjust the bcp in ESI by the displacement in EDX
1530 __ addptr(rsi, rdx);
1532 assert(UseLoopCounter || !UseOnStackReplacement, "on-stack-replacement requires loop counters");
1533 Label backedge_counter_overflow;
1534 Label profile_method;
1535 Label dispatch;
1536 if (UseLoopCounter) {
1537 // increment backedge counter for backward branches
1538 // rax,: MDO
1539 // rbx,: MDO bumped taken-count
1540 // rcx: method
1541 // rdx: target offset
1542 // rsi: target bcp
1543 // rdi: locals pointer
1544 __ testl(rdx, rdx); // check if forward or backward branch
1545 __ jcc(Assembler::positive, dispatch); // count only if backward branch
1547 // increment counter
1548 __ movl(rax, Address(rcx, be_offset)); // load backedge counter
1549 __ incrementl(rax, InvocationCounter::count_increment); // increment counter
1550 __ movl(Address(rcx, be_offset), rax); // store counter
1552 __ movl(rax, Address(rcx, inv_offset)); // load invocation counter
1553 __ andl(rax, InvocationCounter::count_mask_value); // and the status bits
1554 __ addl(rax, Address(rcx, be_offset)); // add both counters
1556 if (ProfileInterpreter) {
1557 // Test to see if we should create a method data oop
1558 __ cmp32(rax,
1559 ExternalAddress((address) &InvocationCounter::InterpreterProfileLimit));
1560 __ jcc(Assembler::less, dispatch);
1562 // if no method data exists, go to profile method
1563 __ test_method_data_pointer(rax, profile_method);
1565 if (UseOnStackReplacement) {
1566 // check for overflow against rbx, which is the MDO taken count
1567 __ cmp32(rbx,
1568 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1569 __ jcc(Assembler::below, dispatch);
1571 // When ProfileInterpreter is on, the backedge_count comes from the
1572 // methodDataOop, which value does not get reset on the call to
1573 // frequency_counter_overflow(). To avoid excessive calls to the overflow
1574 // routine while the method is being compiled, add a second test to make
1575 // sure the overflow function is called only once every overflow_frequency.
1576 const int overflow_frequency = 1024;
1577 __ andptr(rbx, overflow_frequency-1);
1578 __ jcc(Assembler::zero, backedge_counter_overflow);
1580 }
1581 } else {
1582 if (UseOnStackReplacement) {
1583 // check for overflow against rax, which is the sum of the counters
1584 __ cmp32(rax,
1585 ExternalAddress((address) &InvocationCounter::InterpreterBackwardBranchLimit));
1586 __ jcc(Assembler::aboveEqual, backedge_counter_overflow);
1588 }
1589 }
1590 __ bind(dispatch);
1591 }
1593 // Pre-load the next target bytecode into EBX
1594 __ load_unsigned_byte(rbx, Address(rsi, 0));
1596 // continue with the bytecode @ target
1597 // rax,: return bci for jsr's, unused otherwise
1598 // rbx,: target bytecode
1599 // rsi: target bcp
1600 __ dispatch_only(vtos);
1602 if (UseLoopCounter) {
1603 if (ProfileInterpreter) {
1604 // Out-of-line code to allocate method data oop.
1605 __ bind(profile_method);
1606 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method), rsi);
1607 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1608 __ movptr(rcx, Address(rbp, method_offset));
1609 __ movptr(rcx, Address(rcx, in_bytes(methodOopDesc::method_data_offset())));
1610 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1611 __ test_method_data_pointer(rcx, dispatch);
1612 // offset non-null mdp by MDO::data_offset() + IR::profile_method()
1613 __ addptr(rcx, in_bytes(methodDataOopDesc::data_offset()));
1614 __ addptr(rcx, rax);
1615 __ movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rcx);
1616 __ jmp(dispatch);
1617 }
1619 if (UseOnStackReplacement) {
1621 // invocation counter overflow
1622 __ bind(backedge_counter_overflow);
1623 __ negptr(rdx);
1624 __ addptr(rdx, rsi); // branch bcp
1625 call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), rdx);
1626 __ load_unsigned_byte(rbx, Address(rsi, 0)); // restore target bytecode
1628 // rax,: osr nmethod (osr ok) or NULL (osr not possible)
1629 // rbx,: target bytecode
1630 // rdx: scratch
1631 // rdi: locals pointer
1632 // rsi: bcp
1633 __ testptr(rax, rax); // test result
1634 __ jcc(Assembler::zero, dispatch); // no osr if null
1635 // nmethod may have been invalidated (VM may block upon call_VM return)
1636 __ movl(rcx, Address(rax, nmethod::entry_bci_offset()));
1637 __ cmpl(rcx, InvalidOSREntryBci);
1638 __ jcc(Assembler::equal, dispatch);
1640 // We have the address of an on stack replacement routine in rax,
1641 // We need to prepare to execute the OSR method. First we must
1642 // migrate the locals and monitors off of the stack.
1644 __ mov(rbx, rax); // save the nmethod
1646 const Register thread = rcx;
1647 __ get_thread(thread);
1648 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin));
1649 // rax, is OSR buffer, move it to expected parameter location
1650 __ mov(rcx, rax);
1652 // pop the interpreter frame
1653 __ movptr(rdx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
1654 __ leave(); // remove frame anchor
1655 __ pop(rdi); // get return address
1656 __ mov(rsp, rdx); // set sp to sender sp
1659 Label skip;
1660 Label chkint;
1662 // The interpreter frame we have removed may be returning to
1663 // either the callstub or the interpreter. Since we will
1664 // now be returning from a compiled (OSR) nmethod we must
1665 // adjust the return to the return were it can handler compiled
1666 // results and clean the fpu stack. This is very similar to
1667 // what a i2c adapter must do.
1669 // Are we returning to the call stub?
1671 __ cmp32(rdi, ExternalAddress(StubRoutines::_call_stub_return_address));
1672 __ jcc(Assembler::notEqual, chkint);
1674 // yes adjust to the specialized call stub return.
1675 assert(StubRoutines::x86::get_call_stub_compiled_return() != NULL, "must be set");
1676 __ lea(rdi, ExternalAddress(StubRoutines::x86::get_call_stub_compiled_return()));
1677 __ jmp(skip);
1679 __ bind(chkint);
1681 // Are we returning to the interpreter? Look for sentinel
1683 __ cmpl(Address(rdi, -2*wordSize), Interpreter::return_sentinel);
1684 __ jcc(Assembler::notEqual, skip);
1686 // Adjust to compiled return back to interpreter
1688 __ movptr(rdi, Address(rdi, -wordSize));
1689 __ bind(skip);
1691 // Align stack pointer for compiled code (note that caller is
1692 // responsible for undoing this fixup by remembering the old SP
1693 // in an rbp,-relative location)
1694 __ andptr(rsp, -(StackAlignmentInBytes));
1696 // push the (possibly adjusted) return address
1697 __ push(rdi);
1699 // and begin the OSR nmethod
1700 __ jmp(Address(rbx, nmethod::osr_entry_point_offset()));
1701 }
1702 }
1703 }
1706 void TemplateTable::if_0cmp(Condition cc) {
1707 transition(itos, vtos);
1708 // assume branch is more often taken than not (loops use backward branches)
1709 Label not_taken;
1710 __ testl(rax, rax);
1711 __ jcc(j_not(cc), not_taken);
1712 branch(false, false);
1713 __ bind(not_taken);
1714 __ profile_not_taken_branch(rax);
1715 }
1718 void TemplateTable::if_icmp(Condition cc) {
1719 transition(itos, vtos);
1720 // assume branch is more often taken than not (loops use backward branches)
1721 Label not_taken;
1722 __ pop_i(rdx);
1723 __ cmpl(rdx, rax);
1724 __ jcc(j_not(cc), not_taken);
1725 branch(false, false);
1726 __ bind(not_taken);
1727 __ profile_not_taken_branch(rax);
1728 }
1731 void TemplateTable::if_nullcmp(Condition cc) {
1732 transition(atos, vtos);
1733 // assume branch is more often taken than not (loops use backward branches)
1734 Label not_taken;
1735 __ testptr(rax, rax);
1736 __ jcc(j_not(cc), not_taken);
1737 branch(false, false);
1738 __ bind(not_taken);
1739 __ profile_not_taken_branch(rax);
1740 }
1743 void TemplateTable::if_acmp(Condition cc) {
1744 transition(atos, vtos);
1745 // assume branch is more often taken than not (loops use backward branches)
1746 Label not_taken;
1747 __ pop_ptr(rdx);
1748 __ cmpptr(rdx, rax);
1749 __ jcc(j_not(cc), not_taken);
1750 branch(false, false);
1751 __ bind(not_taken);
1752 __ profile_not_taken_branch(rax);
1753 }
1756 void TemplateTable::ret() {
1757 transition(vtos, vtos);
1758 locals_index(rbx);
1759 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1760 __ profile_ret(rbx, rcx);
1761 __ get_method(rax);
1762 __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1763 __ lea(rsi, Address(rsi, rbx, Address::times_1,
1764 constMethodOopDesc::codes_offset()));
1765 __ dispatch_next(vtos);
1766 }
1769 void TemplateTable::wide_ret() {
1770 transition(vtos, vtos);
1771 locals_index_wide(rbx);
1772 __ movptr(rbx, iaddress(rbx)); // get return bci, compute return bcp
1773 __ profile_ret(rbx, rcx);
1774 __ get_method(rax);
1775 __ movptr(rsi, Address(rax, methodOopDesc::const_offset()));
1776 __ lea(rsi, Address(rsi, rbx, Address::times_1, constMethodOopDesc::codes_offset()));
1777 __ dispatch_next(vtos);
1778 }
1781 void TemplateTable::tableswitch() {
1782 Label default_case, continue_execution;
1783 transition(itos, vtos);
1784 // align rsi
1785 __ lea(rbx, at_bcp(wordSize));
1786 __ andptr(rbx, -wordSize);
1787 // load lo & hi
1788 __ movl(rcx, Address(rbx, 1 * wordSize));
1789 __ movl(rdx, Address(rbx, 2 * wordSize));
1790 __ bswapl(rcx);
1791 __ bswapl(rdx);
1792 // check against lo & hi
1793 __ cmpl(rax, rcx);
1794 __ jccb(Assembler::less, default_case);
1795 __ cmpl(rax, rdx);
1796 __ jccb(Assembler::greater, default_case);
1797 // lookup dispatch offset
1798 __ subl(rax, rcx);
1799 __ movl(rdx, Address(rbx, rax, Address::times_4, 3 * BytesPerInt));
1800 __ profile_switch_case(rax, rbx, rcx);
1801 // continue execution
1802 __ bind(continue_execution);
1803 __ bswapl(rdx);
1804 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1805 __ addptr(rsi, rdx);
1806 __ dispatch_only(vtos);
1807 // handle default
1808 __ bind(default_case);
1809 __ profile_switch_default(rax);
1810 __ movl(rdx, Address(rbx, 0));
1811 __ jmp(continue_execution);
1812 }
1815 void TemplateTable::lookupswitch() {
1816 transition(itos, itos);
1817 __ stop("lookupswitch bytecode should have been rewritten");
1818 }
1821 void TemplateTable::fast_linearswitch() {
1822 transition(itos, vtos);
1823 Label loop_entry, loop, found, continue_execution;
1824 // bswapl rax, so we can avoid bswapping the table entries
1825 __ bswapl(rax);
1826 // align rsi
1827 __ lea(rbx, at_bcp(wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1828 __ andptr(rbx, -wordSize);
1829 // set counter
1830 __ movl(rcx, Address(rbx, wordSize));
1831 __ bswapl(rcx);
1832 __ jmpb(loop_entry);
1833 // table search
1834 __ bind(loop);
1835 __ cmpl(rax, Address(rbx, rcx, Address::times_8, 2 * wordSize));
1836 __ jccb(Assembler::equal, found);
1837 __ bind(loop_entry);
1838 __ decrementl(rcx);
1839 __ jcc(Assembler::greaterEqual, loop);
1840 // default case
1841 __ profile_switch_default(rax);
1842 __ movl(rdx, Address(rbx, 0));
1843 __ jmpb(continue_execution);
1844 // entry found -> get offset
1845 __ bind(found);
1846 __ movl(rdx, Address(rbx, rcx, Address::times_8, 3 * wordSize));
1847 __ profile_switch_case(rcx, rax, rbx);
1848 // continue execution
1849 __ bind(continue_execution);
1850 __ bswapl(rdx);
1851 __ load_unsigned_byte(rbx, Address(rsi, rdx, Address::times_1));
1852 __ addptr(rsi, rdx);
1853 __ dispatch_only(vtos);
1854 }
1857 void TemplateTable::fast_binaryswitch() {
1858 transition(itos, vtos);
1859 // Implementation using the following core algorithm:
1860 //
1861 // int binary_search(int key, LookupswitchPair* array, int n) {
1862 // // Binary search according to "Methodik des Programmierens" by
1863 // // Edsger W. Dijkstra and W.H.J. Feijen, Addison Wesley Germany 1985.
1864 // int i = 0;
1865 // int j = n;
1866 // while (i+1 < j) {
1867 // // invariant P: 0 <= i < j <= n and (a[i] <= key < a[j] or Q)
1868 // // with Q: for all i: 0 <= i < n: key < a[i]
1869 // // where a stands for the array and assuming that the (inexisting)
1870 // // element a[n] is infinitely big.
1871 // int h = (i + j) >> 1;
1872 // // i < h < j
1873 // if (key < array[h].fast_match()) {
1874 // j = h;
1875 // } else {
1876 // i = h;
1877 // }
1878 // }
1879 // // R: a[i] <= key < a[i+1] or Q
1880 // // (i.e., if key is within array, i is the correct index)
1881 // return i;
1882 // }
1884 // register allocation
1885 const Register key = rax; // already set (tosca)
1886 const Register array = rbx;
1887 const Register i = rcx;
1888 const Register j = rdx;
1889 const Register h = rdi; // needs to be restored
1890 const Register temp = rsi;
1891 // setup array
1892 __ save_bcp();
1894 __ lea(array, at_bcp(3*wordSize)); // btw: should be able to get rid of this instruction (change offsets below)
1895 __ andptr(array, -wordSize);
1896 // initialize i & j
1897 __ xorl(i, i); // i = 0;
1898 __ movl(j, Address(array, -wordSize)); // j = length(array);
1899 // Convert j into native byteordering
1900 __ bswapl(j);
1901 // and start
1902 Label entry;
1903 __ jmp(entry);
1905 // binary search loop
1906 { Label loop;
1907 __ bind(loop);
1908 // int h = (i + j) >> 1;
1909 __ leal(h, Address(i, j, Address::times_1)); // h = i + j;
1910 __ sarl(h, 1); // h = (i + j) >> 1;
1911 // if (key < array[h].fast_match()) {
1912 // j = h;
1913 // } else {
1914 // i = h;
1915 // }
1916 // Convert array[h].match to native byte-ordering before compare
1917 __ movl(temp, Address(array, h, Address::times_8, 0*wordSize));
1918 __ bswapl(temp);
1919 __ cmpl(key, temp);
1920 if (VM_Version::supports_cmov()) {
1921 __ cmovl(Assembler::less , j, h); // j = h if (key < array[h].fast_match())
1922 __ cmovl(Assembler::greaterEqual, i, h); // i = h if (key >= array[h].fast_match())
1923 } else {
1924 Label set_i, end_of_if;
1925 __ jccb(Assembler::greaterEqual, set_i); // {
1926 __ mov(j, h); // j = h;
1927 __ jmp(end_of_if); // }
1928 __ bind(set_i); // else {
1929 __ mov(i, h); // i = h;
1930 __ bind(end_of_if); // }
1931 }
1932 // while (i+1 < j)
1933 __ bind(entry);
1934 __ leal(h, Address(i, 1)); // i+1
1935 __ cmpl(h, j); // i+1 < j
1936 __ jcc(Assembler::less, loop);
1937 }
1939 // end of binary search, result index is i (must check again!)
1940 Label default_case;
1941 // Convert array[i].match to native byte-ordering before compare
1942 __ movl(temp, Address(array, i, Address::times_8, 0*wordSize));
1943 __ bswapl(temp);
1944 __ cmpl(key, temp);
1945 __ jcc(Assembler::notEqual, default_case);
1947 // entry found -> j = offset
1948 __ movl(j , Address(array, i, Address::times_8, 1*wordSize));
1949 __ profile_switch_case(i, key, array);
1950 __ bswapl(j);
1951 LP64_ONLY(__ movslq(j, j));
1952 __ restore_bcp();
1953 __ restore_locals(); // restore rdi
1954 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1956 __ addptr(rsi, j);
1957 __ dispatch_only(vtos);
1959 // default case -> j = default offset
1960 __ bind(default_case);
1961 __ profile_switch_default(i);
1962 __ movl(j, Address(array, -2*wordSize));
1963 __ bswapl(j);
1964 LP64_ONLY(__ movslq(j, j));
1965 __ restore_bcp();
1966 __ restore_locals(); // restore rdi
1967 __ load_unsigned_byte(rbx, Address(rsi, j, Address::times_1));
1968 __ addptr(rsi, j);
1969 __ dispatch_only(vtos);
1970 }
1973 void TemplateTable::_return(TosState state) {
1974 transition(state, state);
1975 assert(_desc->calls_vm(), "inconsistent calls_vm information"); // call in remove_activation
1977 if (_desc->bytecode() == Bytecodes::_return_register_finalizer) {
1978 assert(state == vtos, "only valid state");
1979 __ movptr(rax, aaddress(0));
1980 __ movptr(rdi, Address(rax, oopDesc::klass_offset_in_bytes()));
1981 __ movl(rdi, Address(rdi, Klass::access_flags_offset_in_bytes() + sizeof(oopDesc)));
1982 __ testl(rdi, JVM_ACC_HAS_FINALIZER);
1983 Label skip_register_finalizer;
1984 __ jcc(Assembler::zero, skip_register_finalizer);
1986 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::register_finalizer), rax);
1988 __ bind(skip_register_finalizer);
1989 }
1991 __ remove_activation(state, rsi);
1992 __ jmp(rsi);
1993 }
1996 // ----------------------------------------------------------------------------
1997 // Volatile variables demand their effects be made known to all CPU's in
1998 // order. Store buffers on most chips allow reads & writes to reorder; the
1999 // JMM's ReadAfterWrite.java test fails in -Xint mode without some kind of
2000 // memory barrier (i.e., it's not sufficient that the interpreter does not
2001 // reorder volatile references, the hardware also must not reorder them).
2002 //
2003 // According to the new Java Memory Model (JMM):
2004 // (1) All volatiles are serialized wrt to each other.
2005 // ALSO reads & writes act as aquire & release, so:
2006 // (2) A read cannot let unrelated NON-volatile memory refs that happen after
2007 // the read float up to before the read. It's OK for non-volatile memory refs
2008 // that happen before the volatile read to float down below it.
2009 // (3) Similar a volatile write cannot let unrelated NON-volatile memory refs
2010 // that happen BEFORE the write float down to after the write. It's OK for
2011 // non-volatile memory refs that happen after the volatile write to float up
2012 // before it.
2013 //
2014 // We only put in barriers around volatile refs (they are expensive), not
2015 // _between_ memory refs (that would require us to track the flavor of the
2016 // previous memory refs). Requirements (2) and (3) require some barriers
2017 // before volatile stores and after volatile loads. These nearly cover
2018 // requirement (1) but miss the volatile-store-volatile-load case. This final
2019 // case is placed after volatile-stores although it could just as well go
2020 // before volatile-loads.
2021 void TemplateTable::volatile_barrier(Assembler::Membar_mask_bits order_constraint ) {
2022 // Helper function to insert a is-volatile test and memory barrier
2023 if( !os::is_MP() ) return; // Not needed on single CPU
2024 __ membar(order_constraint);
2025 }
2027 void TemplateTable::resolve_cache_and_index(int byte_no, Register Rcache, Register index) {
2028 assert(byte_no == 1 || byte_no == 2, "byte_no out of range");
2030 Register temp = rbx;
2032 assert_different_registers(Rcache, index, temp);
2034 const int shift_count = (1 + byte_no)*BitsPerByte;
2035 Label resolved;
2036 __ get_cache_and_index_at_bcp(Rcache, index, 1);
2037 __ movl(temp, Address(Rcache,
2038 index,
2039 Address::times_ptr,
2040 constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::indices_offset()));
2041 __ shrl(temp, shift_count);
2042 // have we resolved this bytecode?
2043 __ andptr(temp, 0xFF);
2044 __ cmpl(temp, (int)bytecode());
2045 __ jcc(Assembler::equal, resolved);
2047 // resolve first time through
2048 address entry;
2049 switch (bytecode()) {
2050 case Bytecodes::_getstatic : // fall through
2051 case Bytecodes::_putstatic : // fall through
2052 case Bytecodes::_getfield : // fall through
2053 case Bytecodes::_putfield : entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_get_put); break;
2054 case Bytecodes::_invokevirtual : // fall through
2055 case Bytecodes::_invokespecial : // fall through
2056 case Bytecodes::_invokestatic : // fall through
2057 case Bytecodes::_invokeinterface: entry = CAST_FROM_FN_PTR(address, InterpreterRuntime::resolve_invoke); break;
2058 default : ShouldNotReachHere(); break;
2059 }
2060 __ movl(temp, (int)bytecode());
2061 __ call_VM(noreg, entry, temp);
2062 // Update registers with resolved info
2063 __ get_cache_and_index_at_bcp(Rcache, index, 1);
2064 __ bind(resolved);
2065 }
2068 // The cache and index registers must be set before call
2069 void TemplateTable::load_field_cp_cache_entry(Register obj,
2070 Register cache,
2071 Register index,
2072 Register off,
2073 Register flags,
2074 bool is_static = false) {
2075 assert_different_registers(cache, index, flags, off);
2077 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2078 // Field offset
2079 __ movptr(off, Address(cache, index, Address::times_ptr,
2080 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f2_offset())));
2081 // Flags
2082 __ movl(flags, Address(cache, index, Address::times_ptr,
2083 in_bytes(cp_base_offset + ConstantPoolCacheEntry::flags_offset())));
2085 // klass overwrite register
2086 if (is_static) {
2087 __ movptr(obj, Address(cache, index, Address::times_ptr,
2088 in_bytes(cp_base_offset + ConstantPoolCacheEntry::f1_offset())));
2089 }
2090 }
2092 void TemplateTable::load_invoke_cp_cache_entry(int byte_no,
2093 Register method,
2094 Register itable_index,
2095 Register flags,
2096 bool is_invokevirtual,
2097 bool is_invokevfinal /*unused*/) {
2098 // setup registers
2099 const Register cache = rcx;
2100 const Register index = rdx;
2101 assert_different_registers(method, flags);
2102 assert_different_registers(method, cache, index);
2103 assert_different_registers(itable_index, flags);
2104 assert_different_registers(itable_index, cache, index);
2105 // determine constant pool cache field offsets
2106 const int method_offset = in_bytes(
2107 constantPoolCacheOopDesc::base_offset() +
2108 (is_invokevirtual
2109 ? ConstantPoolCacheEntry::f2_offset()
2110 : ConstantPoolCacheEntry::f1_offset()
2111 )
2112 );
2113 const int flags_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2114 ConstantPoolCacheEntry::flags_offset());
2115 // access constant pool cache fields
2116 const int index_offset = in_bytes(constantPoolCacheOopDesc::base_offset() +
2117 ConstantPoolCacheEntry::f2_offset());
2119 resolve_cache_and_index(byte_no, cache, index);
2121 __ movptr(method, Address(cache, index, Address::times_ptr, method_offset));
2122 if (itable_index != noreg) {
2123 __ movptr(itable_index, Address(cache, index, Address::times_ptr, index_offset));
2124 }
2125 __ movl(flags , Address(cache, index, Address::times_ptr, flags_offset ));
2126 }
2129 // The registers cache and index expected to be set before call.
2130 // Correct values of the cache and index registers are preserved.
2131 void TemplateTable::jvmti_post_field_access(Register cache,
2132 Register index,
2133 bool is_static,
2134 bool has_tos) {
2135 if (JvmtiExport::can_post_field_access()) {
2136 // Check to see if a field access watch has been set before we take
2137 // the time to call into the VM.
2138 Label L1;
2139 assert_different_registers(cache, index, rax);
2140 __ mov32(rax, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2141 __ testl(rax,rax);
2142 __ jcc(Assembler::zero, L1);
2144 // cache entry pointer
2145 __ addptr(cache, in_bytes(constantPoolCacheOopDesc::base_offset()));
2146 __ shll(index, LogBytesPerWord);
2147 __ addptr(cache, index);
2148 if (is_static) {
2149 __ xorptr(rax, rax); // NULL object reference
2150 } else {
2151 __ pop(atos); // Get the object
2152 __ verify_oop(rax);
2153 __ push(atos); // Restore stack state
2154 }
2155 // rax,: object pointer or NULL
2156 // cache: cache entry pointer
2157 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access),
2158 rax, cache);
2159 __ get_cache_and_index_at_bcp(cache, index, 1);
2160 __ bind(L1);
2161 }
2162 }
2164 void TemplateTable::pop_and_check_object(Register r) {
2165 __ pop_ptr(r);
2166 __ null_check(r); // for field access must check obj.
2167 __ verify_oop(r);
2168 }
2170 void TemplateTable::getfield_or_static(int byte_no, bool is_static) {
2171 transition(vtos, vtos);
2173 const Register cache = rcx;
2174 const Register index = rdx;
2175 const Register obj = rcx;
2176 const Register off = rbx;
2177 const Register flags = rax;
2179 resolve_cache_and_index(byte_no, cache, index);
2180 jvmti_post_field_access(cache, index, is_static, false);
2181 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2183 if (!is_static) pop_and_check_object(obj);
2185 const Address lo(obj, off, Address::times_1, 0*wordSize);
2186 const Address hi(obj, off, Address::times_1, 1*wordSize);
2188 Label Done, notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2190 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2191 assert(btos == 0, "change code, btos != 0");
2192 // btos
2193 __ andptr(flags, 0x0f);
2194 __ jcc(Assembler::notZero, notByte);
2196 __ load_signed_byte(rax, lo );
2197 __ push(btos);
2198 // Rewrite bytecode to be faster
2199 if (!is_static) {
2200 patch_bytecode(Bytecodes::_fast_bgetfield, rcx, rbx);
2201 }
2202 __ jmp(Done);
2204 __ bind(notByte);
2205 // itos
2206 __ cmpl(flags, itos );
2207 __ jcc(Assembler::notEqual, notInt);
2209 __ movl(rax, lo );
2210 __ push(itos);
2211 // Rewrite bytecode to be faster
2212 if (!is_static) {
2213 patch_bytecode(Bytecodes::_fast_igetfield, rcx, rbx);
2214 }
2215 __ jmp(Done);
2217 __ bind(notInt);
2218 // atos
2219 __ cmpl(flags, atos );
2220 __ jcc(Assembler::notEqual, notObj);
2222 __ movl(rax, lo );
2223 __ push(atos);
2224 if (!is_static) {
2225 patch_bytecode(Bytecodes::_fast_agetfield, rcx, rbx);
2226 }
2227 __ jmp(Done);
2229 __ bind(notObj);
2230 // ctos
2231 __ cmpl(flags, ctos );
2232 __ jcc(Assembler::notEqual, notChar);
2234 __ load_unsigned_word(rax, lo );
2235 __ push(ctos);
2236 if (!is_static) {
2237 patch_bytecode(Bytecodes::_fast_cgetfield, rcx, rbx);
2238 }
2239 __ jmp(Done);
2241 __ bind(notChar);
2242 // stos
2243 __ cmpl(flags, stos );
2244 __ jcc(Assembler::notEqual, notShort);
2246 __ load_signed_word(rax, lo );
2247 __ push(stos);
2248 if (!is_static) {
2249 patch_bytecode(Bytecodes::_fast_sgetfield, rcx, rbx);
2250 }
2251 __ jmp(Done);
2253 __ bind(notShort);
2254 // ltos
2255 __ cmpl(flags, ltos );
2256 __ jcc(Assembler::notEqual, notLong);
2258 // Generate code as if volatile. There just aren't enough registers to
2259 // save that information and this code is faster than the test.
2260 __ fild_d(lo); // Must load atomically
2261 __ subptr(rsp,2*wordSize); // Make space for store
2262 __ fistp_d(Address(rsp,0));
2263 __ pop(rax);
2264 __ pop(rdx);
2266 __ push(ltos);
2267 // Don't rewrite to _fast_lgetfield for potential volatile case.
2268 __ jmp(Done);
2270 __ bind(notLong);
2271 // ftos
2272 __ cmpl(flags, ftos );
2273 __ jcc(Assembler::notEqual, notFloat);
2275 __ fld_s(lo);
2276 __ push(ftos);
2277 if (!is_static) {
2278 patch_bytecode(Bytecodes::_fast_fgetfield, rcx, rbx);
2279 }
2280 __ jmp(Done);
2282 __ bind(notFloat);
2283 // dtos
2284 __ cmpl(flags, dtos );
2285 __ jcc(Assembler::notEqual, notDouble);
2287 __ fld_d(lo);
2288 __ push(dtos);
2289 if (!is_static) {
2290 patch_bytecode(Bytecodes::_fast_dgetfield, rcx, rbx);
2291 }
2292 __ jmpb(Done);
2294 __ bind(notDouble);
2296 __ stop("Bad state");
2298 __ bind(Done);
2299 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2300 // volatile_barrier( );
2301 }
2304 void TemplateTable::getfield(int byte_no) {
2305 getfield_or_static(byte_no, false);
2306 }
2309 void TemplateTable::getstatic(int byte_no) {
2310 getfield_or_static(byte_no, true);
2311 }
2313 // The registers cache and index expected to be set before call.
2314 // The function may destroy various registers, just not the cache and index registers.
2315 void TemplateTable::jvmti_post_field_mod(Register cache, Register index, bool is_static) {
2317 ByteSize cp_base_offset = constantPoolCacheOopDesc::base_offset();
2319 if (JvmtiExport::can_post_field_modification()) {
2320 // Check to see if a field modification watch has been set before we take
2321 // the time to call into the VM.
2322 Label L1;
2323 assert_different_registers(cache, index, rax);
2324 __ mov32(rax, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2325 __ testl(rax, rax);
2326 __ jcc(Assembler::zero, L1);
2328 // The cache and index registers have been already set.
2329 // This allows to eliminate this call but the cache and index
2330 // registers have to be correspondingly used after this line.
2331 __ get_cache_and_index_at_bcp(rax, rdx, 1);
2333 if (is_static) {
2334 // Life is simple. Null out the object pointer.
2335 __ xorptr(rbx, rbx);
2336 } else {
2337 // Life is harder. The stack holds the value on top, followed by the object.
2338 // We don't know the size of the value, though; it could be one or two words
2339 // depending on its type. As a result, we must find the type to determine where
2340 // the object is.
2341 Label two_word, valsize_known;
2342 __ movl(rcx, Address(rax, rdx, Address::times_ptr, in_bytes(cp_base_offset +
2343 ConstantPoolCacheEntry::flags_offset())));
2344 __ mov(rbx, rsp);
2345 __ shrl(rcx, ConstantPoolCacheEntry::tosBits);
2346 // Make sure we don't need to mask rcx for tosBits after the above shift
2347 ConstantPoolCacheEntry::verify_tosBits();
2348 __ cmpl(rcx, ltos);
2349 __ jccb(Assembler::equal, two_word);
2350 __ cmpl(rcx, dtos);
2351 __ jccb(Assembler::equal, two_word);
2352 __ addptr(rbx, Interpreter::expr_offset_in_bytes(1)); // one word jvalue (not ltos, dtos)
2353 __ jmpb(valsize_known);
2355 __ bind(two_word);
2356 __ addptr(rbx, Interpreter::expr_offset_in_bytes(2)); // two words jvalue
2358 __ bind(valsize_known);
2359 // setup object pointer
2360 __ movptr(rbx, Address(rbx, 0));
2361 }
2362 // cache entry pointer
2363 __ addptr(rax, in_bytes(cp_base_offset));
2364 __ shll(rdx, LogBytesPerWord);
2365 __ addptr(rax, rdx);
2366 // object (tos)
2367 __ mov(rcx, rsp);
2368 // rbx,: object pointer set up above (NULL if static)
2369 // rax,: cache entry pointer
2370 // rcx: jvalue object on the stack
2371 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification),
2372 rbx, rax, rcx);
2373 __ get_cache_and_index_at_bcp(cache, index, 1);
2374 __ bind(L1);
2375 }
2376 }
2379 void TemplateTable::putfield_or_static(int byte_no, bool is_static) {
2380 transition(vtos, vtos);
2382 const Register cache = rcx;
2383 const Register index = rdx;
2384 const Register obj = rcx;
2385 const Register off = rbx;
2386 const Register flags = rax;
2388 resolve_cache_and_index(byte_no, cache, index);
2389 jvmti_post_field_mod(cache, index, is_static);
2390 load_field_cp_cache_entry(obj, cache, index, off, flags, is_static);
2392 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2393 // volatile_barrier( );
2395 Label notVolatile, Done;
2396 __ movl(rdx, flags);
2397 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2398 __ andl(rdx, 0x1);
2400 // field addresses
2401 const Address lo(obj, off, Address::times_1, 0*wordSize);
2402 const Address hi(obj, off, Address::times_1, 1*wordSize);
2404 Label notByte, notInt, notShort, notChar, notLong, notFloat, notObj, notDouble;
2406 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2407 assert(btos == 0, "change code, btos != 0");
2408 // btos
2409 __ andl(flags, 0x0f);
2410 __ jcc(Assembler::notZero, notByte);
2412 __ pop(btos);
2413 if (!is_static) pop_and_check_object(obj);
2414 __ movb(lo, rax );
2415 if (!is_static) {
2416 patch_bytecode(Bytecodes::_fast_bputfield, rcx, rbx);
2417 }
2418 __ jmp(Done);
2420 __ bind(notByte);
2421 // itos
2422 __ cmpl(flags, itos );
2423 __ jcc(Assembler::notEqual, notInt);
2425 __ pop(itos);
2426 if (!is_static) pop_and_check_object(obj);
2428 __ movl(lo, rax );
2429 if (!is_static) {
2430 patch_bytecode(Bytecodes::_fast_iputfield, rcx, rbx);
2431 }
2432 __ jmp(Done);
2434 __ bind(notInt);
2435 // atos
2436 __ cmpl(flags, atos );
2437 __ jcc(Assembler::notEqual, notObj);
2439 __ pop(atos);
2440 if (!is_static) pop_and_check_object(obj);
2442 __ movptr(lo, rax );
2443 __ store_check(obj, lo); // Need to mark card
2444 if (!is_static) {
2445 patch_bytecode(Bytecodes::_fast_aputfield, rcx, rbx);
2446 }
2447 __ jmp(Done);
2449 __ bind(notObj);
2450 // ctos
2451 __ cmpl(flags, ctos );
2452 __ jcc(Assembler::notEqual, notChar);
2454 __ pop(ctos);
2455 if (!is_static) pop_and_check_object(obj);
2456 __ movw(lo, rax );
2457 if (!is_static) {
2458 patch_bytecode(Bytecodes::_fast_cputfield, rcx, rbx);
2459 }
2460 __ jmp(Done);
2462 __ bind(notChar);
2463 // stos
2464 __ cmpl(flags, stos );
2465 __ jcc(Assembler::notEqual, notShort);
2467 __ pop(stos);
2468 if (!is_static) pop_and_check_object(obj);
2469 __ movw(lo, rax );
2470 if (!is_static) {
2471 patch_bytecode(Bytecodes::_fast_sputfield, rcx, rbx);
2472 }
2473 __ jmp(Done);
2475 __ bind(notShort);
2476 // ltos
2477 __ cmpl(flags, ltos );
2478 __ jcc(Assembler::notEqual, notLong);
2480 Label notVolatileLong;
2481 __ testl(rdx, rdx);
2482 __ jcc(Assembler::zero, notVolatileLong);
2484 __ pop(ltos); // overwrites rdx, do this after testing volatile.
2485 if (!is_static) pop_and_check_object(obj);
2487 // Replace with real volatile test
2488 __ push(rdx);
2489 __ push(rax); // Must update atomically with FIST
2490 __ fild_d(Address(rsp,0)); // So load into FPU register
2491 __ fistp_d(lo); // and put into memory atomically
2492 __ addptr(rsp, 2*wordSize);
2493 // volatile_barrier();
2494 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2495 Assembler::StoreStore));
2496 // Don't rewrite volatile version
2497 __ jmp(notVolatile);
2499 __ bind(notVolatileLong);
2501 __ pop(ltos); // overwrites rdx
2502 if (!is_static) pop_and_check_object(obj);
2503 NOT_LP64(__ movptr(hi, rdx));
2504 __ movptr(lo, rax);
2505 if (!is_static) {
2506 patch_bytecode(Bytecodes::_fast_lputfield, rcx, rbx);
2507 }
2508 __ jmp(notVolatile);
2510 __ bind(notLong);
2511 // ftos
2512 __ cmpl(flags, ftos );
2513 __ jcc(Assembler::notEqual, notFloat);
2515 __ pop(ftos);
2516 if (!is_static) pop_and_check_object(obj);
2517 __ fstp_s(lo);
2518 if (!is_static) {
2519 patch_bytecode(Bytecodes::_fast_fputfield, rcx, rbx);
2520 }
2521 __ jmp(Done);
2523 __ bind(notFloat);
2524 // dtos
2525 __ cmpl(flags, dtos );
2526 __ jcc(Assembler::notEqual, notDouble);
2528 __ pop(dtos);
2529 if (!is_static) pop_and_check_object(obj);
2530 __ fstp_d(lo);
2531 if (!is_static) {
2532 patch_bytecode(Bytecodes::_fast_dputfield, rcx, rbx);
2533 }
2534 __ jmp(Done);
2536 __ bind(notDouble);
2538 __ stop("Bad state");
2540 __ bind(Done);
2542 // Check for volatile store
2543 __ testl(rdx, rdx);
2544 __ jcc(Assembler::zero, notVolatile);
2545 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2546 Assembler::StoreStore));
2547 __ bind(notVolatile);
2548 }
2551 void TemplateTable::putfield(int byte_no) {
2552 putfield_or_static(byte_no, false);
2553 }
2556 void TemplateTable::putstatic(int byte_no) {
2557 putfield_or_static(byte_no, true);
2558 }
2560 void TemplateTable::jvmti_post_fast_field_mod() {
2561 if (JvmtiExport::can_post_field_modification()) {
2562 // Check to see if a field modification watch has been set before we take
2563 // the time to call into the VM.
2564 Label L2;
2565 __ mov32(rcx, ExternalAddress((address)JvmtiExport::get_field_modification_count_addr()));
2566 __ testl(rcx,rcx);
2567 __ jcc(Assembler::zero, L2);
2568 __ pop_ptr(rbx); // copy the object pointer from tos
2569 __ verify_oop(rbx);
2570 __ push_ptr(rbx); // put the object pointer back on tos
2571 __ subptr(rsp, sizeof(jvalue)); // add space for a jvalue object
2572 __ mov(rcx, rsp);
2573 __ push_ptr(rbx); // save object pointer so we can steal rbx,
2574 __ xorptr(rbx, rbx);
2575 const Address lo_value(rcx, rbx, Address::times_1, 0*wordSize);
2576 const Address hi_value(rcx, rbx, Address::times_1, 1*wordSize);
2577 switch (bytecode()) { // load values into the jvalue object
2578 case Bytecodes::_fast_bputfield: __ movb(lo_value, rax); break;
2579 case Bytecodes::_fast_sputfield: __ movw(lo_value, rax); break;
2580 case Bytecodes::_fast_cputfield: __ movw(lo_value, rax); break;
2581 case Bytecodes::_fast_iputfield: __ movl(lo_value, rax); break;
2582 case Bytecodes::_fast_lputfield:
2583 NOT_LP64(__ movptr(hi_value, rdx));
2584 __ movptr(lo_value, rax);
2585 break;
2587 // need to call fld_s() after fstp_s() to restore the value for below
2588 case Bytecodes::_fast_fputfield: __ fstp_s(lo_value); __ fld_s(lo_value); break;
2590 // need to call fld_d() after fstp_d() to restore the value for below
2591 case Bytecodes::_fast_dputfield: __ fstp_d(lo_value); __ fld_d(lo_value); break;
2593 // since rcx is not an object we don't call store_check() here
2594 case Bytecodes::_fast_aputfield: __ movptr(lo_value, rax); break;
2596 default: ShouldNotReachHere();
2597 }
2598 __ pop_ptr(rbx); // restore copy of object pointer
2600 // Save rax, and sometimes rdx because call_VM() will clobber them,
2601 // then use them for JVM/DI purposes
2602 __ push(rax);
2603 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2604 // access constant pool cache entry
2605 __ get_cache_entry_pointer_at_bcp(rax, rdx, 1);
2606 __ verify_oop(rbx);
2607 // rbx,: object pointer copied above
2608 // rax,: cache entry pointer
2609 // rcx: jvalue object on the stack
2610 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_modification), rbx, rax, rcx);
2611 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx); // restore high value
2612 __ pop(rax); // restore lower value
2613 __ addptr(rsp, sizeof(jvalue)); // release jvalue object space
2614 __ bind(L2);
2615 }
2616 }
2618 void TemplateTable::fast_storefield(TosState state) {
2619 transition(state, vtos);
2621 ByteSize base = constantPoolCacheOopDesc::base_offset();
2623 jvmti_post_fast_field_mod();
2625 // access constant pool cache
2626 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2628 // test for volatile with rdx but rdx is tos register for lputfield.
2629 if (bytecode() == Bytecodes::_fast_lputfield) __ push(rdx);
2630 __ movl(rdx, Address(rcx, rbx, Address::times_ptr, in_bytes(base +
2631 ConstantPoolCacheEntry::flags_offset())));
2633 // replace index with field offset from cache entry
2634 __ movptr(rbx, Address(rcx, rbx, Address::times_ptr, in_bytes(base + ConstantPoolCacheEntry::f2_offset())));
2636 // Doug Lea believes this is not needed with current Sparcs (TSO) and Intel (PSO).
2637 // volatile_barrier( );
2639 Label notVolatile, Done;
2640 __ shrl(rdx, ConstantPoolCacheEntry::volatileField);
2641 __ andl(rdx, 0x1);
2642 // Check for volatile store
2643 __ testl(rdx, rdx);
2644 __ jcc(Assembler::zero, notVolatile);
2646 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2648 // Get object from stack
2649 pop_and_check_object(rcx);
2651 // field addresses
2652 const Address lo(rcx, rbx, Address::times_1, 0*wordSize);
2653 const Address hi(rcx, rbx, Address::times_1, 1*wordSize);
2655 // access field
2656 switch (bytecode()) {
2657 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2658 case Bytecodes::_fast_sputfield: // fall through
2659 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2660 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2661 case Bytecodes::_fast_lputfield:
2662 NOT_LP64(__ movptr(hi, rdx));
2663 __ movptr(lo, rax);
2664 break;
2665 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2666 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2667 case Bytecodes::_fast_aputfield: __ movptr(lo, rax); __ store_check(rcx, lo); break;
2668 default:
2669 ShouldNotReachHere();
2670 }
2672 Label done;
2673 volatile_barrier(Assembler::Membar_mask_bits(Assembler::StoreLoad |
2674 Assembler::StoreStore));
2675 __ jmpb(done);
2677 // Same code as above, but don't need rdx to test for volatile.
2678 __ bind(notVolatile);
2680 if (bytecode() == Bytecodes::_fast_lputfield) __ pop(rdx);
2682 // Get object from stack
2683 pop_and_check_object(rcx);
2685 // access field
2686 switch (bytecode()) {
2687 case Bytecodes::_fast_bputfield: __ movb(lo, rax); break;
2688 case Bytecodes::_fast_sputfield: // fall through
2689 case Bytecodes::_fast_cputfield: __ movw(lo, rax); break;
2690 case Bytecodes::_fast_iputfield: __ movl(lo, rax); break;
2691 case Bytecodes::_fast_lputfield:
2692 NOT_LP64(__ movptr(hi, rdx));
2693 __ movptr(lo, rax);
2694 break;
2695 case Bytecodes::_fast_fputfield: __ fstp_s(lo); break;
2696 case Bytecodes::_fast_dputfield: __ fstp_d(lo); break;
2697 case Bytecodes::_fast_aputfield: __ movptr(lo, rax); __ store_check(rcx, lo); break;
2698 default:
2699 ShouldNotReachHere();
2700 }
2701 __ bind(done);
2702 }
2705 void TemplateTable::fast_accessfield(TosState state) {
2706 transition(atos, state);
2708 // do the JVMTI work here to avoid disturbing the register state below
2709 if (JvmtiExport::can_post_field_access()) {
2710 // Check to see if a field access watch has been set before we take
2711 // the time to call into the VM.
2712 Label L1;
2713 __ mov32(rcx, ExternalAddress((address) JvmtiExport::get_field_access_count_addr()));
2714 __ testl(rcx,rcx);
2715 __ jcc(Assembler::zero, L1);
2716 // access constant pool cache entry
2717 __ get_cache_entry_pointer_at_bcp(rcx, rdx, 1);
2718 __ push_ptr(rax); // save object pointer before call_VM() clobbers it
2719 __ verify_oop(rax);
2720 // rax,: object pointer copied above
2721 // rcx: cache entry pointer
2722 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_field_access), rax, rcx);
2723 __ pop_ptr(rax); // restore object pointer
2724 __ bind(L1);
2725 }
2727 // access constant pool cache
2728 __ get_cache_and_index_at_bcp(rcx, rbx, 1);
2729 // replace index with field offset from cache entry
2730 __ movptr(rbx, Address(rcx,
2731 rbx,
2732 Address::times_ptr,
2733 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2736 // rax,: object
2737 __ verify_oop(rax);
2738 __ null_check(rax);
2739 // field addresses
2740 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2741 const Address hi = Address(rax, rbx, Address::times_1, 1*wordSize);
2743 // access field
2744 switch (bytecode()) {
2745 case Bytecodes::_fast_bgetfield: __ movsbl(rax, lo ); break;
2746 case Bytecodes::_fast_sgetfield: __ load_signed_word(rax, lo ); break;
2747 case Bytecodes::_fast_cgetfield: __ load_unsigned_word(rax, lo ); break;
2748 case Bytecodes::_fast_igetfield: __ movl(rax, lo); break;
2749 case Bytecodes::_fast_lgetfield: __ stop("should not be rewritten"); break;
2750 case Bytecodes::_fast_fgetfield: __ fld_s(lo); break;
2751 case Bytecodes::_fast_dgetfield: __ fld_d(lo); break;
2752 case Bytecodes::_fast_agetfield: __ movptr(rax, lo); __ verify_oop(rax); break;
2753 default:
2754 ShouldNotReachHere();
2755 }
2757 // Doug Lea believes this is not needed with current Sparcs(TSO) and Intel(PSO)
2758 // volatile_barrier( );
2759 }
2761 void TemplateTable::fast_xaccess(TosState state) {
2762 transition(vtos, state);
2763 // get receiver
2764 __ movptr(rax, aaddress(0));
2765 debug_only(__ verify_local_tag(frame::TagReference, 0));
2766 // access constant pool cache
2767 __ get_cache_and_index_at_bcp(rcx, rdx, 2);
2768 __ movptr(rbx, Address(rcx,
2769 rdx,
2770 Address::times_ptr,
2771 in_bytes(constantPoolCacheOopDesc::base_offset() + ConstantPoolCacheEntry::f2_offset())));
2772 // make sure exception is reported in correct bcp range (getfield is next instruction)
2773 __ increment(rsi);
2774 __ null_check(rax);
2775 const Address lo = Address(rax, rbx, Address::times_1, 0*wordSize);
2776 if (state == itos) {
2777 __ movl(rax, lo);
2778 } else if (state == atos) {
2779 __ movptr(rax, lo);
2780 __ verify_oop(rax);
2781 } else if (state == ftos) {
2782 __ fld_s(lo);
2783 } else {
2784 ShouldNotReachHere();
2785 }
2786 __ decrement(rsi);
2787 }
2791 //----------------------------------------------------------------------------------------------------
2792 // Calls
2794 void TemplateTable::count_calls(Register method, Register temp) {
2795 // implemented elsewhere
2796 ShouldNotReachHere();
2797 }
2800 void TemplateTable::prepare_invoke(Register method, Register index, int byte_no, Bytecodes::Code code) {
2801 // determine flags
2802 const bool is_invokeinterface = code == Bytecodes::_invokeinterface;
2803 const bool is_invokevirtual = code == Bytecodes::_invokevirtual;
2804 const bool is_invokespecial = code == Bytecodes::_invokespecial;
2805 const bool load_receiver = code != Bytecodes::_invokestatic;
2806 const bool receiver_null_check = is_invokespecial;
2807 const bool save_flags = is_invokeinterface || is_invokevirtual;
2808 // setup registers & access constant pool cache
2809 const Register recv = rcx;
2810 const Register flags = rdx;
2811 assert_different_registers(method, index, recv, flags);
2813 // save 'interpreter return address'
2814 __ save_bcp();
2816 load_invoke_cp_cache_entry(byte_no, method, index, flags, is_invokevirtual);
2818 // load receiver if needed (note: no return address pushed yet)
2819 if (load_receiver) {
2820 __ movl(recv, flags);
2821 __ andl(recv, 0xFF);
2822 // recv count is 0 based?
2823 __ movptr(recv, Address(rsp, recv, Interpreter::stackElementScale(), -Interpreter::expr_offset_in_bytes(1)));
2824 __ verify_oop(recv);
2825 }
2827 // do null check if needed
2828 if (receiver_null_check) {
2829 __ null_check(recv);
2830 }
2832 if (save_flags) {
2833 __ mov(rsi, flags);
2834 }
2836 // compute return type
2837 __ shrl(flags, ConstantPoolCacheEntry::tosBits);
2838 // Make sure we don't need to mask flags for tosBits after the above shift
2839 ConstantPoolCacheEntry::verify_tosBits();
2840 // load return address
2841 {
2842 ExternalAddress table(is_invokeinterface ? (address)Interpreter::return_5_addrs_by_index_table() :
2843 (address)Interpreter::return_3_addrs_by_index_table());
2844 __ movptr(flags, ArrayAddress(table, Address(noreg, flags, Address::times_ptr)));
2845 }
2847 // push return address
2848 __ push(flags);
2850 // Restore flag value from the constant pool cache, and restore rsi
2851 // for later null checks. rsi is the bytecode pointer
2852 if (save_flags) {
2853 __ mov(flags, rsi);
2854 __ restore_bcp();
2855 }
2856 }
2859 void TemplateTable::invokevirtual_helper(Register index, Register recv,
2860 Register flags) {
2862 // Uses temporary registers rax, rdx
2863 assert_different_registers(index, recv, rax, rdx);
2865 // Test for an invoke of a final method
2866 Label notFinal;
2867 __ movl(rax, flags);
2868 __ andl(rax, (1 << ConstantPoolCacheEntry::vfinalMethod));
2869 __ jcc(Assembler::zero, notFinal);
2871 Register method = index; // method must be rbx,
2872 assert(method == rbx, "methodOop must be rbx, for interpreter calling convention");
2874 // do the call - the index is actually the method to call
2875 __ verify_oop(method);
2877 // It's final, need a null check here!
2878 __ null_check(recv);
2880 // profile this call
2881 __ profile_final_call(rax);
2883 __ jump_from_interpreted(method, rax);
2885 __ bind(notFinal);
2887 // get receiver klass
2888 __ null_check(recv, oopDesc::klass_offset_in_bytes());
2889 // Keep recv in rcx for callee expects it there
2890 __ movptr(rax, Address(recv, oopDesc::klass_offset_in_bytes()));
2891 __ verify_oop(rax);
2893 // profile this call
2894 __ profile_virtual_call(rax, rdi, rdx);
2896 // get target methodOop & entry point
2897 const int base = instanceKlass::vtable_start_offset() * wordSize;
2898 assert(vtableEntry::size() * wordSize == 4, "adjust the scaling in the code below");
2899 __ movptr(method, Address(rax, index, Address::times_ptr, base + vtableEntry::method_offset_in_bytes()));
2900 __ jump_from_interpreted(method, rdx);
2901 }
2904 void TemplateTable::invokevirtual(int byte_no) {
2905 transition(vtos, vtos);
2906 prepare_invoke(rbx, noreg, byte_no, bytecode());
2908 // rbx,: index
2909 // rcx: receiver
2910 // rdx: flags
2912 invokevirtual_helper(rbx, rcx, rdx);
2913 }
2916 void TemplateTable::invokespecial(int byte_no) {
2917 transition(vtos, vtos);
2918 prepare_invoke(rbx, noreg, byte_no, bytecode());
2919 // do the call
2920 __ verify_oop(rbx);
2921 __ profile_call(rax);
2922 __ jump_from_interpreted(rbx, rax);
2923 }
2926 void TemplateTable::invokestatic(int byte_no) {
2927 transition(vtos, vtos);
2928 prepare_invoke(rbx, noreg, byte_no, bytecode());
2929 // do the call
2930 __ verify_oop(rbx);
2931 __ profile_call(rax);
2932 __ jump_from_interpreted(rbx, rax);
2933 }
2936 void TemplateTable::fast_invokevfinal(int byte_no) {
2937 transition(vtos, vtos);
2938 __ stop("fast_invokevfinal not used on x86");
2939 }
2942 void TemplateTable::invokeinterface(int byte_no) {
2943 transition(vtos, vtos);
2944 prepare_invoke(rax, rbx, byte_no, bytecode());
2946 // rax,: Interface
2947 // rbx,: index
2948 // rcx: receiver
2949 // rdx: flags
2951 // Special case of invokeinterface called for virtual method of
2952 // java.lang.Object. See cpCacheOop.cpp for details.
2953 // This code isn't produced by javac, but could be produced by
2954 // another compliant java compiler.
2955 Label notMethod;
2956 __ movl(rdi, rdx);
2957 __ andl(rdi, (1 << ConstantPoolCacheEntry::methodInterface));
2958 __ jcc(Assembler::zero, notMethod);
2960 invokevirtual_helper(rbx, rcx, rdx);
2961 __ bind(notMethod);
2963 // Get receiver klass into rdx - also a null check
2964 __ restore_locals(); // restore rdi
2965 __ movptr(rdx, Address(rcx, oopDesc::klass_offset_in_bytes()));
2966 __ verify_oop(rdx);
2968 // profile this call
2969 __ profile_virtual_call(rdx, rsi, rdi);
2971 __ mov(rdi, rdx); // Save klassOop in rdi
2973 // Compute start of first itableOffsetEntry (which is at the end of the vtable)
2974 const int base = instanceKlass::vtable_start_offset() * wordSize;
2975 assert(vtableEntry::size() * wordSize == (1 << (int)Address::times_ptr), "adjust the scaling in the code below");
2976 __ movl(rsi, Address(rdx, instanceKlass::vtable_length_offset() * wordSize)); // Get length of vtable
2977 __ lea(rdx, Address(rdx, rsi, Address::times_4, base));
2978 if (HeapWordsPerLong > 1) {
2979 // Round up to align_object_offset boundary
2980 __ round_to(rdx, BytesPerLong);
2981 }
2983 Label entry, search, interface_ok;
2985 __ jmpb(entry);
2986 __ bind(search);
2987 __ addptr(rdx, itableOffsetEntry::size() * wordSize);
2989 __ bind(entry);
2991 // Check that the entry is non-null. A null entry means that the receiver
2992 // class doesn't implement the interface, and wasn't the same as the
2993 // receiver class checked when the interface was resolved.
2994 __ push(rdx);
2995 __ movptr(rdx, Address(rdx, itableOffsetEntry::interface_offset_in_bytes()));
2996 __ testptr(rdx, rdx);
2997 __ jcc(Assembler::notZero, interface_ok);
2998 // throw exception
2999 __ pop(rdx); // pop saved register first.
3000 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3001 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3002 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3003 __ call_VM(noreg, CAST_FROM_FN_PTR(address,
3004 InterpreterRuntime::throw_IncompatibleClassChangeError));
3005 // the call_VM checks for exception, so we should never return here.
3006 __ should_not_reach_here();
3007 __ bind(interface_ok);
3009 __ pop(rdx);
3011 __ cmpptr(rax, Address(rdx, itableOffsetEntry::interface_offset_in_bytes()));
3012 __ jcc(Assembler::notEqual, search);
3014 __ movl(rdx, Address(rdx, itableOffsetEntry::offset_offset_in_bytes()));
3015 __ addptr(rdx, rdi); // Add offset to klassOop
3016 assert(itableMethodEntry::size() * wordSize == (1 << (int)Address::times_ptr), "adjust the scaling in the code below");
3017 __ movptr(rbx, Address(rdx, rbx, Address::times_ptr));
3018 // rbx,: methodOop to call
3019 // rcx: receiver
3020 // Check for abstract method error
3021 // Note: This should be done more efficiently via a throw_abstract_method_error
3022 // interpreter entry point and a conditional jump to it in case of a null
3023 // method.
3024 { Label L;
3025 __ testptr(rbx, rbx);
3026 __ jcc(Assembler::notZero, L);
3027 // throw exception
3028 // note: must restore interpreter registers to canonical
3029 // state for exception handling to work correctly!
3030 __ pop(rbx); // pop return address (pushed by prepare_invoke)
3031 __ restore_bcp(); // rsi must be correct for exception handler (was destroyed)
3032 __ restore_locals(); // make sure locals pointer is correct as well (was destroyed)
3033 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
3034 // the call_VM checks for exception, so we should never return here.
3035 __ should_not_reach_here();
3036 __ bind(L);
3037 }
3039 // do the call
3040 // rcx: receiver
3041 // rbx,: methodOop
3042 __ jump_from_interpreted(rbx, rdx);
3043 }
3045 //----------------------------------------------------------------------------------------------------
3046 // Allocation
3048 void TemplateTable::_new() {
3049 transition(vtos, atos);
3050 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3051 Label slow_case;
3052 Label done;
3053 Label initialize_header;
3054 Label initialize_object; // including clearing the fields
3055 Label allocate_shared;
3057 ExternalAddress heap_top((address)Universe::heap()->top_addr());
3059 __ get_cpool_and_tags(rcx, rax);
3060 // get instanceKlass
3061 __ movptr(rcx, Address(rcx, rdx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3062 __ push(rcx); // save the contexts of klass for initializing the header
3064 // make sure the class we're about to instantiate has been resolved.
3065 // Note: slow_case does a pop of stack, which is why we loaded class/pushed above
3066 const int tags_offset = typeArrayOopDesc::header_size(T_BYTE) * wordSize;
3067 __ cmpb(Address(rax, rdx, Address::times_1, tags_offset), JVM_CONSTANT_Class);
3068 __ jcc(Assembler::notEqual, slow_case);
3070 // make sure klass is initialized & doesn't have finalizer
3071 // make sure klass is fully initialized
3072 __ cmpl(Address(rcx, instanceKlass::init_state_offset_in_bytes() + sizeof(oopDesc)), instanceKlass::fully_initialized);
3073 __ jcc(Assembler::notEqual, slow_case);
3075 // get instance_size in instanceKlass (scaled to a count of bytes)
3076 __ movl(rdx, Address(rcx, Klass::layout_helper_offset_in_bytes() + sizeof(oopDesc)));
3077 // test to see if it has a finalizer or is malformed in some way
3078 __ testl(rdx, Klass::_lh_instance_slow_path_bit);
3079 __ jcc(Assembler::notZero, slow_case);
3081 //
3082 // Allocate the instance
3083 // 1) Try to allocate in the TLAB
3084 // 2) if fail and the object is large allocate in the shared Eden
3085 // 3) if the above fails (or is not applicable), go to a slow case
3086 // (creates a new TLAB, etc.)
3088 const bool allow_shared_alloc =
3089 Universe::heap()->supports_inline_contig_alloc() && !CMSIncrementalMode;
3091 if (UseTLAB) {
3092 const Register thread = rcx;
3094 __ get_thread(thread);
3095 __ movptr(rax, Address(thread, in_bytes(JavaThread::tlab_top_offset())));
3096 __ lea(rbx, Address(rax, rdx, Address::times_1));
3097 __ cmpptr(rbx, Address(thread, in_bytes(JavaThread::tlab_end_offset())));
3098 __ jcc(Assembler::above, allow_shared_alloc ? allocate_shared : slow_case);
3099 __ movptr(Address(thread, in_bytes(JavaThread::tlab_top_offset())), rbx);
3100 if (ZeroTLAB) {
3101 // the fields have been already cleared
3102 __ jmp(initialize_header);
3103 } else {
3104 // initialize both the header and fields
3105 __ jmp(initialize_object);
3106 }
3107 }
3109 // Allocation in the shared Eden, if allowed.
3110 //
3111 // rdx: instance size in bytes
3112 if (allow_shared_alloc) {
3113 __ bind(allocate_shared);
3115 Label retry;
3116 __ bind(retry);
3117 __ movptr(rax, heap_top);
3118 __ lea(rbx, Address(rax, rdx, Address::times_1));
3119 __ cmpptr(rbx, ExternalAddress((address)Universe::heap()->end_addr()));
3120 __ jcc(Assembler::above, slow_case);
3122 // Compare rax, with the top addr, and if still equal, store the new
3123 // top addr in rbx, at the address of the top addr pointer. Sets ZF if was
3124 // equal, and clears it otherwise. Use lock prefix for atomicity on MPs.
3125 //
3126 // rax,: object begin
3127 // rbx,: object end
3128 // rdx: instance size in bytes
3129 __ locked_cmpxchgptr(rbx, heap_top);
3131 // if someone beat us on the allocation, try again, otherwise continue
3132 __ jcc(Assembler::notEqual, retry);
3133 }
3135 if (UseTLAB || Universe::heap()->supports_inline_contig_alloc()) {
3136 // The object is initialized before the header. If the object size is
3137 // zero, go directly to the header initialization.
3138 __ bind(initialize_object);
3139 __ decrement(rdx, sizeof(oopDesc));
3140 __ jcc(Assembler::zero, initialize_header);
3142 // Initialize topmost object field, divide rdx by 8, check if odd and
3143 // test if zero.
3144 __ xorl(rcx, rcx); // use zero reg to clear memory (shorter code)
3145 __ shrl(rdx, LogBytesPerLong); // divide by 2*oopSize and set carry flag if odd
3147 // rdx must have been multiple of 8
3148 #ifdef ASSERT
3149 // make sure rdx was multiple of 8
3150 Label L;
3151 // Ignore partial flag stall after shrl() since it is debug VM
3152 __ jccb(Assembler::carryClear, L);
3153 __ stop("object size is not multiple of 2 - adjust this code");
3154 __ bind(L);
3155 // rdx must be > 0, no extra check needed here
3156 #endif
3158 // initialize remaining object fields: rdx was a multiple of 8
3159 { Label loop;
3160 __ bind(loop);
3161 __ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 1*oopSize), rcx);
3162 NOT_LP64(__ movptr(Address(rax, rdx, Address::times_8, sizeof(oopDesc) - 2*oopSize), rcx));
3163 __ decrement(rdx);
3164 __ jcc(Assembler::notZero, loop);
3165 }
3167 // initialize object header only.
3168 __ bind(initialize_header);
3169 if (UseBiasedLocking) {
3170 __ pop(rcx); // get saved klass back in the register.
3171 __ movptr(rbx, Address(rcx, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()));
3172 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()), rbx);
3173 } else {
3174 __ movptr(Address(rax, oopDesc::mark_offset_in_bytes ()),
3175 (int32_t)markOopDesc::prototype()); // header
3176 __ pop(rcx); // get saved klass back in the register.
3177 }
3178 __ movptr(Address(rax, oopDesc::klass_offset_in_bytes()), rcx); // klass
3180 {
3181 SkipIfEqual skip_if(_masm, &DTraceAllocProbes, 0);
3182 // Trigger dtrace event for fastpath
3183 __ push(atos);
3184 __ call_VM_leaf(
3185 CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_object_alloc), rax);
3186 __ pop(atos);
3187 }
3189 __ jmp(done);
3190 }
3192 // slow case
3193 __ bind(slow_case);
3194 __ pop(rcx); // restore stack pointer to what it was when we came in.
3195 __ get_constant_pool(rax);
3196 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3197 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::_new), rax, rdx);
3199 // continue
3200 __ bind(done);
3201 }
3204 void TemplateTable::newarray() {
3205 transition(itos, atos);
3206 __ push_i(rax); // make sure everything is on the stack
3207 __ load_unsigned_byte(rdx, at_bcp(1));
3208 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::newarray), rdx, rax);
3209 __ pop_i(rdx); // discard size
3210 }
3213 void TemplateTable::anewarray() {
3214 transition(itos, atos);
3215 __ get_unsigned_2_byte_index_at_bcp(rdx, 1);
3216 __ get_constant_pool(rcx);
3217 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::anewarray), rcx, rdx, rax);
3218 }
3221 void TemplateTable::arraylength() {
3222 transition(atos, itos);
3223 __ null_check(rax, arrayOopDesc::length_offset_in_bytes());
3224 __ movl(rax, Address(rax, arrayOopDesc::length_offset_in_bytes()));
3225 }
3228 void TemplateTable::checkcast() {
3229 transition(atos, atos);
3230 Label done, is_null, ok_is_subtype, quicked, resolved;
3231 __ testptr(rax, rax); // Object is in EAX
3232 __ jcc(Assembler::zero, is_null);
3234 // Get cpool & tags index
3235 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3236 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3237 // See if bytecode has already been quicked
3238 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3239 __ jcc(Assembler::equal, quicked);
3241 __ push(atos);
3242 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3243 __ pop_ptr(rdx);
3244 __ jmpb(resolved);
3246 // Get superklass in EAX and subklass in EBX
3247 __ bind(quicked);
3248 __ mov(rdx, rax); // Save object in EDX; EAX needed for subtype check
3249 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3251 __ bind(resolved);
3252 __ movptr(rbx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3254 // Generate subtype check. Blows ECX. Resets EDI. Object in EDX.
3255 // Superklass in EAX. Subklass in EBX.
3256 __ gen_subtype_check( rbx, ok_is_subtype );
3258 // Come here on failure
3259 __ push(rdx);
3260 // object is at TOS
3261 __ jump(ExternalAddress(Interpreter::_throw_ClassCastException_entry));
3263 // Come here on success
3264 __ bind(ok_is_subtype);
3265 __ mov(rax,rdx); // Restore object in EDX
3267 // Collect counts on whether this check-cast sees NULLs a lot or not.
3268 if (ProfileInterpreter) {
3269 __ jmp(done);
3270 __ bind(is_null);
3271 __ profile_null_seen(rcx);
3272 } else {
3273 __ bind(is_null); // same as 'done'
3274 }
3275 __ bind(done);
3276 }
3279 void TemplateTable::instanceof() {
3280 transition(atos, itos);
3281 Label done, is_null, ok_is_subtype, quicked, resolved;
3282 __ testptr(rax, rax);
3283 __ jcc(Assembler::zero, is_null);
3285 // Get cpool & tags index
3286 __ get_cpool_and_tags(rcx, rdx); // ECX=cpool, EDX=tags array
3287 __ get_unsigned_2_byte_index_at_bcp(rbx, 1); // EBX=index
3288 // See if bytecode has already been quicked
3289 __ cmpb(Address(rdx, rbx, Address::times_1, typeArrayOopDesc::header_size(T_BYTE) * wordSize), JVM_CONSTANT_Class);
3290 __ jcc(Assembler::equal, quicked);
3292 __ push(atos);
3293 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::quicken_io_cc) );
3294 __ pop_ptr(rdx);
3295 __ movptr(rdx, Address(rdx, oopDesc::klass_offset_in_bytes()));
3296 __ jmp(resolved);
3298 // Get superklass in EAX and subklass in EDX
3299 __ bind(quicked);
3300 __ movptr(rdx, Address(rax, oopDesc::klass_offset_in_bytes()));
3301 __ movptr(rax, Address(rcx, rbx, Address::times_ptr, sizeof(constantPoolOopDesc)));
3303 __ bind(resolved);
3305 // Generate subtype check. Blows ECX. Resets EDI.
3306 // Superklass in EAX. Subklass in EDX.
3307 __ gen_subtype_check( rdx, ok_is_subtype );
3309 // Come here on failure
3310 __ xorl(rax,rax);
3311 __ jmpb(done);
3312 // Come here on success
3313 __ bind(ok_is_subtype);
3314 __ movl(rax, 1);
3316 // Collect counts on whether this test sees NULLs a lot or not.
3317 if (ProfileInterpreter) {
3318 __ jmp(done);
3319 __ bind(is_null);
3320 __ profile_null_seen(rcx);
3321 } else {
3322 __ bind(is_null); // same as 'done'
3323 }
3324 __ bind(done);
3325 // rax, = 0: obj == NULL or obj is not an instanceof the specified klass
3326 // rax, = 1: obj != NULL and obj is an instanceof the specified klass
3327 }
3330 //----------------------------------------------------------------------------------------------------
3331 // Breakpoints
3332 void TemplateTable::_breakpoint() {
3334 // Note: We get here even if we are single stepping..
3335 // jbug inists on setting breakpoints at every bytecode
3336 // even if we are in single step mode.
3338 transition(vtos, vtos);
3340 // get the unpatched byte code
3341 __ get_method(rcx);
3342 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::get_original_bytecode_at), rcx, rsi);
3343 __ mov(rbx, rax);
3345 // post the breakpoint event
3346 __ get_method(rcx);
3347 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::_breakpoint), rcx, rsi);
3349 // complete the execution of original bytecode
3350 __ dispatch_only_normal(vtos);
3351 }
3354 //----------------------------------------------------------------------------------------------------
3355 // Exceptions
3357 void TemplateTable::athrow() {
3358 transition(atos, vtos);
3359 __ null_check(rax);
3360 __ jump(ExternalAddress(Interpreter::throw_exception_entry()));
3361 }
3364 //----------------------------------------------------------------------------------------------------
3365 // Synchronization
3366 //
3367 // Note: monitorenter & exit are symmetric routines; which is reflected
3368 // in the assembly code structure as well
3369 //
3370 // Stack layout:
3371 //
3372 // [expressions ] <--- rsp = expression stack top
3373 // ..
3374 // [expressions ]
3375 // [monitor entry] <--- monitor block top = expression stack bot
3376 // ..
3377 // [monitor entry]
3378 // [frame data ] <--- monitor block bot
3379 // ...
3380 // [saved rbp, ] <--- rbp,
3383 void TemplateTable::monitorenter() {
3384 transition(atos, vtos);
3386 // check for NULL object
3387 __ null_check(rax);
3389 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3390 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3391 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3392 Label allocated;
3394 // initialize entry pointer
3395 __ xorl(rdx, rdx); // points to free slot or NULL
3397 // find a free slot in the monitor block (result in rdx)
3398 { Label entry, loop, exit;
3399 __ movptr(rcx, monitor_block_top); // points to current entry, starting with top-most entry
3400 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3401 __ jmpb(entry);
3403 __ bind(loop);
3404 __ cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD); // check if current entry is used
3406 // TODO - need new func here - kbt
3407 if (VM_Version::supports_cmov()) {
3408 __ cmov(Assembler::equal, rdx, rcx); // if not used then remember entry in rdx
3409 } else {
3410 Label L;
3411 __ jccb(Assembler::notEqual, L);
3412 __ mov(rdx, rcx); // if not used then remember entry in rdx
3413 __ bind(L);
3414 }
3415 __ cmpptr(rax, Address(rcx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3416 __ jccb(Assembler::equal, exit); // if same object then stop searching
3417 __ addptr(rcx, entry_size); // otherwise advance to next entry
3418 __ bind(entry);
3419 __ cmpptr(rcx, rbx); // check if bottom reached
3420 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3421 __ bind(exit);
3422 }
3424 __ testptr(rdx, rdx); // check if a slot has been found
3425 __ jccb(Assembler::notZero, allocated); // if found, continue with that one
3427 // allocate one if there's no free slot
3428 { Label entry, loop;
3429 // 1. compute new pointers // rsp: old expression stack top
3430 __ movptr(rdx, monitor_block_bot); // rdx: old expression stack bottom
3431 __ subptr(rsp, entry_size); // move expression stack top
3432 __ subptr(rdx, entry_size); // move expression stack bottom
3433 __ mov(rcx, rsp); // set start value for copy loop
3434 __ movptr(monitor_block_bot, rdx); // set new monitor block top
3435 __ jmp(entry);
3436 // 2. move expression stack contents
3437 __ bind(loop);
3438 __ movptr(rbx, Address(rcx, entry_size)); // load expression stack word from old location
3439 __ movptr(Address(rcx, 0), rbx); // and store it at new location
3440 __ addptr(rcx, wordSize); // advance to next word
3441 __ bind(entry);
3442 __ cmpptr(rcx, rdx); // check if bottom reached
3443 __ jcc(Assembler::notEqual, loop); // if not at bottom then copy next word
3444 }
3446 // call run-time routine
3447 // rdx: points to monitor entry
3448 __ bind(allocated);
3450 // Increment bcp to point to the next bytecode, so exception handling for async. exceptions work correctly.
3451 // The object has already been poped from the stack, so the expression stack looks correct.
3452 __ increment(rsi);
3454 __ movptr(Address(rdx, BasicObjectLock::obj_offset_in_bytes()), rax); // store object
3455 __ lock_object(rdx);
3457 // check to make sure this monitor doesn't cause stack overflow after locking
3458 __ save_bcp(); // in case of exception
3459 __ generate_stack_overflow_check(0);
3461 // The bcp has already been incremented. Just need to dispatch to next instruction.
3462 __ dispatch_next(vtos);
3463 }
3466 void TemplateTable::monitorexit() {
3467 transition(atos, vtos);
3469 // check for NULL object
3470 __ null_check(rax);
3472 const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
3473 const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset * wordSize);
3474 const int entry_size = ( frame::interpreter_frame_monitor_size() * wordSize);
3475 Label found;
3477 // find matching slot
3478 { Label entry, loop;
3479 __ movptr(rdx, monitor_block_top); // points to current entry, starting with top-most entry
3480 __ lea(rbx, monitor_block_bot); // points to word before bottom of monitor block
3481 __ jmpb(entry);
3483 __ bind(loop);
3484 __ cmpptr(rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes())); // check if current entry is for same object
3485 __ jcc(Assembler::equal, found); // if same object then stop searching
3486 __ addptr(rdx, entry_size); // otherwise advance to next entry
3487 __ bind(entry);
3488 __ cmpptr(rdx, rbx); // check if bottom reached
3489 __ jcc(Assembler::notEqual, loop); // if not at bottom then check this entry
3490 }
3492 // error handling. Unlocking was not block-structured
3493 Label end;
3494 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
3495 __ should_not_reach_here();
3497 // call run-time routine
3498 // rcx: points to monitor entry
3499 __ bind(found);
3500 __ push_ptr(rax); // make sure object is on stack (contract with oopMaps)
3501 __ unlock_object(rdx);
3502 __ pop_ptr(rax); // discard object
3503 __ bind(end);
3504 }
3507 //----------------------------------------------------------------------------------------------------
3508 // Wide instructions
3510 void TemplateTable::wide() {
3511 transition(vtos, vtos);
3512 __ load_unsigned_byte(rbx, at_bcp(1));
3513 ExternalAddress wtable((address)Interpreter::_wentry_point);
3514 __ jump(ArrayAddress(wtable, Address(noreg, rbx, Address::times_ptr)));
3515 // Note: the rsi increment step is part of the individual wide bytecode implementations
3516 }
3519 //----------------------------------------------------------------------------------------------------
3520 // Multi arrays
3522 void TemplateTable::multianewarray() {
3523 transition(vtos, atos);
3524 __ load_unsigned_byte(rax, at_bcp(3)); // get number of dimensions
3525 // last dim is on top of stack; we want address of first one:
3526 // first_addr = last_addr + (ndims - 1) * stackElementSize - 1*wordsize
3527 // the latter wordSize to point to the beginning of the array.
3528 __ lea( rax, Address(rsp, rax, Interpreter::stackElementScale(), -wordSize));
3529 call_VM(rax, CAST_FROM_FN_PTR(address, InterpreterRuntime::multianewarray), rax); // pass in rax,
3530 __ load_unsigned_byte(rbx, at_bcp(3));
3531 __ lea(rsp, Address(rsp, rbx, Interpreter::stackElementScale())); // get rid of counts
3532 }
3534 #endif /* !CC_INTERP */
